Molecular Characterization of Endometrial Carcinomas in Black and White Patients Reveals Disparate Drivers with Therapeutic Implications

Endometrial carcinoma (EC) is the most common type of gynecologic malignancy in the United States, with an estimated 66,200 new cases and 13,030 deaths expected in 2023 (1). The incidence and mortality of this disease have been steadily increasing over the past decades (2). Importantly, hysterectomy-corrected mortality rates are highest among Black patients and are 2-fold higher overall than for White patients despite similar incidence (2–4), which represents one of the largest racial disparities among common cancers (5).

EC is a clinically, histologically, and molecularly heterogeneous disease. Two thirds of cases are diagnosed at an early stage and are generally associated with a good prognosis (1). Although endometrioid EC, particularly of low tumor grade (i.e., grades 1 and 2), is the predominant histologic type, the more aggressive subtypes of serous histology and carcinosarcoma, which represent 15% to 20% of ECs, account for the majority of deaths (2, 6). Studies have found that serous and carcinosarcoma histology is more prevalent in Black than in White patients (7, 8), potentially contributing to disparities in survival outcomes. The Cancer Genome Atlas (TCGA) landmark study revealed that ECs can be classified into four molecular subtypes (9), namely, (i) POLE (ultramutated) ECs, characterized by POLE exonuclease domain hotspot mutations and very high tumor mutational burden (TMB); (ii) microsatellite instability (MSI)–high (MSI-H) ECs with a high TMB (TMB-H); (iii) copy number (CN)–high (CN-H) ECs with high levels of somatic CN alterations and recurrent TP53 mutations; and (iv) CN-low (CN-L) ECs, which lack the defining molecular characteristics of the other three molecular subtypes and are also referred to as of no specific molecular profile (NSMP; ref. 10). This molecular classification has proven to be both prognostic, with POLE EC patients having the best and CN-H EC patients having the worst outcomes, and predictive, for example, in selecting patients with ECs of MSI-H subtype for treatment with immune-checkpoint blockade. To date, comprehensive tumor–normal multigene sequencing studies of ECs from self-identified Black patients have been limited [n = 109 pan-cancer TCGA/Pan-Cancer Analysis of Whole Genomes (PCAWG); refs. 11, 12].

At Memorial Sloan Kettering Cancer Center (MSK), the Gynecology Disease Management Team routinely offers clinical tumor and matched normal MSK-IMPACT sequencing, an FDA-authorized targeted next-generation sequencing panel encompassing up to 505 cancer-associated genes (13), to newly diagnosed patients with EC presenting for care. We leveraged this resource of an unselected series of prospectively clinically sequenced ECs to define the distribution of molecular and histologic subtypes as well as the landscape of somatic genetic alterations affecting cancer-related genes in ECs occurring in self-identified Black and White patients.

From a total of 2,248 unique EC patients with high-quality, FDA-authorized clinical tumor–normal MSK-IMPACT sequencing data from January 2014 to December 2021, those self-identifying as Black (n = 259, 12%) and White (n = 1,623, 72%) were included in this study (n = 1,882; Table 1; Supplementary Fig. S1). The prevalence of patients of Hispanic ethnicity was similar between the two groups (n = 13, 5.2% Black vs. n = 92, 5.9% White, P = 0.6; Table 1). The vast majority of MSK-IMPACT sequencing assays were performed on the primary tumor specimen (n = 222, 86% Black vs. n = 1,356, 84% White, P = 0.3; Table 1).

We first assessed the distribution of histologic types based on expert gynecologic pathologist review in this unselected group of EC patients from a large referral cancer center. Although almost half of White patients had ECs of low-grade endometrioid histology (n = 766, 47%), followed by serous (13%), carcinosarcoma (11%), grade 3 endometrioid histology (11%), and mixed/high-grade not otherwise specified (NOS; 8.5%), Black patients more commonly had ECs of high-risk histology (P < 0.01). Serous histology (29%), carcinosarcoma (20%), mixed/high-grade NOS (12%), and grade 3 endometrioid histology (6.6%) made up 68% of ECs in Black patients, whereas only 21% were low-grade endometrioid ECs (Fig. 1A; Table 1). This difference in the distribution of histologic types occurring in Black versus White patients was maintained when analyzing primary and recurrent/metastatic ECs separately (Supplementary Fig. S2A).

Akin to the difference in histologic types, the distribution of EC molecular subtypes was distinct. Although the prevalence of CN-H/TP53 abnormal (TP53abn; 35%), CN-L/NSMP (34%), and MSI-H (25%) molecular subtypes was similar in White patients, 69% of ECs in Black patients were of CN-H/TP53abn molecular subtype (Fig. 1B; P < 0.001). In contrast, ECs of the POLE molecular subtype, associated with a favorable prognosis (9), were found in 5.8% of White patients but were rarer among Black patients (3/259, 1.2%; Table 1 and Fig. 1B). Akin to the histologic types, this distribution of molecular subtypes was also found when analyzing primary and recurrent/metastatic ECs separately (Supplementary Fig. S2B). Within a given histologic type, the molecular subtype distribution was similar between ECs occurring in Black compared with White patients, except for the POLE molecular subtype (Fig. 1C). Collectively, these results demonstrate that a high proportion of ECs among unselected Black patients have tumors of aggressive histologic and molecular subtypes.

Our institution recently described a method to assess global ancestral contributions that can be quantitatively inferred from the clinical MSK-IMPACT sequencing panel used in the current study (14). We applied this method to our data in a hypothesis-generating exploratory analysis, which revealed that 373 of 1,623 (23%) self-identified White patients were of Ashkenazi Jewish (ASJ) ancestry. Although there was no statistically significant difference in the EC molecular subtype distribution (P = 0.08), we found the distribution of histologic types to be distinct (P = 0.014), with a numerically higher proportion of carcinosarcomas in self-identified White patients with ASJ ancestry (57/373, 15% vs. 116/1,250, 9.3%) and of low-grade endometrioid ECs in White patients without ASJ ancestry (610/1,250, 49% vs. 156/373, 42%; Supplementary Fig. S3A and S3B).

In addition, our genetic ancestry analyses revealed that 182 of 259 (70%) and 77 of 259 (30%) self-identified Black patients were of African and admixed ancestry, respectively, and that none were from European, East Asian, Native American, South Asian, or ASJ ancestral populations (Fig. 1D). Importantly, no statistically significant differences in the distribution of histologic types and molecular subtypes of EC between self-identified Black patients with African ancestry (n = 182) as compared with those with admixed ancestry (n = 77) were found (P = 0.4; Fig. 1D). Finally, we compared the histologic and molecular subtype distribution of ECs from the pan-cancer TCGA study for which self-reported race was available along with the genetically inferred ancestry (11, 15). This analysis, despite the sample inclusion biases stemming from the TCGA study design, confirmed our findings and demonstrated no statistically significant differences in the histologic types and molecular subtypes of ECs in self-identified Black patients with African ancestry (n = 50) as compared with those with admixed ancestry (n = 29; Supplementary Fig. S4A and S4B).

Consistent with the lower prevalence of ECs of hypermutated MSI-H and POLE molecular subtypes in Black patients in our MSK-IMPACT series (15.1% vs. 30.7% White patients, P < 0.001), as a group, the median TMB was significantly lower in ECs from Black (median 4, mean 14 mutations (Mut)/megabase (Mb); range, 1–611) than White patients (median 7, mean 24 Mut/Mb; range, 0–668; P < 0.001; Table 1; Fig. 2A; Supplementary Fig. S5A). In fact, 33% (532/1,623) of ECs from White patients had a TMB ≥10, a cutoff used for the tumor-type agnostic eligibility to immune-checkpoint inhibitors (ICI; ref. 16), compared with only 15% (39/259) of ECs in Black patients (P < 0.001; Table 1). On the other hand, with the majority of ECs in Black patients being of CN-H/TP53abn molecular subtype, we found the fraction of genome altered (FGA), a measure of genomic and chromosomal instability (17), to be significantly higher in ECs from Black patients (median 30%; range, 0%–90%) than in ECs occurring in White patients (median 10%; range, 0%–100%, P < 0.001; Fig. 2B; Supplementary Fig. S5A). Whole-genome doubling (WGD), which involves the duplication of a complete set of chromosomes and has been linked to accelerated cancer genome evolution, chromosomal instability, and worse prognosis (18, 19), was significantly more commonly found in ECs of Black (32%) compared with White patients (18%, P < 0.001; Table 1). Of note, no significant differences in TMB and FGA between self-identified Black patients with African genetic ancestry (n = 182) as compared with those with admixed ancestry (n = 77) were observed (Supplementary Fig. S4C).

Given the distinct distribution in histologic and molecular subtypes, we next performed direct comparisons by histology and by molecular subtype between Black and White patients. When TMB was assessed according to histologic type, a significantly higher TMB was detected in ECs from White than from Black patients in the majority of histology types, including low-grade endometrioid (P = 0.044), serous (P = 0.005), carcinosarcoma (P = 0.022), and mixed/high-grade NOS (P < 0.001); no difference in TMB in grade 3 endometrioid and clear-cell ECs was observed, likely due to the small sample sizes (Fig. 2C). In contrast, the level of chromosomal instability as assessed by FGA did not statistically significantly differ in ECs by histologic type between Black and White patients (Fig. 2D). Upon stratification by molecular subtype, TMB and FGA levels were similar within ECs from Black versus White patients, with the exception of TMB in CN-H/TP53abn ECs, which was lower in Black than in White patients (median 3.5 Mut/Mb; range, 1–11 vs. White 4.4 Mut/Mb; range, 1–24; P < 0.001; Supplementary Fig. S5B and S5C). Notably, when excluding ECs of POLE and MSI-H molecular subtype, which are expected to have an elevated mutational burden, and focusing on the nonhypermutated CN-L/NSMP and CN-H/TP53abn ECs, tumors with TMB ≥10 were less common in Black compared with White patients (2.7% vs. 10%, P < 0.001).

We next defined the landscape of the somatic pathogenic mutations and gene CN alterations of ECs and observed significant differences in the recurrently altered genes. As expected, the cancer-related genes affected by pathogenic mutations and gene CN alterations across ECs are the same between Black and White patients. Consistent with the enrichment of the CN-H/TP53abn molecular subtype, as a group, TP53 mutations were more commonly found in ECs from Black than from White patients (72% vs. 42%, q <0.001; Fig. 3A). In addition, genetic alterations previously associated with high-risk histologic types (9, 20, 21), including ERBB2 amplification/oncogenic mutations (15% vs. 6%, q <0.001) and CCNE1 amplification (15% vs. 4%, q <0.001), were more frequently altered in ECs from Black patients (Fig. 3A). It is noteworthy that this difference in ERBB2 alteration incidence in ECs from Black compared with White patients was driven by an enrichment in ERBB2 gene amplification (12% vs. 3%, q <0.001) rather than somatic ERBB2 point mutations (3% vs. 3%; q >0.1). On the other hand, genes commonly altered in low-grade endometrioid ECs (9), including PTEN (26% vs. 55%, q <0.001), ARID1A (19% vs. 47%, q <0.001), and CTNNB1 (9% vs. 17%, q = 0.01), were less commonly observed in tumors in Black than in White patients. Taken together, these results demonstrate that the collective molecular make-up of ECs among self-identified Black patients differs from that of White patients, having tumors with recurrent TP53 mutations, high levels of chromosomal instability, and lower TMB.

Activation of the PI3K/AKT/mTOR signaling pathway plays a key role in EC (9). Analysis of the entire cohort revealed that oncogenic alterations in the PI3K pathway–related genes PTEN (26% vs. 55%, q <0.001), PIK3R1 (13% vs. 28%, q <0.001), and KRAS (12% vs. 21%, q = 0.01) were significantly less common in ECs from Black than from White patients (Fig. 3A and B). PIK3CA genetic alterations were similar between the two groups (38% Black vs. 46% White, q >0.1); however, the mechanism of genetic alteration/activation was distinct: Although 12% of PIK3CA alterations in ECs from Black patients were due to amplification (12/100), PIK3CA amplification was found in only 4% of PIK3CA-altered ECs in White patients (31/767; P = 0.002, Fisher exact test; Fig. 3A and B). In addition, AKT2 gene amplification was more common in ECs from Black compared with White patients (9% vs. 3%, q <0.01; Fig. 3A and B).

When focusing on ECs of CN-H/TP53abn and CN-L/NMSP molecular subtypes, we found that in these nonhypermutated ECs, PIK3CA-activating genetic alterations were similar between groups (CN-H/TP53abn: 30% Black vs. 38% White, q >0.1; CN-L/NMSP: 52% Black vs. 45% White, q >0.1; Fig. 3B). PTEN mutations (6% vs. 15%, q <0.1) and mutations affecting PPP2R1A encoding the protein phosphatase type 2A (PP2A; 15% vs. 28%, q <0.1), a regulator of PI3K downstream signaling and potential therapeutic target (22), were less frequent in CN-H/TP53abn ECs from Black than from White patients (Fig. 3B). No differences in the genetic alterations affecting components of the PI3K pathway were detected in CN-L/NSMP ECs (Fig. 3B). Also, no significant differences in oncogenic alterations affecting PI3K pathway–related genes in endometrioid ECs between the groups were found (Fig. 3B).

Next, we compared the rates of clinically actionable genetic alterations identified in ECs occurring in Black and White patients. Using the FDA-recognized clinical knowledge database OncoKB (23), we annotated patient EC somatic mutations and CN alterations with their level of evidence of being a predictive biomarker of drug response. Notably, the frequency of a clinically actionable alteration making an EC patient eligible for treatment with an FDA-approved drug (level 1 or 2, including TMB ≥10 Mut/Mb and MSI-H) or enrollment into a clinical trial (level 3A) was significantly less in Black than in White EC patients (22.4% vs. 39.7%; P < 0.001; Fig. 3C).

We determined whether the somatic mutations and gene CN alterations of ECs from a given molecular or histologic subtype would be similar between groups. Although no significant differences in the somatic genetic alterations were identified in ECs of CN-L/NSMP, MSI-H, and POLE molecular subtypes, those of the CN-H/TP53abn molecular subtype were distinct. CN-H/TP53abn ECs from Black patients harbored significantly less frequent PPP2R1A (15% vs. 28%, q <0.1) and PTEN (6% vs. 15%, q <0.1; Fig. 3D) mutations, both being negative regulators of the PI3K pathway (9, 22, 24). Of note, PPP2R1A and PTEN mutations were mutually exclusive (P < 0.01; Supplementary Fig. S6A–S6C). Conversely, CN-H/TP53abn ECs from Black patients more frequently harbored CCNE1 amplification (21% vs. 11%, q <0.1), KMT2B mutations (16% vs. 8%, q <0.1) and BRCA1 somatic mutations (3% vs. 0.3%, q <0.1), among others, than ECs from White patients (Fig. 3D).

When analyzing the most common histologic types, the molecular profiles of low-grade endometrioid, high-grade endometrioid, and serous ECs were similar between groups, except for pathogenic MED12, including p.G44A/D/S/R/V hotspot mutations, which were more prevalent among serous ECs from Black versus White patients (11% vs. 0%, q <0.1). Among carcinosarcomas, which comprised 28% of the CN-H/TP53abn molecular subtype in Black patients, CCNE1 amplification, KMT2B mutations, and NF1 mutations were more frequently observed in carcinosarcomas from Black compared with White patients (CCNE1 29% vs. 10%, q <0.1; KMT2B, 31% vs. 10%, q <0.1; NF1, 12% vs. 1%, q <0.1; Fig. 3E).

In this study, leveraging an unselected series of prospectively clinically sequenced ECs with central expert gynecologic pathology review at a comprehensive cancer center, we demonstrate that the spectra of genetic alterations, molecular subtypes, and histologic types observed in ECs from Black patients are distinct from those in White patients. There is a significantly higher prevalence of high-risk histologic and molecular subtypes among Black patients coupled with a lower prevalence of MSI-H/TMB-H ECs. Strikingly, even when comparing CN-H/TP53abn molecular subtype ECs or carcinosarcomas head-to-head, differences at the genetic level are seen.

The distinct proportion of high-risk histologic and molecular subtypes in ECs occurring in Black as compared with White patients is reflected in the overall comparisons assessing molecular features between these two patient groups, with higher levels of TP53 mutations and chromosomal instability seen in ECs of Black patients. In contrast, direct comparisons by histologic type and molecular subtype revealed that TMB was lower in ECs from Black as compared with White patients, even within specific histologic subtypes or CN-H/TP53abn ECs, whereas the FGA was similar. Gene-level analysis of CN-H/TP53abn ECs and carcinosarcomas, however, demonstrated a higher incidence of genetic alterations, such as CCNE1 amplification, associated with poor outcomes and chemoresistance (25), in Black as compared with White patients. On the other hand, PTEN and PPP2R1A mutations were found at lower frequencies in CN-H/TP53abn ECs from Black than from White patients. The co-occurrence of PTEN and PPP2R1A somatic alterations has been associated with negative conditional selection and improved patient survival in high-grade ECs, suggesting that mutual exclusivity between these two genes could result from synthetic lethality (26). Thus, our finding of a lower frequency of these somatic alterations in CN-H/TP53abn ECs from Black patients might suggest a more aggressive clinical behavior.

Molecular classification of ECs into the four TCGA-defined groups has been shown to have strong prognostic value (9) as well as potential predictive value, including for the benefit of adding chemotherapy to standard-of-care adjuvant radiotherapy (27), the use of ICIs (28–31), or of HER2-targeted therapy (32). The variations we observed in the distribution of molecular subtypes and somatic features between ECs from Black and White patients represent important findings that portend prognosis and differential levels of availability and susceptibility to therapies. For example, the TCGA study estimated that 25% to 30% of ECs exhibit an MSI-H phenotype (9), and although a similar prevalence was observed in the present study in ECs from White patients (25%), this rate was significantly lower in Black patients (14%). MSI-H ECs show robust response rates to ICIs, and both pembrolizumab and dostarlimab are approved in recurrent mismatch repair–deficient (dMMR)/MSI-H ECs. Recently, two randomized, placebo-controlled, phase III clinical trials found that the addition of ICIs to standard chemotherapy in patients with advanced or recurrent EC resulted in significantly improved progression-free survival—most pronounced and practice-changing in the dMMR/MSI-H group (30, 31). Additionally, TMB is also a marker of response to ICIs, and pembrolizumab is approved for unresectable or metastatic solid tumors with TMB ≥10 Mut/Mb (33). We detected a lower frequency of TMB-H in ECs from Black compared with White patients (15% vs. 33%). Furthermore, our OncoKB analysis revealed that the frequency of a clinically actionable alteration, namely, MSI-H and TMB ≥10 Mut/Mb, was significantly lower in Black versus White patients, potentially leading to disparate eligibility for available therapies. Importantly, it has been recently shown that tumor–normal sequencing, like the MSK-IMPACT assay used in this study, is not related to an overestimation of TMB in Asian/African populations, whereas tumor-only sequencing is (34). Our findings are consistent with prior studies showing differential TMB in nongynecologic cancers of Black compared with White patients, after adjustment for ancestry, and associations with immunotherapy response (34). Although pembrolizumab with lenvatinib is approved in microsatellite-stable EC, the addition of lenvatinib introduces more toxicity, including hypertension and proteinuria (35). Although rare overall, the POLE molecular subtype is associated with very favorable outcomes and susceptibility to ICIs (36), and only 1.2% of Black patients had ECs of the POLE molecular subtype. Further studies are required to understand the mechanistic basis for the low occurrence of good-outcome POLE ECs in Black patients. Our data suggest, collectively, that Black patients may have differential access to immune-based therapies compared with White patients, as their tumors less commonly have molecular subtypes or markers that confer the most benefit.

We observed fewer low-grade endometrioid ECs, which are often hormonally driven, and thus fewer alterations in the PI3K/AKT signaling pathway in tumors from Black compared with White patients, suggesting fewer Black patients may benefit from hormonal therapies or may be eligible for precision medicine trials targeting the PI3K/AKT pathway. In particular, we found fewer PTEN mutations in ECs occurring in Black patients in the overall cohort and in those of the CN-H/TP53abn molecular subtype as compared with White patients. In colorectal cancer, it has been reported that the PTEN mutation frequencies are highest in tumors bearing POLE mutations (37). This is consistent with previous observations from our team, in which we found PTEN mutations most commonly in ECs of POLE (92%) and MSI-H molecular subtype (87%) compared with CN-L/NSMP (68%) or CN-H/TP53abn tumors (16%; ref. 38). One could hypothesize that the lower PTEN mutation frequency in ECs from Black patients may be associated, at least in part, with the lower occurrence of POLE alterations.

In contrast, we found more serous carcinomas and carcinosarcomas, corresponding to the CN-H/TP53abn molecular subtype, in Black compared with White patients. Although 23% of self-identified White patients with EC in our study were inferred to be of ASJ genetic ancestry and showed a numerically higher proportion of carcinosarcomas compared with those of non-ASJ genetic ancestry, the prevalence of carcinosarcomas occurring in self-identified Black patients was higher than that seen in self-identified White patients overall. Serous carcinomas and carcinosarcomas generally display aberrant expression of p53 with corresponding mutations in TP53 and not uncommonly have WGD; given the aggressive nature of serous ECs and carcinosarcomas, these patients often require chemotherapy even for early-stage disease (27). We noted that TP53 mutations tended to co-occur with PPP2R1A and FBXW7 mutations in the ECs studied, while being mutually exclusive with genes preferentially recurrently mutated in endometrioid ECs (9), including KRAS, ARID1A, and CTNNB1 (Supplementary Fig. S6). Black patients with EC have been reported to be more likely to experience disease recurrence compared with White patients, even after adjustment for stage and treatment (39), and there is a need to precisely define the differences in tumor phenotype to deploy effective targeted therapies and overcome this disparity. In addition, coupled with the higher prevalence of CN-H/TP53abn ECs, our data demonstrate that ECs occurring in Black patients have significantly higher rates of ERBB2 amplification (12% Black vs. 3% White), although differences are not observed within subtypes. Several anti-HER2 therapies, including trastuzumab, are now being added to conventional chemotherapy for advanced serous ECs harboring HER2 overexpression/ERBB2 amplification (40). Ongoing clinical trials are also testing novel anti-HER2 agents such as antibody–drug conjugates (i.e., trastuzumab deruxtecan, NCT04482309), and clinical activity in previously treated patients with advanced or recurrent carcinosarcoma has been recently demonstrated, including those with low HER2 expression (41). Additionally, we found CCNE1 amplification, a marker of aggressive disease (25), to be more common in CN-H/TP53abn ECs and carcinosarcomas of Black as opposed to White patients, representing a potential target in this aggressive histology that may benefit Black patients. Cyclin E1, the protein encoded by CCNE1, plays a role in cell-cycle transition, and in cells with TP53 inactivation, CCNE1 amplification leads to replication stress, chromosome instability, and aneuploidy (42). Consistent with these observations, we found higher levels of genomic instability as assessed by WGD and FGA in ECs of Black compared with White patients. Several ongoing clinical trials are investigating targeted agents that capitalize on CCNE1 amplification, including WEE1 and ATR inhibitor combinations and synthetically lethal PKMYT1 kinase inhibition (43, 44).

Our study has several limitations. We used self-reported race/ethnicity, which reflects self-defined membership in a sociopolitical construct (45), to define groups that may still have substantial biological heterogeneity. This measure considers social determinants of health and structural factors that may influence the racial inequities observed in EC (46). Ancestry inference, on the other hand, is a critical tool for identifying genetic contributors to differences among populations that may contribute to health disparities. We made use of a method recently described by some of the coauthors to assess global ancestral contributions quantitatively inferred using markers captured by the clinical MSK-IMPACT sequencing (14). In this hypothesis-generating, exploratory analysis, we found no statistically significant differences in the distribution of histologic subtypes, molecular subtypes, TMB, or FGA of EC between self-identified Black patients with African as compared with admixed ancestry; however, the sample size is small. The finding of the concordance of histologic and molecular EC subtypes between self-identified Black patients with African and admixed ancestry was independently validated in ECs from the pan-cancer TCGA study. The gene panel used was developed based on actionable mutations in genes reported to play a critical role in cancer; it should be noted that sequencing studies that informed the selection of targeted cancer gene panels predominantly included European ancestry populations, which, by design, do not reflect the diversity across African populations. Furthermore, although the sample size of the ECs from Black patients included in this study is limited, and further work in larger representative cohorts is required, it constitutes the largest cohort with comprehensive clinical multigene panel genomic testing available in the literature to date. To mitigate health inequity, however, future work incorporating diverse populations into molecular characterization studies and trials is essential.

Here, we demonstrate that the prevalence of ECs of clinically aggressive histologic and molecular subtypes is significantly higher in Black compared with White patients, and that even within ECs of the same molecular subtype (i.e., CN-H/TP53abn ECs), molecular features associated with poor outcome, including CCNE1 amplification, are more prevalent in Black patients. On the other hand, POLE ECs associated with favorable outcomes are rarer in Black patients. Our work further reveals potential disparities in therapeutic opportunities for EC patients given differences in MSI-H/POLE subtypes and TMB levels within tumors of Black compared with White patients, and demonstrates that novel targets, such as ERBB2 and CCNE1 amplification, are particularly prevalent drivers in tumors in Black patients. Our findings further highlight that greater diversity and more emphasis on equity and inclusion may be required in the development of novel clinical EC trials, given the fundamental differences in the prevalence of targetable alterations among tumors from Black and White patients, and confirm the importance of genomic profiling of tumors from patients of diverse ancestry. Given current low levels of underrepresented minority recruitment into trials, to combat disparities, concerted efforts are needed to expand access to clinical trials to the populations that would benefit the most.

This study was approved by MSK's Institutional Review Board at MSK, and written informed consent was obtained from all patients. All patients with EC who consented to clinical tumor–normal MSK-IMPACT sequencing (13, 14, 17) between January 2014 and December 2021 were included (N = 2,665; Supplementary Fig. S1). Demographic data, including self-identified race/ethnicity, were extracted by review of the medical record. This study was conducted in accordance with the Declaration of Helsinki.

Histopathologic data were obtained from pathology reports, which are based on the pathology review of departmental gynecologic pathologists through a uniform diagnostic approach with biweekly diagnostic consensus conferences, as previously described (20, 38). All EC histologic types were included. Based on the patients’ initial pathologic diagnosis, histologic type, including endometrioid, serous, clear cell, undifferentiated/dedifferentiated, carcinosarcoma, and mixed/high-grade NOS, as well as tumor grade for endometrioid ECs and FIGO 2009 stage, were recorded. Histology re-review was performed by two gynecologic pathologists (L.H.E. and/or A.M.-B.) for ECs of ambiguous or unclassified histology, and, if unresolved, were categorized as “unclassified,” as previously described (38). In addition, results of the immunohistochemistry (IHC) analysis of the DNA MMR proteins MLH1, MSH2, MSH6, and PMS2 were recorded. MMR deficiency was defined by loss of expression of one or more MMR proteins in all tumor cell nuclei in the presence of an internal control (47).

The MSK-IMPACT assay targets between 341 (2014) and 505 (2021) cancer-related genes (14, 17). Genomic data extracted from MSK-IMPACT included somatic mutations; pathogenicity of somatic mutations as defined by OncoKB, CIViC, and Cancer Hotspots (23, 48, 49); TMB (Mut/Mb); MSIsensor score (50); tumor purity; and FGA (i.e., the percentage of the genome affected by CN alterations; refs. 13, 17, 20, 38). Variants of unknown significance, defined as alterations not classified as oncogenic, likely oncogenic, or predicted oncogenic by OncoKB (23), were excluded from the analysis. We used FACETS to define the allele-specific gene CN alterations (51), which were manually reviewed for the identification of amplifications and homozygous deletions, tumor purity, and ploidy, as previously described (52). WGD status was inferred from MSK-IMPACT sequencing data, as previously described (18). Briefly, tumor samples were considered to have undergone WGD if the fraction of major allele >1 was >50%. For molecular analyses, only high-quality samples were included, and those with a tumor purity of <20% or absence of any somatic mutations were excluded (Supplementary Fig. S1), as previously described (38). Molecular subtype assignment was performed using an integrated clinical sequencing, IHC approach, as previously described (38). ECs were defined as (i) POLE (ultramutated) based on the presence of a known POLE hotspot exonuclease domain mutation (53) or POLD1 pathogenic mutation (54); (ii) MSI-H based on a high MSIsensor score ≥10 inferred from MSK-IMPACT sequencing (50, 55) and/or MMR deficiency based on IHC; (iii) CN-H/TP53abn based on the presence of a pathogenic/driver TP53 mutation or TP53 homozygous deletion; or (iv) CN-L/NSMP based on the absence of the defining features of the other molecular subtypes. In addition, CN-L/NSMP and CN-H/TP53abn ECs with a TMB above the median TMB of MSI-H tumors (i.e., >35 Mut/Mb) were deemed unclassifiable and excluded from further analysis.

Mutual exclusivity analysis was performed using combinations of mutually exclusive alterations (CoMET) with the use of a pairwise Fisher exact test to detect the presence of significant pairs of genes (56). Results of the mutual exclusivity analysis were represented using the somaticInteractions function from maftools (57). Pathways found to be significantly enriched (P < 0.01) were selected in PathwayMapper, as previously reported in Sanchez-Vega and colleagues (58). Clinically actionable somatic genetic alterations were defined as those with at least one level 1, level 2 or level 3A–B alteration as defined in OncoKB (23). OncoKB level 1 includes MSI-H and TMB ≥10 Mut/Mb predictive of response to ICIs, as previously described (14). For a given EC, the highest OncoKB level was used for comparison.

Genetic ancestry was inferred from clinical MSK-IMPACT sequencing panel data using more than 3,000 common single-nucleotide polymorphism (SNP) markers, as previously described (14). Briefly, ADMIXTURE v1.3 (59) was run in supervised mode using the 1000 Genomes Project (1KGP) cohort (60) as a reference to infer ancestral proportions of African, European, East Asian, Native American, and South Asian populations. Patients with an ancestral fraction of >0.8 for any single population were assigned that population label; otherwise, they were considered admixed. ASJ ancestry was inferred using 282 additional SNP markers captured by MSK-IMPACT, and the sensitivity and specificity of the ASJ genetic ancestry inference were tested in >8,000 cases for which ASJ status was manually curated through genetic counseling, as previously described (14). For validation, histologic types and molecular subtypes of self-identified Black EC patients were obtained from the pan-cancer TCGA study (11), and compared with the distribution based on genetic ancestry of the same cases, as described in Carrot-Zhang and colleagues (15).

Summary statistics were used to describe the study population. Fisher exact test or Pearson X² test was used to evaluate associations between race and the categorical variables whenever appropriate. Mann–Whitney U test or Wilcoxon rank sum test was used to evaluate associations between race and the continuous variables. Comparisons of frequencies of genes altered by somatic genetic and CN alterations were performed using the Fisher exact test and logistic regression. Multiple testing correction using the Benjamini–Hochberg method was applied to control for the FDR whenever appropriate. All P values are two-tailed, and 95% confidence intervals were adopted for all analyses.

The MSK-IMPACT sequencing dataset is available through the cBioPortal for Cancer Genomics at www.cBioPortal.org under “Endometrial Carcinoma Ancestry (MSK, Cancer Discov 2023)” (https://www.cbioportal.org/study/summary?id=ucec_ancestry_cds_msk_2023).

B. Weigelt reports grants from Repare Therapeutics outside the submitted work. M.F. Berger reports personal fees from Eli Lilly, AstraZeneca, and Paige.AI outside the submitted work. N.R. Abu-Rustum reports grants from GRAIL outside the submitted work. L.H. Ellenson reports an NIH/NCI Cancer Center Support Grant (P30 CA008748) during the conduct of the study. Y.L. Liu reports grants from AstraZeneca, GSK, and Repare Therapeutics outside the submitted work. C. Aghajanian reports grants from Cycle for Survival during the conduct of the study, as well as grants from AbbVie, AstraZeneca, Clovis, and Genentech/Roche, and personal fees from Roche/Genentech, Eisai/Merck, AstraZeneca/Merck, and Repare Therapeutics outside the submitted work. No disclosures were reported by the other authors.

B. Weigelt: Conceptualization, data curation, supervision, investigation, visualization, writing–original draft, writing–review and editing. A. Marra: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. P. Selenica: Data curation, software, formal analysis, investigation, writing–review and editing. E. Rios-Doria: Data curation, writing–review and editing. A. Momeni-Boroujeni: Writing–review and editing, pathology review. M.F. Berger: Data curation, supervision, writing–review and editing. K. Arora: Formal analysis, writing–review and editing. D. Nemirovsky: Formal analysis, writing–review and editing, statistical analysis. A. Iasonos: Formal analysis, writing–review and editing, statistical analysis. D. Chakravarty: Methodology, writing–review and editing. N.R. Abu-Rustum: Resources, writing–review and editing. A. Da Cruz Paula: Data curation, visualization, writing–review and editing. K. Dessources: Conceptualization, writing–review and editing. L.H. Ellenson: Conceptualization, investigation, writing–review and editing, pathology review. Y.L. Liu: Conceptualization, methodology, writing–original draft, writing–review and editing. C. Aghajanian: Conceptualization, resources, writing–review and editing. C.L. Brown: Conceptualization, supervision, funding acquisition, project administration, writing–review and editing.

We gratefully acknowledge members of the Marie-Josée and Henry R. Kravis Center for Molecular Oncology and of the Molecular Diagnostics Service in the Department of Pathology and Laboratory Medicine for their contributions. This work was supported in part by the NIH/NCI Cancer Center Support Grant P30 CA008748 (to MSK) and by Cycle for Survival (to C.L. Brown, B. Weigelt, L.H. Ellenson, A. Iasonos, and P. Selenica).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.