Scott A. Tomlins

Purpose: We sought to determine whether combining prostate health index (Phi) with urinary prostate cancer antigen 3 (PCA3) and TMPRSS2:ERG (T2:ERG) could improve selection of men for prostate biopsy. These biomarkers have been validated in prostate cancer (PCa) detection separately, but their combination has not previously been developed. Materials and Methods: Prebiopsy blood and post–digital rectal examination urine specimens were assayed to predict subsequent biopsy outcomes from training and validation cohorts (1073 participants across 11 academic centers). Clinical algorithms for combining Phi and PCA3-T2:ERG to predict Grade Group ≥ 2 (GG ≥ 2) PCa were formulated using the training cohort (N = 512). Prediction rules and hypotheses were locked before validation using biopsy-naïve men from the NCI Early Detection Research Network urinary PCA3 trial (N = 561). Rules were compared in weighted sum of specificity and sensitivity with weights specified a priori, and P values were obtained through bootstrap in the validation study. Results: Primary validation analysis showed that Phi combined with urinary PCA3 outperformed Phi alone ( P = .002). Furthermore, serum Phi combined with urinary PCA3-T2:ERG outperformed urinary PCA3-T2:ERG in each of the 3 algorithms reflecting different potential clinical workflows: (1) serum Phi and urine PCA3-T2:ERG tested simultaneously, either exceeding its own threshold ( P = .04); (2) urine PCA3-T2:ERG first and those in the grey zone resolved by subsequent serum Phi ( P = .03); and (3) serum Phi first and those in the grey zone resolved by subsequent urine PCA3-T2:ERG ( P = .002). Conclusions: Combining serum Phi with urinary PCA3 RNA alone or together with urinary T2:ERG RNA, simultaneously or sequentially, improves selection of men for initial prostate biopsy and represents an avenue to improve early detection of aggressive PCa.

Clinically relevant mutations and copy number alterations for HGSC and CRC cohorts

Comparison of aneuploidy between our HGSC models and other models

MiMouse for CGP and genomic credentialing across the neoplastic spectrum of two large cohorts of inducible fallopian tube (ovary) and colorectal (colorectal carcinoma) tumors. A, Promoters and transgenes used in the cohorts. Ovary and colorectal carcinoma tumor cohorts tested by MiMouse. Genes in bold were activated, and those in regular were inactivated; Trp53 (bold gray) was either inactivated through deletion or a point mutation was expressed (see main text for specific models). Abbreviations for genotypes (as used in the heatmap) are indicated. B, MiMouse cohort characteristics and CGP results from the ovary (predominantly HGSC; top, total n = 172) and colorectal carcinoma (bottom, total n = 100) cohorts are shown. Genotype, histology (CRC, carcinoma; eHGSC, early HGSC; HP, hyperplastic; Met., metastasis; MMMT, carcinosarcoma) and model type (GEMM, GEMM tumor tissue) are shown. Gene-level FGA (see main text) is shown according to the color scale (gray indicates no value due to QC failure, see Supplementary Fig. S9). All detected somatic, prioritized, amplifications (red), deep (homozygous) deletions (blue; none detected), and prioritized somatic mutations (black) across the cohorts are shown. C and D, CN plots of two colorectal carcinoma samples with focal amplifications in Myc or Pik3ca (indicated by arrows) are shown. The log2 CNR (vs. pseudo-normal) of individual amplicons are plotted, with genes (different colors) in genome order (chrs on bottom of plot) and gene-level log2 CNR indicated by black bars (black bars). E, CN plots of two histologically distinct HGSC tumors arising from both ovaries of the same mouse (1628LOT_X5 = left; MG1_X33 = right). Shared Myc amplifications are indicated. The remaining somatic CN profile (including gains in Arid1a and Csmd3 and loss of Mtor) is consistent with clonal origin [Trp53 (gray), which includes the engineered flox amplicons lost after induction and is shown for estimation of relative tumor content]. CRC, colorectal carcinoma.

Aneuploidy comparison between HGSC and CRC models

Highly correlated detection of focal, sub‐gene deep (homozygous) deletions by MiMouse

Aneuploidy and admixture detection considerations for mouse CGP by MiMouse. A, To simulate the effects of reduced coverage by CGP vs. near genome-wide technologies, we developed a metric, Rc, in which c is the breadth of coverage for a chr arm (gold) centered at the midpoint of the chr arm, and Rc is the proportion of the chr arm altered (hatched maroon region) by a CNA (maroon) specific to c. The resulting Rc for the same CNA measured by assays with different c (R1.0 = near-genome wide) is shown. Centromeres are indicated by an X. B, Dotplot of PPA for aneuploidy detection across human tumors (n = 6,591 diploid tumors from TCGA) by Rc thresholds for detecting aneuploidy, with variable c (x-axis) and Rc thresholds (color gradient per scale). C, Pearson correlation (r) between varying Rc thresholds (x-axis) and PPA of aneuploidy detection for autosomal aneuploidy arms (TCGA data). D, Ideogram [Mus musculus assembly GRCm38 (mm10)] of autosomal MiMouse genic targets (green) and SNPs (gold). Centrosomes are indicated. E, To determine a minimal set of SNPs capable of distinguishing five common mouse strains (see legend) from the 1,638 identified by Petkov and colleagues (as discriminatory of more than 100 strains; ref. 69), we performed factor analysis. Biplots (left) of the dimensions from the factor analysis (factor scores plotted) are shown, along with structure plots (right) showing in silico pairwise crossing of the mouse strains of interest and the resulting ADMIXTURE strain estimations (mean) using reduced sets of 1,000 and 500 SNPs from factor loadings.

Focused NGS panels overestimate FGA vs. CGP and WES in human HGSC

Table of sequencing statistics, sample information, and strain information for colorectal (CRC) and high grade serous cancer (HGSC) samples

A, Cross-species analysis incorporating synteny prioritizes regions for driver gene localization in colorectal carcinoma. To identify colorectal carcinoma specific, we compared arm-level gains (aneuploidy) in human TCGA colorectal carcinoma samples vs. all (n = 23) other tumor types. Boxplots of FDR (Q value)-corrected statistical significance (negative log Q values from Fisher exact test) of arm-level gains (left) and co-occurring aneuploidy events (right) in colorectal carcinoma vs. other tumor types. Those in bold were more frequent in colorectal carcinoma than at least 50% of other tumor types. B, Utilization of synteny to prioritize regions of aneuploidy for driver gene identification. Through WGS of human colorectal carcinoma, Cornish and colleagues identified three “peaks” (from 13q12.2-33.3) in the recurrent 13q aneuploidy event. Circos plot showing that due to synteny, this human region (peach) maps to five different mouse (lavender) chrs (1, 3, 5, 8, and 14; syntenic mapping colored separately). Three scenarios could be expected in mouse models of human colorectal carcinoma–specific/enriched arm-level gains (example for human 13q is shown): gain of all syntenic chrs if genes across the arm are functionally significant (scenario 1); gain of only a subset of syntenic chrs (if only one or a limited number of genes are functionally significant; scenario 2); or no gain of any syntenic chrs (if that event is not conserved in the mouse model). C, To prioritize potential driver genes from peak regions and the fidelity of human-specific colorectal carcinoma aneuploidy gains (from A) to MiMouse-identified aneuploidy gains in our colorectal carcinoma tumors (Fig. 5C), we performed permutation tests to evaluate scenarios 1–3. Circos plots for each human chr (peach) show all syntenic mouse chromosomes (lavender). The percentage of synteny of each mouse chr to the human chr is shown inside the Circos plot. The aneuploidy (gain) frequency of each syntenic mouse chr in colorectal carcinoma samples is shown outside the Circos plot; those significant (from Fig. 5C) are shown in bold with syntenic mapping in black. By permutation testing, only scenario 2 for mouse chr 5 (syntenic chr 7p, 7q, and 13q) was found to be more likely than scenario 1 or 3 (scenarios 1 and 3 were equally likely). Candidate driver oncogenes (from colorectal carcinoma or pan-tumor studies as reported by Cornish and colleagues ; ref. 46), as shown in yellow, were prioritized by intersection of the syntenic human blocks with peaks of CN gains (from these chr) in the WGS study of >2,000 human colorectal carcinoma. Most notably, CDX2 (human 13q.12.2), which maps to the gained region of mouse chr 5, is both a reported colorectal carcinoma lineage-specific oncogene but is also the source of the promoter (Cdx2-P) used to drive transgene expression in our colorectal carcinoma models. CRC, colorectal carcinoma.