Daniel Cortland Chapman

Transcription

1 Identification of Targeting Factors Involved in the US2- and US11- Mediated Degradation of Major Histocompatibility Complex Class I Molecules by Daniel Cortland Chapman A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Immunology University of Toronto Copyright by Daniel Cortland Chapman 2016

2 Identification of Targeting Factors Involved in the US2- and US11- Mediated Degradation of Major Histocompatibility Complex Class I Molecules Abstract Daniel Cortland Chapman Doctor of Philosophy Department of Immunology University of Toronto 2016 Human cytomegalovirus uses a variety of mechanisms to evade immune recognition by major histocompatibility complex class I molecules. One mechanism mediated by the immunoevasion proteins US2 and US11 causes rapid disposal of newly synthesized class I molecules by the endoplasmic reticulum-associated degradation pathway. Although several components of this degradation pathway have been identified, evidence suggests that there are still additional host cell proteins that are required by US2 and US11 to target major histocompatibility complex class I molecules for degradation. Two approaches were taken to identify proteins involved in the function of these viral immunoevasion molecules. First was a lentiviral screen for cellular factors involved in US11-mediated degradation. Infection of U373- MG cells expressing US11 with a lentiviral shrna library and screening for those shrnas that increased surface class I levels led to the identification of several novel proteins. ERLIN2, PDIA5, DNAJC3, NACA, and SAE1/UBE2I were all shown to impact US11 function and are promising avenues for future research. Second, I identified cyclophilin C, a peptidyl prolyl isomerase of the endoplasmic reticulum, as a component of US2-mediated immune evasion. ii

3 Cyclophilin C could be co-isolated with US2 and with the class I molecule HLA-A2. Furthermore, it was required at a particular expression level since depletion or overexpression of cyclophilin C impaired the degradation of class I molecules. Overexpression of catalytically inactive cyclophilin C produced a similar effect to the wildtype molecule, suggesting a role in complex formation between US2 and components of the ER-associated degradation machinery, rather than a requirement for peptidyl prolyl isomerase activity. To better characterize the involvement of cyclophilin C in class I degradation, I used LC-MS/MS to detect US2-interacting proteins that were influenced by cyclophilin C expression levels. I identified malectin, P5, and TMEM33 as proteins that increased in association with US2 upon cyclophilin C knockdown. In subsequent validation all were shown to play a functional role in US2 degradation of class I molecules. This was specific to US2 rather than general ER-associated degradation since depletion of these proteins did not impede the degradation of a misfolded substrate, the null Hong Kong variant of α1-antitrypsin. iii

4 Acknowledgments There are many people who I must acknowledge and thank for their support. First of all, I thank my supervisor Dr. David Williams. He has provided such amazing guidance and advice during my time in his laboratory. A fantastic mentor, David has led me to become the scientist I am today, and I value all the lessons he has provided and the opportunities he given to me over the years. I also thank my committee members Dr. Tania Watts and Dr. Allen Volchuk, whose expertise and feedback was critical to these studies. Their patience and consideration was always appreciated. I am grateful for the friendship of the many members of the Williams laboratory that I have worked with during my studies. From the graduate students who made me feel so welcome when I first arrived in Toronto, to those whom I had the pleasure of working with as I began work on this manuscript, I am thankful I had the opportunity to know them all. While I learned many lessons from the Williams lab, I must especially acknowledge Myrna Cohen-Doyle for taking the time to train me in many of the techniques I would apply to my research when I first arrived in the lab and for instilling so many good habits in my work. I also acknowledge the many other members of the Departments of Immunology and Biochemistry. I am happy to consider you not just my colleagues but also my friends, and I am grateful for the many scientific discussions and commiserations. I also value all the encouragement my family has given to me in my studies, as well as the assistance that has made those studies possible. Finally, I must express my gratitude to my wife Laura for all her understanding and patience during my graduate studies. Without her support my training at the University of Toronto would not have been possible. iv

5 Table of Contents Abstract... ii Acknowledgments... iv Table of Contents... v List of Abbreviations... ix List of Tables... xii List of Figures... xiii List of Appendices... xv Chapter 1 Human cytomegalovirus and the evasion of major histocompatibility complex class I peptide presentation Chapter Overview Structure and function of major histocompatibility complex class I Broader role of MHC I within the immune system MHC class II and Non-classical MHC I molecules Synthesis and maturation of MHC I ER export and other trafficking of MHC class I Antigen cross-presentation ER Quality Control Glycosylation and the lectin chaperones Cnx and Crt BiP Grp Protein disulphide isomerases Proline isomerases Unfolded protein response v

6 1.6.7 ER-associated degradation Human cytomegalovirus Epidemiology and pathology Viral structure and genome Subversion of major histocompatibility complex class I peptide presentation Evasion of NK cells US2 and US Rationale and Approach Chapter 2 Materials and Methods Chapter Cell lines and antibodies Plasmids and Constructs Lentivirus production and transduction Plasmid transfection and virus production Lentiviral Infection RNA Interference Flow Cytometry and FACS TRC library lentiviral screen Titre optimization Genomic DNA Isolation Microarray Gene Modulation Array Platform (GMAP) Preliminary microarray data analysis GO Annotation Enrichment and Filtering of Datasets Determination of q-value-based False Detection Rates RNAi Gene Enrichment Ranking (RIGER) analysis Gene Set Enrichment Analysis (GSEA) vi

7 2.6.9 Overlap analysis Xbp-1 splicing assay qpcr Metabolic labeling Pulse chase Pulse protocol with MG132/lactacystin treatment LC-MSMS Mass Spectrometry Analysis Chapter 3 Use of a genome-wide lentiviral shrna screen to identify novel factors involved in US11-mediated degradation of major histocompatibility complex class I Chapter Introduction Screen Design Results Selection of U373-MG cells and US11 for the shrna library screen Full-scale screen of U373-MG + US11 cells for human cell factors involved in US11 function Initial ranking of screen targets for magnitude of effect on US11 function Addressing the issue of multiple hypothesis testing Overlap of screen data with other published ERAD datasets Application of false detection rates to identify statistically significant shrna hits Functional annotation filters for relevant targets Pooling Individual shrna enrichment scores into datasets for single targets Survey of DNAJ proteins Validation of selected targets with individual shrna knockdown Discussion vii

8 Chapter 4 Cyclophilin C assists in US2-mediated degradation of major histocompatibility complex class I molecules by promoting interactions with the ER-associated degradation machinery Chapter Introduction Results Depletion of Cnx, Crt, and DNAJC10 in US2+ and US11+ cells CypC, but not CypB, plays a role in US2-mediated degradation of MHC class I CypC depletion impedes the US2-mediated degradation of newly synthesized MHC I Overexpression of CypC disrupts US2-mediated degradation of MHC class I molecules CypC interacts with US2 and MHC class I US2 and MHC class I interact with a diverse array of proteins Identification of TMEM33, PDIA6, and malectin as novel participants in US2-mediated degradation of MHC I CypC, malectin, PDIA6, and TMEM33 do not affect the degradation of a soluble ERAD substrate Discussion Chapter 5 Discussion and Future Directions Chapter Characteristics of proteins functioning with US2 and US Overlap between US11 lentiviral screen and US2 LC-MSMS data Future Avenues of Research Conclusions References Appendices viii

9 List of Abbreviations ADP ATP Cnx Crt CsA Β2m DMEM EDEM ER FBS FKBP HC HCMV Ig IRL IRS mab MHC I MHC II MICA adenosine diphosphate adenosine diphosphate calnexin calreticulin cyclosporine A beta-2-microglobulin Dulbecco s modified eagle s medium ER degradation enhancing α-mannosidase endoplasmic reticulum fetal bovine serum FK506 binding protein heavy chain human cytomegalovirus immunoglobulin internal repeat (adjacent to UL region) internal repeat (adjacent to US region) monoclonal antibody major histocompatibility complex class I major histocompatibility complex class II MHC class I chain-related A ix

10 MICB MIIC NBD NK NKT OST pab PBS PCR PDI PLC RT-PCR SBD UDP UGGT MHC class I chain-related B MHC II compartment nucleotide binding domain natural killer natural killer T oligosaccharyl transferase polyclonal antibody phosphate-buffered saline polymerase chain reaction protein disulfide isomerase protein loading complex reverse transcription polymerase chain reaction substrate binding domain uridine diphosphate UDP1-glucose:glycoprotein glucosyltransferase UL unique long region UPR unfolded protein response US unique short region SDS SDS-PAGE sodium dodecyl sulfate sodium dodecyl sulfate-polyacrylamide gel electrophoresis x

11 TCR T cell receptor Th T helper TRL TRS terminal repeat (adjacent to UL region) terminal repeat (adjacent to US region) xi

12 List of Tables Table 2.1. sirna pairs used for depletion experiments Table 2.2. Individual shrnas used for depletions Table 3.1. Individual shrnas excluded from further analysis due to unusually low p-values. 101 Table 3.2. Overlap of lentiviral screen data with proteomic results of Christianson et al. [87]. 106 Table 3.3. All shrna hits with q-values less than Table 3.4. GO and other annotations enriched in the top 5% of shrna hits Table 3.5. GO and other annotations enriched in the top 5% of shrna hits identified with 2 or more shrnas Table 3.6. Hits from the top 5% of enriched shrnas annotated as involved in a proteasomal ubiquitin-dependent catabolic process Table 3.7. GO annotated shrna hits identified with 2 or more shrnas from the top 5% of shrnas identified as enriched in the high MHC class I group Table 3.8. Top forty enriched gene targets identified through RIGER analysis Table 3.9. Gene sets identified through GSEA from a collection of curated databases in the RIGER results with a q-value less than Table 5.1. Proteins identified in analysis of both US2- and US11-linked hits xii

13 List of Figures Figure 1.1. Structure of MHC I molecules Figure 1.2. Interactions between antigen presenting cells and CD8+ T cells Figure 1.3. Structure of MHC II molecules Figure 1.4. Maturation of MHC class I Figure 1.5. Pathways of protein quality control within the ER Figure 1.6. Sugar structure of newly added N-linked glycans Figure 1.7. Selected structures of ER quality control family members Figure 1.8. ER-associated degradation Figure 1.9. Structure of US2 and US Figure 3.1. Testing of HeLa cells for use in a lentiviral screen Figure 3.2. Testing of U373-MG cells for use in a lentiviral screen Figure 3.3. General workflow of the U373-MG + US11 cell lentiviral screen Figure 3.4. Visual Summaries of the lentiviral screen output and basic filtering processes Figure 3.5. Statistical analysis of screen output through q-values and FDR Figure 3.6. Preliminary validation of selected targets with individual shrna knockdown Figure 3.7. Knockdown of PDIA5 disrupts US11-mediated ERAD of MHC class I Figure 3.8. ERLIN2 shrna depletion indicates involvement in US11-mediated MHC class I ERAD xiii

14 Figure 3.9. Depletion of several other host cell factors increases surface MHC class I in U373- MG + US11 cells Figure DNAJC3 and not DNAJA4 depletion increases MHC I levels in US11+ cells Figure 4.1. Depletion of DNAJC10, calnexin, and calreticulin using specific sirnas Figure 4.2. Depletion of Crt, Cnx, or DNAJC10 does not disrupt US2- or US11-mediated degradation of MHC I Figure 4.3. Depletion of CypC but not CypB increases MHC class I surface expression in US2- expressing cells Figure 4.4. Depletion of CypC but not CypB increases total MHC class I in US2-expressing cells Figure 4.5. Increased MHC I expression upon CypC depletion is not due to US2 instability or ER stress Figure 4.6. Both depletion and overexpression of CypC stabilize MHC I in cells expressing US Figure 4.7. CypC co-isolates with US2 and MHC class I molecules Figure 4.8. Identification of US2-associated proteins that are affected by CypC depletion or overexpression Figure 4.9. STRING analysis of proteins identified in US2-3xHA isolations Figure STRING analysis of proteins identified in HA-HLA-A68 isolations Figure TMEM33, malectin, and PDIA6 co-isolate with US Figure US2-associated proteins identified by modulating CypC expression participate in US2 degradative functions Figure Involvement of novel US2-associated proteins in other ERAD pathways xiv

15 List of Appendices Appendix 1. Lentiviral screen dataset of targets identified in the top 5% of enriched shrnas with two or more independent shrnas Appendix 2. Q-values calculated from the complete shrna dataset to determine false detection rates Appendix 3. RIGER analysis of shrna enrichment screen data Appendix 4. Peaks Studio 7 data from LC-MS/MS analysis of US2-3xHA interacting proteins following background subtraction and normalization to bait protein Appendix 5. Peaks Studio 7 data from LC-MS/MS analysis of HA-HLA-A68 interacting proteins following background subtraction Appendix 6. Analysis of Appendix 4 results using the GO Ontology database and PANTHER Overrepresentation Test Appendix 7. Analysis of Appendix 5 results using the GO Ontology database and PANTHER Overrepresentation Test Appendix 8. Publications xv

16 Chapter 1 Human cytomegalovirus and the evasion of major histocompatibility complex class I peptide presentation 1

17 1 Chapter Overview This thesis will describe experiments studying the cellular proteins required for the degradation of major histocompatibility complex class I (MHC I) molecules by the human cytomegalovirus (HCMV) immunoevasion molecules US2 and US11. Experiments utilizing a lentiviral screen to identify novel proteins involved in US11 function and the role of cyclophilin C in US2-mediated degradation of MHC I will be described. To these ends, a number of concepts and ideas required to understand the context of these experiments will be introduced in this Chapter. As the experimental data in this thesis focuses on human cell culture systems and HCMV evasion molecules, the information presented here will predominantly be restricted to literature on human antigen presentation and immune function. Classical MHC I molecules are present at the cell surface and bind and present endogenously derived peptides to T cells. Immune recognition of a foreign/non-self peptide can lead to killing of the presenting cell. As MHC I molecules will be the focus of this Chapter, their structure and function will be described first, followed by an introduction to MHC I s broader role within the immune system. A special focus will be made on CD8+ T cells, with a description of their maturation and activation. Major histocompatibility complex class II (MHC II) and non-classical MHC class I molecules will be described next to illustrate the diverse roles filled by this family of proteins. This will be followed by a discussion of MHC I maturation, from initial translation of the MHC I heavy chain (HC) and loading with a high affinity peptide by the peptide loading complex (PLC), to export from the ER to the cell surface. 2

18 As with other proteins synthesized within the ER, MHC I is subject to strict quality control and requires specific protein machinery for proper protein folding and maturation. The molecular chaperones, thiol oxidoreductases, and proline isomerases that act within the ER will be presented to place the MHC I-specific machinery into the context of broader ER quality control processes that support the proper folding of nascent proteins and that correct misfolded proteins. As MHC I is also subject to degradation when it is unable to fold correctly, an overview of the ER-associated degradation (ERAD) pathway and its recognition and clearance of misfolded substrates and other targets will be provided. The presence of multiple processes to ensure MHC I is correctly assembled and successfully exported to the cell surface is not surprising when one considers its importance in presenting viral and other foreign peptide antigens to CD8+ cytotoxic T cells. However, this does not mean that pathogens are unable to combat this system. To counter MHC I peptide presentation, viruses such as HCMV have evolved numerous immunoevasion mechanisms to disrupt this presentation and avoid recognition. For the final portion of this Chapter HCMV will be introduced and a number of the immunoevasion mechanisms relating to MHC I will be discussed. This Chapter will then conclude with a detailed description of the immune evasion molecules US2 and US11, as these two proteins will be the focus of the experiments described in Chapters 3 and Structure and function of major histocompatibility complex class I Major histocompatibility complex class I molecules present endogenously derived peptides to CD8+ cytotoxic T cells. Recognition of a foreign or non-self epitope leads to 3

19 activation of the CD8+ T cell and killing of the presenting cell. This process of presentation and recognition is critical for adaptive immunity against numerous viral pathogens, intracellular bacteria, and oncogenic cells. In addition, normal CD8+ T cell production and function requires the presence of peptide-loaded MHC I on the surface of cells in its developmental environment [1-3]. MHC I consists of three separate components: a heavy chain (HC) anchored in the membrane through a single transmembrane domain, a soluble subunit known as β2- microglobulin (β2m), and a short peptide produced by the proteasome and processed by other proteases. All are required for normal MHC I stability and export to the cell surface. The heavy chain itself contains three ER lumenal domains, termed α1, α2, and α3, the transmembrane domain, and a short cytosolic tail (Figure 1.1A). The α1 and α2 domains fold together to form a peptide-binding groove, consisting of two alpha helices and 8 anti-parallel beta sheets that form the floor of the groove. The α3 domain adopts an immunoglobulin fold (Figure 1.1B) that interacts with both the binding groove and β2m. The MHC I molecules are encoded within the MHC locus on chromosome 6 in humans. There are three polymorphic loci, HLA-A, HLA-B, and HLA-C, that are each co-dominantly expressed in any one individual [4]. This leads to up to 6 different MHC I molecules being present on the cell surface, with most nucleated cells expressing MHC I to some degree. In mice there are also three polymorphic loci, known as H-2D, H-2K, and H-2L, which are present on chromosome 17. In both species the alleles present at each locus tend to be closely linked, and a group of alleles commonly inherited together are referred to as a haplotype, with one individual expressing the full complement of alleles from their two inherited haplotypes. 4

20 Figure 1.1. Structure of MHC I molecules. (A) MHC class I consists of a heavy chain containing the α1, α2, and α3 extracellular domains, a single transmembrane domain and a short cytoplasmic tail. The α1 and α2 domains fold together to form a peptide-binding groove with a preference for 8-10 amino acid peptides that contain the correct anchor resides. A second protein subunit, the soluble β2m, associates with the heavy chain and is required for its stability. (B) Crystal structure of HLA-B27 (PDB: 1HSA) [5]. 5

21 In inbred mouse strains these haplotypes are referred to by a superscript. For example, the d haplotype MHC I molecules are referred to as H-2D d, H-2K d, and H-2L d. In humans, a system of numbering has developed to track different alleles, following the format of HLA- A*02:101:01:02N. The sequence of four numbers separated by colons refers to the allele group, the specific HLA protein, any synonymous mutations within the coding region, and any mutations within the non-coding regions, respectively. The letter following the fourth number is a suffix used to describe the expression status of the particular allele, in this case a null allele that is not expressed. Although the full designation is the most accurate, it is common to use a shortened version that only refers to one or both of the first two numbers, such as HLA-A02. In addition to the HLA-A, B, and C molecules just described, there are also many other nonclassical MHC I family members encoded within the human genome. In contrast to the classical MHC I that present peptides, these non-classical MHC I molecules play diverse roles in both immune and non-immune processes. Some of these non-classical molecules will be introduced and described later in this Chapter to provide a better overview of the MHC I protein family. The peptide binding groove of MHC I plays a key role in its function in presenting peptides to the immune system. The grooves of MHC I molecules have a preference for peptides approximately 8-10 amino acids in length. When peptides are bound, the N- and C-terminal ends tend to be buried further into the groove than the central portion, which bulges out. Each MHC I molecule is able to bind a diverse range of peptides, although MHC I alleles have different binding preferences for peptides. This is accomplished through the binding of anchor residues. Within each MHC I molecule are a number of binding pockets that recognize particular amino acid side chains, leading to recognition of specific peptide motifs. For instance, one common 6

22 motif involves a C-terminal hydrophobic residue and an aromatic residue at the P2 position (one away from the N-terminus). For the other amino acids within the peptide, interactions may depend on the peptide backbone rather than the side chains, thereby allowing variation in the bound sequence. As the MHC I binding grooves are very polymorphic, different MHC I alleles have evolved to bind different peptide motifs, while still allowing a large range of peptides to be bound and presented to T cells [6]. On the cell surface, an MHC I molecule loaded with peptide can interact with the T-cell receptor (TCR) on the surface of CD8+ T cells. This protein heterodimer is composed of an α- chain and β-chain and is associated with the CD3 ε, γ, δ, and ζ chains involved in signal transduction following TCR binding. The αβ TCR binds on top of the MHC I α1 and α2 domains, interacting with the MHC I helices creating the binding groove border and with the peptide itself. Specificity to MHC I is created by an associated co-receptor on the T cell known as CD8, which binds to the α3 domain of MHC I. In discriminating between different peptides and MHC I molecules, variable loops on the TCR created by RAG-mediated recombination tend to interact with the peptide itself, whereas more conserved regions interact with the alpha helices, creating a situation where variations in bound peptide can lead to TCR discrimination of different MHC I molecules [7]. 1.3 Broader role of MHC I within the immune system MHC I is one component within a complex network of molecular and cellular interactions but still plays roles within both the innate and adaptive immune systems. In particular, MHC I is important in the function of antigen presenting dendritic cells, CD8+ T cells, and natural killer cells. 7

23 Dendritic cells (DCs) specialize in immune surveillance with many specialized lineages [8-10] performing specific and unique immune roles. DCs are often identified with their role of phagocytizing and engulfing foreign antigens from infected and dead cells at peripheral locations and presenting these exogenous antigens to other immune cells. Tissue-resident DCs are scattered throughout the periphery and survey the antigen environment until they are activated through exposure to pathogen-associated molecular patterns (PAMPs) derived from a pathogen or from tissue damage and inflammation. This signalling through toll-like and other receptors causes the DC to enter an activated state and mature, leading it to upregulate co-stimulatory ligands on its surface and migrate to the lymph tissues. Here DCs interact with the cells of the adaptive immune system and initiate an adaptive response to the pathogen. Importantly, this presentation includes the display of antigens on MHC I molecules to the local immune cell population, leading to CD8+ T cytotoxic cell activation [8-10]. CD8+ T cells possess a specific T cell receptor (TCR) able to interact with MHC I and interrogate the MHC I-peptide complex. In response to TCR signals reaching an appropriate signaling threshold, a naïve T cell is activated, and in the case of CD8+ T cells differentiates and becomes competent for cell killing, becomes anergic and unresponsive as a form of peripheral tolerance, or becomes apoptotic. Activated CD8+ T cell s primary mechanism of activity is killing of cells presenting a recognized peptide on MHC I, although they can also secrete cytokines. This cell death occurs either through release of perforin/granzyme granules, or through FasL signaling on the T cell to Fas on the target, both of which trigger the caspase cascade and eventual apoptosis [11]. In addition to cytolysis, CD8+ T cells can also express and secrete a number of cytokines to promote the immune response. IFN-γ and TNF from CD8+ T cells can recruit neutrophils and macrophages as well as activate them, induce general inflammation, or interfere with pathogen biology [11]. 8

24 Activation of CD8+ T cells is often described through a three-signal model (Figure 1.2). Initially TCR and CD8 binding to peptide-mhc I and signaling through CD3 constitutes signal one. Signal two is usually provided by binding of the co-stimulatory receptor CD28 to CD80 and CD86, which are up regulated on DCs following their activation. Finally, a third signal is typically provided by the secreted cytokines IL-12 or type I interferon [12], with IL-2 playing a downstream role in proliferation and differentiation [12,13]. It is important to note that this is an oversimplification of T cell activation. There are many additional stimulatory and inhibitory signals which can modulate this process and influence the downstream effector responses of the activated CD8+ T cells. Whether the activating CD8+ T cell receives the correct signals depends in large part on the activation state of the antigen presenting cell. Once the naïve T cell has encountered a DC or other APC presenting its ligand peptide-mhc I complex, several outcomes are possible depending on the activation state of the DC. Presentation by an immature DC leads to apoptosis, anergy, or both, allowing for elimination of self-reactive T cells within the periphery [14]. In contrast, interaction with an activated DC leads to a productive CD8+ T cell that is capable of killing presenting targets. CD4+ T cells, to be discussed specifically within the next section, are able to further support CD8+ T cell activation by allowing the CD8+ T cell to bind CD40L on CD4+ T helper cell [15], which allows for cytotoxic CD8+ T cells that are able to be stimulated repeatedly and lead to long term immune memory. 9

25 Figure 1.2. Interactions between antigen presenting cells and CD8+ T cells. Activation of CD8+ T cells through presentation of foreign peptides on MHC class I can be thought to occur following three signals. First, the TCR engages the MHC I-β2m-peptide complex, leading to signaling through the TCR/CD3 complex, the kinases LCK, Fyn, and ZAP70, and the phosphatase CD45 (1). Second, engagement of CD28 by CD80 or CD86 on the APC results in secondary signaling required for T cell activation and avoidance of anergy (2). Third, cytokine signals from the APC, such as IL-12 or type I interferon, provide additional activating signals to the naïve T cell, ensuring its activation, survival, and cell number expansion [7,12,15]. 10

26 A more innate immune cell that can also recognize MHC I is the natural killer (NK) cell. NK cells can be rapidly activated without prior priming and quickly migrate to sites of inflammation. While traditionally thought of as short-lived cells without immune memory, recent work has shown that steady state populations of NK cell subsets are expanded following activation, providing longer-term protection from specific pathogens [16]. These bone marrowderived cells do not express TCR receptors and instead possess a host of activating and inhibitory receptors, many of which recognize self proteins. The integration of these signals leads to a number of potential outcomes from an NK cell-target cell interaction. In the case of missing self, the lack of inhibitory signals results in NK cell activation and target killing. Alternatively, in induced self recognition, the upregulation of cellular stress markers can also lead to target cell death, and survival of the target cell requires both sufficient inhibitory signals as well as lack of activating signals [17,18]. Once activated, NK cells possess a number of effector functions. They possess cytolytic activity granted through granules containing perforin and granzyme B, which can be released upon activation. Alternatively, NK cells can induce apoptosis in target cells through receptor ligand interactions, such as the TRAIL/TRAIL-R pathway [19] and the Fas/FasL pathway. NK cells can also release anti-microbial peptides to act against bacteria, often by disrupting their cell walls, and produce a number of cytokines, in particular IFN-γ [18]. 1.4 MHC class II and Non-classical MHC I molecules Major histocompatibility complex class II (MHC II) molecules perform a related function to MHC I. Unlike MHC I, they do not bind β2m and instead consist of the α and β polypeptides, each anchored to the membrane through a single transmembrane domain followed by a short cytoplasmic tail (Figure 1.3). 11

27 Figure 1.3. Structure of MHC II molecules. (A) MHC class II consists of two heavy chains, each containing two extracellular domains (α1 and α2, or β1 and β2), a transmembrane domain, and short cytoplasmic tail. The α1 and β1 domains form the peptide-binding groove, which can bind more extended peptides, usually amino acids in length. (B) Crystal structure of HLA-DR1 (1DLH) [20]. 12

28 The N-terminal α1 and β1 domains form a binding groove similar to that in MHC I, though it is capable of accommodating longer peptides greater than 13 amino acids in length. MHC II displays peptides derived from within the endosomal compartments, such as those from phagocytosed extracellular pathogens. However, MHC II is not limited to internalized peptides, as epitopes have been identified bound to MHC II that derive from nearly all cellular compartments [21]. These proteins enter the endocytic pathway either through internalization from the plasma membrane or through autophagic degradation of intracellular material [21]. MHC II is presented on a more limited range of cell types than MHC I. MHC II also does not interact with CD8+ cytotoxic T cells, and instead is recognized by CD4+ helper T cells, assisted by the CD4 co-receptor that has a function parallel to that of CD8 (CD4 binds to the β chain β2 domain whereas CD8 binds to the MHC I heavy chain α3 domain [22]). Following activation CD4+ T cells proliferate, driven by autocrine and paracrine secretion of IL-2. The progenitors of these activated CD4+ T cells will differentiate into various subtypes, including effector Th cells, memory Th cells, follicular Th cells, and in some cases regulatory Th cells. The many subtypes [15,23-27] of effector Th cells actively secrete cytokines, proteins, or peptides to carry out an immune response, upregulate stimulatory molecules on their cell surface, and drive activation and differentiation of other immune cell types. Memory Th cells persist after the initial inflammatory event and act as a latent effector cell that retains the antigen affinity of the originally activated naïve CD4+ T cell. Follicular Th cells fulfill several important roles in the generation of plasma and memory B cells, including initiation of germinal centre formation, providing survival and proliferation signals to B cells, and signaling to follicular dendritic cells. 13

29 Finally, regulatory Th cells act to negatively regulate the immune response rather than activate it, and play an important role in self-limiting the immune response and preventing autoimmune reactivity [15]. In additional to the classical MHC I (HLA-A, B, and C) and MHC II (HLA-DP, DQ, and DR) molecules that have been described, there are a number of non-classical MHC class I that play other immune-related roles [28,29]. Non-classical MHC I molecules are relatively diverse in structure and function. Two examples of non-classical MHC I molecules that strongly resemble the classical MHC I are HLA-E and HLA-G, both of which bind and present peptides. Others are more diverse in structure, as seen by the non-peptide binding HLA-F, the glycolipidbinding CD1, and the peptide- and β2m-lacking MHC class I chain-related A (MICA) and MHC class I chain-related B (MICB). HLA-E is a non-classical MHC I molecule that still presents peptide and is β2m-associated. It can be bound by the inhibitory NK cell receptor CD94/NKG2A and the activating receptor CD94/NKG2C/E [30], suggesting some discrimination of peptide by the innate immune system. It can also interact with TCR receptors. The peptide groove of this molecule only binds to particular peptides, particularly the signal sequences from certain HLA-A, -B, -C, and -G molecules. Normally expressed at low levels on the cell surface, HLA-E is able to strongly block NK cell activation through CD94/NKG2A signalling [29-31]. HLA-G plays a tolerogenic role within the placenta. It binds β2m and peptide, and like HLA- E can interact with the TCR, as well as the NK receptors ILT-2 (LIR-1), ILT-4 (LIR-2), and KIR2DL4. In fact, HLA-G has a stronger affinity for these receptors than does HLA-A, -B, or - C. Some isoforms of HLA-G are membrane bound, but others are missing the transmembrane domain and are secreted. As the expression of this non-classical MHC I can be regulated on 14

30 myeloid cells by IL-10 and interferons, it is also a potential inhibitory ligand to NK cells, monocytes, macrophages, B cells, and some subsets of CD4+ and CD8+ T cells [29,31]. HLA-F is similar to HLA-E in structure and is a type I transmembrane protein. Much less has been uncovered about its role compared to HLA-E and -G, though it is known that most HLA-F is located intracellularly with some detected on the surface of activated B, T, and NK cells. Unlike HLA-E or HLA-G, HLA-F does not bind to peptide. This molecule may play a role in protection of the fetus from the maternal immune system, and has also been shown to bind to ILT-2 and ILT-4 when it is present at the cell surface. Both ILT-2 and ILT-4 are members of the LIR family of receptors, and when expressed on the surface of immune cells can provide an inhibitory signal following binding to MHC I molecules [29,31]. CD1 molecules differ from most other classical and non-classical MHC I molecules as they bind to glycolipids rather than peptide, though they are still β2m-associated [31]. As with MHC I, the CD1 heavy chain consists of three ER lumenal/extracellular domains, a transmembrane domain, and a short cytosolic tail. There are five CD1 family members, CD1a-c (group 1), CD1d (group 2), and CD1e (group 3). CD1 molecules are recognized by particular subsets of T cells. CD1d is recognized by γδ T cells [28] and natural killer T cells, CD1a, CD1b, and CD1c are recognized by other non-invariant αβ T cell populations, and CD1c is also recognized by certain γδ T cell subsets [31]. The natural killer T cells (NKTs) typically bear αβ TCRs, but their function is distinct from that of the T cells mentioned previously. The NKT cells can be broken into two groups in humans, referred to as type I and type II. Typically type I NKT cells (referred to here as invariant NKT cells) express semi-invariant TCR receptors that are restricted to recognition of glycolipid antigens presented on CD1d instead of MHC I or MHC II [32]. Type II NKTs also recognize 15

31 lipids in the context of CD1d, but they do not recognize the model antigen α-galactosylceramide and they have a broader repertoire of TCRs [33]. The γδ T cells are most common in the gut mucosa, forming part of the population of intraepithelial lymophocytes, and are also present within the dermis and other mucosal sites [34]. While bearing TCRs, they do not seem to be dependent on antigen processing in the same manner as the classical T cells, or on presentation of peptide epitopes by MHC I or II. In humans γδ T cells are capable of expanding rapidly during acute infection, and may sense danger signals associated with particular classes of pathogens [35]. Despite the fact that molecules such as CD1d resemble MHC I and MHC II, the TCRs of the NKT cells that interact with them do not dock in the same orientation as classical MHC I, adopting a more parallel conformation [36]. CD1d presentation of lipid antigens was originally characterized for its presentation of the α-galactosylceramide (αgalcer) epitope isolated from a marine sea sponge, though the normal epitope is suspected of being of endogenous origin. The TCR receptors recognizing it do not appear to have the same specificity for presentation of a particular antigen as that observed for MHC I and MHC II. Rather, different antigens presented within CD1d induce differing degrees of downstream signaling in the NKT cell, with only strong signaling leading to activation and a response. However, the strength of signal needed may also be influenced by cytokines such as IL-12 [37]. The non-classical MHC I proteins are also well-represented among a group of molecules that can be upregulated by stressed, infected, or malignantly transformed cells, acting as markers of cellular dysfunction. These proteins act as ligands for a number of immune cells, particularly NK cells, and can target cells towards apoptosis and cell killing [29]. MICA and MICB do not bind peptide and are not β2m-associated, though they do have the standard α1 to α3 16

32 domains and are type I transmembrane proteins. Each is capable of interacting with TCRs of γδ lineage cells and with NKG2D dimers on NK cells [29]. They typically function as stressinduced activating ligands for NK cells and populations of CD8+ T cells that express NKG2D [29]. As with MICA and MICB, the UL16 binding proteins (ULBPs) do not present peptide, but can interact with the TCR of γδ cells or with NKG2D. ULBP1-3,5, and 6 lack an α3 domain and are anchored to the membrane through a glycosyl-phosphatidylinositol anchor, while ULBP4 is predicted to be anchored by a transmembrane sequence [29]. The ULBPs are also like MICA and MICB in that they are induced by cell stresses, including stress induced by human cytomegalovirus infection (excluding ULBP4) [29]. While not an exhaustive list of non-classical MHC I molecules, these representatives provide an idea of the roles played by these molecules in the immune response. 1.5 Synthesis and maturation of MHC I Now that the classical HLA-A, B, and C molecules have been placed into the context of the larger MHC I protein family and of T cell and NK cell immune function, it is worthwhile to take a closer look their synthesis. Maturation of MHC I in human cells is a complex process, with numerous chaperones and quality control components acting cooperatively and in a stepwise manner, portrayed in Figure 1.4. Briefly, following translocation of the nascent MHC I heavy chain (HC) into the endoplasmic reticulum (ER), it rapidly associates with the membrane anchored chaperone calnexin (Cnx) and the associated protein disulphide isomerase (PDI) ERp57. Initial domain folding and disulphide formation is followed by binding to β2m. In human cells Cnx is then replaced with a soluble Cnx ortholog, calreticulin (Crt). 17

33 Figure 1.4. Maturation of MHC class I. (Panel A) MHC I maturation begins in the ER following translocation of the heavy chain. Initially, the molecular chaperone calnexin and thiol oxidoreductase ERp57 bind the heavy chain (HC) and assist with early domain folding and disulphide bond formation (1). This is followed by association with the soluble subunit β2m and entry into the peptide-loading complex (2). The PLC consists of the molecular chaperone calreticulin, ERp57, tapasin, and peptide transporter TAP (consisting of the TAP1 and TAP2 ABC transporter proteins). Peptide editing is carried out by the ERp57-tapasin heterodimer, exchanging low affinity peptides for high affinity ones. Loading with a high affinity peptide is associated with release from the peptide-loading complex (3). (Panel B) Following exit from the PLC, MHC I loaded with peptide clusters at ER exit sites (1) prior to exit from the ER in COP II vesicles (2). Following export to the Golgi the MHC I molecules either continue on to the cell surface (3) or are recycled back to the ER for further quality control. 18

34 These initial folding steps and β2m association allows a peptide-receptive conformation of the HC and entry into the peptide loading complex (PLC), which consists of MHC I associated with β2m, Crt, ERp57, tapasin, TAP1/TAP2, and Bap31. MHC I peptides are predominantly produced by the proteasomal degradation of cytosolic proteins. Peptides 8-16 amino acids long are transported into the ER by the ABC transporter associated with antigen processing (TAP, formed by the two subunits TAP1 and TAP2). Within the PLC high affinity peptides are exchanged for low affinity ones in the HC binding groove. Binding with high affinity peptide is associated with MHC I release from the PLC and ER export involving Bap31 [1-3]. MHC I then passes through the Golgi apparatus on its way to the cell surface. The C-terminal amino acid of HLA-A,B, and C acts as an export signal for MHC I molecules [38], with differing interactions between MHC I and ER matrix proteins producing the different export rates observed for MHC I molecules expressed from different alleles [39]. Following the initial translocation of MHC I into the ER, disulphide bond formation within the heavy chain occurs in the α2 and α3 domains. Formation of these bonds is enhanced by the association of β2m [40], and both of these steps occur rapidly following translocation [41,42]. They are assisted by several proteins that transiently associate with MHC I early in its maturation process, namely the molecular chaperones BiP and calnexin, and the thiol oxidoreductase ERp57 [43]. Calnexin has been a particular focus of studies on MHC I maturation. This lectin chaperone recognizes monoglucosylated N-linked glycans, which are present on MHC I. An important tool in studying calnexin s role has been castanospermine (CST), a glucosidase inhibitor which prevents formation of the sugar structure recognized by calnexin. When calnexin recognition of MHC I HC was disrupted with CST there was a modest effect on the rate of disulphide bond formation [40]. However, additional studies in human Cnxdeficient cells showed no defect in the assembly or quality control of MHC I [44], suggesting it 19

35 is dispensable for human MHC I maturation, though upregulation of other chaperones in these cells may be compensating for its loss [45]. The thiol oxidoreductase ERp57 is one member of the large PDI family of proteins involved in disulphide bond formation and rearrangement. ERp57 has been found in several disulphide-bonded reaction intermediates with MHC I heavy chains [46,47], and loss of ERp57 did slow formation of the second of the two MHC I disulphide bonds in mouse cells [42,48]. For MHC I at least, ERp57 appears able to recognize early MHC I directly, without assistance of calnexin or calreticulin (with which it associates) [48]. However, it has been more difficult to show a functional effect in human cells, as one study instead showed an effect of PDIA1 knockdown rather than ERp57 [49]. This may indicate that in human cells PDIA1 may be better able to substitute for ERp57, or may suggest that in humans PDIA1 plays a more significant role in MHC I maturation. Following these initial folding and assembly events, the MHC I HC and β2m heterodimer associate with a new group of ER proteins that assist in its loading with a high affinity peptide. Together, these ER proteins form the peptide loading complex, or PLC. The PLC contains the ATP-binding cassette transporter TAP (consisting of the two multi-pass transmembrane proteins TAP1 and TAP2), tapasin, the MHC I HC-β2m heterodimer, Crt and ERp57. Bap31 and PDIA1 have also been found to be associated [1-3]. The TAP transporter is the largest component of the PLC and is bridged to MHC I by tapasin, which acts as a central component of the PLC [1-3]. TAP transports peptides from the cytosol to the ER lumen in an ATP-dependent process. It does not transport all peptides equally, and shows a preference for particular sequences and lengths. In particular, human TAP discriminates based on the first three N-terminal amino acids and the C- terminal amino acid of cytosolic peptides, and prefers peptides that are 8-16 residues in length 20

36 [50]. Importantly, the N- and C-terminal amino acids preferred by TAP are similar to the anchor residues bound by the peptide binding groove of many MHC I alleles. When the TAP transporter is disrupted or absent, MHC I is unable to present most peptides at the cell surface to CD8+ T cells [1-3]. MHC I molecules bound to low affinity peptides (or empty MHC I) are not as stable as properly loaded MHC I and are largely retained in the ER or cycle between the ER and cis- Golgi. If not loaded with a high affinity peptide, they can eventually be destroyed through ERassociated degradation [51-53]. The nature of the peptides available to MHC I is also influenced prior to and after transport by TAP. In the cytosol the standard proteasome is capable of producing peptides suitable for transport by TAP and binding by MHC I, but a different proteasome subunit composition is present within haematopoietic cells as well as cells exposed to IFN-γ or TNF-β. These induced subunits, known as LMP2, LMP7, and MECL-1, replace the β1, β2, and β5 constitutive subunits to form the immunoproteasome. The immunoproteasome can have a dramatic impact on peptide presentation by MHC I. In mice lacking these alternate subunits the levels of MHC I on the cell surface of a range of different immune cell types is decreased by approximately 50% compared to wildtype mice. CD8+ T cell eptitopes are protected from degradation through alterations of the binding pockets and cleavage profile of the proteasome, increasing the supply of peptides suitable for transport by TAP and binding by MHC I, as well as significantly altering the repertoire of peptides available [54]. These repertoire differences are significant enough to lead to rejection of wildtype splenocytes by mice deficient in these three immunoproteasome subunits [55]. Restricted expression of the immunoproteasome subunits by non-immune cells until exposure to an inflammatory trigger may act as a tolerogenic mechanism and help restrict innapropriate immune responses. 21

37 Following transport by TAP the peptide repertoire may still be altered. Some peptides translocated into the ER are longer than that preferred by MHC I, as the preference of TAP is for peptides 8-16 amino acids in length. Importantly, they are still able to potentially bind to MHC I, as after transport into the ER lumen they may still be trimmed by peptidases such as endoplasmic reticulum aminopeptidase (ERAP1). This IFN-γ-induced protease trims amino acids at the N- terminus, leaving the C-terminal anchor residues preferred by most MHC I alleles and thereby increasing the pool of peptides available for binding [56]. The next member of the PLC that will be discussed is tapasin, which bridges MHC I to the TAP transporter. In addition to its bridging role, tapasin is also critical for PLC function. Loss of tapasin reduces the stability and surface expression of MHC I and decreases loading with high affinity peptides [1-3,57]. Tapasin both stabilizes TAP1 and TAP2, increasing their steady state levels in the ER and thus increasing peptide supply [58], and acts as a scaffold to assemble the PLC around MHC I (though a soluble tapasin mutant was also able to accomplish much of the function of wildtype despite no longer interacting with or stabilizing TAP [59,60]). Tapasin also helps prevent the export of poorly-assembled MHC I out of the cis-golgi, retaining it in the early secretory pathway for additional attempts at loading a high affinity peptide [57,61]. However, the key function of tapasin is that of a peptide-editor; exchanging low affinity peptides in the binding groove of MHC I for high affinity ones [62] that exhibit a long half-life in the binding groove [63]. This function requires association with the thiol oxidoreductase ERp57, and without it tapasin is not as effective at exchanging peptides. Even when an excess of low affinity peptides was supplied, tapasin-erp57 was still able to favour loading with high affinity peptides, indicating that it selects for high affinity peptides [64] as well as increasing the rate of peptide exchange [64,65]. 22

38 Tapasin forms an unusual disulphide bond with ERp57 [66] and has an extensive interface with the protein [67]. In addition, this interaction between ERp57 and tapasin is stable, unlike most of ERp57 s interactions with client substrates [68]. Mouse ERp57-deficient cells demonstrated the importance of ERp57, with reduced surface expression and impaired loading of peptides onto MHC I in these cells [69]. However, this appears to be a structural role, as mutagenesis of the CXXC active sites did not impair PLC function [48,70], and in fact the catalytic sites are inaccessible in the tapasin-erp57 crystal structure [67]. ERp57 appears to act instead to maintain proper stability and function of tapasin in editing peptides within the MHC I binding groove. As mentioned, PDIA1 has also been found associated with the PLC and may play a role in MHC I maturation at this stage. PDIA1 may affect the oxidation state of MHC I heavy chains, which may influence peptide loading of the binding groove (through PDI-catalyzed oxidation of the α2 disulphide proximal to the peptide binding groove [71]. Problematically, other groups have not been able to confirm a role for PDIA1 within MHC I maturation [46,72], and as such its association may not play a major role in the outcome. Calreticulin is the final protein forming the PLC. This molecular chaperone is related to calnexin, and also binds to substrates using a lectin domain that recognizes monoglucosylated N- linked glycosylations. Attached to this lectin domain is an extended arm that is known to interact and recruit other ER proteins. Calreticulin is recruited to the PLC through the tip of its arm domain by binding ERp57 and through its lectin domain with MHC I [73]. Lower surface expression and defects in peptide loading were observed for Crt-deficient mouse cells and there is unusually rapid export of MHC I through the Golgi [74]. This rapid export phenotype suggests that calreticulin helps prevent the premature escape of MHC I into the secretory pathway. 23

39 1.5.1 ER export and other trafficking of MHC class I PLC function is critical for MHC I maturation. In cells with disrupted peptide supply to the ER, MHC I accumulates at the ER or Golgi apparatus [1-3,52]. This suggests that peptidereceptive MHC I is retained within the cell by various quality control mechanisms. Both tapasin and Crt possess ER retrieval sequences at their C-termini, and removal of these sequences produces more rapid export of MHC I [57,74]. The PLC itself appears to play a role in this retention, as providing a supply of high affinity peptides to the ER resulted in release of MHC I from the PLC and clustering at ER exit sites [75]. However, this release due to peptide was not the rate limiting step, as addition of peptide did not accelerate the export rate of MHC I out of the ER [75]. Instead, it seems more likely that association with factors promoting MHC I export is more likely. Consistent with this, the addition of peptide increased the association of MHC I with the putative export receptor Bap31 [76]. Bap31 interacts with both tapasin and MHC I [77], suggesting it is present within the PLC, and the export rate of MHC is correlated with the amount of Bap31 present [77,78]. Rather than affecting peptide supply, Bap31 instead appears to influence the export rate of MHC I by regulating access to ER-exit sites. In fact, release from ER matrix proteins was recently shown to be the primary driver of the differential export rates observed for MHC I molecules of different allotypes [39,79]. This leads to a model where MHC I interactions with a broad network of interacting ER proteins determines whether MHC is incorporated into COP II vesicles (lipid vesicles surrounded by a protein scaffold that are involved in anterograde protein export, versus retrograde transport by COP I vesicles) following PLC release. Despite this regulation, export of MHC I from the ER may not be an irreversible step. MHC I loaded with poor quality peptides due to TAP deficiency can still be incorporated into COP II vesicles for export from the ER [80]. This study suggested instead that peptide-receptive 24

40 MHC I can cycle between the ER and cis-golgi until they are loaded with high affinity peptide. Both Crt and tapasin are potentially regulating this recycling. Mutant Crt lacking its ER-retrieval signal (a C-terminal KDEL sequence) was no longer detected in the ER-Golgi intermediate compartment or cis-golgi, and it was also not able to support loading of high affinity peptide onto MHC I even though it was still present in the PLC [81]. Tapasin is also present in the Golgi and associates there with peptide-receptive MHC I [82]. When tapasin s ER retrieval signal (a C- terminal KKXX) was mutated, the Golgi interactions between both tapasin and MHC I with COP I (Golgi-to-ER) vesicle proteins were lost [61]. Although this increased the amount of MHC I exported to the cell surface, these molecules were less stable Antigen cross-presentation The previous section discussing peptide loading of MHC I deals with the typical pathway taken by MHC I to the cell surface. It is relevant for the vast majority of cells in the body as well as most immune cells. However, there are alternative routes for loading of antigen onto MHC I, namely cross-presentation. Cross-presentation is important to allow dendritic cells to present exogenously-derived antigens on MHC I, in order to activate CD8+ cytotoxic T cells and allow them to carry out killing of target cells. Without a means to load exogenous antigens onto MHC I, which is normally loaded with peptides derived from the cytosol, the only way a DC could activate a CD8+ T cell would be to become infected or transformed itself. There are two proposed pathways for antigen cross-presentation. First, phagocytized antigens may be allowed to escape the endosomal compartment and enter the cytosol. Here they would be accessible to the proteasome and broken down into peptide fragments, allowing entry into the ER in the same manner as other endogenously-derived peptides. The second pathway involves trafficking of peptide-receptive MHC I molecules to the endosomal compartment, 25

41 where they could be loaded with peptides produced from the lysosomal breakdown of phagocytized proteins [83]. It has proven challenging to conclusively identify which of these two pathways is more likely to exist, or if in fact both serve some function by acting in parallel, in different DC subtypes, or in presentation of antigen from different sources. The primary challenge to the first, or cytosolic, pathway is to explain how protein antigens are able to cross the endosomal membrane into the cytosol. Without a means to do so, the exogenous antigens would not be accessible to the proteasome to allow for their degradation and peptide generation [84]. The most well-studied route for proteins to exit from a membrane compartment into the cytosol is the ERassociated degradation pathway, which transports misfolded ER proteins into the cytosol for degradation by the proteasome. Intriguingly, there is evidence for incorporation of ER membrane into phagocytic vesicles, potentially providing the necessary cellular machinery for retrotranslocation of phagocytized proteins into the cytosol [84]. TAP and other PLC components were also present in these vesicles, providing a means for localized retrotranslocation, degradation, and re-translocation of peptides back into endocytic pathway [84]. There are also examples of proteins able to enter the cytosol from outside of the cell, as shown for translocation of gelsolin protein toxin coupled to beads into the cytosol. Crosspresentation in this case was also inhibited by both TAP and proteasome inhibitors [83]. Finally, loss of some immunoproteasome subunits decreases the efficiency of cross-presentation of some protein antigens in vivo, which would also be consistent with this pathway [83]. In regards to the second, or vesicular, pathway, there is the important issue of how endosomally-degraded peptides can be produced with the same epitope profile as the proteasome. If not able to do so, there would be little point to cross-presentation as the peptides 26

42 presented by MHC I on cross-presenting DCs would not resemble those produced by other nucleated cells expressing MHC I, preventing recognition of intracellular pathogens in these cells [84]. Publications have shown TAP- and proteasome-independent cross-presentation, which argue against the cytosolic pathway for cross-presentation. Additionally, some endosomal compartments can be accessed by both PLC components and MHC I, putting the correct players in place for endosomal-derived peptide loading [83]. Although this may be a potential route of loading for some selected peptides, it still seems implausible that it could support and replicate the broad range of proteasomally-generated epitopes, and the presence of MHC I and PLC components in endosomal vesicles only provides circumstantial evidence at best. Overall the cytosolic pathway is more likely to be dominant based on existing evidence, although particular peptides may be produced and loaded in endocytic vesicles. One final route for the DC presentation of exogenous antigens is the intercellular transfer of MHC I from one cell to the DC. This has been proposed to occur through a number of mechanisms. Both trogocytosis, or the transfer of membrane patches from one cell to another, and exosome uptake, the incorporation of small extracellular vesicles released by cells into the DC membrane, have been examined, in addition to other mechanisms [85]. This type of transfer also appears to occur in vivo, although the degree it contributes to cross-presentation remains to be seen [85]. 1.6 ER Quality Control MHC I requires a number of specialized proteins to ensure its correct folding and loading with peptide. However, as with many other proteins assembled within the ER it folds in a cellular compartment densely packed with other newly-synthesized polypeptides that have not yet 27

43 reached their native structure. A host of cellular enzymes and factors act on these nascent proteins, sensing their folding state, inhibiting aggregation, preventing misfolding, forming disulphide bonds, and adding other modifications to the polypeptide. This entire process of ensuring that proteins reach their native conformation and functional state is referred to as ER quality control (Figure 1.5). One of the first modifications made to a nascent polypeptide as it enters the ER is the addition of an oligosaccharide structure onto asparagine residues in the polypeptide chain. The oligosaccharide itself can play a role in protein folding through its interactions with water and sterically hindering the folding protein. In addition, the glycans can function as signals for chaperone recruitment, ER export, and degradation. Several chaperone systems also act within the endoplasmic reticulum to help ensure proper folding of nascent polypeptides into their native state. These can involve binding to proteins to prevent aggregation of hydrophobic patches, isolating the protein from the high density of other proteins present within the ER, and unfolding portions of proteins to allow further attempts at exploring the folding landscape. Some of these processes are energy dependent, while others appear not to require ATP. In a broad sense, these chaperones can be divided into two categories: the lectin-dependent (typified by calnexin and calreticulin) and the polypeptide-dependent (represented by BiP and Grp94) systems. Disulphide bonds are also common within the domains of proteins synthesized within the ER. Formed by the oxidation of two cysteine residues, they occasionally require assistance to form. Incorrect disulphide bonds can also form, and require either isomerization or reduction to allow the nascent polypeptide to escape terminal misfolding. This assistance is provided by the protein disulphide isomerases, of which a number exist. 28

44 Figure 1.5. Pathways of protein quality control within the ER. Following translocation of a nascent polypeptide into the ER through the Sec 61 translocon, it is rapidly glycosylated on asparagine residues by the oligosaccharyltransferase complex. Trimming of the three terminal glucose residues down to one allows binding by lectin chaperones such as calnexin and calreticulin, and exposed hydrophobic patches allow for polypeptide-mediated chaperone interactions with BiP and Grp94. Cycles of binding and release by chaperones provide multiple opportunities to obtain the proper molecular fold. Thiol oxidoreductases and proline isomerases also assist with folding by forming disulphide bonds and isomerizing proline residues, resepectively. For substrates that persist in the ER, mannose trimming can lead to degradation of terminally misfolded substrate through ER-associated degradation. 29

45 The thiol oxidoreductase ERp57 has already been mentioned for its role in MHC I maturation, but thiol oxidoreductases act broadly on a number of different substrates. Finally, proline residues within a polypeptide may require modification. Due to their structure they can confer either a cis or trans conformation to the polypeptide chain. In some cases, this can act as a molecular switch to activate or suppress molecular signalling pathways. One of the most common examples of this is cyclophilin A, which is implicated in a number of immune signalling pathways. However, proline conformation also plays an important role in protein folding. While some prolines may spontaneously isomerize, in other cases they must be catalytically modified to allow correct folding to occur or to correct an incorrect earlier isomerization event. In the ER, these isomerization reactions are carried out by two types of proteins, the peptidyl-proline isomerases also known as cyclophilins, as well as the FK506 binding proteins (FKBPs). Each of these are named for their ability to bind to different immunosuppressant drugs, cyclosporine A for the cyclophilins and FK506 for the FKBPs. While these different groups of proteins each possess separate functions that are applied to the goal of assisting nascent polypeptides in reaching a stable native fold, they do not act independently of each other within the ER environment. Instead, ER chaperones form multiprotein complexes that form networks of interactions within the ER lumen [86]. Not only do these protein networks link proteins involved in folding, they also connect to those that degrade misfolded proteins [87], which are also present in complex protein networks. Taken together, it appears likely that rather than a collection of independent systems, the ER is a network of proteins that can act on substrates and potentially transfer target proteins to other areas of the network as needed. 30

46 1.6.1 Glycosylation and the lectin chaperones Cnx and Crt Before any of these quality control systems can act, a protein destined for the ER must first be translocated across the ER. There is one modification made during this import that is required for some molecular chaperones to associate with the nascent polypeptide, known as N- linked glycosylation. As previously mentioned, many proteins entering into the endoplasmic reticulum are rapidly modified through addition of pre-formed oligosaccharides. This process normally occurs co-translationally, but can also occur post-translationally for some proteins and sites. The oligosaccharide itself can increase protein solubility and aid proteins in reaching the correct native conformation. Modification of the oligosaccharide chains also creates signals that recruit ER lectin chaperones to the nascent polypeptide, retain it within the ER, promote its export, or target it for degradation [88,89]. The glycosylation event is catalyzed by the oligosaccharyltransferase (OST) complex. This enzymatic complex transfers the preformed Glc3Man9GlcNAc2 oligosaccharide (shown in Figure 1.6) from a lipid carrier onto the asparagine acceptor site. The catalytic subunit of the eukaryotic OST is known as STT3. The mammalian OST complexes come in two forms, containing either STT3A or STT3B. In addition, both forms of the OST complex associate with ribophorin I, ribophorin II, OST48, DAD1, and OST4. Depletion of subunits such as STT3A or STT3B can lead to hypoglycosylation of substrate proteins [90]. Ribophorin I appears to play a less critical role, although some studies suggest that depletion of ribophorin I impacts the glycosylation of specific substrates [91]. It may also play a role in recruiting proteins to the OST [90]. 31

47 Figure 1.6. Sugar structure of newly added N-linked glycans. The newly added N-linked glycan consists of a 14 sugar structure. N-acetylglucosamines form the initial base added onto an asparagine residue, followed by 9 mannose residues formed into three branches, A, B, and C. The final sugars consist of three glucose residues on the terminal end of the A branch [92]. 32

48 Following addition of the glycan structure shown in Figure 1.6, the first glucose residue is rapidly trimmed from the A branch by glucosidase I. This di-glucosylated oligosaccharide can then be recognized by the ER lectin malectin. Malectin is unique in recognizing the diglucosylated form of oligosaccharides [93]. In addition to this binding, malectin has been shown to associate with ribophorin I, and may play a role in retaining misfolded proteins within the ER [94]. Next, trimming of a second glucose residue by glucosidase II [95] produces the binding site for the ER lectins and molecular chaperones calnexin and calreticulin, which act to prevent protein aggregation and misfolding through sequential cycles of binding and release. Calnexin and calreticulin can be considered the prototypical example of lectin-based chaperones, recognizing N-linked oligosaccharides on folding proteins. Calnexin and calreticulin both consist of a globular domain and an extended arm domain attached [96]. Both bind to ERp57 and other proteins through a site at the tip of the arm domain [97], and both recognize particular glycan motifs through a binding site on the globular domain. Not surprisingly there is also significant substrate overlap between the two proteins, though some differences exist. This may be partially explained by the differing localization of these two chaperones, as Cnx is membrane anchored by a transmembrane domain, whereas Crt is soluble within the ER lumen, but there is evidence that differences in substrate specificity cannot be solely explained by their localization [74]. Regardless, Cnx and Crt in vivo are largely selective for glycoprotein substrates. As the glycoprotein selectivity is determined by the lectin binding site, it is important to understand its selectivity. The lectin binding site recognizes mono-glucosylated oligosaccharides with the structure Glc1Man5-9GlcNAc2 (Figure 1.6). 33

49 Figure 1.7. Selected structures of ER quality control family members. Representative crystal structures of the Hsp70, Hsp90, PDI, and cyclophilin protein families. (A) Crystal structure of the E. coli Hsp70 homolog DnaK (4B9Q) [98]. The SBD (both subdomains) and NBD are labeled. (B) Crystal structure of dimeric yeast Hsp90 (2CG9) [99]. The NTD, MD, and CTD domains are labeled. ATP analog is shown in orange. (C) Crystal structure of human PDI (4EL1) [100]. The a, b, b`, and a` domains are labeled. (D) Crystal structure of human cyclophilin A (1CWA) [101]. Cyclophilin A is shown in magenta, cyclosporine A in yellow. 34

50 The base of this glycan structure attaches to the glycoprotein at an asparagine residue in the motif Asn-X-Ser/Thr motif, where X can be any amino acid other than proline. It is a terminal glucose (in addition to three underlying mannose residues) that is key for recognition by Cnx/Crt, which is formed following cleavage of two other glucose residues by α-glucosidase I and II [102,103]. Binding of Cnx and Crt helps to prevent aggregation with nearby proteins and can protect the folding protein from terminal misfolding by stabilizing conformational intermediates in the bound state. The lectin binding site positions the bound protein in such a way that the arm domain wraps around it to provide steric protection from interactions with other unfolded proteins [104]. The arm domain provides additional support for protein folding by recruiting other proteins to the bound protein. As mentioned earlier, both Cnx and Crt interact with the thiol oxidoreductase ERp57. They also both interact with the proline isomerase cyclophilin B [86], Crt interacts with the thiol oxidoreductase PDIA5 [105] and Cnx interacts with additional thiol oxidoreductases [86]. Overall, by interacting with these enzymes Cnx and Crt can recruit their activities towards bound substrates. The exchange of one bound enzyme for another may also provide specialized activities that can be brought to bear on select subsets of substrate proteins. The terminal glucose on the N-linked glycan is only present transiently. Dissociation of the glycoprotein from Cnx or Crt can lead to de-glucosylation by glucosidase II. The glycan may then be glucosylated by UDP-glucose:glycoprotein glucosyltransferase (UGGT1), an enzyme that recognizes hydrophobic patches on non-native proteins [ ]. Re-glucosylation again allows recognition by Cnx and Crt and re-entry into the cycle of binding and dissociation. Released proteins that are properly folded will no longer have the exposed hydrophobic patches recognized by UGGT1 and will avoid further cycles of chaperone activity. Proteins that persist in 35

51 a non-native state may eventually undergo further mannose trimming, generating targeting signals for ERAD BiP Cnx and Crt are generally focused on glycosylated substrates. There are other chaperones that specifically recognize and bind to non-native polypeptide segments and that operate alongside the lectin chaperones. The ER-localized Hsp70 family member BiP (immunoglobulin heavy chain binding protein) is a central player in ER protein quality control. BiP is the major chaperone for non-glycosylated proteins, as well as some glycosylated ones. BiP s chaperone activity is complex and involves cycles of regulated ATP binding, hydrolysis, and nucleotide exchange. In an unstressed ER, BiP exists in pools of either a post-translationally modified form (ADP-ribosylated/phosphorylated) that is inactive, or an unmodified form capable of binding substrates. BiP contains two domains, a nucleotide binding domain (NBD) and a substrate binding domain (SBD) (Structure of related Hsp70 shown in Figure 1.7A). BiP acts through cycles of binding and release of substrate proteins. When the active form of BiP binds to ATP and potassium the two domains increase in proximity, opening a lid on the SBD and allowing substrate binding. Next, hydrolysis of the bound ATP to ADP triggers tighter binding of BiP to the substrate and closes the lid. Finally, exchange of ADP for ATP allows for substrate release, restarting the cycle [109]. Similar to binding and release by calnexin and calreticulin, these cycles of binding and release segregate hydrophobic patches of polypeptides from the rest of the folding environment to help prevent protein aggregation. The cycle of ATP hydrolysis and exchange is governed by a number of co-chaperones and nucleotide exchange factors, which play an important role in selecting substrates for BiP. 36

52 Substrate transfer to BiP is regulated by a group of proteins containing a particular fold known as a J domain. These J domain-containing proteins, known as DnaJs, Hsp40s, or ERdjs, interact with BiP through their J domains to increase the ATP hydrolysis activity of BiP. They also regulate which substrates are bound by it by activating BiP only in particular ER locations and under particular conditions, or by directly handing off substrates [109]. BiP s nucleotide exchange factors (NEFs) also play an important role in recycling BiP for additional binding events, and in mammalian cells the proteins Grp170 and BiP-associated protein (BAP) have been found to stimulate nucleotide release and preferentially bind to the ADP-bound state of BiP [109] Grp94 A second significant chaperone involved in polypeptide-based recognition of substrates within the ER is the Hsp90 chaperone Grp94. Grp94 forms a dimeric structure required for its normal chaperone functions. Like BiP it is an ATPase, but possesses a different structure than BiP. Grp94 expression is co-regulated with a number of chaperones as part of the unfolded protein response [110] (UPR, which will be described shortly), and perturbations of the protein folding machinery can lead to upregulation of Grp94 and other chaperones [110]. There are four domains, an N-terminal domain (NTD), acidic linker domain (LD), middle domain (MD), and C- terminal domain (CTD) (Structure of related family member shown in Figure 1.7B). The nucleotide binding domain is present in the NTD, with the middle domain and linker also required for hydrolysis of ATP. The C-terminal domain controls dimerization [110]. Unlike cytosolic Hsp90 there is less known about co-chaperones that may function with Grp94 [110]. Some client protein-specific co-factors have been identified, but it is still not clear if the paucity of data is due to Grp94 not needing co-chaperones to function or if they are yet to be characterized [111]. 37

53 Grp94 appears to fulfill four major functions within the ER. Similar to BiP, Grp94 primarily functions as a molecular chaperone, but in contrast to the broad substrate range of BiP it possesses a much smaller substrate list [110]. In addition, whereas BiP acts early on during protein folding, Grp94 seems to act much later, binding to proteins with oxidized disulphides after BiP has dissociated [112]. Grp94 is also an important calcium binder, helping to store Ca 2+ ions in the ER through binding and to regulate levels of free calcium. Third, Grp94 appears to play a role in selecting misfolding proteins for ERAD [110]. Finally, Grp94 interacts with many of the other ER quality control proteins, suggesting a functional role as part of a folding network [86] Protein disulphide isomerases Although calnexin, calreticulin, BiP, and Grp94 all assist folding proteins in obtaining the correct three-dimensional fold, they are not always sufficient on their own. As previously mentioned, disulphide bonds play an important role in supporting and stabilizing this structure, as well as restricting the folding landscape that needs be explored. Protein disulphide isomerases act to create, rearrange, or undo these bonds. The catalytic domain of PDIs is the thioredoxin domain, which contains at its core the CXXC motif. There are over 20 PDI family members within the ER that vary in domain arrangement, number of catalytic domains, and substrate specificity. To introduce a new disulphide bond into a protein, the two cysteines within this motif are oxidized to form an intramolecular disulphide bond. A nucleophilic attack from a reduced cysteine in a substrate protein leads to formation of a thiol disulphide intermediate between the isomerase and the substrate protein, followed by a second exchange to release the PDI and leave the disulphide bond on the folding protein. This is made possible through deprotonation of the reduced cysteine, 38

54 forming a thiolate anion that attacks the oxidized cysteine [92]. Essentially, PDIs form disulphide bonds through a series of electron transfer events in an elegant example of cellular redox chemistry. In isomerizing or breaking a bond, the PDI must instead be in a reduced conformation, but a similar series of enzymatic steps take place. How PDI members select their substrates is an important part of their function. The most common thiol oxidoreductase referred to is PDIA1. PDIA1 has been shown to act on a wide range of substrates. It consists of a four domain structure, with both N- and C-terminal catalytic thioredoxin domains (referred to as a and a`) which flank two non-catalytic thioredoxin-like domains (known as b and b`). These form a U structure, with the catalytic domains orienting towards a hydrophobic surface on the b` domain that binds to a wide range of hydrophobic and misfolded polypeptide sequences (Structure shown in Figure 1.7C). Importantly, the b domain appears to modulate this binding, as does an X-linker region between the b` and a` domains that can transiently block the binding site [92]. Due to its apparent importance in ER quality control, PDIA1 was considered essential for some time. However, depletion experiments showed that while PDIA1 was important, its loss only delayed the folding of many ER substrates. This was hypothesized to be due to compensation by other protein disulphide isomerase family members, and was supported by combined depletion experiments of PDIA1 with other family members [113]. As mentioned, ERp57 is another ER thiol oxidoreductase playing a role in the folding of ER proteins, especially MHC I. ERp57 has a similar domain structure to PDIA1, but acts on a different subset of substrates due to its interaction with calnexin and calreticulin. By associating with lectin chaperones, it shows a preference for glycosylated nascent proteins. This was 39

55 confirmed through depletion experiments similar to those conducted for PDIA1, and identified a number of glycosylated proteins whose maturation was impacted by loss of ERp57 [113]. In addition to aiding in protein folding, some PDI family members appear to act as reductases, reducing disulphide bonds to aid in the degradation of misfolded proteins. Both ERdj4 (DNAJB9) and ERdj5 (DNAJC10) were shown to be important for degradation of surfactant protein C [114]. Importantly, BiP was important in triggering this degradation as well. When both of these J domain containing proteins were mutated to be unable to stimulate the ATPase activity of BiP, the degradation of the surfactant protein was inhibited [114]. This example emphasizes the role protein disulphide isomerases play in both protein folding and degradation, and also highlights how the different J domain proteins are able to adapt BiP towards distinct purposes Proline isomerases Just as PDI family members act on a nascent polypeptide to ensure its cysteines are arranged in the correct manner, so do cyclophilins and FKBPs act on a protein s proline residues. There are 16 cyclophilins identified in humans, with the cytosolic cyclophilin A acting as the classic example [115]. Cyclophilin A (CypA) is most known for the downstream effects of its binding by cyclosporine A (CsA) (Structure of CypA bound to CsA shown in Figure 1.7D). Once bound by CsA, CypA interacts with calcineurin, binding between the catalytic and regulatory domains to disrupt its phosphatase activity and block its function. This is not the only effect of CsA binding though, as it also disrupts the catalytic activity of cyclophilin A, and binding of CsA to other cyclophilins is able to block their activity as well. There are two cyclophilins present within the ER: cyclophilin B (CypB) and cyclophilin C (CypC). These cyclophilins have been linked to several roles within the ER. Several examples 40

56 of CypB s role in protein folding have been reported. Secretion of IgG antibody is dependent on isomerization of a conserved proline residue by CypB [116]. CypB is also required for proper folding of the Na+-dicarboxylate co-transporter, though in this case it is a chaperone activity of the cyclophilin that is needed, rather than any isomerization activity [117]. This suggests that cyclophilins may both act catalytically, as well as to bind misfolded proteins and prevent their aggregation or promote a correct fold. In addition, CypB and CypC are both implicated in maintaining the oxidation status of the ER, with combined depletion of cyclophilin B and C leading to ER hyperoxidation [118]. Finally, cyclophilin B has also been implicated in ERassociated degradation. It was shown to catalytically enhance the degradation of soluble ERAD substrates containing cis-prolines [119]. FKBPs share a similar catalytic activity to the cyclophilins, but carry out this function using structurally different domains. Instead of being inhibited by CsA, they can be inhibited by another small molecule known as FK506. The prototypical member of the FKBPs is cytosolic FKBP12. When bound by FK506, the complex interacts with calcineurin and inhibits its phosphatase activity [120,121]. FKBP12 is also the target of rapamycin, another immunosuppressant, though this drug-protein complex instead inhibits the signalling kinase mtor instead of calcineurin [120,121]. Importantly, both of these drugs act to inhibit the catalytic activity of FKBP12 upon binding [120,121]. Within humans there are six FKBP proteins, FKBP13, FKBP19, FKBP22, FKBP23, FKBP60, and FKBP65 (with the discordant gene names FKBP2, FKBP11, FKBP14, FKBP7, FKBP9, and FKBP10, respectively) that possess an ER localization signal, and five that also possess ER retention signals [121]. In general, these have been suggested to have functions in protein folding and trafficking of proteins out of the ER. This can also include interactions with 41

57 other ER quality control components. Interaction of BiP with FKBP23 can lead to modulation of BiP s ATPase activity in a mechanism that involves FKBP23 s catalytic activity [122]. FKBPs have also been linked to ERAD. FKBP65 (FKBP10) depletion was shown to prevent degradation of mutant glucocerebrosidase, whereas overexpression of FKBP65 accelerated its degradation through ER-associated degradation [123]. In addition to these functions, various ER-localized FKBPs have been shown to interact with extracellular cytoskeletal proteins and influence hepatic tumor development, membrane trafficking of proteins, and collagen folding [121]. However, as these somewhat scattered targets suggest, no firm overarching role has been adequately proposed for the ER FKBPs as yet. As with the ER-localized cyclophilins, we are still searching for clear, general functions to ascribe to them Unfolded protein response Many of the molecular chaperones, thiol oxidoreductases, and proline isomerases introduced here are expressed constitutively within the cell, but this does not mean that they are always expressed at the same levels, as the protein folding load on the ER can vary dramatically between cell types, cellular environments, current functions. To respond the ER contains a signalling system that recognizes the presence of unfolded proteins and upregulates molecular chaperones and quality control components, slows translation of proteins to give the ER time to catch up, and accelerates removal of misfolded proteins from the ER. This is known as the unfolded response (UPR). Three main signalling pathways feed into the UPR, which can be represented by the three transmembrane proteins that serve as sensors for unfolded protein within the ER: activating transcription factor 6 (ATF6), protein kinase RNA-like ER kinase (PERK), and inositol-requiring protein 1α (IRE1α) [124,125]. Under normal conditions each of these proteins is associated with the molecular chaperone BiP. With the accumulation of misfolded proteins within the ER, BiP shifts its association to them and is released from the 42

58 sensors, activating their downstream signalling pathways. In addition to release of BiP, direct binding to unfolded proteins may also play a role in UPR activation [126]. Activation of theses three UPR signalling pathways drives a number of downstream events with differing kinetics [124,125]. For instance, PERK dimerization first signals for an inhibition of protein translation driven by phosphorylation of eif2α, and splicing and subsequent translation of Xbp1 mrna, driven by IRE1α dimerization, triggers the degradation of ER-bound mrnas [125]. Autophagy and pre-emptive quality control also play a role and together these mechanisms slow the influx of proteins into the ER and remove protein aggregates/damaged ER microdomains, giving time for the existing protein load to be cleared. Following this initial pause in synthesis there follows a coordinated program of gene expression, driven by all three pathways, to upregulate chaperones and other proteins involved in promoting protein folding, redox balance, amino acid metabolism, lipid synthesis, protein secretion, and ER-associated degradation [125]. Finally, if the stress persists and cannot be resolved it can lead to an apoptotic response induced by upregulated expression of pro-apoptotic proteins such as CHOP, BIM, PUMA, and NOXA [124,125]. Overall, this system helps cells respond to varying degrees of protein misfolding and is essential to ensure the normal function of cell types with heavy protein folding burdens ER-associated degradation While there are numerous systems to assist nascent polypeptides and misfolded proteins in reaching their correct native fold, these pathways are not always successful. Induction of the UPR may be able to compensate for an increased burden of misfolded protein, but at some point if a cell wishes to avoid apoptosis it may simply remove the terminally misfolded proteins and start fresh. The common pathway for dealing with these incorrectly folded proteins is known as 43

59 ER-associated degradation (Figure 1.8A). ER-associated degradation (ERAD) is a normal cellular process for the removal and degradation of proteins from the endoplasmic reticulum [ ], and consists of recognition and targeting of misfolded proteins towards the ERAD pathway, retrotranslocation of the protein across the ER membrane to the cytosol, and ubiquitination and proteasomal degradation of the protein. Proteins can be degraded for a number of reasons. The most primary cause is terminal misfolding, but can also include regulation of cellular signaling pathways, as is the case for ERLIN1 and ERLIN2 degradation of inositol-3-phosphate receptors [130], or maintenance of subunit stoichiometry, as with TCR subunits [131]. When terminally misfolded, MHC I is not exempted from a fate of degradation. When MHC (both free heavy chain and to a lesser extent HC-β2m) is deprived of β2m or a supply of peptide it is retained within the ER and eventually degraded [51,52]. In addition, viruses are known to exploit the ERAD pathway to evade the immune system. Human cytomegalovirus (HCMV) targets MHC I heavy chains for rapid degradation through the immunoevasion molecules US2 [132] and US11 [133], mouse γ- herpesvirus 68 also targets MHC I using the protein mk3 [134,135], and HIV targets CD4 through Vpu [136,137]. Misfolded proteins are thought to be targeted to ERAD through changes in protein glycosylation, exposure of misfolded protein lesions, or a combination of both signals. How the cell recognizes terminal misfolding of a protein is an intriguing question, as one can imagine that there is little to distinguish a terminally misfolded protein from a nascent polypeptide still exploring the folding landscape to reach a native conformation. 44

60 Figure 1.8. ER-associated degradation. (A) In a broad sense, ERAD can be thought of as proceeding in four stages. First, misfolded proteins are recognized by the ER quality control machinery as terminally misfolded or ready for degradation. This can also include substrates targeted for degradation, as is the case for MHC I degradation by the human cytomegalovirus evasion molecules US2 and US11. Recognition is followed by targeting of the polypeptide towards the ERAD machinery. This leads to association with ERAD complexes, typified by a core E3 ligase. While the mechanism of the process is still not entirely understood, polypeptides are then retrotranslocated out of or across the ER membrane, with ubiquitination playing a role, though the exact timing is unclear for some ERAD substrates. The final step is proteasomal degradation of the ERAD substrate in the cytosol. (B) The core yeast ERAD complex for degradation of ER lumenal substrates. The core E3 ligase Hrd1 and known associated players are shown. Human homologs of the yeast proteins are shown in brackets [129]. 45

61 One of the better understood routes to ERAD is trimming of N-linked glycans on the lumenal glycoproteins in S. cerevisiae. Following addition of the Glc3Man9GlcNAc2 oligosaccharide (Figure 1.6) to the protein, glucosidases I and II and ER-mannosidase I rapidly trim down the oligosaccharide to a Glc1Man5-9GlcNAc2 structure. While this would be recognized by Cnx and Crt in mammalian cells, it is not in yeast as there does not appear to be a functioning Cnx/Crt cycle. For terminally-misfolded glycoproteins, mannosidases act to further trim the glycan, allowing the glycoprotein to be recognized by a new group of ERAD-promoting lectins. Htm1 was also identified as a mannosidase, in addition to ER-mannosidase I, [138] that acts on glycosylated substrates [139] to generate a glycan moiety that is recognized by the lectin Yos9 [ ]. The specific signal recognized in this case is exposure of a terminal α-1-6 mannose on the C-branch of the oligosaccharide [144,145], which is labeled in Figure 1.6. The slow rate of trimming by Htm1 has been described as a timer, providing nascent polypeptides with time to fold before they are targeted for degradation. The situation in mammalian cells is less clear than in yeast. Htm1 has three homologs in human cells, EDEM1, 2, and 3. EDEM1 has been most studied, and plays a role in extracting substrates from the calnexin/calreticulin cycle. EDEM2 and EDEM3 are related proteins that both localize to the ER [146]. A recent question has been whether the EDEM proteins possess mannosidase activity or whether they are simply lectins. Initial studies suggested that EDEM1 acted only as a lectin [147], although it did seem to be an ERAD-promoting factor [147,148]. However, overexpression of EDEM1 accelerates mannose-trimming, and mutation of predicted catalytic residues in EDEM1 blocks this effect, arguing for a catalytic role of EDEM1 in mammalian cells [149,150]. While this could still be due to EDEM1 stimulating mannosidase activity in another as yet unidentified protein, the simpler explanation is that EDEM1 itself possesses it. Overexpression of wildtype or mutant EDEM3 also suggested the presence of 46

62 mannosidase activity [146]. Finally, EDEM2 (along with EDEM1 and EDEM3) was suggested to have mannosidase activity in a gene knockout assay [151], with wildtype activity being recoverable upon expression of wildtype EDEM2 but not EDEM2 with the E11Q mutation (mutation of this catalytic residue in ER-mannosidase I disrupts catalytic activity [152]). This study also suggested that EDEM2 targeted the first terminal mannose of the C branch, with EDEM1 and EDEM3 then acting on the second mannose. In yeast, Yos-9 recognizes the glycan signal produced by Htm1 and targets misfolded substrates for ERAD [144,145]. Although the lectin Yos-9 is important for recognition of ERAD substrates in yeast, its binding does not appear sufficient for degradation on its own. The adapter protein Hrd3p (yeast homolog of the human SEL1L) was shown to bind directly to an ERAD substrate independently of Yos9 [153]. In this study, Hrd3p and Yos9 were suggested to act together to provide specificity for recognition of terminally-misfolded substrates. Additionally, Der1p has been observed interacting with the misfolded CPY* substrate independent of Hrd3p and Yos9, suggesting another pathway of recognition [153,154]. An important caveat is that although a direct interaction was found between Der1p and the substrate, it is not known if and how it was recruited to Der1p, and Hrd3p and Yos9 both seem to be required for normal ERAD along with Der1p. Regardless, the end result of binding of misfolded protein by these ERAD sensors is the recruitment of the substrate to membrane-bound ERAD machinery. The role of the human homologs of Yos-9, OS-9 and XTP3-B, may be more complicated. These two lectins bind to both substrate and SEL1L, and appear to recruit misfolded proteins to the ERAD machinery. How they recognize substrates has been controversial, as some studies have reported that OS-9 and XTP3-B require their lectin domain to interact with SEL1L, not misfolded proteins, suggesting they are binding polypeptide-based motifs on the misfolded 47

63 protein instead of lectin signals [155]. They also showed that OS-9 and XTP3-B cooperated with Grp94 to degrade misfolded substrates [155], which supports reports that OS-9 and Grp94 are required for degradation of some non-glycosylated substrates [156]. However, other studies contradicted lectin-driven recruitment of OS-9/XTP3-B to SEL1L, instead showing that the lectin activity of XTP3-B was not required for the SEL1L interaction [157], and that OS-9 used its lectin domain to interact with a misfolded substrate [158]. Further complicating the situation is the report that XTP3-B may prevent degradation rather than accelerate it, as overexpression of XTP3-B blocked ERAD of the NHK mutant of alpha-1 antitrypsin [157]. That both XTP3-B and OS-9 have been independently shown to use their lectin domains for substrate recognition argues in favour of this model, though it does not exclude the possibility that they may bind to glycans on SEL1L under some circumstances. Setting aside details of the exact mechanism for now, XTP3-B and OS9 appear interact with and recruit misfolded proteins to Hrd3/SEL1L, which functions as an adapter to localize misfolded substrates to the ERAD machinery. Now that the routes onto the ERAD pathway have been described it is appropriate to mention the protein complexes that form the core of the membrane-localized ERAD machinery. Depending on the nature of the protein lesion, the substrate may be degraded by a distinct complex of proteins (at least in yeast, in mammalian cells the situation is less clear). Each distinct complex possesses an E3 ligase that is responsible for the ubiquitination of the ERAD substrates. Ubiquitination is the covalent attachment of the 76 amino acid long polypeptide ubiquitin onto lysine, cysteine, serine, or threonine residues of the target protein, and will be described in more detail shortly. These complexes also assist with retrotranslocation of ER lumenal proteins into the cytoplasm. However, ubiquitination does not necessarily preceed retrotranslocation, even though the complex thought to provide the driving power to extract the misfolded protein from the ER membrane, the p97 AAA-ATPase complex, binds to factors that 48

64 recognize ubiquitin moieties. Instead, it seems that the p97 complex may be involved in the final separation of the misfolded protein from the ER membrane, allowing the misfolded polypeptide to reach the proteasome for degradation. When considering how a protein is targeted to ERAD, it is important to consider both the topology of the protein and the location of the misfolded lesion. In yeast, there are two E3 ligases that typify two separate pathways for degradation. Lesions in the cytosolic domains of ER membrane proteins are targeted towards the Doa10 complex, which also commonly contains Ubx2, Ubc7, Ubc6, Cue1, and the p97 complex subunits cdc48, Ufd1, and Npl4. ER lumenal or transmembrane lesions are recognized by the Hrd1 complex (Figure 1.8B), which consists of Hrd3, Der1, Yos9, Kar2, Usa1, Ubx2, Ubc7, and the p97 subunits [129]. Homologous complexes to those found in yeast also exist in mammalian cells. The yeast Hrd1 E3 ligase exists and associates with SEL1L (the human homolog to yeast Hrd3). Doa10 is also found in mammalian cells, and is known as Teb4 or MARCH-VI in Homo sapiens. As well as these equivalent complexes, there are additional E3 ligases and ERAD complexes, many of which have poorly understood substrate specificities. There also seems to be more overlap between ERAD complexes for some substrates, though others have very specific degradation pathways. AMFR (also known as RNF45 or gp78), RNF5 (or RMA1), RFP2 (also known as Leu5 or Trim13), RNF170, Nixin/ZNRF4 [159], RNF139 (or TCR8) [160], TMEM129 [161,162], and RNF185 [163] all have been linked to various ERAD substrates. Kf-1 (or RNF103) has also been shown to have an ER localization and to interact with the ERAD components Derlin-1 and VCP, though it does not yet have an identified substrate [164]. However, with this expansion of E3 ligases comes a lack of clearly defined substrate pathways, and we do not yet understand the details of substrate sorting between them. 49

65 As mentioned, the E3 ligases involved in ERAD recognize and act on large yet distinct groups of protein substrates. However, they all carry out a similar enzymatic conjugation of ubiquitin onto proteins targeted for degradation. Ubiquitination occurs through covalent attachment of the C-terminus of ubiquitin to what is typically a lysine residue on the target protein, though some models have shown that serine, threonine, and cysteine residues can be targeted under particular conditions [165,166]. Prior to this ligation, the ubiquitin subunit must first be acted on by an activating enzyme (E1). This involves the ATP-dependent attachment of ubiquitin to a cysteine on the E1 through its C-terminus. Next, the ubiquitin is transfered to a cysteine on a conjugating enzyme (E2). Finally the ubiquitin moiety is transferred to the substrate protein by a ligating enzyme (E3) [166]. The ERAD E3s are thought to be more promiscuous in selecting which protein residues are ubiquitinated, owing to the partially folded status of the substrates they typically encounter [167]. As polyubiquitin chains are preferentially recognized by the p97 AAA-ATPase complex and cytosolic degradatory machinery, additional ubiquitin subunits must be added onto the initially conjugated ubiquitin molecule to form a chain structure. This can occur through conjugation to either the N-terminus of ubiquitin or to one of seven lysine residues. The ERAD E3s polymerize ubiquitin by linking the C-terminus of free ubiquitin onto lysine 48 of ubiquitin within the growing chain [167]. Chains of 4-6 ubiquitin resuides linked in such a manner are known to provide an efficient signal for proteasomal degradation [166], although other chain structures containing branches, alternate linkages (for example, lysine 11 linkages) or mixtures of linkages may also be added to ERAD substrates [166,167]. There is also still debate as to how a misfolded protein crosses the ER membrane. A number of potential pore complexes have been proposed, but no conclusive evidence yet exists as to what proteins form this pore. Sec61 was one of the originally proposed pore proteins, based 50

66 on work studying MHC I degradation by the HCMV immunoevasion molecule US2 [132]. Other evidence has suggested that one or more of the Derlins may be involved, or that an E3 ligase may form the pore, with most studies focusing on Hrd1p in yeast [159]. Finally, for some substrates there may not even be a pore, with misfolded proteins being directly extracted from or through the membrane, although such a pore-less retrotranslocation process seems more probable for proteins containing a transmembrane domain rather than soluble ER lumenal proteins. While much still needs to be addressed, a recent in vitro reconstitution study has provided insight into the pore s identity in yeast. By reconstructing the minimal ERAD components needed for retrotranslocation in a purified protein system, The Hrd1p E3 ligase was identified as the primary protein required for retrotranslocation of a misfolded substrate [168]. As the only membrane protein present in the reconstituted proteoliposomes, Hrd1p is clearly able to mediate transfer of a substrate across the membrane. Future studies will hopefully inform on whether human Hrd1 and other E3 ligases involved in mammalian ERAD have a similar ability to support retrotranslocation. Based in a large part on yeast studies many of the core players in ERAD have been identified, but the mammalian system is far more complex than that in S. cerevisiae. A larger number of E3 ligases and a large network of interactions between ERAD players add to the complexity of the system. In addition, new ERAD components are still being identified, and it seems probable that additional players in ERAD remain to be identified. There is a need to identify novel factors to better understand the transition between protein folding and protein degradation, and to characterize the specialized ERAD components targeting particular subsets of misfolded substrates. 51

67 1.7 Human cytomegalovirus Viruses can provide an excellent means to better understand cellular processes and mechanisms. This is particularly true for immune processes, as the evolutionary process has refined many viral immunoevasion molecules to be able to disrupt immune processes with scalpel-like precision and efficiency. This may be a direct targeting of an immune function, or an exploitation of the cellular or immune machinery to do the virus s dirty work for it. Now that MHC I and its important role have been introduced, as well as the cellular quality control processes that are important to its assembly, it is appropriate to discuss how one particular virus, human cytomegalovirus, manipulates ER quality control to disrupt surface presentation of peptides by MHC I Epidemiology and pathology Human cytomegalovirus (HCMV), or human herpesvirus 5 (HHV-5) is endemic around the world, with infection rates ranging from %. In those that are healthy infection usually results in no overt symptoms; however, in vulnerable populations clinically relevant pathologies can develop. These vulnerable groups include HIV-infected individuals, pregnant women, transplant patients, immune deficient patients, or those who are otherwise immunosuppressed [169]. Prior to the introduction of anti-retroviral therapy, cytomegalovirus infections in HIV+ individuals were a common complication. CMV end-organ disease was typically rare until CD4+ T cell counts dropped below 50 cells/mm 3, or when other opportunistic infections were present. CMV retinitis was a common result, typically causing a painless and progressive loss of vision. Gastrointestinal infections and other pathologies are also potential consequences of infection in 52

68 this case [170]. HCMV is also a concern for pregnant women, as the fetus may be infected across the placenta in women who seroconvert during pregnancy or become infected due to a latent maternal infection reactivating. These types of congenital infections affect approximately 0.5-3% of all newborns and results in a risk of 20% for serious disease, such as deafness, learning disabilities, and mental retardation [171]. HCMV infection is characterized by enlarged cells with intranuclear and paranuclear inclusions surrounded by a halo, often referred to as owls eyes when observed under a light microscope with an acid stain. CMV can infect many tissues, but has affinities for the CNS, eyes and liver, and particularly salivary glands. As with other herpesviruses, CMV can lie dormant in host tissue indefinitely and is reactivated periodically. Latency seems to occur following viral entry into a cell, with suppression of lytic gene expression playing some part in the process. [172]. HCMV displays a wide tropism. Two proteins, glycoprotein B (gb) and the gh-gl dimer, are critical for cell entry. The virus then begins a complex, carefully regulated program of gene expression, with approximately 200 reading frames being expressed in a particular temporal pattern. This includes a number of immune evasion molecules expressed through the viral life cycle. Despite the presence of these immunoevasins, the virus is typically controlled but not eliminated in healthy hosts, leading to a stable but active stand-off between the immune system and the virus. In a normal lytic life cycle, the virus first attaches to and penetrates the cell, with uncoating occurring in the cytoplasm. The nucleocapsid then enters into nucleus and releases the viral DNA to initiate expression of the immediate early-1 and -2 genes. This leads to viral DNA replication, encapsulation, and nuclear egress. Final envelopment occurs in the cytoplasm by 53

69 budding into a specialized vesicle structure [173] at the ERGIC (ER-Golgi intermediate compartment) and virus release through exocytosis [174] Viral structure and genome HCMV is unusual among viruses for many reasons, one of which is its large genome. The HCMV virion contains at its core an approximately 250-kb dsdna genome with a complex gene structure, located within a proteinaceous capsid [175]. The capsid is surrounded by a tegument layer consisting of viral phosphoproteins. Lastly, an outer membrane envelopes the virus and contains a high density of glycoprotein complexes. Of particular interest are two gene regions close to the terminal ends of the genome which are unique to CMV among other betaherpesviruses, and are designated UL and US. These are flanked on either side by direct repeats (TRL and IRL denoting the terminal and internal repeats around the UL region, and TRS (terminal repeats) and IRS (internal repeats) denoting repeats around the US gene region) [176]. The UL (unique long) and US (unique short) gene regions are particularly noteworthy for the large number of immunoevasin genes located within them Subversion of major histocompatibility complex class I peptide presentation The UL and US gene regions contain many genes coding for products which target both the adaptive and innate responses. The sheer number that specifically target MHC I is noteworthy, suggesting that evading or disrupting the presentation of viral peptides on MHC I has been strongly selected for during HCMV evolution. Specifically, the HCMV proteins US2, US3, US6, US10, and US11 all inhibit peptide presentation by MHC class I and target it early on during its maturation. These immunoevasins are important for escaping CD8+ T cell responses in 54

70 vivo and are critical for superinfection, or the establishment of secondary HCMV strains into an already infected host [177]. US6 acts on the ER lumenal face of the TAP transporter to inhibit translocation of peptides into the ER [178], although unlike some other TAP-inhibiting immune evasion molecules it does not block peptide binding to TAP [179]. Instead, it induces a conformational change that inhibits the binding of ATP to the cytosolic nucleotide binding domain [180]. Studies of US3 first reported a role in retaining MHC I within the ER but not disrupting peptide loading [181]. This was suggested to be due to a transient interaction that blocked export (potentially through oligomerization of US3 [182]) rather than a retrieval of MHC I from the secretory pathway, since US3 itself was unstable and poorly retained in the ER [183]. The poor retention of US3 led Hegde et al., to investigate an additional role for US3, where they found evidence that US3 bound to MHC II heterodimers and disrupted their interaction with the invariant chain which escorts the heterodimers to appropriate compartments for peptide loading [184]. The activity of US3 cannot be considered entirely independently of the other HCMV immunoevasion mechanisms. US3 has been suggested to act in concert with both US2 [185] and US11 [186]. US3 was found to retain MHC I and increase its binding to US2 for subsequent degradation, resulting in further decreases in surface MHC I [185]. US3 and US11 co-expression also resulted in decreased levels of surface MHC I, but with increased ER retention and decreased degradation [186]. It is not yet clear what the significance of this US3-mediated regulation of US11 and US2 is, but it does support a more complex interplay between immunoevasins then has been previously appreciated. Like US3, US10 was also originally thought to delay trafficking of MHC I, though contradictorily it did not appear to influence the function of US2 or US11 [187] despite 55

71 presumably increasing MHC I availability within the ER by delaying its export. This role was later discarded by a paper that showed US10 degrading HLA-G molecules rather than the classical HLA-A, B, and C [188]. Not listed earlier was US8, another immunoevasion molecule that has no detectable impact on MHC I in HCMV-infected cells, but does bind to MHC I [189]. Discarding this immunoevasin molecule as non-functional is likely premature, as it may have cooperative functions with US2, US10, US11, or other evasins that are yet to be discovered. Finally, US2 [132,190] and US11 [133,190] interact with newly synthesized MHC I HCs, targeting them to the ERAD machinery for degradation within minutes of synthesis. This is also not an exhaustive summary, as there are even more immunoevasion molecules targeting MHC I that are located outside of the US gene region, such as UL82, UL83, and the microrna US4-1[191] Evasion of NK cells Human cytomegalovirus does not restrict itself to manipulating and/or suppressing T cell responses. NK cells also play an important role in HCMV infection, as patients who possess a compromised NK response exhibit increased susceptibility to HCMV infection. Although disrupting MHC class I is important to the survival and replication of many viruses, it can expose infected cells to recognition by natural killer cells. These NK cells may be activated, at least in part, by low surface levels of MHC I. In response, human cytomegalovirus has evolved a number of proteins and micro RNAs that either mimic MHC class I or compensate for the decreased MHC class I (and loss of NK cell inhibitory signals) by blocking activating signals and promoting inhibitory ones [192]. 56

72 One example of NK evasion is provided by the MHC I mimic UL18. This evasion molecule adopts a similar fold to MHC I, associates with β2m, and presents peptide in a binding groove. However, the presence of approximately 13 N-linked glycosylation sites blocks access to the binding groove by the TCR receptor while still allowing the LIR1/ILT2 inhibitory receptor to bind [193] (with approximately a 1000-fold higher affinity than that observed for LIR1/ILT2 binding to MHC I [194]). UL18 is also resistant to the action of other HCMV immunoevasins that normally target MHC I, despite requiring peptide for its maturation. In fact, UL18 inactivates US6, allowing the TAP transporter to function when UL18 is present in the PLC, and competes with endogenous MHC I for access to the TAP transporter [195]. The story behind UL18 is complicated further by the observation that while LIR1+ NK cells were inhibited by UL18, LIR1- NK cells were stimulated by UL18 expression, suggesting a LIR1-independent means of UL18 recognition [196]. While UL18 completely replaces MHC I on the cell surface, another HCMV protein, UL40, restores HLA-E surface expression by providing peptide in a TAP-independent manner. In healthy cells HLA-E presents peptides derived from the signal sequences of HLA-A, -B, -C, and -G molecules. HLA-E can then be recognized by the inhibitory receptor CD94/NKG2A. The inhibition of TAP by US6 would normally disrupt this peptide source, but the UL40-derived peptide can substitute for endogenous HLA-E peptides in a TAP-independent mechanism [197]. Rather than providing a signal for an inhibitory receptor, the immunoevasin UL16 antagonizes the activating NKG2D receptor by removing selected NKG2D ligands from the cell surface. NKG2D recognizes a number of ligands (MICA, MICB, ULBP1-4, and RAET1G) which can lead to a NK cell activation-promoting signal. UL16 binds to ULBP1 and ULBP2 (the 57

73 ULBP family was named for their affinity for UL16) and sequesters these NKG2DLs within the ER, leading to suppression of NK cell activation [198]. Lastly, as with the description of molecules targeting classical MHC I, this is not an exhaustive list, and other molecules are known that manipulate NK cell function, both through direct action on inhibitory or activating receptors or their ligands, or through other immune modulatory functions that have been attributed to proteins of both the US and UL gene regions [199] US2 and US11 As previously mentioned, US2 [132,190] and US11 [133,190] are both type I transmembrane proteins that localize to the ER and degrade MHC I. They bind directly to MHC I through their ER lumenal domains (Figure 1.9A) and trigger rapid degradation of HC early after its translocation into the ER. There is evidence that different portions of MHC I and US2/US11 are required for binding and for degradation. The cytosolic tail of MHC I heavy chains does not appear to be involved in binding to US2/US11, but is required for MHC I degradation [200]. In addition, alteration of the cytosolic tails of MHC I molecules resistant to degradation can confer sensitivity to US2/US11 [201]. The observations that US2/US11 binding does not confer degradation of MHC I on its own suggests that US2 and US11 are recruiting additional factors to MHC I to trigger its destruction. 58

74 Figure 1.9. Structure of US2 and US11. (A) US2 is a type I transmembrane protein consisting of a single ER lumenal immunoglobulin superfamily domain, a transmembrane domain, and a short cytoplasmic tail. A crystal structure of US2 bound to MHC class I indicates extensive contacts with the α3 domain, as well as with the α2 domain. Binding to MHC I appears to localize to the ER lumenal domains, the cytoplasmic tail has been shown to be critical for degradation of MHC I, and is predicted to form a 310 helix that interacts with other cellular machinery. (B) The overall domain structure of US2 and US11 is similar, with each consisting of an ER lumenal domain, transmembrane domain, and short cytoplasmic tail at the C-terminus. 59

75 Similarly, the luminal, transmembrane, and cytosolic domains of US2 are fulfill parallel functions to the domains of US11 (Figure 1.9B). The lumenal domain interacts directly with MHC I [192], does not require any additional factors for its interaction [202], and is responsible for retaining US2 within the ER [203]. The cytosolic tail has been shown to be equally important for function and has been suggested to form a 3(10) helix with an interaction face important to MHC I degradation [204]. Finally, the transmembrane (TM) domain plays additional roles separate from both the ER lumenal and cytosolic portions, as replacement with the TM of CD4 disrupts US2 function [205]. Similar to US2, the ER lumenal domain of US11 is responsible for binding to MHC I [206] and its transmembrane domain and cytosolic tail are involved in recruiting the degradation machinery [203,206]. What this suggests for US2 and US11 is that they act through a multistep process, whereby initial binding is followed by recruitment of additional players that are needed for them to function. Both the US2- and US11-mediated degradation mechanisms have been shown to require the molecular chaperone BiP [207] and the translocon-associated protein TRAM1 [208]. Additionally, other dependencies that are unique to US2 or US11 have been identified, suggesting that while both US2 and US11 are ERAD-mediated evasins, they use different pathways to achieve the same outcome. US11 requires the ERAD machinery defined by the E2- recruitment protein AUP1, UBXD8 (ubiquitin regulatory X domain 8) [209], the E3 ligase TMEM129 [161,162], the E2 ubiquitin-conjugating enzyme Ube2J2 [161,162], and the polytopic membrane protein Derlin1 [210,211]. Although the ERAD adapter SEL1L [212] had previously been linked to US11 function, later work suggested that SEL1L may instead be involved in the degradation of US11 itself [213]. 60

76 US2 operates with a different set of cellular factors. The E3 ligase RNF139 (TRC8) [160] is required for US2 function and the thiol oxidoreductase PDIA1 is thought to aid in the release of MHC I heavy chains from US2 prior to their degradation [214]. Additionally, signal peptide peptidase (SPP) has been shown to interact with PDIA1 [214], US2 [215] and RNF139 [160] but conflicting SPP depletion studies raise questions concerning its functions in US2-mediated degradation of MHC I [215] [216]. Finally, the molecular chaperones calnexin and calreticulin have been shown to interact with MHC I in a US2-dependent manner, although a functional contribution to MHC I degradation remains to be established [217]. Beyond that, the US2 and US11 pathways converge, both utilizing the Cdc48-p97-Ufd1- Npl4 AAA-ATPase complex for extraction, deglycosylation in the cytoplasm by peptide N- glycanase [218], heavy chain ubiquitylation, and proteasomal degradation. As previously mentioned, in vitro studies with recombinant US2 proteins showed that other factors are not needed for recognition of MHC I [202], and this is presumably the case for US11 as well. Both US2 and US11 are also able to recruit their required E3 ligases, RNF139 and TMEM129 respectively, independently of the lumenal domains. TMEM129 interacts with US11 through Derlin-1 [161]. The interaction between Derlin-1 and US11 depends on key glutamine residue within US11 s transmembrane domain [161], suggesting that US11 can directly crosslink MHC I to the ERAD machinery. Similarly, the cytosolic tail of US2 was shown to recruit RNF139 [219]. One conclusion from these observations is that US2 and US11 induce degradation of MHC I by simply localizing the MHC I molecule to E3 ligases involved in ERAD. By bypassing the normal ER quality control steps the conformational state of MHC I would have an inconsequential effect on its eventual degradatory fate. Consistent with this view, US2 and US11 61

77 do not seem to require disruption of MHC I structure to degrade it, as shown by the lack of any obvious misfolding of MHC I by US2 [192]. MHC I degradation rates are also much faster in US2- or US11-expressing cells [132,133] than those observed for misfolded MHC I in cells lacking these viral proteins [220]. suggesting that the normal processes of ER quality control are not being followed. However, there are key observations that argue against this model fully explaining the mechanism of action for US2 and US11. Both BiP [207] and PDIA1 [214] have been shown to be required for US2/US11 function. This suggests that soluble ER quality components do play some role in regulating entry into ERAD, even when the substrate is forced into association with the downstream retrotranslocation and ubiquitination machinery. The discovery of additional additional US2 targets beyond MHC I also disrupted this model of highly specific targeting. US2 has been found to target other cellular and plasma membrane proteins for degradation, including integrin α-chains, CD112, IL12RB1, PTPRJ, and thrombomodulin [219]. While all of these factors may bind directly to US2, it is also probable that US2 is influencing the network of ER quality control proteins to affect the maturation outcome of these proteins, strengthening an argument that there are additional proteins recruited by US2 and US11 to manipulate the fate of their targets. 1.8 Rationale and Approach Both US2 and US11 target MHC I for degradation through ERAD, but neither does so through obvious misfolding of their target. This suggests that they instead recruit ERAD machinery to MHC I and/or exclude MHC I from the ER environment required for its normal folding. While some factors, such as the required E3 ligases, have been identified as playing a 62

78 role in US2/US11 degradation of MHC I, it is probable that other proteins also play an important role in targeting heavy chains for the rapid degradation caused by US2 or US11. Therefore, a search will be undertaken to identify novel proteins that contribute to US2/US11 function. Finding these other players is a challenging prospect, as the ER is a complex environment with a high density of protein and a complex web of interactions between interacting chaperones and quality control components. Much of the search for US2 and US11-involved proteins has also focused on the ER, which may mean additional cytosolic players remain to be identified as well. However, identifying these players in MHC I ERAD is important to study how HCMV is manipulating the cellular environment (to both provide potential drug targets to disrupt the virus and to better understand this potential viral vaccine vector), as well as improving our understanding of ER-associated degradation of MHC I and other proteins. The hypothesis that will be explored in this manuscript is that US2- and US11-mediated cross-linking of MHC I E3 ligases alone is insufficient to explain the degradation observed, and that US2 and US11 manipulate additional proteins to fully target MHC I for ERAD. Two approaches will be used to identify likely factors recruited by US2 and US11 to induce ERAD of MHC I. A genomic lentiviral shrna screen will be used to identify transcripts for proteins that result in increased levels of surface MHC I when depleted from US11-expressing cells (Chapter 3). Second, proteins with activities suspected to be involved in either US2 or US11 function based on relevant functions elsewhere in the cell will be investigated by sirna depletion (Chapter 4). In both cases, novel targets will be validated and their effects on surface and total MHC I examined further. 63

79 Chapter 2 Materials and Methods 64

80 2 Chapter Cell lines and antibodies All studies were conducted using the human glioblastoma astrocytoma cell line U373- MG as well as variants that stably express US2 or US11. These cells were provided by Dr. Hidde Ploegh (Whitehead Institute, MIT) and were maintained in high glucose DMEM (Life Technologies, Carlsbad, CA), supplemented with 10% fetal bovine serum, antibiotics (penicillin/streptomycin) and 2 mm glutamine. The mouse monoclonal antibodies (mab) HCA2 and HC10 were used to detect denatured MHC I heavy chains [221,222], and the mouse mab W6/32 was used to detect folded (β2massociated) MHC I [223]. A mouse mab against DNAJC10 was also used (ab573376, Abcam, Cambridge, UK). Rabbit polyclonal antisera specific for the following proteins were used in this study: anti-calnexin generated as described previously [224], anti-calreticulin as described previously [225], anti-cyclophilin C which cross-reacts with cyclophilins A and B ( AP, Proteintech Group, Chicago, IL), anti-cyclophilin B (ab16045, Abcam, Cambridge, UK) anti- PDIA6 (PA3008, Affinity Bioreagents, Golden, CO), anti-malectin (PA , Thermo Scientific, Waltham, MA), anti-tmem33 (SAB , Sigma Aldrich, St Louis, MO), anti- SPFH2 (Erlin2, ab84691, Abcam, Cambridge, UK), anti-dnajc3 (Abcam, Cambridge, UK), anti-pdia5 (ab97328, Abcam, Cambridge, UK), anti-pdia1 (AssayDesigns), anti-erp72 (AssayDesigns), anti-bip (610978, BD Biosciences), anti-us2 and anti-us11 provided by Dr. Domenico Tortorella (Mount Sinai School of Medicine, New York, NY), anti-dnaja4 provided by Dr. J. Young (McGill University), and anti-human α1-antitrypsin (A0409, Sigma Aldrich, St Louis, MO). Mouse anti-gapdh was obtained from Millipore (MAB374, Billerica, MA) and mouse anti-ha mab 12CA5 was provided by Dr. Tania Watts (University of Toronto, Toronto, ON). Goat anti-mouse Alexa 647-conjugated antibody (A21235, Life Technologies) was used 65

81 for secondary staining in flow cytometry. 2.2 Plasmids and Constructs Wildtype US2 and US11 expression constructs in a retroviral vector and C-terminally HA tagged US2 and US11 constructs in pcdna3.1 were provided by Dr. Domenico Tortorella (Mount Sinai School of Medicine, New York). A pcdna3.1 plasmid containing cdna encoding the NHK variant of α1-antitrypsin [226] was obtained from Dr. Richard Sifers (Baylor College of Medicine, Houston). The HA-CypC construct in pcdna3.1+ consisted of wildtype CypC with an HA tag inserted at the N-terminus following the signal sequence. An additional CypC construct lacking the HA tag was synthesized by Life Technologies, and was designed to be resistant to sirnas CypC-(i) and CypC-(ii) (see below). Each sirna recognition site was modified with at least 6 synonymous mutations. Site directed mutagenesis was then used to create two additional mutations, either R89A using forward primer 5`-CGGTTATAAAG- GAAGCAAGTTTCATGCTGTCATCAAGGATTTCATGATT-3` and reverse primer 5`- AATCATGAAATCCTTGATGACAGCATGAAACTTGCTTCCTTTAT-AACCG-3`, or K123A using forward primer 5`-ACATTTCCAGATGAGAACTTCGCACT- GAAGCACTATGGCATTGGGT-3` and reverse primer 5`-ACCCAATGCCATAGTGCT- TCAGTGCGAAGTTCTCATCTGGAAATGTC-3`. These were based on similar mutations in CypB which abolished catalytic activity [119,227] or calnexin binding [228], respectively. The sirna-resistant wildtype CypC and two mutant constructs were subcloned into the plasmid plvx-zsgreen1, resulting in plvx-cypc WT, plvx-cypc R89A, and plvx-cypc K123A. HLA- A68 with an N-terminal HA tag (HA-HLA-A68) [229] was provided by Dr. Kwangseog Ahn (Seoul National University, Seoul, South Korea) and subcloned into plvx-tdtomato. A wildtype P5 construct lacking any epitope tag and wildtype US2 construct with three HA tags 66

82 added to the C-terminus (US2-3xHA) were synthesized by Life Technologies and subcloned into pcdna3.1+/zeo and plvx-ires-tdtomato, respectively. Lastly, GIPZ shrna vectors containing either a nonsense shrna (GIPZ nonsense shrna) or shrnas specific for CypC (GIPZ305215, GIPZ305217, GIPZ , and GIPZ305219) were obtained from Open Biosystems (Huntsville, AL). 2.3 Lentivirus production and transduction Plasmid transfection and virus production Lentiviral production and transduction protocols were based on those described by the RNAi Consortium (Broad Institute, MIT and Harvard University). For production, HEK293T cells were seeded in 5 ml antibiotic-free growth medium into T25 filter cap flasks and transfected with 1000 ng of the appropriate expression or shrna lentiviral construct, 900 ng pcmv-pax2, and 100 ng VSV-G-envelope plasmid pmd2.g using the TransIT-LT1 or Fugene6 transfection reagents. This results in replication incompetent virus due to the separation of viral genes across three separate plasmids, only one of which is packaged into the viral particle itself. Media was switched to high FBS medium (DMEM + glutamine + antibiotics + 30% FBS) at 18 h following transfection. Approximately 48 h later, the medium was harvested and filtered through a 0.22 µm syringe filter Lentiviral Infection Lentiviral transductions were carried out by diluting viral supernatant approximately 1/5 or 1/10 in normal DMEM medium and adding polybrene to a final concentration of 8 µg/ml. The diluted supernatant was added to the target cells, cultured for 24 h, and then the medium was 67

83 replaced with normal DMEM. Infection efficiency was monitored by flow cytometry and detection of fluorescent markers (typically GFP, ZsGreen1, or tdtomato) or by drug selection for resistance markers with puromycin or hygromycin. Infection efficiencies greater than 95% were typically achieved in U373-MG cells. 2.4 RNA Interference. Stealth sirnas (Life Technologies) were used for all knockdowns. CypC-(i) sirna (PPICHSS108319) specifically targeted cyclophilin C, whereas CypC-(ii) sirna (PPICHSS108320) targeted both CypC and CypB. The Stealth negative sirna (medium GC content) was used as a negative control ( , Life Technologies). For depletion of proteins identified by LC-MS/MS (other than PDIA6, for which a previously optimized sirna was available [113]), two to three sirnas were pooled at equimolar ratios (see Table 2.1). Table 2.1. sirna pairs used for depletion experiments. A number of different sirna pairs were used to deplete individual targets. Gene target is the mrna product depleted by the sirna, Stealth sirna ID is the unique reference number used to identify each sirna pair. The sirna sequence of the sense sirna is also shown, not shown is the sequence of the corresponding anti-sense sirna. Gene Target Stealth sirna ID sirna sequence of sense sirna (5` to 3`) PPIC (i) PPICHSS #1-CAAGGUCUUCUUUGAUGUGAGGAUU #2-AAUCCUCACAUCAAAGAAGACCUUG PPIC (ii) PPICHSS #1-GCAACAGGAGAGAAAGGAUAUGGAU #2-AUCCAUAUCCUUUCUCUCCUGUUGC PPIB PPIBHSS #1-GGUCUCUUCGGAAAGACUGUUCCAA #2-UUGGAACAGUCUUUCCGAAGAGACC P5 PDIA6HSS #1-CAAGGCAGAAGUGAUAGUUCAAGUA #2-UACUUGAACUAUCACUUCUGCCUUG CCT8 CCT8HSS #1-CGUUGGAUUAGAUAUUGAGGCUGAA #2-UUCAGCCUCAAUAUCUAAUCCAACG CCT8 CCT8HSS #1-GGACAUGCUGGAAGCUGGUAUUCUA #2-UAGAAUACCAGCUUCCAGCAUGUCC CCT8 CCT8HSS #1-UGUGACAAACGAUGCAGCAACUAUU #2-AAUAGUUGCUGCAUCGUUUGUCACA Malectin MLECHSS #1-CCCUGAGGACCAGAUCCUGUAUCAA #2-UUGAUACAGGAUCUGGUCCUCAGGG Malectin MLECHSS #1-GGACUACGUGCUGGUCUUGAAAUUU 68

84 #2-AAAUUUCAAGACCAGCACGUAGUCC Malectin MLECHSS #1-GGACUUGGAUAUCUUUGAUCGUGUU #2-AACACGAUCAAAGAUAUCCAAGUCC TMEM33 TMEM33HSS #1-ACGUGCUUUGCUGGCAAAUGCUCUU #2-AAGAGCAUUUGCCAGCAAAGCACGU TMEM33 TMEM33HSS #1-CCCAUAUUGUCGGACCUUAUUUAAU #2-AUUAAAUAAGGUCCGACAAUAUGGG TMEM33 TMEM33HSS #1-CAAAGAUUACCACACUUCCAGUUAA #2-UUAACUGGAAGUGUGGUAAUCUUUG Rpn1 RPN1HSS #1-CCCAGAUGAGCUGCACUACACCUAU #2-AUAGGUGUAGUGCAGCUCAUCUGGG Rpn1 RPN1HSS #1-GAGCUACCUCUUUCCUGCUGGCUUU #2-AAAGCCAGCAGGAAAGAGGUAGCUC ERp44 ERP44HSS #1-CCCAGUGAAUAUAGGUAUACUCUAU #2-AUAGAGUAUACCUAUAUUCACUGGG ERp44 ERP44HSS #1-CCCAACCCUCAAAUUGUUUCGUAAU #2-AUUACGAAACAAUUUGAGGGUUGGG ERp44 ERP44HSS #1-CAGAAUGAAGUAGCUCGGCAAUUAA #2-UUAAUUGCCGAGCUACUUCAUUCUG Cnx CNXHSS Not available Crt CALRHSS Not available ERp72 PDIA4HSS #1-ACACUUUCAGCACAGAAAUAGCAAAA #2-TGTGUUUGTCGTGTCTTTUTCGTTTT DNAJC10 (i) DNAJC10HSS Not available DNAJC10 (ii) DNAJC10HSS Not available PDIA1 PDIA1HSS #1-GGGAAGAACUUUGAAGACGUGGCUU #2-CCCTTCTTGUUUCTTCTGCUCCGUU Knockdowns were carried out as described previously [113] using Stealth sirnas (Life Technologies). A 4 µl volume of 20 µm sirna was mixed with 490 µl of OPTI-MEM I and incubated 5 min at room temperature in each well of a 6-well plate. A 6 µl volume of Oligofectamine (Life Technologies) was added, mixed, and incubated 20 min at room temperature. Following the incubation, 2.5 x 10 5 cells in 1.5 ml DMEM + 2 mm glutamine were added, gently mixed, and cultured 4 h before addition of 1 ml DMEM + 2 mm glutamine + 30% FBS. For all knockdowns the procedure was repeated on day 3 to ensure sufficient knockdown of targets at the protein level. 69

85 Knockdown of targets using shrnas was conducting using lentiviral expression vectors carrying the shrna sequences. Lentiviral particles containing the shrna expression vectors were produced as described in Section Table 2.2. Individual shrnas used for depletions. A number of different shrna constructs were used to deplete individual targets. Gene target is the mrna product depleted by the shrna, clone ID is a unique reference number for each shrna, target sequence is the region bound by the shrna within the target mrna, and target region indicates whether the shrna binds within protein coding sequence (CDS), 5` untranslated region (5UTR), or 3` untranslated region (3UTR) of the target mrna. The shrnas with clone IDs beginning with TRC originate from The RNAi Consortium shrna library, and those beginning with GIPZ were obtained from Open Biosystems (GE Healthcare, Lafayette, CO). Gene Target Clone ID Target Sequence (5` to 3`) Target Region HSPA5 TRCN CAAGGTCTATGAAGGTGAAAG CDS HSPA5 TRCN CTTGTTGGTGGCTCGACTCGA CDS HSPA5 TRCN AGATTCAGCAACTGGTTAAAG CDS VCP TRCN CCTATCAACAGCCATTCTCAA CDS VCP TRCN CCTAGCCCTTATTGCATTGTT CDS VCP TRCN CCTGATGTGAAGTACGGCAAA CDS VCP TRCN GATGGATGAATTGCAGTTGTT CDS CUL2 TRCN CGTTTGCAGTTGATGTGTCTT 3UTR CUL2 TRCN GCCCTTATTCAAGAGGTGATT CDS CUL2 TRCN GCAAGCTACATCGGATGTATA CDS CUL2 TRCN CCCTTGGAGAAAGACTTTATA CDS CBLB TRCN CCACATCAACAGCTAAATCAT 3UTR CBLB TRCN CCGGTTAAGTTGCACTCGATT CDS CBLB TRCN GCCCAGAATAATGTCGAAGTT CDS CBLB TRCN CGGGATGTGTTTGGGACTAAT CDS CBLB TRCN CCCTTATTTCAAGCCCTGATT CDS DNAJA4 TRCN CCTGTGTATGTGTTCAGCATT 3UTR DNAJA4 TRCN GCGAGAAGTTTAAACTCATAT CDS DNAJA4 TRCN CCTCGACAGAAAGTGAGGATT CDS DNAJA4 TRCN CAGAAGGATCATAGTGTCTTT CDS DNAJA4 TRCN CACCAGTTATCTGTAACTCTT CDS PDIA5 TRCN CCTGGCAGAAAGATTCCACAT CDS PDIA5 TRCN CGGAATAATGTACTGGTGCTT CDS PDIA5 TRCN GCTCCTGAAGAAGGAAGAGAA CDS PDIA5 TRCN CCACACTGTAAGAAGGTCATT CDS MAN1A1 TRCN GCAGAGTGAATGGAGGCTATT CDS MAN1A1 TRCN CCTTTATCCTAACTATCTGAA CDS MAN1A1 TRCN GCTGGTATTCAGCGCCTTCAT CDS CD160 TRCN CCCAGCTTCATCTAAATACTT 3UTR 70

86 CD160 TRCN CGCGACTAAACTTAATCTGTA CDS CD160 TRCN CTACACAGTGACGGGATTGAA CDS CD160 TRCN TCTCAGTTGATGTTCACCATA CDS CD160 TRCN CTGTACTGTATGGCATAAGAA CDS DNAJC3 TRCN CAGTCGCAGAAACGAGATTAT CDS DNAJC3 TRCN CCAGATAACTTCCAGAATGAA CDS DNAJC3 TRCN GAGCCAAGCATTGCTGAATAT CDS ERLIN2 TRCN CCGCAGAAACTACGAGTTGAT CDS ERLIN2 TRCN CGCAGTGTATGATATAGTGAA CDS ERLIN2 TRCN CACAAGATAGAAGAGGGACAT CDS ERLIN2 TRCN CCAGACAGATGAGGTGAAGAA CDS EDEM3 TRCN CCCAATTTACAGCCGTATATT 3UTR EDEM3 TRCN CCAGAGGCATTTACCACAGAT CDS EDEM3 TRCN GCTAGTTCAATCGACGCTGAA CDS EDEM3 TRCN CAGCTTCAAGAACAATCAGAA CDS MAN1A1 TRCN TCTAGCAGCGGACTAACTTAT CDS MAN1A1 TRCN GACCTAAATTCCTTATCATAT 3UTR MAN1A1 TRCN GACTACTCTCAGCCTACTATC CDS LCK TRCN TTCATTGAAGAGCGGAATTAT CDS LCK TRCN GGGATCCTGCTGACGGAAATT CDS LCK TRCN GCATGAACTGGTCCGCCATTA CDS LCK TRCN GAATGGGAGTCTAGTGGATTT CDS LCK TRCN GCCATTAACTACGGGACATTC CDS LCK TRCN TCACATGGCCTATGCACATAT 3UTR LCK TRCN TGTGTAGCCTGTGCATGTATG 3UTR LCK TRCN CACTGCAAGACAACCTGGTTA CDS Luciferase TRCN CAAATCACAGAATCGTCGTAT - RNF139 TRCN CCTTTCTGTTAGCTGCAACTT CDS SEL1L TRCN GCAGAGTAGTTGCTGGTCAAA CDS UBE2I TRCN AGCAGAGGCCTACACGATTTA CDS UBE2I TRCN AGAAGTTTGCGCCCTCATAAG CDS UBE2I TRCN GCCTACACGATTTACTGCCAA CDS PIAS1 TRCN CATGGCAGTATATCTTGTAAA CDS SAE1 TRCN GCTATGTTGGTCCTTTGTTTA 3UTR SAE1 TRCN GAACAGGTAACTCCAGAAGAT CDS PPIC GIPZV3LHS_ PPIC GIPZV3LHS_ PPIC GIPZV3LHS_ PPIC GIPZV3LHS_ GIPZns Flow Cytometry and FACS Cells were trypsinized and washed in PBS + 3% FBS % sodium azide prior to staining with 1.5 µg/100 µl mab or 1-3 µl/100 µl pab for 30 min on ice. The cell pellet was 71

87 washed and stained for 20 min with 1 µg/100 µl goat anti-mouse Alexa647. The cell pellet was washed, fixed in 0.5% paraformaldehyde in PBS, and strained, before analysis on a BD FACSCalibur flow cytometer (Faculty of Medicine Flow Cytometry Facility, University of Toronto). A similar protocol to that above was used for live cell sorting. Approximately 5 x 10 6 cells were harvested by trypsinization and washed 2x with PBS. Cells were then washed 1x with PBS + 1 mm EDTA + 25 mm HEPES ph % BSA (FACS wash buffer) and then incubated in approximately 1 ml PBS + 1% BSA with µg (for monoclonal) or µl (for polyclonal) primary antibody for 30 min. The cells were washed 1x with FACS wash buffer and then incubated 20 min with 10 µg fluorophore-conjugated polyclonal secondary antibody. The cells were washed 2x with FACS wash buffer and suspended in 1 ml of FACS wash buffer. Cells were then sorted on a BD FACSAria flow cytometer (488 nm, 633 nm, and 407 nm lasers) in the Flow Facility of the Faculty of Medicine at the University of Toronto. 2.6 TRC library lentiviral screen In conducting a genome-wide shrna knockdown screen, we obtained The RNAi Consortium (TRC) human lentiviral shrna library containing 78,432 shrnas, kindly provided by Dr. Jason Moffat (University of Toronto, Toronto, ON). The library is based on the plko.1 vector that contains a puromycin resistance gene for drug selection of transduced cells and uses the U6 promoter to drive shrna production. Viral particles produced for this library are replication incompetent owing to separation of viral genes across three separate plasmids, only one of which is packaged into the viral particle itself. The shrnas are designed to recognize a large number of human transcripts and trigger their depletion and subsequent loss of protein expression. Additionally, multiple independent shrnas are present for each target to allow for 72

88 differing efficiencies of depletion, degrees of toxicity, and off-target effects. Each individual library vector contains a hairpin sequence containing a 21-mer stem sequence and a 6-mer loop sequence containing an XhoI restriction site. The shrnas were inserted into the plko.1 vector, transformed into bacteria, and grown in sets of pooled transformants, and robotic colony picking was used to select clones for re-sequencing to confirm the identity of the inserted shrnas [230]. To construct the lentiviral library from this plasmid collection, a high throughput method of transfecting HEK293T cells in 96-well plates with each library plasmid and two separate packaging plasmids was used to produce lentiviral supernatants [230]. While scale and conditions will differ, this was similar to the three-plasmid protocol described in Section Approximately 300 µl of supernatant was collected from each well, and then a portion pooled to create the TRC lentiviral library used here. U373-MG + US11 cells were plated in T175 flasks with the TRC lentiviral library diluted to obtain a multiplicity of infection of 30% (day 0). After 24 h the lentiviral supernatant was removed and replaced with DMEM supplemented with 10% FBS, 2 mm glutamine, penicillin/streptomycin, and 1.5 μg/ml puromycin. On day 3 following transduction cells were re-plated into Corning 5-layer cellstack vessels (Sigma Aldrich, St. Louis, MO). On day 6 following transduction the cells were trypsinized and counted, yielding approximately 5 x 10 7 cells per flask. The cells were sorted using a FACSAria cell sorter as previously described, with modifications to the FACS buffer (1x PBS, 20 mm Hepes, 5 mm EDTA, 1% FBS). The mab W6/32 primary antibody and Alexa647 goat anti-mouse secondary antibody were used for surface staining of MHC class I. Following staining, cells were suspended at a density of 1 x 10 7 per ml for sorting. Gating was designed to obtain two populations of cells: the 5% of cells expressing the highest levels of MHC class I, and a separate sample consisting of the remaining 73

89 95% of cells. Collected cells were rinsed in PBS and frozen at -80 C for genomic DNA isolation and precipitation. A total of three separate biological replicates were prepared, each consisting of two separate technical replicates of the cell samples. The technical replicates were treated independently from time of transduction to the genomic DNA isolation, after which they were pooled Titre optimization Cells were plated in T175 flasks at 6 x 10 6 cells per flask and transduced with 10-fold dilutions of relevant viral supernatants (typically 100%, 10%, 1%, 0.1%, 0.01%, and 0% dilutions). After 24 h viral supernatants were exchanged for regular DMEM media + 2 mm glutamine + 1x penicillin/streptomycin + 10% FBS with 1 μg/ml puromycin. After 48 h of selection with 1.5 µg/ml puromycin, flasks were trypsinized and live cells counted using trypan blue stain. The same experiment was carried out with duplicate flasks, minus puromycin selection to determine total cell numbers. Data for all viral dilutions were plotted and used to calculate the viral dilution required for incorporation of virus into approximately 30% of target cells Genomic DNA Isolation The Qiagen DNA BloodAmp Maxi kit (Qiagen, 51192) was used to isolate genomic DNA (gdna) following the manufacturer s protocol. To precipitate the isolated DNA, a NaCl solution was added to a final concentration of 0.2 M, followed by 2 volumes of -20 C anhydrous 100% ethanol (Medstore House Brand, P006-EAAN), which was mixed by inverting 15 times. The solution was centrifuged for 15 min at 13K rpm, 4 C, and the supernatant removed by aspiration. The DNA pellet was washed with 500 μl of 70% ethanol at -20 C by mixing 10 times through inversion, sedimented by centrifuging 10 min at 13,000 rpm, 4 C, and the 74

90 supernatant removed through aspiration. A final 1 min centrifugation was used to allow aspiration of any remaining liquid. The DNA pellet was air dried in a sterile biosafety cabinet for 5-10 min and dissolved in 10 mm Tris-HCl, ph 7.4 with a final DNA concentration of 450 ng/μl. The DNA samples were heated for 1 h at 50 C and pipetted/vortexed repeatedly to fully resuspend the DNA. Before providing the samples to the lab of Dr. Jason Moffat for further processing, the DNA concentration was confirmed and adjusted to 400 ng/μl Microarray Gene Modulation Array Platform (GMAP) Final preparation and analysis of the samples by microarray were carried out by the lab of Dr. Jason Moffat and Dr. Troy Ketela, as previously described [231,232] and as listed on the TRC website < The shrna hairpins were amplified from genomic DNA using a PCR master mix containing 160 μl H2O, 24 μl dntp mixture (2.5 mm of each dntp), 30 μl 10X Ex Taq buffer, 30 μl 10 μm LKO-6738 biotinylated 5` primer (5`-bio-AATGGACTATCATATGCTTACCGTAACTTGAA-3`), 30 μl LKO ` primer (5`-TGTGGATGAATACTGCCATTTGTCTCGAGGTC-3`), and 4.5 μl TaKaRa Ex Taq DNA polymerase (5 units/μl). Each genomic DNA sample was split into six tubes consisting of 5 μl of DNA and 45 μl of master mix. PCR conditions consisted of an initial incubation at 95 C for 5 min, followed by 35 cycles to amplify DNA hairpins (95 C for 30 sec, 50 C for 30 sec, and 72 C for 1 min), and a final 10 min incubation at 72 C. Following the above amplification, a second round of amplification was carried out. A 45 μl volume of fresh master mix was added to each previous 50 μl reaction and incubated at 95 C for 7 min, 55 C for 2 min, and 72 C for 1 h, after which the six replicate aliquots were pooled. Following PCR amplification, hairpins were digested into half-hairpins with XhoI. DNA samples were purified using Qiagen s QIAquick PCR purification kit and a QIAvac 24 plus 75

91 vacuum manifold (Qiagen, Hilden, Germany) for hybridization to an Affymetrix Gene Modulation Array Platform (GMAP) microarray chip [232]. A 35 μl volume of purified hairpin was mixed with 4.5 μl of 100 μm blocking primer LKO-6810 (5`- GTCCTTTCCACAAGATATATAAAGCCAAGAAATCGAAATA-3`) and 38.5 μl H2O. The mixture was heated at 99 C for 5 min and 45 C for 5 min. The heated DNA mixture was mixed with hybridization cocktail (6 μl fragmentation buffer, 15 μl GeneChip control oligonucleotide B2 (3 nm), 15 μl GeneChip 20X Eukaryotic hybridization controls (biob, bioc, biod, cre), 3 μl herring sperm DNA, 3 μl BSA (50 mg/ml), 150 μl 2X hybridization buffer, and 30 μl DMSO. The mixture was heated 99 C for 5 min followed by 45 C for 5 min, then kept at room temperature. The probe array was prepared and then incubated with the hybridization DNA mixture in the hybridization oven for 16 h. Following this incubation, the hybridization DNA mixture was removed and replaced with non-stringent wash buffer. GeneChip Operating Software (GCOS) was used to run the microarray wash and stain protocol (FlexGE_WS2v5 Biao30) on an Affymetrix fluidics station. The array was stained with a labeling mix consisting of 1x MES staining buffer, 2 mg/ml BSA, and 10 µg/ml streptavidin-phycoerythrin. The array was then scanned and binding of the fluorophore-labeled streptavidin to the biotinylated shrna amplicons was detected. Raw data preparation of Affymetrix microarray data was carried out by the laboratory of Dr. Jason Moffat. Affymetrix Power Tools v < were used to extract signal for each chip feature (essentially, each dot on the chip). As there were triplicate features per shrna, the median value of the three was used. Following a background correction for non-specific probe binding, replicate arrays were normalized using the Bioconductor affy 76

92 package v in the statistical platform R. This is desribed in futher detail at < and < in [232] Preliminary microarray data analysis Data was provided by the laboratory of Dr. Moffat in excel format with the signal abundance (post-background correction and chip normalization) of each shrna hairpin expressed as a log2 value. Preliminary analysis of the lentiviral screen data involved calculation of the fold difference between each paired high MHC I (top 5% of cells) sample and control (remaining 95% of cells) sample for each of the three biological replicates. The mean fold difference in shrna signal was tested for enrichment, or a departure from a log fold difference of 0, using Student s t-test. The test was conducted for paired data to calculate p-values for each individual shrna, at this point without any multiple testing corrections applied. From this, a list of the top 5% of shrna hits was produced, consisting of those with the greatest fold enrichment in the high MHC I samples. An additional step identified targets in the top 5% of hits that were represented by two or more shrnas, as the independent identification increased our confidence that the target was not present due to off-target shrnas effects GO Annotation Enrichment and Filtering of Datasets In examining groups of targets identified from the lentiviral screen dataset, a number of shrnas were often found that, according to existing literature, are more likely to localize to a cellular location distant from where US11 functions or is known to play a role that makes it unlikely to operate in an ERAD-related process. Alternatively, a group of hits may be found scattered throughout the list that when considered individually may not appear relevant, but when considered as a whole become more significant. By probing what is already known about 77

93 the shrna targets through existing GO annotations, hits that are less likely to be involved can be filtered from consideration, and families or pathways represented by multiple hits can be identified. Not only can novel pathways or families be identified, but the overall quality of the lentiviral screen data can be evaluated by looking for overrepresentation of annotations relating to ERAD. Examination of gene ID lists for enrichment of particular GO annotations was carried out using the Gene Ontology Consortium [233] < and Panther Classification System [234] < analysis tools. For gene IDs that were not matched to the database used by the Gene Ontology software, the gene name was substituted where possible to ensure a higher percentage of genes were included in the analysis. The lentiviral screen hits identified by two or more independent shrnas in the top 5% of shrnas enriched in the high MHC I cells are being used in this analysis. The Gene Ontology algorithms compared the frequency of particular GO annotations in this subset of the lentiviral screen data (390 hits, of which 378 were matched) to the frequency observed in the reference database (20,814 entries). GO terms that appear to be enriched in our screen dataset were assigned a p- value based on the probability of observing enrichment of that particular GO term by random chance. These p-values then underwent a Bonferonni correction to account for the testing of multiple GO annotations. Additionally, annotations were examined using The Database for Annotation, Visualization and Integrated Discovery (DAVID) Bioinformatics Resources (version 6.7) website functional annotation tool < This resource operated similarly to the GO Ontology Consortium tools, but consults a wider range of gene set databases [235,236], including gene ontology, protein domain, pathway, protein interaction, disease, tissue 78

94 expression, and published literature. As with the GO Ontology Consortium tools, it can also be used to either filter hits annotated (or lacking) particular terms from consideration, or it can be used characterize which terms are overrepresented in the lists of hits. We used two lists of hits in this analysis: a dataset consisting of hits from the top 5% of shrnas enriched in the high MHC I cell sample, and a second dataset composed of hits identified by 2 or more shrnas within the top 5% of shrnas enriched in the high MHC I cell sample. To evaluate enrichment of a particular annotation in the analysis results the platform calculated p-values estimating the probability of observing enrichment of particular database categories by random chance. It also applied corrections to compensate for multiple hypothesis testing (Bonferroni and others). P-values were used to rank and identify the enriched terms with the lowest probability of being observed by random chance (rather than applying a threshold of p-values of 0.05 or lower, though in most cases p-value met this requirement) Determination of q-value-based False Detection Rates The statistical platform R (version GUI 1.66 Snow Leopard build 6956) along with the Bioconductor q-value package (release 3.1) were used to estimate q-values from p-values [237]. Q-values are produced as a method of taking the distribution of p-values into account when determining the significance of a particular measurement, in order to account for multiple hypothesis testing in large datasets. It asks what the false discovery rate (FDR) would be when a particular result is considered to be significant. The FDR is the number of false positives as a fraction of the total number of features (shrnas) considered, and is typically determined when looking at the list of q-values in ranked order. The technique takes advantage of several properties of p-values for tests described by the null hypothesis, which in this case is that an shrna has shown significant enrichment in the high MHC I cell sample did so by random 79

95 chance. When large numbers of tests are performed, the p-values for tests described by the null hypothesis should be uniformly distributed between 0 and 1, whereas those for the alternative hypothesis should be biased towards 0 [238]. If one is able to estimate the proportion of p-values falling under the null hypothesis it is then possible to extrapolate to how many are not due to random chance, and to determine how many false positives are present at different statistical cutoffs. This is the essence of the information contained with q-values. Detailed below is the approach used to calculate q-values for the lentiviral screen dataset. It focuses on the inputs set for this analysis, but leaves out the mathematical details underlying the calculations as a full explanation of the statistical mathematics is left to better qualified specialists [237]. In calculating q-values, we used the p-values calculated in the preliminary analysis of the microarray data. A small number of shrnas exhibited identical log2 values for all three replicates (to an extremely high precision that was not technically realistic). These were likely artifacts and as such excluded from the analysis. Calculating q-values requires an estimation of π0, which is defined as: Where the number of truly null features is the number of shrnas for which we fail to discard the null hypothesis of no enrichment in the high MHC I cell sample, and the total number of features is the total number of shrnas. The value for π0 can be estimated by looking at the distribution of p-values for the dataset. Taking advantage of the fact that the histogram of null p- values should be randomly and uniformly distributed (or flat) between 0 and 1, we can estimate π0 for the lentiviral screen data by looking at the portion of the histogram corresponding to p- values greater than 0.5 (which should be mostly null p-values, and therefore relatively flat). The 80

96 alternative p-values (for those shrnas with which we can discard the null hypothesis) should be clustered at the other end of the p-value histogram, closer to 0. A bootstrap method [239] (a method of random sampling with replacement) was applied to estimate π0. In estimating π0, it is also required to determine another variable, λ. Lambda is the p- value cutoff above which all p-values are thought to come from data reflecting the null distribution (from t-tests that are describing data that is randomly different). Lambda was restricted to be a fixed range of 0 to 0.95, and a lambda step size (used to determine the number of points in the range of potential lambda values) of 0.05 was set. Finally, a local FDR estimate is calculated for the p-values. A probit transformation method was applied to the p-values. This method allows estimation of a local FDR that avoids edge effects of the 0 to 1 range the p-values are contained in. To rephrase, it eliminates some mathematical issues that can occur at the extreme maximum and minimum p-values. Several other values were adjusted to aid in the analysis. Any local FDR values greater than 1 were set to 1, to avoid the issue of having false detection rates greater than 100%. An adjust value that influences the function estimating local FDR was set to 1.5, the recommended default value. A threshold value for the tails of the p-value distribution was also set to Once the analysis is complete, a false detection rate must be selected. As with choosing a suitable p-value in single statistical tests, competing concerns as well as common practice plays a role. The cutoff must not be so stringent, represented by a small q-value or FDR, that useful hits are discarded. It must also not be so relaxed, represented by a large q-value or FDR, that the list of significant results is flooded with false positives. As the lentiviral screen dataset was hoped to enrich for significant hits that would be used for further study, we did not wish to discard too 81

97 many results out of hand. Therefore, an FDR/q-value cutoff of 0.2 was used for significance, representing a statistical likelihood of only 1 false positive in every 5 results RNAi Gene Enrichment Ranking (RIGER) analysis The above q-value analysis treats all shrnas independently. However, the lentiviral screen library contains multiple shrnas that target a single gene transcript. An alternative approach that incorporates the data from all shrnas specific for a single target is known as RNAi Gene Enrichment Ranking (RIGER) [240]. This approach also has the advantage of accounting for multiple hypotheses testing and not requiring a normal distribution of data. The GENE-E software was used with the RIGER extension to perform this analysis [241]. The approach uses Kolmogorov-Smirnov (KS)-based statistics to construct gene scores from the shrna construct datasets, and is based on GSEA methodology [242]. By considering the entire shrna construct profile for each target it calculates gene scores using the entire list of shrnas and does not use any arbitrary cutoffs, among other benefits [240]. As for calculation of q- values, the inputs used for RIGER (and the GSEA below) are described, but detailed discussion of the mathematics underpinning the statistics are left to better-qualified individuals. Briefly, shrnas are scored using the signal-to-noise metric [243] to quantify the differential enrichment between control and high MHC I cell samples. This is followed by calculation of a raw enrichment score, which is then normalized to account for the differing number of shrnas specific for screen targets. A null distribution generated from 10,000 random permutations of a similarly sized hairpin set was used to carry out this normalization. The shrna identifiers were then converted to the appropriate gene identifier (in this case a fusion of the gene name, gene ID, and gene description). The primary outputs used were a normalized enrichment score, gene rank, and RIGER p-value associated with each gene score. 82

98 2.6.8 Gene Set Enrichment Analysis (GSEA) While the RIGER analysis provides a convenient ranking of gene targets from the lentiviral screen dataset, without any arbitrary cutoffs. However, it does not provide an of the biological information present in the early annotation filtering and enrichment analyses previously described. Rather than creating an arbitrary cutoff for the RIGER scores, a technique known as a Gene Set Enrichment Analysis can be conducted to capture which functional groups of genes are clustered towards higher RIGER scores, or are overrepresented in the high MHC I cell samples. Essentially, this technique is an attempt to incorporate the positive elements of the q-value, RIGER, and GO annotation analyses into a single technique. Gene Set Enrichment Analysis (GSEA) software is a common approach to characterizing gene expression profiles from microarray data [244,245]. Data output from the RIGER analysis was used as a ranked gene list for GSEA, with the goal of identifying gene sets that were enriched in the high MHC I samples versus control cell samples. The ranked normalized enrichment score and Entrez gene IDs were used as input for comparison to the GO gene sets and curated gene sets molecular signature databases obtained from GSEA website (Broad Institute, MIT and Harvard, Cambridge, MA). The GO gene set (all GO gene sets, Entrez IDs, c5.all.v5.0.entrez.gmt) consists of lists of genes annotated with particular GO terms. This included GO biological processes, GO cellular components, and GO molecular functions. The curated gene sets (all curated gene sets, Entrez IDs, c2.all.v5.0.entrez.gmt) included information from publications in PubMed, knowledge of domain experts, data on chemical and genetic perturbations, canonical pathways, BioCarta gene sets, KEGG gene sets, and Reactome gene sets, as well as other database sources. These are described in more detail online < 83

99 In conducting the GSEA analysis 1000 permutations were used to produce null data sets, a classical enrichment statistic was used (the most conservative option within the software), and gene sets larger than 500 or smaller than 10 were excluded (as this analysis can have difficulty in properly determining significance of or normalizing these small and large datasets, respectively Overlap analysis Validation of lentiviral screen results was carried out by comparing the hits identified from our GMAP data to those from an independent study of ERAD-linked proteins. We compared the hits from two different sets of data, the top 5% of hits enriched in the MHC I high samples, and the top 5% of hits enriched in the MHC I high samples that were also identified with 2 or more independent shrnas. In contrast to the approaches described previously, we did not assume that the control and high MHC I sample data were paired. These gene lists of hits were overlapped with the high confidence immunoprecipitated proteins (HCIPs) identified in a digitonin solubilization by LC-MSMS following immunoisolation of various ERAD-related bait proteins, as determined and listed in Supplementary Table 2A of the 2012 paper by Christianson, et al. [87]. Processing of this list of LC-MSMS hits resulting in 148 unique Entrez Gene IDs, of which 134 were present in the TRC 80K shrna library. These 134 LC-MSMS hits were compared to the 3518 hits in the top 5% shrna hit group and the 390 hits in the top 5% and greater than 2 shrna hit group. Statistical significance of overlap was tested by a bootstrap approach using 10,000 random overlaps without replacement to determine the degree of overlap expected by random chance. Estimated p-values were corrected using the Bonferroni method. Calculations were performed using the R software platform [246]. 84

100 2.7 Xbp-1 splicing assay Detection of the unfolded protein response through Xbp-1 splicing was carried out as described previously [113]. Briefly, total RNA was isolated from cells using the RNeasy mini kit (Qiagen, Valencia, CA). Primers specific for both the spliced and unspliced forms of Xbp-1 mrna (5`-GGAGTTAAGACAGCGCTTGG-3` and 5`-GAGATGTTCTGGAGGGGTGA-3`) were used in a one-step reverse transcriptase-pcr (Qiagen) to obtain fragments of differing sizes depending on the spliced state of the transcript. The PCR program consisted of an initial 30 min. incubation at 50 C to synthesize cdna, 15 min at 95 C to denature templates, and subsequently 30 cycles (1 min at 94 C, 1 min at 60 C, and 1 min at 72 C) to amplify transcript fragments. A final 10 min. incubation at 72 C completed any partial extensions. Following RT- PCR, DNA fragments were analyzed by 2% agarose gel electrophoresis. 2.8 qpcr For quantitative real-time PCR of mrna levels, total RNA was first isolated using the Qiagen RNeasy Mini kit. The Life Technologies Superscript VILO cdna Synthesis Kit with random primers was used to synthesize cdna. TaqMan real-time PCR using TaqMan Gene Expression Master Mix (Applied Biosystems, Waltham, MA) and pre-optimized TaqMan gene expression assays (PPIC (CypC) Hs , MLEC _m1, TMEM33 Hs _m1, CCT8 Hs _mH, PDIA6 Hs _m1, HLA-A Hs _g1, RPN1 Hs _m1, and actin Hs , Life Technologies) were used to quantify mrna depletion and HLA-A transcript levels. In all cases, a 7500 Real-Time PCR System was used for analysis (Applied Biosytems) using the standardized protocol for these reagents. The delta delta threshold cycle (ΔΔCt) method was used to determine mrna levels relative to the reference transcript beta actin [247]. Briefly, ΔΔCt = (Ct(target, untreated) 85

101 (Ct(reference, untreated)) (Ct(target, treated) Ct(reference, treated)). This can then be expressed as 2 ΔΔCt to calculate the ratio of the target transcript to the reference transcript. 2.9 Metabolic labeling Pulse chase U373-MG cells were starved of methionine and radiolabeled for 5-10 min with 100 µci/ml of [ 35 S] methionine. Following radiolabeling, cells were either chased with 2 mm methionine-supplemented media or immediately prepared for lysis as follows. Cells were washed twice with chilled PBS and lysed in RIPA buffer (10 mm HEPES ph 7.4, 150 mm NaCl, 1% Nonidet P-40, 0.25% sodium deoxycholate, 0.1% SDS, 10 mm iodoacetamide, and 1x proteasome inhibitor cocktail (P8340, Sigma Aldrich)). Lysates were rocked for 30 min and centrifuged for 10 min at 12K rpm to remove insoluble debris. The supernatants were collected and relevant proteins immunoisolated through incubation with 30 μg of the appropriate monoclonal antibody or 20 µl of the appropriate polyclonal serum for 2 h, followed by incubation with 30 μl packed protein A or protein G beads (as appropriate for monoclonal antibody isotype) for 1 h. Beads were washed 4x with either 0.2% digitonin or 0.5% NP40, in 10 mm Hepes ph mm NaCl. Material was eluted off the beads through boiling for 5 min with 1x SDS-PAGE sample buffer + DTT and run on large SDS-PAGE gels overnight. To image the gels, they were first fixed with 10% TCA, 30% ethanol, and 10% acetic acid for 30 min. The TCA fix was removed through 2x washes with water, followed by an additional 2x10 min washes with water. Lastly, gels were incubated with 1 M sodium salicylate, dried, and exposed to autoradiography film. 86

102 2.9.2 Pulse protocol with MG132/lactacystin treatment U373-MG cells were starved of methionine and radiolabeled with 100 µci/ml of [ 35 S]- methionine for 10 min. The cells were then washed twice with chilled PBS and lysed in RIPA buffer. The lysates were rocked for 30 min then centrifuged for 10 min at 11,000 rpm to remove nuclei and insoluble material. Supernatant fractions were treated with 10 µg w6/32, 10 µg HC10, and 10 µg HCA2 to isolate MHC class I and with 20 µl anti-calnexin antiserum as a radiolabeling and immunoisolation recovery control. After 2 h a mixture of protein A-Sepharose (GE Healthcare) and protein G-Sepharose beads (Life Technologies) were added and incubated for 1 h. Beads were washed 4x with 0.5% Nonidet P-40, 10 mm Hepes ph 7.4, 150 mm NaCl and then proteins were eluted, separated by SDS-PAGE under reducing conditions and visualized by fluorography. For inhibition of proteasomal activity, cells were pre-treated for 60 min with 25 µm lactacystin (Abcam), starved of methionine in the presence of 25 µm MG132 (Boston Biochem), radiolabeled in the presence of both inhibitors, and chased in the presence of 25 µm MG LC-MSMS Mass spectrometry to identify bait protein interaction partners was based on a protocol developed by Christianson et al. [87]. Three cell lines were created for these experiments: 1) U373-MG + nonsense shrna, which expresses the plvx-empty and GIPZ nonsense shrna vectors and was used to detect non-specific proteins isolated by this protocol, 2) U373-MG + US2-3xHA, which expresses US2 with a triple HA tag at the C-terminus, and 3) U373-MG + US2 + HA-HLA-A68, which expresses both untagged US2 and N-terminally HA-tagged HLA- A68. Furthermore, the U373-MG + US2-3xHA cells were transduced with lentivirus 87

103 expressing either a GIPZ nonsense shrna vector (U373-MG + US2-3xHA + nonsense shrna), a lentiviral pool of four shrnas (GIPZ305215, GIPZ305217, GIPZ , and GIPZ305219) to deplete CypC (U373-MG + US2-3xHA + CypC KD), or a construct for overexpression of CypC (U373-MG + US2-3xHA + CypC OE). The U373-MG + US2 + HA-HLA-A68 cells were transduced only with lentivirus containing the nonsense knockdown shrna vector (U373-MG + US2 + HA-HLA-A68 + nonsense shrna). Cells were cultured in 5-layer CellSTACK vessels (Corning, Cat. No. 3319) with a surface area of 3,280 cm 2. Approximately 1 x 10 8 cells were incubated for 8 h with 25 µm MG132 and then were lysed in 10 mm HEPES ph 7.4, 150 mm NaCl, 1% digitonin, 10 mm iodoacetamide, and 1x proteasome inhibitor cocktail (P8340, Sigma Aldrich). Nuclei and insoluble debris were removed by centrifugation and lysates were incubated for 2 h with anti-ha conjugated agarose beads (Thermo Fisher Scientific, Waltham, MA) to isolate the tagged bait protein and any associated prey proteins. Beads were washed 3 times with 10 mm HEPES ph 7.4, 150 mm NaCl, 0.2% digitonin and 3 times with 50 mm ammonium bicarbonate ph 7.4. To elute bound proteins, the beads were incubated overnight at 37 C with 50 mm ammonium bicarbonate, ph 7.4 and 0.1% Rapigest (Waters, Cat. No ), a mass spectrometry compatible detergent that is hydrolyzed at low ph. Following the incubation, the eluate was removed and any remaining material was eluted from the beads by boiling 5 min with 0.1% Rapigest, 50 mm ammonium bicarbonate, ph 7.4. Samples were reduced, alkylated with iodoacetamide and digested overnight with sequencing grade chemically modified trypsin (V5111, Promega, Madison, WI). The ph was reduced to 2 by the addition of trifluoroacetic acid to a final concentration of 0.5%, hydrolyzing the Rapigest detergent during a 45 min incubation at 37 C. Samples were centrifuged and the aqueous fraction was collected (discarding the insoluble portion of the detergent). Samples were concentrated in a SpeedVac concentrator 88

104 (Savant) and desalted with C18 ZipTips (Millipore) for analysis on an Orbitrap Elite LC-MS/MS by the SPARC Biocentre (Sick Kids, Toronto) Mass Spectrometry Analysis The Peaks 7 software package (Bioinformatics Solutions Inc., Waterloo, ON) was used to analyze mass spectrometry results [248]. Peaks Studio calculates a p-value for each peptide, representing the probability that a false identification has occurred. These p-values are then represented as a significance score (-10log10(p-value)) for ease of consideration. For the US2-3xHA immunoisolations, only peptides with a significance score greater than 20 (p-value less than 0.01) were considered for subsequent analysis. Furthermore, peptides that were detected in only a single immunoisolation sample were not included. Identified peptides were then matched to the most likely originating protein, again with a stringent significance score of 20, and with a minimum of 2 unique peptides contributing to the match. HA-HLA-A68 immunoisolations were analyzed in a similar way but using a significance score of 5. With the high precision mass spectra available from an Orbitrap mass analyzer, peptide area derived from the ion current versus retention time plot (quantified area) is used by the Peaks Studio software to estimate relative abundance of identical peptides between samples. Multiple unique peptides are considered for each protein and are used to provide a semi-quantitative estimate of protein abundance. This approach has been validated in a number of applications [ ]. We obtained protein abundance data for US2-3xHA immunoisolations (each consisting of two biological replicates) from cells expressing normal, depleted or overexpressed levels of CypC. To remove contaminating proteins from these results, protein abundances obtained from the anti-ha immunoisolation of U373-MG + nonsense shrna cells were subtracted from the three US2-3xHA immunoisolations. The same background subtraction was applied to protein abundances obtained from the HA-HLA-A68 immunoisolation. Finally, to 89

105 account for variations in immunoisolation and sample preparation between US2-containing samples, all protein abundances were normalized to those of the US2 bait protein. 90

106 Chapter 3 Use of a genome-wide lentiviral shrna screen to identify novel factors involved in US11-mediated degradation of major histocompatibility complex class I 91

107 3 Chapter Introduction A major challenge in studying complex biological systems is identifying the individual components that function in specific processes. Major factors with essential roles are often identified early while other proteins playing important but not essential roles can be more difficult to discover. While proteins have been identified that are of importance to US11 function, our objective was to conduct a broader search for novel factors required by US11 by depleting cellular proteins and examining the impact of their loss on MHC I steady state expression levels. Initially we began by depleting chaperones of interest and examining their role in US2- and US11-mediated degradation, either by western blot of total MHC I or by monitoring surface expression following depletion through flow cytometry. In depletion experiments for calnexin, calreticulin, cyclophilin B, ERp57, ERp72, PDIA1, and ERdj5 only very modest effects were observed at best. There were some depletion targets that did show a more significant effect, namely PDIA5 and cyclophilin C, and these will be discussed in more detail in Chapter 4. However, we were aware that these targets were selected based on our specific interest in studying the role of molecular chaperones in US2- and US11-mediated MHC I degradation, and that our investigations were restricted by our assumptions and previous knowledge when taking such a directed approach. Some studies have taken a broader approach recently, [160,162,219], but most have focused on directed approaches to examining the role of host cell proteins involved in US2 or US11 function. In addition to the more obvious targets examined by sirna depletion, we were interested in identifying factors required by viral evasion molecules with less 92

108 obvious links to MHC I or ERAD. To broaden the net cast by our search, we decided to combine our approach of depleting proteins of interest with an unbiased search for relevant players throughout the expressed genome. Therefore, we designed, validated, and carried out a screen of expressed transcripts using a lentiviral library of shrnas. This approach using a pool of lentiviral shrna constructs consisting of 78,432 different shrna hairpins corresponding to approximately 15,982 human targets allowed a functional evaluation of factors involved in US11 function. A pooled library was more convenient compared to individual shrnas, as it did not require robotics or automation to screen more than several hundred potential targets. While it is possible to use an arrayed format of the library that consists of multiple sub-pools of library shrnas, we chose to use a single combined pool to simplify sample collection and analysis. The stable expression of shrnas also allowed the establishment of cell lines expressing individual shrnas, which was useful during later validation steps Screen Design The selected lentiviral library (TRC 80K) was provided by Dr. Jason Moffat and was a version of The RNAi Consortium (TRC) library which was previously developed and validated in part by Dr. Jason Moffat s group at the University of Toronto [230,231]. Surface expression of MHC, as measured by flow cytometry, was chosen as the form of selection for use in identifying shrna hits in the screen. In both US2 and US11-expressing cells, MHC I surface expression is low. Depletion of a host cell factor required by an immunoevasion molecule by a lentiviral shrna would be expected to increase total and surface MHC I levels, and shrnas that increase MHC I would be expected to be enriched in the high MHC I cells. Conversely, shrnas that decreased MHC I levels would be less likely to be found in the high MHC I-expressing cells. Lastly, shrnas that had no effect would be present at similar frequencies in either group of cells. 93

109 A large number of cells is required for this type of library screen. To ensure sufficient detection of the effect of each shrna we chose to infect a sufficient number of cells such that there would be approximately 100 cells that take up each shrna. To minimize the possibility of multiple shrnas entering the same cells, a multiplicity of infection (MOI) of 0.3 was chosen. Consideration of these requirements leads to an estimate for the starting number of cells needed for transduction of 2.7 x 10 7 cells. Once the transduction is complete, untransduced cells are eliminated through puromycin selection. On day 6 following transduction the cells are analyzed by fluorescence-assisted cell sorting (FACS). The cells are sorted into two groups based on MHC I expression. Those cells expressing the highest levels of MHC I (the top 5%) are sorted into one group, and the remaining 95% are sorted into the other. Genomic DNA is then isolated for each sample, precise PCR conditions are used to amplify the shrna sequences in a quantitative manner using biotinylated primers, and the amplified DNA is applied to the microarray chips for analysis. From the genomic DNA isolation onwards, standardized protocols based on previous procedures in use by the Moffat laboratory [230] are applied to ensure equivalent handling of the DNA samples and appropriate data output from the microarray analysis. 3.2 Results Selection of U373-MG cells and US11 for the shrna library screen. A critical step in setting up the lentiviral screen was selection of a suitable cell system. Ideally the cells would (1) allow stable expression of US2 or US11, (2) exhibit a strong phenotype of MHC I loss from the cell surface upon US2 or US11 expression, (3) be easy to culture and expand to large numbers, (4) be easily infected by lentiviruses, (5) be amenable to fluorescence-assisted cell sorting (FACS), and (6) exhibit an increase in MHC I expression levels 94

110 following disruption of US2 or US11-mediated ERAD of MHC I, such as that seen with proteasome inhibition. Initially, HeLa cells stably expressing a GFP-tagged HLA-A2 construct followed by introduction of stably expressed US2 or US11 were evaluated for their suitability for the planned screen. These had been used previously by the laboratory of Dr. Paul Lehner in an sirna-based screen to identify the E3 ligase utilized by US2 to degrade MHC class I [160], and therefore appeared to meet the first three criteria. Lastly, the breadth of studies utilizing HeLa cells in combination with flow cytometry, as well as specific examples of their use in screens [252], supported their use in our genome-wide screen. The class I degradation phenotypes of the HeLa cells co-expressing GFP-HLA-A2 with US2, US11, or neither viral protein were confirmed. The presence of either viral immunoevasion molecule resulted in a loss of GFP-HLA-A2 expression, as shown by flow cytometry (Figure 3.1A). In addition, treatment of either US2- or US11-expressing HeLa + GFP-HLA-A2 cells with the proteasome inhibitor MG132 increased MHC I levels in both cell lines, and to a lesser degree in the HeLa + GFP-HLA-A2 cells without either viral protein (Figure 3.1B). To confirm an impact on surface MHC I, cells were transduced with shrnas specific for targets known to be required by US2 and/or US11. Two shrnas were selected, one specific for the E3 ligase RNF139 and one specific for SEL1L, as well as a control shrna that targeted luciferase that is not expressed in these cells. It was expected that depletion of RNF139 would increase MHC I levels specifically in US2+ cells, and that the SEL1L depletion would specifically impact US11 function. 95

111 Figure 3.1. Testing of HeLa cells for use in a lentiviral screen. (A) HeLa cells stably expressing an N-terminally GFP-tagged HLA-A2 construct show reduced expression of MHC I due to US2 or US11 expression. Cells were examined by flow cytometry for GFP expression. (B) Treatment of HeLa cells +/- US2 or US11 with the proteasomal inhibitor MG132 exhibit increased HLA-A2 levels, as shown by GFP fluorescence. (C) HeLa cells resist depletion of target proteins with lentiviral-based transductions. HeLa + GFP-HLA-A2 + US2 and HeLa + GFP-HLA-A2 + US11 cells were transduced with lentiviral particles that express shrnas specific for RNF139, SEL1L, or luciferase (as a control). Approximately 2 days following transduction, cells were selected with puromycin for 4 days and analyzed by flow cytometry for expression of GFP on day 6 following transduction. 96

112 However, there were difficulties in efficiently infecting the HeLa cells with unconcentrated lentiviral test stocks. In addition, little or no change in MHC I levels following transduction with either shrna in any of the HeLa + GFP-HLA-A2 cell lines (Figure 3.1C). This would necessitate larger quantities of lentiviral library supernatant to boost the MOI and concentration of the lentiviral particles. As the procedure to create and validate the lentiviral library supernatant is quite complex and laborious, it is preferable to minimize the amount of this complex reagent. The poor transduction efficiency and its subsequent impact on protein depletion also complicated efforts to validate the suitability of these cell lines for the lentiviral screen. Therefore, an additional system based on the U373-MG cell line was tested for suitability. U373-MG cells are a glioblastoma astrocytoma derived from a malignant tumor. The U373-MG cell lines lack a stable GFP-tagged MHC I construct but do stably express either US2 or US11. This resulted in a loss of endogenous surface MHC I expression as shown by flow cytometry (Figure 3.2A). In contrast to the HeLa cell system, U373-MG + US2 cells exhibited a strong increase in surface MHC I following depletion of RNF139, but little to no effect following depletion of SEL1L (Figure 3.2B). Conversely, U373-MG + US11 cells demonstrated an increase in surface MHC I following depletion of SEL1L. The dependency of US2 on RNF139 and US11 on SEL1L was expected based on previously published reports [160,209]. All depletions were carried out using lentiviral transductions with diluted lentiviral supernatant, and U373-MG cells were easily titrated with the lentiviral library supernatant to determine the concentration required for an MOI of 0.3 (Figure 3.2C). 97

113 Figure 3.2. Testing of U373-MG cells for use in a lentiviral screen. (A) U373-MG cells expressing either US2 or US11 exhibit decreased levels of surface MHC I. Surface MHC I was monitored by flow cytometry using mab W6/32. (B) Depletion of RNF139 or SEL1L in U373-MG cells stably expressing US2 or US11 results in increased surface MHC I in a viral immunevasin-specific manner. U373-MG cell lines were transduced with lentiviral shrnas as carried out previously using HeLa cells, and surface MHC I was monitored by flow cytometry using mab W6/32. (C) The TRC 80K shrna lentiviral library was titred for transduction of U373-MG + US11 cells. Cells were transduced at varying dilutions, selected with puromcyin to remove untransduced cells, and live cell numbers and MOI values determined using crystal violet staining. Conditions similar to those planned for the large scale lentiviral screen were used in order to determine a suitable dilution to obtain an MOI of

114 For these reasons, as well as the ease with which these cells could be cultured, and the previous history of study of US2 and US11 function that existed with these cells, the U373-MG cell line was selected for use in the lentiviral screen. During experiments examining the baseline surface expression levels of MHC I, it was noted that the U373-MG + US11 cells possessed lower surface MHC I staining than the U373-MG + US2 cells, compared to the control line (Figure 3.2A). This larger gap in MHC I expression resulted in a larger potential functional range of screen output, and as such US11 was selected as the first candidate viral protein for our screen, with examination of factors involved with US2 set aside for later studies Full-scale screen of U373-MG + US11 cells for human cell factors involved in US11 function Moving forward with the U373-MG + US11 cell line, an experimental design was developed for the full-scale screen (Figure 3.3). U373-MG + US11 cells were transduced with the lentiviral library with a MOI of 0.3. Following drug selection to eliminate untransduced cells, the cells were stained with W6/32 mab to detect levels of β2m-associated MHC I on the cell surface (Figure 3.3). FACS was then used to sort the transduced U373-MG + US11 cells into two groups: the top 5% of cells expressing the highest levels of MHC I on the surface, and a second control population made up of the remaining 95% of cells with normal or lower levels of surface MHC I (a ratio selected based on early validation and ease of application during sorting). Genomic DNA extracted from the top 5% of cells expressing high surface MHC I and from the remaining 95% of cells was prepared for analysis by microarray with the assistance of Dr. Jason Moffat s group, and the analysis was carried out as done previously [231,232]. The abundance of shrnas in each sample was determined to calculate the relative enrichment of shrnas within the top 5% of cells expressing high MHC I compared to the remaining 95% of cells. 99

115 Figure 3.3. General workflow of the U373-MG + US11 cell lentiviral screen. U373-MG + US11 cells were transduced with the ~80,000 shrna lentiviral library at a multiplicity of infection of approximately 0.3. Untransduced cells were eliminated by puromycin selection. Following several days of culture to allow for depletion of shrna targeted transcripts and proteins, the U373-MG + US11 cells were stained with the MHC I antibody W6/32 and sorted into two populations, those expressing high MHC I levels (the top 5% of cells) and the remaining normal/low expressing cells (remaining 95% of cells). Following genomic DNA isolation and PCR amplification of incorporated shrnas, microarray analysis was used to characterize shrna abundance in each cell population. 100

116 3.2.3 Initial ranking of screen targets for magnitude of effect on US11 function A total of three biological replicates were prepared, consisting of paired samples of the top 5% of high MHC I cells and the remaining 95% of cells. Each pair of samples was prepared and submitted for microarray separately. The abundances of each shrna were obtained, expressed on a Log2 scale, and from these a basic analysis was conducted to examine the relative fold difference in abundance for each shrna. These values were averaged and the standard deviation determined. This allowed a Student s t-test to examine if the average result was statistically different from the null hypothesis, which states that there was no fold difference in shrna enrichment between the high MHC I cells and control cells (Figure 3.4A). In conducting this analysis, 9 shrnas were found to have unusual and most likely artefactual patterns between replicates (Table 3.1). The signal values obtained between replicates were essentially identical to an unrealistic level of precision (for instance, decimal values identical to nine places), resulting in extremely low p-values (10-30 or lower). Therefore, the results obtained with these shrnas were determined artefactual and excluded from subsequent analyses. Table 3.1. Individual shrnas excluded from further analysis due to unusually low p- values. Clone ID Gene Symbol Gene ID TRCN NAGK TRCN GLT25D TRCN RAB5A 5868 TRCN HOXB TRCN PIK3AP TRCN MGAT TRCN CPT1A 1374 TRCN ADAMTS TRCN

117 Figure 3.4. Visual Summaries of the lentiviral screen output and basic filtering processes. (A) A Volcano plot showing fold enrichment (within the high surface MHC I cells) on the x- axis and Log (p-value) on the y-axis was used to graphically display the results of the lentiviral screen in U373-MG + US11 cells. Fold enrichment is an average of the triplicate result for each shrna (expressed as a Log2 value) and the p-value was calculated using Student s t-test comparing the mean of triplicate results to the expected null (no enrichment) result (expressed as Log10 value). The top 5% of shrnas with the highest enrichment in the high MHC I cells are shown in red. (B) The average fold enrichment for all ~80,000 shrnas plotted in ranked order (arranged from left to right in order of highest fold enrichment in the high surface MHC I cells, to lowest fold enrichment). The top 5% of shrnas (3,922) most enriched in the high surface MHC I cells are shown in red, and the 5% of shrnas most depleted from the high surface MHC I cells are shown in green. (C) The number of shrnas for each unique target present in the top 5% of enriched shrnas was determined. In most cases only a single shrna for a particular target was present in the top 5% of enriched shrnas, but there were 368 targets identified by two independent shrnas, 21 targets by three independent shrnas, and a single target identified by four shrnas. 102

118 The average enrichment of shrnas in the top 5% of high MHC I cells were examined in preliminary analyses. For this initial examination the p-values were not considered as the Student s t-test applied earlier is not truly suitable for multiple testing conducted in this screen. Most shrnas appeared to have very minor or no effect on surface MHC I, but a selected set of shrnas correlated with increased MHC I surface expression (Figure 3.4B). Similarly, a number correlated with decreased MHC I surface levels, but we were less confident in the validity of these observations as our protocol was not designed to detect depletion of shrnas from the relatively smaller population of cells making up the high MHC I sample. To continue this analysis, it was necessary to filter the dataset as it is difficult to consider over 78,432 shrnas at once (Figure 3.4A). The 5% of shrnas most enriched in the high MHC I cell sample are highlighted in red in Figure 3.4B. Restricting analyses to only these shrnas reduces the portion of the dataset under consideration from 78,432 down to 3,922, representing 3,518 unique gene targets. However, over three thousand hits is still a difficult number to manage and evaluate individually. Within the lentiviral library each gene is targeted on average by 4-5 different shrnas. Therefore, a second filter of only considering those hits with 2 or more shrnas independently identifying it within the top 5% of hits (excluding any target in the top 5% that is only identified by a single shrna), compressing the dataset down to 390 unique targets. While this had the possible cost of increasing the rate of false negative results, this was outweighed by an increased likelihood of eliminating false positives caused by single shrnas that increased surface MHC I levels in US11+ cells through a non-specific or off-target mechanism (Figure 3.4C and Appendix 1). Several gene targets of note stood out in a cursory examination of these 390 gene targets (Appendix 1). This evaluation generally focused on those shrnas with a Log2 enrichment value 103

119 of 4 or greater, though some particular families of proteins were investigated more extensively. The most enriched shrna was specific for CD160 (Log2 enrichment of 7.47), with a second shrna specific for this target also being present in the top 5% of enriched shrnas (4.25). A number of proteasome subunits were identified (including PSMC1, which had a total of 4 shrnas specific for it in the top 5% of shrnas enriched in the high surface MHC I sample, with Log2 enrichments of 4.50, 3.59, 2.87, and 2.81), MAN1A1 (5.35 and 4.47), a mannosidase involved in processing N-linked glycans, and VCP, the AAA-ATPase involved in retrotranslocation (3.64 and 2.57). The thiol oxidoreductase PDIA5, also known as PDIR, was also identified (4.04 and 2.58). PDIA5 is an ER-localized thiol oxidoreductase containing three catalytic domains. The DnaJ protein DNAJC3 (3.58 and 2.66) was also noted, along with EDEM3, or ER degradation enhancer, mannosidase alpha-like 3 (3.29 and 2.87). Other shrnas of note that were highly enriched were linked to ubiquitin or SUMO processing. These included two shrnas (Log2 enrichment of 6.13 and 3.53) against SAE1, UBE2J2 (Log2 enrichment of 4.99 and 3.76), UBA1 (4.22 and 4.15), and UBA6 (5.08 and 4.91). SAE1 is involved in protein SUMOylation, UBE2J2 is an E2 enzyme previously linked to US11 function [161], and UBA1 and UBA6 (also known as UBE1L2) are E1 enzymes. An important caveat with this type of approach is the arbitrary nature of the cutoffs applied. No consideration of p-values is given when only the top 5% of shrnas highly enriched in the high surface MHC I cell sample are examined. To increase confidence in shrna hits it is first necessary to consider this statistical information. Of course, even if the p-values obtained from Student s t-tests for individual shrnas are considered, significant issues arise in interpreting these p-values due to multiple hypothesis testing and a corresponding increase in the number of false positives. In order to account for the number of statistical tests being performed, 104

120 it is necessary to not consider each shrna result individually but to consider it alongside all the other shrnas within the dataset, leading to the next approaches to be discussed Addressing the issue of multiple hypothesis testing The generally accepted p-value cutoff for Student s t-test is an α = 0.05, which represents a likelihood of observing a particular result by random chance of one in twenty. This rate of false positives becomes problematic when applied to the 78,432 Student s t-tests that would need to be conducted for the lentiviral screen dataset, as one would expect upwards of 3,922 false positives (null results that were incorrectly assigned to the alternate hypothesis of enrichment in the high MHC I cell sample). One of the most basic approaches to deal with this issue is the Bonferroni correction. In this method the desired α (essentially the threshold p-value being applied) is divided by the number of tests. For the lentiviral screen dataset, an α = 0.05 would be divided by 78,432, leading to a Bonferroni-corrected α = When applied to an individual statistical test on individual shrnas this very conservative approach leads to a large number of positive results being discarded as false negatives, although it can be of benefit when applied to group analyses with a smaller number of individual statistical tests. For this reason it was not a suitable method to apply to the lentiviral screen dataset, though it will be used in the next section and elsewhere when evaluating groups of genes Overlap of screen data with other published ERAD datasets Before going through more complex steps to increase confidence in individual shrna hits it was prudent to conduct a validation of the lentiviral screen dataset as a whole. To do this the lentiviral data was compared to other published results, specifically that from a group which recently characterized proteins involved in ERAD through a large scale proteomic study[87]. The lentiviral data here asked what proteins US11 is functionally dependent on, while another 105

121 group using LC-MSMS asked what proteins physically interacted with known components of the ERAD machinery. By comparing our functional screen results to their physical interaction data we could ask if there was significant overlap between the two, as would be expected if our screen was returning hits related to ERAD as well. To determine what would be significant, the amount of overlap that could be expected by random chance was determined by computationally sampling the datasets to create a random overlap group. By repeating this sampling many times in a process known as bootstrapping, a measure of the degree of overlap that could be expected by random chance can be obtained, and the likelihood of seeing the actual observed degree of overlap determined. This bootstrap analysis was conducted without replacement. Explained with a metaphor, each card (gene) drawn from a deck (dataset) was not replaced before the next card for a hand is dealt (each gene could only be selected once in each round of sampling). Reassuringly, this bootstrap analysis of the top 5% of shrna hits, or of hits with >2 shrnas in the top 5% of shrnas showed significant overlap with the high confidence hits of the proteomic study conducted by Christianson et al. (Table 3.2). Table 3.2. Overlap of lentiviral screen data with proteomic results of Christianson et al. [87]. Lentiviral screen data from the top 5% of hits in the high MHC I shrna results or from the top 5% of hits in the high MHC I shrna results with 2 or more independent shrnas were compared to data from Supplementary Table 2A of the paper by Christianson et al [87]. This table of high confidence interacting proteins consisted of 148 unique Entrez Gene IDs, 134 of which were present in the TRC 80K lentiviral library. Overlaps Expected describes the number of intersections between datasets expected by random chance. Random overlaps, or bootstraps, were performed 10,000 times without replacement, and p-values referring to the probability of seeing the number of Overlaps Observed due to random chance were Bonferroni corrected. Calculations were performed with the assistance of Dr. Kevin Brown using the R software platform [246]. Subset Size Overlaps Overlaps Expected p-value Observed (mean +/- S.D.) Top 5% of hits in high MHC I shrna / results Top 5% of hits in high MHC I shrna /

122 results with 2 or more shrnas This overlap analysis provided further confidence that our lentiviral screen was successfully probing for targets functionally involved in US11-mediated ERAD prior to embarking on more advanced statistical approaches Application of false detection rates to identify statistically significant shrna hits In order to properly account for the statistical problems raised by testing 78,432 shrnas it is important to apply some sort of correction, but the very conservative Bonferroni correction is excessively stringent and deals with the issue in a crude manner. Fortunately, the Bonferroni correction is not the only one available and more flexible methods that account for multiple hypothesis testing exist, such as False discovery rates (FDR). FDRs use a converted p-value that attempts to restrict the expected number of false positives in the set of shrnas that appear significantly different to a set value, rather than dealing with each shrna individually, as is done in standard calculations of p-values. Importantly, it also takes into account the distribution of p-values and size of the dataset when determining the q-values used in determining the FDR, as it considers the FDR in the context of the number of genes with q-values equal to or less than a particular cutoff value. As we intended to use our lentiviral screen as a preliminary screen, we considered two different arbitrary FDR cutoffs, 20% and 25%. Details of how q-values were determined and their meaning are described in the Methods section (Chapter 2). In considering the histogram distribution of p-values for all 78,432 shrnas it is clear that there is not a uniform distribution of p-values; instead there appears to be a bias towards values below 0.5 (Figure 3.5A). Figure 3.5B outlines the estimation of π0 used to ultimately calculate q-values. The value π0 reflects the number of truly null test results (shrnas with no significant enrichment in the high surface MHC I cell sample) as a fraction of the total number of tests (shrnas). 107

123 Figure 3.5. Statistical analysis of screen output through q-values and FDR. (A). Histogram of the p-values for all 78,432 shrnas. This distribution is used by the q-value software package for the R programming environment to estimate π0, the estimated proportion of p-values following the null hypothesis (p-values derived from t-tests for which the null hypothesis is true). This value combined with the distribution of p-values is used to calculate q- values for each t-test result (blue). Importantly, q-values relate to the estimated rate of false positives for a group of results; the local FDR refers to the probability of an individual shrna being a false positive (red). (B) Lambda plot used to estimate mathematically the value for π0 applied in (A). (C) Relationship between q-values and p-values for p-values between 0 and 0.2. (D) Number of results that would be considered significant for various q-value cut-offs. (E) Number of false positives that would be expected based on the number of tests considered significant (as determined from q-values and the selected FDR threshold). 108

124 The relationship between p-values and q-values becomes clear when the two are plotted against each other (Figure 3.5C). Q-values increase far more rapidly than p-values, reflecting the increased likelihood of a false positive in our large dataset. The number of tested shrnas that would be considered significant for various q-value cutoffs is shown in Figure 3.5D, and the relationship between the number of tests considered significant and the number of expected false positives is shown in Figure 3.5E. After deriving the q-values for all 78,432, 24 shrnas it was possible to identify 24 hits corresponding to a FDR (q-value) less than 0.2 (Table 3.3), and 766 shrna hits with an FDR less than 0.25 (Appendix 2). Reconsidering this result in the context of the number of false positives expected when various numbers of tested shrnas are considered significant, if we choose a q-value cutoff (FDR) of 0.25, one would expect 192 of the 766 shrnas tested and considered significant to be false positives. In contrast, if we wished to ultimately have a FDR of only 20%, only 24 shrnas would be considered significant, and 5 of these shrna hits would be predicted to be false positives. Table 3.3. All shrna hits with q-values less than 0.2. All shrna hits with q-values less than 0.2 as determined through the R Qvalue package. This represents an expected false positive rate for this list of 24 shrnas of 20%. Gene name Gene description Mean Enrichment (Log 2 ) q-value RPL8 ribosomal protein L TPCN1 two pore segment channel TAF9B TAF9B RNA polymerase II, TATA box binding protein (TBP)-associated factor, 31kDa SEPT12 septin C1orf77 chromosome 1 open reading frame MRPL37 mitochondrial ribosomal protein L ARPC1A actin related protein 2/3 complex, subunit 1A, 41kDa ZNF579 zinc finger protein ELK3 ELK3, ETS-domain protein (SRF accessory protein 2) CCL4 chemokine (C-C motif) ligand SFRS4 splicing factor, arginine/serine-rich

125 ANKRD49 ankyrin repeat domain CTDSPL CTD (carboxy-terminal domain, RNA polymerase II, polypeptide A) small phosphatase-like SSH2 slingshot homolog 2 (Drosophila) SULT1B1 sulfotransferase family, cytosolic, 1B, member PTN pleiotrophin SCGB2A2 secretoglobin, family 2A, member DUSP1 dual specificity phosphatase RERG RAS-like, estrogen-regulated, growth inhibitor LUZP4 leucine zipper protein DEFB1 defensin, beta CARS2 cysteinyl-trna synthetase 2, mitochondrial (putative) GYG1 glycogenin NFATC3 nuclear factor of activated T-cells, cytoplasmic, calcineurin-dependent An important consideration of this type of analysis is that while it allows a statistically robust method of deciding which enriched shrnas are more likely to be true positives, it does not consider the biological plausibility of the results. Some hits appeared to have a plausible link to US11 or immune function, as was the case for NFATC3, a regulator of transcriptional activation in T-cells [253] involved in IL-2 induction. Other hits that were enriched, such as ZNF579 (potential zinc finger transcriptional regulator), SULT1B1 (sulfotransferase that targets hormones [254]), RERG (calcium channel subunit [255]), GYG1 (enzyme involved in glycogen synthesis [256]), and CARS2 (cysteinyl t-rna synthetase), had unclear roles linking them to MHC class I ERAD by US11. Still other results were for shrnas that were depleted in the high surface MHC I cells rather than enriched (those with negative enrichment scores). While we cannot immediately exclude these results from the lentiviral screen dataset, their characteristics and previous studies make a role in US11 function less likely. This statistical approach utilizing false detection rates is more rigorous than simply ranking the shrnas based on their mean enrichment. Unfortunately, the results of this analysis were of questionable value when the biological context of the genes identified was considered. 110

126 Therefore, it was worthwhile to examine other approaches that could more appropriately filter for those hits with more plausible biological significance. By considering the pre-existing biological information available for each shrna target it was hoped that the significant hits with unclear relevance to ERAD may be eliminated, leading to a cleaner list of gene targets more likely worth the effort of validation Functional annotation filters for relevant targets As mentioned, in many cases existing knowledge of the characteristics of a protein can argue against a role in ERAD despite statistical significance. In addition, an shrna target without statistical significance may still be of interest due to its known biological properties. The problem that must be addressed is how to obtain a useful list of genes from the lentiviral data, which is not necessarily the same as the statistically significant results. To this end we examined the frequency of various GO and other annotations in the datasets consisting of targets identified by the top 5% of hits enriched in the high MHC I group or the top 5% of hits enriched in the high MHC I group with 2 or more independent shrnas. The Database for Annotation, Visualization and Integrated Discovery (DAVID) Bioinformatics Resources (version 6.7) website functional annotation tool [235,236] allowed for analysis of statistically significant enrichment of annotations within the hit lists, providing another means of validating the overall quality of the data (Table 3.4 and Table 3.5). In reviewing these results, we noted more terms of interest in the gene targets identified by 2 or more shrnas enriched in the top 5% of hits (Table 3.5). These included annotations for proteasome and proteasome complex members, protein and cellular catabolic processes, and proteolysis, providing further support for enrichment of US11-related factors by the lentiviral screen. 111

127 Table 3.4. GO and other annotations enriched in the top 5% of shrna hits. The database annotations enriched in the top 5% of shrna hits were determined using the DAVID Bioinformatics Resources website. Shown are the top 20 annotations, ranked based on Bonferroni-corrected p-values, with the number of shrna hits annotated with a particular term (Hits with Term) and the fold enrichment in the top 5% of shrna hits versus the total annotation database for the term being considered (Fold Enrichment). P-values are shown with a Bonferroni correction to compensate for the large number of annotations queried. SP_PIR_KEYWORDS refers to the Swiss-Prot (SP) and Protein Information Resource (PIR) Keywords database, SP_PIR_KEYWORDS refers to data from the UniProt database, KEGG_PATHWAY refers to data from the KEGG PATHWAY database, GOTERM_MF_FAT refers to GO terms from the molecular function dataset, GOTERM_BP_FAT refers to the biological process dataset, and GOTERM_CC_FAT refers to the cellular component dataset. Database Term Hits with Term Fold Enrichment p-value, Bonferroni SP_PIR_KEYWORDS phosphoprotein E-11 SP_PIR_KEYWORDS alternative splicing E-09 GOTERM_CC_FAT GO: ~plasma membrane part E-08 UP_SEQ_FEATURE splice variant E-08 GOTERM_CC_FAT GO: ~proteasome complex E-06 GOTERM_CC_FAT GO: ~intrinsic to plasma membrane E-05 KEGG_PATHWAY hsa03050:proteasome E-05 SP_PIR_KEYWORDS proteasome E-05 GOTERM_CC_FAT GO: ~integral to plasma membrane E-05 SP_PIR_KEYWORDS cytoplasm E-05 SP_PIR_KEYWORDS nucleotide-binding E-05 SP_PIR_KEYWORDS atp-binding E-05 UP_SEQ_FEATURE sequence variant E-04 SP_PIR_KEYWORDS polymorphism E-04 UP_SEQ_FEATURE mutagenesis site E-04 GOTERM_MF_FAT GO: ~purine nucleotide binding E-03 GOTERM_MF_FAT GO: ~purine nucleoside binding E-03 GOTERM_MF_FAT GO: ~nucleotide binding E-03 GOTERM_MF_FAT GO: ~nucleoside binding E-02 GOTERM_BP_FAT GO: ~negative regulation of ubiquitin-protein ligase activity during mitotic cell cycle E

128 Table 3.5. GO and other annotations enriched in the top 5% of shrna hits identified with 2 or more shrnas. The database annotations enriched in the top 5% of shrna hits with 2 or more shrnas were determined using the DAVID Bioinformatics Resources website. Shown are the top 20 annotations, ranked based on Bonferroni-corrected p-values, with the number of shrna hits annotated with a particular term (Hits with Term) and the fold enrichment in the list of hits with 2 or more shrnas within the top 5% of shrna hits versus the total annotation database for the term being considered (Fold Enrichment). P-values are shown with a Bonferroni correction to compensate for the large number of annotations queried. SP_PIR_KEYWORDS refers to the Swiss-Prot (SP) and Protein Information Resource (PIR) Keywords database, SP_PIR_KEYWORDS refers to data from the UniProt database, KEGG_PATHWAY refers to data from the KEGG PATHWAY database, GOTERM_MF_FAT refers to GO terms from the molecular function dataset, GOTERM_BP_FAT refers to the biological process dataset, and GOTERM_CC_FAT refers to the cellular component dataset. Database Term Hits with Term Fold Enrichment p-value, Bonferroni GOTERM_MF_FAT GO: ~nucleotide binding E-04 KEGG_PATHWAY hsa03050:proteasome E-03 SP_PIR_KEYWORDS proteasome E-03 GOTERM_MF_FAT GO: ~ribonucleotide binding E-03 GOTERM_MF_FAT GO: ~purine ribonucleotide binding E-03 GOTERM_MF_FAT GO: ~purine nucleotide binding E-02 SP_PIR_KEYWORDS nucleotide-binding E-02 GOTERM_CC_FAT GO: ~proteasome complex E-02 GOTERM_MF_FAT GO: ~purine nucleoside binding E-02 GOTERM_MF_FAT GO: ~nucleoside binding E-02 GOTERM_BP_FAT GO: ~protein catabolic process E-02 GOTERM_MF_FAT GO: ~ATP binding E-02 GOTERM_MF_FAT GO: ~adenyl ribonucleotide binding E-01 GOTERM_MF_FAT GO: ~adenyl nucleotide binding E-01 GOTERM_BP_FAT GO: ~modificationdependent macromolecule E-01 catabolic process GOTERM_BP_FAT GO: ~modificationdependent protein catabolic E-01 process SP_PIR_KEYWORDS atp-binding E-01 GOTERM_BP_FAT GO: ~proteolysis involved in cellular protein catabolic E

129 GOTERM_BP_FAT GOTERM_MF_FAT process GO: ~cellular protein catabolic process GO: ~protein domain specific binding E E-01 The DAVID platform was also used to create sub-lists of targets annotated with particular GO terms or other annotations of interest, narrowing the scope of hits to be considered, although this approach would remove proteins from consideration that lacked relevant annotations. For instance, the GO term corresponding to ER localization (GO: ) revealed 154 unique hits from the top 5% of enriched shrnas, though to only use this GO term would remove from consideration all cytosolic proteins. A large number of GO terms were investigated and a filtered list was developed for the hits annotated with GO: (proteasomal ubiquitin-dependent protein catabolic process) or GO: (protein folding), shown in Table 3.6. This filter produced a number of proteins of potential interest; Derlin-1, Derlin-2, EDEM3, ERLIN2, a large number of proteasome subunits, and VCP were all identified. While the Derlin proteins and EDEM3 are known ERAD players, ERLIN2 has been suggested to have a more specific in the regulation of inositol 1,4,5- triphosphate receptors (IP3Rs) [130]. It has been shown to interact with many other components of the ERAD pathway [87,257], and some evidence has also suggested a potential role in ERAD of other ER proteins [257], although its appearance in our screen dataset is the first time it has been linked to degradation of MHC class I molecules. In addition to these, PDIA5, cyclophilin E, and Hsp90 were noted as potential chaperones involved with US11, though only PDIA5 is localized to the ER. 114

130 Table 3.6. Hits from the top 5% of enriched shrnas annotated as involved in a proteasomal ubiquitin-dependent catabolic process. The list of unique hits identified in the top 5% of shrna were used to query the DAVID Bioinformatics website to identify groups of hits with common GO annotations. Shown are those hits annotated with GO: (proteasomal ubiquitin-dependent protein catabolic process) or GO: (protein folding). Gene Gene Name ID Der1-like domain family, member Der1-like domain family, member ER degradation enhancer, mannosidase alpha-like ER lipid raft associated F-box protein F-box protein MAD2 mitotic arrest deficient-like 1 (yeast) cell division cycle 26 homolog (S. cerevisiae); cell division cycle 26 homolog (S. cerevisiae) pseudogene 5700 proteasome (prosome, macropain) 26S subunit, ATPase, 1; similar to protease (prosome, macropain) 26S subunit, ATPase proteasome (prosome, macropain) 26S subunit, ATPase, proteasome (prosome, macropain) 26S subunit, ATPase, proteasome (prosome, macropain) 26S subunit, ATPase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) activator subunit 2 (PA28 beta) proteasome (prosome, macropain) activator subunit 3 (PA28 gamma; Ki) 5682 proteasome (prosome, macropain) subunit, alpha type, proteasome (prosome, macropain) subunit, alpha type, proteasome (prosome, macropain) subunit, alpha type, proteasome (prosome, macropain) subunit, alpha type, proteasome (prosome, macropain) subunit, alpha type, proteasome (prosome, macropain) subunit, beta type, proteasome (prosome, macropain) subunit, beta type, proteasome (prosome, macropain) subunit, beta type, proteasome (prosome, macropain) subunit, beta type, proteasome (prosome, macropain) subunit, beta type, protein phosphatase 2, regulatory subunit B', gamma isoform 5704 similar to 26S protease regulatory subunit 6B (MIP224) (MB67-interacting protein) 115

131 (TAT-binding protein 7) (TBP-7); proteasome (prosome, macropain) 26S subunit, ATPase, synovial apoptosis inhibitor 1, synoviolin 6907 transducin (beta)-like 1X-linked 7415 valosin-containing protein AHA1, activator of heat shock 90kDa protein ATPase homolog 1 (yeast) heat shock protein 90kDa alpha (cytosolic), class A member 2; heat shock protein 90kDa alpha 3320 (cytosolic), class A member peptidylprolyl isomerase E (cyclophilin E) 5203 prefoldin subunit protein disulfide isomerase family A, member tubulin folding cofactor A Similar filters were also applied to the targets identified with 2 or more shrnas from the top 5% of enriched shrnas. A slightly different subset of GO terms than Table 3.6 were used in this filtered list (Table 3.7), specifically GO terms GO: (proteasomal ubiquitindependent protein catabolic process), GO: (positive regulation of protein ubiquitination), and GO: (proteolysis). From this list we found several hits from Table 3.6, as well as the E3 ligases CBLB and CBLC, the SUMOylation enzyme SAE1, the E3- activator NDFIP2, and the ERAD scaffold protein cullin 2. Table 3.7. GO annotated shrna hits identified with 2 or more shrnas from the top 5% of shrnas identified as enriched in the high MHC class I group. The list of unique hits identified in the top 5% of shrna hits by 2 or more independent shrnas were used to query the DAVID Bioinformatics website to identify groups of hits with common GO annotations. Shown are those hits annotated with GO: (proteasomal ubiquitindependent protein catabolic process), GO: (positive regulation of protein ubiquitination), or GO: (proteolysis) from the biological process ontology. Gene ID Gene Name ER degradation enhancer, mannosidase alpha-like proteasome (prosome, macropain) 26S subunit, ATPase, 1; similar to protease (prosome, macropain) 26S subunit, ATPase proteasome (prosome, macropain) 26S subunit, ATPase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) 26S subunit, non-atpase, proteasome (prosome, macropain) subunit, alpha type, proteasome (prosome, macropain) subunit, beta type, proteasome (prosome, macropain) subunit, beta type, valosin-containing protein 116

132 54602 Nedd4 family interacting protein ADAM metallopeptidase with thrombospondin type 1 motif, Cas-Br-M (murine) ecotropic retroviral transforming sequence b Cas-Br-M (murine) ecotropic retroviral transforming sequence c DDI1, DNA-damage inducible 1, homolog 2 (S. cerevisiae) DET1 and DDB1 associated ADAM metallopeptidase with thrombospondin type 1 motif, F-box and leucine-rich repeat protein F-box protein F-box protein SUMO1 activating enzyme subunit YME1-like 1 (S. cerevisiae) 842 caspase 9, apoptosis-related cysteine peptidase 8453 cullin delta/notch-like EGF repeat containing lysine (K)-specific demethylase 2B 4485 macrophage stimulating 1 (hepatocyte growth factor-like) matrix metallopeptidase microtubule-associated protein 1 light chain 3 beta 8554 protein inhibitor of activated STAT, ring finger protein split hand/foot malformation (ectrodactyly) type transferrin receptor ubiquitin fusion degradation 1 like (yeast) ubiquitin-conjugating enzyme E2, J2 (UBC6 homolog, yeast) 3093 ubiquitin-conjugating enzyme E2K (UBC1 homolog, yeast) 7317 ubiquitin-like modifier activating enzyme ubiquitin-like modifier activating enzyme von Hippel-Lindau tumor suppressor In general, applying GO term annotations to filter large datasets to more manageable levels appears to be a useful approach. One risks discarding potentially interesting hits using these filtering approaches, but the ability to pick out hits of interests based on broad characteristics is promising. At this point three methods of identifying relevant shrnas have been trialed, a simple ranking based on mean enrichment in the top 5% of cells with high MHC I, a rigorous statistical approach that focused solely on false detection rates, and filtering gene subsets from the lentiviral dataset for biological database annotations that suggested relevancy towards ERAD 117

133 processes. Each approach possessed benefits, but also contained flaws that appeared to be restricting its usefulness. In many cases the lists of relevant genes still contained large numbers of genes without any clear link to ERAD, MHC I, viral evasion, or the ER. In addition, although on average 4-5 shrnas are present that are specific for each target, each shrna was usually treated independently of the others. This approach discards a portion of the information available on each gene target within the lentiviral screen. By considering multiple shrnas specific for a single target it is possible to increase the number of data points available to evaluate it, as one would expect to see a correlation in the fold enrichment observed for each of that target s shrnas. Therefore, it was desirable to adopt an approach to the analysis that would deal with the statistical issues introduced by multiple hypothesis testing and could incorporate existing biological information, and would do so using the information provided by all shrnas that are specific for a particular gene target. It was hoped that this combined approach, discussed in Section 3.2.8, would provide superior enrichment for interesting gene targets compared to the more focused approaches attempted thus far Pooling Individual shrna enrichment scores into datasets for single targets In preparing to conduct this combined analysis it was decided to abandon the q-value and FDR-based approach. Although evaluation of large datasets using q-values is a valid statistical approach, there are potential flaws and complications. In addition to the issue of treating each shrna as an independent test, there are signs of potential skewing of p-values that could be affecting the validity of the q-values. Figure 3.5A showed a histogram distribution of p-values with a clear bias towards lower values. This non-normal distribution suggests the presence of some results following the alternative hypothesis that there is enrichment or depletion of an shrna within the 5% of cells expressing high MHC I (versus the null hypothesis of no 118

134 difference). However, there is a decrease in the proportion of p-values in the range of p = compared to p = This type of histogram pattern may be due to violations of the assumptions that must be met for a valid q-value determination [238]. It may suggest that the microarray data may still display a Log-normal skew even after conversion to Log2 values. Additionally, it may indicate that the t-tests used variables that showed some correlation to the other variables. While most shrnas would be expected to be independent of other shrnas in the screen, small groupings of shrnas (those targeting the same transcript) would show some correlation to each other. The phenomena could also be due to correlations present from the microarray chip processing and analysis that were not completely compensated for. These issues do not necessarily invalidate the q-value analysis of the screen data, but they do illustrate some potential complications to the analysis as performed. To account for these issues, a RNAi Gene Enrichment Ranking (RIGER) analysis approach was conducted. This approach considers all of the screen data without arbitrary cutoffs, incorporates information present in the pattern for shrnas specific for a single target, and handles data with a skewed p-value distribution (such as that potentially observed in the lentiviral screen data in Figure 3.5A). A RIGER analysis examines the fold enrichment or depletion of shrnas in the high MHC I cell group and constructs a ranked list of the shrnas, and then determines which shrna targets are enriched near the top or bottom of the list. Looking at the distribution of multiple shrnas removes biases present from individual shrna analyses, such as skewing effects from a single shrna with an unusually large enrichment value. The analysis also normalizes results for the number of shrnas specific for each target. Finally, the likelihood of observing an overrepresentation of shrnas for a particular target at the top of the ranked list by random 119

135 chance can also be estimated, giving a statistical measure of significance to the results. In effect, it condenses the data from the level of shrnas to the level of gene targets, simplifying analysis. The output of this RIGER analysis is an enrichment score for each gene target, normalized for the number of shrnas present in the screen library for each gene target (Appendix 3 and Table 3.8) Table 3.8. Top forty enriched gene targets identified through RIGER analysis. Shown is the top 40 targets most enriched in the high MHC I cell group obtained from RIGER analysis of the ranked list of shrnas from the lentiviral screen. A normalized enrichment score (NES) was used to quantify the relative enrichment of shrnas near the top or bottom of ranked list. Normalization accounts for the differing number of shrnas specific for different gene targets in the overall lentiviral screen dataset. NES represents the raw enrichment score adjusted for differences in the number of shrnas specific for each gene. Gene Name Gene ID Gene Description NES p-value BTG B-cell translocation gene C6orf chromosome 6 open reading frame TRAK trafficking protein, kinesin binding FCGR2B 2213 Fc fragment of IgG, low affinity IIb, receptor (CD32) DECR ,4-dienoyl CoA reductase 1, mitochondrial DEC deleted in esophageal cancer PSMB proteasome (prosome, macropain) subunit, beta type, PSMC proteasome (prosome, macropain) 26S subunit, ATPase, FBLN fibulin VCPIP valosin containing protein (p97)/p47 complex interacting protein ZNF zinc finger protein TRIT trna isopentenyltransferase PPAP2B 8613 phosphatidic acid phosphatase type 2B HNRNPF 3185 heterogeneous nuclear ribonucleoprotein F RBL retinoblastoma-like 1 (p107) TCF transcription factor UROS 7390 uroporphyrinogen III synthase SULT1B sulfotransferase family, cytosolic, 1B, member RNF ring finger protein C4orf chromosome 4 open reading frame PVRL poliovirus receptor-related PTPN protein tyrosine phosphatase, non-receptor type TANK TRAF family member-associated NFKB activator OR52L olfactory receptor, family 52, subfamily L, member

136 CGA 1081 glycoprotein hormones, alpha polypeptide FABP fatty acid binding protein 7, brain MERTK c-mer proto-oncogene tyrosine kinase XIRP xin actin-binding repeat containing NKX NK2 transcription factor related, locus 3 (Drosophila) PF4V platelet factor 4 variant DERL Der1-like domain family, member ALS2CR amyotrophic lateral sclerosis 2 (juvenile) chromosome region, candidate 8 F2RL coagulation factor II (thrombin) receptor-like CPSF cleavage and polyadenylation specific factor 2, 100kDa CD CD160 molecule DCC 1630 deleted in colorectal carcinoma KCNN potassium intermediate/small conductance calcium-activated channel, subfamily N, member 1 LIN lin-37 homolog (C. elegans) SNTB syntrophin, beta 2 (dystrophin-associated protein A1, 59kDa, basic component 2) PRRX paired related homeobox The next stage of this analysis is to incorporate pre-existing biological information into the results. Previously in Section the functional annotation filtering was reliant on first deciding on an arbitrary cutoff to construct the list of shrnas to be examined (i.e. taking the top 5% of shrnas enriched in the high MHC I cells and other subsets based on this group). By combining the RIGER analysis with a Gene Set Enrichment Analysis (GSEA) the power of both approaches can be applied to the complete lentiviral dataset, while avoiding arbitrary cut-offs for shrna consideration. The GSEA examines the enrichment of particular GO terms within the gene target dataset produced by the RIGER analysis. Rather than use an arbitrary cutoff, this approach looks for groups of genes whose expression levels (or in this case level of enrichment in the high MHC I group) are correlated with each other within the entire list of gene targets from the lentiviral screen dataset. In the analysis, an annotation database consisting of annotations drawn from KEGG, Reactome, BIOCARTA, PubMed, and curated gene lists of canonical pathways (c2.all.v5.0.entrez.gmt, or C2) was used as a reference for comparison with our RIGER data. 121

137 Out of 4,208 gene sets from the curated database C2, there were 2,387 sets that were overrepresented within those gene targets enriched, as determined by RIGER analysis, within the top 5% of high MHC I cells. Of these, 93 gene sets were enriched with a p-value of 0.01 or less and 247 gene sets with a p-value of 0.05 or less. To consider a more manageable subset, the 58 gene sets enriched with an FDR q-value less than 0.2 are shown in Table 3.9. Table 3.9. Gene sets identified through GSEA from a collection of curated databases in the RIGER results with a q-value less than 0.2. In addition to the earlier gene set enrichment analysis conducts using GO annotation databases and the RIGER results, a GSEA using a collection of curated databases was also produced. Shown are all enriched gene sets with an FDR q-value less than 0.2. Normalized enrichment score (NES) represents the enrichment score adjusted for gene set size and correlations between gene sets. NAME NES FDR q-value kegg_proteasome E-03 biocarta_proteasome_pathway E-03 smith_tert_targets_up E-03 reactome_destabilization_of_mrna_by_auf1_hnrnp_d E-03 reactome_cdk_mediated_phosphorylation_and_removal_of_cdc E-03 reactome_scf_beta_trcp_mediated_degradation_of_emi E-03 reactome_regulation_of_mrna_stability_by_proteins_that_bind_au_rich_ elements E-03 reactome_vif_mediated_degradation_of_apobec3g E-03 reactome_cross_presentation_of_soluble_exogenous_antigens_endosomes E-03 reactome_autodegradation_of_the_e3_ubiquitin_ligase_cop E-02 reactome_regulation_of_apoptosis E-02 reactome_p53_independent_g1_s_dna_damage_checkpoint E-02 reactome_regulation_of_ornithine_decarboxylase_odc E-02 reactome_signaling_by_wnt E-02 reactome_autodegradation_of_cdh1_by_cdh1_apc_c E-02 dazard_uv_response_cluster_g E-02 sartipy_blunted_by_insulin_resistance_up E-02 reactome_p53_dependent_g1_dna_damage_response E-02 zhong_secretome_of_lung_cancer_and_fibroblast E-02 reactome_ctla4_inhibitory_signaling E-02 reactome_scfskp2_mediated_degradation_of_p27_p E-02 reactome_dna_replication E-02 reactome_downstream_signaling_events_of_b_cell_receptor_bcr E-02 reactome_mitotic_m_m_g1_phases E-02 fridman_immortalization_dn E-02 reactome_cdt1_association_with_the_cdc6_orc_origin_complex E-02 reactome_apc_c_cdc20_mediated_degradation_of_mitotic_proteins E

138 reactome_antigen_processing_ubiquitination_proteasome_degradation E-02 finak_breast_cancer_sdpp_signature E-02 reactome_er_phagosome_pathway E-02 reactome_apc_c_cdh1_mediated_degradation_of_cdc20_and_other_apc_c_ cdh1_targeted_proteins_in_late_mitosis_early_g E-02 reactome_assembly_of_the_pre_replicative_complex E-02 reactome_activation_of_nf_kappab_in_b_cells E-02 reactome_adaptive_immune_system E-02 sengupta_nasopharyngeal_carcinoma_with_lmp1_up E-02 wang_response_to_forskolin_up E-01 reactome_orc1_removal_from_chromatin E-01 pid_cdc42_pathway E-01 reactome_class_i_mhc_mediated_antigen_processing_presentation E-01 pellicciotta_hdac_in_antigen_presentation_up E-01 biocarta_epo_pathway E-01 browne_hcmv_infection_24hr_up E-01 reactome_regulation_of_mitotic_cell_cycle E-01 akl_htlv1_infection_dn E-01 reactome_cyclin_e_associated_events_during_g1_s_transition_ E-01 reactome_m_g1_transition E-01 biocarta_pyk2_pathway E-01 mori_emu_myc_lymphoma_by_onset_time_dn E-01 kegg_other_glycan_degradation E-01 biocarta_fcer1_pathway E-01 amit_egf_response_40_mcf10a E-01 reactome_synthesis_of_dna E-01 reactome_regulatory_rna_pathways E-01 reactome_synthesis_of_very_long_chain_fatty_acyl_coas E-01 bandres_response_to_carmustin_mgmt_48hr_up E-01 wu_silenced_by_methylation_in_bladder_cancer E-01 reactome_glucose_metabolism E-01 reactome_costimulation_by_the_cd28_family E-01 Reassuringly, the top two enriched gene sets referred to the proteasome, a not surprising observation in a screen searching for depletion targets that disrupt ERAD of MHC I. Gene sets specific for MHC I antigen presentation, HCMV infection, and the adaptive immune system were also promising. There does appear to be a significant amount of noise in the results, as shown by the numerous apparently unrelated gene sets observed, but these may be due to other processes impacting on MHC class I expression levels or other more non-specific cellular effects. By reviewing the top gene targets identified by RIGER in Table 3.8, we found some previously mentioned proteins as well as several new factors. CD160 and several proteasome 123

139 subunits were again identified, in addition to the VCP-associating VCPIP1, the ERAD component DERL2 (Derlin 2), and the E3 ligase RNF111, all of which provide support for this methodology Survey of DNAJ proteins Considering their role in substrate selection for the molecular chaperone BiP and other Hsp70s [258], a special search for Hsp40 proteins was made to determine which were prominent within the lentiviral screen data. In addition to DNAJC3, which was mentioned earlier during analysis of the top 5% of shrnas enriched in high MHC I cells (Section 3.2.3), the Hsp40s DNAJA4, DNAJC7, DNAJB6, DNAJC5B, DNAJB11, DNAJB12, DNAJC14, DNAJC30, DNAJB9, and DNAJA1 were also found in the various analyses that have been presented. Of these, DNAJA4 had the highest single shrna Log2 fold enrichment score (4.42), but only had a single shrna present in the top 5% of shrnas enriched in high MHC I cells. Analyzing these J domain proteins for relevant annotations (using the Gene Ontology Consortium enrichment analysis tools [233] and Panther Classification System [234]) highlighted DNAJB9, DNAJA1, DNAJB11, and DNAJC3 as linked to the response to unfolded protein (GO: ), and DNAJB9, DNAJB11, and DNAJC3 as ER lumenal (GO: ). Finally, DNAJB9 (or ERdj4) has also been linked to ERAD of another substrate, surfactant protein C [114] Validation of selected targets with individual shrna knockdown In the previous seven sections a number of rankings, statistical analyses, and biological filters were applied. While analyses could have continued on the bulk dataset, at some point the gene targets identified must be independently validated. A straightforward approach to this was taken by simply conducting depletions on a number of identified targets using the individual shrnas from the TRC 80K lentiviral library instead of the pooled format. A total of 12 targets 124

140 were selected. This was designed to include a positive control consisting of a known ERAD player, a likely false positive, and a small selection of targets from several of the analytical approaches discussed previously (Figure 3.6). None were selected from the shrna hits identified through a q-value-based false detection rate analysis. In addition to validating the presence of these genes in the screen results, it allowed a more thorough evaluation of shrna toxicity and a chance to observe minor effects due to inclusion of less-optimal shrnas, that may not have been below the detection threshold of the lentiviral screen. The positive control selected was Vasolin-containing protein (VCP), as it is a common requirement in most ERAD pathways as a component of the p97-aaa-atpase complex. Of the four VCP-specific shrnas present in the library, most showed significant toxicity, which was not surprising considering its central role in ERAD and other processes (Figure 3.6A). The one remaining shrna increased surface MHC I as expected. LCK has a well characterized role in TCR signaling but was considered an unlikely player in US11 function. In initial data collection LCK appeared as an extremely strong hit in the first replicate, and it remained in the list of hits targeted by 2 or more shrnas in the top 5% of enriched shrnas (with both producing a Log2 enrichment of 2.65). With individual shrna transductions, LCK had only one shrna with sufficient cell viability which had an impact on surface MHC I in US11+ cells (Figure 3.6L), although only for a subset of cells. Despite this LCK is still thought to be artefactual, as LCK transcripts only appear to be present in cells of lymphoid origin [9] and were not expected to be present in U373-MG cells, owing to their astrocyte origin. This type of result cautions that false positive results may escape through this validation process and require further validation using other techniques. 125

141 Figure 3.6. Preliminary validation of selected targets with individual shrna knockdown. A number of individual shrnas corresponding to selected validation targets from the TRC 80K showed a significant increase in surface MHC class I in U373-MG + US11 cells. While a number of shrnas were found to be toxic at the lentiviral titres used, in most cases at least one shrna had a sizable impact on surface MHC I of the remaining living cells, as shown by staining with W6/32 and analyzing by flow cytometry 6 days post-transduction and 5 days postselection with puromycin. For each target, the number of shrnas showing toxicity were (A) 3/4 for VCP, (B) 1/4 for CBLB, (C) 2/4 for CUL2, (D) 0/3 for HSPA5, (E) 1/4 for ERLIN2, (F) 3/5 for CD160, (G) 2/3 for DNAJC3, (H) 1/6 for MAN1A1, (I) 1/4 for PDIA5, (J) 3/5 for DNAJA4, (K) 2/3 for EDEM3. The shrnas showing toxicity were not included in the representative flow plots. This toxicity is likely not representative of all transduction conditions however. (L) Only the one LCK-specific shrna was tested in this validation. 126

142 Other ERAD or ERAD-related proteins tested were CUL2 (Figure 3.6C), EDEM3 (Figure 3.6K), and MAN1A1 (Figure 3.6H). CUL2, or cullin 2, was identified from the functional annotation filters. It is a member of a family of proteins that typically serve as a scaffold in a number of E3 ubiquitin-protein ligase complexes [259]. Interestingly, it has also been reported to interact with the Von Hippel-Lindau disease tumor suppressor protein [260], which was identified with more than 2 shrnas in the top 5% of shrna hits following filtering for GO annotations (Table 3.7). However, little effect was observed when two shrnas specific for CUL2 were individually tested (Figure 3.6C). EDEM3 is an ER lumenal protein with evidence supporting a role in ERAD [261] and was identified in both the initial rankings of shrnas based on mean Log2 fold enrichment scores as well as the functional annotation filtering. MAN1A1 (alpha 1,2-mannosidase IA) was also identified from the initial rankings of screen hits and has been linked to ERAD of glycosylated substrates, with a potential role in retrieving ERAD substrates from the Golgi [262]. Both of these targets were much more promising for a role in US11-mediated MHC I degradation, with 1 EDEM3 shrna (Figure 3.6K) and 5 MAN1A1 shrnas (Figure 3.6H) showing an effect on surface MHC I levels in U373-MG + US11 cells (with some showing varying degrees of a bimodal distribution). Several proteins with potential chaperone roles were also tested, as they could be involved in the initial targeting of MHC I to ERAD. These included PDIA5 (Figure 3.6I), DNAJC3 (Figure 3.6G), and DNAJA4 (Figure 3.6J). PDIA5 was identified during the intial shrna rankings and the functional annotation filtering. It has been shown to interact with calreticulin [86,105] and with BiP [86], which has been implicated with both US2 and US11 function [207,217]. DNAJC3 was identified during the initial shrna rankings and the survey of DNAJ proteins, and has been linked to BiP [86] and is implicated in attenuation of the unfolded protein response [263]. DNAJA4 is cytosolic protein identified during the survey of DNAJ 127

143 proteins, and was suggested to be involved in cholesterol biosynthesis [264]. Interestingly, one shrna for DNAJC3 (Figure 3.6G), two for DNAJA4 (Figure 3.6J), and three for PDIA5 (Figure 3.6I) showed an effect on surface MHC I. All three of these genes were also targeted by shrnas which produced significant toxicity, but cell death does not necessarily support or contradict a potential effect of the shrna on surface MHC I in the context of the screen. The absence of these shrnas in Figure 3.6 does not eliminate these proteins from consideration, but caution should be applied to any early conclusions. The chaperone HSPA5 (also known as grp78 or BiP) was included as another positive control (Figure 3.6D), as it has been previously tied to both US2 and US11 function [207]. It was also logical to include due to the number of J domain proteins in the results of the lentiviral screen. Of three shrnas, one showed a clear increase in surface MHC I, one showed a significant fraction of cells with increased MHC I (as a bimodal population), and a third showed a very minor subset of cells demonstrating an effect (Figure 3.6D). With all three shrnas for HSPA5 showing some effect this chaperone would be strongly implicated in US11 function (as expected based on the previous reports [207]). Finally, we examined CD160 (Figure 3.6F), which was identified in the initial ranking of shrnas and in the RIGER analysis, CBLB (Figure 3.6B), which was identified in the initial ranking of shrnas and the functional annotation filtering, and ERLIN2 (Figure 3.6E) which was identified in the functional annotation filtering. CD160 is a cell surface protein that binds to MHC class I with low affinity [265], plays an important role in NK cell interferon production [266], and inhibits activation of CD4+ T cells through binding to herpes virus entry mediator (HVEM) [267]. Two individual shrnas specific for CD160 increased surface MHC I (Figure 3.6F), though it is unclear by what mechanism this may be driven. CBLB is an E3 ligase linked 128

144 to negative regulation of the T-cell [268] and B-cell receptors [269], which had two shrnas showing a strong effect on MHC I, and one with a bimodal effect (Figure 3.6B). Lastly, ERLIN2 s role in degradation of IP3 receptors [257] was intriguing, as this involved induced degradation of a substrate that did not appear to be terminally misfolded. ERLIN2 (also known as SPFH2) had 3 separate shrnas showing an increase in surface MHC I in U373-MG + US11 cells (Figure 3.6E). Overall, the majority of these validation targets showed evidence of involvement with US11 function, other than CUL2 (cullin 2), which showed little evidence for an effect on surface MHC I. However, it is important to note that only a single shrna showed a positive effect on surface MHC I in the US11-expressing U373-MG cells for VCP, LCK, DNAJC3, EDEM3, and DNAJA4. For VCP, EDEM3, and DNAJC3 this was due to cell toxicity with some specific shrnas, which may indicate an important role in ERAD or protein homeostasis or could be due to some other specific or non-specific effect of the shrnas. The effect of LCK depletion was also suspect due to its pattern of tissue expression. Although all of these validation targets could be examined further to validate their role in ERAD, only a selected group was selected for further study at this stage PDIA5 PDIA5 was found to be present in complexes with calreticulin, ERp72 [86,105], and BiP [86], and has also been found in complex with ERp57, Grp94, and P5 through proteomic approaches [86]. It was intriguing that this particular thiol oxidoreductase was identified in our results, as it had seemed more likely that others would be found, such as ERdj5 (DNAJC10) which has been linked to ERAD of surfactant protein C [114] and alpha-1-antitrypsin (NHK variant) [270], as well as binding to EDEM proteins and BiP [270]. 129

145 Four individual shrnas from the lentiviral screen were used to deplete PDIA5 in U373- MG + US11 cells (Figure 3.7A). Three of these shrnas (TRCN , TRCN , and TRCN ) showed some depletion of PDIA5, as determined by western blot, with TRCN only producing a modest decrease in PDIA5 (Figure 3.7B). However, all four PDIA5-specific shrnas increased MHC I levels (Figure 3.7C). The fourth shrna, despite increasing total MHC I, produced poor PDIA5 depletion and also increased levels of BiP, suggesting it was inducing an unfolded protein response through an offtarget effect. Additionally, the shrna TRCN increased MHC I to a greater degree than TRCN and TRCN , despite these two shrnas resulting in greater PDIA5 depletion (Figure 3.7C). Another graduate student in the laboratory of Dr. David Williams (University of Toronto), Seo Jung Hong, also depleted PDIA5 and assessed surface levels of MHC I by flow cytometry and total MHC I levels by western blot in U373-MG + US11 cells. She also observed an increase in surface MHC I following PDIA5 depletion [271]. However, this increase was extremely modest, and the degree of PDIA5 depletion observed for each shrna did not directly correlate with the magnitude of MHC class I increase observed. Overall, that three shrnas were able to produce an effect would argue against a simple off-target effect to explain the increase in MHC I. Rather, an effect of PDIA5 depletion on US11-mediated degradation may concurrently influence the cell in some other way (potentially through induction of an ER stress response) and produce a more variable phenotype. 130

146 Figure 3.7. Knockdown of PDIA5 disrupts US11-mediated ERAD of MHC class I. (A) Depletion of PDIA5 increases MHC class I in US11+ cells. Lentiviral vectors containing four different shrna constructs specific for PDIA5 (TRCN , TRCN , TRCN , and TRCN ) were stably transduced into U373-MG + US11 cells and selected with puromycin for stable incorporation of the constructs. Levels of PDIA5, MHC class I, BiP, and GAPDH were assessed by western blot. An unidentified band appearing in the anti-pdia5 immunoblot above PDIA5 is denoted by *. (B) Quantification of PDIA5 depletion shown in (A), normalized to GAPDH. (C) Quantification of fold increase in MHC I shown in (A), normalized to GAPDH. 131

147 ERLIN2 The possibility that US11 may manipulate ERLIN2 to induce MHC I degradation in the same manner that ERLIN2 degrades IP3 receptors was intriguing. IP3 receptors are involved in the release of calcium from the ER downstream of signaling events that result in IP3 cleavage and release from the membrane. The receptors are large proteins approximately 2,700 amino acids in length and are integrated into the ER membrane through 6 transmembrane domains that form a number of different tetrameric structures. Activation is associated with a change in conformation and channel opening [130]. Down-regulation of IP3 receptors follows soon after their activation [272] leading to a down-regulation of signaling. This down-regulation is carried out by ERLIN1 and ERLIN2 (also referred to as SPFH1 and SPFH2), which target the activated IP3 receptors for degradation through ERAD [130,257]. Lastly, ERLIN1 and ERLIN2 have been linked to other components of the ERAD pathway and to other ERAD substrates [87,257]. If MHC I was recruited to ERLIN1 and ERLIN2 by US11 it would provide a convenient mechanism for how US11 is able to rapidly induce degradation of MHC I. To examine the role of ERLIN2, ERLIN2 was depleted in U373-MG control, U373-MG + US2 and U373-MG + US11 cells by Franziska Teusel, a visiting student under my supervision. Depletion of ERLIN2 with the ERLIN2-specific shrna TRCN (Figure 3.8A) produced a strong up regulation of surface MHC I in US11+ cells (Figure 3.8B). In contrast, ERLIN2 depletion resulted in a slight decrease in surface MHC I in the US2+ and U373-MG control cells, suggesting that the increase in MHC I was specific to US11-mediated degradation and not a more general mechanism. Seo Jung Hong also re-investigated the role of ERLIN2. 132

148 Figure 3.8. ERLIN2 shrna depletion indicates involvement in US11-mediated MHC class I ERAD. (A) Depletion of ERLIN2 in U373-MG cells + US11 with lentivirus carrying the ERLIN2- specific shrna TRCN results in loss of ERLIN2 protein. U373-MG + US11 cells were transduced with lentivirus carrying a luciferase shrna or an individual ERLIN2 shrna and protein levels examined by western blot 8 days post-transduction. (B) Depletion of ERLIN2 leads to an increase in surface MHC I in U373-MG + US11 cells. U373-MG control, U373-MG + US2, and U373-MG + US11 cells transduced with lentivirus carrying luciferase or ERLIN2 shrnas were stained with the mab W6/32 and examined by flow cytometry. An increase in surface MHC I was observed in U373-MG + US11 cells but not U373-MG control cells, and a modest drop in MHC I was observed in U373-MG + US2 cells following ERLIN2 knockdown. Unpublished data courtesy of Franziska Teusel. 133

149 She confirmed that two of the four ERLIN2 shrnas contributed to increased surface MHC I levels, above that observed for PDIA5 [271]. Unfortunately, while increases were also observed by western blot there was again a lack of correlation between degree of ERLIN2 depletion and the magnitude of effect on total MHC I [271], with one of the shrnas showing little increase in total MHC I but producing excellent depletion of ERLIN2. As with PDIA5, there is enough evidence of a role for ERLIN2 to warrant further studies. However, it would be critical to ensure that the effect of ERLIN2 depletion on MHC I levels is validated further to account for the contradictory effects observed SUMOylation and US11 While not included in the earlier validation experiments, GO annotation filtering of the lentiviral screen results had identified SAE1 as a potential hit from the initial ranking of screen results and the functional annotation filtering. SAE1 (SUMO activating enzyme subunit 1) plays an important role in protein SUMOylation as a component of the E1 activating enzyme complex. SUMOylation is the post-translational modification of proteins with a ubiquitin-like modifier often involved in regulating processes such as transcription and DNA repair. Similar to ubiquitin, SUMO is added to substrates through an enzymatic pathway involving an activating enzyme (E1), conjugating enzyme (E2), and a protein ligase (E3) [273]. Based on the presence of SAE1, Seo Jung Hong conducted a search for other SUMO-related components and identified PIAS1, an E3 ligase for SUMO [273], within the initial rankings of shrnas. Finally, if SUMOylation were playing a role in US11-mediated degradation of MHC I, it would be expected that the SUMO-specific E2 enzyme UBE2I would also be involved as it is also required for SUMO addition to substrates. 134

150 Seo Jung Hong proceeded to examine the effect of depletion of these components on MHC I in US11+ cells. Individual depletion of these hits and analysis of surface MHC I level by flow cytometry showed little effect for PIAS1, but a 2.8 fold increase in MHC I with SAE1 depletion (Figure 3.9A) and 2.6 fold increase with UBE2I depletion [271]. The absence of effect for PIAS1 depletion could be explained due to the presence of several alternative SUMO E3s that may be required instead of PIAS1. Similar effects of SAE1 and UBE2I depletion were observed by western blot for total MHC I, with depletion of SAE1 or UBE2I leading to increased MHC I levels in U373-MG + US11 cells (Figure 3.9, Panels B and C), though not all shrnas showed the same magnitude of effect. Follow-up western blots examining US11 expression levels in these depletion experiments indicated a potential mechanism, as UBE2I depletion led to a dramatic drop in US11 protein levels and corresponding increase in total MHC I (Figure 3.9C). This suggests that SUMOylation involving SAE1 and UBE2I is potentially involved in regulating US11 protein levels in cells, and that disruption of SUMOylation leads to loss of US11 protein and subsequent rescue of MHC I from ERAD NACA In addition to these SUMOylation components, Seo Jung Hong studied a number of other hits from the screen, attempting to validate their phenotypes by depleting them in U373-MG + US11 cells and U373-MG control cells [271]. She identified a number of targets that increased surface MHC class I following knockdown with specific shrna. LMO7 (5.4 fold increase), NDFIP2 (2.5 fold increase), and NACA (2.8 fold increase) all appeared to be influencing US11 function (Figure 3.9A). However, complications arose upon further analysis, with LMO7 expression (and depletion) in U373-MG cells unable to be confirmed by either western blot or real-time PCR, and NDFIP2 affecting surface MHC I levels in U373-MG control cells as well as U373-MG + US11 cells [271]. 135

151 136

152 Figure 3.9. Depletion of several other host cell factors increases surface MHC class I in U373-MG + US11 cells. (A) Selected shrnas were chosen for a number of targets identified from the lentiviral screen data by Seo Jung Hong [271]. U373-MG + US11 cells were transduced with lentiviral supernatants carrying each shrna and surface MHC I levels characterized through staining with the mab W6/32 and flow cytometry. Median fluorescence values from each individual replicate were taken and averaged and compared to control U373-MG + US11 cells treated with lentivirus carrying a luciferase-specific shrna, using paired Student s t-tests, * indicates p-value < Cells transduced with lentivirus carrying a GIPZ depletion vector with a non-specific shrna were included as a negative control. Where multiple shrnas were available, the shrna showing the greatest effect on MHC I levels is shown (TRCN for Cbl-B, TRCN for Cbl-C, TRCN for SAE1, TRCN for UBE2I, TRCN for PIAS1, TRCN for DNAJC10, TRCN for DNAJA4, TRCN for RNF167, TRCN for LMO7, TRCN for NACA, TRCN for STX5, and TRCN for NDFIP2. SEL1L shrna was used as a positive control. (B) Depletion of SAE1 and UBE2I increases total MHC I levels in U373-MG + US11 cells. Western blots of MHC I levels following depletion of PIAS1, UBE2I, Sae1, or Cbl-B through transduction with lentiviral shrna vectors into U373-MG + US11 cells or U373-MG control cells. GAPDH was blotted as a loading control. (C) Treatment of U373-MG + US11 cells with lentivirus containing shrnas specific for UBE2I results in an increase in total MHC class I by western blot and a corresponding decrease in US11 protein. U373-MG + US11 cells transduced with lentivirus carrying UBE2I-specific shrnas and lysed cells following depletion of the target protein. Loss of UBE2I appeared to correlate with a corresponding loss of US11. Data courtesy of Seo Jung Hong [271]. Interestingly, NACA did appear to be specific to US11+ cells [271], though a small increase in MHC I was observed for control cells. As NACA prevents entry of non-signal peptide-bearing proteins into the ER [274], it has been proposed that relaxation of this regulation could increase MHC I translocation into the ER, potentially allowing an increased amount of MHC I to escape US11 degradation [271] J domain-containing proteins DNAJA4 and DNAJC3 Potential involvement of J domain proteins were of particular interest, due to the previously studied role of BiP in US11 degradation of MHC I [207]. DnaJ/Hsp40 family members play an important role in protein quality control. They act to regulate Hsp70 chaperone activity by binding specific proteins for delivery to the Hsp70, and they stimulate the ATP hydrolysis activity of Hsp70s to allow the chaperone to cycle on and off these substrates [258]. A 137

153 number of DnaJ s were observed in the lentiviral screen results, including the ER-localized DNAJB9, DNAJB11, and DNAJC3. From these Hsp40s two were selected for further study. DNAJC3 is an ER localized J domain protein. It has been linked to the unfolded protein response within the ER [263], plays a role in decreasing activity of eif2α [263,275], and regulates the RNA-dependent protein kinase (PKR) [275]. This made DNAJC3 appear to be a promising candidate for involvement in US11- mediated degradation of MHC I considering the role of PKR in blocking viral replication and the link between the unfolded protein response to ERAD and to US11 function [276]. DNAJA4 was another Hsp40 present in the gene targets validated in Section While not fully characterized, it does appear to play a role in the cholesterol biosynthesis pathway [264]. When surface MHC I levels were measured by Seo Jung Hong [271], a significant increase in surface MHC I was observed following depletion of DNAJC3 (3.3 fold increase) or DNAJA4 (6.3 fold increase) (Figure 3.9A). The effect of DNAJA4 depletion was observed primarily with a single shrna, and a similar effect could also be observed by western blot (Figure 3.10A). In addition, Seo Jung Hung observed that the effect of DNAJA4 depletion on MHC I levels appeared to be specific to U373- MG + US11 cells and not U373-MG control cells (Figure 3.10B). Unfortunately, upon attempting to confirm shrna-mediated depletion of DNAJA4, no drop in transcript or protein levels was observed which led Seo Jung Hong to conclude that an off-target effect was likely the cause of the increase in MHC I (Figure 3.10C). Combined with the fact that only a single shrna supported its role, DNAJA4 s presence in the lentiviral screen results appears to be artefactual. 138

154 Figure DNAJC3 and not DNAJA4 depletion increases MHC I levels in US11+ cells. (A) Depletion of DNAJA4 in U373-MG + US11 cells increased total MHC I levels. Following treatment with the DNAJA4-specific shrna TRCN , MHC I levels in cell lysates were measured by western blot with the mab HC10. (B) Depletion of DNAJA4 in U373-MG control and U373-MG + US2 cells had no effect on MHC I. (C) No decrease in DNAJA4 transcript levels were detected by semi-quantitative PCR. DNAJA4 and β-actin transcripts were amplified from serial dilutions of cdna synthesized from RNA by PCR. RNA was obtained from U373-MG + US11 cells expressing either a luciferase-specific shrna or a DNAJA4- specific shrna. (D) The specific shrnas TRCN (DNAJC3-i) and TRCN (DNAJC3-ii) deplete DNAJC3 protein levels, as determined by western blot. (E) Depletion of DNAJC3 with the sirna DNAJC3-i increases MHC I levels as measured by western blot. GAPDH was blotted as a loading control. Data courtesy of Seo Jung Hong [271]. 139

155 DNAJC3 depletion also showed an impact on surface MHC I levels in U373-MG + US11 cells (Figure 3.9A) as well as modest effects on total MHC I by western blot (Figure 3.10E). Importantly, depletion of DNAJC3 could be confirmed by western blot (Figure 3.10D), unlike DNAJA4. However, as depletion with only one of the shrnas tested resulted in increased MHC I these results must be treated with caution until additional depletion tests are performed. DNAJC3 s role in the unfolded protein response [263] provides a convenient link to ERAD that may be exploited by US11 to degrade MHC I, and when combined with the available data concerning the effect of its depletion on US11 function makes a case for further study. 3.3 Discussion Previous studies examining the degradation of MHC I by US11 have taken a relatively directed approach to identify functionally important host cell factors required by this viral immunoevasin molecule. While these have improved our understanding of some of the later stages of MHC I degradation by these viral proteins, there are still uncertainties in how this immunoevasion (as well as the related protein US2) actually triggers the degradation of MHC I. This is compounded by some studies indicating that US2 and/or US11 bind MHC I in a number of different folding states, and that binding alone is not sufficient for degradation [192,277,278]. We hypothesized that US11-mediated degradation of MHC I depended on more than crosslinking of MHC I to the E3 ligase TMEM129 [161]. Rather than attempting to make an educated guess as to the identity of these other players, we carried out a broad genome-wide lentiviral depletion screen to find novel proteins required by US11. The depletion screen was carried out using the TRC 80K lentiviral shrna library. This pool of 78,432 shrna constructs targeting almost 16,000 gene targets was used to deplete protein expression levels in U373-MG + US11 cells. Following shrna depletion of gene 140

156 expression, the cells were sorted by FACS into two groups: the top 5% of cells expressing high MHC I, and the remaining 95% of cells. Finally, the shrnas incorporated into the genomic DNA of these cell samples was isolated by PCR and the relative abundances measured by microarray. The initial data from this screen, a measure of the level of shrna enrichment following lentiviral-shrna transduction in a high MHC I cell group versus a control cell group, were characterized in a number of ways. These included relatively arbitrary cutoffs (targets identified with 2 or more independent shrnas in the top 5% of enriched shrnas), statistical methods to account for multiple hypothesis testing problems (q-values and RIGER analysis), and filtering for groups of hits tied to relevant biological processes or characteristics (GO annotation filtering/enrichment analysis and GSEA). The full lentiviral dataset is too overwhelming to try and analyze as a whole. With 78,432 measurements completed in triplicate there is simply too much data to be able to pick out what is relevant from the noise. The initial approach taken to address this was an arbitrary cut-off. Following calculation of the mean fold enrichment of each shrna within the 5% of cells expressing high MHC I the shrnas were ranked from highest enrichment to lowest. By only considering the top 5% of enriched shrnas, the 78,432 data points was reduced to 3,922. By restricting this list to only those gene targets identified by 2 or more shrnas within it, the number of potential genes of interested was reduced from the thousands to the hundreds. The primary advantage of this approach is the ease with which it may be carried out. Unfortunately, its simplicity is its main weakness. The use of a simple cutoff does not take into account any statistical testing, discards the great majority of the lentiviral screen data available, and makes no consideration of the biological relevance of the shrnas identified. Despite these flaws, with knowledge of ERAD and ER quality control it was still possible to review this list of 141

157 shrnas and identify proteins with interesting links to US11, ERAD and protein folding, a number of which (EDEM3, MAN1A1, PDIA5, DNAJC3, CD160, and CBLB) were tested in validation depletion experiments. What these proteins have in common is that they were all already linked in some way to protein folding or degradation. In a somewhat obvious conclusion, the proteins we tend to see as potentially relevant are those we already know to be potentially relevant. The first improvement we sought to make was to incorporate a statistical measure of confidence in the shrna fold enrichment values. Calculating q-values and using them to estimate false detection rates allowed for the earlier arbitrary cutoffs to be replaced with ones that better reflected the nature of the lentiviral screen dataset. This approach also dealt with the multiple testing issues present in this type of screen. However, upon review of shrnas with a 20% rate of false positives there were few that appeared interesting enough to pursue further. Many of the shrnas were specific for gene targets localized to the nucleus or mitochondria, or had clear functions in cellular metabolism unrelated to ERAD. Our conclusion from this was that statistical relevance did not correlate with biological relevance, and the highest confidence hits were not necessarily those that should be selected for further study. Whether this conclusion would have been made if the lentiviral screen had been expanded with additional replicates and larger cell populations to increase its sensitivity and power is unknown, but it seems to hold for the experiment as conducted here. The q-value analysis demonstrated that there was a significant amount of noise present in our results. This led to consideration of the lentiviral screen data as more of an enrichment for potential players, rather than a method of exclusively identifying them. Adopting a filtering approach, the genes targeted by the top 5% of shrnas enriched in the high MHC I cells were 142

158 submitted to databases containing a range of biological annotations. By searching this list of genes for particular GO or other annotations, smaller lists made up of subsets of genes could be produced and examined. While the annotations selected are not unbiased, the short length of the genes present with each annotation led to the identification of additional targets of interest, such as CUL2, ERLIN2, and SAE1. Unfortunately, the same lack of statistical rigor that plagued the arbitrary cutoffs of the intial shrna rankings also applied here. While a measure of statistical significance could be applied to enrichment of a particular GO annotation overall, it was not available for the individual proteins within the list. Additionally, the biological annotation searches still discarded most of the lentiviral screen data by focusing on the top 5% of shrnas enriched in high MHC I cells. The final approaches were the RIGER analysis and GSEA. RIGER solved several problems of the previous techniques in that it accounted for the multiple hypothesis testing and did not discard any screen data. It appeared to produce valid results based on the enrichment of relevant gene annotations as determined by the GSEA. Finally, it did include CD160, one of the gene targets selected for further validation, as well as other players in ERAD. However, it still suffered from an apparently high background of confusing or irrelevant gene targets that had to be manually filtered in interpreting the results. In conclusion, while RIGER and GSEA are likely the most rigorous and statistically valid approach to interpreting large data sets that was applied in this study, a simple arbitrary cutoff combined with close hands on review by experimenters often proved equally useful. In many cases the most interesting hits were still chosen based on the background knowledge of the experimenter on molecular processes and proteins likely to be related to US11 function. 143

159 Despite these difficulties, we were able to identify several proteins and cellular pathways of interest from the lentiviral screen dataset, and a number of these survived additional validation tests. In particular, the translocation regulator NACA is a potential candidate for further research. ERLIN2, DNAJC3, and PDIA5 all possess functional characteristics that may be exploited by US11, though there would need to be further investigations to confirm the validity of the shrna depletions performed for them. Finally, SAE1 and UBE2I provide a means of studying an apparent role for SUMOylation in regulating US11 levels within transduced cells. At this point these six proteins appear to be the most promising prospects for future study of US11 function, and a better understanding of their role will improve our understanding of MHC I degradation as a means of viral immune evasion as well as the molecular processes involved in ERAD. Routes these future studies could take to examine the role of these proteins will be described in more detail in Section

160 Chapter 4 Cyclophilin C assists in US2-mediated degradation of major histocompatibility complex class I molecules by promoting interactions with the ER-associated degradation machinery Work in this Chapter has been previously published in the following manuscript: Daniel C. Chapman, Pawel Stocki, and David B. Williams. Cyclophilin C Participates in the US2-Mediated Degradation of Major Histocompatibility Complex Class I Molecules. PLoS One. 10 e (2015). 145

161 4 Chapter Introduction In Chapter 3, a broad, unbiased lentiviral screen was applied to identify novel factors involved in US11-mediated degradation of MHC I. From this a number of potential candidate proteins were identified and validated. It is noteworthy that the hits identified as worthy of further study included the thiol oxidoreductase PDIA5 and the J domain protein DNAJC3, among others. PDIA5 (also known as PDIR) is a ER-localized protein disulphide isomerase that interacts with other ER chaperones, such as calreticulin and BiP, and the thiol oxidoreductases ERp72 and ERp57 [86,105]. DNAJC3 is an ER-localized Hsp40 that has been shown to regulate the unfolded protein response [263]. Neither of these had been previously linked to ERAD and their presence suggested that a broad range of proteins may act in concert with US2 and US11 to target MHC I. In addition to the lentiviral screen, a directed approach to uncover components involved in ERAD was also carried out. A number of molecular chaperones and thiol oxidoreductases were depleted by sirna and their impact on US2- and US11-mediated degradation of MHC assessed. Calnexin and calreticulin were the first examined, as previous work demonstrated they increase in association with MHC I in the presence of US2 [217], although a functional link was not shown. The thiol oxidoreductases PDIA1, ERp72, and DNAJC10 (ERdj5) were also depleted by sirna. Proteins targeted towards the ERAD pathway can have their disulphide bonds reduced to allow efficient retrotranslocation, as has been shown for DNAJC10 in the degradation of the NHK variant of α1-antitrypsin [270]. Finally, the peptidyl proline isomerases cyclophilin B (CypB) and cyclophilin C (CypC) were depleted in US2+ and US11+ cells. CypB and CypC 146

162 have been shown to play a role in ER hyperoxidation [118]. A role was also recently demonstrated for CypB in the degradation of soluble ERAD substrates [119]. This study showed that the catalytic activity of CypB was required, suggesting that it may isomerize cis-prolines to remove bends in the substrate before retrotranslocation. In light of this, we chose to investigate the potential role of ER-localized cyclophilins in MHC I ERAD induced by US2 and US11. We found that CypC, but not CypB, participates in US2-mediated degradation of MHC I and that neither cyclophilin was involved in degradation by US11. Surprisingly, CypC does not require its catalytic activity for its role in MHC class I degradation by US2. We hypothesized that CypC may promote complex formation between MHC class I, US2, and other ERAD components [86]. In investigating this model, we identified several ER proteins including malectin, PDIA6, and TMEM33 whose associations with US2 were altered upon CypC depletion or overexpression, and whose presence enhanced US2-mediated degradation of MHC I. These findings indicate that US2 co-opts a diverse array of host proteins to accomplish the efficient degradation of MHC I and subsequent evasion of immune surveillance by cytotoxic T cells. 4.2 Results Depletion of Cnx, Crt, and DNAJC10 in US2+ and US11+ cells Cnx and Crt were the first depletion targets to be investigated, as a study linking them to US2 [217] that first led to the pursuit of the research described in this manuscript. The first step was to validate that the sirnas available did in fact deplete these proteins efficiently, which was the case for both calnexin ((Figure 4.1B) and calreticulin (Figure 4.1C). DNAJC10 (ERdj5) could also be specifically depleted with two different sirnas (Figure 4.1A). 147

163 Figure 4.1. Depletion of DNAJC10, calnexin, and calreticulin using specific sirnas. U373-MG control, US2+, or US11+ cells were treated with specific sirnas on day 0 and day 3, and were replated on day 5. Western blot was used to evaluate protein depletion on day 6. (A) DNAJC10 depletion with sirna DNAJC10 (i) or DNAJC10 (ii). (B) Calnexin depletion. (C) Calreticulin depletion. 148

164 Although PDIA1 and ERp72 were initially considered for depletion, they were not pursued beyond initial sirna testing as PDIA1 depletion had been shown to impact on US2 function and ERp72 depletion did not impact MHC I levels in US2+ or US11+ cells [214]. Following sirna depletion with a Crt-specific sirna, U373-MG, U373-MG + US2, and U373-MG + US11 cells were examined by flow cytometry for surface MHC I or by metabolic labeling. For both calnexin (Figure 4.2B) and DNAJC10 (Figure 4.2C) metabolic labeling for 5 min with a chase of up to 20 min demonstrated little effect of depletion on MHC I degradation rates in US2+ or US11+ cells. For Crt, surface MHC I levels characterized by flow cytometry actually decreased following sirna-mediated depletion (Figure 4.2A). Overall, these depletions demonstrated relatively minor effects on surface or total MHC I in US2- or US11-expressing cells CypC, but not CypB, plays a role in US2-mediated degradation of MHC class I Since previous work had demonstrated a role for CypB in ERAD [119], we were interested in determining if either of the ER-localized cyclophilins (CypB or CypC) was involved in MHC I ERAD mediated by US2 or US11. Human U373-MG cells stably expressing either US2 or US11 were used to examine the role of ER cyclophilins by RNA interference. As shown in Figure 4.3A, expression of either US2 or US11 resulted in the expected reduction of MHC class I at the cell surface. Depletion of CypC in US2-expressing cells increased surface levels of MHC I (Figure 4.3B). However, depletion of CypC in US11-expressing cells or in U373-MG control cells did not affect surface MHC I (Figure 4.3C). This effect was specific to CypC, since depletion of CypB had little effect on MHC I expression in any of these cell lines (Figure 4.3C). 149

165 Figure 4.2. Depletion of Crt, Cnx, or DNAJC10 does not disrupt US2- or US11-mediated degradation of MHC I. (A) U373-MG control, U373-MG + US2, or U373-MG + US11 cells were treated with an sirna specific for Crt on day 0 and day 3 and replated on day 5. On day 6 cell surface MHC I levels were measured by flow cytometry with the mab W6/32. (B and C) U373-MG + US2, or U373-MG + US11 cells were treated with sirnas specific for Cnx or DNAJC10 (DNAJC10-ii) on day 0 and day 3 and replated on day 5. On day 6 samples were radiolabeled for 5 min with [ 35 S] Met. Cells were lysed following labeling or after chasing in Met-containing medium for the indicated times. MHC I was immunoisolated using mab HC10 or mab HCA2. 150

166 151

167 Figure 4.3. Depletion of CypC but not CypB increases MHC class I surface expression in US2-expressing cells. (A) U373-MG control cells or cells expressing US2 or US11 were stained with mab W6/32 to detect surface MHC I molecules followed by goat anti-mouse Alexa647 secondary Ab and analyzed by flow cytometry. (B and C) U373-MG control, US2+, or US11+ cells were treated with CypB sirna, CypC-(i) sirna, or a cocktail of CypC-(ii) and CypB sirnas mixed in a 3:1 ratio. Following treatments on day 0 and day 3, cells were replated on day 5 and analyzed by flow cytometry on day 6 (mab W6/32). Data from panel B is included in and expanded on in panel C. (D) Average fold increase in surface MHC I (mab W6/32) upon depletion of Cyp B (n = 8), Cyp C (n = 15), and CypB + CypC-(ii) (n = 8) in U373-MG + US2 cells. Student s T-test was used to assess statistically significant differences from the Neg sirna-treated cells, and between the CypC-(i) and CypB + CypC-(ii)-treated cells (* indicates p-value < 0.05; n.s., not significant). MFI = median fluorescence index. We also performed a combined knockdown using a mixture of CypB sirna and an alternative CypC sirna (that also targets CypB). Although there appeared to be a modest increase in surface MHC I compared to that obtained by depleting CypC alone (Figure 4.3C), this was not statistically significant in replicate experiments (Figure 4.3D). Similar results were obtained when total MHC I levels were analyzed by immunoblotting following efficient depletion of CypB, CypC or a combination of CypB and CypC (Figure 4.4A). An increase in MHC I was observed upon CypC depletion in US2+ cells but not in control or US11+ cell lines. (Figure 4.4B). Furthermore, the effect of combined depletion of CypB and CypC was not significantly different from CypC depletion alone (Figure 4.4B and quantified in Figure 4.4C). The specificity to US2 suggested that CypC was involved in US2-driven MHC I ERAD rather than participating in a more general pathway of MHC I homeostasis. In the latter case, CypC depletion would be expected to affect MHC I levels in U373-MG control or US11+ cells. Furthermore, this stabilization of MHC class I by CypC depletion was not due to loss of US2 protein expression since US2 protein levels remained unchanged after knockdown of CypC (Figure 4.5A). 152

168 Figure 4.4. Depletion of CypC but not CypB increases total MHC class I in US2-expressing cells. (A) U373-MG control, US2+, or US11+ cells were treated with CypB sirna, CypC-(i) sirna, or a cocktail of CypB and CypC-(ii) sirna. Cyclophilin depletion was evaluated on day 6 using anti-cypc antiserum which also detects CypA and CypB. Due to differing glycosylation states, CypC was typically observed as three distinct bands [118]. GAPDH was used as a loading control. (B) For the conditions described in panel A, MHC class I levels were assessed by immunoblot using mab HC10; a representative background band was included as a loading control. (C) Average fold increase in MHC I in U373-MG + US2 cells upon depletion of Cyp B (n = 4), Cyp C (n = 6), and CypB + CypC-(ii) (n = 5). Student s T-test was used to assess statistically significant differences from the Neg sirna-treated cells, and between the CypC- (i) and CypB + CypC-(ii)-treated cells (* indicates p-value < 0.05; n.s., not significant). 153

169 Considering the potential for CypC to be involved in protein folding, we tested whether depletion of CypC was disrupting normal maturation of proteins within the ER. As an accumulation of misfolded proteins would be expected to induce an unfolded protein response, we used a PCR-based assay to monitor stress-induced mrna splicing of the Xbp-1 transcription factor. No increase in spliced Xbp-1 mrna was observed in our experiments (Figure 4.5B) which is consistent with our previous studies in human hepatoma cells [118]. In contrast, treatment of cells with cyclosporin A, an inhibitor of the catalytic activity of all cellular cyclophilins, did induce an unfolded protein response (Figure 4.5B) and hence cyclosporin A was not used in subsequent experiments CypC depletion impedes the US2-mediated degradation of newly synthesized MHC I US2 has previously been shown to down-regulate MHC class I through ERAD [132] and we wished to confirm that the increased MHC I levels observed with CypC depletion were consistent with impaired ERAD. As shown in Figure 4.6A, when US2+ cells were radiolabeled for 10 min and chased for periods up to 30 min, a rapid loss in newly synthesized MHC I heavy chain was observed. Consistent with ERAD disposal, this loss could be largely blocked by combined treatment with the proteasome inhibitors lactacystin and MG132. Proteasome inhibition also resulted in the accumulation of deglycosylated class I heavy chains that are normally degraded following retrotranslocation to the cytosol. To assess the effect of CypC depletion on this process, US2+ cells were subjected to RNA interference with negative control or CypC sirnas, radiolabeled for 5 min and chased for up to 20 min. In cells depleted of CypC there was a striking increase in the amount of newly synthesized MHC I which was then degraded throughout the chase (Figure 4.6B). This suggested that CypC depletion mainly stabilizes MHC I during or shortly following synthesis. 154

170 Figure 4.5. Increased MHC I expression upon CypC depletion is not due to US2 instability or ER stress. (A) U373-MG cells were treated with CypC-(i) sirna and US2 protein was monitored by immunoblotting on day 6 using anti-us2 antiserum. Calnexin and GAPDH were used as loading controls. CypC depletion and increased MHC I levels were verified by immunoblotting with anti-cypc antiserum and mab HC10, respectively. (B) U373-MG + US2 cells were treated with CypB or CypC-(i) sirna and RNA was isolated 6 days later. Primers specific to both spliced and unspliced Xbp1 mrna were used to identify activation of the unfolded protein response in a one-step reverse transcriptase PCR. The PCR products corresponding to the unspliced and spliced forms are indicated. Cells were treated overnight with 20 μg/ml cyclosporine A as a positive control for Xbp1 splicing. 155

171 156

172 Figure 4.6. Both depletion and overexpression of CypC stabilize MHC I in cells expressing US2. (A) U373-MG + US2 cells were starved of methionine, radiolabeled with [ 35 S] Met for 10 min and chased in Met-containing medium for the indicated times. Cells at each time point were treated either with DMSO in the starvation, pulse, and chase media, or with a combination of MG132 and lactacystin. Cells were lysed and MHC I was isolated using a cocktail of mabs W6/32, HC10, and HCA2. Anti-calnexin was also included in the same immunoisolation to control for differences in radiolabeling and immunoisolation efficiencies. (B) U373-MG + US2 cells were treated with negative control or CypC-(i) sirnas. On day 6 samples were radiolabeled for 5 min with [ 35 S] Met. Cells were lysed following labeling or after chasing in Met-containing medium for the indicated times. MHC I was immunoisolated using mab HC10. Cell lysate (lower panel) served as a control for differences in radiolabeling efficiency. (C) U373-MG and U373-MG+US2 cells were treated with negative control or CypC-(i) sirnas and on day 6 triplicate samples were radiolabeled for 10 min with [ 35 S] Met. Cells were lysed and MHC class I was immunoisolated as in panel A. (D) U373-MG cells were transduced with lentiviruses prepared with either empty vector or vectors encoding sirna-resistant forms of CypC WT, CypC R89A, or CypC K123A. Following transduction, all cells were sorted for low expression of the plasmid marker (ZsGreen1). Cells were treated with CypC-(i) sirna to deplete endogenous CypC and lysed on day 6. MHC class I levels were assessed by immunoblotting with mab HC10. GAPDH served as a loading control. (E) U373-MG + US2 cells were transduced with lentivirus prepared with empty vector and treated with either negative control or CypC-(i) sirnas. Additional cells were transduced with lentiviruses expressing either CypC WT, CypC R89A, or CypC K123A and were treated with negative control sirna. On day 6 following knockdown all cells were stained with mab W6/32 for determination of surface MHC class I expression by flow cytometry. To confirm this, we radiolabeled both control and US2-expressing cells for 10 min without a subsequent chase and analyzed in triplicate. As shown in Figure 4.6C, comparison of the negative sirna lanes in the two cell lines revealed that substantially less newly synthesized MHC I was recovered from the US2+ cells, suggesting that in a 10 min pulse a substantial portion of MHC I was being degraded. Upon depletion of CypC, this degradation was impeded consistent with impairment in US2-mediated ERAD. Furthermore, in keeping with the previous immunoblotting experiments, CypC depletion had no effect on newly synthesized MHC I in control cells, indicating the specificity of CypC in US2-mediated degradation as opposed to general effects on MHC I synthesis and folding processes within the ER (Figure 4.6C). 157

173 4.2.4 Overexpression of CypC disrupts US2-mediated degradation of MHC class I molecules To further validate the involvement of CypC in US2-mediated ERAD and to evaluate functional sites, we reintroduced sirna-resistant versions of wildtype CypC as well as functional site mutants into US2+ cells depleted of CypC. For functional site mutants, we tested R89A and K123A which correspond to mutations in CypB previously demonstrated to disrupt catalytic activity [119,227] and a calreticulin/calnexin-binding site [228], respectively. CypC has demonstrable peptidyl prolyl isomerase activity [279] but its ability to bind calreticulin or calnexin is less clear. Unexpectedly, overexpression of wildtype CypC did not restore the rapid degradation of MHC I in US2+ cells depleted of CypC (Figure 4.6D, middle), which raised the possibility that the impaired MHC I degradation was due to an off-target effect of the CypCdirected sirna. However, this was not the case as we found that simple overexpression of CypC without any knockdown caused a dramatic increase in MHC I levels as assessed by flow cytometry (Figure 4.6E, left). This suggested that US2-mediated degradation is sensitive to the expression level of CypC, and that both CypC depletion and overexpression impacts US2 function. Furthermore, overexpression of both CypC mutants caused a similar increase in MHC I expression (Figure 4.6D and Figure 4.6E) indicating that the catalytic activity and potential interactions with lectin-chaperones are dispensable for the impaired MHC I degradation caused by CypC overexpression CypC interacts with US2 and MHC class I Since both overexpression and depletion produced similar phenotypes, we hypothesized that CypC may bridge the US2-MHC class I complex with ERAD machinery. In this scenario, depletion of CypC removes the bridge, whereas overexpression disrupts the stoichiometry, or saturates, the link between US2 and ERAD components. Another possibility is that altered 158

174 expression of CypC, as a component of a complex network of ER folding and quality control machineries [86], may indirectly impact US2 function by affecting the levels or spectrum of proteins associated with the US2-MHC I complex. In either scenario, CypC might be expected to be present in a complex with US2 and MHC I. Indeed, co-expression of HA-tagged CypC with US2 resulted in robust coimmunoisolation of US2 with anti-ha mab (Figure 4.7A). In contrast, when a myc-tagged CypB construct was expressed under similar conditions, no co-immunoisolation of US2 could be detected (Figure 4.7B), consistent with the lack of phenotype associated with CypB knockdown (Figure 4.3B and Figure 4.4B). In addition, overexpressed HLA-A2 could be co-immunoisolated with HA-CypC (Figure 4.7C). Since this was observed in U373-MG cells (not expressing US2), it suggests that CypC is able to interact with MHC I under normal MHC I maturation conditions. Importantly, although CypC could be detected in association with both US2 and MHC I, it was not required for their interaction. As shown in Figure 4.7D, depletion of CypC did not impair the recovery of US2-MHC I complexes immunoisolated from cells transfected with both US2-HA and HLA-A2 (Figure 4.7D), Rather, there appeared to be an increased association of HLA-A2 with US2 consistent with reduced MHC I degradation in CypC-depleted cells US2 and MHC class I interact with a diverse array of proteins If CypC depletion/overexpression is modulating the quality control/erad machinery recruited by US2 to effect rapid MHC I disposal, one would expect to see alterations in the spectrum of proteins associated with US2 upon CypC depletion or overexpression. To this end, we used lentiviral transduction to establish three stable cell lines for use in an unbiased LC- MS/MS proteomic screen. These included U373-MG cells co-expressing US2 with a triple HA tag along with either a nonsense shrna, a mixture of four CypC-specific shrnas to deplete CypC, or a construct that overexpresses CypC. 159

175 160

176 Figure 4.7. CypC co-isolates with US2 and MHC class I molecules. U373-MG cells were transiently transfected for 48 h with the indicated constructs in pcdna3.1+/zeo and then lysed in 1% digitonin lysis buffer followed by immunoisolation of epitope-tagged proteins. (A) CypC-US2 interaction. U373-MG cells expressing HA-CypC, US2, or HA-CypC + US2 were subjected to immunoisolation using anti-ha mab and associated US2 was detected by immunoblot with anti-us2 antiserum. A background signal from the anti-ha mab light chain is denoted by *. An unidentified band in anti-cypc immunoblots appearing below HA-CypC in cells transfected with both HA-CypC and US2 is denoted by **. (B) CypB does not interact with US2. U373-MG cells expressing myc-cypb, US2, myc-cypb + US2, or HA-CypC + US2 were lysed and immune complexes isolated with anti-myc or anti-ha mabs as appropriate. US2 was detected by immunoblotting with anti-us2 antiserum. The myc- CypB construct produced two non-specific bands in an anti-us2 blot which are denoted by **, and can be observed most clearly in the Myc-CypB + Vector control lane. While these bands overlap with the upper US2 band (1), the lower species (2) can be clearly distinguished. The single * corresponds to the light chain of the anti-ha or anti-myc mab. (C) CypC-HLA-A2 interaction. U373-MG cells transiently transfected with plasmids encoding HLA-A2, HA-CypC, or HLA-A2 + HA-CypC were subjected to immunoisolation with anti- HA mab. Co-isolated HLA-A2 was detected by immunoblot with mab HCA2. (D) Depletion of CypC does not disrupt the US2-MHC I interaction. U373-MG cells were treated with negative control or CypC-(i) sirnas on day 0 and again on day 3. Cells were then transfected on day 4 with plasmids encoding US2-HA, HLA-A2, or US2-HA + HLA-A2. Immunoisolations of US2-HA were conducted on day 6 using anti-ha mab and associated HLA-A2 was identified using mab HCA2. CypC depletion was detected using anti-cypc antiserum with GAPDH as a loading control. We also prepared two control cell lines: U373-MG cells transduced with control lentiviral vectors (U373-MG + non-sense shrna) to identify proteins that interact non-specifically with anti-ha conjugated beads, and U373-MG expressing both US2 and HA-tagged HLA-A68. This latter control was used to filter hits from the US2-3xHA pulldown for those also present in an MHC class I pulldown, reasoning that such proteins might be more likely to play a role in US2- mediated degradation of MHC I. All cell lines were incubated for 8 h with MG132 to inhibit MHC I degradation and to stabilize associated proteins, and then were lysed in digitonin, subjected to immunoisolation using immobilized anti-ha mab, and recovered proteins were digested with trypsin and subjected to LC-MS/MS. The results were analyzed using the Peaks 7 software package [248] which provides a semi-quantitative estimate of protein abundance (see Section ). By 161

177 normalizing protein abundances to the US2-3xHA bait protein in the various samples, the fold change in abundance of US2-associated proteins between samples with different CypC levels (endogenous, depleted and overexpressed) could be calculated. This data is shown in Appendix 1 for 177 identified prey proteins. Similarly, we identified 59 proteins in the HA-HLA-A68 isolations by LC-MS/MS and their relative abundances are shown in Appendix 5. The results from the combined US2-3xHA immunoisolations and the HA-HLA-68 immunoisolation were analyzed using the Gene Ontology Consortium [233] and Panther Classification System [234] analysis tools. Identified proteins were examined for statistical enrichment of particular GO terms when compared to a database of 20,814 GO annotated proteins. By comparing the number of occurrences of a particular GO term in the LC-MS/MS results (#Observed) to the number of occurrences in the database (# Database), the likelihood of seeing a particular term by random chance in the LC-MS/MS dataset could be calculated (# Expected). A number of GO terms were enriched for proteins recovered in the US2-3xHA and HA-HLA-A68 isolations (Figure 4.8A compare #Observed and # Expected; Appendix 6 and Appendix 7). These included terms for protein folding, MHC I antigen presentation, cellular responses to stress, protein glycosylation, and endoplasmic reticulum localization of proteins. In addition, the US2-3xHA immunoisolations had a statistically significant number of proteins involved in ERAD and viral processes. As an alternative means to visualize the types of proteins present in each LC-MS/MS dataset, we used STRING to identify relationships between these proteins [280]. By analyzing US2-3xHA-and HA-HLA-A68-associated proteins for experimental, database, text mining, and homology links, clusters of proteins tied to related functions or complexes could be identified. 162

178 163

179 Figure 4.8. Identification of US2-associated proteins that are affected by CypC depletion or overexpression. (A) Gene Ontology classification of US2-3xHA- and HA-HLA-A68-interacting proteins. The frequency of proteins associated with particular GO terms in the LC-MS/MS datasets (# Observed, out of 177 identified proteins for US2-3xHA and 59 proteins for HA-HLA-A68) was compared to the number of proteins with that particular GO term in the GO Ontology database (# Database, out of 20,814 entries) to determine the frequency that a particular GO term would appear in the LC-MS/MS dataset by random chance (# Expected). The p-value indicates the probability that the # Observed for each GO term appears in the LC-MS/MS dataset by random chance. All p-values use the Bonferonni correction for multiple testing. (B) Many US2 interacting proteins change in abundance upon CypC depletion or overexpression. Quantified abundances of the various proteins detected in U373-MG + US2-3xHA isolations and their fold enrichment or depletion following CypC depletion (left panel) or overexpression (right panel) were plotted to evaluate overall changes. HNRPQ (SYNCRIP), CLH1 (CLTC), NEST (NES), CMC2 (SLC25A13), and THY1 are not plotted, as they are located outside the axis limits (Appendix 4). Of the 174 identified proteins examined, 164 increased in association with US2-3xHA following CypC depletion and 135 increased in association following CypC overexpression. (C) Proteins detected in association with US2-3xHA were filtered against those detected in association with HA-HLA-A68 to include results common to each. The fold change in their abundance following depletion (black bars) or overexpression (white bars) of CypC is shown. Identification of selected GO terms is indicated below each set of bars as coded dots. US2 (which was used for normalization) is boxed, and hits selected for further validation are shaded in grey. As expected, clusters related to ER-associated degradation and the peptide loading complex were present in both the US2-3xHA (Figure 4.9) and HA-HLA-A68 (Figure 4.10) datasets. Note that not all proteins identified by LC-MS/MS were displayed due to the high stringency required for links. Additionally, the US2-3xHA dataset contained a concentration of proteasome-related proteins. Surprisingly, a cluster of oligosaccharyl transferase (OST) complex and TRiC chaperone complex proteins were observed in the STRING analysis. 164

180 Figure 4.9. STRING analysis of proteins identified in US2-3xHA isolations. Proteins identified by LC-MS/MS from immunoisolations of US2-3xHA (Appendix 4) were submitted for characterization using the STRING10 web software that displays protein relationships based on experimental evidence, database links, and homology links [280]. 165

181 Figure STRING analysis of proteins identified in HA-HLA-A68 isolations. Proteins identified by LC-MS/MS from immunoisolations of HA-HLA-A68 (Appendix 5) were characterized using the STRING10 web software [280] in a similar manner to that described in Figure

182 While the presence of individual subunits could be present by chance, the representation of these complexes in both datasets suggested a potential role in US2-mediated degradation. When the quantitative data from the US2-3xHA immunoisolations was compared between samples from cells expressing normal levels of CypC versus those depleted of CypC, a large number of hits increased in association with US2 upon CypC depletion rather than disappearing (Figure 4.8B, left panel). A similar trend was observed for CypC overexpression (Figure 4.8B, right panel). To focus the analysis of US2-interacting proteins to those most relevant to MHC I degradation, a list was constructed of proteins that were recovered in common from both the US-3xHA and HA-HLA-A68 isolations. The fold change in abundance of these hits was then examined following CypC depletion or overexpression (Figure 4.8C). A number of proteins involved in protein folding, such as molecular chaperones that may be involved in early targeting of MHC I for ERAD, as well as redox factors (isomerases, oxidoreductases) that could help unfold MHC I prior to retrotranslocation, were increased in association with US2 following knockdown or overexpression of CypC. Several of these, such as BiP, calnexin, calreticulin and the thiol oxidoreductase PDIA1, have previously been linked to US2 function [207,214,217]. In addition, we observed a large number of OST complex subunits as well as cytosolic TRiC chaperonin complex subunits (TCPG and TCPZ) that had also been identified in the STRING analysis. The increase in the association of these proteins with US2 was intriguing, since we expected with the bridging model to see a reduction in associated ERAD/quality control components upon CypC depletion or overexpression. Rather, the overall increase in levels of these hits combined with the involvement of many of them in protein maturation processes suggested that modulation of CypC levels may stall US2 in association with its interaction partners. In such a scenario, depletion of these interaction partners might also affect US2-167

183 mediated degradation of MHC I Identification of TMEM33, PDIA6, and malectin as novel participants in US2-mediated degradation of MHC I To test whether the proteins that were enriched in association with US2 following alteration of CypC expression were indeed playing a functional role in MHC class I degradation, we selected a number of hits for further analysis. TMEM33 was the top hit that was enriched upon CypC depletion or overexpression (Figure 4.8C). This predicted three-pass transmembrane protein is ER localized and is potentially involved in the regulation of reticulon proteins [281]. It is also induced upon ER stress and its overexpression modulates the activities of the unfolded protein response sensors IRE1 and PERK [282]. PDIA6, also known as P5 [283], was another prominent hit and as a thiol oxidoreductase could potentially be involved in reducing MHC I disulfides prior to retrotranslocation and degradation [270]. Malectin was also of interest, as this lectin has been linked to retention of misfolded proteins within the ER [93,284]. The oligosaccharyltransferase (OST) subunit ribophorin I (RpnI) was also selected due to its published interactions with malectin [94,285] and the fact that another OST subunit, ribophorin 2 (Rpn2), also exhibited increased association with US2 (Figure 4.8C). Lastly, the TRiC chaperone complex was targeted. Multiple subunits of this complex were detected in the US2-3xHA and HA-HLA-A68 immunoisolations (Figure 4.9 and Figure 4.10). One of the subunits, TCPQ (or CCT8), was chosen to examine potential involvement of the complex in US2 function. Although the identification of this chaperone complex may be artifactual due to its cytoplasmic localization and ability to interact with non-native substrates, the fact that multiple subunits were represented in our results bore further investigation. We first confirmed by IP-western the associations between US2 and malectin, TMEM33 and PDIA6, as well as the increased association between US2 and either TMEM33 or malectin 168

184 following CypC overexpression or depletion (Figure 4.11A and Figure 4.11B). To assess whether these or any of the other selected proteins influenced US2-mediated degradation of MHC I, two to three sirnas for each target were pooled for use with U373-MG + US2 cells. For PDIA6, a previously validated sirna used in studies of its ER quality control functions was used [113] and CypC sirna was included as a positive control. In all cases, efficient knockdown was observed at the mrna level, as assessed by qpcr (Figure 4.12A). Immunoblotting to detect MHC class I protein in these US2+ sirna-treated cells revealed significant increases in MHC I upon depletion of CypC as expected, but also upon depletion of PDIA6, TMEM33, and malectin (Figure 4.12C, quantified in Figure 4.12D). No increases were observed in the levels of HLA-A transcript for any of these depletions, suggesting that any observed increases in MHC I protein levels were not a consequence of increased MHC I transcription or altered transcript stability (Figure 4.12B). In contrast, depletions of RpnI or CCT8 were not associated with an increase in steady state MHC I protein levels. No statistically significant increase in MHC I expression was observed for any of the knockdowns in U373MG control cells, consistent with PDIA6, TMEM33 and malectin being specifically involved in US2- mediated degradation (Figure 4.12C and Figure 4.12D). These effects of target depletion were confirmed by flow cytometry. As expected, depletion of TMEM33, PDIA6, or malectin increased surface MHC I in US2+ cells, with no impact of any of these knockdowns on surface MHC I in U373-MG cells (Figure 4.12E). Further investigation into PDIA6 revealed that in addition to playing a role in US2+ degradation of MHC I, it also stabilized MHC class I molecules in U373-MG cells expressing US11 (Figure 4.13A). Remarkably, in a manner reminiscent of CypC, over-expression of PDIA6 also increased the level of MHC class I in US2+ cells (and also in US11+ cells, Figure 4.13B). 169

185 Figure TMEM33, malectin, and PDIA6 co-isolate with US2. (A) US2-3xHA stably expressed in U373-MG cells or in cells with stable depletion (CypC KD) or overexpression (CypC OE) of CypC was immunoisolated from digitonin lysates using anti- HA mab and immunoblotted for associated malectin and TMEM33. * denotes background bands present in immunoblots of cell lysates. ** denotes a background band observed in all IP samples following immunoblotting for TMEM33. (B) U373-MG cells transiently transfected with plasmids encoding US2, PDIA6-HA, or US2 + PDIA6-HA were subjected to immunoisolation with anti-ha conjugated beads. Co-isolated US2 was detected by immunoblot with polyclonal anti-us2 serum. Two different exposures of the US2 immunoblot are shown. 170

186 171

187 Figure US2-associated proteins identified by modulating CypC expression participate in US2 degradative functions. (A) Determination of sirna knockdown efficiency. The indicated targets of interest were depleted in U373-MG + US2 cells using 2-3 individual sirnas transfected together on day 0 and boosted on day 3. On day 6 cells were lysed and RNA isolated for real-time PCR analysis. All knockdowns resulted in 90% or greater depletion of target mrnas. Results were normalized to beta actin using the delta delta Ct method for comparison of knockdown treatments. An unpaired Student s T-test was used to show statistically significant differences from the negative control sirna (* indicates p-value < 0.05, n = 5). (B) Using a similar method, HLA-A transcripts were quantified under various knockdown conditions. (C) U373- MG + US2 cells and U373-MG control cells were treated with the indicated sirna pools and on day 6 MHC class I levels were determined by immunoblotting with mab HC10. (D) Quantified MHC I band intensities from the experiment in panel C. All results were normalized to the negative control sirna sample for each cell line (* indicates p-value < 0.05, n = 4). (E) U373-MG + US2 cells and U373-MG control cells were subjected to knockdown as in panel (C) and surface MHC I levels were determined by flow cytometry following staining with mab W6/ CypC, malectin, PDIA6, and TMEM33 do not affect the degradation of a soluble ERAD substrate Following the observation that PDIA6 participates in both US2- and US11-induced degradation of MHC I, it was of interest to determine whether any of the proteins that were identified might be more broadly involved in ERAD. Consequently, we assessed the effects of their depletion on the degradation of a misfolded soluble ERAD substrate, the null Hong Kong (NHK) variant of α1-antitrypsin that is degraded by the HRD1 branch of ERAD [155]. As shown in Figure 4.13C, depletion of PDIA6, CypC, malectin, and TMEM33 had no effect on steady state NHK levels suggesting that they do not participate in ERAD of soluble glycoprotein substrates. In contrast, depletion of Rpn1 was associated with the appearance of more rapidly migrating forms of NHK, which were determined to be underglycosylated species by endo H digestion (data not shown), and there was also an increase in the total amount of NHK present. 172

188 Figure Involvement of novel US2-associated proteins in other ERAD pathways (A) US11-mediated degradation. U373-MG + US2 or + US11 cells were treated on day 0 and on day 3 with an sirna specific for PDIA6 or a negative control. On day 6 cells were lysed and MHC I levels were determined by immunoblotting with mab HC10. GAPDH served as a loading control. (B) U373-MG, U373-MG + US2, or U373-MG + US11 cells were transiently transfected with either empty pcdna vector or pcdna + PDIA6 WT. After 48 h, MHC I levels were determined by immunoblot; BiP served as a loading control. (C) U373-MG cells were treated with the indicated panel of pooled sirnas on days 0 and 3. On day 4 cells were replated and transiently transfected with an α1-antitrypsin NHK variant expression plasmid. Approximately 48 h later cells were lysed and NHK levels were determined by immunoblotting. 173