Imprint of parity and age at first pregnancy on the genomic landscape of subsequent breast cancer

All analyses were performed on a publicly available dataset comprising 560 breast cancer patients referred to as BRCA560 [14]. Clinical data, sequencing coverage, and mutational load were obtained from Additional file 1: Table S1–S3 in that reference. Coding driver mutation events and the contribution of mutational signatures were obtained from Additional file 1: Table S14 and S21 in that reference. Raw count data from RNA sequencing were obtained from the authors. Results from HRDetect classifier were obtained from Additional file 1: Table S4 in reference [15].

Eligible patients from BRCA560 were those with samples collected from primary tumor only (patients with local recurrence or metastasis samples were excluded, n = 8) who had available information on parity. There were only two available HER2+ patients (both parous) in the transcriptomic analysis so we preferred to exclude them from this analysis. For each patient, we determined the breast cancer intrinsic subtype (PAM50) using genefu R/Bioconductor package [16]. Nulliparous patients were defined as women with breast cancer who had no full-term pregnancy at the time of breast cancer diagnosis. Parous patients were defined as women with breast cancer who had at least one full-term pregnancy at the time of breast cancer diagnosis. Early parous patients were defined as ≤ 25 years of age at first full-term pregnancy, while late parous patients were defined as > 25 years of age at first full-term pregnancy. PABC patients were defined as women diagnosed not during pregnancy but up to 10 years after the first pregnancy. As referred to the publicly available BRCA560 dataset [14], the Internal Review Boards of each participating institution approved the collection and use of samples of all patients in this study.

The percentage of TILs was independently evaluated by two pathologists (R.S. and H.M.H.) on hematoxylin and eosin slides using the International TILs Working Group 2014 methodology as described before [17]. There were 242 original samples with evaluable TILs from 239 patients. For the three patients with two samples, the arithmetic averages were obtained. We obtain a final set of 231 patients with primary tumor only (patients with local recurrence or metastasis samples only were excluded, n = 8). TIL information for patients for which evaluation from only one pathologist was available were discarded (n = 3). The TILs were scored in the context of an International Cancer Genome Consortium (ICGC) project on the immune characterization of this series, and the scoring is available for all institutions having the ICGC Data Access Compliance Office (DACO) approval. The data will be made broadly available when this global project will be finished or before upon request to the authors.

Except for age at diagnosis that was considered as a continuous variable and therefore compared using the non-parametric Mann–Whitney U test, differences in other clinicopathological characteristics of breast cancer between groups were analyzed using the χ² test or the Fisher exact test when appropriate. All statistical tests comparing groups were done using the non-parametric Mann–Whitney U test and the χ2 test or the Fisher exact test when appropriate for continuous and categorical variables, respectively. For the multivariate analysis, we used a linear and logistic regression to assess the independent association of continuous (log transformed) and categorical variables respectively with—parity (nulliparous vs parous) or—age at first pregnancy (≤ 25 years vs. > 25 years) controlling for age at diagnosis, pathological stage, molecular subtypes by IHC, and histological subtypes. For WGS results, we also corrected for log-transformed sequence coverage of tumor and normal samples (continuous). All interaction and multivariate tests (P_adj) were done using the analysis of variance to compare the models with and without the extra term. Because continuous variables contain zeros, the logarithmic transformation was applied as follows: log₁₀(x + 1). The Kruskal–Wallis test was used to test if MYC expression originates from the same population according to the status of MYC/TP53 alterations. All correlations were calculated using the non-parametric Spearman’s rho coefficient. All reported P values were two-tailed. Multiple testing correction was performed using the false discovery rate method (FDR) [18], and differences were considered significant when the FDR was < 0.05. All analyses were done in R software version 3.3.3 (available at www.​r-project.​org) and Bioconductor version 3.6. Differential expression analysis was performed with DESeq2 v.1.14.1 R/Bioconductor package [19] on raw count data (20,724 genes). Significantly differentially expressed genes were selected with a FDR of < 0.1, independent filtering was performed using default parameters to select a set of genes for multiple test correction which maximizes the number of adjusted P values less than a given critical value alpha (by default 0.1) and differentially expressed genes were identified by using the default cutoff of P value adjusted for multitesting < 0.1. We used gage v.2.24.0 R/Bioconductor package [20] to identify significantly enriched pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG) [21] and biological process from Gene Ontology with the log2FoldChange from DEseq2 results as input data.

From a publicly available dataset comprising 560 breast cancer patients [14], a total of 313 with available information on parity were included. We identified 264 (84.3%) parous and 49 (15.7%) nulliparous patients (Additional file 2: Figure S1). In the parous group, 153 patients (57.9%) had available information on age at first pregnancy (median of 25 years, range 16–46 years). Parous patients were divided into two groups: 82 early and 71 late parous patients by using the median age at first pregnancy as a cutoff value. All patients had available somatic mutations and SCNAs data, 182 patients (58.1%) had available transcriptomic data, and 170 patients (54.3%) had information on TIL levels (Additional file 2: Figure S2).

Table 1 summarizes the clinicopathological features of patients. Compared to parous patients, nulliparous patients had larger tumors (tumor size > 2 cm, 59.2% vs. 37.1%, P = 0.006), higher frequency of lymph node positive disease (40.8% vs. 27.7%; P = 0.027), and lower frequency of triple negative breast cancer (TNBC) (4.1% vs. 23.2%; P = 0.001). Compared to early parous patients, late parous patients had a younger age at breast cancer diagnosis (median, 49 years; range, 28–81 years vs. median, 59 years; range, 34–81 years; P = 2.58 × 10⁻⁵) and were more often pre-menopausal (45.3% vs. 20.9%; P = 0.005). Late parous patients also had smaller tumors (tumor pathological size > 2 cm, 40.8% vs. 51.2%, P = 0.012), lower frequency of TNBC (19.7% vs. 39%, P = 0.026), and higher frequency of lobular histological subtype (14.5% vs. 2.5%, P = 0.01).

In the following sections, we investigated the imprint of parity and age at first pregnancy on breast cancer biology by using a systematic multivariate analysis adjusted for potential confounders, namely age at diagnosis, pathological stage, molecular subtypes by IHC, and histological subtypes.

We first sought to investigate the imprint of parity and age at first pregnancy on somatic mutational load (Additional file 1: Table S1). There was no significant difference in the total number of substitutions (SNVs) according to parity nor age at first pregnancy (P_adj = 0.097, P_adj = 0.075, respectively, Fig. 1a). There was no significant difference in the total number of insertions or deletions (Indels) according to parity (P_adj = 0.464, Fig. 1a). In contrary, compared to tumors from late parous patients, tumors from early parous patients were significantly associated with a higher Indels load (P_adj = 0.002, FDR = 0.007, Fig. 1a). There was no significant difference between the total number of rearrangements according to parity nor age at first pregnancy (Additional file 1: Table S1).

We next interrogated the influence of parity and age at first pregnancy on the frequency of mutations in breast cancer driver genes. Among the driver mutated genes, seven had at least one non-silent mutation with a frequency of > 5% across the whole cohort (Additional file 1: Table S2). As expected, PIK3CA and TP53 were the most frequently mutated genes (Fig. 1b). None of the driver mutated genes were associated with parity in the multivariate analysis. However, in the parous group, early age at first pregnancy was independently associated with higher frequency of TP53 mutations (41/82 (50%) vs. 16/71 (22.5%); P_adj = 0.010; FDR = 0.046, Fig. 1b) and lower frequency of CDH1 mutations (1/82 (1.2%) vs. 9/71 (12.7%); P_adj = 0.013; FDR = 0.046, Fig. 1b). Considering the distribution of TP53 mutations type, early parous patients had a significantly higher frequency of truncating mutations as compared to late parous patients (21/82 (25.6%) vs. 5/71 (7%); P_adj = 0.014, Additional file 2 : Figure S3). Altogether, our results show that age at first pregnancy is associated with biological differences in the mutational landscape of subsequent breast tumors with early parity associated with higher Indels burden and higher frequency of deleterious mutations in TP53 gene.

Somatic copy number alterations (SCNAs) play a major role in breast cancer biology [22, 23]. We identified five driver genes with a frequency of SCNAs > 5% across all patients (Additional file 1: Table S3). MYC tended to be more frequently amplified in parous than in nulliparous patients (49/264 (18.6%) vs. 2/49 (4.1%); P_adj = 0.052; FDR = 0.26, Fig. 1b). In the parous group, MYC amplification was significantly more frequent in the early parous group than in the late parous group (23/82 (28%) vs. 5/71 (7%); P_adj = 0.008; FDR = 0.040, Fig. 1b). When evaluating the co-occurrence of SCNAs and somatic mutations, we found that co-occurrence of MYC amplification and TP53 mutations was independently associated with age at first pregnancy, with early parous patients having a higher frequency of simultaneous alterations of MYC and TP53 genes (15/82 (18.3%) vs. 3/71 (4.2%); P_adj = 0.087, Fig. 2a and Additional file 2: Figure S3). Taken together, these results suggest that age at first pregnancy may also shape the somatic copy number alteration profiles of subsequent breast cancer.

We investigated the BRCAness status of tumors according to reproductive history. Since germline BRCA1/2 mutation status was not available for all samples, we used HRDetect to identify BRCA1/BRCA2-deficient samples [15]. We did not find any significant differences in the proportion of BRCA1/BRCA2-deficient patients between nulliparous and parous groups (6/49 (12.2%) vs. 50/264 (18.9%), respectively, P_adj = 0.386) nor between early and late parous group (25/82 (30.5%) vs. 14/71 (19.7%), respectively, P_adj = 0.473). Thus, reproductive history and age at first pregnancy do not seem to affect homologous recombination DNA repair capacity in subsequent breast cancer.

RNA sequencing data were available for a subset of 182 patients, of which 34 were nulliparous (Additional file 2: Figure S2 and Additional file 1: Table S4). We first determined the intrinsic molecular subtype distribution of breast cancer using the PAM50 classifier [24]. We did not find a significant difference in the distribution of the PAM50 subtypes between nulliparous and parous (Fig. 1d). In contrast, early parous patients had a higher proportion of basal-like subtype tumors (15/51 (29.4%) vs. 4/45 (8.9%); P = 0.009, Fig. 1d, Additional file 1: Table S4). In order to identify de novo gene expression profiles that might be associated with parity and age at first pregnancy, we performed a multivariate differential expression analysis using DEseq2 [19] controlling for age at diagnosis, pathological stage, molecular subtypes by IHC, and histological subtypes. A total of 62 genes were differentially expressed between nulliparous and parous (Additional file 1: Table S5). Among these genes, three were associated with mammary development; OXTR and ATP2B2 were downregulated whereas NRG3 was upregulated in nulliparous. Pathway analysis using the generally applicable gene-set enrichment (GAGE) analysis [20] revealed an enrichment of genes related to extracellular matrix (ECM) receptor interaction in parous (Additional file 1: Table S6). When comparing early and late parous patients, 466 genes were differentially expressed, among which 305 were upregulated in early parous (Additional file 1: Table S7). However, pathway analysis did not reveal any significant enrichment of relevant biological processes (Additional file 1: Table S8).

Due to the higher frequency of MYC amplification in early parous patients, we determined if this would also impact MYC at the mRNA expression levels. Early parous patients were independently associated with an upregulation of MYC expression (P_adj = 0.0113, Fig. 2b). We also evaluated the expression of MYC according to TP53 mutations and MYC amplification and found that MYC expression was the highest in tumors harboring concurrent TP53 mutation and MYC amplification (Fig. 2c).

Previous reports have hypothesized that the pregnancy-induced tumor protection could be attributable to an improved anti-tumor immunity [25‐29]. Therefore, we assessed whether reproductive history could be associated with tumor-infiltrating lymphocyte (TIL) levels that is considered as a surrogate of tumor immunogenicity (Additional file 1: Table S9). We did not find any significant difference in the proportion of stromal TILs according to parity or according to the age at first pregnancy (P_adj = 0.655; P_adj = 0.200, respectively, Fig. 2d). Similarly, no differences were observed when comparing intratumoral TILs according to parity or age at first pregnancy (P_adj = 0.240; P_adj = 0.889, Additional file 1: Table S9). Thus, reproductive history does not seem to influence breast tumor immune microenvironment.

Pregnancy-associated breast cancer (PABC) can be defined as cases diagnosed up to 10 years postpartum [30]. In this cohort, we identified 17 PABC patients and compared them with 49 nulliparous patients. Compared to nulliparous, PABC patients had a younger age at diagnosis (median, 38 years; range, 28–48 years vs. median, 54 years; range, 30–81 years; P = 5.79 × 10⁻⁶, Additional file 1: Table S10) and higher frequency of TNBC (5/17 (29.4%) vs. 2/49 (4.1%); P = 0.021, Additional file 1: Table S10). We did not find any significant differences between PABC and nulliparous in the pattern of somatic mutations or somatic copy number alterations (Additional file 1: Table S1-S3). Nine PABC had available gene expression and TIL scoring. At the transcriptomic level, we found that PABC patients were associated with enrichment of biological processes related to immune function (Fig. 3a and Additional file 1: Table S11). Moreover, PABC patients had an increased lymphocytic infiltration both for stromal and intratumoral TIL levels (P_adj = 0.040 and P_adj < 0.0001, respectively; Fig. 3b, c). Taken together, these results indicate that cancer occurring in the postpartum breast is associated with increased TIL levels.