Inter-tumor genomic heterogeneity of breast cancers: comprehensive genomic profile of primary early breast cancers and relapses

The study population included a mono-institutional cohort of 128 patients with early-stage breast cancer that underwent surgery at the European Institute of Oncology of Milan between 1999 and 2019. All the patients had at least one loco-regional or distant relapse in a timeframe of 17 years. The median age at diagnosis was 50 years (range 24–79 years). The median time to relapse was 42.5 months, range 1–200 months. Primary and relapsed tumors were classified according to immunohistochemical surrogates, following the St. Gallen criteria [15]. Estrogen receptor and progesterone receptor (PR), HER2, and proliferation index (Ki-67) were evaluated by immunohistochemistry, and for HER2 equivocal cases (immunohistochemical score 2+), reflex fluorescence in situ hybridization (FISH) analysis was performed. The tumors were classified as luminal B HER2− (hormone receptors+/HER2−), luminal B HER2+ (hormone receptors+/HER2+), HER2+ non-luminal (hormone receptors−/HER2+), and triple negative (hormone receptors−/HER2−). The recurrence sites were divided into loco-regional (axillary lymph-node, skin or soft tissue of chest wall) or distant (liver, lung, distant lymph node, pleura, distant soft tissue, ovary, bone) relapses. The clinico-pathological characteristics of the study population are reported in Table 1. All patients gave written informed consent regarding the storage of any biological specimens collected in the course of diagnosis and the use of these samples for research purposes. The study was conducted in accordance with the 1964 Helsinki Declaration and later amendments.

Primary tumors and recurrences were evaluated using large multi-genes NGS panels detecting different types of genetic alterations such as single nucleotide variants (SNVs), insertion/deletions (InDels), copy number variants (CNVs), and fusion genes. Representative formalin-fixed paraffin-embedded (FFPE) tumor tissue blocks of primary tumors and recurrences were retrieved from the archives of the Division of Pathology of the European Institute of Oncology. In the first phase of the study, 202 FFPE blocks were analyzed with the Foundation One test, including 315 genes (Roche Pharma AG, Grenzach-Wyhlen, Germany). We then performed a NGS panel in-house (Oncomine Comprehensive Assay v.3, ThermoFisher Scientific, Waltham, MA, USA), for 87 additional samples, following the manufacturer’s instructions. Briefly, the nucleic acids (DNA and RNA) were extracted automatically using Promega Maxwell RSC DNA or RNA FFPE kit (Promega, Madison, WI, USA) and then quantified, as previously reported [16]. Ten nanograms of genomic DNA and 10 ng of RNA were used for the library preparation and the subsequent chip loading, both performed automatically on the Ion Chef System (ThermoFisher Scientific, Waltham, MA, USA). The sequencing run was done on Ion S5 System (ThermoFisher Scientific, Waltham, MA, USA) and data were analyzed using the Ion Reporter Analysis Software v5.10. Only mutations with a variant allele frequency (VAF) equal/superior to 5% and with adequate quality metrics (read depth > 100; VAF × read depth > 25; p value = 0.00001) were reported. Copy number variants were evaluated for samples with a Median of the Absolute values of all Pairwise Difference (MAPD) < 0.5 [17].

The mutations were classified as driver alterations, including all the alterations belonging to level I, II, and III class, as previously described [18] or variants of uncertain significance (VUS) if they were annotated as unknown in cancer gene mutation databases, including Catalogue of Somatic Mutations in Cancer (COSMIC) [19], cBioPortal for Cancer Genomics [20], and ClinVar–NCBI–NIH [21], or considered damaging by “in silico” predictors only, available at VarSome website [22]. Furthermore, we evaluated the clinical actionability of driver alterations using OncoKB levels of evidence V2 ranking [23]. Variants classified as polymorphism, benign, likely benign, or neutral were not reported. The co-occurrence of selected gene alterations was evaluated using the mutual exclusivity analysis of cBioPortalbioinformatics tool [20].

For this analysis, only alterations occurring in genes targeted by both the FoundationOne and Oncomine Comprehensive Assays (Additional file 1 for the complete gene list) were evaluated. Overall, 188 of 289 (65.1%) samples analyzed with large NGS panels achieved the quality parameters required by the specific assay, including 106 primary and 82 recurrence tumors. Moreover, for 70 of 128 (54.7%) patients, both primary and matched relapse sample were successfully profiled with the same NGS panel.

Patient clinico-pathological characteristics were reported with median and interquartile range (IQR) for continuous variables and absolute and relative frequencies for categorical variables. Cohen’s Kappa test was used to assess the driver gene alteration agreement between primary and relapse samples. Genes with a Kappa coefficient over 60% and a mutation frequency over 10% were considered in this analysis. Hamming distance for binary variables was applied to implement a heatmap of the selected driver genes. Univariate logistic models were used to assess the associations of gene aberrations with molecular subtype (triple negative vs luminal B) and recurrence site (distant vs loco-regional). The results were shown using the R package “EnhancedVolcano” for Volcano plot implementation. The association of driver gene alterations with time to recurrence was analyzed with a Log-rank test. We also evaluated the association of the total number of mutations in primary breast cancer samples as categorical variables, with the median value as cutoff. Multivariate Cox proportional hazard models were chosen considering backward and forward selection of variables and adjusting for known prognostic factors (molecular subtype and pT). Hazard ratios (HR) with 95% confidence intervals (95% CI), from multivariate models, were reported. Since the survival analyses were carried out only for patients with a relapse, hazard ratios represented a measure of the association with time to relapse and they should not be interpreted as probability of relapse. Molecular subtype changes between primary breast cancer and relapse samples were displayed in a Chord Diagram from the R package “circlize.” Finally, Fisher exact tests were employed to estimate whether gene aberrations in primary and recurrence samples were significantly different by molecular subtype (triple negative vs luminal B). Only gene alterations with a frequency over 5% in both primary breast and relapse groups were used in this analysis. Due to the explorative nature of this study, a multiplicity correction was omitted.

Overall, 106 primary tumors were analyzed, including 56 (52.8%) luminal B HER2−, 45 (42.5%) triple negative, 4 (3.8%) luminal B HER2+, and 1 (0.9%) HER2+ non-luminal. All primary samples harbored at least one gene alteration (driver or VUS), with a median number of six alterations per sample (range 1–16). Driver alterations were detected in 102 of 106 (96%) samples, with a median number of three driver alterations per sample (range 0–10). In detail, we found 721 gene aberrations (driver or VUS), including 472 (65.5%) SNV, 75 (10.4%) InDels, 162 (22.5%) CNV, and 12 (1.7%) fusion genes (Additional file 2). 352/721 (48.8%) were driver alterations, including 135 (38.4%) SNV, 51 (14.5%) InDels, 160 (45.5%) CNV, and 6 (1.7%) fusion genes (Additional file 2).

The most frequently mutated genes were TP53 and PIK3CA (Fig. 1a), with alterations spanning the whole coding sequence of TP53 and involving hotspot regions in PIK3CA (Additional file 3). Moreover, recurrent copy number gains were identified in MYC, CCND1, FGF19, FGF3, and FGFR1 (Fig. 1a). Among them, CCND1, FGF19, and FGF3 genes mapped on the same cytogenetic band, 11q13.3, and showed a statistically significant co-occurrence (p value < 0.001 and q < 0.01). No other significant driver alterations co-occurrence was found. Distinctive genomic alterations were significantly associated with molecular subtypes (Additional files 4 and 5). In particular, TP53, PIK3R1, and NF1 mutations were detected more frequently in triple negative tumors (odds ratio > 2.71, p value < 0.05). CCND1, FGF3, and FGFR1 copy number gains were recurrently identified in luminal cases (odds ratio < 0.36, p value < 0.05) (Additional file 4).

Moreover, TP53 mutations and MYC copy number gain were significantly associated with shorter time to relapse, both in univariate and multivariate analyses, adjusted for known prognostic factors (p value < 0.05) (Fig. 2, Table 2). Also, an increased number of alterations, including both driver and VUS variants, was associated with a shorter time to relapse in univariate analysis, even if not statistically significant (p value 0.06, Fig. 2).

TP53 mutations and MYC copy number gain findings in primary breast cancers were prognostic factors independently associated with shorter time to relapse, along with a high primary tumor pathologic stage (pT = 2–4) and triple negative molecular subtype.

Molecular subtype changes between primary tumors and relapses were seen in 10 of 128 (7.8%) cases (Additional file 6). In detail, 5 patients with luminal B HER2− primary tumors had triple negative recurrences, 3 patients with triple negative primary tumor relapsed with luminal HER2− tumor, and 2 patients with luminal HER2+ primary disease had a luminal HER2− recurrence. All the 82 recurrence samples that underwent a comprehensive genomic profile harbored at least one alteration, with a median number of six alterations per sample (range 1–16). Driver alterations were identified in 78 of 82 (95.1%) samples, with a median number of three driver alterations per sample (range 0–11). Overall, in the recurrence samples, we found 549 aberrations (driver and VUS), including 352 SNV (64.1%), 59 InDels (10.7%), 126 CNV (23%), and 12 fusion genes (2.2%). Considering the driver alterations only (n = 291; 53%), we detected 112 (38.5%) SNV, 46 (15.8%) InDels, 125 (43%) CNV, and 8 (2.7%) fusion genes. A higher number of driver alterations, including SNVs, CNVs, InDels, and fusion genes, were seen in the relapses as compared to the primary tumors, although not statistically significant (Additional file 2).

As in the primary tumors group, the most frequently mutated genes were TP53 and PIK3CA (Fig. 1b), with alterations spanning the whole coding sequence of TP53 and involving hotspot regions in PIK3CA (Additional file 3). An increased frequency of alterations in a subset of genes was seen in the recurrence (R) as compared to primary (P) samples, including FGFR1 (13% P–17% R), ESR1 (9% P–17% R), NF1 (9% P–11% R), BRCA1 (8% P–10% R), and PTEN (7% P–10% R), even if not statistically significant (Fig. 1).

Similarly to the primary tumors, specific alterations identified in the recurrence samples were associated with molecular subtypes (Additional files 4 and 5), including TP53 and NF1 alterations in triple negative tumors (odds ratio > 2.71 p value < 0.05) and CCND1, FGFR1, and ESR1 aberrations in luminal cases (odds ratio < 0.36 p value < 0.05).

In 70 cases, we successfully analyzed primary tumors and matched recurrences with the same NGS panel. Among these, 61 patients had a single primary and relapse specimen (Fig. 3). We found that 55.8% of driver alterations were shared between primary tumors and recurrences, with a median number of two aberrations in common per sample (range 0–7). Including the variants of unknown significance, the prevalence of shared alterations was 61.2% (median = 5; range 0–11) (Additional file 7). Moreover, the most recurrent driver alterations identified in the primary samples were maintained in the recurrence, including mutations of TP53 and PIK3CA and copy number gain of CCND1, FGF19, FGF3, and FGFR1 (Fig. 4). The number of shared aberrations was significantly associated with the time to relapse. Considering the median relapse time (50 months) of the cases successfully profiled, those with early and late recurrences showed a different proportion of shared aberrations (66.7% versus 56%, respectively) and driver (62.9% versus 49.3% respectively) alterations (Additional file 8). Indeed, in 39 of 61 patients (63.9%), additional private alterations were detected in the recurrence samples only, affecting breast cancer-related genes as ERBB2, ESR1, FGFR1, or NF1, and including 12 patients (19.7%) with clinically relevant alterations according to the OncoKB levels of evidence V2 ranking (levels 1–3) (Additional files 9 and 10). Nine cases had multiple primary and/or recurrence samples available for the analysis (Additional file 11). Although spatial and temporal heterogeneity was seen, most driver alterations were retained in the different recurrence samples.