Tumor mutational profile of triple negative breast cancer patients in Thailand revealed distinctive genetic alteration in chromatin remodeling gene

Study population

The study protocol was approved by the Siriraj and King Chulalongkorn Memorial Hospital Institutional Review Boards (Protocol No. 175/2559 and 642/2557). The study was conducted according to the Good Clinical Practice and the Declaration of Helsinki. All participants provided written informed consent. One hundred and sixteen Thai patients who were diagnosed with primary TNBC and treated at both hospitals between 2014 and 2017 were included. All 116 patients were female. The average age at diagnosis was 56.47 ± 11.90 years (±SD) with an age range between 25 and 79 years. The BMI was 26.28 ± 5.89 kg/m2. Majority of the patients (87%) were categorized as early stage breast cancer (20% as stage I and 67% as stage IIa). Fifty-five patients had follow-up period up to 3 years. There were six clinical relapses within the 3-years follow-up period; four cases were in stage IIa and two cases were in stage IV. One patient died during follow-up period whose cause of death did not appear to be cancer-related.

Primary tumor tissues and lymph nodes were surgically removed as a standard treatment at Siriraj and King Chulalongkorn Memorial Hospitals and examined by board-certified pathologists. The tissues were dissected for histological diagnosis and immunohistochemistry staining then archived as FFPE tumor block. TNBC subtype was defined by absence of estrogen receptor (ER), progesterone receptor (PR) and human epithelial growth factor receptor 2 (HER2) by immunohistochemistry staining with appropriate positive control. No amplification of HER2 was also confirmed by in-situ hybridization. Independent pathologists determined only the samples of primary TNBC tumors, which had more than 50% tumor content after dissection, to be analyzed in this study. The FFPE primary tumors were sectioned into 10 micrometers using a new blade and preserved in 1.5 mL Eppendorf tubes. Blade was changed for every tissue block to prevent the contamination of DNA.

Tumor genome sequencing and variant calls

The genomic DNA (gDNA) was extracted using Qiagen DNeasy DNA Isolation Kit (Hilden, Germany). FFPE gDNA (50–150 ng) was converted into libraries and enriched for whole exome sequencing using Agilent’s SureSelect Human All Exon V5 + UTR Sample Prep kit. Sequencing was performed on the Illumina HiSeq 2500/4000 platform with average 300× sequencing depth. Genomic data were then processed through bioinformatics platform and knowledge base (Wuxi NextCODE Genomics, Boston, MA, USA) to identify genomic alterations including single nucleotide polymorphisms/substitutions (SNPs) and small insertions/deletions (indels). A threshold of 5% allelic fraction was used for SNPs and indels. Any variants presented at allele frequency above 1% in dbSNP, 1,000 Genomes and ExAC databases were removed. To assess somatic status of mutations in a tumor-only setting, we used both MuTect2 (Broad Institute, Boston, MA, USA) and VarScan2 (Washington University, St. Louis, MO, USA) on the aligned sequence data to determine somatic variants. All variants are further annotated with the extensive pipeline including COSMIC data annotations. Additional annotation includes HGMD Professional, ClinVar, OMIM and multiple missense functional predictors including polyphen2, SIFT, LRT, MutationTaster, MutationAssessor and CADD. The effect of the sequence variants on all protein-coding genes in the RefSeq database was further annotated using the Variant Effect Predictor, which predicts the consequence of each sequence variant on all neighboring RefSeq genes based on a set of 35 consequence terms defined by the Sequence Ontology (McLaren et al., 2016). Only variants predicted to cause strong and moderate alteration on gene functions, such as stop gained/lost variants, frameshift, indels, donor/acceptor splice variants, initiator codon variants, missense variants, in-frame indels and splice region variants were selected for analysis. Finally, the results were manually reviewed by molecular geneticists. These analysis methods applied to the cancer genome atlas (TCGA) tumor variants and expression data have distinguished different molecular or histologic subtypes of breast cancer at over 97% accuracy.

Data comparison with other breast cancer cohorts

To compare frequency of cancer gene alterations between this study and previously published data, breast cancer mutation data from TCGA, METABRIC and French cohorts were downloaded from the cBioPortal for Cancer Genomics (http://www.cbioportal.org). TCGA cohort consisted of cancer genome data form primary breast cancer patients in the United States (Cancer Genome Atlas Network, 2012). METABRIC cohort data were collected from primary breast cancer patients in the United Kingdom and Canada (Pereira et al., 2016). French cohort data were primarily from the patients with metastatic breast cancers from four different prospective trials in France (Lefebvre et al., 2016). Cancer gene mutation data from TCGA database were selected only from breast invasive ductal carcinoma that was classified as PAM50 basal subtype, which is closely related to triple negative subtype of breast cancer, while data from METABRIC and French cohorts were selected from samples with negative ER, PR and HER2, similar to this study. The comparison of mutation frequency of each cancer genes was trimmed to 173 genes due to data limitation from 173-gene panel in METABRIC cohort.

Statistical analysis and data visualization

Genomic characteristics were compared across cohorts using one-way analysis of variance for continuous variables. The prevalence of somatic mutations was compared across cohorts. Descriptive statistics was used to show the average age and BMI of population. Statistical analyses were conducted using SAS software (version 9.4; SAS Institute, Cary, NC, USA). A two-tailed p value less than 0.05 was considered significant. Landscape of co-occurrence and mutual exclusion of cancer gene mutations was generated with OncoPrinter (version 1.0.1; cBioPortal for Cancer Genomics, New York, NY, USA). Mutational spectrum of three most commonly mutated genes in lollipop plot format was generated with MutationMapper (version 1.0.1; cBioPortal for Cancer Genomics, New York, NY, USA) (Gao et al., 2013).

Comparison of gene alterations between Thai TNBC and other TNBC cohorts

A total of 1,088,237 variants were detected from the whole exome sequencing data of 116 patients. Variant filtering with MuTect2 and VarScan2 identified 81,460 somatic variants that passed data analysis algorithms. When all 969 known or potential cancer-associated genes were examined, only 5,906 somatic variants were identified. We found an average of 222 variants per sample (range 104–388 variants) with an average of 20 altered cancer genes (range 5–35 genes). The 10 most commonly mutated genes were TP53, KMT2C, SYNE1, PIK3CA, BRCA1, NF1, BRCA2, PTEN, RB1 and ARID1B (Fig. 1; Tables 1 and 2). Compared to the cancer gene mutation frequencies from TCGA, METABRIC and French cohorts, Thai TNBC had significantly higher mutation frequencies in KMT2C, SYNE1, PIK3CA, NF1, PTEN, BRCA1 and BRCA2. On the contrary, significant differences were observed only in SYNE1 and BRCA1 among TCGA, METABRIC and French cohorts.

Driver gene analysis

We performed analysis of the filtered cancer genes to identify potential genes of interest. We first examined for genes mutated in multiple samples and found that 14 genes were mutated in at least 10% of the samples. SYNE1 was excluded from analysis due to unclear role in cancer. TP53 was the most frequently mutated gene with variants found in 88 samples. KMT2C, which encodes for the lysine methyltransferase mixed-lineage leukemia (MLL3), was the next most frequently mutated gene with mutations in 67 samples. The third most frequently mutated gene was PIK3CA, which was shared by 27 samples. Together, 111 of the 116 samples had an alteration in at least TP53, KMT2C or PIK3CA (Fig. 2). The most commonly recurring mutation was the PIK3CA H1074R mutation, which was found in the majority of the PIK3CA mutant samples (16 out of 27 samples; 59%). Within these genes, we also found a number of recurrent variants including the previously reported mutations in KMT2C S338L (21 samples; 31%), TP53 R175H (10 samples; 11%), R248Q (five samples; 6%), Y220C (four samples; 5%) and R273H (four samples; 5%) and PIK3CA K733R (six samples; 22%) and E545K (four samples; 15%) (Fig. 3).

TP53 mutation status and co-occurring mutations

As previously shown, TP53 was the most commonly mutated gene in the cohort with 88 of the 116 samples (76%) containing a mutation. To identify roles of TP53 and its association with the other two most commonly mutated genes; KMT2C and PIK3CA, we subdivided the cohort as TP53 wild-type and TP53 mutant groups. In the TP53 wild-type samples (n = 28), the PI3K pathway appeared to be a predominant driver with 19 samples (68%) containing a mutation in either the PIK3CA gene or PI3K pathway members, including PIK3C2B, PIK3CG and PTEN. In the TP53 mutant group, 67 samples (76%) had one or more mutations in genes encoding chromatin remodeling proteins, including ATRX, DNMT3A and KMT2C, which have been reported to be involved in cancer. This mutation co-occurrence may suggest a complex interplay in TNBC. However, there was no association between TP53, PIK3CA or KMT2C mutation status and cancer staging.

Tumor mutation burden

One of the biological hallmarks in cancer is genome instability. Genome alterations, which either involve in carcinogenesis or occur as a result of widespread genome instability, can create neoantigens and trigger immune response. We used the variant data to calculate the tumor mutation burden for each sample and employed the same filtering scheme accounted for all variants including both nonsynonymous and synonymous calls. Tumor mutation burden was defined as the number of somatic base substitutions, and indels per megabase of coding genome sequence examined. Synonymous mutations are counted in order to reduce sampling noise. Though majority of synonymous mutations are not likely to cause tumor immunogenicity, they may reflect mutational processes that will also result in nonsynonymous mutations and neoantigens elsewhere in the genome (Diederichs et al., 2016; Supek et al., 2014). We found that Thai TNBC has an average tumor mutation 7.3 variants per megabase (95% CI [6.9–7.6]), which consists of nonsynonymous mutation 3.9 variants per megabase (95% CI [3.7–4.1]) and synonymous mutation 3.4 variants per megabase (95% CI [3.3–3.6]). This data showed significantly higher tumor mutation burden than median value (3.6 variants per megabase) found in invasive ductal breast carcinoma cohort (Chalmers et al., 2017).