Clonal hematopoiesis of indeterminate potential influences breast cancer outcomes in a genotype-specific manner

Abstract

Purpose:

Clonal hematopoiesis of indeterminate potential (CHIP) has been associated with adverse outcomes in some solid tumor settings, but its impact on breast cancer remains unclear. We sought to investigate the genotype-specific effects of CHIP on breast cancer outcomes and the tumor microenvironment.

Experimental Design:

We examined a retrospective cohort of 125 patients with breast cancer, using targeted sequencing to identify CHIP. Metastatic events were recorded, and distant metastasis-free survival probability was analyzed. In parallel, we developed chimeric mouse models of the two most mutated CHIP genes, DNMT3A and TET2. CHIP and control mice were orthotopically injected with syngeneic breast cancer cells. Tumor growth was measured, and immune infiltrate was profiled via mass cytometry.

Results:

CHIP was present in 18.4% of patients. High-burden CHIP and non-DNMT3A CHIP were associated with significantly shorter distant metastasis-free survival. In vivo, mice with Tet2-CHIP developed larger primary tumors and were more likely to experience lung metastasis, while Dnmt3a-CHIP did not differ from controls. The general immune subsets observed in both CHIP models were similar, but immunophenotyping revealed clonal expansion and immune cell subset skewing specific to the Tet2-CHIP model.

Conclusions:

Our findings demonstrate a genotype-specific impact of CHIP on breast cancer across human and mouse data. Further, the chimeric mouse models we generated offer a clinically relevant tool to study solid tumors in a CHIP background. This work underscores the need for further functional studies and personalized risk assessment to clearly define the impact of various CHIP genotypes on breast cancer.

Introduction

Clonal hematopoiesis of indeterminate potential (CHIP) is an age-associated (1–3) somatic process in which leukemia-associated mutations in hematopoietic stem cells (HSCs) promote clonal expansion of blood cells (4–6). Though it has long been appreciated that hematopoietic stem cells undergo clonal division (7) and differentiate into mature hematopoietic lineages (8), advances in sequencing technologies have enabled the discovery of somatic mutations in HSCs that provide a proliferation or survival advantage to individual HSCs and thus result in clonal skew. CHIP is defined by the presence of these clones at a variant allele frequency (VAF) of at least 2% in people who otherwise do not meet the criteria for hematologic malignancy (9). DNMT3A and TET2, both epigenetic regulators, are the two most commonly mutated genes in CHIP (10,11). While CHIP is most studied for its risk of transformation to hematologic neoplasm (4,5,12) and its association with cardiovascular disease (13–15), its association with myriad conditions and health outcomes is rapidly expanding.

Recent studies have connected CHIP to solid tumors through numerous approaches (as reviewed in 16,17). Analyses of retrospective clinical cohorts of patients with solid tumors have uncovered evidence that CHIP is associated with worse outcomes in some contexts (18–23). Interrogation of additional clinical cohorts has revealed that the incidence of CHIP is increased among patients with solid tumors (18,24,25), with certain cancer therapies promoting expansion of CHIP clones (19,26,27). For example, studies have found that CHIP mutations provide a selective advantage in the presence of genotoxic stress, such as radiation therapy and chemotherapy, and that the incidence and clonal VAF of CHIP are increased in patients who have received anti-tumor therapies (18,27–30). Other studies have highlighted common risk factors between CHIP and solid tumors (3,10,31–35) and observed that CHIP may lead to an increased risk of some solid tumors (11,36–39). Finally, some groups have started to describe the inflammatory and immune-dysregulating effects of CHIP clones, both in peripheral circulation (40–45) and within the tumor microenvironment (23,46–50).

Despite these advances, much remains unknown about how CHIP impacts clinical outcomes and the tumor microenvironment. Few studies have examined the genotype-specific effects of CHIP mutations on solid tumor outcomes. Similarly, only a small handful of studies have utilized chimeric mouse models to study CHIP in a solid tumor setting, which allows for wildtype and CHIP-mutant cells to interact and infiltrate into the tumor microenvironment. An exciting recent study (51) utilized a chimeric CHIP model in the setting of lung adenocarcinoma, but it assessed only a single CHIP mutation and did not evaluate the effects on tumor growth. To our knowledge, no study has directly compared the effects of different CHIP genotypes on tumor growth and the tumor immune microenvironment.

Here, we present orthogonal mouse and human data demonstrating a genotype-specific effect of CHIP on breast cancer growth and metastasis. Analysis of a real-world cohort of patients presenting with primary breast cancer identified a higher incidence of metastatic events in patients with high-burden CHIP and non-DNMT3A CHIP. To investigate these associations, we then generated and validated chimeric mouse models of DNMT3A-CHIP and TET2-CHIP, which are the two most common CHIP genotypes. Again, we observed a genotype-specific effect, whereby Tet2-CHIP, but not Dnmt3a-CHIP, resulted in larger primary tumors and increased lung metastasis in a mouse breast cancer model. Interrogation of the tumor immune microenvironment via mass cytometry was used to define how immune cell composition and clonal skew differ between CHIP genotypes. Together, this report sheds light on how the most common CHIP mutations differentially influence breast cancer growth and biology.

Materials and Methods

Clinical cohort

This retrospective cohort study was approved by the Institutional Review Board at Vanderbilt University Medical Center (VUMC) (IRB #220474) and utilized participants with blood samples in an institutional biobank (BioVU) (52,53). Under the BioVU framework, patients are consented to provide their de-identified electronic health record (EHR) data to VUMC’s synthetic derivative (SD). Written informed consent was obtained from participants, and all studies were conducted in accordance with recognized ethical guidelines. If blood is available after routine clinical testing at VUMC, samples are used for DNA collection tied to the patient’s SD record.

Patients diagnosed with cancer who received radiation therapy at least 6 months prior to their BioVU blood sample were identified as previously described (medRxiv https://doi.org/10.1101/2024.09.27.24314321). Only patients with a diagnosis of breast cancer, as confirmed by manual review, were included. CHIP was identified by sequencing BioVU DNA samples with an institutional Clonal Hematopoiesis Sequencing Assay (54). Briefly, the assay was designed to detect 95% of common CHIP mutations (10) via a targeted next-generation sequencing assay. After sequence alignment and error correction, Mutect2 (GATK, Broad Institute) was used to identify putative somatic mutations. Mutations were then filtered to require at least 2% VAF, a total read depth ≥ 100, and a variant allele read depth ≥ 3.

All relevant clinical data were manually curated from de-identified patient data in the BioVU SD. Information gathered included age of DNA sample used for CHIP calling, chart-reported race, breast cancer stage and subtype, and date of metastasis (if applicable). Patients with unknown date of diagnosis, staging, or subtype were excluded from further analysis. Patients who presented with metastatic disease or who had a known hematologic malignancy were also excluded.

Distant-metastasis free survival probabilities were estimated using Kaplan-Meier (KM) curves, and group differences were assessed with the log-rank test. Time-to-event was defined from the date of cancer diagnosis to the occurrence of the event or censoring. For multivariable analysis, Cox proportional hazards regression was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs), with adjustment for relevant covariates. P-values were calculated using two-tailed t tests, and values less than 0.05 were considered statistically significant.

CHIP mouse model

All animal procedures followed Institutional Animal Care and Use Committee (IACUC)-approved protocols and were performed in accordance with VUMC and IACUC regulatory guidelines. All parent mouse strains used were purchased from The Jackson Laboratory (Bar Harbor, ME). Crosses of Mx-Cre (Jax Strain 003556, RRID:IMSR_JAX:003556) with Tet2f (Jax Strain 017573, RRID:IMSR_JAX:017573) or Dnmt3a fl-R878H (Jax Strain 032289, RRID:IMSR_JAX:032289) were generated at VUMC. The resulting crosses, which express CD45.2, are referred to as Tet2^fl/fl Mx1-Cre and Dnmt3a^fl-R878H/+ Mx1-Cre. Both crosses had one copy of Mx1-Cre. The floxed littermate controls had no copies of Mx1-Cre. In both experimental and control donor animals, the Tet2f allele was maintained as homozygous Tet2^fl/fl while the Dnmt3a fl-R878H allele was maintained as heterozygous Dnmt3a^fl-R878H/+.

Mutant genotype mice and floxed littermates were aged to 8 weeks and then intraperitoneally injected with 15 mg/kg poly (I:C) (Millipore Sigma Cat# P0913) every other day for 5 total injections to induce knockout of Tet2^fl/fl or knock in of Dnmt3a^fl-R878H/+ via Cre recombination. Control floxed littermates were also treated with poly (I:C). Donor marrow was harvested from mutant and floxed littermates and 1 million marrow cells were transplanted into CD45.1 JAXBoy (Jax strain 033076, RRID:IMSR_JAX:033076) recipients. The control group of recipient mice received an injection of 50% CD45.2 floxed (wildtype) cells and 50% CD45.1 JaxBoy support (wildtype) cells. The experimental group received an injection of 50% CD45.2 mutant cells (knockout Tet2^−/− or knock in Dnmt3a^R878H/+) and 50% CD45.1 JaxBoy support (wildtype) cells. These resulting experimental mice were referred to Tet2-CHIP and Dnmt3A-CHIP, respectively.

JAXBoy recipients were given two 550 rad doses of radiation: one 18 to 24 hours and another 2 to 4 hours prior to transplant injection. Transplant injections were done via tail vein. Recipient mice were allowed to recover, and engraftment was validated up to 9 weeks post-transplant with flow cytometry. Blood was collected from the saphenous vein, treated with ACK lysis buffer (Gibco Cat# A1049201), stained with 1:1000 dilution of Zombie NIR (Biolegend Cat# 423106), Fc blocked (BioLegend Cat# 101339, RRID:AB_2616683), and stained with fluorescent antibodies. CD45.1 PE/Dazzle 594 (BioLegend Cat# 110748, RRID:AB_2564295) and CD45.2 BV711 (BioLegend Cat# 109847, RRID:AB_2616859) antibodies were both used at 1:200 working dilutions. Engraftment samples were analyzed on an Attune NxT acoustic flow cytometer (Life Technologies, Carlsbad, CA, RRID:SCR_019590). Differences in engraftment, as measured by peripheral blood CD45.2+ chimerism, were calculated via t-test with alpha = 0.05.

Primary breast tumor mouse models

Twelve weeks after bone marrow transplant, mice were injected with the syngeneic breast cancer cell line E0771 (ATCC Cat# CRL-3461, RRID:CVCL_GR23; authenticated and tested for mycoplasma by the John Hopkins Genetic Resources Core Facility or ATCC before use and always injected at passage 4–8). The day prior to injection, depilatory cream was applied to the lower abdomen of each mouse to allow for visualization of the nipple. The day of injection, mice were anesthetized. 5 × 10⁴ cells in 100 μL PBS were injected into the 4^th mammary fat pad (either right side or bilateral) with a 27G needle.

Tumor endpoint was defined as a tumor volume of 1000 mm³ or tumor ulceration. Tumor volumes were calculated with the formula ${\mathrm{L}}^2\times \mathrm{W}\times 0.5$. Mice were closely monitored throughout for health concerns and weight loss. At endpoint, differences in tumor volume and tumor weight between CHIP and littermate control mice were statistically evaluated via t-test with alpha = 0.05. Experimenters who measured tumors were blind to mouse genotype (CHIP versus control) until after all data were collected.

Tissue harvesting, tumor dissociation and immune cell enrichment

When the first mouse in a cohort achieved tumor endpoint, all mice were euthanized for tissue harvesting. Immediately after euthanasia, peripheral blood was collected via cardiac puncture into EDTA (Corning Cat# 46–034-CI) coated tubes. Tumors were carefully excised from the mammary fat pad, weighed, and then placed on ice in 2% FBS for downstream processing. Lungs, spleen, and femurs were also collected and held in 2% FBS on ice for further processing as described below. For tissues to be fixed via FFPE, tissue was placed into fresh 10% neutral buffered formalin for 24 hours at room temperature. Further processing and paraffin embedding was performed by the Vanderbilt Translational Pathology Shared Resource. For fresh frozen specimens, tissue was snap frozen and stored in liquid nitrogen until processing.

Blood processing:

50–100 μL of whole blood was processed for CyTOF. Red blood cell lysis was performed by adding 1 mL ACK lysis buffer and incubating for 5 minutes at room temperature. The reaction was quenched by adding 5 mL cold PBS and the white blood cells were pelleted by centrifugation at 300 × g for 5 minutes. If needed, a second round of ACK lysis was performed to remove any remaining red cells.

Cell isolation for downstream analysis:

Tumors were manually minced on ice and then transferred to gentleMACS C tubes (Miltenyi Cat# 130–093-237, RRID:SCR_020270) containing 217.5 U ml⁻¹ DNase I (Sigma-Aldrich Cat# D5025) and 109 U ml⁻¹ collagenase (Sigma-Aldrich Cat# C2674). Tubes were placed on the Miltenyi gentleMACS Octo Dissociator with Heaters (RRID:SCR_020271) for mechanical dissociation with mouse implant tumor 1 setting followed by enzymatic incubation with the 37C_m_TDK2 program. Tumor dissociates were then passed through a 70 μm filter. Spleens were mechanically disrupted with the back of a 1mL syringe and strained through a 70 μm filter. Bone marrow was flushed from mouse femurs and run through a 70 μm filter. Cell suspensions of all three tissues were then ACK lysed for red blood cell removal and stored in 2% FBS at 4 °C for downstream processing.

Immune cell enrichment:

Single cell suspensions of dissociated tumors were mixed by gentle inversion with 90% Percoll (Cytiva Cat# 17089102) and 500 U heparin (Fresenius Kabi NDC 63323–540-05) and then centrifuged at 500 × g for 10 minutes at 4 °C. The remaining pellet was resuspended in 2% FBS.

Cell counting:

Cells were counted using the trypan blue-based Vi-Cell BLU Cell Viability Analyzer (Beckman Coulter, RRID:SCR_026900), using settings optimized for immune cell detection.

Histology

Immunohistochemistry (IHC) on frozen tissue:

Frozen tumor sections were cut at 5 μm. Endogenous peroxidases were inhibited with 3% hydrogen peroxide (Fisher Scientific Cat# BP2633), and protein block solution (Agilent Cat# X0909) was applied. Avidin block (Biocare Cat# AB972H-A), Biotin block (Biocare Cat# AB972H-B) and background punisher (Biocare Cat# BP974H) were applied for CD45.1 and CD45.2. Sections were then incubated with the primary antibody (Supplemental Table S1). Visualization was performed with the corresponding system, DAB (Agilent, K3468) as the chromogen and hematoxylin (Fisher Scientific, 220–100) as the counterstain.

Immunohistochemistry (IHC) on fixed tissue:

FFPE tumor tissue sections were cut at 4 μm and deparaffinized. Antigen retrieval was performed with citrate buffer pH 6 (Agilent, S2369). Endogenous peroxidases were inhibited with 3% hydrogen peroxide (Fisher Scientific, BP2633), and protein block solution (Agilent, X0909) was applied. Avidin block (Biocare, AB972H-A), Biotin block (Biocare, AB972H-B) and background punisher (Biocare, BP974H) were applied for CD4 and CD8. Sections were then incubated with the primary antibody (Supplemental Table S1). Visualization was performed with the corresponding system, DAB (Agilent Cat# K3468) as the chromogen and hematoxylin (Fisher Scientific Cat# 220–100) as the counterstain.

Optimization of assay conditions was performed on mouse tonsil, spleen and small bowel, which were subsequently used as controls for each batch. Slides were evaluated by a pathologist for the presence and localization of immune cells.

Evaluation of lung micrometastases:

FFPE lung tissue sections harvested from mice at primary tumor endpoint were stained with hematoxylin and eosin (H&E). Multiple levels were evaluated for the presence of metastasis by a pathologist. Relative risk for developing lung metastasis in CHIP mice compared to controls was calculated by MedCalc (55).

Mass Cytometry

CyTOF data acquisition:

Matched ACK-lysed blood and immune cell-enriched tumor dissociate were prepared from mice at tumor endpoint (see above). Each sample was stained with both a myeloid-focused panel (Supplemental Table S2) and a T cell-focused panel (Supplemental Table S3). Up to 500,000 cells per sample were pelleted in PBS at 300 × g for 5 minutes, resuspended, stained with 5 mM Cell ID Cisplatin (Standard Biotools Cat# 201064) for 2.5 minutes at room temperature (RT), and then quenched with cell staining media (CSM). Cells were spun again at 300 × g for 5 minutes cells and then fixed with 1.6% paraformaldehyde (PFA) in the dark at RT for 10 minutes, washed in PBS, and spun at 800 × g for 5 minutes. Fixed cells were resuspended in permeabilization buffer twice and stained with Cell ID palladium barcoding kit (Standard Biotools Cat# 201060). Samples were combined per the manufacturer’s instructions prior to antibody staining. Cells were spun once more at 800 × g for 5 minutes in CSM and Fc blocked for 15 minutes with a 1:150 dilution of unconjugated CD16/32 antibody (Biolegend Cat# 101339, AB_2616683), or with a metal-conjugated CD16/32 antibody when included as part of the staining panel (Supplemental Table S2). The staining panel was added, and the combined cells were stained for 30 minutes. Cells were quenched with CSM, spun down, and resuspended in methanol to be stored at −80 °C for a minimum of 30 minutes until ready to acquire on a Helios CyTOF mass cytometer (Standard Biotools, South San Francisco, CA, RRID:SCR_019916).

Prior to acquisition, cells in methanol were washed with a 2X volume of PBS, spun at 800 × g for 5 minutes, and resuspended in CSM. For intracellular staining, cells were spun again, resuspended, and stained for 30 minutes. Following intracellular staining (or washing out of methanol if no intracellular stain), cells were resuspended in 1.6% PFA and a 1:50 dilution of Cell ID Intercalator (Standard Biotools Cat# 201192A) for 30 minutes. Cells were washed in PBS and resuspended in a solution of 9:1 MaxPar Cell Acquisition Solution (Standard Biotools Cat# 201240) to EQ Four Element Calibration beads (Standard Biotools Cat# 201078). Cells were resuspended at ~1000 cells per μl to run at 500 events per second on the Helios through the wide bore injector. Mass cytometry antibodies, including staining volumes, are found in Supplemental Tables S2 and S3.

CyTOF data analysis:

Briefly, collected CyTOF files were gated for live, intact cells using standard procedures. Live cells were gated and mass cytometry FCS files were debarcoded and compensated using CATALYST (56). Dual ion count signal intensity for each channel was arcsinh-transformed with a cofactor of 5. Spillover was corrected using a spillover matrix to estimate the degree of isotopic spillover between channels, and then non-negative least-squares (NNLS) was used to apply the compensation to remove contaminating signal per standard (56). Principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) were used for dimension reduction, followed by Leiden clustering, resulting in the expected number of clusters (57). Expert annotation was used to annotate cell types by virtue of presence and absence of canonical cell surface protein markers. All analysis was performed in R 4.4.1.

CyTOF comparisons and statistical analysis:

Cells were first assigned to one of three major immune subsets: myeloid (CD3−; thus includes B cells and NK cells), CD4+ T cells (CD3+, CD4+, CD8−), and CD8+ T cells (CD3+, CD4−, CD8+). Within each major immune subset, cells were further subclustered and expertly annotated. CD45 isoform identity was considered only after all subsets and clusters were defined. Total CD45.1+ cells and CD45.2+ cells were quantified per cluster. For general immune infiltrate comparisons, immune subset composition relative to all CD45+ cells was calculated. For comparisons of chimerism, the percentage of CD45.2+ cells out of a given subset was calculated. Statistical comparisons across multiple groups were performed using two-way ANOVA, followed by Bonferroni’s post hoc test for pairwise comparisons. For the determination of chimeric skew between immune subsets, the percent CD45.2+ of each subset was compared to the percent CD45.2+ in all immune cells using ordinary one-way ANOVA followed by Dunnett’s post hoc test for pairwise comparisons.

Statistical Analysis

GraphPad Prism (v10.3.1) was used for plotting and statistical analyses. Tests were considered statistically significant if the p-value was less than 0.05. Additional relevant detail is outlined in the figure legends and respective methods sections.

Data Availability

Data for the BioVU cohort are available in Supplemental Table S4 in the form of variant-level sequencing calls. In accordance with BioVU policies and patient consent forms, to protect patient privacy and ensure compliance, raw data cannot be deposited in a public database. Upon reasonable request, qualified researchers may request access to de-identified data necessary to verify the conclusions of this study, specifically to analyze and reproduce the variant calls. Access will be provided within a secure, controlled environment after the execution of appropriate legal agreements, and will be limited to the sole purpose of reproducing the published results. All other data generated in this study are available upon request from the corresponding author

Results

CHIP burden and genotype stratify metastatic risk in primary breast cancer

BioVU (52), an electronic health record-based biorepository at Vanderbilt University Medical Center, was leveraged to investigate the clinical impact of CHIP on breast cancer. To enrich for the presence of CHIP, we evaluated patients with breast cancer who received radiation therapy (medRxiv https://doi.org/10.1101/2024.09.27.24314321). Whole blood drawn at least six months after radiation therapy was processed, and CHIP calling was performed via a validated targeted sequencing panel that identifies >95% of observed CHIP mutations (54). Manual chart review was used to identify clinical characteristics and metastatic outcomes, resulting in a final cohort of 125 patients (Table 1).

Table 1.

Inclusion and exclusion criteria for the retrospective cohort study using BioVU blood samples and matched deidentified health records (see methods for details), with total numbers of patients as indicated.

Criteria	Number of Patients
Included
Patients with breast cancer BioVU sample available History of radiation treatment BioVU sample dated ≥ 6 months after radiation	144
Excluded
Unknown date of diagnosis	1
Unknown staging	7
Unknown subtype	12 (5 with unknown staging and unknown subtype)
Presented with metastatic disease	4
Final cohort	125

CHIP was identified in 18.4% of samples. VAFs ranged from 2.2% to 44.4%, with a median of 4.5% (Supplemental Table S5). A total of 23 unique CHIP variants were detected, with each patient having only a single CHIP clone. Not surprisingly, patients with CHIP were older than those without CHIP (mean age at blood draw 64.2 versus 57.3 years, respectively). Otherwise, the groups did not differ by chart-reported race, smoking status, disease subtype, or exposure to chemotherapy (Table 2). Irrespective of CHIP status, distant metastasis-free survival (DMFS) varied as expected by stage (Supplemental Figure S1A) and by subtype (Supplemental Figure S1B), with stage 3 disease and triple negative breast cancer showing the highest incidence of metastasis.

Table 2.

Clinicopathologic characteristics of patients with and without CHIP. Significance of association for each variable was calculated with Fisher’s Exact test except for age, which was calculated with Mann-Whitney U test. Race was chart-reported.

	No CHIP (n = 102)	CHIP (n = 23)	p-value
Age			0.0313*
Mean (SD)	57.3 (12.0)	64.2 (11.8)
Median [Min, Max]	57.6 [29.3, 90.8]	60.6 [48.6, 86.4]
Race			>0.9999
Caucasian	80 (78.4%)	18 (78.3%)
Non-Caucasian	22 (21.6%)	5 (21.7%)
Smoking			0.8143
Never	65 (63.5%)	14 (60.9%)
Former or Active	37 (36.3%)	9 (39.1%)
Stage			0.0035**
0	3 (2.9%)	6 (26.1%)
1	43 (42.2%)	9 (39.1%)
2	38 (37.3%)	4 (17.4%)
3	18 (17.6%)	4 (17.4%)
Subtype			0.7943
HR+	66 (66.7%)	11 (64.7%)
HER2+	16 (16.2%)	2 (11.8%)
TNBC	17 (17.2%)	4 (23.5%)
Chemotherapy			0.9465
None	68 (66.7%)	15 (65.2%)
1 line	10 (9.8%)	2 (8.7%)
2 lines	23 (22.5%)	6 (26.1%)
3 lines	1 (1.0%)	0 (0.0%)

Stats: Mann-Whitney U (age); Fisher’s Exact (all others)

Race is chart reported

DMFS did not significantly differ between CHIP and non-CHIP groups (Figure 1A). Given findings in other disease settings that demonstrate poor outcomes generally increase as CHIP VAF increases (5,18,58), we further grouped patients with CHIP into low-burden CHIP (VAF ≤ 10%) and high-burden CHIP (VAF > 10%). Examination of metastatic outcomes after accounting for CHIP burden demonstrates a striking difference, where patients with high-burden CHIP had significantly shorter DMFS compared to no-CHIP counterparts while patients with low-burden CHIP experienced no metastatic events (Figure 1B). To understand the impact of CHIP genotype on metastatic outcomes, we additionally grouped patients with CHIP into those with DNMT3A mutations (n = 12; median VAF 4%) and those with non-DNMT3A mutations (n = 11; median VAF 6.4%) (Figure 1C). We observed a marked difference in metastatic outcomes and a significantly shorter DMFS in patients with non-DNMT3A CHIP (Figure 1D). On multivariable analysis, CHIP burden > 10% VAF was independently associated with increased odds of metastasis (Figure 1E). DNMT3A-CHIP status was not included in analysis because none of the patients with DNMT3A-CHIP developed metastasis. These data suggest that in addition to VAF burden, CHIP genotype is an important variable to consider when understanding the effects of CHIP on solid tumor outcomes.

Figure 1.

DMFS in the patient cohort. Kaplan-Meier curves plotting the probability of DMFS with patients stratified by A, no CHIP versus any CHIP and by B, no CHIP, low-burden CHIP (VAF 2–10%), and high-burden CHIP (VAF >10%). C, Among those with any CHIP, patients were grouped by genotype. D, Kaplan-Meier curve plotting the probability of DMFS among patients with CHIP by DNMT3A status. E, Multivariable analysis of DMFS in patient cohort. Subgroups that had no metastatic event were excluded.

Development of chimeric mouse models enables study of CHIP in breast cancer

To better understand how CHIP influences solid tumor growth and alters the tumor immune microenvironment, we generated mouse models of the two most commonly mutated CHIP genes, Dnmt3a and Tet2 (Figure 2A). Briefly, sub-lethally irradiated female mice were engrafted with a mixture of bone marrow harvested from wildtype and either Dnmt3a-mutant (Dnmt3a^R878H/+) or Tet2-mutant (Tet2^−/−) donor mice. Transplants were allowed to engraft for 12 weeks to allow for full hematopoietic lineage reconstitution (59) and then injected with the syngeneic triple-negative breast cancer line E0771 into the mammary fat pad. At endpoint, tumors were harvested for histology and immunophenotyping. Peripheral blood, spleen, and lungs were also collected. Crucially, this model results in a biologically relevant chimeric hematopoietic system containing both wildtype cells and CHIP clones. Each population is congenically marked, with CHIP clones expressing CD45.2 and wildtype cells expressing CD45.1 (Figure 2B). Control mice were littermates lacking Cre recombinase and thus also expressed CD45.2 but were 100% wild type.

Figure 2.

Generation of Dnmt3a-CHIP and Tet2-CHIP mouse models enables characterization of breast tumor growth in vivo. A, Schematic of chimeric models of CHIP that were injected with orthotopic breast tumors (Created in BioRender. Reed, S. (2025) https://BioRender.com/t77gyyt). B, Representative plots showing CD45.1+ and CD45.2+ cells in mice following bone marrow transplant, along with C, chimeric burden in peripheral blood of all CHIP and control mice at 1–2 months post-transplant (Dnmt3A model: n = 9 control, 7 CHIP; Tet2 model: n = 19 control, 20 CHIP). Differences in chimeric burden evaluated via t test (Dnmt3a-CHIP vs control not significant; Tet2-CHIP vs control p = 0.0053). D, Chimeric burden of mice in peripheral tissues at tumor endpoint (n ≥3 per group). E, Breast tumor volume (mm³) at endpoint in Dnmt3a-CHIP and Tet2-CHIP mice, with differences compared to littermate controls evaluated via t test (Dnmt3A model: n = 14 control, 8 CHIP; Tet2 model: n = 23 control, 27 CHIP). F, Number of mice experiencing lung metastasis as evaluated by pathologist review. RR = relative risk. G, H&Es of a lung micrometastasis and matched primary breast tumor.

In these models, CD45.2+ chimerism can be used as a proxy for the level of CHIP present in each tissue. Chimerism at one to two months post-transplant was assessed by CD45.2+ frequency in peripheral blood (Figure 2C). Prior to tumor injection, the Dnmt3a-CHIP model produced similar chimeric burdens to controls (median = 33.0% versus 23.8%, respectively). In contrast, the Tet2-CHIP model resulted in significantly higher chimeric burdens compared to controls (median = 70.3% versus 50.1%, respectively). This aligns with observations from other groups that models of TET2-CHIP tend to show greater clonal expansion than DNMT3A-CHIP (60).

Following tumor endpoint, peripheral tissues were again assessed for chimeric burden (Figure 2D). Both models exhibited stable chimerisms between peripheral blood and spleen, confirming that wildtype and CHIP cells distribute evenly in circulation and that we can reliably measure chimeric burden. Though similar methods are more widely used in models of hematopoietic disease, they have infrequently been utilized in investigations of solid tumors. Collectively, these data demonstrate that we can successfully model CHIP as a chimeric mixture of mutant clones and wildtype cells in mice.

Tet2-CHIP, but not Dnmt3a-CHIP, alters primary tumor growth and metastasis in vivo

After establishing that a chimeric bone marrow transplantation model can be applied to a breast cancer model, we then measured primary breast tumor growth among mice with either Dnmt3a-CHIP or Tet2-CHIP. Orthotopic primary tumors were grown for four weeks, and tumor volume was measured at endpoint (Figure 2E). Tumors in mice with Dnmt3a-CHIP did not significantly differ in size from controls. In contrast, tumors in mice with Tet2-CHIP were significantly larger than controls. The same pattern was observed when measuring tumor weight at endpoint (Supplemental Figure S2).

We next examined the lungs harvested from mice at tumor endpoint. A breast pathologist evaluated multiple levels stained with hematoxylin and eosin (H&E) per lung for the presence of micrometastases (Figure 2F). No mice with Dnmt3a-CHIP experienced lung metastasis, while 22.2% of their control counterparts had detectable lung metastasis (relative risk = 0.25). The opposite trend was observed among mice with Tet2-CHIP, with metastasis identified in 14.3% of control mice and 37.5% of Tet2-CHIP mice (relative risk = 2.625). The identified lung micrometastases were compared with their matched primary tumors to confirm that each represented a metastatic event rather than a de-novo or radiation-induced tumor (Figure 2G). These data reflect the pattern observed in our clinical cohort, in which there were no metastatic events among patients with DNMT3A-CHIP and a significantly shorter DMFS among patients with non-DNMT3A-CHIP (including TET2-CHIP).

Taken together, these mouse and human data present a striking picture of genotype-specific effects of CHIP on solid tumor outcomes. While TET2-CHIP appears to promote increased primary breast cancer growth and increased metastasis, DNMT3A-CHIP does not have an appreciable effect on breast tumor size or metastatic burden.

CHIP genotypes exhibit similar immune composition but distinct clonal dynamics in the breast tumor microenvironment

Breast tumors from CHIP and control mice were further investigated to characterize their immune infiltrate. We first confirmed that both wildtype cells (CD45.1+) and CHIP clones (CD45.2+) infiltrate into the tumor via immunohistochemistry (IHC) on frozen tumor tissue (Figure 3A). We then performed additional hematoxylin and eosin (H&E) staining and IHC staining on formalin-fixed paraffin-embedded (FFPE) tumor tissue. Pathologist review of the tumor tissues confirmed that the histological architecture of the tumor was consistent across Tet2-CHIP, Dnmt3A-CHIP, and control mice. IHC staining revealed both myeloid and lymphoid infiltrate, with strong macrophage (F4/80+) infiltrate throughout the tumor and more prominent T cell infiltrate (CD4+ or CD8+) at the tumor periphery (Figure 3B, Supplemental Figures S3 and S4). The observed immune cell infiltrate and tissue histology demonstrate that this model faithfully recapitulates key features of breast tumor biology. Importantly, the presence of both wildtype and CHIP cells in tumors with robust macrophage and lymphocyte infiltrate provides further evidence that our chimeric mouse models enable exploration of CHIP in the context of solid tumors and the tumor immune microenvironment.

Figure 3.

Evaluation of breast tumor immune infiltrate in CHIP models via IHC and CyTOF. A, IHC on frozen breast tumors showing CD45.1+ and CD45.2+ infiltrate in the Tet2-CHIP model and matched wildtype control. B, H&E staining and IHC for CD4, CD8, and F4/80 on FFPE breast tumors demonstrating expected breast tumor histology, with both myeloid and lymphoid infiltrate present. Tumor periphery of Tet2-CHIP and control mice shown (Dnmt3a-CHIP mice and tumor centers can be found in supplemental figures). C, Uniform Manifold Approximation and Projection (UMAP) plot depicting the immune cells profiled by CyTOF. Each point represents a single cell, and Leiden clustering was applied after dimension reduction to group cells into broad immune subsets based on similarity in their surface marker expression profiles (Tet2 model: n = 5 control mice, 5 CHIP mice; Dnmt3a model: n = 6 control mice, 3 CHIP mice). D, Composition of peripheral blood (PB) and tumor immune infiltrate (T) by general immune subsets (Tet2 model: n = 5 per group; Dnmt3a model: n = 6 per control group, 3 per CHIP group). Differences in composition for each immune subset across all four tissues (control PB, control T, CHIP PB, CHIP T) were evaluated by two-way ANOVA with Bonferroni’s multiple comparisons test. All comparisons were not significant. E, CD45.2+ burden in all immune cells and in each general immune subset (Tet2 model: n = 5 per group; Dnmt3a model: n = 6 per control group, 3 per CHIP group). Comparisons across all four tissues (control PB, control T, CHIP PB, CHIP T) were evaluated by two-way ANOVA with Bonferroni’s multiple comparisons test (comparisons were not significant except where noted). F, CD45.2+ burden in each general immune subset as compared to the CD45.2+ burden in all immune cells, plotted by genotype and tissue (Tet2 model: n = 5 per group; Dnmt3a model: n = 6 per control group, 3 per CHIP group). Differences in chimeric burden were evaluated by one-way ANOVA with Dunnett’s multiple comparisons test (comparisons were not significant except where noted). * = p < 0.05, ** = p < 0.01, **** = p < 0.0001, ns = not significant

Tumors were then harvested and dissociated into single cells. After enriching the single cell suspension for immune cells, we performed cytometry by time-of-flight (CyTOF) to profile the tumor-infiltrating immune cells. In parallel, CyTOF was performed on matched peripheral blood. Dimensional reduction and clustering were performed. Visualization with Uniform Manifold Approximation and Projection (UMAP) revealed myeloid and lymphocyte tumor infiltrate from both wildtype cells and CHIP clones (Figure 3C). Cells were then further analyzed to identify subclusters of immune cells present in blood and tumor.

To define the immune cell makeup in tumor infiltrate and peripheral blood, we considered all immune cells agnostic of CD45 isoform (Supplemental Figure S5). Cells were classified as belonging to one of three major immune subsets: myeloid cells (includes B cells and NK cells; see methods), CD4+ T cells, and CD8+ T cells. Among both Dnmt3a-CHIP and Tet2-CHIP, the general immune composition of CHIP mice mirrored that of control mice (Figure 3D). This pattern of similar immune composition was true for both peripheral blood and tumor infiltrate. Importantly, the immune makeup was also similar across both CHIP genotypes. Next, we labeled and compared the subclusters comprising each immune subtype (Supplemental Figures S6 and S7). As expected, we observed significant differences in subcluster abundance between peripheral blood and tumor infiltrate that were consistent between Tet2-CHIP and Dnmt3a-CHIP (Supplemental Figures S8A and S8B). In both CHIP models as well as their controls, there were significantly more monocytes, activated monocytes, PD1+ CD4+ T cells, and PD1+ CD8+ T cells in tumor infiltrate compared to blood. Conversely, there were significantly more B cells, naïve CD4+ T cells, and naïve CD8+ T cells in peripheral blood when compared to tumor infiltrate.

We subsequently re-analyzed the major immune cells present in tumor and peripheral blood by calculating chimerism, as measured by CD45.2+ percentage, within each immune subset (Supplemental Figure S9). We observed a significant expansion of CHIP clones in both peripheral blood and tumor infiltrate in mice with Tet2-CHIP compared to controls (Figure 3E). This finding was observed when comparing Tet2-CHIP chimerism in all immune cells and in each subset individually. In contrast, there was no significant difference in Dnmt3a-CHIP chimerism in each tissue compared to controls. The same patterns were observed in most immune subclusters (Supplemental Figure S10).

Finally, we evaluated the major immune subsets within each individual tissue to compare the chimerism burden of each subset to the overall chimerism burden in that tissue (Figure 3F). This comparison allowed us to detect skews in chimeric burden between major immune subsets and evaluate whether individual subsets were over- or underrepresented in their CHIP clone composition. Consideration of chimeric skew revealed a marked difference between CHIP genotypes. Mice with Tet2-CHIP, but not control mice, displayed a significant skew in most major immune subsets in both peripheral blood and tumor infiltrate. In this setting, Tet2-CHIP clones were overrepresented in myeloid cells and underrepresented in CD8+ T cells compared to the overall chimeric burden. We did not observe any significant skewing among any of the immune subsets in mice with Dnmt3a-CHIP.

In summary, the overall composition of immune cells when considering both CD45 isoforms together did not differ significantly between either CHIP genotype and control mice. However, the amount of CHIP-mutant versus wildtype cells making up each subset did differ significantly, again conferring a genotype-specific difference. Mice with Tet2-CHIP had higher chimeric burdens in their tissues compared to controls and displayed significant skews in chimeric burden between immune subsets. Mice with Dnmt3a-CHIP did not exhibit differences in chimeric burden or chimeric skew.

Discussion

In this study, we present evidence from human data and experimental models that the effects of CHIP on breast cancer are both genotype-specific and clinically relevant. Using a retrospective cohort of patients with primary breast cancer along with novel chimeric mouse models of CHIP, we show that while DNMT3A-CHIP is not associated with increased tumor growth or metastasis compared to no-CHIP controls, TET2-CHIP is associated with enhanced primary tumor growth and increased metastatic burden. We further demonstrate that TET2-CHIP is associated with distinct patterns of immune cell infiltration and clonal skew within the tumor microenvironment. These results extend the current understanding of the impact of CHIP on solid tumors, supporting a mutation-specific influence on tumor biology and clinical outcomes.

This is the first report to demonstrate a clear difference in breast cancer outcomes and microenvironment between the two most common CHIP genotypes. In line with these findings, previous groups have used retrospective cohort studies to identify genotype-specific differences in outcomes in other disease settings (12,13,39,58,61–64). A key strength of our approach is the development of chimeric mouse models of DNMT3A-CHIP and TET2-CHIP, which enabled a side-by-side comparison of the two most common CHIP mutations in a biologically relevant environment containing both wildtype and mutant hematopoietic cells. This method expands on previous studies of CHIP in solid tumors that have used models in which 100% of hematopoietic cells (or the lineage of interest) carry the CHIP mutation, thus allowing for consideration of how wildtype and mutant cells interact and how CHIP burden skews across tissues.

Our study has a few limitations. First, the clinical cohort is relatively modest in size, particularly for genotype-stratified analysis. Validation in independent cohorts, including those not pre-treated with radiation therapy, is needed and will be essential to determine whether these findings are generalizable across other settings. Larger clinical datasets will also allow direct comparison of multiple specific CHIP genotypes and a powered analysis of how the variables of CHIP VAF, CHIP genotype, and tumor stage interact. Second, we introduced a relatively high burden of CHIP into our mouse models because our clinical data demonstrated high VAF is more associated with worse tumor outcomes. While the chimeric burden used in our models can occasionally be seen in patient samples, it will be important to repeat similar experiments with a lower CHIP burden to determine the range of VAFs at which the described effects are observed. Nevertheless, we anticipate these findings will be generalizable to patients with lower VAFs, particularly those with a VAF greater than 10%. In addition to our patient cohort that demonstrated shorter distant metastasis-free survival among patients with lower-than-modeled CHIP burdens, other groups have similarly observed measurable effects of CHIP at real-world VAFs (11,18,65,66). It is possible that the effects of low-burden CHIP (VAF ≤ 10%) are more subtle than can be measured in our models or with retrospective clinical cohorts. It will still be important to identify and study these low-burden cases, because CHIP burden often increases over time and states of cellular stress may heighten the effects of mutant clones.

Finally, our interrogation of the tumor immune microenvironment was primarily descriptive. In combination with already-published work, these findings suggest that functional consequences of CHIP mutations on immune cells, such as cytokine production, phagocytosis, antigen presentation, and exhaustion phenotypes, may vary by genotype. While such functional consequences and the means by which they alter primary tumor growth and metastasis require further investigation, work by other groups points to a few possible mechanisms by which TET2-CHIP may promote tumor progression Studies have shown that HSCs with TET2 loss can better expand and resist apoptosis in conditions of inflammatory stress (67,68). Additionally, multiple studies have demonstrated alterations in cytokine or cell signaling pathways among TET2 knockout cells, including increased activation of the NLRP3 inflammasome, higher production of IL-1β and IL-6, and increased NF-κB signaling (42,43,69,70). Both activation of the NLRP3 inflammasome and NF-κB signaling have been linked to breast cancer progression, tumor growth, and metastasis (71,72). Other groups have documented opposing effects of TET2 and DNMT3A signaling on myeloid derived suppressor cells (MDSCs), with TET2 loss resulting in expansion of MDSCs and a reduction of anti-tumor T cells in a murine breast cancer model (73,74). Researchers have also observed expansion of myeloid populations, enhanced monocyte migration into tumors, M2-polarization of tumor-associated macrophages, and increased numbers of tumor-infiltrating NK cells, neutrophils, and regulatory T cells in some settings of TET2 loss (47,51,75). It has long been appreciated that myelopoiesis often correlates with disease stage and can create a pro-tumor immunosuppressive microenvironment, and loss of TET2 may exacerbate this cycle of aberrant immune response (76,77). Lastly, recent analysis of bulk RNA sequencing on tumors with CHIP infiltrate revealed differential expression of multiple gene sets implicated in tumor immunity, including interferon response and antigen presentation (23). These findings and others representing exciting avenues of future study to more comprehensively understand how TET2-CHIP influences breast cancer growth and metastasis.

Despite these limitations, this study improves our understanding of CHIP in solid tumors by using mouse and human data to demonstrate a striking deleterious effect of TET2-CHIP but not DNMT3A-CHIP in breast cancer. It further highlights the need to consider both CHIP burden and CHIP genotype when investigating the effect of CHIP on solid tumors and other disease states. Our chimeric mouse models provide a system for studying CHIP in a more relevant hematopoietic context. Future work can leverage these models to study mutation-specific effects in other tumor types, to enable mechanistic and functional studies into the CHIP-tumor microenvironment, and to assess the effects of treatments such as immunotherapy on solid tumor outcomes in the setting of CHIP. Ultimately, this and future work will be critical to inform our understanding of risk assessment and personalized therapy in patients with solid tumors who have CHIP.

Translational relevance.

Clonal hematopoiesis of indeterminate potential (CHIP) is increasingly recognized as a condition that associates with and augments many disease states. Reports to date on the impact of CHIP on solid tumors have demonstrated variable effects and primarily utilized retrospective cohorts that require additional investigation. In this study, we probe the effects of different CHIP genotypes on breast cancer. We further introduce biologically relevant mouse models that can be used to interrogate the solid tumor microenvironment in a CHIP context. This work reinforces the need to take a gene-specific approach in evaluating CHIP while laying the groundwork for better understanding the mechanisms by which CHIP differentially alters primary tumor growth and metastasis. Defining more precise relationships between different CHIP genotypes and breast cancer biology will enable better personalized decision making for the large number of oncology patients who present with or develop CHIP during their disease course.