A generalizable machine learning framework for classifying DNA repair defects using ctDNA exomes.

Specific classes of DNA damage repair (DDR) defect can drive sensitivity to emerging therapies for metastatic prostate cancer. However, biomarker approaches based on DDR gene sequencing do not accurately predict DDR deficiency or treatment benefit. Somatic alteration signatures may identify DDR deficiency but historically require whole-genome sequencing of tumour tissue. We assembled whole-exome sequencing data for 155 high ctDNA fraction plasma cell-free DNA and matched leukocyte DNA samples from patients with metastatic prostate or bladder cancer. Labels for DDR gene alterations were established using deep targeted sequencing. Per sample mutation and copy number features were used to train XGBoost ensemble models. Naive somatic features and trinucleotide signatures were associated with specific DDR gene alterations but insufficient to resolve each class. Conversely, XGBoost-derived models showed strong performance including an area under the curve of 0.99, 0.99 and 1.00 for identifying BRCA2, CDK12, and mismatch repair deficiency in metastatic prostate cancer. Our machine learning approach re-classified several samples exhibiting genomic features inconsistent with original labels, identified a metastatic bladder cancer sample with a homozygous BRCA2 copy loss, and outperformed an existing exome-based classifier for BRCA2 deficiency. We present DARC Sign (DnA Repair Classification SIGNatures); a public machine learning tool leveraging clinically-practical liquid biopsy specimens for simultaneously identifying multiple types of metastatic prostate cancer DDR deficiencies. We posit that it will be useful for understanding differential responses to DDR-directed therapies in ongoing clinical trials and may ultimately enable prospective identification of prostate cancers with phenotypic evidence of DDR deficiency.

TP53 Alterations Are Associated With Poor Survival in Patients With Primary Mediastinal Nonseminoma Germ Cell Tumors.

BACKGROUND: Primary mediastinal nonseminoma germ cell tumors (PMNSGCT) are a subgroup of nonseminoma germ cell tumors (GCT) with poor prognosis. In this study, PMNSGCT-specific genomic landscape was analyzed and correlated with clinical outcomes. METHODS: DNA was extracted and sequenced from 28 archival tumor tissue of patients with mediastinal GCT (3 seminoma and 25 nonseminoma). Overall survival (OS) and association with gene alterations were estimated using the Kaplan-Meier and univariate Cox regression methods. RESULTS: Three patients (11%) had a karyotype XXY, 17/28 (61%) tumor samples presented chromosome 12p amplification. Somatic mutations were detected in 19/28 (68%) samples. The most frequently mutated genes were: TP53 (13/28; 46%), KIT (5/28; 18%), and KRAS (5/28; 18%). Deleterious TP53 alterations were associated with significantly reduced overall survival (HR: 7.16; P = .012). CONCLUSIONS: TP53 alterations are common in PMNSGCT and are associated with reduced overall survival, potentially underlying the poor sensitivity to chemotherapy observed in these patients.

Deep whole-genome ctDNA chronology of treatment-resistant prostate cancer.

Circulating tumour DNA (ctDNA) in blood plasma is an emerging tool for clinical cancer genotyping and longitudinal disease monitoring1. However, owing to past emphasis on targeted and low-resolution profiling approaches, our understanding of the distinct populations that comprise bulk ctDNA is incomplete2-12. Here we perform deep whole-genome sequencing of serial plasma and synchronous metastases in patients with aggressive prostate cancer. We comprehensively assess all classes of genomic alterations and show that ctDNA contains multiple dominant populations, the evolutionary histories of which frequently indicate whole-genome doubling and shifts in mutational processes. Although tissue and ctDNA showed concordant clonally expanded cancer driver alterations, most individual metastases contributed only a minor share of total ctDNA. By comparing serial ctDNA before and after clinical progression on potent inhibitors of the androgen receptor (AR) pathway, we reveal population restructuring converging solely on AR augmentation as the dominant genomic driver of acquired treatment resistance. Finally, we leverage nucleosome footprints in ctDNA to infer mRNA expression in synchronously biopsied metastases, including treatment-induced changes in AR transcription factor signalling activity. Our results provide insights into cancer biology and show that liquid biopsy can be used as a tool for comprehensive multi-omic discovery.

Somatic Features of Response and Relapse in Non-muscle-invasive Bladder Cancer Treated with Bacillus Calmette-Guérin Immunotherapy.

BACKGROUND: High-risk non-muscle-invasive bladder cancer (NMIBC) is treated with bacillus Calmette-Guérin (BCG), but relapse is common. Improvement of patient outcomes requires better understanding of links between BCG resistance and genomic driver alterations. OBJECTIVE: To validate the prognostic impact of common genomic alterations in NMIBC pretreatment and define somatic changes present in post-BCG relapses. DESIGN, SETTING, AND PARTICIPANTS: We retrieved tumour tissues and outcomes for 90 patients with BCG-naive NMIBC initiating BCG monotherapy. Post-BCG tissue was available from 34 patients. All tissues underwent targeted sequencing of tumour and matched normal. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Associations between clinical outcomes and genomics were determined using Cox proportional hazard models. RESULTS AND LIMITATIONS: Of the patients, 58% were relapse free at data cut-off, 24% had NMIBC recurrence, and 18% experienced muscle-invasive progression. The risk of relapse was associated with ARID1A mutation (hazard ratio [HR] = 2.00; p = 0.04) and CCNE1 amplification (HR = 2.61; p = 0.02). Pre- and post-BCG tumours shared truncal driver alterations, with mutations in TERT and chromatin remodelling genes particularly conserved. However, shifts in somatic profiles were common and clinically relevant alterations in FGFR3, PIK3CA, TSC1, and TP53 were temporally variable, despite apparent clonal prevalence at one time point. Limitations include the difficulty of resolving the relative impact of BCG therapy versus surgery on genomics at relapse and biopsy bias. CONCLUSIONS: Somatic hypermutation and alterations in CCNE1 and ARID1A should be incorporated into future models predicting NMIBC BCG outcomes. Changes in tumour genomics over time highlight the importance of recent biopsy when considering targeted therapies, and suggest that relapse after BCG is due to persisting and evolving precursor populations. PATIENT SUMMARY: Changes in key cancer genes can predict bladder cancer relapse after treatment with bacillus Calmette-Guérin. Relapses after treatment can be driven by large-scale genetic changes within the cancer. These genetic changes help us understand how superficial bladder cancer can progress to be treatment resistant.

Plasma Circulating Tumor DNA and Clonal Hematopoiesis in Metastatic Renal Cell Carcinoma.

BACKGROUND: There is a lack of molecularly-informed biomarkers for patients with metastatic renal cell carcinoma (RCC). Plasma cell-free DNA (cfDNA) sequencing is a minimally-invasive alternative to tissue for profiling the genome in other cancers but relevance in metastatic RCC remains unclear. MATERIALS AND METHODS: Whole blood was collected from 55 patients with metastatic RCC. Plasma cfDNA and leukocyte DNA were subjected to targeted sequencing across 981 cancer genes. Matched tumor tissue from 14 patients was analyzed. RESULTS: Thirty-three percent of patients had evidence for RCC-derived circulating tumor DNA (ctDNA), significantly lower than patients with metastatic prostate or bladder cancer analyzed using the same approach. Among ctDNA-positive patients, ctDNA fraction averaged only 3.9% and showed no strong association with clinical variables. In these patients, the most commonly mutated genes were VHL, BAP1, and PBRM1, and matched tissue concordance was 77%. Evidence of somatic expansions unrelated to RCC, such as clonal hematopoiesis of indeterminate potential, were detected in 43% of patients. Pathogenic germline mutations in DNA repair genes were detected in 11% of patients. CtDNA-positive patients had shorter overall survival and progression-free survival on first-line therapy. Patients with evidence of clonal hematopoiesis of indeterminate potential had an intermediate prognosis compared with ctDNA-positive and -negative patients. CONCLUSIONS: CfDNA sequencing enables straightforward characterization of the somatic RCC genome in a minority of patients with metastatic RCC. Owing to low ctDNA abundance, and the presence of non-RCC derived somatic clones in circulation, cfDNA sequencing may not be a simple pan-patient alternative to tissue biopsy in metastatic RCC.

Immunocytochemistry for ARID1A as a potential biomarker in urine cytology of bladder cancer.

BACKGROUND: Mutations of AT-rich interactive domain 1 (ARID1A) have been associated with a worse outcome after intravesical treatment with bacille Calmette-Guérin in patients with non-muscle-invasive bladder cancer (NMIBC). Loss of ARID1A protein expression in urine cytology may serve as an indication of an ARID1A mutation. Therefore, the authors examined the expression of ARID1A in urine cytology and histological specimens of bladder cancer for correlation with ARID1A mutational status. METHODS: The authors constructed a tissue microarray containing samples from 164 tissue samples from 150 patients with NMIBC and 100 tissue samples from 81 patients with muscle-invasive bladder cancer. A second cohort consisted of archived cytological specimens and matched tissue sections from 62 patients with high-grade NMIBC. The authors established immunohistochemistry and immunocytochemistry (ICC) protocols, respectively, for the analysis of ARID1A protein expression in histological and cytological specimens. Confirmatory next-generation sequencing (NGS) was performed on tumor specimens using a targeted NGS panel containing all exonic regions of ARID1A. RESULTS: The prevalence of ARID1A loss of expression on the tissue microarray was 3.6% in NMIBC (6 of 164 tissue samples) and 10% in muscle-invasive bladder cancer (10 of 100 tissue samples) (P = .059). Loss of ARID1A expression in cytology was concordantly immunohistochemistry negative in 6 of 8 matched tissue specimens. NGS confirmed an ARID1A mutation on all 6 histology samples with loss of ARID1A expression. When NGS demonstrated an absence of ARID1A mutation, histology was concordantly positive (16 of 16 cases). CONCLUSIONS: The authors have suggest ARID1A ICC as a promising surrogate marker for ARID1A mutational status in patients with urothelial carcinoma. Pitfalls in ICC scoring include benign umbrella cells that often are negative for ARID1A. Further prospective studies are needed to determine the clinical relevance of ARID1A ICC in urinary cytology.

Frequent mutation of the FOXA1 untranslated region in prostate cancer.

Prostate cancer has a low somatic mutation rate but non-coding regions remain underexplored. We sequenced the untranslated regions (UTRs) of 72 established driver genes in 428 patients with metastatic prostate cancer and identified FOXA1 3'-UTR mutations in 12% of patients. The mutations were predominantly insertions or deletions, covered the entire UTR without motif enrichment, and were not detected in other cancers. FOXA1 lies in head-on orientation with the androgen-regulated non-coding gene AL121790.1, resulting in strong prostate lineage-specific bidirectional transcription across the FOXA1 3'-UTR. This suggests transcriptional activity as a cause for the localized hypermutation. The indel-dominant pattern of somatic mutation extends into the FOXA1 coding region, where it is shaped by clonal selection to yield a cluster of non-frameshift indels inside the forkhead domain. Somatic FOXA1 3'-UTR mutations may prove useful for diagnostic and screening approaches, given their high frequency and lineage specificity.