A Molecular Atlas of Ovarian Cancer Maps the Genomic Terrain Across 82 Tumor Samples

Memorial Sloan Kettering's SPECTRUM project profiles high-grade serous ovarian cancer at multiple body sites per patient, splitting the cohort into DNA-repair categories that determine which women might benefit from targeted therapies.

High-grade serous ovarian cancer kills more American women than any other gynecologic malignancy, in part because it is almost never caught early. By the time a surgeon opens the abdomen, the disease has usually seeded itself across the peritoneal cavity — omentum, pelvis, bowel surfaces — creating a scattered landscape of tumors that may look alike under the microscope but carry different molecular vulnerabilities. The question that shapes treatment is deceptively simple: can a patient's tumor repair its own damaged DNA, or can't it? The answer determines eligibility for a class of drugs called PARP inhibitors, which trap cancer cells in a lethal cycle of unrepaired breaks. But measuring DNA-repair status is not straightforward, and different assays sometimes disagree.

The MSK SPECTRUM atlas confronts that complexity head-on. Forty-two women with high-grade serous ovarian cancer (HGSOC) contributed 82 tumor specimens — many patients giving tissue from more than one anatomical site — all profiled by whole-genome sequencing at Memorial Sloan Kettering Cancer Center. The result is a richly annotated cohort in which every specimen carries not just a mutational profile but a surgical context, an anatomical address, a treatment history, and competing classifications of its DNA-repair machinery. What follows is a walk through the atlas's structure, its headline findings, and the quality-control checks that determine how far those findings can be trusted.

Who Is in the Cohort

The 42 patients span an age range consistent with the epidemiology of HGSOC, a disease that typically strikes women in their fifties and sixties. The age-at-diagnosis distribution peaks in exactly that window. But the tails matter too: any cluster of women diagnosed before age 45 raises the possibility of inherited BRCA1 or BRCA2 mutations — germline variants that would place those patients in the DNA-repair-deficient category from birth and could influence how their tumors behave in every downstream analysis.

Because many patients contributed multiple specimens, the same age can appear several times in a specimen-level histogram, inflating certain age bins. Comparing patient-level and specimen-level distributions is a basic but necessary sanity check: if one 55-year-old patient contributed five tissue samples, she should not be mistaken for a cluster of five separate patients.

Microsatellite Stability: Confirming the Expected

One of the first quality checks in any cancer genomics cohort is microsatellite instability, or MSI — a measure of whether the cell's mismatch-repair machinery is broken. In colon cancer, high MSI is common and clinically actionable, flagging patients for immunotherapy. In HGSOC, it is vanishingly rare. The SPECTRUM cohort confirms this: MSI scores cluster near zero across nearly all 82 specimens. Any specimen with an elevated score would be unusual enough to warrant individual review — it could signal a misclassified tumor type, a Lynch syndrome carrier, or simply a technical artifact. The overwhelmingly stable MSI profile here clears the way for interpreting mutational signatures without the confound of mismatch-repair deficiency.

Mapping the Body: Where the Specimens Come From

HGSOC does not stay in one place. It spreads through the peritoneal cavity with a characteristic pattern: primary tumors in the adnexa (ovary and fallopian tube), early metastases to the omentum — the fatty apron draping the intestines — and further deposits across the pelvis and peritoneum. The SPECTRUM atlas mirrors this biology in its sampling strategy, with specimens drawn from adnexal, omental, pelvic, and peritoneal supersites.

The balance matters. If omental specimens vastly outnumber adnexal ones, any cohort-wide molecular average becomes effectively an average of omental biology. And within each supersite, the mix of primary versus metastatic tissue determines which comparisons are possible. Adnexal sites should be overwhelmingly primary; omental and peritoneal sites should be metastatic. Deviations from that expectation would call the clinical annotations into question.

Treatment Context and Enrollment Arms

Not all specimens were collected under the same clinical circumstances. Some were taken at primary debulking surgery before any chemotherapy; others came from interval surgery after neoadjuvant treatment. This distinction is critical because platinum-based chemotherapy reshapes a tumor's genome — inducing new mutations, selecting for resistant clones, and potentially altering the apparent mutational signature. Mixing pre-treatment and post-treatment specimens without accounting for therapy context would confound every molecular comparison.

The atlas tracks treatment arm at both the patient and specimen level. Balanced arms are essential for powered comparisons; any arm with fewer than roughly ten patients should be treated as exploratory. Similarly, the split between surgical procedure types — debulking versus biopsy versus interval surgery — affects tissue quality and tumor cellularity in ways that ripple through every downstream measurement.

Platinum-based chemotherapy reshapes a tumor's genome — inducing new mutations, selecting for resistant clones, and potentially altering the apparent mutational signature.

The Headline Finding: DNA-Repair Status Splits the Cohort

The central question in HGSOC treatment today is whether a tumor's homologous recombination repair pathway is intact or broken. Homologous recombination deficiency — HRD — means the cell cannot accurately fix double-strand DNA breaks, leaving it vulnerable to PARP inhibitors like olaparib and niraparib. Roughly half of all HGSOC tumors are HRD, and the SPECTRUM cohort reflects that proportion.

But arriving at a definitive HRD or HRP (homologous recombination proficient) label is harder than it sounds. The atlas reconciles two independent measurement approaches: whole-genome sequencing–derived mutational signatures and the Myriad myChoice assay, which is the FDA-approved companion diagnostic for olaparib. Each method captures a different facet of genomic instability. Some patients are unambiguously HRD by both assays. Others fall into a gray zone where one test says yes and the other says no. The consensus classification in SPECTRUM adjudicates these discordances, providing a single label per patient — but the discordant cases are precisely the ones that deserve scrutiny in any clinical analysis.

The Mutational Landscape: TP53 and Beyond

If there is one near-universal truth about high-grade serous ovarian cancer, it is that the tumor suppressor gene TP53 is mutated in virtually every case. The SPECTRUM oncoprint — a grid showing which genes are altered in which patients — confirms this defining feature: TP53 tiles across nearly every column.

Beyond TP53, the landscape is sparser but revealing. BRCA1 and BRCA2 mutations, the best-known drivers of HRD, appear in a meaningful fraction of patients and are largely mutually exclusive — a patient carrying one rarely carries the other, consistent with their redundant roles in the same DNA-repair pathway. Other homologous recombination genes like RAD51C and PALB2 appear at lower frequencies. Patients with very few alterations outside TP53 likely harbor copy-number–driven tumors, a biology not fully captured by mutation-level analysis alone. At the other extreme, patients with unusually dense alteration patterns across many genes should be cross-referenced with tumor mutational burden and MSI status to rule out hypermutator phenotypes.

The nonsynonymous tumor mutational burden distribution fits the expected range for HGSOC: a peak around 3 to 5 mutations per megabase, with a modest right tail. Any specimen exceeding roughly 10 mutations per megabase is a candidate for closer inspection — it may reflect prior platinum exposure, mismatch repair deficiency, or a POLE mutation.

The SPECTRUM atlas does not end with a single answer. It ends with a framework — a way of asking precise questions about a disease that spreads widely and mutates unevenly. The cohort is large enough to split by DNA-repair status, by anatomical site, by treatment history, and still retain statistical traction in most strata. But the atlas also reveals its own limits: gray-zone patients whose repair status depends on which assay you trust, low-purity specimens from omental sites that may be hiding true mutations, and treatment arms whose small size constrains interpretation. These are not failures of the data; they are the honest boundaries of a 42-patient cohort confronting a disease with enormous molecular diversity. The next step is to layer on the atlas's other modalities — single-cell transcriptomics, spatial profiling, histological imaging — and ask whether the genomic categories defined here correspond to distinct tumor microenvironments. If they do, the molecular split between HRD and HRP may turn out to be not just a predictor of drug response but a window into fundamentally different ways that ovarian cancer organizes itself in the body.