Yingjie Zhu

Supplementary Figure 1: Characteristics of the IMPACT cohort. Violin plots of (A) age at sample acquisition and (B) age at patient’s first diagnosis separated by amplification status of MYCf genes. Bar plot showing (C) the stage of the cancer at the patient’s first diagnosis and (D) patient sex separated by MYCf amplification status. (E-G) Scatter plot showing the correlation between fraction of genome altered (FGA) and segment size for amplifications in (E) MYC, (F) MYCN, and (G) MYCL. On each plot is shown Pearson’s R and p-value. (H) Violin plots showing the clonality (cellular fraction divided by sample purity derived from FACETS) of MYCf amplifications.

Supplementary Figure 2: Expression of MYC and MYCL in various cancer types. (A) Boxplot of the expression of MYCL separated by cancer type. Cancer types were included if they contained at least five samples with MYCL amplifications. Each point represents one sample, and samples are separated by amplification status of MYCL. BH-adjusted p-values are shown. (B-C) Boxplots of ssGSEA for samples in (Figure 3A) for two Hallmark pathways: MYC Targets V1 (B) and MYC Targets V2 (C). Nominal p-values are shown when below 0.05. (D-H) Boxplots comparing expression of MYC across cancer types, separated by alteration status of five genes that were identified through cross-validation GLM-NET as correlating with MYC expression: CDKN2A, ARID4A, PIK3CA, TERT, and TP53, respectively. (I) Boxplots comparing expression of MYC in HR positive breast carcinoma, separated by amplification status of MYC and alterations in PIK3CA. For (A-I), expression is shown as natural log of RSEM, the size of each point represents the ploidy-corrected total copy number of MYC, and p-values are derived from Wilcoxon tests.

Supplementary Figure 3: Overall survival analysis for six cancer types. Multivariate analysis of overall survival for patients with MYC amplifications. Cancer types are as follows: (A) hormone receptor (HR) positive breast cancer, (B) colorectal cancer, (C) lung adenocarcinoma, (D), ovarian cancer, (E) prostate cancers, (F) and uterine cancer. Covariates include fraction of genome altered (FGA), patient sex, age at sample acquisition, age and stage at first diagnosis, and binary variables for focality of the amplification, the amplitude of the total copy number compared to the median, and alterations in cooccurring genes. Hazard Ratios (HR), 95% confidence intervals (CI) and nominal p-values are shown to the right. BH-adjusted p-values can be found in Supplementary Table 7.

Summarizes considerations relevant to the representativeness and generalizability of patient cohorts in studies of the cancer types evaluated.

Number of samples from TCGA cohort used in expression analysis separated by type of MYCf amplification. If cancer type was not used in a comparison, values appear as NA.

Values derived from the cross-validated GLM-NET modeling expression by ploidy-corrected copy number, cancer type, and genes with alterations co-occurring with MYC amplifications across cancer types in TCGA samples.

Values derived from the cross-validated GLM-NET modeling expression by ploidy-corrected copy number, and genes with alterations co-occurring with MYC amplifications in each of six cancer types in TCGA samples.

Comparison of focal and broad MYC amplification frequency across the most common cancer types in Figure 1A. For each cancer type, the ratio was calculated as the number of focal divided by the number of total MYC amplifications. The background, expected ratio was based on all cancer types in Figure 1A, excluding the cancer type in question. Binomial test was used for comparison with nominal and BH-corrected p-values shown.

Co-occurrence and mutual exclusivity of MYCf genes as depicted in Figure 2. Columns include gene or gene focality from Figure 2A and B, respectively, alteration being tested, the number of samples per alteration combination, odds ratio (OR), q-value, and cancer type.

Results from modeling overall survival in each of seven cancer types from Figure 3D and Supplementary Figure 3. Variables include sex, age of sample acquisition, stage and age at first diagnosis, fraction of genome altered (FGA), and genes with commonly occurring alterations in a given cancer type. The coefficient is derived from each variable in the Cox regression model. The p-value for each variable was BH-adjusted in column “padj”. The “likelihoodP” and “logrankP” are the p-values for the entire model for each cancer type derived from the likelihood ratio test and log rank test, respectively. The total shows the number of samples included in the model.