Associations Between Cancer Predisposition Mutations and Clonal Hematopoiesis in Patients With Solid Tumors

Abstract

PURPOSE

Clonal hematopoiesis (CH), the expansion of clones in the hematopoietic system, has been linked to different internal and external features such as aging, genetic ancestry, smoking, and oncologic treatment. However, the interplay between mutations in known cancer predisposition genes and CH has not been thoroughly examined in patients with solid tumors.

METHODS

We used prospective tumor-blood paired sequencing data from 46,906 patients who underwent Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) testing to interrogate the associations between CH and rare pathogenic or likely pathogenic (P/LP) germline variants.

RESULTS

We observed an enrichment of CH-positive patients among those carrying P/LP germline mutations and identified a significant association between P/LP germline variants in ATM and CH. Germline and CH comutation patterns in ATM, TP53, and CHEK2 suggested biallelic inactivation as a potential mediator of clonal expansion. Moreover, we observed that CH-PPM1D mutations, similar to somatic tumor-associated PPM1D mutations, were depleted in patients with P/LP germline mutations in the DNA damage response (DDR) genes ATM, CHEK2, and TP53. Patients with solid tumors and harboring P/LP germline mutations, CH mutations, and mosaicism chromosomal alterations might be at an increased risk of developing secondary leukemia while germline variants in TP53 were identified as an independent risk factor (hazard ratio, 36; P < .001) for secondary leukemias.

CONCLUSION

Our results suggest a close relationship between inherited variants and CH mutations within the DDR genes in patients with solid tumors. Associations identified in this study might translate into enhanced clinical surveillance for CH and associated comorbidities in patients with cancer harboring these germline mutations.

Inherited mutations in cancer predisposition genes are associated with clonal hematopoiesis mutations in patients with solid tumor.

INTRODUCTION

Cancer is a genetic disease characterized by the multistep acquisition of DNA damage due to cell intrinsic or extrinsic factors and the inability to repair it.¹ Mutations in cancer driver genes confer increased fitness to cells, leading to clonal expansion without overt transformation.^1,2 Several recent studies have identified mutations in cancer genes in various, otherwise normal tissues, in an age-dependent manner, suggesting these genomic changes could prime cells for neoplastic transformation.^3-7 Specifically, clonal hematopoiesis (CH) arises when somatic mutations in blood cells provide a fitness advantage and lead to clonal expansion.^8-11 Individuals carrying CH mutations have a higher risk of transformation to hematological cancers and detrimental cardiovascular and pulmonary health through proinflammatory conditions.^12-15 Moreover, previous research studies in our group and others have shown that the presence of CH further increases the risk of developing treatment-related myeloid neoplasms in patients with solid tumor.^16-18 These studies also demonstrated that external factors, such as smoking and cancer therapy, are associated with CH mutations in specific genes, suggesting that different mutagenic processes other than aging might be responsible for mutations in hematopoietic stem cells and, eventually, CH.

CONTEXT

• Key Objectives

• Previous studies of interplay between germline variation and clonal hematopoiesis (CH) have identified strong associations in cohorts of individuals without cancer; however, how CH mutations interact with inherited cancer risk genes leading to hematological malignancies has not been studied thoroughly before.

• Knowledge Generated

• Using prospective clinical tumor-blood genomics data from 46,906 patients with solid tumors, we found that CH is enriched in patients harboring germline variants and that germline ATM mutations are significantly associated with CH, specifically with the acquisition of biallelic mutations in ATM. Moreover, we found that germline variants add to the risk of developing secondary hematological malignancies.

• Relevance

• These findings provide valuable insights into the underlying mechanisms of cancer development and the potential role of germline genetic variations in the pathogenesis of CH with implications for clinical surveillance and risk assessment in cancer patients with CH mutations and germline variants in known cancer predisposition genes.

Cancer develops not only because of external factors but also from inherited genetic variations that lead to cellular transformation.^19,20 Previous research studying associations between germline variants and CH in healthy individuals have yielded inconclusive results. Although a comparison of CH prevalence between monozygotic and dizygotic pairs failed to detect genetic predisposition to CH,^21,22 genome-wide association analyses using large-scale sequencing data, such as those from UK-Biobank initiative, identified common germline genetic variants in TERT, ATM, and TET2 as associated loci with the presence of CH.^23-29 In contrast, rare germline variants in CHEK2 were enriched in patients with CH mutations.^25,29 However, the interplay between mutations in known cancer predisposition genes and CH has not been thoroughly examined in patients who developed solid tumors.

Here, we use data from a prospective clinical genomic profiling program to interrogate the associations of mutations in inherited cancer risk genes and CH from 46,906 patients with solid tumors. We show that CH is enriched in patients harboring rare and pathogenic or likely pathogenic (P/LP) germline variants and that germline ATM mutations are significantly associated with CH, specifically with the acquisition of mutations in ATM itself. Moreover, patients with solid tumor with CH mutations and germline variants in TP53 might be at a higher risk of developing secondary hematological malignancies. We believe the associations between germline variants and CH identified in this study might improve clinical surveillance for CH and its associated comorbidities in patients with cancer.

METHODS

Identification of Clonal Hematopoiesis and Germline Mutations

We analyzed sequencing data from 46,906 patients with solid tumors who have undergone prospective Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) testing using paired tumor and blood samples. This study was approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board, and all patients provided written informed consent for tumor sequencing and review of medical records for demographic, clinical, and pathology information. CH, P/LP germline variants, and clinical variables for each patient were annotated similarly to our previous research work^16,17,30 and anonymized to prevent patient traceability or reidentification. The details of mutation calling, mosaic chromosomal alterations (mCAs) detection using FACETS-CH,³¹ clinical data annotation, and CH confounding variables assessment are provided in the Data Supplement.

Follow-up clinical data regarding the development of secondary leukemia were available for 31,352 patients in our cohort, as described in the previous work.³¹ We annotated different types of leukemias: AML, chronic lymphocytic leukemia (CLL), chronic myeloid leukemia (CML), myelodysplastic syndrome (MDS), and myeloproliferative neoplasm (MPN).

MSK-IMPACT clinical cohort containing somatic tumor sequencing data from 70,250 patients was obtained through cBioPortal³² as of January 2023 to observe germline mutation frequencies in ATM, CHEK2, and TP53 and the comutation patterns when PPM1D tumor mutations were present in patients with solid tumor.

Associations Between Clonal Hematopoiesis and Germline Mutations

Multivariable logistic regression models were run to assess germline variants association with CH at different levels by fitting generalized logistic models using the glm() R function. Overall association between the presence of rare and P/LP germline variants with the presence of CH was studied by fitting generalized logistic models accounting for the following and already evaluated CH-confounding variables in previous studies in our laboratory^16,17: patient's age when the blood sample was drawn for sequencing, sex, inferred ancestry,³³ smoking status, major cancer type, and treatment status (Data Supplement). False discovery rate (FDR) correction by Benjamini and Hochberg³⁴ was applied to control for multiple-testing assessment, annotating significance at FDR P value < .1 (Q < .1).

RESULTS

Integrated Analysis of Germline and CH Variants

We began our analysis by identifying CH and germline variants from clinical sequencing data from 46,906 patients with solid tumors who have undergone MSK-IMPACT testing using paired tumor and blood samples. Joint anonymization of germline variants and CH mutations was performed to prevent patient reidentification (Fig 1A). Nonsilent CH variants were detected as described previously^16,17 in 13,988 patients (29.8%), with 33.5% (n = 4,684) of those harboring two or more CH mutations (Fig 1B and Data Supplement). DNMT3A was the most recurrently mutated gene (n = 5,358; 38.3%), followed by TET2 (n = 1,674; 12%), PPM1D (n = 1,208; 8.6%), ASXL1 (n = 681; 4.9%), and TP53 (n = 543; 3.9%) similar to previous results¹⁷ (Data Supplement). The median variant allele frequency (VAF) for all CH mutations was 4.5% in patients' blood (IQR, 6.1%) (Data Supplement). Half of the CH mutations detected were identified as putative drivers (CH-PD) (n = 10,743; 50.7%) (Data Supplement). Previous studies have identified several factors that are significantly associated with the presence of CH mutations. Therefore, we used a multivariable regression model to show the effects of these confounding variables on the rate of CH across the cohort (Data Supplement). Patient age, genetic ancestry, smoking history, solid tumor type, and the treatment history of the patients before blood draw were all associated with the presence of CH in this data set.

FIG 1.

Integrated analysis of pathogenic germline variants and CH mutations in 46,906 patients with cancer. (A) Study diagram. Identification of nonsilent CH mutations and rare P/LP germline variants was assessed on 90 CPGs in our cancer patient cohort data where the patient's tumor and blood were sequenced using MSK-IMPACT gene panel testing. Joint data anonymization step was performed on CH and P/LP germline mutations to impede patient reidentification. (B) Fraction of patients with pathogenic germline variants in the entire cohort (orange) and the fraction of patients presenting CH mutations (red). Venn diagram showing the overlap between patients with CH somatic mutations and patients with predicted pathogenic germline variants. (C) Percentage of patients in each of the age bins in our cohort showing the total of patients (gray bar plot), patients with germline variants (orange), and patients harboring CH mutations (red). (D) Forest plot showing log odds ratio values resulting from logistic regression models assessing the effect of the presence of germline variants toward different CH genotypes (ie, presence of CH, no. of CH mutations per patient, presence of CH-PD or CH mutations with 10% VAF or higher) when accounting for known CH confounders (ie, patient's age when blood was drawn, sex, inferred ancestry, smoking history, solid tumor cancer type, and if a patient was treated with cancer therapy or not). Denominators indicate the no. of patients with P/LP germline variants that were either negative for CH or positive for the CH criteria listed. P values from multivariable regression models are expressed as follows: *P < .05; **P ≤ .01. (E) Association between most recurrently germline-mutated genes in our cohort (genes mutated in 50 or more patients in the cohort) with CH presence: (from left to right) orange-colored bar plot showing the percentage of patients harboring germline variants in each gene from y-axis; red-colored bar plot presenting the percentage of patients with CH among those with germline variants in each gene form y-axis; forest plot with log odds ratio values indicating the level of association of germline variants in each gene from y-axis with CH presence. Logistic regression models were performed accounting for known CH confounders (ie, patient's age when blood was drawn, sex, inferred ancestry, smoking history, solid tumor cancer type, and if the patient was treated with cancer therapy or not). FDR-adjusted P value significance from multivariable regression models are expressed as follows: *Q < .1; **Q ≤ .05; ***Q ≤ .01. CH, clonal hematopoiesis; CPGs, cancer predisposition genes; FDR, false discovery rate; MSK-IMPACT, Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets; P/LP, pathogenic or likely pathogenic; PD, putative driver; VAF, variant allele frequency.

P/LP germline variants in 90 cancer predisposition genes (Data Supplement) were identified in 13.6% (n = 6,397) of patients in our cohort (Fig 1B), with the vast majority harboring a single P/LP variant (n = 5,643; 88.2%) (Data Supplement). The most frequently mutated genes were MUTYH (n = 764; 11.9%), APC (n = 710; 11.1%), BRCA2 (n = 708; 11.1%), CHEK2 (n = 677; 10.5%), and BRCA1 (n = 526; 8.2%) (Data Supplement). Altogether, 1,924 patients with CH harbored at least one P/LP germline variant (Fig 1B). Although CH prevalence increased with patient age, the rate of P/LP germline variants was generally stable across age groups (min: 8% [age: 90-100] and max: 17.9% [age: 30-40]; Fig 1C).

We next sought to assess whether carriers of P/LP germline variants with solid tumors were more likely to develop CH and whether the resulting CH exhibited any characteristic features. Overall, we observed a small but significant enrichment of patients with CH among those harboring P/LP germline mutations (odds ratio [OR], 1.09, P = .019; Fig 1D) even when taking into account confounders known to be associated with CH mutations (Data Supplement). However, we did not observe a significant association when only considering high penetrance P/LP germline variants in cancer predisposition genes (OR, 1.07, P = .204). Although we identified a subtly significant association between P/LP variants and CH-PD mutations (OR, 1.09, P = .045), other features of CH mutations were not significant such as total number of CH mutations (OR, 1.06, P = .354) or large clone size as measured by CH mutations with at least 10% VAF (OR, 1.08, P = .175). These results suggest that although P/LP germline variants might drive the presence of CH, the effect is not seen in the global features of CH.

Gene-Level Associations

To assess whether P/LP germline variants in specific cancer predisposition genes were associated with CH, we performed multivariate regression analysis at the gene level while accounting for known CH confounders. We observed a higher than the overall cohort CH-positive fraction (29.8%) in patients with mutations in certain predisposition genes such as g.APC (35.1%), g.CHEK2 (33.4%), g.ATM (35.8%), and g.ERCC3 (33.6%) (Fig 1E). Nevertheless, only germline variants in ATM (OR, 1.58, Q = .028) showed a significant association with the presence of CH in our cohort after accounting for age and other confounders (Fig 1E), as has previously been reported.^26,27 Although genome-wide comparative studies between CH-positive and CH-negative populations have identified an association with common germline variants in TERT,^25,26 only one patient of seven harboring rare P/LP germline mutations in TERT was identified as CH-positive in our cohort. Interestingly, although rare pathogenic variants in CHEK2 have also previously been linked to germline predisposition to CH,^25,29 we only observed significant enrichment for patients with CH among those with g.CHEK2 mutations in univariate analysis (Fisher test P = .037) while multivariate analysis was not significant after taking into account all CH confounders (OR, 1.07, Q > .1) (Fig 1E). The inclusion or exclusion of the CHEK2 I157T variant did not change the association results. However, recently highlighted CHEK2 frameshift deletion by Kessler et al²⁹ was also found to be associated with CH (OR, 1.73, P = .047) in our data. Furthermore, we did not observe any significant associations between the number of CH variants, the presence of CH-PD or high-VAF CH mutations, and P/LP germline mutations in ATM or any other predisposition gene evaluated (Data Supplement).

Biallelic Inactivation of Cancer Predisposition Genes

If P/LP germline variants were playing a biological role in the emergence of CH, we might detect evidence of positive selection for somatic CH mutations leading to biallelic inactivation of the corresponding genes. Overall, we found that patients carrying a P/LP germline variant were significantly more likely to harbor at least one CH mutation in the same gene (OR, 16.7, P < .001) after considering all CH confounding variables. We next conducted a gene-by-gene regression analysis and observed a strong selection for ATM CH mutations in patients with g.ATM mutations (ATM CH rate, 5.8%; OR, 12.05; and Q < .001; Fig 2A and Data Supplement), suggesting biallelic inactivation as a possible mediator of clonal expansion. Similarly, TP53 CH mutations were significantly enriched in g.TP53 mutation carriers (TP53 CH rate, 4.4%; OR, 9.11; and Q < .001). Germline carriers of P/LP variants occasionally harbored multiple independent CH mutations affecting the same gene. We identified two g.ATM patients with three separate ATM CH mutations each (and additional CH variants in TET2 and ERBB3, respectively) and one patient with g.TP53 with two independent CH-TP53 mutations (Data Supplement). Although enrichment analysis for multiple CH mutations in the same gene in patients with g.ATM or g.TP53 did not reach significance (Fisher test P = .65 and .26, respectively); this provides sporadic evidence of clones independently undergoing biallelic inactivation through the acquisition of different somatic CH mutations.

FIG 2.

Comutation states between germline variants and CH mutations and DDR genes relations. (A) Percentages of CH mutation rate in ATM, CHEK2, and TP53 genes when each gene is also germline-mutated (+) or not mutated (−). FDR-adjusted P value significance from multivariable regression models are expressed as follows: ***Q ≤ .01 or NS. (B) Patients carrying germline P/LP mutations in CHEK2, ATM or TP53 (orange), or CH mutations (dark red) in PPM1D are plotted to visualize mutual exclusivity between mutation status in patients with alterations in those genes (n = 2,352; 5.01%) of all patients in our cohort (N = 46,906). Gray indicates no alteration (germline or CH) in the corresponding patient for that specific gene. (C) Patients carrying germline P/LP mutations in CHEK2, ATM, or TP53 (orange) and somatic mutations (teal) in PPM1D are plotted to visualize the mutual exclusivity between mutation status in patients with alterations in those genes. Gray indicates no alteration in the corresponding patient for that specific gene. No. of patients with a variant in an indicated gene is shown under the gene name. CH, clonal hematopoiesis; DDR, DNA damage response; FDR, false discovery rate; MSK-IMPACT, Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets; NS, not significant; P/LP, pathogenic or likely pathogenic.

Interestingly, patients carrying P/LP germline variants in another DNA damage response (DDR) gene, CHEK2, showed a negative trend toward CH-CHEK2 mutation presence (CHEK2 CH rate, 0.14%; OR, 0.2; Q = .314; Fig 2A and Data Supplement). However, we reasoned that CH mutations alone fail to account for an important mechanism of gene inactivation and driver of clonal evolution, namely somatic copy number variation.^3,35-37 Following previous studies using MSK-IMPACT clinical data,^31,38 we characterized the presence of mosaic chromosomal alterations (mCAs) in 40,349 patients from our current cohort (see the Methods section). This revealed three additional patients with g.CHEK2 and CH mCAs leading to loss of heterozygosity (LOH) and enrichment of the CHEK2 mutant allele, suggesting a positive selection for CHEK2 biallelic inactivation with LOH-mCA as a second hit (CHEK2 mCA rate, 0.4%, P < .001 and Q = .011) (Data Supplement). Higher g.CHEK2 mutation VAF values in the three patients carrying the LOH-mCA supported this observation (Data Supplement). More broadly, we identified 888 patients harboring at least one mCA event, differentiated between deletions (n = 305), amplifications (n = 392), and copy neutral-LOH (n = 228) (Data Supplement). Approximately 50% of mCA-positive patients (n = 434) also harbored nonsilent CH mutations, although regression models accounting for CH confounders produced no significant results when assessing associations between P/LP germline mutations and the presence of mCAs (Data Supplement). However, taken together, our results suggest that both mutations and mCA are potential mechanisms that can lead to biallelic inactivation in DDR genes by cooperating with most inactivating germline mutations.

CH-PPM1D Mutation Depletion in Patients With g.DDR Mutations

We next sought to determine whether carriers of germline P/LP variants in specific genes tended to develop CH-associated mutations in other genes. Pairwise analysis revealed several associations between P/LP variants and CH genes. For example, DNMT3A CH mutations were significantly enriched among patients with g.ATM mutations (P = .002 and Q = .008), TET2 CH mutations were enriched in patients with g.FANCA (P = .005 and Q = .05), and SF3B1 CH mutations were associated with g.MUTYH (P = .01 and Q = .01) (Data Supplement). Conversely, we identified a significant depletion of CH-PPM1D variants in patients with g.CHEK2 alterations (three patients of 672; OR, 0.15; P = .008; and Q = .08) (Fig 2B and Data Supplement). This observation further extended to patients with g.ATM and g.TP53 variants (zero patients identified in both scenarios), suggesting a negative selection toward CH-PPM1D clones when there are germline defects in CHEK2, ATM, and TP53. Intriguingly, this negative association between germline variants in DDR-related genes (ie, CHEK2, TP53, and ATM) was also observed in somatic tumor mutations identified through the same clinical sequencing platform (n = 68,185). We found no somatic PPM1D tumor mutations in patients with g.CHEK2 (n = 360), despite a pan-cancer somatic prevalence of 1.2% (Fig 2C). This depletion was equally persistent in comparisons with germline mutations in other DDR pathway genes (ie, no patients with somatic PPM1D mutations of the 222 with g.ATM or 55 patients for g.TP53). However, these patterns of mutual exclusivity were not present between CH-PPM1D and CH mutations affecting CHEK2, ATM, or TP53.

We considered the possibility that mutations in the DDR genes (ie, CHEK2, ATM, and TP53) and mutations in PPM1D might produce a fitness disadvantage in carrier clones and performed functional experiments using CRISPR-induced PPM1D + CHEK2-knockout cell lines (refer the Methods section and Data Supplement). Flow cytometry experiments tracking clonal competition showed no difference between wild-type (WT) and PPM1D + CHEK2-knockout cell lines (Data Supplement), suggesting that cells with both PPM1D and CHEK2 genes disrupted are also capable of proliferation. However, we cannot rule out the possibility that compensatory mechanisms occur when both genes are mutated or that functional consequences might be restricted to specific cancer types.^39,40

Germline Variants and Risk of Secondary Leukemias

Finally, we sought to explore how P/LP germline variants might contribute to the risk of developing secondary leukemias and their presence in composite genotypes in patients with CH mutations or mCAs. Clinical follow-up data were available for 31,352 patients in our solid tumor cancer patient cohort. Among those, 59 patients developed secondary leukemias (AML, CLL, CML, MDS, and MPN) and 42 of them harbored at least one P/LP germline variant, CH mutation, or mCA event (Data Supplement).

As we previously reported, patients with CH composite genotypes (ie, harboring CH mutation and mCAs) showed a higher cumulative incidence of secondary leukemias; moreover, patients also carrying P/LP germline mutations (ie, CH mutation, mCA, and germline mutation) appeared to be at slightly higher risk (Fig 3A). Patients with g.ATM mutations were not observed among those developing secondary leukemias (Fisher test P = 1). However, germline mutations in TP53 and CHEK2 were identified in five and three patients, respectively (Fisher test P < .001 and P = .056, respectively). Multivariable Cox proportional hazard model assessing the contribution of different genomic composites showed that CH mutations (HR, 3.39; 95% CI, 1.92 to 6; P = .0000246) and mCAs (HR, 14; 99% CI, 7.86 to 28.6; P < .00001) were contributing to the incidence model, as it was already reported³¹ (Fig 3B). However, germline variants in TP53 showed a specific contribution to the hazard model (HR, 36; 99% CI, 10.18 to 134.4; P = .0000000411), whereas variants in CHEK2 did not (HR, 14; 95% CI, 0.27 to 6.7; P = .709).

FIG 3.

Risk of secondary leukemia in CH patients. (A) Cumulative incidence of secondary leukemia in patients with solid tumor grouped according to individual genotype: patients carrying rare and P/LP germline events, CH mutations, and mCA events (germ + CH + mCA, orange); patients carrying CH mutations and mCA events (CH + mCA, purple); patients with CH mutations only (CH, red); and patients with neither of the previous mutational events (none, gray). These groups are exclusive of each other. (B) Multivariable cause-specific Cox regression analysis to assess mutational events identified in patients contributions: presence of CH mutations (CH), mCAs (mCAs), rare and P/LP germline events (Germline.Event), in TP53 (g.TP53) and CHEK2 (g.CHEK2). No. under y-axis genotype labels indicate patients developing leukemia of those with the specific genotype among the subset analyzed in the Cox regression analysis (n = 31,352). P values from multivariable cause-specific Cox regression analysis are indicated on the right side of the plot for each subset and expressed as follows in the plot: *P ≤ .001. CH, clonal hematopoiesis; mCA, mosaic chromosomal alterations; P/LP, pathogenic or likely pathogenic.

DISCUSSION

Genetic susceptibility to cancer has been extensively studied, but only a few studies have examined the effects of inherited variants on the genetic predisposition to CH identifying risk loci in healthy populations and sometimes with conflicting results.^21-25,28 Here, by leveraging results from MSK-IMPACT, which sequences tumor and matched blood samples to enable accurate identification and differentiation of germline variants and CH, we identified associations between pathogenic germline variants and the presence of CH mutations in patients with solid tumor.

Our results show that patients with CH mutations are enriched in those harboring rare P/LP germline variants and that ATM germline mutations might play a role in this association. While previous studies had highlighted higher rates of CH mutations among patients with germline variants in certain genes, most of these observations were performed by applying genome-wide approaches studying common variant loci.^25,26,41-43 For instance, common germline variants in TERT were identified in genome-wide comparative studies between CH-positive and CH-negative healthy individuals.^26,44 In this study, we only identified one patient of seven carrying rare P/LP germline mutations in TERT and also harboring CH mutations. The TopMed study by Bick et al²⁵ reported some degree of association between the higher prevalence of CH and germline mutations in CHEK2 when aggregating rare putative loss-of-function germline variants, while a recent work by Kessler et al²⁹ identified a specific germline frameshift variant in the CHEK2 gene that was significantly associated with CH. Although we were able to confirm the association signal for this specific CHEK2 frameshift deletion, we did not find an overall association between P/LP germline mutations in CHEK2 and the presence of CH mutations in our data after correcting for multiple known CH confounders. However, we did observe an elevated number of CH mutations in patients with these pathogenic CHEK2 germline variants. Therefore, we believe our highly annotated cohort for different CH confounders could be closer to elucidating the actual contribution of rare and pathogenic germline variants in CHEK2 toward CH presence. Similarly, several genes were observed to surpass the overall fraction of patients with CH (eg, APC, CHEK2, ATM, ERCC3, and FANCC) but did not reach significance after accounting for the multiple CH confounding variables, thus reflecting the importance of adequately annotated data.

We further report a significant association between germline mutations in ATM and CH after considering different CH confounders. Moreover, we observed a positive enrichment of ATM-CH mutations in patients carrying ATM germline mutations. Slavin et al²⁷ assessed P/LP germline variants in individuals through a cancer predisposition gene panel testing, identifying P/LP germline variants and the presence of likely somatic mutations (ie, CH variants). Although they did not find an overall association between both, they reported enrichment of CH mutations among patients with germline variants in ATM. Interestingly, they also reported that, among g.ATM carriers, most CH mutations were in ATM.

We observed different comutation patterns in patients with germline mutations in DDR genes (ie, CHEK2, ATM, or TP53). While comutation states in ATM and TP53 were positively enriched suggesting biallelic inactivation, CHEK2 germline variants were negatively associated with CH mutations in CHEK2. However, we have identified three patients with germline mutations in CHEK2 and harboring mCAs in the form of LOH affecting the genomic locus of CHEK2, pointing toward a biallelic inactivation in those clones through copy number alteration. Germline mutations in ATM, CHEK2, and TP53 have been associated with CH and other hematopoietic malignancies.^28,45 Moreover, germline CHEK2 mutations have been linked to somatic chromosomal mosaicism.³⁵ However, to our knowledge, biallelic inactivation of DDR genes in CH was not reported yet, and we currently lack functional proof of their significance.

Similarly intriguing was the depletion of PPM1D-CH mutations in patients with rare P/LP germline variants in the DDR genes. This same pattern of mutual exclusivity was also observed in solid tumors, showing a negative association toward PPM1D tumor mutations in patients with CHEK2, ATM, or TP53 germline mutations. However, functional experiments did not identify whether genetically modified leukemia cells with impeded CHEK2 and PPM1D genes showed a less growth-favorable phenotype than those with only PPM1D knockout.

We also reported that rare P/LP germline mutations in TP53 could increase the risk of developing secondary leukemias in patients with solid tumors, given the association observed in the Cox multivariable model assessing for different mutational composites. Although overall germline events have shown a neutral contribution to this model, g.TP53 variants were observed as one of the main effectors, along with CH and mCAs. Germline variants in TP53 have been associated with an increased risk of developing leukemias, especially with hypodiploid ALL, characterized by aneuploidy and poor outcome.^46,47

Our curated identification of rare P/LP germline variants while using blood and tumor samples to characterize CH variants efficiently has produced a suitable data set to evaluate the association between germline events and CH mutations. However, efforts toward growing this cohort are crucial to further study associations at the gene level. The gene relations observed here point to potential biological interactions that will be worth exploring further. Associations identified in this work might translate into enhanced clinical surveillance for CH and associated comorbidities in patients with cancer harboring these germline mutations and helping advance clinical care best practices for patients with cancer with CH. While long-term clinical follow up data are currently lacking, we anticipate that the recent establishment of CH clinics at many academic centers will enable longitudinal monitoring of large cohorts of patients to inform future strategies for prognostication and potential preventative treatment modalities.

AUTHOR CONTRIBUTIONS

Conception and design: Sebastià Franch-Expósito, Ryan N. Ptashkin, Pablo Sánchez-Vela, Elizabeth Comen, Larry Norton, Barry Taylor, Ross Levine, Kenneth Offit, Michael F. Berger, Ahmet Zehir

Financial support: Michael F. Berger

Provision of study materials or patients: Pablo Sánchez-Vela, Barry Taylor, Kenneth Offit

Collection and assembly of data: Sebastià Franch-Expósito, Ryan N. Ptashkin, Kelly L. Bolton, Chaitanya Bandlamudi, Linda Zhang, Erika Gedvilaite, Pablo Sánchez-Vela, Elli Papaemmanuil, Ross Levine, Kenneth Offit, Michael F. Berger, Ahmet Zehir

Data analysis and interpretation: Sebastià Franch-Expósito, Miika Mehine, Ryan N. Ptashkin, Kelly L. Bolton, Chaitanya Bandlamudi, Preethi Srinivasan, Margaret A. Goodell, Kamal Menghrajani, Pablo Sánchez-Vela, Diana Mandelker, Larry Norton, Ryma Benayed, Teng Gao, Elli Papaemmanuil, Ross Levine, Kenneth Offit, Zsofia Stadler, Michael F. Berger, Ahmet Zehir

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).