Exome sequencing reveals a distinct somatic genomic landscape in breast cancer from women with germline PTEN variants

Summary

Germline PTEN variants (PTEN hamartoma tumor syndrome [PHTS]) confer up to 85% lifetime risk of female breast cancer (BC). BCs arising in PHTS are clinically distinct from sporadic BCs, including younger age of onset, multifocality, and an increased risk of second primary BCs. Yet, there is no previous investigation into the underlying genomic landscape of this entity. We sought to address the hypothesis that BCs arising in PHTS have a distinct genomic landscape compared to sporadic counterparts. We performed and analyzed exome sequencing data from 44 women with germline PTEN variants who developed BCs. The control cohort comprised of 497 women with sporadic BCs from The Cancer Genome Atlas (TCGA) dataset. We demonstrate that PHTS-derived BCs have a distinct somatic mutational landscape compared to the sporadic counterparts, namely second somatic hits in PTEN, distinct mutational signatures, and increased genomic instability. The PHTS group had a significantly higher frequency of somatic PTEN variants compared to TCGA (22.7% versus 5.6%; odds ratio [OR] 4.93; 95% confidence interval [CI] 2.21 to 10.98; p < 0.001) and a lower mutational frequency in PIK3CA (22.7% versus 33.4%; OR 0.59; 95% CI 0.28 to 1.22; p = 0.15). Somatic variants in PTEN and PIK3CA were mutually exclusive in PHTS (p = 0.01) but not in TCGA. Our findings have important implications for the personalized management of PTEN-related BCs, especially in the context of more accessible genetic testing.

Germline PTEN variants cause PTEN hamartoma tumor syndrome (PHTS), a hereditary cancer predisposition syndrome conferring up to 85% lifetime risk of female breast cancer. Brewer et al. report that the somatic mutational landscape of PHTS-derived breast cancers is distinct from sporadic counterparts, providing insights into distinctive biology in this group.

Introduction

Hereditary cancer syndromes are caused by germline variation in cancer susceptibility genes. Cancers arising in this context tend to be more aggressive and multifocal and occur at a younger age than sporadic counterparts. PTEN, encoding phosphatase and tensin homolog (MIM: 158350), is a tumor suppressor gene¹ and is among the most commonly somatically altered genes in diverse sporadic malignancies including breast cancer (BC).² PTEN hamartoma tumor syndrome (PHTS) is a molecular diagnosis that encompasses individuals harboring a germline PTEN mutation, which causes heritable predisposition to multiple types of cancer including breast, thyroid, kidney, endometrial, and colon cancers and melanoma.³

In women with PHTS, BC represents the malignancy with the highest lifetime risk, up to 85%, compared to 12.9% in the general population.^{3, 4, 5, 6} Furthermore, women with PHTS have a significantly higher incidence of second primary malignant neoplasms, particularly BC.⁷ With a tremendous increase in clinical genetic testing, the prevalence of PHTS is expected to increase. Yet, we know nothing about the clinical characteristics and somatic tumor landscape of PHTS-BC. In order to identify distinct underlying biological and molecular processes, we analyzed exome sequencing data from 44 primary BCs arising in the setting of germline PTEN variants (PHTS) and compared them to 497 sporadic BC samples from the Cancer Genome Atlas (TCGA).

Subjects and methods

Research participants

Approved by the Cleveland Clinic’s institutional review boards (IRBs), written informed consents were obtained from each individual enrolled under the study protocol. Among 6,934 research participants prospectively accrued from September 1, 2005 to September 10, 2020, we identified 3,066 female participants with a personal history of breast cancer (BC). Of these, 130 had germline PTEN variants. We then identified 44 women with appropriate consents for acquisition of biospecimens and whose tissues representing BC were available for sequencing.

Germline PTEN mutation and deletion/duplication status was confirmed by clinical genetic testing and verified by polymerase chain reaction (PCR) and/or multiplex ligation-dependent probe amplification (MLPA). PTEN pathogenicity predictions were derived from genetic test reports from orthogonal testing at CLIA-certified facilities, ClinVar database classifications, and/or the ClinGen gene-specific criteria for PTEN variant curation.⁸ For unreported variants, Franklin by Genoox, an online variant interpretation tool based on machine learning and the American College of Medical Genetics and Genomics (ACMG) criteria, was used in conjunction with expert opinions at the Genomic Medicine Institute at the Cleveland Clinic (Cleveland, OH, USA). We classified variants in the PTEN promoter region as mutation positive if the clinical presentations were associated with Cowden syndrome or if the variants are known to affect PTEN functions.^{3,9, 10, 11} We further classified germline PTEN variants into two groups: (1) Tier 1 variants, which are classified as pathogenic or likely pathogenic by ClinVar or Genoox or either large deletions, nonsense, or frameshift mutations that are predicted to be damaging either by impairing PTEN function or transcript stability, and (2) tier 2 variants, which are classified as variants of uncertain significance or likely benign by ClinVar or Genoox or they have never been reported with sufficient supporting evidence. Tier 2 variants were initially classified as mutations or determined as potentially deleterious on the basis of clinical presentations and functional evidence at the time of the individuals’ enrollment into the study. We further defined tier-1-derived tumors as TIER-1 and tier-2-derived tumors as TIER-2. The baseline Cleveland Clinic score (CC score), which is a semi-quantitative surrogate of age-related PHTS phenotypic burden,⁹ was extracted from the Cleveland Clinic Genomic Medicine Institute’s relational database (Table S1).

Original formalin-fixed, paraffin-embedded (FFPE) samples representing primary BC were obtained from respective healthcare institutions where the pathology specimens were collected. Two of the 44 cases had exposure to systemic chemotherapy prior to sample collection from which the source DNA originated. Clinical information including tumor type, grade, stage, and hormone receptor status was obtained by reviewing surgical pathology reports and pertinent medical records (Table S1).

DNA extraction

DNA was extracted from the FFPE samples with QIAamp DNA FFPE Tissue kit (Qiagen, MD, USA). Briefly, tissues from FFPE blocks were deparaffinated with xylene and crude DNA was precipitated with 100% ethanol. Following complete proteolysis of the samples with Proteinase K at 56°C, DNA was extracted and purified with the column method according to the manufacturer’s protocol with slight reagent volume modifications. For matched germline samples, we obtained blood-derived genomic DNA originating from whole blood from the PHTS-affected individuals from the Genomic Medicine Biorepository of the Genomic Medicine Institute at the Cleveland Clinic (Cleveland, OH, USA) following standard procedures.

Processing of extracted DNA samples

Samples with sufficient DNA yields and quality were subjected to exome sequencing. The DNA concentration was measured with the Qubit Fluorometer dsDNA HS (High Sensitivity) Assay kit (Thermo Fisher Scientific, Waltham, MA, USA). While the ideal DNA concentration for sequencing library preparation was considered to be 30–40 ng/μL, the range of DNA concentrations of submitted samples was 9.6–68.4 ng/μL and 19.0–98.4 ng/μL, and the range of sample volumes submitted was 30–45 μL and 30–60 μL for tumor and germline/blood samples, respectively.

Exome sequencing

The tumor-blood DNA pairs were sent to the Broad Institute Genomic Services (Cambridge, MA, USA) for next-generation sequencing (NGS). The Broad Institute created libraries from the submitted DNA samples and used the Illumina HiSeq platform to generate NGS data. Of the 44 tumor-blood samples, 28 were processed with the Illumina Somatic Exome protocol and the remaining 16 with the TWIST Somatic Exome protocol (pair-end sequencing with read length range of 67 bp to 140 bp). The Illumina Somatic Exome protocol had target depths of 20× and 50× for the blood and tumor samples, respectively. For the TWIST Somatic Exome protocol, the target depth was 100× for both blood and tumor samples. The raw data were quality controlled, aligned, and sorted through a standard NGS pipeline at the Broad Institute. Reads were aligned to the reference human genome GRCh37/hg19 with the Burrows-Wheeler aligner (BWA)-ALN aligner (version 0.5.9).¹² Local realignment, duplicate removal, and base quality score recalibration were performed with the Genome Analysis Toolkit and Picard per the Broad Institute standard protocol.¹³ The processed sequencing data, derived from both tumor and blood samples, were delivered as binary alignment map (BAM) files.

Sporadic breast cancer cohort

The control cohort data were derived from The Cancer Genome Atlas (TCGA) breast cancer dataset from the Genomic Data Commons (GDC). BC cases with available exome sequencing data were selected. Cases with germline mutations in known cancer susceptibility genes (Table S2) were identified on the basis of previously published data¹⁴ and excluded. Pertinent clinical information of the selected cases was obtained from Nationwide Children’s Hospital dataset, which is publicly available from the GDC portal(universally unique identifier [UUID] 8162d394-8b64-4da2-9f5b-d164c54b9608). The final control cohort comprised of 497 women with no known germline mutations in cancer susceptibility genes and, thus, who developed sporadic BC (estrogen receptor [ER] positive/human epidermal growth factor 2 [HER2] negative: n = 308; ER+/HER2+: n = 80; ER−/HER2+: n = 23; triple negative breast cancer [TNBC]: n = 86). The original input files (BAMs) of tumor and matched blood samples, aligned to reference human genome GRCh38/hg38, were downloaded from the GDC archive website for bioinformatics analyses.

Somatic variant calling

We analyzed the BAM files by using our in-house bioinformatics pipeline, which is composed of the following steps: (1) variant calling for somatic single-nucleotide variants (SNVs) and small insertions and deletions (indels); (2) variant annotation; (3) data visualization; and (4) downstream analyses. To optimize recall and precision in the variant calling step, we used three variant callers: (1) the Genome Analysis Toolkit (GATK) Mutect2 (version 4.1.9.0)¹³; (2) Strelka (version 2.9.10);¹⁵ and (3) VarDict (version 1.8.2).¹⁶ For each variant caller, we employed the default settings. Outputs from these variant callers were combined by SomaticCombiner,¹⁷ which generates a consensus set of somatic variants as a single variant call format (VCF) file. Somatic variants that passed all the SomaticCombiner filters with a read depth of at least 20× for normal (blood) and 20× for tumor samples were included for visual inspection with the Integrative Genomics Viewer (IGV) (version 2.6.3). The cut-off of 20 (Phred-scaled quality score) was used for both mean base call quality and mean mapping quality. Following the filtering step, the detected variants were annotated with GATK4/Funcotator (version 4.1.9.0) according to the GATK best practice workflow of the Broad Institute, which generates mutation annotation format (MAF) files.

Variant filtration and selection

For targeted analysis, we aggregated lists of genes associated with BC from the TCGA publication,¹⁸ NCCN Genetic/Familial High-Risk Assessment: Breast and Ovarian guidelines (version 1.2022-August 11, 2021) and 22 previously reported gold standard (GS) genes for BC.¹⁹ We further searched for preliminary BC susceptibility genes and targetable BC-associated genes from the literature and compiled a list to examine mutational status in these genes.^{19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38} A total of 84 BC-associated genes were included in the final list (Table S3). Variants in all BC-associated genes were manually reviewed with the IGV with the original BAM files as input. The determination whether the variant reviewed should pass or fail was made with modified criteria based on the standard operating procedure (SOP) for somatic variant refinement of tumor-normal pair sequencing data.³⁹ During visual inspection, SNVs and indels were allowed to pass if the variant allele frequency (VAF) was greater than 5% and they were not classified as benign or likely benign. Variants with VAFs between 2% and 5% were retained if they were previously reported as pathogenic or likely pathogenic. We used an online variant interpretation tool, Franklin by Genoox, to assist with variant classification. To remove variants found at relatively high frequencies in the general population (minor allele frequency [MAF] greater than 0.001 [0.1%]), we used the allele frequency information for female samples from a large population database (field “gnomAD_exome_AF_female” from the genome Aggregation Database [gnomAD] version 2.1, Funcotator Data Source version 1.7.20200429 s). Sex determination for this gnomAD dataset is based on both X heterozygosity and Y coverage and it contains a wide range of ancestries.

Identification of frequently mutated genes and gene interactions

In order to identify the most frequently mutated BC-associated genes, we used the maftools package (version 2.10.0)⁴⁰ in R to summarize and visualize our dataset. R version used was R-4.1.2 for windows. Samples containing variants in BC-associated genes that passed the manual review on IGV were included in the analyses. We used metafor package in R (version 3.4.0)⁴¹ to identify which frequently altered genes are statistically significantly different in frequency between PHTS and sporadic BC groups from TCGA. Forest plots were created with GraphPad Prism version 9.0 (GraphPad Software, San Diego, CA, USA). We performed the subset analyses to compare TIER-1 (n = 31) and TIER-2 (n = 13) against TCGA BCs (n = 497). To investigate the presence of gene sets that were co-occurring or mutually exclusive in PHTS-all (TIER-1 and TIER-2 BCs combined), we used the “somaticInteractions” function in maftools to create a pairwise correlation matrix plot based on Fisher’s exact test to identify significant pairs of genes.

Driver gene identification

To uncover cancer driver genes, we performed a cancer-type agnostic gene ranking analysis by using cDriver (version 0.4.2)¹⁹. The input files included both synonymous and non-synonymous variants that passed all the SomaticCombiner filters, with read depth of greater than 20× for both tumor and blood, and with mean base call quality and mean mapping quality above 20 (Phred-scaled quality). An optimized VAF cutoff of 2% was used for this analysis because we observed that known pathogenic variants in genes such as PIK3CA (MIM: 114480), although rare, can occur with VAF lower than 5%. To generate purity and ploidy information, we used FACETS (version 0.6.1), an R package.⁴²

Mutational signature identification

To characterize the genome-wide mutational landscape, we investigated mutational signatures, including apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC) enrichment, and single base substitution (SBS) signatures. Variants that passed all the SomaticCombiner filters with minimal read depths of 20× for both tumor and blood samples and a VAF greater than 5% were included. We used the BSgenome package (version 1.62.0) to create a matrix of nucleotide substitutions that served as input for APOBEC enrichment and SBS signature analyses. APOBEC enrichment scores were estimated with a previously described method.⁴³ We then used maftools “plotApobecDiff” function to create a plot showing the difference in mutational load of tCw (W = A or T) motif as well as differentially mutated genes between APOBEC-enriched and non-APOBEC-enriched samples. For SBS signature analyses, we used non-negative matrix factorization, NMF package in R (version 0.24.0),⁴⁴ to measure the goodness of fit in order to determine the number of best match clusters. Then, we used maftools “extractSignatures” function to extract mutational signatures to compare against known SBS signatures in the Catalogue of Somatic Mutations in Cancer (COSMIC) legacy and SBS signature database. Finally, we used maftools “plotSignatures” function to plot best match SBS signatures detected in our dataset.

Tumor mutational burden estimation

Tumor mutational burden (TMB) was calculated by taking the total number of non-synonymous variants divided by a capture size of 30. This capture size is based on the size of the coding region, for which 30 Mb was employed. This approximation is a generally accepted assumption that approximately 1% of the whole human genome (an approximate total of 3 billion base pairs) is protein coding.⁴⁵ Non-synonymous variants that passed all the SomaticCombiner filters, had minimal read depths of 20× for both tumor and blood samples, and had VAF greater than 5% were included. The cut-off of 20 (Phred-scaled quality) was used for both mean base call quality and mean mapping quality. The TMB was calculated for each sample, which was then log2 transformed for optimum scatterplot visualization. We used Kruskal-Wallis test to perform an overall comparison of all eight BC groups (PHTS-all, TIER-1, TIER-2, all sporadic BC, ER+/HER2−, ER+/HER2+, ER−/HER2+, and TNBC). For post-hoc, pairwise comparisons, we employed Mann-Whitney test with adjusted p values < 0.05 to be considered statistically significant.

TCGA data analysis

We applied the same variant calling algorithm to the raw TCGA sporadic BC dataset to make a head-to-head comparison with our PHTS series data.

Sample size estimation

We performed sample size calculations to determine the minimum number of cases we need to show statistically significant genomic differences between the PHTS and TCGA sporadic BC groups. In order to detect characteristic driver mutations at the variant level, we used the two proportions derived from the somatic PTEN mutation rate in the preliminary PHTS group with 29 samples (21.0%) and that of sporadic luminal subtypes in literature (4.0%).^18,46 We estimated that 30 samples from PHTS and 250 samples from TCGA should be sufficient to achieve a power of 81.0% with an alpha of 0.05 (two-sided) to detect a significant difference.

Statistical analysis

To compare clinical characteristics and somatic mutational landscapes between PHTS and TCGA BCs, we applied a two-tailed Chi-squared or Fisher’s exact test to categorical variables. For continuous variables, analysis of variance was used. A logistic regression model was used for a binary outcome with continuous independent variables. Statistical analyses were performed with GraphPad Prism version 9.0 (GraphPad Software, San Diego, CA, USA), except for statistical analyses incorporated in maftools (version 2.10.0).⁴⁰ p values < 0.05 were considered statistically significant unless stated otherwise.

Results

Clinical characteristics

In the all combined PHTS group (PHTS-all, n = 44), the majority of cases had ductal histology (86.4%) and were ER positive (84.1%), progesterone receptor (PR) positive (81.8%), and HER2 negative (84.1%). The most common grade was 2 (intermediate grade, 52.3%), and the majority of cases were early stage, namely stages 0, I, and II (95.5%, including cases whose stages are unknown and presumed to be early stage). The median ages of diagnosis were 49.5 years (range, 34–85 years) and 47 years (range, 34–75 years) in PHTS and tier 1, respectively, both of which were significantly younger than that of TCGA (median, 56 years; range, 34–85 years; p = 0.003 [PHTS versus TCGA] and p < 0.001 [tier 1 versus TCGA]; Table 1). Compared to TCGA, PHTS-all displayed a significantly higher proportion of ductal histology (86.4% versus 66.8%; p = 0.007) and of earlier clinical stages (95.5% versus 75.3%; p = 0.001; Table 1).

Table 1.

Clinical characteristics of women with PHTS-derived breast cancers compared to those with sporadic breast cancer from TCGA

Characteristics	PHTS (n = 44)	Tier 1 (n = 31)	Tier 2 (n = 13)	TCGA (n = 497)	p valuea
Median age of diagnosis – years	49.5	47.0	53.0	56	0.003b
Diagnosis age range – years	34–85	34–75	40–85	34–85	[<0.001]b
IQR (95% CI of median)	14.5 (45–54)	16 (40–56)	17.5 (46–63.5)	17.5 (55–59)	(>0.99)b
Age of diagnosis – no. (%)
≤40 years	9 (20.5)	8 (25.8)	1 (7.7)	35 (7.0)	–
41–49 years	13 (29.5)	10 (32.3)	3 (23.1)	111 (22.3)	–
50–59 years	13 (29.5)	9 (29.0)	4 (30.8)	139 (28.0)	–
≥60 years	9 (20.5)	4 (12.9)	5 (38.5)	212 (42,7)	–
Tumor type – no. (%)
Ductal	38 (86.4)	27 (87.1)	11 (84.6)	332 (66.8)	0.007c
Lobular	1 (2.3)	1 (3.2)	0 (0)	117 (23.5)	[0.02]c
Mixed histology	5 (11.4)	3 (9.7)	2 (15.4)	15 (3.0)	(0.24)c
Other/unknown	0 (0)	0 (0)	0 (0)	33 (6.6)	–
Stage - no. (%)
0	1 (2.3)	1 (3.2)	0 (0)	0 (0)	0.001d
I	26 (59.1)	18 (58.1)	8 (61.5)	88 (17.7)	[0.3]d
II	11 (25.0)	7 (22.6)	4 (30.8)	286 (57.5)	(0.5)d
III	1 (2.3)	1 (3.2)	0 (0)	119 (23.9)	–
IV	1 (2.3)	1 (3.2)	0 (0)	4 (0.8)	–
Unknown	4 (9.1)	3 (9.7)	1 (7.7)	0	–
ER status – no. (%)
Positive	37 (84.1)	26 (83.9)	11 (84.6)	388 (78.1)	0.25e
Negative	6 (13.6)	4 (12.9)	2 (15.4)	109 (21.9)	[0.37]e
Unknown	1 (2.3)	1 (3.2)	0 (0)	0 (0)	(0.74)e
PR status – no. (%)
Positive	36 (81.8)	26 (83.9)	10 (76.9)	338 (68.0)	0.03e
Negative	7 (15.9)	4 (12.9)	3 (23.1)	159 (32.0)	[0.03]e
Unknown	1 (2.3)	1 (3.2)	0 (0)	0 (0)	(0.77)e
HER2 status – no. (%)
Positive	5 (11.4)	3 (9.7)	2 (15.4)	101 (20.3)	0.33f
Negative	37 (84.1)	26 (83.9)	11 (84.6)	396 (79.7)	[0.49]f
Unknown	2 (4.5)	2 (6.5)	0 (0)	0 (0)	(>0.99)f
Grade – no. (%)
I	12 (27.3)	7 (22.6)	5 (38.5)	unknown	–
II	23 (52.3)	17 (54.8)	6 (46.2)	unknown	–
III	9 (20.5)	7 (22.6)	2 (15.4)	unknown	–

Abbreviations: PHTS, all PTEN hamartoma tumor syndrome breast cancer cases combined; tier 1, breast cancer cases arising from germline PTEN variants classified as pathogenic or likely pathogenic; tier 2, breast cancer cases arising from germline PTEN variants classified as variants of unknown significance or likely pathogenic; TCGA, The Cancer Genome Atlas breast cancer cohort (all subtypes combined); IQR, interquartile range; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2.

^aComparison between PHTS and TCGA is shown, followed by tier 1 compared to TCGA (in the brackets) and tier 2 compared to TCGA (in the parentheses). Differences were considered statistically significant with p < 0.01 (two-tailed, Bonferroni corrected) unless stated otherwise.

^bWhile the PHTS group is normally distributed, TCGA cohort was not. Thus, a nonparametric test (the Kruskal-Wallis) was performed. Adjusted p value < 0.05 is considered statistically significant.

^cFisher’s exact test was performed between pure ductal and non-pure ductal (lobular, mixed, other, unknowns) to meet the criteria of all expected values greater than 1 and at least 20% of the expected values greater than 5.

^dFisher’s exact test was performed between early stage (0, I, II, unknown) and advanced (III and IV) because Chi-squared test calculations on 2 × 6 require all expected values to be greater than 1.

^eFisher’s exact test was performed between positive marker status (ER or PR positive and unknown) and negative marker status (ER negative or PR negative) to meet the criteria of all expected values greater than 1 and at least 20% of the expected values greater than 5. Unknown cases were considered ER and PR positive.

^fFisher’s exact test was performed between HER2 positive and HER2 negative. One unknown case was considered positive (for a case of ductal carcinoma in situ) and the other negative (for a case HER2-directed treatment history is not documented).

Somatic variants in BC-associated genes

In PHTS-all BCs, PTEN and PIK3CA were the most frequently somatically mutated BC-associated genes, each affecting 22.7% of the samples, followed by MAP3K1 (MIM: 600982 [13.6%]), TP53 (MIM: 114480 [11.4%]), GATA3 (MIM: 131320 [9.1%]), and TBX3 (MIM: 601621 [9.1%]) (Figure 1A and Table 2). All the somatic hits in PTEN were distinct from the germline PTEN variants (Tables S1 and S4), representing second hits to PTEN, while the underlying germline PTEN variants represent the first hits per Knudson’s two-hit hypothesis.⁴⁷ Most of these somatic variants have been previously reported and considered pathogenic or likely pathogenic (Table S4). Two individuals had two somatic hits in PTEN. The relative risk for the presence of somatic PTEN variants was 4.03 (95% confidence interval [CI], 2.1 to 7.75; p < 0.001) in PHTS-all BCs compared to TCGA. The relative risk for the presence of somatic PTEN variants was 5.15 in TIER-1 BCs (95% CI, 2.7 to 9.9; p < 0.01), while it was 1.36 in TIER-2 BCs (95% CI, 0.2 to 9.3; p = 0.75) compared to TCGA. The majority of somatic variants detected in PIK3CA were pathogenic hotspot mutations (Table S4). Somatic variants in PTEN and PIK3CA were found to be mutually exclusive in PHTS-all BCs, and this was statistically significant (p = 0.01; Figure 1B).

Figure 1.

Somatic mutational landscape of PHTS-related breast cancers (BCs) and comparison to sporadic BCs from TCGA

(A) Oncoplot showing the distribution of non-synonymous somatic mutations in BC-associated genes among the PHTS-all samples (TIER-1 and TIER-2 BCs combined). Each column represents a sample, and each row represents a BC-associated gene with at least one somatic variant detected in any sample. The gene names are listed in order of the highest to the lowest mutational frequency, which is the proportion of all samples affected (out of 44 samples) shown as a percentage. The top bar plot shows the number of variants detected in each sample. The bar plot on the far right shows the number of samples harboring variants in each gene. The bottom bar plot shows transition and transversion patterns for each sample. Abbreviations: ins, insertion; del, deletion.

(B) The correlation plot shows co-occurring and mutually exclusive gene pairs in PHTS-all (TIER-1 and TIER-2 BCs combined). Pairwise Fisher’s exact test (two-tailed) was performed to identify statistically significant pairs. The numbers in brackets represent the number of samples harboring non-synonymous variants in each gene. The plot shows variants in PTEN and PIK3CA occur in a mutually exclusive manner in PHTS-all BCs, and this was statistically significant (p = 0.01). The p values for statistically significant gene pairs are p = 0.006 for NF1 and AFF2, p = 0.01 for KMT2C and TBX3, p = 0.01 for PTEN and PIK3CA, p = 0.03 for FOXA1 and APC, and p = 0.03 for GATA3 and CBFB.

(C) The forest plot shows odds ratio (OR) to compare the PHTS-all series (n = 44) and TCGA cohort (n = 497) for statistically significantly mutated BC-associated genes. OR > 1 indicates there were more somatic variants in the PHTS-all group (ATM [OR = 3.97; 95% CI, 1.03 to 15.23; p = 0.04], CBFB [OR = 3.97; 95% CI, 1.03 to 15.23; p = 0.04], and PTEN [OR = 4.93; 95% CI, 2.21 to 10.98; p < 0.001]). OR < 1 indicates there were more somatic variants in the TCGA cohort (CDH1 [OR = 0.11; 95% CI, 0.01 to 0.80; p = 0.03] and TP53 [OR = 0.24; 95% CI, 0.09 to 0.63; p = 0.04]). All BC subtypes from both groups are included in this analysis. The horizontal bar represents 95% confidence interval (CI).

(D) The forest plot shows OR to compare TIER-1 BCs (n = 31) and TCGA (n = 497) cohort for statistically significantly mutated BC-associated genes. OR > 1 indicates there were more somatic variants in the TIER-1 BC group (AFF2 [OR = 5.22; 95% CI, 1.36 to 20.03; p = 0.02], AR [OR = 8.50; 95% CI, 1.49 to 48.34; p = 0.02], CBFB [OR = 5.81; 95% CI, 1.49 to 22.66; p = 0.01], ESR1 [OR = 16.53; 95% CI, 1.01 to 270.85; p = 0.049], and PTEN [OR = 6.85; 95% CI, 2.89 to 16.26; p < 0.001]). OR < 1 indicates there were more somatic variants in the TCGA cohort (TP53 [OR = 0.28; 95% CI, 0.10 to 0.82; p = 0.02] and PIK3CA [OR = 0.30; 95% CI, 0.10 to 0.86; p = 0.03]). All BC subtypes from both groups are included in this analysis. The horizontal bar represents 95% CI.

(E) The forest plot shows odds ratio (OR) to compare TIER-2 BCs (n = 13) and TCGA cohort (n = 497) for statistically significantly mutated BC-associated genes. OR > 1 indicates there were more somatic variants in the TIER-2 BC group (TBX3 [OR, 6.48; 95% CI, 1.66 to 25.22; p = 0.007] and CHEK2 [OR, 13.72; 95% CI, 1.33 to 141.67; p = 0.03]). All BC subtypes from both groups are included in this analysis. The horizontal bar represents 95% CI.

(F) The forest plot shows odds ratio (OR) to compare TIER-1 series (n = 31) and TIER-2 series (n = 13) for statistically significantly mutated BC-associated genes. OR < 1 indicates there were less somatic variants in the TIER-1 group (PIK3CA [OR, 0.17; 95% CI, 0.04 to 0.79; p = 0.02]). The horizontal bar represents 95% CI.

Table 2.

Most frequently mutated breast cancer-associated genes in breast cancers derived from PHTS and TCGA

Genes	PHTS			TCGA
	All cases (n = 44) no. (%)	TIER-1 (n = 31) no. (%)	TIER-2 (n = 13) no. (%)	All cases (n = 497) no. (%)	ER+/HER2− (n = 308) no. (%)	ER+/HER2+ (n = 80) no. (%)	ER−/HER2+ (n = 23) no. (%)	TNBC (n = 86) no. (%)
PTEN	10 (22.7)	9 (29.0)	1 (7.7)	28 (5.6)	17 (5.5)	5 (6.3)	0 (0)	6 (7.0)
PIK3CA	10 (22.7)	4 (12.9)	6 (46.2)	166 (33.4)	125 (40.6)	29 (36.3)	2 (8.7)	10 (11.6)
MAP3K1	6 (13.6)	3 (9.7)	3 (23.1)	45 (9.1)	32 (10.4)	8 (10.0)	3 (13.0)	2 (2.3)
TP53	5 (11.4)	4 (12.9)	1 (7.7)	171 (34.4)	62 (20.1)	21 (26.3)	17 (73.9)	71 (82.6)
GATA3	4 (9.1)	4 (12.9)	0 (0)	56 (11.3)	48 (15.6)	8 (10.0)	0 (0)	0 (0)
TBX3	4 (9.1)	1 (3.2)	3 (23.1)	22 (4.4)	19 (6.2)	1 (1.3)	0 (0)	2 (2.3)
AFF2	3 (6.8)	3 (9.7)	0 (0)	10 (2.0)	3 (1.0)	4 (5.0)	1 (4.4)	2 (2.3)
ATM	3 (6.8)	2 (6.5)	1 (7.7)	9 (1.8)	5 (1.6)	2 (2.5)	0 (0)	2 (2.3)
CBFB	3 (6.8)	3 (9.7)	0 (0)	9 (1.8)	7 (2.3)	2 (2.5)	0 (0)	0 (0)
KMT2C	2 (4.5)	1 (3.2)	1 (7.7)	51 (10.3)	38 (12.3)	8 (10.0)	0 (0)	5 (5.8)
NF1	2 (4.5)	2 (6.5)	0 (0)	16 (3.2)	10 (3.3)	3 (3.8)	0 (0)	3 (3.5)
RB1	2 (4.5)	1 (3.2)	1 (7.7)	15 (3.0)	3 (1.0)	3 (3.8)	0 (0)	9 (10.5)
RUNX1	2 (4.5)	1 (3.2)	1 (7.7)	21 (4.2)	14 (4.6)	5 (6.3)	0 (0)	2 (2.3)
AR	2 (4.5)	2 (6.5)	0 (0)	4 (0.8)	3 (1.0)	1 (1.3)	0 (0)	0 (0)
FANCC	2 (4.5)	2 (6.5)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)

Abbreviations: PHTS, PTEN hamartoma tumor syndrome (germline PTEN variant positive); TCGA, The Cancer Genome Atlas; TIER-1, breast cancers arising from germline PTEN variants classified as pathogenic or likely pathogenic; TIER-2, breast cancers arising from germline PTEN variants classified as variants of unknown significance or likely pathogenic; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; TNBC, triple negative breast cancer.

The table shows a comparison of the most frequently mutated BC-associated genes between PHTS (all cases combined), TIER-1, TIER-2, and TCGA sporadic breast cancer groups. The genes are listed in descending order by the percentage of samples harboring somatic variants (single-nucleotide variants and indels) in each gene.

Subset analysis with TIER-1 BC cases (n = 31) revealed a higher somatic mutational frequency in PTEN (29.0%) and a lower mutational frequency in PIK3CA (12.9%) compared to all the PHTS samples combined (PHTS-all, n = 44). Compared to TIER-1 BCs, TIER-2 had a higher mutational frequency in PIK3CA (46.1%) and a lower frequency in PTEN (7.7%; Table 2). On the basis of the data above, TIER-1 and TIER-2 BCs appear to be two genomically different types of tumors. Thus, we performed TIER-1 and TIER-2 analyses separately, along with our overall PHTS series (PHTS-all).

CC score is a clinically useful tool to provide pretest probability of detecting a germline PTEN variant and is a useful surrogate of phenotypic burden.⁹ We performed logistic regression to examine whether the CC score was associated with the presence of somatic PTEN or PIK3CA variants in the PHTS groups. In PHTS-all BCs, no significant association was found between the CC score and the occurrence of somatic PTEN variants (odds ratio [OR] per 5-point increase in CC score = 1.28; 95% CI, 0.98 to 1.72; p = 0.08; Figure 2A). There was, however, a statistically significant relationship between a lower CC score and an increased likelihood of detecting somatic PIK3CA mutations (OR per 5-point increase in CC score = 0.54; 95% CI, 0.31 to 0.78; p = 0.007; Figure 2B). This association persisted in TIER-1 (Figures 2C and 2D) but not in TIER-2 BCs (Figures 2E and 2F).

Figure 2.

Logistic regression and receiver operating characteristic (ROC) curves for CC scores to predict somatic variants in PHTS, TIER-1, and TIER-2 breast cancers

(A) The relationship between the CC score (per 5-point increase) and the presence of somatic PTEN variants in PHTS-all BCs. Top shows a logistic regression curve with the CC score (per 5-point increase) on the x axis and probability of detecting somatic PTEN variants on the y axis. Odds ratio (OR) was 1.28 (95% confidence interval [CI], 0.98 to 1.72; p = 0.08). Bottom shows ROC curve from the logistic regression model. The area under the curve was 0.69 (95% CI, 0.50 to 0.88; p = 0.08).

(B) The relationship between the CC score (per 5-point increase) and the presence of somatic PIK3CA variants in PHTS-all BCs. Top shows a logistic regression curve with the CC score (per 5-point increase) on the x axis and probability of detecting somatic PIK3CA variants on the y axis. Odds ratio was 0.54 (95% CI, 0.31 to 0.78; p = 0.007). Bottom shows ROC curve from the logistic regression model. The area under the curve was 0.83 (95% CI, 0.70 to 0.96; p = 0.002).

(C) The relationship between the CC score (per 5-point increase) and the presence of somatic PTEN variants in TIER-1 BCs. Top shows a logistic regression curve with the CC score (per 5-point increase) on the x axis and probability of detecting somatic PTEN variants on the y axis. Odds ratio was 1.23 (95% CI, 0.89 to 1.80; p = 0.24). Bottom shows ROC curve from the logistic regression model. The area under the curve was 0.63 (95% CI, 0.39 to 0.86; p = 0.27).

(D) The relationship between the CC score (per 5-point increase) and the presence of somatic PIK3CA variants in TIER-1 BCs. Top shows a logistic regression curve with the CC score (per 5-point increase) on the x axis and probability of detecting somatic PIK3CA variants on the y axis. Odds ratio was 0.42 (95% CI, 0.15 to 0.79; p = 0.03). Bottom shows ROC curve from the logistic regression model. The area under the curve was 0.86 (95% CI, 0.64 to 1.00; p = 0.02).

(E) The relationship between the CC score (per 5-point increase) and the presence of somatic PTEN variants in TIER-2 BCs. Top shows a logistic regression curve with the CC score (per 5-point increase) on the x axis and probability of detecting somatic PTEN variants on the y axis. Odds ratio (OR) was 0.89 (95% confidence interval [CI], −2.66 to 0.83; p = 0.87). Bottom shows ROC curve from the logistic regression model. The area under the curve was 0.50 (95% CI, 0.22 to 0.78; p > 0.99).

(F) The relationship between the CC score (per 5-point increase) and the presence of somatic PIK3CA variants in TIER-2 BCs. Top shows a logistic regression curve with the CC score (per 5-point increase) on the x axis and probability of detecting somatic PIK3CA variants on the y axis. OR was 0.82 (95% CI, −1.12 to 0.43; p = 0.57). Bottom shows ROC curve from the logistic regression model. The area under the curve was 0.52 (95% CI, 0.20 to 0.85; p = 0.89).

For the entire sporadic BC cohort from TCGA (n = 497), including all the four subtypes, TP53 was identified as the most frequently mutated BC-associated gene, affecting 34.4% of the samples, followed by PIK3CA (33.4%), CDH1 (MIM: 192090 [17.7%]), GATA3 (11.3%), and KMT2C (MIM: 606833 [10.3%]). PIK3CA was the most frequently mutated gene in ER+ subgroups (40.6% in ER+/HER2− and 36.3% in ER+/HER2+), while TP53 was the most frequently mutated in ER− subgroups (73.9% in ER−/HER2+ and 82.6% in TNBC). Somatic variants in PTEN were only found at low frequencies (5.6%–7.0%), with the highest of 7.0% in the TNBC subgroup and 5.6% in the combined sporadic BC cohort (Table 2).

We found that several BC-associated genes were significantly more somatically altered in PHTS-all BCs than in TCGA: ATM (MIM: 607585 [OR = 3.97; 95% CI, 1.03 to 15.23; p = 0.04]), CBFB (MIM: 121360 [OR = 3.97; 95% CI, 1.03 to 15.23; p = 0.04]), and PTEN (OR = 4.93; 95% CI, 2.21 to 10.98; p < 0.001). In contrast, we found two significantly less altered BC-associated genes in PHTS-all BCs than in TCGA: CDH1 (OR = 0.11; 95% CI, 0.01 to 0.80; p = 0.03) and TP53 (OR = 0.24; 95% CI, 0.09 to 0.63; p = 0.04) (Figure 1C). Compared to TCGA, TIER-1 BCs had five BC-associated genes that were more significantly altered—AFF2 (MIM: 300806 [OR = 5.22; 95% CI, 1.36 to 20.03; p = 0.02]), AR (MIM: 313700 [OR = 8.50; 95% CI, 1.49 to 48.34; p = 0.02]), CBFB (OR = 5.81; 95% CI, 1.49 to 22.66; p = 0.01), ESR1 (MIM: 114480 [OR = 16.53; 95% CI, 1.01 to 270.85; p = 0.049]), and PTEN (OR = 6.85; 95% CI, 2.89 to 16.26; p < 0.001)—while two BC-associated genes were significantly less altered—PIK3CA (OR = 0.30; 95% CI, 0.10 to 0.86; p = 0.03) and TP53 (OR = 0.28; 95% CI, 0.10 to 0.82; p = 0.02; Figure 1D). TIER-2 BCs had two BC-associated genes that were more significantly altered compared to TCGA: CHEK2 (MIM: 604373 [OR, 13.72; 95% CI, 1.33 to 141.67; p = 0.03]) and TBX3 (OR, 6.48; 95% CI, 1.66 to 25.22; p = 0.007) (Figure 1E). When TIER-1 and TIER-2 BCs were compared, PIK3CA was significantly less somatically mutated in TIER-1 compared to TIER-2 BCs (OR, 0.17; 95% CI, 0.04 to 0.79; p = 0.02)

Somatic mutational signatures

In order to identify genome-wide differences between PHTS and TCGA BCs, we performed analyses including all BC-associated genes and non-BC-associated gene variants detected in our pipeline. Previous studies showed apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC), a deaminase, to be one of the most important endogenous sources of mutagenesis in human cancer, especially in BC.^43,48 APOBEC-related somatic mutations were found enriched in PHTS-all (9.1%), TIER-1 (3.2%), TIER-2 (23.1%), and TCGA (31.0%) BCs. The mutational load between APOBEC-enriched and non-APOBEC-enriched samples had no statistically significant difference in PHTS-all BCs (p = 0.62; Figure 3A). In contrast, there was a statistically higher mutational load in APOBEC-enriched than in non-APOBEC-enriched samples in TCGA (p < 0.001; Figure 3B). PIK3CA was one of the differentially mutated genes in TCGA (OR = 2.22; 95% CI 1.43 to 3.45; p < 0.001) but not in PHTS-all, TIER-1, or TIER-2 BCs (Figures 3A and 3B). Indeed, no differentially mutated genes were found in TIER-1 or TIER-2 BCs.

Figure 3.

Genome-wide somatic mutational signature analyses

(A and B) Apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC) enrichment estimates are shown for the PHTS-all BCs (A) and TCGA BC cohorts (B). Box plots on the left compare the differences in mutational burden between APOBEC-enriched (magenta) and non-APOBEC-enriched (blue) samples. The vertical bar represents 95% CI. The pie charts on the right show the proportion of C>T transition events occurring in the TCW motif (tCw load), indicated by dark blue shadowing. APOBEC-related mutations were found enriched in 9.1% of samples in the PHTS-all group compared to 31.0% in TCGA. For PHTS-all BCs, CCDC88C (MIM: 611204) showed a mutational burden that was significantly higher in APOBEC-enriched than non-APOBEC-enriched samples by Fisher’s exact test (odds ratio [OR] = 30.68; 95% interval [CI], 1.19 to 2,298.87; p = 0.02) but PIK3CA did not (OR = 5.35, 95% CI, 0.33 to 87.36; p = 0.15). For TCGA, genes showing statistically higher mutational burdens in APOBEC-enriched than non-APOBEC-enriched samples by Fisher’s exact test are shown in the right lower corner with corresponding bar plots indicating differences in mutational loads. Odds ratio, 95% CI, and p values for the top eight differentiated genes in (B) are HMCN1 (MIM: 608548 [OR = 4.85; 95% CI, 2.30 to 10.50; p < 0.001]); MXRA5 (MIM: 300938 [OR = 7.94; 95% CI, 2.51 to 29.06; p < 0.001]); AKAP13 (MIM: 604686 [OR = 4.42; 95% CI, 2.93 to 139.34; p < 0.001]); PIK3CA (OR = 2.22; 95% CI, 1.43 to 3.45; p < 0.001); ATP10B (MIM: 619791 [OR = 9.60; 95% CI, 2.35 to 56.03; p < 0.001]); SETD2 (MIM: 612778 [OR = 9.60; 95% CI, 2.35 to 56.03; p < 0.001]); UBR5 (MIM: 608413 [OR = 16.16; 95% CI 3.37 to 153.71; p < 0.001]); and SMG1 (MIM: 607032 [OR = 12.7; 95% CI 2.50 to 124.68, p < 0.001]). ^∗∗∗p value < 0.001, ^∗∗p value < 0.01, ^∗p value < 0.05.

(C−E) Single base substitution (SBS) signatures in PHTS-all (C), TIER-1 BCs (D), and TCGA (E). SBS signatures were extracted and compared against known SBS signatures from the Catalogue of Somatic Mutations in Cancer (COSMIC) database. Cosign similarity was calculated to identify best matched signatures and identified signatures are plotted.

Single base substitution (SBS) signature analysis identified five best-match clusters in PHTS-all BCs: SBS1 (spontaneous or enzymatic deamination of 5-methylcytosine), SBS2 (APOBEC cytidine deaminase [C>T]), SBS5 (unknown etiology), SBS6 (defective DNA mismatch repair), and SBS26 (defective DNA mismatch repair) (Figure 3C). Three out of these signatures (SBS1, SBS2, and SBS5) persisted in TIER-1 BCs (Figure 3D). In contrast, the TCGA cohort clustered into three best-matched groups: SBS2, SBS6, and SBS29 (exposure to tobacco [chewing] mutagens) (Figure 3E). SBS29 was identified in TCGA but not in PHTS-all or TIER-1 BCs. SBS29 signifies mutagens like tobacco use⁴⁹ and may indicate environmental and lifestyle-related causes contributing to carcinogenesis in sporadic BCs but not in PHTS-all or TIER-1 BCs. In PHTS and TIER-1 BCs, we identified the SBS1 signature, which is strongly correlated with older age of cancer onset.^49,50 No significant SBS was found for conversions in TIER-2 BCs.

Driver gene ranking

cDriver identified PTEN and TP53 as the top ranked driver genes in PHTS-all and TCGA BCs, respectively (Tables S5 and S6). Other high-ranking BC-associated genes in PHTS-all BCs included PIK3CA, MAP3K1, GATA3, CBFB, and TBX3. While the top ten genes ranked for TCGA samples were all known BC-associated genes, some non-BC-associated genes were ranked as top driver genes in PHTS-all BCs (ZNF253 [MIM: 606954], TCHH [MIM: 190370], STAP2 [MIM: 607881], NPIPA5, ZNF676, MAGEC1 [MIM: 300223], and MDC1[MIM: 607593]). PTEN remained to be the top driver gene but PIK3CA ranked number six in TIER-1 BCs (Table S7). On the other hand, PIK3CA was identified as the top driver gene, but PTEN failed to be ranked within the top 14 in TIER-2 BCs (Table S8).

Tumor mutational burden

TIER-2 BCs had the highest TMB with a median of 2.57 (interquartile range [IQR], 3.22; 95% CI, 0.90 to 4.83), followed by TNBC (median, 2.30; IQR, 2.72; 95% CI, 1.90 to 3.00), PHTS-all (median, 2.23; IQR 3.53; 95% CI 1.43 to 3.20), TIER-1 (median, 2.07; IQR, 3.83; 95% CI, 0.77 to 3.47), ER+/HER2+ (median, 1.88; IQR, 2.19; 95% CI, 1.47 to 2.23), ER−/HER2+ (median, 1.73; IQR, 1.57; 95% CI, 1.53 to 2.43), TCGA groups all combined (median, 1.70; IQR, 1.97; 95% CI, 1.60 to 1.83), and ER+/HER2− (median, 1.47; IQR, 1.63; 95% CI, 1.33 to 1.67). Although TIER-2 BCs had the highest TMB value, there were no statistically significant differences compared to any of the other groups (Figure 4). Among TCGA BC subtypes, TNBC had a statistically higher median TMB than the ER+/HER2− cohort and all sporadic BC combined (p < 0.001). The failure to demonstrate statistically significant difference in TMB between PHTS groups (especially TIER-2 BCs) and the sporadic BC cohorts may be due to the limited sample size.

Figure 4.

Comparative analysis of tumor mutational burden (TMB)

Scatter plot showing log2-transformed tumor mutational burden (TMB) values from each sample derived from the counts of non-synonymous variants with variant allele frequency greater than 5%. In this plot, eight groups were compared with a Kruskal-Wallis test. Each dot in the graph represents one sample. The black horizontal bars near the center of the scatter plots represent log2-transformed median TMB values. The vertical lines running perpendicular to the median bars represent 95% confidence intervals (CIs). The raw TMB median values in mutations per megabase (Mut/Mb), interquartile ranges (IQRs), and 95% CIs are shown in the top right. Post-hoc sub-analyses were performed with Mann-Whitney test and the results are shown in the bottom right. Adjusted p values < 0.05 are considered statistically significant. Abbreviations: ns, not statistically significant; IQR, interquartile range; Mut/Mb, mutations per megabase.

Discussion

We identified notable differences in the clinical characteristics and somatic variant spectra between PHTS-related BCs and sporadic counterparts from TCGA. Our data reveal several key findings, providing insights into distinct BC biology in the background of germline PTEN mutations and their clinical implications.

First, our findings are consistent with the hypothesis that high penetrance germline variants influence somatic phenotypes.⁵¹ Germline variants of high penetrance in cancer susceptibility genes tend to acquire bi-allelic inactivation of the gene harboring them and are less likely to have other independent gain-of-function driver mutations compared to tumors retaining heterozygosity in the germline variants. Our data support this observation with the high frequency of second (somatic) hits in PTEN (presumably resulting in bi-allelic inactivation), as well as mutual exclusivity of somatic PTEN and PIK3CA mutations in PHTS-all BCs, but not in TCGA. Data restricted to BCs arising in the background of pathogenic or likely pathogenic PTEN germline variants (tier 1) further support this hypothesis.

Second, our overall data point to somatic variants in PTEN as the predominant, somatic oncogenic drivers in PHTS-related BCs, especially in TIER-1. This is mainly supported by cDriver’s ranking PTEN as the top driver on the basis of cancer cell fraction, mutational frequency, and functional impacts.¹⁹ Interestingly, there were two PHTS breast samples with two somatic hits in PTEN. In the case of PHTS-AE, one somatic variant is a pathogenic frameshift deletion (c.855_856del [p.Glu285Aspfs^∗12] [GenBank: NM_000314.4]) and the other is a likely pathogenic variant at the last nucleotide of exon 8 (c.1026G>C [p.Lys342Asn] [GenBank: NM_000314.4]). In the other case (PHTS-L), both somatic variants are classified as pathogenic. One is a frameshift in exon 8 (c.741dup [p.Pro248Thrfs^∗5] [GenBank: NM_000314.4]) and the other one is a missense mutation in exon 6 (c.448G>A [p.Glu150Lys] [GenBank: NM_000314.4]). It is therefore challenging to distinguish which may be the driver mutation in either of these cases, and all of them indeed appear damaging. Importantly, missense mutations in PTEN can have dominant negative effects,⁵² potentially making them as damaging or even more damaging than frameshift truncating or nonsense mutations. Also, it is possible we are detecting somatic variants from different cell populations (i.e., different clonal populations) with two distinct somatic hits in PTEN for either of these cases. Further investigations, including gene expression analyses, epigenetic studies, and examination of the tumor microenvironment, are warranted for mechanism resolution and for elucidation of the functional impacts of the second hit somatic variants in PTEN.

Third, our results corroborate that germline PTEN variants trend to increased genomic instability, a hallmark of cancer,⁵³ reflecting one of the PTEN’s non-canonical roles. Accumulating evidence shows that normal PTEN functions are important in the maintenance of genome integrity.⁵⁴ Despite lack of statistical significance due to limited sample size, the numerically higher median TMB values in PHTS-all, TIER-1, and TIER-2 BCs compared to TCGA also suggest increased genomic instability due to germline PTEN variants and are in line with PTEN’s role in maintaining genome stability.^54,55 Whether this may lead to multiple global alterations in the genome warrants further investigation. We also showed through mutational signature analyses that the mechanism of mutational burden appears different between PHTS and TCGA BCs.

Fourth, we observed notable differences in mutational signatures between PHTS-all and TIER-1 BCs compared to TCGA. The sporadic breast cancers from TCGA dataset had a signature associated with mutagens such as tobacco-use, indicating environmental and lifestyle etiologies, while PHTS-all and TIER-1 groups lacked this signature (Figures 3C and 3D). Interestingly, one of the SBS clusters in PHTS-all and TIER-1 BCs was SBS1, which is strongly correlated with older age of cancer onset. Knowing the PHTS-all and TIER-1 groups had significantly younger median ages of BC diagnosis, detecting this particular signature may indicate that mutational patterns that usually accumulate with time tend to occur much faster in PHTS-related BCs, especially in the background of pathogenic germline PTEN mutations (tier 1). Of note, no significant SBS signatures were found for conversions in the TIER-2 group, which is attributable to its small sample size.

Fifth, all the PIK3CA somatic mutations detected in PHTS BCs have been reported previously and classified as pathogenic or likely pathogenic (Table S4). Furthermore, we found that lower CC scores may predict the presence of somatic PIK3CA mutations arising in PHTS-all and TIER-1 BCs, illustrating that germline variants in the setting of phenotypic burden (CC score) may inform somatic mutations in the BC tissue, which can provide clinically useful information and precision management.

Finally, our genomic data reveal TIER-1 and TIER-2 BCs are two genomically different types of tumors. This provides further evidence to support the different pathogenicity predictions between tier 1 and tier 2 germline PTEN variants, which are expected to influence the biology of carcinogenesis in BC differently. Further in-depth research is warranted to investigate whether this proposed hypothesis has clinical significance in determination of pathogenicity in germline variant calling.

This study has several limitations. First, this study may be limited by a small sample size for PHTS, owing to the difficulty in identifying PHTS. Yet, we have a well-annotated and homogeneous clinical cohort of PHTS-affected individuals, enabling us to provide clinically useful information for this apparently rare but important population. Second, heterogeneity of each BC sample is an inherent limitation of this study design. Third, not being whole-genome sequencing, our pipeline did not identify all genomic variations such as non-coding ones. Fourth, two different exome sequencing platforms were used for the PHTS series. The older platform, the Illumina Somatic Exome protocol, had a lower sequencing coverage goal (20× for tumor) and the read distribution along target regions tended to be less uniform, which may have affected sensitivity of variant calling in certain areas. However, the average coverage for tumor samples was 129× (ranges from 42× to 261×) even for the older platform and as high as 261× (range from 182× to 368×) for the newer platform, which appear appropriate. Fifth, this study focused on small genomic changes, including single-nucleotide variants and indels. Thus, our pipeline did not detect large deletions, insertions, and chromosomal rearrangements. Sixth, this study was not designed for examining epigenetic alterations.

In conclusion, we demonstrate a characteristic genomic landscape in PHTS-related BCs. Currently, this entity is treated similarly to sporadic BCs, according to the standard of care in the absence of somatic landscape data in PHTS-related tumors. Although targeting PTEN-associated mutations and alterations for therapeutic purposes has tremendous challenges,^56,57 our findings call for more targeted, personalized strategies to effectively treat PHTS-related cancers, a population that will only rise in incidence as clinical genetic testing becomes more widely accessible in the clinic.

Published: August 4, 2022

Web resources

Biorepository of the Genomic Medicine Institute at the Cleveland Clinic, http://www.lerner.ccf.org/gmi/gmb

Broad Institute, GATK Best Practice workflow, https://gatk.broadinstitute.org/hc/en-us

Catalogue of Somatic Mutations in Cancer (COSMIC), https://cancer.sanger.ac.uk/cosmic

ClinVar, https://www.ncbi.nlm.nih.gov/clinvar

Franklin by Genoox, https://franklin.genoox.com/clinical-db/home

GDC Data Portal, https://portal.gdc.cancer.gov

gnomAD, https://gnomad.broadinstitute.org

Integrative Genomics Viewer (IGV), https://igv.org

NCCN guideline, https://www.nccn.org/home

Surveillance, Epidemiology, and End Results Program (SEER), https://seer.cancer.gov

TCGA Portal, https://cancergenome.nih.gov

Data and code availability

Data will be available on reasonable request to the corresponding author. All the bioinformatics and statistical tools are publicly available and are described in the subjects and methods.