Expanded View Figures

Transcription

1 Molecular Systems iology Tumor CNs reflect metabolic selection Nicholas Graham et al Expanded View Figures Human primary tumors CN CN characterization by unsupervised PC Human Signature Human Signature Pan-Cancer & Cross-Species Consistency Consistently altered regions: elimination of passenger genes via synteny mapping - human core tumors - human expanded Figure EV1. Integration of patient and model system CN in a pan-cancer and cross-species bioinformatic and experimental analysis. Overall workflow for the analyses described in the Results section. EV1 Molecular Systems iology 13: ª 217 The uthors

2 Nicholas Graham et al Tumor CNs reflect metabolic selection Molecular Systems iology Figure EV2. Principal component analysis of TCG copy number data reveals a shared CN signature enriched for p3 mutation (related to Fig 1). C Copy number profiles of 39 lung (LU) squamous carcinomas (LUSC) and 368 lung adenocarcinomas (LUD) (), 83 ovarian serous cystadenocarcinoma (OV) (), and 492 uterine corpus endometrial carcinomas (UCEC) (C) tumors sorted by their PC1 score. Signature tumors are sorted to the top of the heatmap. PC1-ranked tumors were analyzed for enrichment of known mutations or molecular subtypes (e.g., basal/luminal in breast, see D for all subtypes tested). The red horizontal bars to the right of each tumor type indicate the tumor subtype (e.g., LUD), or the presence of the indicated mutation (e.g., TP3). Normalized enrichment score and permutation P-value for each significantly enriched mutation or clinical subtype are shown below the tick plots. D Enrichment analysis of molecular subtypes in PC-defined signatures and. The molecular subtypes tested were defined by the TCG (The Cancer Genome tlas Research Network, 211, 212a,b, 213, 214). ecause LUD and LUSC were combined into a single dataset LU, all LUSC and LUD subtypes were associated with signatures and, respectively. Within LU, signature was most associated with the classical subtype of LUSC, and signature was most associated with the bronchioid subtype of LUD. Permutation-based P-values are indicated. E mplification of MDM2 and deletion of CDKN2 are mutually exclusive in signature human RC. Each point represents a single patient sample. Samples are colored red or blue according to their membership in signatures or, respectively, with intermediate colors representing intermediate PC1 scores between signatures and. Samples colored white represent nearly diploid tumors, as indicated by the triangle in Fig 1. ª 217 The uthors Molecular Systems iology 13: EV2

3 Molecular Systems iology Tumor CNs reflect metabolic selection Nicholas Graham et al LU LUSC LUD TP3 CDKN2 NFE2L2 MUC4 HRS PTEN PIK3C R1 SLC178 RM MET EGFR STK11 KRS OV Proliferative Mesenchymal TP3 TM RC1 Principal Component 1-1. C Principal Component 1 Log2 ratio 1. UCEC Normalized enrichment score Serous TP3 DCF8L2 PIK3C MUC4 CDC27 RC2 RC1 NF1 PC MTOR SETD2 TPTE SI Principal Component 1-1. Log2 ratio RM POLE CSP8 TRX MP3K1 NOTCH1 TM PIK3R1 KRS CTNN1 PTEN 1. Normalized enrichment score Log2 ratio 1. Normalized enrichment score D Molecular sub-type NES Signature E RC RC basal RC claudin-low RC HER2-enriched RC luminal RC luminal RC luminal LUSC classical LUSC primitive LUSC basal LUSC secretory LUD magnoid LUD squamoid LUD bronchioid CDKN2 CN 2 OV proliferative OV immunoreactive OV differentiated OV mesenchymal MDM2 CN UCEC copy number high UCEC ultra-mutated UCEC hyper-mutated UCEC copy number low Figure EV2. EV3 Molecular Systems iology 13: ª 217 The uthors

4 Nicholas Graham et al Tumor CNs reflect metabolic selection Molecular Systems iology Figure EV3. Hierarchical clustering confirms the existence of the shared pan-cancer CN signatures across multiple tumor types, and distinct signature subtypes within tumor types (related to Fig 1). Pan-cancer clustering of 3,37 copy number profiles using a balanced, random sampling of tumors from 1 tumor types. The dendrogram reveals the presence of a cluster containing multiple tumor types and enriched in PC-defined signature tumors (hypergeometric P-value = for concordance between the PC and hierarchical clustering results). There is also a cluster enriched in the generally less uniform PC-defined signature tumors (hypergeometric P-value = ), which contains other tumor types as well (e.g., SKCM). In the key, the individual tumor types and pan-cancer PC1 scores are indicated. dditionally, for the core signature tumors (OV, RC, UCEC, and LU), tumor-specific subtypes (e.g., RC basal and luminal) and tumor type-specific PC1 scores are indicated. The gene loci CN levels in the heatmap are ordered based on their chromosomal locations (columns). ª 217 The uthors Molecular Systems iology 13: EV4

5 Molecular Systems iology Chr Low CN / Diploid Signature enriched Core Signature enriched Type LU PC1 OV PC1 RC PC1 UCEC PC1 Subtype OV RC UCEC LU LC RED SKCM COD HNSC STD GM PRD LGG THC KIRC Pan PC1 Pan-Cancer Nicholas Graham et al Tumor CNs reflect metabolic selection KIRC GM (LGG) 1q mp LGG COD RED OV Subtypes Proliferative Mesenchymal Other RC Subtypes UCEC Subtypes LU Subtypes Signature asal Luminal Other Serous Non-serous LUSC LUD Sig Sig Pan PC1-2 4 RC PC1 OV PC UCEC PC1 LU PC Log2 ratio 2 Figure EV3. EV Molecular Systems iology 13: ª 217 The uthors

6 Nicholas Graham et al Tumor CNs reflect metabolic selection Molecular Systems iology Figure EV4. CN-defined signature tumors exhibit RN expression signatures consistent with increased glycolytic activity (related to Fig 3). KEGG metabolism pathway enrichment analysis based on mrn expression in RC and LU signature and signature tumors. RN expression data are from RC and LU tumors with paired TCG DN and RN data. To evaluate the similarity to RC tumors with high FDG-PET uptake, we included a gene set consisting of genes from the glycolysis and pentose phosphate pathway that were upregulated in FDG-high RC tumors (Palaskas et al, 211). To combine RC and LU into one enrichment analysis, genes were ordered by their average rank in RC and LU tumor types. Enrichment scores for all KEGG metabolism pathways and for individual tumor types can be found in Table EV4. D RN expression signatures of high glycolysis (Palaskas et al, 211) were used to predict the glycolytic phenotype of RC (), LU (C), and OV (D) tumors with paired TCG DN and RN data. UCEC tumors were excluded due to a lack of sufficient samples with paired RN and DN data. The color scale was normalized for each tumor type by the mean FDG score for tumors with integrated CN score >.2. Non-normalized values are shown in the color legend and in (E). E RN-based WGV predictions of glycolysis show that RC and LU signature tumors (top % PC1) are predicted to be more glycolytic than the tissue typematched signature tumors (bottom % PC1). Mann Whitney U-test P-values are shown. Data are presented in box (median, first and third quartiles) and whisker (extreme value) plots. F Spearman correlation values between RN-based WGV glycolysis predictions and CN-based PC1 values, calculated for different range windows of integrated CN levels to control for the general increase in glycolysis predictions at increased levels of genomic instability. P-values shown are calculated by performing 1 6 permutations where the WGV scores were randomly shuffled, and the P-value was calculated as the percentage of random permutations that achieved a higher correlation value for any range window than did the experimentally observed relationship. ª 217 The uthors Molecular Systems iology 13: EV6

7 Molecular Systems iology Tumor CNs reflect metabolic selection Nicholas Graham et al KEGG metabolism pathway gene set (n = 76) Permuta on FDG-PET High RC Palaskas et al Pyrimidine metabolism Pentose phosphate pathway Glycolysis & Pentose phosphate pathway Glycosphingolipid biosynthesis - lacto and neolacto series Cysteine and methionine metabolism Core glycolysis Purine metabolism NES verage gene rank RC+LU RN Sig v. Sig FDR q-value RC C LU D OV..4 Integrated CN score PC1 2 2 PC1 2 2 PC RN-based FDG score RN-based FDG score RN-based FDG score E F RN-based FDG score Spearman rho values RC (p < 1 x -6 ) LU (p < 1 x -6 ) OV (p =.) Tumor type: : RC LU OV 2.4 x x Range of Integrated CN score Figure EV4. EV7 Molecular Systems iology 13: ª 217 The uthors

8 Nicholas Graham et al Tumor CNs reflect metabolic selection Molecular Systems iology Figure EV. Metabolic patterns are predictive of increased glycolysis in MEF lines (related to Fig 6). In cross-prediction tests, signature -based predictions of signature MEF line metabolic phenotypes perform better than signature -based predictions. Glucose consumption was measured using a bioanalyzer. In cross-prediction tests, PC1 scores of signature MEFs are more predictive of average growth fold change in signature MEFs, while scores from signature MEFs are more predictive in signature MEFs. C G Mass spectrometry-based metabolite measurements. Samples are arranged left to right from the strongest signature (blue) PC1 scores to the strongest signature PC1 (red) scores. Sample D (purple) is a mixed signature / line. (C) Relative intracellular metabolite levels. (D) Isotopomer distributions for intracellular metabolites (M, monoisotopic molecular weight). (E) Percent label for intracellular metabolites defined as percent of metabolite molecules with isotopomer mass greater than the monoisotopic molecular weight. Note, the high to low range for percent metabolite molecules with incorporated label varies for each metabolite and generally does not extend from to %. (F) Relative metabolite consumption from and secretion into media normalized to initial media levels (i.e., cellular metabolite footprint). Here, positive and negative values indicate metabolite secretion into and consumption from the media, respectively. In (D G), the correlation of each metabolic parameter with glucose consumption (as measured by a bioanalyzer, and as a positive variable, Fig 6) is indicated (Pearson correlation, q). In the isotopomer cases (D), the summary metric of percent label is used for the correlation determination. Notably, cells with high glucose consumption rates tend to exhibit high consumption of serine, glycine, and glutamine; and high secretion of lactate (with concordant lactate secretion results obtained by bioanalyzerbased measurement of the same spent cell culture media). (G) Pathway schematics of differences in relative media levels (footprint, from F) between signature and signature MEFs. Each node is colored according the signal-to-noise (SNR) ratio. Red indicates higher post-culture media levels (lower consumption or higher secretion rates) in signature MEFs, and blue indicates higher post-culture media levels in signature MEFs. Error bars indicate standard deviations of biological replicates. Footprinting data were normalized to the integrated cell number of the time course of labeling (24 h), and intracellular metabolite concentrations were normalized to the number of cells present at the time of extraction. Full metabolomic data can be found in Table EV. ª 217 The uthors Molecular Systems iology 13: EV8

9 Molecular Systems iology Tumor CNs reflect metabolic selection Nicholas Graham et al Correlation with average fold change cell number Training set C. D E F G Figure EV. EV9 Molecular Systems iology 13: ª 217 The uthors