Introduction

Molecular profiling of endometrioid and serous carcinomas of the endometrium by The Cancer Genome Atlas (TCGA) revealed four genomic groups, which include ultramutated tumors with POLE exonuclease domain mutations, hypermutated tumors with microsatellite instability (MSI)/ DNA mismatch repair (MMR) deficiency, copy number-high tumors with TP53 mutations, and copy number-low tumors, which lack all of the above-mentioned alterations1. The latter group was shown to be composed of tumors with predominately endometrioid morphology and considerable molecular heterogeneity, which has since been referred to as endometrial carcinoma (EC) of no specific molecular profile (NSMP)2. Follow-up studies suggested an exclusion surrogate approach (ProMisE) for identifying NSMP-EC, whereby ECs harboring any characteristic of the other TCGA groups are excluded, i.e. tumors lacking POLE exonuclease domain mutations, are MMR-proficient, and are p53/TP53 wild-type are designated as NSMP-ECs3,4. The designation of NSMP to this group of ECs derives from considerable molecular heterogeneity of this group and absence of defining molecular features. Additionally, the NSMP-EC have been shown to be a clinically and histologically diverse.

The molecular heterogeneity of NSMP-ECs has been shown to have clinical relevance5. For example, it has been reported that NSMP-ECs harboring 1q32.1 high level gain6,7, CTNNB1 hotspot mutations7 or expression of L1CAM8 are associated with adverse outcomes.

Given the heterogeneity observed in NSMP-EC, one of the largest molecular subgroups of EC, in-depth studies are needed to further refine and understand the molecular underpinnings of these tumors9. In this study, we sought to define the clinicopathologic and molecular characteristics of a large group of NSMP-ECs subjected to clinical sequencing of 410–468 cancer-related genes and attempt to subclassify these tumors based on characteristic molecular findings.

Materials and methods

Case selection and data extraction

This study, including review and analysis of data, was approved by the Institutional Review Board at Memorial Sloan Kettering Cancer Center (MSK). ECs of all histologic types that underwent clinical tumor-normal targeted DNA next-generation sequencing (NGS) of 410–468 cancer-related genes using MSK-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT)10,11 between 2014 and 2019 were evaluated (n = 1,239). ECs of the NSMP-EC molecular subtype class in this MSK-IMPACT cohort were identified by employing a surrogate of the ProMisE model12, as previously described13: tumors with hotspot POLE exonuclease domain mutations (n = 60), with DNA MMR-deficiency defined either by immunohistochemistry and/or genomic determination (high MSIsensor score; n = 357)14, and with aberrant p53 immunohistochemical expression or TP53 genetic alterations including somatic mutations and homozygous deletions (n = 297) were excluded. In addition, to ensure high quality data, sequenced tumors with very low tumor purity (<10% as estimated by the reviewing pathologist or average variant allele frequency <5%) were excluded from the analysis (n = 53). The remaining cases were included in the downstream analysis (n = 472; see CONSORT diagram Fig. 1).

Fig. 1: CONSORT diagram summarizing the selection process of NSMP endometrial carcinomas included in this study.
figure 1

NSMP-EC, endometrial carcinomas of no special molecular profile; EDM, exonuclease domain mutant; MMR, DNA mismatch repair; MSI, microsatellite instability.

Full size image

Genomic data

The MSK-IMPACT assay assesses somatic mutations, copy number (CN) alterations, structural variants, fraction of genome altered (FGA) and tumor mutation burden (TMB), as previously described13. GISTIC (version 2.0.23) was used to analyze the broad CN data15. The mutational data included chromosomal location, base-pair change, protein change, predicted functional impact of the mutation and the associated variant frequency. Data on allele specific CN alterations and ploidy were extracted using the ‘facets’ R package (version 0.5.14)16. Cancer cell fractions (i.e. clonality) of all somatic mutations were inferred using ABSOLUTE (v1.0.6), as described previously17 using CN and tumor purity estimation information derived from FACETS16. Mutational signatures were determined using the MuSiCa R application (version 1.0)18 in samples harboring ≥5 mutations (both synonymous and non-synonymous). Annotation of the genomic alterations for oncogenic properties was performed using the OncoKB database, as previously described19. Evaluation for mutual exclusivity/co-occurrence was performed using the ‘DISCOVER’ R package (v0.9)20.

Pathology review

The histopathologic and morphologic data were extracted from the synoptic pathology report. The scanned pathology slides of the sequenced tumor were reviewed by a gynecologic pathologist (A.M.-B.) to confirm the findings of the pathology report. For histologic typing, to mitigate the effect of suboptimal interobserver concordance21,22, we performed a single-institution study with a group of experienced gynecologic pathologists. Biweekly diagnostic consensus conferences encouraged a uniform diagnostic approach within the group, as did frequent review of each other’s cases for tumor board and quality assurance, as described13.

Clustering and statistical analysis

A reduced segments matrix was calculated from the FACETS-derived CN segmentation file using the CNTools R package. The CN and the cancer cell fraction of somatic mutations were then used to create a combined data matrix which was subsequently normalized. Principal component (PC) analysis (PCA) was used to reduce the dimensionality of the matrix by employing the fast.prcomp function of the gmodels R package. Thirteen statistically significant PCs were identified using a permutation test. The output was visualized using t-distributed stochastic neighbor embedding (T-SNE; ‘rtsne’ R package; Supplementary Fig. S1)23. To identify tumor clusters in our dataset, partitioning into separate clusters based on their scores along the thirteen significant PCs was performed using the HDBSCAN function of the DBSCAN R package24.

Comparisons of quantitative data between the groups were performed using ANOVA with post-hoc Tukey test and comparison of qualitative data including associations between clinicopathologic features and molecular data were performed using chi-squared test with Fisher’s exact p value calculation. All p value were two tailed and p values <0.05 were considered statistically significant. Disease-free survival (DFS) and overall survival (OS) were evaluated by calculating survival curves using the Kaplan–Meier method, using the Log-rank test to compare subgroups, with the start date set as the date of initial diagnostic biopsy. Univariate and multivariate Cox Proportional Hazards analysis was performed to determine the hazard ratio (HR).

Final case selection

Following genomic clustering, an equal number of endometrioid ECs were selected from each genomic cluster for further downstream analysis using random sampling with stratification (n = ~80 sample for each of three clusters; Fig. 1). In addition, 44 non-endometrioid ECs or endometrioid variants meeting the same criteria used for NSMP-EC were selected and analyzed separately; these tumors consisted of 15 high-grade ECs with ambiguous morphology, nine endometrial clear cell carcinomas, seven mesonephric-like carcinomas, five carcinosarcomas, two corded and hyalinized endometrioid carcinomas, two dedifferentiated ECs, two undifferentiated ECs and two uterine serous carcinomas. The pathology reports and histopathology slides were reviewed for the 240 uterine endometrioid carcinomas and the 44 non-endometrioid/ variant endometrioid ECs, and clinical data relating to disease presentation and course were extracted from the electronic health records (Fig. 1).

Results

A large series of 472 NSMP-ECs subjected to clinical FDA-authorized tumor-normal MSK-IMPACT sequencing of 410–468 cancer-related genes, which included primarily uterine endometrioid carcinomas (UECs; n = 367), but also non-endometrioid ECs and/or tumors with variant endometrioid histology (n = 105), were clustered based on mutational and gene CN alteration data. Specifically, in this initial cohort of NSMP-EC (n = 472), UEC was the most common tumor histologic type (367/472, 77.8%), followed by high-grade ECs with ambiguous morphology (46/472, 9.7%), uterine clear cell carcinoma (23/472, 4.9%), mesonephric-like carcinoma (12/472, 2.5%), uterine carcinosarcoma (9/472, 1.9%) with rare tumors categorized as uterine serous carcinoma (6/472, 1.3%), corded and hyalinized endometrioid carcinoma (4/472, 0.8%), dedifferentiated carcinoma (3/472, 0.6%), and undifferentiated carcinoma (2/472, 0.4%). The majority of UECs were FIGO grade 1 tumors (199/367, 54.2%), followed by FIGO grade 2 (97/367, 26.4%) and FIGO grade 3 (71/367, 19.3%) tumors.

The clustering revealed the presence of 3 distinct clusters: cluster 1 (C1), cluster 2 (C2) and cluster 3 (C3). 240 UECs in total were randomly selected with stratification for further clinical, pathologic and genomic review (C1, n = 80; C2, n = 81; C3, n = 79; see Methods; Fig. 1; Supplementary Fig. S1).

Genomic landscape of endometrioid NSMP-ECs

Overall, in these 240 NSMP-UECs, the most commonly mutated genes were PTEN (n = 154, 64%), ARID1A (n = 108, 45%), PIK3CA (n = 102, 42.5%), CTNNB1 (n = 86, 36%), PIK3R1 (n = 77, 32%), KRAS (n = 70, 29%), CTCF (n = 34, 14%) and BCOR (n = 3, 14%). Other recurrent alterations included FBXW7, SOX17, FGFR2 (all n = 21, 9%) and ESR1 (n = 20, 8%; Fig. 2A).

Fig. 2: Recurrent somatic genetic alterations in NSMP uterine endometrioid carcinomas.
figure 2

A Oncoprint depicting the most recurrent genomic alterations in uterine endometrioid carcinomas of no special molecular profile (NSMP-UEC). Each column represents a tumor with the bar graph at the top depicting the number/distribution of alterations per sample, and the Oncoprint rows showing alterations for each gene. The bottom part of the graph shows the summary of histopathologic and clinical information for each case. The bar graph on the right of the panel shows the number and distribution of alterations for each gene. Mutation types and clinicopathologic features are color-coded according to the legend. B Oncoprint depicting the most recurrent molecular alterations in NSMP-UEC focusing on alterations activating the PI3K/AKT/mTOR pathway in addition to CTNNB1 alterations. The bottom part of the graph shows the summary of molecular cluster information for each case. Note that majority of tumors harbor a combination of PIK3CA/PIK3R1 and PTEN mutations while AKT1 mutations occur in absence of upstream alterations. KRAS, ERBB2 and FGFR2 alterations are mutually exclusive. Mutation types and other features are color-coded according to the legend. C1, Cluster 1; C2, Cluster 2; C3, Cluster 3.

Full size image

Of the 240 NSMP-UECs included in this study, 127 (52.9%) had alterations in combinations of PTEN/PIK3CA and PTEN/PIK3R1, other tumors had pathogenic mutations in AKT1 (6%) and MTOR (3%). KRAS, FGFR2 oncogenic mutations and ERBB2 amplification/activating mutations occurred in a mutually exclusive pattern in 29%, 10%, and 4% of NSMP-UECs, respectively. CTNNB1 hotspot mutations were present in 36% (n = 87) of NSMP-UECs, and there was mutual exclusivity between KRAS/FGFR2/ERBB2 and CTNNB1 alterations (DISCOVER P < 0.001; Fig. 2B).

As a next step, the genomic landscape of NSMP-UECs according to the clusters was assessed and found to be distinct in terms of their genomic instability, mutational burden, CN and mutational landscapes (Fig. 3). The fraction of genome altered, a measure of chromosomal instability, was significantly higher in the C3 NSMP-UECs compared to NSMP-UECs from the other two clusters (mean 0.17 versus 0.07, ANOVA P = 0.0002; Fig. 3B). Overall, amplifications and homozygous deletions were rare in the NSMP-UECs, with ERBB2 (n = 5), AKT3 (n = 4), and NTRK1 (n = 4) amplification and CDKN2A deletions (n = 4) being the most common. In terms of broad chromosomal arm level alterations, 1q gains were common in C3 (n = 35, 44.9%) and C1 tumors (n = 24, 30.4%), and less common in C2 tumors (n = 14, 17.5%, X2 P = 0.001; Figs. 2A, 3A).

Fig. 3: Global genomic landscape of NSMP uterine endometrioid carcinomas.
figure 3

A Summary of broad copy number alterations in the three molecular clusters of uterine endometrioid carcinomas of no special molecular profile (NSMP-UEC) based on GISTIC results. Each column indicates a chromosomal arm as shown in the X-axis labels. The Y-axis represents G-score which considers the amplitude of the aberration as well as the frequency of its occurrence across samples. B Comparison of fraction of genome altered values across the three NSMP-EC clusters. C Comparison of tumor mutational burden (number of mutations/megabase) across the three NSMP-UEC clusters. D Bar-plots comparing the alteration frequency of 8 commonly altered genes among the three molecular clusters of NSMP-UEC. The black horizontal lines marked with a * at the top of each bar-plot represent statistically significant difference in the frequency of the altered gene among the clusters; n.s. not significant. C1, Cluster 1; C2, Cluster 2; C3, Cluster 3.

Full size image

The tumor mutational burden progressively decreased from C1 to C3 NSMP-UECs (median number of somatic mutations per megabase for C1, C2 and C3: 6.9, 5.9, 4.4, respectively ANOVA P: 0.0001; Table 1, Fig. 3C). When assessing specific mutations, we found that while PTEN alterations were common in the C1 and C2 clusters (88 and 100% respectively), they were rare in C3 NSMP-UECs (5%, X2 P < 0.0001; Fig. 3D). PIK3R1 mutations, which were primarily truncating, were almost exclusively found in C1 tumors (91%; versus 0% in C2 and 5% in C3, X2 P < 0.0001). Conversely, PIK3CA mutations, which preferentially affected the hotspot kinase domain codon H1047 (21.4%) or the helical domain hotspot codons E545 and E542 (10.7%, 4.5%), were uncommon in C1 (10%), but had higher frequencies in C2 and C3 NSMP-UECs (74 and 45%, X2 P < 0.0001). AKT1 mutations, primarily the E17K hotspot mutation, were almost exclusively found in C3 NSMP-UECs (17% versus 3% in C1 and 0% in C2, X2 P < 0.0001; Fig. 3D), and they frequently co-occurred with CTNNB1 mutations (see below). No statistically significant difference in the frequency of alterations affecting CTNNB1, KRAS and CTCF was found between the clusters (X2 P > 0.05 for all). We noted that the majority of non-synonymous somatic mutations identified in NSMP-UECs were oncogenic or likely/predicted oncogenic (72.9%; Supplementary Figs S2 and S3), including ESR1 mutations involving codons L536, Y537 and D538 which are associated with resistance to hormone therapy in breast cancer25.

Table 1 Summary of genomic findings in the three clusters of uterine endometrioid carcinomas of no special molecular profile (NSMP-UEC).
Full size table

Finally, we evaluated the cancer cell fractions/clonality of the mutations identified, meaning the bioinformatically inferred percentage of cancer cells harboring a given mutation in a tumor sample. This analysis revealed that in C1 NSMP-UECs PTEN mutations had the highest cancer cell fractions, and were primarily clonal, whereas PIK3R1, ARID1A and KRAS mutations occurred at lower cancer cell fractions. The same trend was observed in C2 NSMP-UECs with PTEN mutations having the highest cancer cell fractions, and PIK3CA, ARID1A and KRAS mutations with lower cancer cell fractions. In C3 NSMP-UECs, however, more heterogeneity was observed, with either AKT1, ARID1A or KRAS being clonal events in these tumors (Fig. 4).

Fig. 4: Clonality of somatic mutations in NSMP uterine endometrioid carcinomas and recurrent genomic alterations in non-endometrioid endometrial cancers of NSMP.
figure 4

A Heatmaps showing the cancer cell fractions of the most commonly altered genes among the three clusters of uterine endometrioid carcinomas of no special molecular profile (NSMP-UEC). Each column represents a tumor. The gradient of the heatmap is based on the calculated cancer cell fraction of the alteration in the sample, color-coded according to the legend. B Oncoprint depicting the most recurrent genomic alterations in non-endometrioid endometrial carcinomas of no special molecular profile (NSMP). Each column represents a tumor with the bar graph at the top depicting the number/distribution of alterations per sample, and the Oncoprint rows showing alterations for each gene. The bottom part of the graph shows the summary of histopathologic and clinical information for each case. The bar graph on the right of the panel shows the number and distribution of alterations for each gene. Mutation types and clinicopathologic features are color-coded according to the legend. C1, Cluster 1; C2, Cluster 2; C3, Cluster 3.

Full size image

Taken together, NSMP-UECs are heterogeneous at the genetic level with C1 tumors being defined by recurrent PIK3R1 and PTEN alterations, C2 tumors by a combination of PTEN and PIK3CA mutations and C3 tumors by KRAS mutations along with 1q high level gain in the absence of PTEN alterations or alternatively AKT1 mutations along with CTNNB1 mutations.

Evaluation of the interaction matrix between the commonly altered genes in NSMP-UECs showed strong mutual exclusivity between PIK3CA and PIK3R1 mutations (DISCOVER P < 0.0001), confirming previous reports1,13. In addition, CTNNB1 and KRAS as well as PTEN and AKT1 mutations were also found to be mutually exclusive (DISCOVER P < 0.01; Supplementary Fig. S4).

In terms of mutational signatures, 202 cases had sufficient mutations (≥5 mutations) to allow for investigation of mutational signatures. Consistent with previous results26, signature 1 associated with aging was the most common mutational signature in NSMP-UECs; no statistically significant difference was observed between the 3 clusters (Supplementary Fig. S5).

Genomic landscape of NSMPs of non-endometrioid and variant endometrioid histology

In addition to the 240 NSMPs of endometrioid subtype, 44 additional ECs subjected to clinical sequencing that fulfilled the criteria of NSMP were of non-endometrioid (n = 19) or variant (n = 21) histologic types. The majority of these non-endometrioid or variant endometrioid NSMP-ECs clustered with C3 UECs (72.7%, n = 32); 15.9% clustered with C2 tumors (n = 7) and 11.4% clustered with C1 UECs (n = 5). One of the two cases of corded and hyalinized endometrioid carcinomas clustered with C2 tumors and the other with C3 tumors.

In terms of global genomic profile, these 44 non-UEC or variant UEC NSMP tumors showed a significantly higher fraction of genome altered (median: 0.0993, range: 0.024–0.76) compared to NSMP-UEC (median: 0.035, range: 0–0.67; Mann-Whitney U P < 0.0001; Fig. 4B). Furthermore, non-UEC or variant UEC NSMP tumors had a higher number of intrachromosomal breakpoints (median: 28, range: 23–74) compared to UEC tumors (median: 24, range: 21–72; Mann–Whitney U P < 0.0001; data not shown) and had a lower overall number of somatic mutations (median: 5 versus 6 respectively; Mann–Whitney U P = 0.009).

The 15 high-grade ECs with ambiguous morphology harbored alterations in PIK3CA (n = 6), ARID1A and KRAS (n = 5). The clear cell carcinomas had ARID1A (n = 4) and PTEN alterations (n = 1; Fig. 4B). The seven mesonephric/mesonephric-like carcinomas all had activating KRAS mutations and chromosome 1q gain, with only two cases harboring concurrent PTEN or PIK3CA mutations27. The carcinosarcomas all had PIK3CA mutations (n = 5) along with either PTEN (n = 3) or FBXW7 (n = 2) alterations. Both undifferentiated ECs showed truncating mutations in SMARCA4. Both variant UEC NSMPs of corded and hyalinized endometrioid carcinoma harbored MED12 mutations, with one also harboring PTEN and PIK3CA mutations and the other case a CTNNB1 mutation. Despite having a variant morphologic phenotype, the molecular phenotype of these tumors closely resembled that of other UECs; however, the limited number of cases prevents any firm conclusion to be drawn (Fig. 4B).

Clinicopathologic features of NSMP ECs

Among the NSMP-UEC patients for whom in-depth clinicopathologic review was performed (n = 240), the median age at diagnosis was 63 years (range: 28–91 years); there was no statistically significant difference between the clusters in terms of age distribution (Fig. 5A). As expected, age had a strong association with the background endometrium with patients with background atrophic/inactive endometrium being older (median age: 65.5 years) compared to patients with either background endometrial hyperplasia (median age: 59 years) or proliferative endometrium (median age: 50.5 years; ANOVA P = 0.0001). Of note, background endometrial hyperplasia was more likely to be observed in C2 (n = 42) compared to C1 (n = 29) and C3 (n = 17) patients (X2 P = 0.02; Fig. 5B), which is in line with prior work showing frequent alteration of PTEN and PIK3CA in endometrial hyperplasia28,29 while PIK3R1 alterations are reportedly rare in hyperplasia30.

Fig. 5: Clinicopathologic features of NSMP uterine endometrioid carcinomas.
figure 5

A Box and whisker plots comparing age distribution between the molecular clusters of uterine endometrioid carcinomas of no special molecular profile (NSMP-UEC). B Bar-plot comparing the background endometrium findings among the molecular clusters. C Bar-plot comparing the tumor FIGO grade among the molecular clusters. D Bar-plot comparing the distribution of myometrial invasion among the molecular clusters. E Box and whisker plots comparing the depth of myometrial invasion among the molecular clusters. F Bar-plot comparing the distribution of tumor FIGO clinical stage among the molecular clusters. G Bar-plot comparing the tumor spread at the time of diagnosis among molecular clusters. H Bar-plot comparing the tumor spread at the time of recurrence among molecular clusters. Molecular clusters are color coded with blue representing Cluster 1, red representing Cluster 2 and yellow representing Cluster 3 tumors.

Full size image

Most of the NSMP-UEC in the cohort were FIGO grade 1 tumors (n = 117, 48.7%), followed by FIGO grade 2 (n = 76, 31.7%) and FIGO grade 3 tumors (n = 47, 19.6%). C3 tumors were more likely to have a high FIGO grade with 30% (24/79) having FIGO grade 3 morphology compared to 14.4% in the other clusters (23/161; X2 P: 0.0001; Fig. 5C).

Estrogen receptor (ER) expression results by immunohistochemical analysis were available for 48 samples; C3 tumors were more likely to have negative or weak and focal ER expression (12/22, 54.5%) compared to C1/C2 tumors (23.1% (6/26); X2 P: 0.025; Supplementary Fig. S6). Progesterone receptor (PR) expression as assessed by immunohistochemistry were available for 33 samples; C3 tumors were more likely to have negative or weak and focal PR expression (9/15, 60%) compared to C1/C2 tumors (16.7% (3/18); X2 P: 0.03; Supplementary Fig. S7).

Patients with C3 NSMP-UECs were more likely to present with higher clinical stage disease with 36.7% (29/79) presenting at FIGO stages III and IV compared to 19.9% (32/161) for C1 and C2 NSMP-UECs (X2 P: 0.005; Table 2). The same was observed for pathologic T stage where 50.6% (40/79) of patients with C3 tumors presented at pathologic T2 stage or above compared to 23.6% (38/161) of C1/C2 tumors (X2 P: < 0.001; Fig. 5F). Follow-up data showed that patients with C3 tumors had a higher mortality rate (49.4%, 39/79) compared to C1 and C2 patients (17.5%, 14/80 and 22.2%, 18/81, respectively; X2 P: 0.0001). Lymphovascular invasion was identified in 78 tumors (32.5%). The C3 tumors were more likely to have lymphovascular invasion compared to the other clusters (41.8%, 33/79 versus 27.9%, 45/161 for C1/C2, X2 P: 0.032). On the other hand, C1/C2 tumors were more likely to have lymph node metastasis (62.7%, 101/161 versus 39.2%, 31/79; X2 P: 0.001) than C3 tumors (Table 2).

Table 2 Summary of clinicopathologic findings in the three clusters of uterine endometrioid carcinomas of no special molecular profile (NSMP-UEC).
Full size table

Patients with C3 NSMP-UECs were more likely to have lung metastasis at recurrence (25/79 versus 16/161 for C1 and C2; X2 P: 0.0001), and to have involvement of extra-pelvic peritoneum and omentum on recurrence (16 and 8 out of 79 cases, respectively) compared to C1/C2 tumors (17 and 4 out of 161 cases, respectively; X2 P = 0.04 and 0.011; Fig. 5G, H).

Among genomic factors, 1q high level gain, absence of PTEN mutation and presence of AKT1 mutation were strongly associated with lung metastasis (X2 P = 0.0001, 0.0001 and 0.005 respectively). Tumors with KRAS mutations in isolation were not associated with lung metastasis (X2 P = 0.167), however in combination with an absence of PTEN mutations the association with lung metastasis was significant. Furthermore, 1q gain was associated with increased likelihood of myometrial invasion (91.3% (n = 63/69) vs. 59.6% (n = 102/171); X2 P = 0.0001). Among the genomic alterations in our entire cohort, ESR1 mutations showed a strong correlation with age; patients with ESR1 mutations were strictly post-menopausal with a median age of 69 years (range 60–82 years) while the patients without ESR1 mutations were younger and included premenopausal patients (median age 62, range: 28–91 years; Mann–Whitney-U P = 0.0001).

From the 284 patients with NSMP-EC, 171 patients were seen at the time of initial treatment planning with surgery performed at MSK, whereas the remainder of the patients (n = 113) were seen at the time of recurrence. As an exploratory, hypothesis-generating analysis, we assessed whether the different clusters of NSMP-ECs identified would be associated with outcome (Fig. 6). Mortality and disease recurrence were more likely in patients with C3 tumors with a 45.3% mortality rate (24/53) and a 66% recurrence rate (35/53). In contrast, patients with C1 and C2 tumors had lower mortality rates (5.3% (3/54) and 18.7% (13/64), respectively) and lower recurrence rates (31.9% (17/54) and 25% (16/64), respectively; Fig. 6E). Log-rank survival analysis showed that the molecular clusters were associated with distinct overall survival (OS), with patients with C1 tumors having the best survival outcomes (Log-rank P: 0.0005; Fig. 6A), while C2 and C3 tumors were statistically similar. In terms of disease-free survival (DFS), patients with C3 tumors had the worse outcomes, however, C1 and C2 tumors together showed similar DFS curves (Log-rank P: < 0.0001; Fig. 6B). Multivariate Cox-regression survival analysis showed that tumor clusters conferred OS disadvantage with C2 tumors having a hazard ratio (HR) of 3.7 and C3 tumors having a hazard ratio of 4.0 (P: 0.049 and 0.032 respectively). However, tumor clusters were not associated with disease-free survival on multivariate analysis. Interestingly, after controlling for stage, the most important prognostic factors for both OS and DFS were presence of FIGO grade 3 morphology (OS HR: 19.6 (P: 0.0003), DFS HR: 10.5 (P < 0.0001)) and non-endometrioid morphology (OS HR: 30.6, P < 0.0001; DFS HR: 12.9, P < 0.0001; Fig. 6C, D; Supplementary Table S1).

Fig. 6: Association of molecular clusters and histologic grade with outcome in NSMP uterine endometrial carcinomas.
figure 6

A, B Kaplan–Meier curves comparing overall survival and disease-free survival among molecular clusters of NSMP-EC. C, D Kaplan–Meier curves comparing overall survival and disease-free survival among histologic grades and type of NSMP-EC. E Distribution of outcome events among the three molecular clusters of NSMP-EC. C1, Cluster 1; C2, Cluster 2; C3, Cluster 3; G1-G3, Grades 1 through 3.

Full size image

Discussion

The NSMP-EC have been shown to have considerable morphologic and clinical diversity: In comparison with copy number-high or POLE ECs, patients with NSMP-EC have an intermediate prognosis comparable to that of women with MSI-H ECs1. However, considerable heterogeneity within the NSMP category exists in terms of clinical outcomes as some patients have an excellent prognosis, whereas others have a more aggressive disease course with associated morbidity and mortality3,6. While the majority of NSMP-EC are endometrioid, other morphologic variants including those with ambiguous morphology, endometrial clear cell carcinoma, mesonephric-like carcinoma, and other rare morphologic variants have been previously reported31. Current management of NSMP-EC is mainly driven by clinical stage and histopathologic features such as FIGO grade, and presence of lymphovascular invasion32. Recently, molecular features such as CTNNB1 alterations or L1CAM expression have been used in risk stratification of the NSMP tumors33.

Our results demonstrate that the clinical and pathologic heterogeneity of the NSMP-ECs can be largely explained with the underlying molecular alterations. In fact, our results suggest that NSMP-ECs, which comprise the largest share of EC molecular subtypes, are composed of several distinct molecular subclades; based on clustering of mutations in cancer-related genes and copy number alterations, we have shown that NSMP-ECs are composed of at least 3 distinct molecular clusters: The first two clusters appear to be to be driven by activating alterations of the PI3K pathway, where mutations in PTEN are often paired with truncating alterations of PIK3R1 (C1) or with activating PIK3CA mutations (C2). Analysis of cancer cell fractions suggests that PTEN alterations are the initiating event for C1 and C2 tumors followed by alterations in either PIK3CA or PIK3R1, ARID1A, CTNNB1 and KRAS.

C3 tumors are markedly different from both C1 and C2 tumors. These tumors have a relative dearth of PTEN alterations. PI3K pathway alterations in C3 tumors were mostly single hits with either PIK3CA or AKT1 or KRAS activating alterations. A subset of C3 tumors harbored KRAS activating mutations along with chromosome 1q high level gain. The latter finding has been well-studied and believed to be associated with a mesonephric-like phenotype34 with often adverse outcomes35. The mesonephric-like carcinomas are known to have a propensity for lung metastasis36, a phenomenon which we also observed in our cohort. Unlike the KRAS mutated/1q gain tumors, which are well documented, characterization of an AKT1 mutated subgroup has only been observed anecdotally37. In this study, we report an even stronger propensity for lung metastasis compared to the KRAS mutated/1q gain tumors.

The genomic landscape of C1 and C2 NSMP-ECs supports the previous findings of the synergistic effect of PTEN and PIK3CA mutations38,39,40, and in the absence of PIK3CA activating mutations, PIK3R1 truncating mutations may exert a similar synergistic effect38,40,41. The mutually exclusive nature of PIK3R1 and PIK3CA mutations has been previously shown in both endometrial and breast cancers1,42. AKT1 p.E17K mutation is another activating mechanism of the PI3K/AKT pathway43,44 and we have shown that activating mutations of AKT1 gene occur in absence of alterations of upstream PTEN/PIK3CA/PIK3R1, thereby defining a novel molecular subcategory of NSMP-ECs.

Of interest CTNNB1 gene alterations were observed in relatively equal proportions in all three clusters (27–49% of samples). While other reports suggest adverse clinical outcomes associated with CTNNB1 alterations45,46,47, in our cohort, CTNNB1 hotspot mutations did not show any adverse survival effect. However, the previous studies did not follow the ProMisE classification and were limited to UECs, which may explain the different results observed in our cohort45,46,47.

The molecular clustering of the NSMP-EC may portend clinical and prognostic significance. In our cohort, multivariate analysis showed a statistically significant OS hazard ratio associated with C2 and C3 tumors. Our findings suggest that NSMP-EC should be considered as separate molecular clades rather than a single group of clinically, pathologically, and molecularly heterogenous tumors. Previously, it was suggested that clustering of NSMP-EC based on 1q high level gain can also successfully separate the tumors into prognostic groups6,9, however, in our cohort, 1q copy number status alone was not prognostically significant in multivariate analysis (see Supplementary Table S1).

In conclusion, we have shown here that the NSMP-EC molecular subset of endometrial carcinomas can be further subclassified based on their molecular landscape and that these molecular clusters are associated with meaningful clinical differences. We suggest that cluster 1 and 2 NSMP-EC can perhaps be defined as a PTEN and PI3K altered NSMP-EC group, and the cluster 3 tumors designated as either 1) PTEN wild-type AKT1 altered NSMP-EC or 2) PTEN wild-type KRAS altered NSMP-EC or 3) PTEN wild-type PIK3CA altered NSMP-EC. Further evaluation is needed to confirm the clinical or prognostic significance of these clusters.