Fourteen patients with endometrial cancer were recruited for this study. DNA was extracted using the QIAamp DNA Micro kit (Qiagen, Heidelberg, Germany) following the manufacturer's protocol. Extracted DNA samples were quantified using NanoDrop 2000 spectrophotometer (Thermal Fisher Scientific, Waltham, MA, USA) and a Qubit fluorometer (Invitrogen, Carlsbad, CA, USA). This study was approved by the Institutional Review Board of Kaohsiung Medical University Hospital (KMUH-IRB-970488).

To generate standard exome capture libraries, we used the Agilent SureSelect XT Reagent kit protocol for the Illumina Hiseq paired-end sequencing library (cat. no. G9611A). In all cases, the SureSelect XT Human All Exon Version 4 (51 Mb) probe set was used. We used 50 ng genomic DNA for library construction with the Agilent SureSelect XT Reagent kit. The adapter-ligated sample was purified using Agencourt AMPure XP beads (Beckman Coulter, Brea, CA, USA) and analyzed on a Bioanalyzer DNA 1000 chip. Part of the sample (750 ng) was prepared for hybridization with the capture baits, and the sample was hybridized for 90 min at 65°C, captured with the Dynabeads MyOne Streptavidin T1 (Life Technologies) and purified using Agencourt AMPure XP beads. The Agilent protocol was used to add index tags by post-hybridization amplification. Finally, all samples were sequenced on an Illumina Hiseq system using 100 PE protocol. Metadata were deposited in the NCBI Sequence Read Archive under the accession no. SRP099176.

To filter low-quality reads, we used the FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit) to process the raw read data files. There were two steps of sequence quality processing. The command 'fastq_quality_filter -Q33 -q 30 -p 70'. '-q 30' indicated that the minimum quality score was 30. '-p 70' meant that bases must have a '-q' quality score of 70% or greater. Sequences were retained if both forward and reverse sequencing reads passed the first step.

An efficient sequence alignment tool, Bowtie2, was used to align the retained reads with human genome (Grch38.p2) (20). Based on the results of sequence alignment, reads having only one chromosome location were retained for further analysis. The Genome Analysis Tool kit, is a widely used discovery tool to identify genetic variants based on the results of sequence alignment (21).

Several databases and tools were used to annotate identified genetic variants. dbSNP (b144) is an archive of genetic variation within different species that provides information about each genetic variant (22). ClinVar is a database of clinically significant genetic variants (23). COSMIC (v73) collects somatic mutation information for human cancers (24). The TCGA project collects genomics, methylomic and transcriptomic data across cancer types (25). Integrated mutation prediction software (PolyPhen-2 and SIFT) were used for analyses of the identified variants (26,27). DrGaP was used to identify driver genes and driver signaling pathways (28).

Potential mutations identified by whole-exome sequencing were confirmed by PCR and Sanger sequencing. Specific PCR primers were designed using Primer3 software (primers can be provided on request). The products were sequenced directly with the ABI PRISM terminator cycle sequencing kit v3.1 on an ABI 3130 DNA sequencer (Applied Biosystems, Carlsbad, CA, USA).

Using massive parallel sequencing on a HiSeq platform, we generated ~45 billion bases of effective sequence data with an average read length of 200 bases. After mapping to the human reference genome (Grch38.p2) using the Bowtie 2 alignment tool, we obtained an average depth of coverage for the target regions of 20× for each sample (Table I). The false positive and false negative rates were estimated to be 2.75% (12/437) and 19.69% (76/386), respectively, after confirmed by target sequencing of cancer-related genes (29). Table II provides an overview of our approach to identify variants.

The 143 non-synonymous mutations identified in the present study occurred in 129 genes, and included 141 missense mutations, one nonsense mutation and one frame shift mutation (details can be provided on request). The most frequently mutated genes were PTEN (35.71%; 5/14), KRAS and PIK3R1 (14.29%, 2/14).

Sixty-eight variants already existed in the dbSNP, COSMIC, or TCGA databases and 75 variants in 71 genes did not. In addition, we found 13 different mutations in the same codon of 12 canonical cancer-related genes compared with the TCGA database (CDK8, DPYD, EPHA3, EPHA6, FLNA, MAPK1, MAPK7, PLK1, PTEN, RAPGEF2, RFC4 and ZNF521) (details can be provided on request).

We ordered the 75 new variants according to sequencing depth. Next, we selected 12 variants for Sanger sequencing. Six variants had relatively high depth (208x, 156x, 131x, 105x, 97x and 89x). Several novel mutations were present only in cancer tissue, including MAPT p.C432Y, IL24 p.G192W, MCM6 p.R263H, TSC1 p.D1054G and BIRC2 p.P152S. However, KAT6B p.S1693N was also detected in paired non-cancerous tissues. Six variants had low depth (22x, 21x, 21x, 20x, 20x and 20x). Novel mutations present only in cancer tissue including CIITA p.E1063D, DST p.Q2014H, CASP8 p.L219M and NOTCH2 p.M2054I. FLNA p.V528L and HSPA1L p.P93H were also detected in paired non-cancerous tissues (Fig. 1).

The 1,271 non-synonymous mutations identified in this study, including 1,270 missense mutations and one nonsense mutation, were located at 1,144 genes. In total, 723 variants had previously been reported in the dbSNP, COSMIC or TCGA databases, while 548 variants in 516 genes had not (details can be provided on request).

However, nearly all mutations had been reported in the 542 TCGA endometrial cancer samples. Only three mutations had never been reported (ARMCX4, CUTA and SAP30L). We selected these mutations for Sanger sequencing. CUTA p.T187R and SAP30L p.H77Y were also detected in paired non-cancerous tissues; ARMCX4 p.P2056H was the sole novel mutation that was detected only in cancer tissue (Fig. 2). We also found 49 different mutations in the same codon of 49 non-canonical cancer-related genes compared with the TCGA database (details can be provided on request).

The 1,144 genes were divided into five groups according to mutation frequency. Fifteen non-canonical cancer-related genes had high frequencies of mutation (35.71–50%; 5–7/14) and four variants in three genes (IGSF10, FBXL13 and PRUNE2) were novel. We selected these new genetic variants for Sanger sequencing. We detected IGSF10 p.R455W, FBXL13 p.G313A and PRUNE2 p.S2439I in paired non-cancerous tissues, while IGSF10 p.K188Q was detected only in cancer tissue (Fig. 2).

Six genes were mutated in 4 (28.57%) of the 14 patients. One gene (VPS13C) had a new genetic variant. Sanger sequencing confirmed that VPS13C p.V2017M was a novel mutation detected only in cancer tissue (Fig. 2).

Twenty-two genes were mutated in 3 (21.43%) of the 14 patients, and there were 12 new genetic variants in 8 genes. We carried out Sanger sequencing on the new variants, and found that ARAP2 p.Y1491C, DNAH14 p.D4225V, TLN1 p.D1325E, TREH p.M196T and ZSCAN29 p.G304V were also present in paired non-cancerous tissues. Sanger sequencing also confirmed that DCT p.Y548S, DNAH14 p.G1566C, TLN1 p.A555V, ZNF605 p.P283Q, p.E652A and ZSCAN29 p.P71S were novel mutations present only in cancer tissue (Fig. 2). One variant (TRANK1 p.W2445L) had an incorrect base call.

One hundred and nine genes were mutated in 2 (14.29%) of the 14 patients, and there were 73 new genetic variants in 54 genes. We ordered the 73 new variants according to sequencing depth. Next, we selected 10 variants for Sanger sequencing. Five variants had high depth (163x, 108x, 104x, 99x and 92x). Several novel mutations were detected only in cancer tissue, including MOCOS p.I320T, CMYA5 p.P2149H and PCDH17 p.P976H (Fig. 2). MAGEE2 p.F249L and CRYBG3 p.Q632E were also detected in paired non-cancerous tissues. Five variants had low depth (all 20x). Novel mutations present only in cancer tissue including UGT1A8 p.Y290C and CYFIP2 p.C959Y (Fig. 2), while APOL4 p.A316V and TRIM26 p.Q197H were also detected in paired non-cancerous tissues. One variant (MYO3B p.G230W) had an incorrect base call.

Among low-frequency mutations (7.14%; 1/14) in 992 non-canonical cancer-related genes, there were 458 new genetic variants in 450 genes. We ordered these 458 variants according to sequencing depth. Next, we selected 10 variants for Sanger sequencing. Five variants had high depth (286x, 244x, 223x, 197x and 187x). Several novel mutations were detected only in cancer tissue, including MACF1 p.Q181H, NUDT5 p.G135S, JAKMIP1 p.R352M and PCDHGB4 p.L154I (Fig. 2). MYH10 p.L1048X was also detected in paired non-cancerous tissues. Five variants had low depth (all 20x). Novel mutations detected only in cancer tissue included FAM178A p.T86I, SNX6 p.L38R, IMP4 p.T163M and PCMTD1 p.A41D (Fig. 2). TNFAIP8L3 p.A38T was also detected in the paired non-cancerous tissues.

Of the 14 endometrial cancer samples, 28.57% (4/14) harbored ACMG mutations, including RYR2, MSH6 and TSC1, but did not have mutations in the remaining 53 genes. Mutations in the RYR2 gene were detected in 14.29% (2/14) of endometrial cancer specimens, with mutations at p.G1209R (rs770286824) and p.F4960L (a novel mutation) identified in our endometrial cancer samples. Both the MSH6 mutation at p.Q572H (rs745772518) and the TSC1 mutation at p.D1054G were detected in 7.14% (1/14) of endometrial cancer samples.

Functional annotation of the 1,273 mutated genes was performed using the DrGaP tool. Thirty-seven of the 1,273 mutated genes were found in the Kyoto Encyclopedia of Genes and Genomics (KEGG) cancer pathways (hsa05200), including APC2, BCR, BIRC2, CASP8, CCNA1, CREBBP, CTNNA2, CTNNA3, CTNNB1, E2F3, ERBB2, FGFR2, FOXO1, FZD6, HHIP, ITGA2, ITGA3, KRAS, LAMA1, LAMA2, LAMA3, LAMA5, LAMC1, LAMC2, MAPK1, MAPK10, MSH6, MTOR, PDGFRA, PIK3CA, PIK3R1, PLCG2, PTEN, SOS2, STAT5A, TCF7 and TRAF5. We used Sanger sequencing validation to confirm new variants Fig. 3). Moreover, we identified several cellular pathways that were altered in endometrial cancer tissues and we found that each sample had at least two pathways involved in the carcinogenesis of endometrial cancer (Fig. 4).