Clonal relatedness in tumour pairs of breast cancer patients

Fresh-frozen tumour specimens for 74 invasive breast carcinomas, corresponding to 37 patients diagnosed in Western Sweden between 1988 and 1998 with multiple breast cancers, were selected from the tumour bank at the Sahlgrenska University Hospital Oncology Lab (Gothenburg, Sweden). The patients were stratified into four groups based on the anatomic location of the multiple breast cancers (ipsilateral or bilateral) and time interval between the diagnoses (synchronous or metachronous). Ipsilateral was defined as tumours occurring in the same breast while bilateral was defined as the occurrence of tumours in both breasts. Metachronicity was defined as a time interval greater than 6 months between the diagnoses of the first and second tumours, while synchronicity specified that the two tumours occurred concurrently. Clinicopathological information was obtained from Regional Cancer Centre West (Gothenburg, Sweden) and the Sympathy and Melior databases (Sahlgrenska University Hospital). A part of the dataset was stratified into the molecular breast cancer subtypes (normal-like, basal-like, luminal subtype A, luminal subtype B/human epidermal growth factor receptor 2 (HER2)+, luminal subtype B/HER2-, and HER2/oestrogen receptor (ER)-) as described elsewhere [23, 24]. Luminal subtype B was further stratified according to HER2 status as determined by aCGH; HER2+ was set to log₂ ratio ≥ + 0.5 and HER2- was set to log₂ ratio < + 0.5 [25]. Routine haematoxylin and eosin-stained slides from formalin-fixed paraffin-embedded (FFPE) blocks were revised by a board-certified breast pathologist. Classification of the subtypes based on immunohistochemistry was not possible due to the lack of information on the Ki-67 status. The patients had an average follow-up time of 7.2 years. None of the patients were diagnosed with distant metastasis at the time of diagnosis of either the first or second tumours. The selection criteria were to use samples from opposite quadrants for ipsilateral cases and no nipple involvement. Representative imprints from each tumour specimen were stained with May-Grünwald Giemsa (Chemicon, Temecula CA, USA) and evaluated for neoplastic cells. Tumour specimens with at least 70% neoplastic cell content were included in downstream analyses.

aCGH and data pre-processing was performed as previously described [24] and summarised in the Additional file 1: Supplementary Methods. Segmented data for segment analysis were generated using the “GLAD” package [26] in R (v3.4.3) [27]. The “Clonality” package [28] was used to define the likelihood ratio with individual comparisons (LR2) and LR2 p value and required copy number data procession with the “DNAcopy” package [29].

Sixteen samples were randomly selected to represent each clinical group with four samples corresponding to two patients per group. Purified genomic DNA was processed at the SNP&SEQ technology platform, Uppsala, Sweden, using Illumina Infinium MethylationEPIC BeadChips (MethylationEPIC_v-1-0; mapped to UCSC Feb 2009 hg19: GRCh37). Raw data (IDAT files) were processed in R using the “RnBeads” package [30]. The probes were normalised with the BMIQ method (beta mixture quantile dilation) [31]. Beta values were obtained with “RnBeads”. The intensity values were extracted using the “ChAMP” package to generate segmented copy number data for the segment analysis [32, 33]. The “conumee” package was used to extract unsegmented information of CNAs on the probe level [34]. The unsegmented CNAs were used for the similarity index (SI), the distance measure and the clustering analysis.

Total RNA samples were processed at the Science for Life Laboratory (National Genomics Infrastructure, Stockholm, Sweden). Illumina TruSeq strand-specific RNA libraries (Ribosomal depletion using RiboZero human) containing 125 bp paired-end reads were obtained for each sample on a HiSeq2000 sequencer (Illumina, San Diego, CA, USA). The computations were performed on resources provided by SNIC through Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX) [35], as described in the Additional file 1: Supplementary Methods.

Genome-wide SNP genotyping analysis was processed with Illumina Infinium HumanOmni2.5–8 v1.3 Beadchips at the SCIBLU Genomics DNA Microarray Resource Center (SCIBLU), Sweden, as described in the Additional file 1: Supplementary Methods.

The 37 breast cancer patients were stratified into four clinical groups based on tumour laterality and the time interval between the diagnoses of the first and second tumours (BM: bilateral-metachronous; BS: bilateral-synchronous; IM: ipsilateral-metachronous; IS: ipsilateral-synchronous). The clinicopathological characteristics are shown in Additional file 2: Table S1. Metastatic spread to the axillary lymph nodes was more prevalent in the synchronous groups (BS: 100%; IS: 85.7%) as compared to the metachronous groups (BM: 61.5%; IM: 14.3%; P = 0.001).

For the clinical classification of clonality, several histopathological and molecular features were taken into consideration, including histological subtype, the status of ER and HER2, and the molecular subtype (Table 1). While the receptor status was available for most samples, the molecular subtype was only defined for about 40% of the tumours. Thirty-two percent of the patients (12/37) showed discordances between the first and the second tumour, with one-fourth of the 12 patients showing two discordant changes. Most changes were found in the histological subtypes (35%; 6/17 patients), while the molecular subtype differed in 25% (2/8 patients), ER status in 11% (4/35 patients), and HER2 status in 8% (3/37 patients). In patients with metachronous cancer, changes in receptor status from positive to negative were observed for patients BM6 and BM7. The discordant changes were equally distributed between the different clinical groups and showed no significance when stratified by group.

DNA copy number analysis using aCGH was performed to identify recurrent regions of DNA copy number gain (blue) and loss (red) in at least 25% of the tumours in the patient cohort. Recurrent DNA gains were identified on chromosomes 1q, 8q, 16p, 17q, and 20q, while DNA loss was detected on 1p, 8p, 11q, 13q, and 16q (Fig. 1a). These results were in line with DNA gains and losses frequently identified in breast cancer [42‐44]. There was very little difference in the DNA copy number profiles when stratified by synchronicity (excluding copy number variations (CNVs) and probes from sex chromosomes) with 59 significantly different genomic regions displaying DNA copy number imbalances (Fig. 1b). Most noticeable were losses of the entire chromosome 14 and the long arm of chromosome 11 in the metachronous subgroup. In contrast, stratification by laterality yielded 134 statistically significant minimal common regions of copy number imbalances, including more fractions of genome altered in the ipsilateral subgroup with prominent losses on 8p and 11p (Fig. 1c).

The variability of the beta values was the highest in the bilateral and synchronous groups and consequently in the BS group, which was in line with patients BS7 and BS8 having the highest variability in methylation patterns between the two respective tumour pairs (Additional file 3: Table S2). Principal component analysis of the methylation data showed a statistically significant association with synchronicity (P = 0.007), while no further associations to other variables were found. Kruskal’s non-metric multidimensional scaling (MDS) demonstrated that most of the synchronous samples were further away from each other, while the metachronous samples formed a distinct cluster, suggesting a higher variability of beta values in synchronous samples (Fig. 2).

Strong consensus in clonality could be found for the tumours of patients BM7, BS8, and IS1, while the tumour pairs for patients BS7 and IS4 were determined to be independent primary tumours (Table 2). In general, DNA methylation intensity values were a more liberal method for clonality classification, in particular the clustering analysis, and frequently classified tumour pairs as similar in comparison with other types of molecular data.

The gene expression cohort consisted of seven patients with ipsilateral tumours (three metachronous and four synchronous). The clonality analyses based on gene expression microarray data showed strong concordance to the clinical groups with all four synchronous cases being similar for all analyses while 2/3 metachronous cases were classified as different entities (Additional file 4: Table S3). All analyses of the gene expression cohort were in line with the aCGH results except for patient IM4, whom was classified as independent in the gene expression analysis and equivocal in the aCGH data set. MDS demonstrated similar gene expression patterns between the tumour pairs of the patients IM3, IS3, and IS10 (Additional file 5: Figure S1A).

RNA-seq and SNP genotyping were performed for both tumours of patients IM4, IS10, and IS11. A total of 64 fusion transcripts were detected in the two tumours of patient IM4, with five fusion transcripts (7.8%) containing the same fusion breakpoints in the 5′- and 3′-gene partners in both tumours (Additional file 6: Table S4). For patients IS10 and IS11, 1/836 (0.1%) and 5/153 (3.3%) fusion transcripts were identical between the two tumours, respectively. No other shared fusion transcripts were found between different tumours. The RNA-seq data was then evaluated to identify shared genetic variants in genomic and exonic (coding) regions. Shared genomic variants (genome-wide and the 254-gene panel) showed similar tendencies that were in line with the aCGH distance measure and SNP shared segment (LRR) data (Additional file 7: Table S5). The shared exonic variants in the 254-gene panel only found two shared mutations in patient IM4, which contradicted most other RNA-seq results. The shared segment and clustering analyses of the SNP array data classified patient IS10 as clonal, which was in line with the aCGH results but contradicted the distance measure and MDS, which classified the LRRs of the tumour pair IM4 as most similar (Additional file 5: Figure S1B). The “Clonality” package applied on the exonic variants classified all tumour pairs as clonal. A circos plot summarising the results of patient IM4 visualised the similarities in copy number profiles of both aCGH-derived log₂ ratio and SNP array-derived LRR and fusion transcripts (Fig. 3).

Calculation of Cohen’s kappa indices was applied to detect the highest agreement between the different statistical methods used to estimate clonality. For the aCGH data, hierarchical clustering and the similarity index (SI) were identified as the most appropriate (0.659 and 0.630, respectively). Since the SI is easier to interpret as a measure and independent of the cohort, it presented the most reasonable definition of clonality and determined 46% (17/37) of the tumour pairs as clonal (Fig. 4). No statistical significance was found to associate the tumour clonality defined by the SI with the clinical classification (Wilcoxon rank sum test: P_Laterality = 0.247; P_{Synchronicity} = 0.095; Analysis of variance (ANOVA): P_{Clinical groups} = 0.229), highlighting the alarming reality that there is very little connection between current clinical guidelines and the biology underlying tumour clonality.

The majority of the analyses conducted were in agreement with the SI except for patients BM1, IM4, IM7, IS1, and IS7 (Fig. 4). Interestingly, the histopathological concordances often showed opposite tendencies compared to the aCGH analysis. The different methods applied to the DNA methylation, gene expression and SNP array data sets displayed strong homogeneity within their type of data regardless of the method applied. The results for the SI and hierarchical clustering were consistent in most data sets. The distance measure also overlapped with these results but seemed to be a more conservative measure since fewer tumour pairs were classified as clonal. The shared segment analysis with the aCGH data clearly favoured the clonality hypothesis with defining 21/37 tumour pairs as clonal along with 4/8 cases in the methylation intensity data and 1/3 in the LRR data. The shared segment analysis was in most cases consistent between the different types of data.