Clonal Hematopoiesis in Young Women Treated for Breast Cancer

Despite significant advances in breast cancer therapy, adjuvant radiotherapy and chemotherapy remain mainstays of treatment for a large proportion of women with localized breast cancer. Although effective at reducing the risk of breast cancer relapse, these treatments carry risks of short- and long-term complications, including increased rates of cardiovascular disease, development of premature menopause, and evolution of therapy-related myelodysplastic syndrome (t-MDS) and acute myeloid leukemia (t-AML; refs. 1–4). These risks are of heightened importance in young patients who, when cured of their breast cancer, have the potential to live for many decades.

Clonal hematopoiesis (CH) is the expansion of clonally related blood cells due to the acquisition of somatic mutations that accumulate during the normal aging of hematopoietic stem cells (5). Clonal hematopoiesis of indeterminate potential (CHIP), a subset of CH in which the somatic mutation in the blood clone is present at variant allele fraction (VAF) of at least 0.02 in an individual with no overt hematologic malignancy, is common and age-associated (6). The mutations found in CHIP are largely restricted to a select number of genes; the most common mutated genes in an aging population are DNTM3A, TET2, and ASXL1 (7–9). Individuals with CHIP are not only at a significantly elevated risk of developing a myeloid malignancy, but also have increased rates of numerous additional ill-health effects including ischemic cardiovascular disease, chronic obstructive pulmonary disease, osteoporosis, and premature menopause (10–16).

In patients receiving cytotoxic cancer therapies, CHIP is both more common and more likely to harbor mutations in genes that regulate the cellular DNA damage response such as TP53 and PPM1D, the former being associated with a particularly elevated risk of developing MDS and AML (17–19). To date, the majority of studies of CHIP in cancer populations have focused on either nonselected populations or specific cohorts, many of which are older or have been heavily pretreated with multiple rounds and types of chemotherapy (17–21). In many of these populations, including patients with solid tumors, lymphoma, and multiple myeloma, CHIP is associated with increased risk of therapy-related MDS and AML and inferior overall survival (18, 19, 22, 23). The prevalence and consequences of CHIP in younger adult patients with cancer, however, have not been assessed. In this study, we investigated the prevalence, mutational landscape, and clinical significance of CHIP in a prospectively collected cohort of young women treated for breast cancer.

The Young Women's Breast Cancer Study (YWS) is a prospective cohort of women with breast cancer aged 40 years or younger at diagnosis between 2006 and 2016 (24). Enrollment occurred at academic and community centers in the United States and Canada. The study was performed in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board at Dana-Farber/Harvard Cancer Center and other participating sites. Participants signed informed consent prior to enrollment. When possible, peripheral blood samples were obtained from patients near the time of diagnosis (“baseline”), approximately 1 year after diagnosis (“1-year”), and approximately 4 years after diagnosis (“4-year”).

YWS participants completed a baseline survey and were then surveyed twice a year for the first 3 years following diagnosis and annually thereafter. Cancer stage at diagnosis was ascertained from pathology reports and medical record review. Primary breast cancer treatment, including surgery, chemotherapy, radiotherapy, and/or endocrine therapy (within the first year after diagnosis), were obtained through a combination of survey data and medical record review. Similarly, events during follow-up including comorbidities (premature menopause, ischemic cardiovascular disease, AML/MDS), distant breast cancer recurrence, and survival events were self-reported during follow-up and/or extracted from medical records.

Targeted sequencing to identify CH was performed on all samples. DNA was extracted from peripheral blood samples (Qiagen) and subject to error-corrected next-generation sequencing. Hybrid capture using RNA probes from Twist Biosciences was performed for all samples with subsequent sequencing on the Illumina platform (Illumina) as described previously (20). The genes sequenced for the 1-year samples are shown in Supplementary Table S1 and for the baseline and 4-year samples in Supplementary Table S2. For the 1-year samples, mutations with three or more alternative reads were considered in analysis, and curated based on likely pathogenicity and frequency in the population as described previously (18). Mutations with a VAF greater than 0.35 were excluded to avoid inclusion of germline variants. The mutations identified in the 1-year samples were then assessed in the paired baseline and 4-year samples without specific alternative read cutoffs.

The primary endpoints were overall survival (OS) among women diagnosed with stage I to IV cancer and distant recurrence-free survival (DRFS) among women diagnosed with stage I to III cancer as evaluated through Cox proportional hazards regression. The coprimary endpoints uniformly share the type I error of 0.10 used for determining significance. OS was defined as time from diagnosis to death; DRFS was time to first distant recurrence and excluded women with stage IV disease at diagnosis. Patients without events were censored at last follow-up.

Secondary endpoints were ischemic cardiovascular disease, t-MDS/AML, and premature menopause. Tertiary analyses included the association between baseline patient characteristics and CHIP at 1 year and adjuvant treatment for breast cancer and CHIP at 1 year. All secondary and tertiary analyses were conducted with a type I error of 0.05. Analyses of outcome associations with CH at VAF 0.02 were considered exploratory. Continuous variables were evaluated using t tests or Wilcoxon rank-sum tests, where appropriate, and categorical variables were evaluated using Fisher exact test. Patients with stage 0 were excluded from the survival analyses, and stages III and IV were combined given the rarity of stage IV disease in this cohort.

The data generated in this study are available upon request from the corresponding author.

To determine the frequency and distribution of clonal hematopoietic mutations in young women treated for breast cancer, we performed deep, error-corrected sequencing on peripheral blood DNA obtained from 1-year samples from 878 individuals (Fig. 1A). The clinical characteristics of the cohort are shown in Table 1. All study participants were women, and the median age at the time of the 1-year blood draw was 38 years old. At the time of diagnosis, 830 (95%) had localized breast cancer (stage 0–III), whereas the remainder had metastatic disease. Within the first year of treatment, 366 women (42%) received radiotherapy and chemotherapy, 308 (35%) received chemotherapy only, 61 (7%) received radiotherapy only, and 493 (56%) were started on hormone therapy (tamoxifen or aromatase inhibitor). At a median follow-up of 9 years, 764 (87%) patients were alive and 728 (83%) were breast cancer free (data not shown).

We identified CHIP, defined by the presence of a somatic mutation with a VAF greater than 0.02, in 24 patients (2.7% of cohort; Fig. 1B; Supplementary Table S3). The most common CHIP mutations were in DNMT3A and TET2, consistent with what has been observed in most studies of patients without prior exposure to cytotoxic therapy (8, 9). In this study, there was one patient with PPM1D-mutant CHIP and no patients with TP53-mutant CHIP. In exploratory analyses of patients with CH below the VAF threshold for CHIP of 0.02, we identified 252 patients (28.7% of cohort) with pathogenic somatic mutations (minimum VAF 0.0002; Supplementary Table S4; Supplementary Fig. S1). Among these 252 patients, DNMT3A mutations were most common (187 mutations in 152 patients), followed by TET2 (23 mutations in 22 patients), PPM1D (22 mutations in 17 patients), ASXL1 (14 mutations in 13 patients), SRCAP (13 mutations in 13 patients), and TP53 (nine mutations in nine patients; Supplementary Table S4; ref. 25).

With a median follow-up of 9 years, we did not detect any significant differences in either clinical characteristics (patient and tumor characteristics, including pathologic subtype) or survival outcomes between individuals with and without CHIP [HR, 0.85; 95% CI, 0.27–2.69) and 1.06 (95% CI, 0.33–3.39) for OS and DRFS, respectively], although given the limited number of CHIP cases observed in this cohort, we were not powered to identify small differences (Fig. 2A and B; Table 2). During the 9-year follow-up period, the rarity of t-MDS/AML (one case overall in the no CHIP group) and ischemic cardiovascular disease (one case overall, occurred in the CHIP group) precluded formal risk analyses, but did suggest that CHIP was not strongly associated with these outcomes in this cohort. Neither the presence of CHIP nor any CH mutation was associated with self-reported cessation of menopause at the 1-year time point after controlling for age and receipt of chemotherapy (8, 11, 13). In an exploratory analysis of patients with other VAF cutoffs including 1% and any detectable CH clone, we again did not find any significant associations (Supplementary Fig. S2; Supplementary Table S5).

Some prior studies of CH in cancer survivors who received cytotoxic therapy reported an increased rate of CHIP and a mutation distribution enriched for genes involved in the DNA damage response, such as PPM1D and TP53 (17, 18, 21). We did not observe a similar distribution (Fig. 1B; Supplementary Fig. S1), nor were there significant associations between treatments administered for breast cancer (chemotherapy, radiotherapy, and endocrine therapy) and the presence of CHIP or the frequency of CHIP mutations within a given gene. Comparing those who received either chemotherapy or radiotherapy (n = 758) with those who did not (n = 119), we found no significant difference in the frequency of CHIP (2.6% vs. 3.4%, P = 0.55) or any CH mutation (29% vs. 26%, P = 0.59). Although there was a higher rate of chemotherapy or radiotherapy receipt among CH patients with mutations of any VAF in TP53 or PPM1D (n = 24) compared with those without such a mutation (n = 228), the results did not reach statistical significance (92% vs. 74%, P = 0.078 for chemotherapy; 67% vs. 46%, P = 0.085 for radiotherapy; 96% vs. 87%, P = 0.33 for chemotherapy or radiotherapy; Supplementary Table S6).

As the YWS includes prospective collection of biospecimens at multiple timepoints, we obtained all available peripheral blood samples collected at baseline and at 4 years after diagnosis for patients in whom any CH mutation was identified in the 1-year sample. Among all the patients for whom a CH mutation was identified at 1-year posttreatment, we sequenced paired baseline, 1-year, and 4-year samples for 125 patients, paired baseline and 1-year samples for 65 patients, and paired 1-year and 4-year samples for 35 patients (Fig. 3A). Mutations identified in the 1-year samples were assessed by targeted sequencing, and VAFs of detectable mutations were compared across timepoints, as described in Materials and Methods.

For each mutation, the VAFs were highly concordant across time points, with no evidence of significant clonal expansion or contraction over the time interval within which the specimens were collected. This was true regardless of the mutated gene or the VAF of the mutation identified at the 1-year timepoint (Fig. 3B–D). Specifically, we did not observe significant changes in clone size over time when stratified by specific gene mutation (Supplementary Fig. S3A–S3D). In addition, when comparing the frequency of mutations between time points, we did not observe significant differences in the prevalence of mutations in specific genes, suggesting that patient exposures in the intervals between sample acquisition did not drive selection for mutations in particular genes (Supplementary Fig. S3E). Taken together, these data suggest that over the longitudinal sample acquisition, mutations in this cohort were stable without evidence of significant clonal expansion or variation.

The distribution of VAFs for all mutations identified revealed a set of mutations with VAFs greater than 0.35 that were suspicious for being germline heterozygous variants, particularly those with VAFs of approximately 0.5 (Fig. 4A). Among these, we identified 15 mutations that were likely to impair protein function (nonsense, splicing, and frameshift), most commonly in the genes ATM and TP53 (Supplementary Table S7), both of which have been implicated as germline-risk alleles (25, 26). We also identified nonsynonymous variants in ATM (n = 37) and TP53 (n = 6), many of which localized to functional domains of the proteins, suggesting that they may in fact impair protein function (Fig. 4B).

In this cohort of 878 young women with breast cancer, we identified CHIP in 2.7% of the patients from blood samples collected an average of 1 year after diagnosis. We did not observe any association between patient characteristics, tumor characteristics, or treatment with the prevalence or mutation pattern of CHIP, nor was there an association between CHIP and overall survival, distant recurrence-free survival, or the risk of adverse outcomes including t-MDS/t-AML, ischemic cardiovascular disease, or premature menopause. In contrast to other published series of CH in patients with cancer, mutations in the DNA damage response genes TP53 and PPM1D were not enriched in this cohort (17–19). In addition, among the individuals with a CH mutation and longitudinal samples, we did not observe any significant change in the size of the mutant clone, suggesting that there is very limited clonal expansion or contraction over the 4-year time-period studied. Finally, we identified a series of likely germline variants, including in the genes ATM and TP53, both of which have been shown to confer a risk of early-onset breast cancer.

The prevalence of CHIP in this cohort was lower than that reported in other studies of patients with cancer treated with cytotoxic therapy, even when considering differences in technical factors such as sequencing depth and limits of detection for small clones (17–19, 21). We hypothesize that this is due to the comparatively young average age of our study subjects, consistent with the low rate of CHIP in healthy adults under age 40 in cross-sectional studies (8–10). This may be related to the stochastic nature of mutation acquisition in hematopoietic stem cells (HSC), such that many adults do not have acquired CH-associated mutations when assessed at a young age. However, emerging evidence suggests that CH mutations can arise at any point in life, even during development in utero (27). These suggest a model in which HSC clones with CH mutations are present in many young adults, but have not yet expanded enough for the mutations to be detectable by current sequencing methods.

Such a model may simply reflect a linear relationship to the time necessary for a sufficient number of cell divisions for a CH clone to reach a detectable size. However, the relative stability of many clones across years of observation, both in our study and others, suggests that other variables could also affect a CH clone's expansion, such as the relative health or attrition of the nonmutant HSC pool.

The lack of clinical impact of CHIP in this study can be explained by several factors. First, because CHIP's effect on clinical outcomes is driven primarily by large clones, the rarity of such clones in this cohort limits our power to detect anything less than the largest effect sizes (8, 10, 11). Second, available evidence suggests that CHIP's impact on nonhematologic outcomes is best conceptualized as an amplifying influence, rather than as a root cause. For young patients in whom the baseline rate of cardiovascular and inflammatory disease is low, it may be unsurprising that CHIP does not measurably increase the incidence of these events. Finally, multiple studies have shown that single DNMT3A and TET2 mutations pose a low risk of evolution to hematologic malignancies (28, 29). The paucity of higher-risk mutations, especially in TP53, is consistent with the low rate of therapy-related AML and MDS reported in previous studies of patients with breast cancer (1).

Most subjects in this cohort with CH had clones too small to be considered CHIP, which in turn had no significant impact on any clinical outcomes evaluated over the study period. However, the large CHIP clones associated with adverse clinical outcomes in other studies presumably began as small clones at some point in the past. Although receipt of cytotoxic therapy did not lead to expansion of CH clones over the 4-year time interval captured in this study, we cannot rule out a model in which this exposure in young adulthood accelerates HSC attrition later in life, leading to rapid expansion of existing CH clones and uncertain impacts on CHIP-associated endpoints. This hypothesis could be addressed by a future study of CH in older populations of long-term cancer survivors.

The strengths of our approach include the focus on a prospectively studied young cohort of patients with robust clinical phenotyping, use of a uniform sequencing platform in the 1-year time point, and availability of serial samples for longitudinal analyses. Although our study of CHIP was the largest to date in young women with breast cancer, we recognize that future studies with even larger sample sizes may be able to identify clinical associations that we were not powered to assess. In addition, although our cohort was relatively homogenous in terms of age, there was a range of treatment approaches. Further efforts to study patients exposed to a defined therapeutic regimen, such as in the context of a randomized clinical trial, may provide insights regarding specific treatment exposures, CHIP, and clinical outcomes.

The routine incorporation of next-generation sequencing into the evaluation of cancer patients will inevitably lead to the incidental discovery of CH in a subset of patients, and some centers have started to establish “CH clinics” that evaluate and monitor these patients for hematologic deterioration and evidence of leukemic transformation (17–19, 30–32). However, there are currently no approved interventions for CH, and its discovery can lead to significant patient anxiety and increased medical costs (33). Taken together, our data show that CHIP is infrequent in young women treated for breast cancer and do not support routine testing for CHIP in this population. Furthermore, in the absence of any hematologic abnormalities, our data show that young patients with breast cancer who are incidentally found to have CHIP via germline or somatic testing can be reassured and may not require long-term monitoring.

T. Sella reports personal fees from Roche and nonfinancial support from Gilead outside the submitted work. A.S. Sperling reports grants from NCI and DOD during the conduct of the study and personal fees from Novartis and Roche outside the submitted work. S.M. Rosenberg reports grants from Pfizer/Conquer Cancer and the ASCO Foundation outside the submitted work. A. Patel reports other support from Natera Inc. outside the submitted work. D.A. Dillon reports grants from Canon outside the submitted work. A.G. Bick reports personal fees from TenSixteen Bio outside the submitted work. D.S. Neuberg reports grants from V Foundation during the conduct of the study. A.H. Partridge reports other support from UpToDate outside the submitted work. P.G. Miller reports personal fees from Foundation Medicine, Inc. and other support from Roche outside the submitted work. No disclosures were reported by the other authors.

C.J. Gibson: Conceptualization, data curation, formal analysis, validation, investigation, writing–original draft, writing–review and editing. G.G. Fell: Data curation, formal analysis, validation, writing–original draft, writing–review and editing. T. Sella: Conceptualization, data curation, formal analysis, validation, investigation. A.S. Sperling: Formal analysis, methodology. C. Snow: Data curation, formal analysis. S.M. Rosenberg: Data curation, formal analysis, methodology. G. Kirkner: Data curation, formal analysis, methodology. A. Patel: Software, writing–original draft, writing–review and editing. D. Dillon: Data curation, formal analysis, validation, methodology. A.G. Bick: Data curation, formal analysis, validation, methodology. D. Neuberg: Conceptualization, formal analysis, writing–original draft, writing–review and editing. A.H. Partridge: Conceptualization, data curation, investigation, methodology. P.G. Miller: Conceptualization, data curation, formal analysis, validation, writing–original draft, writing–review and editing.

This work was supported by grants from the V Foundation, the NIH [K08-CA263181 (P.G. Miller), K08-CA252174 (A.S. Sperling)], the Office of the Director [DP5-OD02958 (A.G. Bick)], the Edward P. Evans Foundation (to P.G. Miller), the Burroughs Wellcome Foundation (to A.G. Bick), and the Pew Charitable Trusts and the Alexander and Margaret Stewart Trust (to A.G. Bick).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.