Molecular and Clinical Characterization of Postpartum-Associated Breast Cancer in the Carolina Breast Cancer Study Phase I–III, 1993–2013

This article is featured in Highlights of This Issue, p. 493

Breast cancer is the most commonly diagnosed cancer among women aged 20 to 49 years (1) and the cancer most frequently temporally-related to pregnancy (2). Several studies have found that breast cancer diagnosed during pregnancy or the postpartum years is associated with worse overall and disease-free survival, and plausible mechanisms have been proposed (3–9). Elevated concentrations of estrogen, progesterone, and prolactin during pregnancy may promote the growth and progression of breast tumors (10–12). Postpartum involution, a breast remodeling process that generates a wound-healing/inflammatory microenvironment (13–16), promotes tumor growth in animal studies (17, 18). However, despite the strength of the preclinical and mechanistic data, epidemiologic and clinical studies of breast tumors diagnosed during pregnancy or the postpartum years have reported incongruent results with respect to molecular characteristics (3–8, 19–24).

Inconsistencies in the molecular phenotypes of postpartum-associated breast cancer (PPBC) may be in part due to small sample size, varying definitions that combine breast cancers diagnosed during pregnancy (PrBC) with PPBC (25), and differences in the menopausal status of the study participants. Use of different reference groups has also complicated comparisons across studies. For instance, previous studies have either used only nulliparous women as the reference group (4, 21–24) or combined nulliparous women and parous women diagnosed with breast cancer remotely (after a given number of postpartum years) in one reference group (5–8, 19, 20). To identify the effects of recent pregnancy, an ideal referent group would restrict to premenopausal women and account for age differences. Furthermore, adequate sample size, clear definitions of recently postpartum women (i.e., ≤10 years prior to diagnosis), and distinct comparison groups, including remotely postpartum (>10 years prior to diagnosis) and nulliparous women, would improve inferences and comparisons across studies.

We examined the distribution of a wide array of biological features including ER, progesterone receptor (PR), HER2, p53, IHC- and RNA-based subtypes, risk of recurrence multigene (ROR-P) scores, and multi-gene immune scores among premenopausal women <50 years of age according to parity and recency of pregnancy using data from a large population-based study. Based on prior literature, we used 10 years as the cutoff for our recently postpartum group (3, 13, 25), and we conducted sensitivity analyses and explored smaller windows of time since last childbirth.

The Carolina Breast Cancer Study (CBCS) is a population-based study of breast cancer in North Carolina (phase I: 1993–1996; phase II: 1996–2001; and phase III: 2008–2013; N = 4,806); details of the study have been described previously (26, 27). Briefly, the primary study enrolled women age 20 to 74 years diagnosed with first primary invasive breast cancer using rapid case ascertainment, with randomized recruitment to oversample Black and younger (age <50 years) women (28). This analysis examined tumor characteristics of premenopausal women under 50 years of age, who contributed tumor tissue (N = 2,081; Supplementary Fig. S1). Written informed consent was obtained from each participant prior to data collection, and the study was conducted in accordance with U.S. Common Rule. The study was approved by the Office of Human Research Ethics/Institutional Review Board at the University of North Carolina at Chapel Hill.

In-person interviews were conducted by trained nurses to collect medical history including detailed information on pregnancy history. Date of breast cancer diagnosis was collected using medical record abstraction. Time since last full-term birth (≥7 months of pregnancy or live birth) was calculated by subtracting date of last full-term birth from date of breast cancer diagnosis. Cases with missing date of last full-term birth were excluded (N = 2). Breast cancers diagnosed during pregnancy were excluded (N = 15). Women diagnosed within 10 years of their last childbirth were included in the “recent” group (N = 618). Women diagnosed more than 10 years after their last childbirth were assigned to the “remote” group (N = 1,086). Women who never had a full-term birth prior to their diagnosis were assigned to the “nulliparous” group (N = 360).

Tumor size, grade, node status, and stage were abstracted from medical records and pathology reports. The multiple imputation plus outcome approach was used to impute missing tumor grade values for CBCS phase II participants (n = 578; ref. 29). Estrogen receptor (ER), progesterone receptor (PR), and HER2 status were abstracted from medical records or determined from formalin-fixed, paraffin-embedded (FFPE) tumor tissue blocks, which were previously sectioned and stained for a panel of IHC markers at the Translational Pathology Laboratory at the University of North Carolina, Chapel Hill; details have been described previously (30). Briefly, ER and PR were considered positive if the percentage of positive cells was ≥10%, based on previous work showing that 10% positivity was most closely aligned with RNA-based subtypes (30), and HER2 positivity was defined as IHC 3+ (31). Details of IHC-based breast cancer subtyping have been published previously (26, 32). Briefly, IHC-based subtype was defined as luminal A-like (ER⁺ or PR⁺ and HER2⁻), luminal B-like (ER⁺ or PR⁺ and HER2⁺), HER2-type (ER⁻ and PR⁻ and HER2⁺), or triple-negative (ER⁻ and PR⁻ and HER2⁻).

Tumor-infiltrating lymphocytes (TIL) scores were calculated by adapting a digital algorithm similar to that previously described by Sandhu and colleagues (33) and optimized based on standardized international guidelines on TIL assessment in breast cancer by the International TILs Working Group (34). Briefly, composition algorithms were trained on several annotated representative hematoxylin and eosin (H&E)-stained tumor slides for each tissue feature (immune, epithelial, stromal, adipose) with accuracy optimized to match visual annotation by a trained pathologist's assistant. Percentage area and nuclei per unit area (nuclear density) were calculated for each region.

Details of RNA isolation and quantification have been described previously (35). Briefly, RNA was isolated and quantified using Qiagen FFPE RNeasy kit and NanoString assay, respectively (36). Data that passed the NanoString nSolver software quality control were normalized using methods previously described by Bhattacharya and Hamilton and colleagues (37), including reference gene normalization, upper-quartile normalization (38), and Remove Unwanted Variation (RUV) using the RUVg function from the RUVSeq Bioconductor package (39). Only FFPE tumor blocks with adequate tumor cellularity by manual review were included in RNA analyses. Relative to participants without RNA data, those with RNA data had larger tumors, higher tumor stage, and higher grade and were more likely to be lymph node positive, but age at diagnosis or time since last childbirth were not significantly different.

P53 status was determined using a previously validated TP53-dependent signature (40). Tumors were classified as p53 mutant-like versus wild-type–like based on a similarity-to-centroid approach (Pearson coefficient) using distance-weighted discrimination (41). RNA-based intrinsic subtype was determined using the PAM50 gene signature to classify tumors as luminal A, luminal B, HER2-enriched, basal-like, and normal-like, as previously described (35). Normal-like samples were excluded because this subtype is believed to arise due to insufficient tumor cellularity. The PAM50 predictor was used to calculate the risk of recurrence (ROR) score, which incorporates subtype and additional weighting by the proliferation signature (ROR-P) (35). To identify the most aggressive tumors, ROR-P was categorized as low/medium versus high.

We curated a 48-gene panel of immune markers that is representative of at least 10 individual immune cell types based on work from Bindea (42) and Danaher and colleagues (43). As described in Hamilton and colleagues (in preparation), cell type scores for 10 immune cell types, including B Cells, T cells, Th cells, T regulatory cells (Treg), T follicular helper cells (Tfh), CD8 T cells, Neutrophils, Eosinophils, natural killer (NK) cells, macrophages, and the immune checkpoint marker, PDL1, were calculated by computing the median of the log₂-transformed expression from each cell's marker genes. For CD8 T cells specifically, we elected to include genes that were both specific and sensitive. These less specific markers, such as GZMM and PRF1, have been shown to be expressed on both NK cells and CD8 T cells; thus, we refer to these cytotoxic cells as CD8 T cells/NK cells in downstream analysis. In addition to cell type scores, overall adaptive and innate immunity scores were calculated, where the median log₂-transformed expression for all B-cell, T-cell, CD8 T-cell, Th cell, Treg, and Tfh markers were computed for the adaptive scores, and the median of all eosinophil, neutrophil, NK cell, and macrophage markers were calculated for the innate scores.

Race was determined by self-report and categorized as non-Black or Black. Less than 2% of non-Black participants self-identified as multiracial, Hispanic, or other race/ethnicities. Age at diagnosis was obtained from the baseline survey and used as a continuous variable in models. Age at last full-term pregnancy was obtained from baseline survey and categorized as <30 years and ≥ 30years. Information on parity was obtained from baseline survey and categories as nulliparous, or 1, 2, and ≥3 full-term births.

Descriptive statistics for patient sociodemographic and tumor characteristics of recently postpartum women were compared with distally postpartum and nulliparous women using χ² tests (Fisher exact test when cell count <5). Binary logistic regression models were used to compute prevalence odds ratios (POR) and 95% confidence intervals (95% CI) as the measure of association between tumor characteristics and recency of pregnancy. The following tumor characteristics were studied in association with recency of pregnancy: tumor grade (low-intermediate, high), stage of disease (I/II, III/IV), lymph node status (positive, negative), tumor size (≤2 cm, >2 cm), ER (negative, positive), PR status (negative, positive), HER2 status (negative, positive), IHC-based subtype (luminal A, luminal B, HER2-type, triple negative), p53 status (wild-type, mutant), PAM50 intrinsic breast cancer subtype (luminal A, luminal B, HER2-enriched, basal-like), and ROR-P score (low, medium, high). Models were adjusted for study phase (I, II, III), age at diagnosis (continuous), and race (Black, non-Black). Sensitivity analyses included using smaller intervals of time since last full-term birth (≤ 2, >2–5, >5–10 years postpartum) and stratifying by age at last full-term pregnancy (<30 years, ≥30 years). We performed multiple sensitivity analyses, including adjusting for body mass index (BMI) at diagnosis (<25, 25-<30, ≥30), parity (1, 2, ≥3), lactation history (ever, never), and oral contraceptive use (never, current, former). All analyses were conducted in SAS version 9.4 (SAS Institute, Cary, NC). P values were two-sided with an α of 0.05 for statistical significance.

For participant confidentiality and due to ethical restrictions, data are available upon request and are subject to data use agreements and other stipulations. Permission to access data from the Carolina Breast Cancer Study may be obtained online (https://unclineberger.org/cbcs/) or by contacting the authors.

Among premenopausal breast cancer cases <50 years at diagnosis in the CBCS (phase I–III, 1993–2013), Table 1 presents the distribution of age at diagnosis, age at last full-term pregnancy, race, parity, and study phase for recently postpartum (diagnosed ≤10 years postpartum), remotely postpartum (diagnosed >10 years postpartum), and nulliparous women diagnosed with breast cancer. Recently postpartum women had a significantly younger age at diagnosis (median = 39 years) compared with remotely postpartum (median = 45 years) and nulliparous (median = 42 years). Compared with remotely postpartum, recently postpartum also had an older age at last full-term pregnancy (median = 33 years vs. 26 years) and a higher frequency of Black race. There was no significant difference in multiparity or study phase between groups.

Although differences for several factors were observed when comparing recently postpartum with nulliparous women, there were few differences between recently and distally postpartum women (Table 2). Compared with nulliparous women, tumors among recently postpartum were more likely to be lymph node–positive, ER/PR-negative, and IHC-based triple negative subtype, after adjusting for age at diagnosis, race, and study phase. Although not statistically significant, recently postpartum women tended to be diagnosed at a later stage and were more frequently IHC-based HER2-type subtype compared with nulliparous women. Sensitivity analysis considering smaller windows of time since last childbirth showed strongest associations with respect to lymph node positivity and later stage for those 0 to 2 years and 2.1 to 5 years postpartum versus nulliparous women (Supplementary Tables S1 and S2). No substantial difference in grade, size, and HER2 status were observed between the recent and nulliparous groups.

Compared with remotely postpartum women, recently postpartum women did not have more aggressive clinical and molecular tumor characteristics (Table 2). After adjusting for age at diagnosis, race, and study phase, no significant differences in tumor grade, stage, lymph node status, and ER/PR/HER2 status were observed between recent and remote groups and recently postpartum women had significantly smaller tumors compared with remotely postpartum women.

No significant differences in RNA-based intrinsic subtypes, TP53 status, and ROR-P scores were observed between the recent and nulliparous groups (Table 3). However, compared with remotely postpartum, recently postpartum women had significantly more frequently wild-type TP53 status, and more frequent non–basal-like subtype. No substantial or significant differences in ROR-P scores were observed between the recent and remote groups. Sensitivity analyses conducted with time since pregnancy as a continuous variable showed similar results. Adjusting for BMI at diagnosis, parity, lactation history, and oral contraceptive use did not substantially alter POR estimates. Sensitivity analysis stratifying on age at last full-term pregnancy or including smaller time intervals of recency did not alter the relationships between parity, recency, and RNA-based features (Supplementary Tables S3 and S4). Similarly, stratifying on prior parity status or restricting to only CBCS phase III participants did not change these results.

Despite few changes in epithelial tumor features, we observed an enrichment for adaptive immune cells and higher immune-cell composition among recently postpartum women compared with nulliparous women (Fig. 1A). After adjusting for intrinsic subtype, race, age, and study phase, we saw an enrichment for overall adaptive immunity and adaptive immune cells, such as T cells, B cells, CD8 T cells, activated CD8 T cells/NK cells, and Tfh cells in recent versus nulliparous women (Fig. 1A). Consistent with these findings, tumors from recently and remotely postpartum participants had higher immune-cell composition than tumors from nulliparous participants (Fig. 1B). When comparing remote versus nulliparous women, CD8 T cells were also significantly enriched in remote versus nulliparous women (Fig. 1C); however, many fewer immune cell scores were significant. Nonetheless, no significant differences in immune cell scores were observed between recently versus remotely postpartum women (Fig. 1D). Box plots for signatures that were significant in Fig. 1A showed that the level of each immune cell score was elevated in both recently and remotely postpartum compared with nulliparous, but were modestly attenuated in remotely postpartum (Fig. 1E).

Among premenopausal women, we observed few differences between the tumors of recently and remotely postpartum women in the CBCS. While parity status seems to be associated with breast cancer clinical and molecular tumor characteristics, recent pregnancy among parous women had weak or null associations with clinical and molecular characteristics, except when considering the most recent postpartum (≤5 years) and lymph node status and stage. However, consistent with the postpregnancy wound healing microenvironment hypothesis (13), we observed an enrichment for adaptive immune cells and higher immune-cell composition among recently postpartum women compared with nulliparous women. CD8 immune changes appeared to persist long term, but other immune response changes were attenuated in remotely postpartum versus nulliparous women.

Three previous studies have reported significant differences in some key tumor characteristics between recently postpartum compared with nulliparous women (3, 21, 22). Consistent with our findings, Goddard and colleagues observed an increased lymph node involvement among PPBC women compared with nulliparous women (3). Pilewskie and colleagues found breast cancers diagnosed within 2 years of pregnancy were more likely to be lymph node–positive, hormone receptor—negative, and triple negative compared with nulliparous women, after conditioning multivariate models on categorical age at diagnosis (21). Similarly, Nagatsuma and colleagues reported ER/PR-negative and triple negative tumors to be more common among recently postpartum (≤2 years before diagnosis) women compared with nulliparous women, in a Japanese hospital-based study (22). While these latter studies used a narrower postpartum window than we did to define recency, we observed that several of the differences between recently postpartum and nulliparous women may continue to persist for many years postpartum similar to Goddard and colleagues (3).

Some previous literature seems to contradict the findings of our research. Specifically, we found ER/PR-negative and IHC-based triple negative subtype tumors to be more common among recently postpartum (≤10 years postpartum) compared with nulliparous women, but two other groups found no difference in ER/PR status or IHC-based subtype between recently postpartum (<5 years postpartum) compared with nulliparous young women (4, 24). However, compared with Callihan and colleagues we had several underlying methodologic differences. For example, we adjusted for age, while Callihan and colleagues did not (4). The population studied in Callihan and colleagues has been expanded and the analysis has been updated by Goddard and colleagues in a recent publication corroborating our findings (3). The study by Collins and colleagues also found no significant differences in ER/PR status and IHC-based subtype distribution among women diagnosed within 5 years of their last pregnancy compared with nulliparous women. However, patient ER status definition varied within their study population as their enrollment period began in 2006 and lasted beyond 2010, when ER positivity classification changed from ≥10% to ≥1% of tumor cells; thus, women who would have been classified as ER-negative before 2010 were classified as ER-positive if enrolled after 2010, leading to a slight reduction in ER-negative and triple negative cases (24). Additionally, the Collins and colleagues study population largely consisted of white women and the distribution of ER status has been shown to vary by race (24). An advantage of the current analysis is that it included a significant representation of Black women, a group that has been underrepresented in previous analyses. Some findings of Collins and colleagues were consistent with ours, notably, neither study observed significant difference in tumor grade or stage comparing recently postpartum women with nulliparous women (24).

Only two previous studies have reported on the distribution of tumor characteristics among women >10 years postpartum (remotely postpartum; refs. 3, 4). Similar to our results, both of these studies have reported no significant difference in breast cancer grade, stage, nodal status, ER/PR status, HER2 status, and IHC-based subtypes between the recently and remotely postpartum groups.

To our knowledge, no other studies have examined RNA-based p53 signature, RNA-based intrinsic subtype, or ROR-P scores with respect to recently postpartum and nulliparous premenopausal women at breast cancer diagnosis. However, one previous study did evaluate p53 status by IHC. Our p53 findings corroborate those reported by Pilewskie and colleagues, who found no significant difference in IHC-based p53 status between recently postpartum and nulliparous women at breast cancer diagnosis (21).

Our study is unique among the aforementioned studies in that we examined the immune microenvironment of tumors diagnosed according to time since last childbirth (recently vs. remotely postpartum women). Our results showed an enrichment in adaptive immune cells and a higher immune-cell composition among recently postpartum women (≤10 years postpartum), which is attenuated among remotely postpartum women, compared with nulliparous women. Our findings are consistent with previous literature, which has noted an upregulation of immune-related genes in tumor-associated stroma of animal models (44, 45) or pregnant women (46). Within normal breast tissue, this enrichment of inflammatory and immune response genes may persist many years after the completion of a pregnancy (47, 48). In animal models, tumors during pregnancy or within the postpartum had an enrichment of immune responsiveness genes and immune-cell infiltration (44, 45). Harvell and colleagues noted that compared with nulliparous or women diagnosed >5 years postpartum, tumors diagnosed during pregnancy overexpressed immune response genes (46). Lastly, in normal breast tissue, genes in the immune response and inflammation pathways have been observed to be upregulated among parous compared with nulliparous women (47, 48).

Previously in the CBCS population, we observed that breast cancer–specific and overall survival were worse in recently postpartum women compared with both remotely postpartum and nulliparous women (49). Several previous studies have also reported worse overall and disease-free survival among recently postpartum women (3–9). Therefore, we hypothesized that tumor characteristics might explain the poor survival observed within this population. However, defining the ideal comparison group for recently postpartum women is complex because age, menopausal status and parity are strongly associated with tumor characteristics. Addressing this complexity was a strength of our analyses as we showed comparisons to two different populations (nulliparous women and remotely postpartum parous women), restricted our analysis to premenopausal women less <50 years of age, and controlled for the variation in age between comparison groups. To fully understand how recency of pregnancy affects outcomes, analyses should examine breast cancer recurrence or disease-specific mortality and incorporate both molecular data and treatment information, but our study population (CBCS phase III) has too few recurrences among recently postpartum women (n = 56) to address these subgroups.

Results suggesting a distinct immune microenvironment in recently postpartum women is an intriguing finding. It is increasingly recognized that immune microenvironment plays a key role in response to treatment, and as the science of immune response and chemotherapy response is better defined, there may be implications for understanding PPBC outcomes. Future studies should examine the interplay between tumor microenvironment, postdiagnostic treatment, and breast cancer recurrence/progression among recently postpartum women.

S.N. Vohra reports grants from NIH and Susan G. Komen for the Cure during the conduct of the study. K.E. Reeder-Hayes reports grants from Pfizer Foundation outside the submitted work. No disclosures were reported by the other authors.

S.N. Vohra: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. Walens: Formal analysis, visualization, writing–original draft, writing–review and editing. A.M. Hamilton: Data curation, methodology, writing–original draft, writing–review and editing. M.E. Sherman: Conceptualization, writing–review and editing. P. Schedin: Conceptualization, writing–review and editing. H.B. Nichols: Conceptualization, writing–review and editing. K.E. Reeder-Hayes: Conceptualization, writing–review and editing. A.F. Olshan: Conceptualization, funding acquisition, writing–review and editing. M.I. Love: Conceptualization, methodology, writing–review and editing. M.A. Troester: Conceptualization, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

S.N. Vohra was supported by the University Cancer Research Fund of North Carolina, the Komen Graduate Training in Disparities Research (grant no. GTDR16381071), the Doctoral Degree Advancement Award by the University of North Carolina (UNC) – Initiative for Minority Excellence, and the UNC Cancer Control Education Program (T32CA057726). M.A. Troester was supported by the NCI of the NIH (P01CA151135) and the NCI Specialized Program of Research Excellence (SPORE) in Breast Cancer (NIH/NCI P50-CA58223). The Carolina Breast Cancer Study was funded by the University Cancer Research Fund of North Carolina and Susan G. Komen for the Cure (OGUNC1202).

This research was supported by a grant from UNC Lineberger Comprehensive Cancer Center, which is funded by the University Cancer Research Fund of North Carolina, the Susan G. Komen Foundation (OGUNC1202), the Komen Graduate Training in Disparities Research Grant (GTDR16381071), the NCI of the NIH (P01CA151135), and the NCI Specialized Program of Research Excellence (SPORE) in Breast Cancer (NIH/NCI P50-CA58223). This research recruited participants and/or obtained data with the assistance of Rapid Case Ascertainment (RCA), a collaboration between the North Carolina Central Cancer Registry and UNC Lineberger. RCA is supported by a grant from the NCI of the NIH (P30CA016086). We are grateful to CBCS participants and study staff. We also acknowledge the late Robert C. Millikan, founder of the CBCS phase III.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.