Profiling and Imaging Mass Spectrometry Compilation of Work from the Caprioli Lab James Mobley, Ph.D. Director of Urologic Research Associate Director of Mass Spectrometry University of Alabama at Birmingham mobleyja@uab.edu
Primary Focus of Today s s Lecture? Brief Overview of Biomarker Discovery (BMD) for Clinical Applications, Why do we do it, Why do we use MALDI-ToF? Understanding Advances in MALDI-ToF Driven Profiling of Tissue Sections for BMD, and the bottlenecks in this newly emerging field. How to produce a Mass Image from a Series of Profiles. What to Do with All that Data (Following the Workflow from Pre-prosessing to Statistical Analysis). How to ID those Peaks, are they really Proteins?
First off.. Why Do Biomarker Discovery? To find associations between biological components (i.e. SM, FA s, Proteins) and any clinical endpoint quickly, non-invasively, affordably. To non-invasively determine.. Pathologic Changes (i.e. early detection of cancer) Aggressiveness/ Stage of Disease Predicting Rx Response Drug Target Discovery Mechanistic Studies (Systems Biology) The Potential Clinical Impact is Tremendous!!
Using Proteins as an Example; How new is BMD? Number of Publications 3000 2000 1000 BMD Publications: 51 real, 29 reviews Proteomics BMD-Proteomics Nearly half are reviews! 0 1 2 3 4 5 6 7 8 9 10 Years Past
Bottlenecks in Biomarker Discovery Quick & Reproducible Protein Isolation Biological Samples Increase Acquisition Speed & Sensitivity Spectral Pre-processing & Data Analysis Streamline Protein Characterization Diagnostic Assay
Can 2D PAGE Get Us There? Primarily HMW Proteins! O. Golaz etal. (1993) Electrophoresis 14 1223-1231. Acidic region IPG, ph 3.5-10 Serum Proteome Anderson, PNAS, 1977 Direct Analysis ~1000 Protein Spots 25yrs later Pieper, Proteome, 2003 Affinity, SCX, Size Exclusion Yielded 74 Fractions - 2D 3700 Proteins Spots 327 Distinct Proteins
What About MALDI-ToF for Biomarker Discovery? MALDI-Tof Crude Protein 2D PAGE, ID by MALDI-Tof Primarily LMW Proteins!
R e la tiv e P e ak Area ( 10 0 % = 1.6 fm o le ) Sensitivity is Inversely Proportional to Mass (MALDI-ToF Example) c 100.0 25.0 6.3 1.6 0.4 0.1 MALDI-Tof Low 1.6 femptomole Attomole 3.9 pg/ μl 1.6 fmole 2D PAGE Relative Peak Area (100 % set equal to 1.6 fmole) 0 10 20 30 40 50 Molecular Weight (kda)
Taking a Closer Look at our Proposed Target Primary Target: Many growth factors and cytokines are secreted into the plasma. Growth factor were previously referred to substances that promote cell growth. Promote/ Inhibit: mitogenesis, chemotaxis, apoptosis, angiogenesis, differentiation. Cytokines were simply known as proteins that exhibited immuno-modulating effects. A generic name for a diverse group humoral regulators. Chemokines are a family of cytokines previously referred to as the SIS, SIG, SCY, PF4, and Intercrine families. 8-10 kda, chemotactic agents, with high homology (contain C, CC, CXC, or CX3C). Known In Late 90 s, the cytokine family was limited mainly consisted of 22 lymphokines; But Now, Cytokines, including those with no names: 491, Separate Chemokines: 345 Primary Source: Cytokines Online Pathfinder Encyclopaedia http://www.copewithcytokines.de/
Advantages of MALDI-ToF for Profiling Biological Soln s and Tissue Sections Resistant to many impurities, robust! Highly sensitive in low mass range, which is just starting to be chartered. Direct Analysis on Tissue; very little protein is required for analysis. Analysis of crude extracts of biological fluids. Primarily 1+ charged proteins/ peptides (less complex specta). HTP!!
Directed Discovery Profiling Slice frozen tissue on cryostat (~12 μm thick) Thaw slice onto MALDI plate, allow to dry Non-Directed Discovery Imaging Droplets Apply matrix Spray coating Laser + + + Acquire mass spectra Laser + + + Protein profiles Protein images m/z 18 388
Protein Expression Profiling by MALDI-MS Human breast tumor needle biopsy 1 mm % Intensity 40 32 24 16 8 10 28085 8 6 4 2 17253 17381 22265 31011 43255 41854 0 17000 27600 38200 x5 48800 59400 70000 3 81934 6434 6633 7564 α2+ 7934 β 2+ 8686 8810 9442 5360 10091 0 2000 5000 8000 11000 14000 17000 44085 11649 199988 ~50500 12344 13760 15127 15869 α + β + 14037 66753 2.4 215149 1.8 78494 1.2 94266 97455 132990 162805 0.6 148196 0 70000 116000 162000 208000 Mass (m/z) 254000 300000
Human Glioma Biopsy 3 mm X 5 C Tumor A B D Non-Tumor Tumor (A) Tumor (A) Tumor (B) Tumor (B) Non Tumor (C) Non Tumor (C) Non Tumor (D) Non Tumor (D) 4000 8000 Schwartz et Al. Clin. Cancer Res., 10, 981-987 (2004). m/z 15000
Principle of MALDI MS Imaging Frozen section MS Images Matrix application Raster of section 4000 6000 8000 10000 m/z MALDI Mass Spectrum
Spray Deposition of Matrix on Tissue Sections Spray nebulizer for TLC plates Matrix solution Tissue section Nitrogen
Comparing Matrix Coatings Spotted 100 5445 9569 80 6575 % Intensity 60 40 20 4965 6228 9906 11208 12239 14005 16773 20746 26835 50um Laser spot 200 μm Sprayed 0 3000 8400 13800 19200 24600 30000 Mass (m/z) 100 5445 80 6228 % Intensity 60 40 4965 6575 9569 9906 20 11208 12239 14005 16773 18310 20746 26838 0 3000 8400 13800 19200 24600 30000 Mass (m/z)
Glioma Mouse Model - Intracranial Injection of GL261 Cancer Cells Tumor 2 mm 3347 4747 5171 5444 5708 6251 6545 6573 7492, α 2+ 7537 7810, β 2+ 9193 9910 9975 10259 8565 6719 4965 Tumor Non tumor NT) T) 2000 3800 5600 7400 9200 11000 x25 11641 12372 12134 19940 20743 22173 16790 14982, α + 15618, β + Relative intensities NT) T) 11000 13800 16600 19400 22200 25000 Mass (m/z)
Glioma Mouse Model. Imaging Resolution: 100 μm 2 mm m/z 6924 m/z 7539 m/z 9910 Acyl CoAbinding protein m/z 11 307 and 11 348 Histone H4 m/z 12 134 Cytochrome C m/z 13 804 and ~15 350 Histones H2B1 and H3 m/z 14 039 m/z 14 982 and 15 617 α and β hemoglobins m/z 18 420 Chaurand et Al. Anal Chem, 76, 86A-93A (2004). m/z 22 173 m/z 24 807 0 100%
Can We Do Better? Tissue Profiling/Imaging How Best to Apply Matrix? Matrix Deposition Variables: Time Repr. Inherent Spot Res. Laser Demand BG Dependent Spot Size Manual Spotting seconds variable N >1mm Y Robotic Spotting minutes good N ~200µ Y Laser Capture hours good Y ~7µ to 100 µ N Current Laser Spot Size for Old STR: 25µ x 50µ
Hand Spotting Vs. Pico Spotting Vs. LCM 4x HandSpot ~2mm 10x PicoSpot ~200µ 100 µ
The Robotic Spotter Computer controlled motorized stage Drop imaging MALDI target with sample Strobe LED for drop illumination Substrate image capturing Acoustic drop ejector with reservoir
Acoustic Drop Ejection Technology x y Sample plate Ejection of microdroplets Matrix reservoir Coupling membrane Transducer Coupling reservoir RF generator Waveform generator Current performances: - Spot size: ~180-200 µm - Drop ejection rate: 10 Hz - Drops per pixel: 60-80
Tissue Sectioning for Protein Profiling
Seamless High Resolution Imaging of Histological Slides Whole Mouse Mammary Gland
Histology Directed Matrix Deposition
Whole Lung Sections Lung Mets
Automating Whole Plate Profiling Rapid Spotting Over Entire Plate MALDI-Tof Plate... A2 Not used A3..... A4 A5 B1. B2 B3.... B4 C4 C3 C2 C1
Breast Cancer Early Vs. Late Disease M.G. - NO Tumor M.G. Early BCa M.G. Late BCa 2 kda 20 kda
Lung Results Mets from Vs. Mass Late Profiling Carcinoma Tissue of the Sections Breast Lung Mets Vs. Late Carcinoma Calgranulin A M.G. Late BCa Lung Mets 2 kda 20 kda
Manual Spotting Vs. The Robotic Spotting A) Mouse brain, Analysis of the corpus callosum B)
MS Imaging of a Mouse Brain Section By Robotic Spotting m/z 17885 m/z 11839 m/z 20688 1000 μm 0 % 100 % m/z 11790 m/z 6759 m/z 7338 Intensity 2014 15579 29144 42708 56273 69838 Mass (m/z)
Schematic Representation of Protein Marker Identification Homogenate Lung Tumor * 5000 8750 12500 16250 20000 m/z MD HPLC Fractions Tryptic Digestion MS/MS MALDI MS * 10000 12500 15000 m/z Mw Matches Mw Matches Add Post- Translational Modification Database Search Protein ID 100 300 500 700 900 1100 1300 m/z
mucosa submucosa Normal Human Colon Biopsy 1000 µm muscle m/z 22535 Combined Image m/z 7930 m/z 8958 Combined Image
cortex Rat Kidney Sagittal Section medula 600 9979 500 9943 400 m/z 9979 m/z 10,935 2.5 mm Intensity 300 200 m/z 9943 m/z 17,514 100 0 9900 9950 10000 10050 Mass (m/z) Overlay Overlay
Molecular Determination of Tumor Margins Histological tumor margin Non-tumor 1 cm 15 x 100 spots 250 µm resolution Tumor Vessel Fibrous stroma Skeletal muscle Robert Caldwell,
B A B A C D E E C D
Histological vs. Molecular Assessment of the Tumor Margin:
Proteins are OK But What About Drug Distribution/ Metabolism? Better..Correlating Drug Effects with Protein Expression. Cut frozen slice (12 μm) Dose animal orally i.v. Remove tissue Apply matrix Reyzer ML et al, J Mass Spectrom, 38, 1081-1092 (2003) Reyzer ML et al, Cancer Res 64, 9093-9100 (2004) Laser Analyze by: MALDI MS and MALDI MS/MS
MMTV/HER2 Transgenic Mouse Mammary Tumors MMTV/HER2 cells transplanted in FVB female mice. Tumor grown to a size of ~200 mm 3. OSI 774 is an intracellular tyrosine kinase EGF receptor inhibitor. Administered orally for 1 week Tumor Volume (mm 3 ) 1000 800 600 400 200 Captisol 200 µl (vector) OSI 774 100 mg/kg 0 0 5 10 15 20 25 30 35 Days post-treatment Contributed by M. Sliwkowski (Genentech, Inc.)
O O MS/MS Analysis of OSI-774 in Tumor Tissue Tumors removed after a single 100 mg/kg dose of OSI-774 O O +H +H m/z 278.1 m/z 336.2 N N NH 278.1 +H + MW = 393.4 Intensity [OSI774 + H] + 336.2 HC C 100 200 300 400 500 m/z 4 hr 1 2 3 5 4 6 330 Intensity 278 Spot #3 Monitored CAD transition m/z 394 278 240 260 280 300 320 m/z
Dose Dependence of Protein Alteration (20 hr after dose) 9 control mice, Av of 54 spectra 3x3 dosed mice, Av of 9 spectra per dose thymosin β 4 Relative intensity 4700 4780 4860 4940 5020 5100 Control 10 mg/kg 30 mg/kg 100 mg/kg 8300 8400 8500 8600 8700 8800 Mass (m/z)
Analysis of MALDI-Tof Data The Workflow Raw Spectra Preprocessing Txt File Export Baseline/ noise subtraction Recalibration/ Realignment Normalization TIC Wavelet Processed Spectra Peak Picking Binning Peak Selection Supervised Feature Selection WMA T-Test Supervised Classification (training set) KNN SVM/ LOOCV Visual Inspection of Peaks rule out noise chemical adducts Biomarker List
Recent Example; No Data Processing Raw Files Check For Saturation! 2.5kDa 13.5kDa
Post Baseline and Normalization Pre Alignment Check Alignment! * 2.5kDa 13.5kDa
* Post-BSL/Norm/Align [No Smoothing] Good Alignment Decreased Error Smaller Bin Sizes!
Testing the Approach: Liver Set Doped with Known Proteins 10 Spectra/ Set 1. Baseline Correct (Efeckta) 2. Smoothing (none) 3. Calibration (Efeckta) 4. Normalize/ Transform (TIC, Wavelet, Log, Ln, CR) 5. Standardization (none) 6. Peak Picking (WMA data dependent cutoff ) m/z mix 1 2 3 4 Insulin (porcine) 5778.6 0.2375 0.11875 0.059375 0.02375 Cytochrome C 12361.2 0.95 0.475 0.2375 0.095 Apomyoglobin 16952.5 2.375 1.1875 0.59375 0.2375 Trypsinogen 23982 2.5 3.75 4.375 4.75 Spectra contain liver extract proteins spiked with the standard proteins listed above. Concentrations are in pmol/ul (μm)
Relative Normalized Intensities 0.8 0.0-0.8 Insulin 2+ cytc ApoM Tryp NonNormalized TIC STD's CR Wavelet WaveletTIC Peaks Identified Using all Techniques with all stds at 2 fold diff except Tryp at 1.3. TIC and TIC with Wavelet work best! 0 10k 20k 30k 40k m/z Relative Normalized Intensities 0.4 0.2 NonNormalized TIC STD's CR Wavelet WaveletTIC 0.0 8.4k 8.5k m/z
-statistically relevant peaks that are not real! Std Set with 10 Fold Difference [Set 1 Vs. Set 4] Relative Normlized Intensities 0.8 0.0-0.8 NonNormalized TIC STD's CR WaveletNorm WaveletSMNorm WaveletTIC Peaks Identified Using all Techniques with all stds at 10 fold diff except Tryp at 2.0. 0 10k 20k 30k 40k m/z Std Set with 10 Fold Difference [Set 1 Vs. Set 4] Relative Normlized Intensities Std Set with 10 Fold Difference [Set 1 Vs. Set 4] NonNormalized TIC STD's CR WaveletNorm WaveletSMNorm WaveletTIC Relative Normlized Intensities 0.8 NonNormalized TIC STD's CR WaveletNorm WaveletSMNorm WaveletTIC 0.0 0.0 8.4k 8.5k 8.5k m/z 17k 18k 19k m/z - logging techniques pick shoulders not peak asymptotes!
Spectral Processing Increases Detection Range and Sensitivity Unprocessed Processed 1e+5 1e+5 Ins (1+) CC (1+) ApoM (1+) Ins (1+) CC (1+) ApoM (1+) log (Absolute Intensity) 1e+4 log (Normalized Intensity) 1e+4 1e+3 1e+3 0.01 0.1 1 10 log (Protein Concentration) 0.01 0.1 1 10 log (Protein Concentration) 4000 4000 3500 3500 Intensity 3000 2500 2000 1500 Normalized Intensity 3000 2500 2000 1500 1000 1000 500 500 0 0 10000 20000 30000 m/z -500 10000 20000 30000 m/z
These Techniques also Greatly Enhance MALDI Generated Images! Raw Data Processed Data Norris J.L., Cornett, D.S., Mobley J.A., Andersson M., Caprioli R.M. Processing MALDI Mass Spectra to Aid Biomarker Discovery and Improve Mass Spectral Image Quality, International Journal of Mass Spectrometry, 2006 (In Print).
What About Statistics? Now we can apply these data sets to really any type of statistics that one might want to employ. But be careful, we have to insure that the peaks are real and not noise and not adducts of sodium, potassium, or matrix. This is very common and the mass spec person needs to be involved again at this point!
Evaluation of Peaks in Pairwise Analysis with a Std. Err Plot 0.0030 0.0015 0.0000 0.0030 0.0015 Healthy Group CaP Patients 0.0000 2000 4000 6000 8000 10000 12000 0.002 0.001 CaP m/z 2935, 2951 0.0009 0.0006 0.0003 CaP m/z 9129 0.000 2910 2925 2940 2955 2970 00075 00050 00025 CaP m/z 5913 0.0016 0.0008 9080 9120 9160 9200 CaP m/z 9420 00000 5760 5820 5880 5940 0.0000 9360 9390 9420 9450 9480
Hierarchical Clustering Analysis Carried out Statistically Differing Peaks from a Training Set M/Z CaP NL Top 280 data Points (36 Peaks) 6886 9420 9129 4567 4707 4592 4921 5913 3906 2951 3919 4032 2366 2349 2283 2935 7327 7194 7130 3565 2871 2384 8448 2675 5672 3666 8950 2978 6220 2437 2230 2513 2305 2532 2066 2786 56C 38C 64C 57C 37C 28C 62C 27C 71C 70C 9C 4C 50C 18C 17C 11C 8C 10C 79N 47N 48N 82N 61N 118N 9N 63N 44N 1N 11N 94N 93N 98N 64N 23N 119N 81N 97N 117N 62N 29N 28N 21N 78N 46N 66N 65N 80N 49N 115N CaP 2935 CaP
Clinical Serum Profiling Experiment; Example of Hierarchical Clustering Analysis Carried out on 168 Patient Sample Set 120 100 80 60 40 g y p CaP NL n=70 n=98 Normal Local CaP Metastatic CaP 20 0 --- -- --- --- --- -- -- -- --- -- --- -- -- --- --- -- -
Clinical Test Based on Specific Protein Peaks Diagnostic Efficiency as Determined through Class Prediction by Weighted Voting Scheme General Stats: N Median Age PSA > 4.0 PSA < 4.0 Normals 98 55 12 (4.2-7.8) 86 CaP 70 64 62 8 (1.9-3.6) Diagnostic Stats: # Variables N Sensitivity Specificity PPV NPV Non-Predicted 280 168 94.1 % 99.0 % 0.99 0.96 5 Sensitivity; TP/ (TP + FN) TP true positive Specificity; TN/ (TN + FP) TN true negative PPV; TP/ (TP + FP) FP false positive FN false negative NPV; TN/ (TN + FN) Note: Non-predicted values were not included in final calculations
Another Example: Biomarker Identification and Classification was Applied to a Single Section; (May be able to differentiate DCIS from IMC??) Cornett D.S., Mobley J.A., Dias, E.C., Andersson M., Arteaga C.L., Sanders M.E., Caprioli, R.M.; Histology Directed MALDI-MS Profiling Improves Throughput and Cellular Specificity in Human Breast Cancer, Journal of Molecular and Cellular Proteomics 2006 (in print).
What to Do with These Peaks? Top Down Proteomics Bottom up Peptidomics Medium Down Approaches Top Down Directed
What to do With these Markers? First and foremost validation using immunodirected or quantitative MS techniques must also be carried out. Mechanistic studies, i.e. knocking out a gene found to be involved in the disease process (LOF). This can be combined with global or directed stable isotope label studies in cell culture (i.e. SILAC, ITRAQ, ICAT).
Ex: HTP Validation of Novel Markers with Multiplex Bead Assays (Luminex( Luminex) Multiplex Bead Assay for cytokines The highlighted area represent populations of fluorescent beads, distinctively labeled, and carrying capture antibodies for sandwich assay of different cytokines. All detection antibodies carry the same fluorophore, which is read in a third channel to quantify sample cytokine concentration Bosch I. et. al. work in progress!
Quantitative Proteomics Stable Isotope Tags: (Mechanism Studies) ICAT (isotope coded affinity tag) SILAC (stable isotope labeled AA in cell culture) itraq (Isotope tags for relative and absolute and quantification) 18 O Digests (labels trypsin digested peptides at lysine and arginine) Many Other Chemical Tags
Conclusions?? Draw your own.. Just Another Tool in the Toolbox??? Too Soon to Know, Let s see where it takes us!
MSRC Richard Caprioli (Director) Shannon Cornett James Mobley Michelle Reyzer Jeremy Norris Robert Caldwell Erin Seeley Stacey Oppenheimer Sheerin Khatib-Shahid Pierre Chaurrand Kristin Burnum Kristen Herring Saturday Puolitaival Hans Reudi Aerni Greg Bowersock (IT) Ken Shriver Others Per Andren, Uppsala University Robert Weil, NIH Walter Korfmacher, Schering-Plough Markus Stoeckli, Novartis Acknowledgments Vanderbilt Collaborators Carlos Arteaga (Breat SPORE Dir.) David Carbone (Lung SPORE Dir.) Robert Coffey Jr. (GI SPORE Dir.) Marie-Claire Orgebin-Crist Ariel Deutch Dennis Hallahan Pat Levitt Robert Matusik Jason Moore Hal Moses Jennifer Pietenpol Ned Porter Yu Shyr Robert Whitehead Ken Schriver Funding NIH (NCI, GMS, NIDA) Vanderbilt University
Questions?