All patients were registered at the Clinical Research of Cancer Gene Panel Analysis of Gynecologic Cancers Study, which was conducted from January 2019 to October 2021 at the Kagoshima University Hospital. The clinical samples used in this study were approved by the Ethics Committees for Clinical and Epidemiologic Research at Kagoshima University (approval number: 180215) and written informed consent was obtained from all participants. Among the 155 cases entered, 36 cases, including 16 G2 and 3 G3 EC cases, 3 DC cases, 4 CS cases, 9 SC cases, and 1 large cell neuroendocrine carcinoma (LCNEC) case with areas of solid proliferation were included in this study. Clear cell carcinoma (CCC) is considerably less frequent in the Gynecologic Cancers Study in our hospital, therefore, CCC cases were excluded from this study. No patients with undifferentiated carcinoma were included.

Resected tissues obtained by hysterectomy were fixed with 10% neutral phosphate-buffered formalin, routinely processed for paraffin embedding, and sectioned for hematoxylin and eosin (HE) staining, IHC, and NGS. Pathological diagnoses were made according to the World Health Organization (WHO) classification system [1]. The following criteria were used for diagnosing SC: carcinoma showing complex papillary, glandular and/or solid growth patterns with marked nuclear pleomorphism; CS: admixture of high-grade müllerian type carcinomatous elements and apparent mesenchymal morphology, as determined using homologous (CD10, desmin, alpha-smooth muscle actin) or heterologous (myogenin, MyoD1, S-100) IHC markers [12, 13]; DC: biphasic tumor consisting of EC and undifferentiated nested or trabecular architecture with no gland formation; EC: glandular proliferation with smooth luminal outline, composed of columnar cells with pseudostratified nuclei. Grading criteria of EC: Grade 1, 2, and 3, respectively, exhibit ≤5%, 6–50%, and >50% solid growth was used.

The dedifferentiated parts of DC are positive for epithelial marker expression in scattered cells, and mesenchymal elements of CS are weakly positive or negative for these markers; hence, diffuse expression of keratins (AE1/AE3 and CAM5.2) and EMA was used to differentiate endometrial cancers [1].

The antibodies used for IHC are listed in Online Resource 1. MMR-proficient (MMR-p) was defined as positive nuclear staining for all MMR proteins (MLH1, PMS2, MSH2, and MSH6). MMR-d was defined as the complete absence of nuclear staining for any MMR proteins [14]. p53 expression resulting in scattered nuclear staining with variable intensity was categorized as wildtype (wt p53) pattern, whereas diffuse and strong nuclear overexpression or a complete loss of expression was defined as mutation (mt p53) pattern. For ER, vimentin, WT-1, CycD1, ARID1A, and PTEN expression, the percentage of marker-positive areas assessed in the tumor section were evaluated in 10% increments. In areas of glandular or carcinomatous and solid proliferation, the expression of ARID1A and PTEN was categorized as very low at <10%, and as lost at <5%.

A cancer panel was redesigned by making a minor modification of the previous panel [11] to include 56 cancer-related genes and 17 microsatellite foci selected from the QIAseq Targeted DNA Custom Panel (Qiagen, Reston, VA, United States), with 2,640 primers for the regions of interest (194,131 bp) and an average exon coverage of 99.87% (Online Resource 2). Whole blood DNA was extracted using QIAamp DNA Blood Mini Kits (Qiagen). Cancer DNA was obtained from three to six sections (10 μm thickness) of FFPE tissues, representing more than 30% of the cancerous tissue. When cancer is composed of various histology, several tissue sections were subjected to NGS analysis if separate sampling of each element was possible by micro or macrodissection. The FFPE sections were first incubated with proteinase K (Promega, Madison, WI, United States) for 15 h at 70°C, followed by incubation at 98°C for 1 h in the lysis buffer (Promega). Following centrifugation (12,000 × g, 5 min at 4°C), the supernatants were applied to a Maxwell^® RSC DNA FFPE kit and the Maxwell 16 system (Promega). The concentrations of extracted DNA were measured using Qubit 3.0 dsDNA BR assay kits (Life Technologies, Grand Island, NY, United States), and the DNA quality was monitored using the QIAseq DNA quantiMIZE kit (Qiagen). DNA with a quality check score <0.04 was considered high-quality DNA [11]. NGS was performed using a MiSeq sequencer (Illumina, San Diego, CA, United States) as previously described [11].

Sequence data were annotated as previously described using the Qiagen Web Portal service (https://www.qiagen.com./us/shop/genes-and-pathways/data-analysis-center-overview-page/) and Mitsubishi Space Software (Amagasaki, Hyogo, Japan) [11, 15]. The COSMIC database and human genome reference GRCh37 (hg19) (https://www.ncbi.nlm.nih.gov/assembly/GCF_000001405.13/) were used as references. The sequence data obtained from whole blood DNA were used only as a reference, and germline analysis was not performed.

To calculate the copy number (CN) from the baseline data used for counting correction per amplicon, the number of reads sequenced in each amplicon was counted, and the reads per million (RPM) value was determined. The RPM coefficient of variation (CV), mean, and median value per amplicon in at least 100 FFPE samples were calculated. The RPM median of the amplicons with a CV < 0.34 and mean > 10 was set as the baseline. The number of reads sequenced in each 56‐panel amplicon was counted in the sample to calculate the CN of each sample. The baseline ratios {log2 ratio [ = log2 (sample RPM/baseline RPM median)]} in amplicons that satisfied the conditions of CV < 0.34 and mean >10 were counted, and the overall SD and median value of the log2 ratio for each gene was calculated. The genes with a log2 ratio median value >2 SD were categorized as amplified, while the genes with a log2 ratio median value <−2 SD were categorized as gene loss [16].

Missense mutations with more than 10% variant allele frequency, including nonsynonymous mutations and internal deletions, were counted as somatic mutations. The TMB was calculated as the number of single nucleotide variants million per base pairs (Mbp) of the DNA sequence [17, 18], and the MSI scores were determined using MSIsensor (ver. 1.0) [19, 20]. To determine the cutoff values of MSI and TMB using receiver operator characteristic curves, 59 samples from 41 endometrial cancer cases were used.

All values were expressed as the mean ± standard error. Significant differences were identified using student’s or Welch’s t-tests and were considered significant at p < 0.05.

The IHC results regarding the MMR proteins, p53, ARID1A, PTEN, WT-1, ER, CycD1, and vimentin are summarized in Table 1 (extracted) and Online Resource 3 (detailed), and representative photomicrographs are shown in Figure 1. Among 19 EC cases, 16 showed the wt p53 pattern, and three showed the mt p53 pattern. Six EC and three DC cases showed loss of MMR protein expression. PTEN expression was very low (<10% area showing expression) or lost (<5% area showing expression) in all DC (3/3) and most EC cases (12/19), while ARID1A was lost less frequently in DC (1/3) and EC cases (7/19). All SC (9/9) and some CS (2/4) cases exhibited the mt p53 pattern in carcinomatous and sarcomatous elements. Three CS cases were MMR-p, whereas one was MMR-d. Almost all SC cases were MMR-p (8/9). PTEN expression was lost in one CS and five SC cases. ARID1A expression was well preserved in CS (3/4) and SC (8/9) cases. One LCNEC exhibited MMR-p and had mt p53 as well as diffuse PTEN and ARID1A expression. The expression of WT-1, vimentin, CycD1, and ER was variable in all carcinoma types. The genomic correlation between CycD1, ER, PTEN, and ARID1A is described later.

The TMB values of wildtype and mutated POLE cases were 36.3 ± 3.8 and 175.1 ± 61.0 (p = 0.026), respectively, and the cutoff value for TMB-ultrahigh (TMB-UH) was calculated as 72. Excluding the POLEmut cases, the TMB values of MMR-p and MMR-d cases were 22.5 ± 3.0 and 63.0 ± 5.1, respectively (p < 0.001), and the cutoff values for TMB-high (TMB-H) and -low (TMB-L) were calculated to be 42. Similarly, the MSI values of MMR-p and MMR-d cases were 3.6 ± 0.5 and 38.7 ± 2.0, respectively (p < 0.001), and the cutoff values for the differentiation of MSI-high (MSI-H) and -low (MSI-L) were estimated to be 13. The distribution of TMB and MSI scores of the tested endometrial cancers for cutoff value estimation is shown in Figure 2.

Genomic profiles of 36 cases are shown in Table 1 (extracted) and Online Resource 4 (detailed). All three DC cases (no. 20–22) carried MLH1 and POLD1 mutations with TMB-H and MSI-H, and exhibited variable mutations in PTEN, ARID1A, PIK3CA, PIK3R1, and CTNNB1. The TERT coding region was mutated in two DC cases, although the promoter was not. All 19 EC cases exhibited common mutations in PTEN, ARID1A, CTNNB1, PIK3CA, and PIK3R1. Among the 19 cases, 16 cases (except for case no. 2, 13, and 19) included MMR-p and MMR-d cases and exhibited higher TMB scores than did the SC cases (Table 2). Six EC cases that were MMR-d had higher TMB and MSI scores than the ten MMR-p cases and were classified as TMB-H and MSI-H (Table 3). Three EC cases (no. 2, 13, and 19) that harbored a POLE mutation were classified as TMB-UH and MSI-L.

Among the four CS cases (no. 32–35), two contained mutations in TP53 and exhibited mt p53 expression. Furthermore, three CS cases, one with mutated and two with wildtype TP53, exhibited amplification of MYC, CCNE1, or CCND1. All CS cases were MMR-p (MSI-L and TMB-L) and showed no mutations in MMR, except for one case that exhibited a loss of MLH1 and PMS2 expression (MSI-H and TMB-H) without any mutations in MLH1 and PMS2. IHC and NGS profiles of eight SC cases, displaying mt p53 expression and MMR-p characteristics were classified as MSI-L and TMB-L. One CS case (no. 32) showed ERBB2 amplification. Among the eight SC cases, five showed amplification in MYC, CCNE1, or CCND1. One SC case (no. 27) exhibited MMR-d with wildtype MMR and was classified as TMB-UH and MSI-H. LCNEC (no. 36) exhibited an SC-like profile with mutations in TP53 and PIK3CA. Other mutations such as BRCA1, BRCA2, and ATM were detected in some cases (nos. 2, 4, 5, 18, 19, 20, 22, 27, and 36) (Online Resource 4).

For the tumors composing of heterogenous elements, the genomic profile was evaluated separately for each element suing two or three FFPE sections in six cases of EC (no. 2, 14, 15, and 17–19), two of CS (nos. 34 and 35), one of SC (no. 26), and three of DC (nos. 20, 21, and 22). The genomic profiles were not exactly matched between the heterogenous elements, but were not so different as much as TCGA classification was revised (Online Resource 4).

The correlation between the genomic and IHC analyses of ARID1A and PTEN is presented in Online Resource 5. Most CS (3/4) and SC cases (8/9) exhibiting high ARID1A expression (>90% of the positive area) had no mutations. Fourteen EC cases (14/19) harbored frameshift, nonsense mutations or splice variants. Among these, eight cases exhibited a loss of ARID1A expression (<5% of the positive area). PTEN expression was lost in one CS case with gene alterations (1/4). Although SC cases did not display PTEN mutations (9/9), five cases did not express PTEN. Among 19 EC cases, 18 carried PTEN mutations and 13 accompanied by a loss of PTEN expression (<5% of the positive area). As previously reported [21, 22], ARID1A expression was almost consistent with ARID1A mutations in EC, CS, and SC cases, and that of PTEN was consistent with PTEN mutations in EC cases. For CycD1 and ER, a correlation between genomic and IHC analysis results could not be established due to the limited number of cases exhibiting these gene mutations (data not shown). The status of TP53 mutation matched well with the p53 IHC results in EC, SC, CS, and LCNEC cases (33/36), except for two EC cases (no. 6 and 11) and one DC case (no.21) that exhibited wt p53 IHC but harbored TP53 mutations.

Based on TCGA classification [23], EC cases were classified as POLEmut (3/19), MMR-d (6/19), NSMP (8/19), and p53mut (2/19), and all DC cases were categorized as MMR-d (3/3). CS cases were classified as MMR-d (1/4), p53mut (2/4), and NSMP (1/4). Most SC cases were classified as p53mut (8/9), and one was classified as MMR-d (1/9) (Table 4).