In 181 Uterine Tumors With Broken DNA Repair, a Mutational Flood That May Explain Why Immunotherapy Works

A genomic atlas of mismatch repair–deficient endometrial cancers at Memorial Sloan Kettering reveals the scale and shape of hypermutation — and the driver genes it leaves in its wake.

Every time a human cell divides, its DNA replication machinery makes mistakes — a wrong base here, a slipped repeat there. Normally, a set of proteins called the mismatch repair system catches and corrects those errors before the next division. When that system fails, mistakes accumulate at extraordinary speed. The result is a hypermutated genome: hundreds or thousands of mutations scattered across the tumor's DNA, many of them in the short repetitive sequences called microsatellites.

Endometrial cancer — cancer of the uterine lining — is the most common gynecologic malignancy in the United States, and roughly a quarter to a third of cases harbor mismatch repair deficiency (MMRD). That deficiency matters clinically: it is one of the strongest predictors of response to immune checkpoint inhibitors, drugs that unleash the patient's own immune system against the tumor. But not all MMRD tumors behave the same way. Some arise from inherited Lynch syndrome mutations, others from acquired somatic hits to repair genes, and still others from epigenetic silencing. Whether those different origins produce different mutational landscapes — and therefore different therapeutic vulnerabilities — is a question that requires looking at a large, well-annotated cohort gene by gene, mutation by mutation. That is what 181 patients profiled by MSK-IMPACT targeted sequencing at Memorial Sloan Kettering now make possible.

Confirming the Baseline: Who Is in the Cohort

Before interrogating the mutational landscape, the first task is confirming that the cohort is what it claims to be. All 181 patients carry mismatch repair–deficient endometrial cancer, but mismatch repair deficiency and microsatellite instability are assessed by different assays — immunohistochemistry (IHC) for protein loss versus PCR or next-generation sequencing for microsatellite shifts. If the two methods disagree in a substantial fraction of cases, every downstream analysis needs to account for that discordance.

The MSI status distribution across the cohort is the first checkpoint. In a cohort selected for MMR protein loss, the overwhelming majority of samples should score as MSI-High (MSI-H) on molecular testing. Any MSI-Low or microsatellite-stable (MSS) cases raise immediate questions: was the IHC read incorrectly, or does the molecular assay miss a subset of MMRD? Conversely, if the cohort is uniformly MSI-H, the downstream TMB and variant-type analyses can proceed without that confounder.

Histology and Biomarker Testing: Setting the Subgroup Boundaries

Endometrial cancer is not one disease. The dominant histologic subtype — endometrioid carcinoma — tends to be lower-grade and more strongly associated with MMRD than its rarer cousins, serous and clear-cell carcinoma. A cohort built around MMR deficiency should therefore skew heavily toward endometrioid tumors. If serous or clear-cell cases appear in meaningful numbers, they may represent Lynch syndrome–associated disease or sporadic MMR loss in aggressive subtypes, both of which behave differently and respond differently to treatment.

The histology breakdown confirms the expected dominance of endometrioid cases. But histology alone does not tell the full story. The method by which each tumor's MMR deficiency was detected — IHC protein loss, molecular MSI testing, or a combination — shapes what downstream comparisons are valid. A cohort tested almost entirely by one method limits the ability to cross-validate results.

The Mutational Flood: Tumor Mutational Burden in MMRD Endometrial Cancer

With the cohort composition established, the central question becomes quantitative: how many mutations do these tumors carry? Tumor mutational burden (TMB) — measured as nonsynonymous mutations per megabase of sequenced DNA — is the metric that links MMRD biology to immunotherapy eligibility. The FDA-approved TMB threshold for pembrolizumab is 10 mutations per megabase. MMRD endometrial cancers are expected to cluster well above that line.

The TMB histogram across all 181 cases confirms a right-shifted distribution consistent with hypermutation. Most tumors land in the range of tens of mutations per megabase, with a tail stretching toward extreme values that may reflect concurrent POLE exonuclease-domain mutations — a second, independent hypermutation engine. Importantly, the left tail is thin: very few MMRD tumors fall near the low-TMB zone, reinforcing the idea that loss of mismatch repair reliably produces the mutational load thought to generate neoantigens and attract immune recognition.

Loss of mismatch repair reliably produces the mutational load thought to generate neoantigens and attract immune recognition — very few MMRD tumors fall near the low-TMB zone.

The Signature of Broken Repair: Frameshifts, Indels, and the Variant Spectrum

Raw mutation count tells you how many errors piled up. The type of mutation tells you how they got there. Mismatch repair deficiency leaves a distinctive fingerprint: because the repair system normally corrects insertion-deletion loops at microsatellite repeats, its absence produces a disproportionate number of frameshift insertions and deletions. These frameshifts are biologically significant — they scramble the reading frame of a protein and often generate novel peptide fragments (neoantigens) that the immune system can recognize.

The variant classification spectrum across the cohort bears this out. Missense mutations — single amino-acid substitutions — still dominate by sheer count, as they do in virtually all cancer sequencing datasets. But the frameshift bars are conspicuously tall relative to what one would see in mismatch repair–proficient tumors. That enrichment is the molecular calling card of MMRD and the mechanistic bridge between broken repair and immune visibility.

Beyond frameshifts, the biomarker status composition within each histology subtype reveals whether certain histologies are detected by different MMR assay methods, introducing a potential confounder. If endometrioid cases are overwhelmingly confirmed by one biomarker category while rarer subtypes split differently, stratified comparisons become unreliable.

The Driver Genes: What Gets Hit, and How Often

A flood of mutations does not mean every gene matters equally. Some genes are recurrently altered because they sit in pathways the tumor needs to deregulate. In MMRD endometrial cancer, the usual suspects are well known: PTEN, the most frequently mutated tumor suppressor in endometrial cancer; PIK3CA, an oncogene in the same PI3K signaling pathway; and ARID1A, a chromatin remodeling gene lost in a wide range of cancers.

The OncoPrint — a sample-by-gene matrix of the 25 most recurrently altered genes — displays these patterns at single-sample resolution. Rows are genes, columns are patients, and colored tiles mark the type of alteration. Several features emerge. PTEN, PIK3CA, and ARID1A are altered in a large fraction of the cohort, often co-occurring in the same sample. That co-occurrence — dual hits in the PI3K pathway — suggests that one alteration alone is insufficient and the tumor requires convergent activation. Meanwhile, genes in the RAS-MAPK pathway (KRAS, BRAF) tend toward mutual exclusivity: if one is hit, the other usually is not, because either alone is enough to activate the pathway.

The chromosomal distribution of variants adds a spatial dimension to this picture. Variant counts per chromosome roughly track chromosome length and gene density, but disproportionate peaks on specific chromosomes may reflect microsatellite-dense regions or biases in panel design — a question that links back to quality control.

Quality Control: Making Sure the Signal Is Real

Targeted sequencing panels like MSK-IMPACT cover hundreds of cancer-related genes but leave most of the genome unexamined. That means variant counts, TMB estimates, and gene recurrence figures are all shaped by what the panel covers. The MSK cohort used three successive panel versions — IMPACT341, IMPACT410, and IMPACT468 — with expanding gene lists and different exonic footprints. If the panel version is not accounted for, TMB denominators shift between samples, and genes covered only by the largest panel will appear less frequently mutated in samples sequenced on smaller panels.

The panel usage chart shows which version was applied to each sample. If a single panel dominates, cross-sample comparisons are straightforward. If the cohort is split across panels, formal analyses must stratify or normalize accordingly.

Mutation position density across genomic coordinates serves as a second quality-control lens. A roughly uniform distribution with modest peaks at gene-dense regions is healthy. Sharp, narrow spikes could indicate recurrent hotspot mutations — or technical artifacts at repetitive loci prone to mismapping. Extended gaps of zero coverage are expected and simply reflect regions outside the panel's target territory.

What this atlas makes concrete is the sheer scale of damage that mismatch repair failure inflicts on a tumor genome — and why that damage, paradoxically, may be the tumor's undoing. The hundreds of frameshift mutations strewn across microsatellite repeats are not just molecular debris; they are potential neoantigens, flags that the immune system can learn to recognize. That recognition is the basis of checkpoint immunotherapy's success in MMRD cancers. But the data also make clear that MMRD is not monolithic. The mechanism of repair loss — germline, somatic, epigenetic — varies; the resulting TMB spans a wide range; and the driver genes hit alongside the repair defect differ from patient to patient. As clinical trials increasingly stratify patients by molecular subtype, cohorts like this one — deeply annotated, panel-sequenced, and publicly accessible — become the reference maps against which each new patient's tumor can be read. The next step is linking these mutational portraits to treatment outcomes: which constellations of driver mutations predict response, resistance, or relapse in the era of immunotherapy.