Clonal hematopoiesis is associated with reduced response to atezolizumab in non-small cell lung cancer

Abstract

Background

The acquisition of somatic mutations associated with the expansion of certain hematopoietic stem cells is known as clonal hematopoiesis (CH). Despite evidence of CH association with age-related pro-inflammatory diseases, little is known about its role in the response to antitumor immune modulatory therapies such as immune-checkpoint inhibition (ICI). In this report, we characterize the effect of CH on ICI objective response in non-small cell lung cancer (NSCLC).

Methods

We performed baseline blood-derived whole-exome sequencing of 1,281 patients with NSCLC from five anti-programmed death ligand-1 (PD-L1) trials (FIR, POPLAR, IMpower110, IMpower150, and IMpower131). Multivariable logistic regression adjusting for age, sex, and smoking history was used to model the odds of being a non-responder in each trial, and random effects meta-analyses were performed in anti-PD-L1 and comparator arms.

Results

CH carriage, defined as the presence of CH with variant allele fraction≥2%, was associated with adverse objective response across the anti-PD-L1 treatment arms (OR=1.69 (95% CI 1.08 to 2.63); p=0.02). No association was observed in the comparator arms (OR=0.61 (95% CI 0.31 to 1.21); p=0.16). CH carriers had lower absolute leukocyte counts (p_Bonferroni=0.01), driven by reduced lymphocytes (p_Bonferroni=0.004).

Conclusions

In this report, we provide evidence of the adverse impact of CH on ICI treatment objective response in NSCLC, highlighting CH as a potential predictive biomarker for patient stratification in ICI. Further experimental validation is needed to better understand the mechanism of action behind this reduction in ICI efficacy.

Introduction

The accumulation of somatic alterations in hematopoietic stem cells (HSCs) is a common feature of aging. While many acquired mutations have no apparent phenotypic effects, certain mutations in leukemia-related genes such as DNMT3A, TET2, and ASXL1 provide proliferative advantage, resulting in the clonal expansion of certain HSC lineages, known as clonal hematopoiesis (CH).^{1 2} Recent studies have found that CH carriers without overt hematological malignancies are at increased risk of inflammatory-mediated disorders, including cardiovascular diseases and stroke.¹ CH has also been implicated as a predictive biomarker, improving the efficacy of immune-modulatory therapeutic agents such as chimeric-antigen receptor (CAR)-T cell therapy in hematological malignancies.^{3 4} Given the role of CH in altered inflammatory signaling, there is emerging interest in understanding the impact of CH in other contexts involving therapeutic immune modulation, such as immune checkpoint inhibition (ICI) for solid tumors.

ICI targeting the programmed death-1/ligand-1 (PD-1/L1) immune checkpoints has significantly improved the historically dismal prognosis of advanced-stage non-small cell lung cancers (NSCLCs).^{5 6} However, most patients do not benefit from the treatment, motivating efforts to identify reliable predictive biomarkers beyond tumor PD-L1 expression and tumor mutational burden (TMB).7,9 While growing evidence points to the adverse effect of CH in cancer prognosis,^{10 11} the role of CH as an ICI predictive biomarker has not been widely explored.

In contrast to the positive impact of CH observed for CAR-T in hematological malignancies,^{3 4} a recent report, focusing on 1,677 ICI-treated patients, provided the first indication of an overall adverse impact of CH in outcomes related to solid tumors.¹² However, in that report, discordant effects were observed in different solid cancer types, possibly reflecting evidence of both ICI-sensitizing and ICI-resistant inflammatory markers enriched in CH-harboring immune cells. Specifically, CH was shown to be associated with improved ICI outcomes in colorectal cancer (n=176), while the opposite effect was observed in glioma (n=112). No statistically significant associations were seen in other cancer types, including NSCLC (n=340), which is a growing target for ICI therapeutics. In the present study, we leveraged data from 1,281 patients with NSCLC participating in multiple anti-PD-L1 trials to characterize the effect of CH on ICI treatment objective response in NSCLC.

Methods

Study sample

The study sample included 1,281 patients with NSCLC (n=857 receiving anti-PD-L1 monotherapy or in combination with chemotherapy, and n=424 on the comparator arms) from five completed randomized clinical trials sponsored by F. Hoffman-La Roche/Genentech. These trials included FIR (GO28625) (anti-PD-L1 arm n=85, no comparator arm), POPLAR (GO28753) (anti-PD-L1 arm n=74, comparator arm n=74), IMpower110 (GO29431) (anti-PD-L1 arm n=85, comparator arm n=78), IMpower150 (GO29436) (anti-PD-L1 arm n=363, comparator arm n=182), and IMpower131 (GO29437) (anti-PD-L1 arm n=250, comparator arm n=90). Objective response was measured by tumor shrinkage using standardized Response Evaluation Criteria in Solid Tumors V.1.1 (RECIST V.1.1).¹³ Complete response (complete tumor shrinkage) and partial response (PR: ≥30% tumor shrinkage) were grouped as responders, while progressive disease (PD: ≥20% increased tumor size) and stable disease (neither PR nor PD) were grouped as non-responders. The study protocol of each clinical trial was approved by the Ethics Committees and the Institutional Review Boards in each country for each study site. All study participants provided informed consent for all biological materials used in this study.

CH somatic variant calling from peripheral blood

We performed whole-exome sequencing (WES) of blood-derived genomic DNA, following protocols as detailed in Nabet et al.¹⁴ The WES library was prepared with Agilent SureSelect V.6 and sequenced at 2×150 bp. The average read depth was approximately 72X, which has been shown to provide reliable CH calling at variant allele fraction (VAF)≥2%.¹⁵ CH calling and quality controls were performed in accordance with the standard recommendations for WES-derived CH calling.¹⁶ In brief, we aligned the raw reads using the Genome Analysis Toolkit¹⁷ with GRCh38. Somatic variant calling of the germline WES data was performed using Mutect2 in “tumor-only” mode for the exonic regions of the 43 CH driver genes described in Kar et al.¹⁶ To account for known recurrent sequencing artifacts and possible germline variants, we used external population references from the 1000 Genomes Project¹⁸ and the Genome Aggregation Database.¹⁹ Putative CH variants were retained if flagged as “PASS” by FilterMutectCalls, had minimum alternate reads of 2, and read depth of 7 for single nucleotide variants or 10 for indels/substitution. The selected high-quality variants were then functionally annotated using Variant Effect Predictor (VEP),²⁰ and an individual was considered a CH carrier if they carried at least one of the previously known CH driver mutations from the CH candidate gene list with VAF≥2%.¹⁶

Statistical analysis

To investigate the impact of CH on anti-PD-L1 treatment objective response in each trial, the odds of being a non-responder versus a responder were modeled using a multivariable logistic regression adjusted for age at treatment, sex, and smoking history. We then performed a random-effects meta-analysis of all the trials for any CH gene carriers, as well as for DNMT3A-specific, and TET2-specific CH carriers. A random-effects model was used in this study, rather than a fixed-effect model, to account for potential heterogeneity across each trial. We leveraged the comparator arm of each trial and performed the same analyses in this patient sample. We further tested if the overall effect of CH on ICI objective response in the anti-PD-L1 arm differed from that in the comparator arm. We used a random-effects meta-analysis with a treatment group (anti-PD-L1 or comparator arm) as a moderator to determine whether the effect of CH is unique to atezolizumab and not merely prognostic, while accounting for potential heterogeneity across studies.

Additionally, to explore if CH impacts ICI efficacy via the modulation of the abundance of ICI-modulating immune cell populations, we performed Wilcoxon’s rank-sum tests to compare the differential status of six available immune cell counts measured in peripheral blood at baseline (detailed in online supplemental table S3).

Results

Patient characteristics

The study sample included a total of 1,281 patients with NSCLC (n=857 on the anti-PD-L1 arms and n=424 on the comparator arms) (table 1) from five randomized clinical trials (see Methods and online supplemental table S1). The median age at treatment was 64 years, and most of the study sample were smokers (61.7% previous and 23.6% current smokers), as expected in patients with NSCLC. The proportion of males was higher than females (66% male), reflecting the sex-biased prevalence of lung cancer.²¹ The prevalence of CH carriers (VAF≥2%) in the study sample was 14.1% (14.4% in the anti-PD-L1 arms, and 13.7% for the comparator arms) (table 1). Consistent with previous reports,¹⁵ CH was absent in patients under 40 years old, with prevalence increasing to 39.1% in those over 80 (online supplemental figure S1A). Most CH carriers (91.7%) had only one CH driver mutation (online supplemental figure S1B), with DNMT3A and TET2 being the most commonly affected genes (online supplemental figure S1C).

Table 1. Clinical characteristics of the study sample in each randomized clinical trial.

	CH status	Response status	Age	Sex	Smoking history
	Yes (%)	Non-responders (%)	Median (range)	Male (%)	Never (%)	Previous (%)	Current (%)
Anti-PD-L1 arm (N)
FIR (85)	17 (20.0)	61 (71.8)	66 (41–85)	49 (57.6)	9 (10.6)	62 (72.9)	14 (16.5)
POPLAR (74)	12 (16.2)	57 (77.0)	61 (42–82)	48 (64.9)	10 (13.5)	51 (68.9)	13 (17.6)
IMpower110 (85)	5 (5.9)	57 (67.1)	64 (46–81)	55 (64.7)	11 (12.9)	55 (64.7)	19 (22.4)
IMpower150 (363)	54 (14.9)	167 (46.0)	63 (31–83)	228 (62.8)	65 (17.9)	207 (57.0)	91 (25.1)
IMpower131 (250)	35 (14.0)	95 (38.0)	65 (25–85)	199 (79.6)	24 (9.6)	158 (63.2)	68 (27.2)
Overall (857)	123 (14.4)	437 (51.0)	64 (25–85)	579 (67.6)	119 (13.9)	533 (62.2)	205 (23.9)
Comparator arm (N)
FIR (NA)	NA	NA	NA	NA	NA	NA	NA
POPLAR (74)	13 (17.6)	53 (71.6)	61.0 (36–80)	36 (48.6)	17 (23.0)	47 (63.5)	10 (13.5)
IMpower110 (78)	8 (10.3)	48 (61.5)	65.0 (44–84)	52 (66.7)	14 (17.9)	49 (62.8)	15 (19.2)
IMpower150 (182)	26 (14.3)	92 (50.5)	63.0 (31–87)	108 (59.3)	33 (18.1)	107 (58.8)	42 (23.1)
IMpower131 (90)	11 (12.2)	47 (52.2)	65.0 (45–79)	70 (77.8)	5 (5.6)	55 (61.1)	30 (33.3)
Overall (424)	58 (13.7)	240 (56.6)	63.5 (31–87)	266 (62.7)	69 (16.3)	258 (60.8)	97 (22.9)
Both arms=1281	181 (14.1)	677 (52.8)	64 (25–87)	845 (66)	188 (14.7)	791 (61.7)	302 (23.6)

CH, clonal hematopoiesis; PD-L1, programmed death ligand-1.

Association between CH and ICI objective response

To elucidate the impact of CH on ICI objective response, we fitted a multivariable logistic regression model adjusted for age, sex, and smoking history. In the random effects meta-analysis (accounting for potential heterogeneity across each trial), CH carriage was associated with reduced RECIST-measured objective response (detailed description of RECIST in Methods) (OR=1.69 (95% CI 1.08 to 2.63); p=0.02). Consistent directions of effects were observed in each tested trial (figure 1). This adverse impact was unchanged when restricted to those patients (97% of the total) with PD-L1 positivity, as derived from tumor-infiltrating immune cells (online supplemental figure S1): OR=1.66 (95% CI 1.06 to 2.59); p=0.03). To address any potential residual confounding from PD-L1 stratification, we performed an additional analysis adjusting for granular PD-L1 tiers among PD-L1 positive samples. Granular PD-L1 data were only available for 39.5% of the samples, but these were representative of the full data set and the effect size (OR=1.86 (95% CI 0.89 to 3.92); p=0.1; online supplemental figure S3A) was consistent with that observed in the overall PD-L1 positive cohort (OR=1.66 (95% CI 1.06 to 2.59); p=0.03). Adjustment for PD-L1 immune cell (IC) 1–3 tiers (1–5%: low, 5–10%: medium, ≥10%: high) in this subset had no impact on the effect size (OR=1.95 (95% CI 0.90 to 4.21); p=0.09; online supplemental figure S3B). These findings indicate that there is minimal residual confounding from PD-L1 stratification.

Figure 1. Forest plot of the association between CH and treatment response in anti-PD-L1 and comparator arms. OR>1 indicated CH was associated with increased odds of non-response, while OR<1 suggested that CH was associated with increased odds of response. Results from each trial were meta-analyzed using a random effect model. We also meta-analyzed the association between CH and response in a subgroup of DNMT3A-specific and TET2-specific CH. The top panel represented results from the anti-PD-L1 treated group, while the bottom panel displayed results from the control arms. FIR was a single-arm anti-PD-L1 trial and was not included in the control arm analysis. CH, clonal hematopoiesis; PD-L1, programmed cell death ligand-1.

To account for the potential effect of ancestry, we repeated these analyses in the full cohort for a subset of patients with self-reported European ancestry, who comprised the largest proportion of our study sample (84% of the total). We observed a consistent adverse effect of CH (online supplemental figure S3): OR=1.80 (95% CI 1.11 to 2.93); p=0.02), indicating that our findings were robust and not confounded by ancestry.

To determine if the effect of CH is similar across CH driver genes, we repeated the full-cohort analyses for the two most prevalent genes (DNMT3A and TET2). Numbers were small—38 with DNMT3A-specific CH and 20 with TET2—and neither subset reached nominal significance. However, a comparison of the OR estimates indicated that most of the signal was attributable to CH due to DNMT3A mutations (OR=1.93 (95% CI 0.92 to 4.07); p=0.08; vs TET2 (OR=0.98 (95% CI 0.39 to 2.48); p=0.94).

We also investigated the effect of CH on objective response in the comparator arm. No associations were observed, either for the overall CH gene (OR=0.61 (95% CI 0.31 to 1.21); p=0.16) or CH gene-specific subgroup analyses (DNMT3A: OR=0.55 (95% CI 0.15 to 2.09); p=0.38 and TET2: OR=0.64 (95% CI 0.14 to 2.93); p=0.57) (figure 1). To test if the overall effect of CH on objective response differed between those in the anti-PD-L1 arm and those in the comparator arm, we performed a random-effect meta-analysis comparing the two treatment groups (anti-PD-L1 vs comparator) as a moderator. The effect of CH on the odds of being a non-responder was statistically significantly different between the anti-PD-L1 versus comparator arm (meta-analysis OR=2.75 (95% CI 1.22 to 6.21); p=0.01), indicating that the adverse impact of CH on objective response is specific to ICI. These conclusions were unchanged if we restricted the analyses to the self-reported European patients (meta-analysis OR=2.90 (95% CI 1.19 to 7.05); p=0.02).

As a sensitivity analysis, we also explored the association between CH and ICI objective response in a pooled analysis combining data from all five clinical trials. In Model 1, we fitted a multivariable logistic regression adjusted for specific anti-PD-L1 treatment groups (anti-PD-L1 alone, anti-PD-L1+ chemotherapy, or anti-PD-L1+chemotherapy+targeted therapy), in addition to age, sex and smoking status (OR _{Model 1}= 1.70 (95% CI 1.11 to 2.64); p _{Model 1}= 0.02) (online supplemental table S2). In Model 2, we added information on prior exposure to chemotherapy (yes/no) as an additional covariate (OR _{Model 2}= 1.66 (95% CI 1.08 to 2.59); p _{Model 2}= 0.02) (online supplemental table S2). We observed consistent findings of the adverse impact of CH on ICI objective response in both models, with no statistically significant associations observed in the comparator arm (online supplemental table S2). Together, these analyses indicated that specific anti-PD-L1 treatment groups and prior exposure to chemotherapy did not significantly impact our main findings.

Association between CH, progression-free survival, and overall survival

Given the observed association with objective response, as measured using RECIST criteria, we additionally examined if these findings translate to the long-term durable response, as assessed by post-treatment overall survival (OS) and progression-free survival (PFS). We observed no statistically significant association with OS and PFS in the anti-PD-L1 treated group (HR_OS=1.17 (95% CI 0.89 to 1.52); p_OS=0.26; HR_PFS=1.09 (95% CI 0.86 to 1.38); p_PFS=0.49) and the comparator arms (HR_OS=1.02 (95% CI 0.74 to 1.42); p_OS=0.88; HR_PFS=1.17 (95% CI 0.73 to 1.86) ; p_PFS=0.51) (online supplemental figure S5 and S6). The data suggested that the adverse impact of CH observed for object response did not result in detectable differences in OS or PFS.

CH and baseline immune cell counts

To further understand how CH may have impacted ICI response, we tested for potential differences in six baseline circulatory immune cell counts by CH status using Wilcoxon’s rank-sum test (online supplemental figure S3). CH carriers had significantly lower absolute levels of leukocytes (Bonferroni adjusted p value=0.01). This difference was strongly driven by low lymphocytes—a key cell population crucial in antitumor immune response—in CH carriers (Bonferroni adjusted p value=0.004) (figure 2).

Figure 2. Baseline circulatory inflammatory markers and CH. Wilcoxon’s rank-sum test was used to compare circulatory inflammatory blood cell counts by CH status. After Bonferroni multiple testing adjustment, two cell populations were statistically significantly different by CH status (p adjusted<0.05). Lower leukocyte (panel A) and lymphocyte absolute levels (panel B) were observed in CH carriers. CH, clonal hematopoiesis.

CH and immune-related adverse events

Prior research has observed intriguing associations between CH and immune-mediated traits, such as ICI-induced autoimmune myocarditis and CAR-T-associated cytokine release syndrome.^{22 23} These findings prompted us to investigate the role of CH in immune-related adverse events (irAEs) within our study sample. We focused on irAEs with a prevalence of>5% in the anti-PD-L1 arm. These included rash, hepatitis, hypothyroidism and pneumonitis (online supplemental table S4). Given the small number of each irAE in each trial, we performed pooled analyses using a multivariable logistic regression model adjusting for age, sex, smoking history, and study IDs. We observed no statistically significant association between CH and irAEs (online supplemental table S5)

Discussion

The adverse impact of CH has been documented in the context of inflammatory-mediated diseases.¹ However, little is known about its role in the ICI therapeutic immune modulation setting. In this report, we provide the first evidence in NSCLC supporting the adverse impact of CH on ICI-associated antitumor immune response, as measured by post-treatment tumor shrinkage using the RECIST criteria. The limited sample size available for CH driver gene-specific analyses reduced the biological interpretability of the effect of specific CH mutated genes on ICI’s objective response, but numerically, the data suggested that most of the observed association was attributable to DNMT3A mutations. The effect of CH on the objective response to treatment in the anti-PD-L1 arm was significantly different from that in the comparator arm (chemotherapy or observation arms), consistent with CH as an anti-PD-L1 specific predictive biomarker of objective response. Interestingly, CH was able to further stratify a subpopulation of patients with tumor PD-L1 positivity who may not benefit from ICI treatment response, thereby complementing the limited utility of PD-L1 as a predictive biomarker.

A recent report explored the role of CH in ICI treatment outcomes in solid cancers,¹² although the pooled analysis suggested that CH negatively impacted ICI, the results varied across cancer types. In NSCLC, no statistically significant link between CH and post-ICI OS was reported, and the impact on objective response was not investigated. Our study identified an adverse impact of CH on ICI objective response, but consistent with the previous report,¹² this did not extend to OS or PFS. CH-related effects, such as elevated pro-inflammatory cytokines or activated inflammatory signaling pathways, may primarily drive early antitumor regression without sustaining a durable response. Long-term endpoints like PFS/OS, measured further downstream from ICI exposure, may be influenced by various factors like tumor adaptation or acquired immune exhaustion, diluting CH’s adverse impact and making detection challenging with the sample size used for objective response.

Another study hinted at a potential benefit of CH on ICI treatment outcomes, as measured by PFS, but the sample size was small (n=115 ICI+chemotherapy-treated) and the association signal was of only marginal significance (p=0.079).²⁴ Comparisons between this study and ours are further complicated by differing cancer types (pancreatic vs NSCLC in our study) and ICI treatment drugs (anti-CTLA4 vs anti-PD-L1 in our study). Additional studies are needed to further clarify these findings and characterize the nuanced roles of CH across cancer types and ICI therapeutic contexts.

Given the growing evidence linking CH to autoimmune and immune-mediated traits,^{22 23} we also investigated its association with common ICI-induced irAEs, such as rash, hepatitis, hypothyroidism, and pneumonitis. As with a prior study, with more limited sample size,²⁴ our study did not reveal statistically significant associations. If these findings are validated in future larger studies, CH-associated inflammatory pathways could be specifically targeted to enhance anti-tumor immunity without negatively affecting irAEs.

The adverse impact of CH on anti-PD-L1 objective response, as observed in our study, may be explained by CH-associated reductions in the abundance of baseline circulatory immune cells, which have been previously linked with unfavorable ICI outcomes.²⁵ Consistent with a prior report that patients with CH-positive cancer exhibited decreased absolute lymphocyte counts,²⁶ we observed that CH carriers had lower overall leukocyte counts, primarily driven by reduced lymphocyte abundance. Experimental data suggest that CH-harboring HSCs preferentially proliferate into the myeloid lineage,²⁷ which aligns with the reduced lymphocyte population we observed in CH carriers. Beyond affecting cell abundance, CH may also influence immune cell states. CH-harboring HSCs express elevated levels of pro-inflammatory cytokines such as IL6 and IL-1β, which are key drivers of ICI resistance.28,30 Additionally, a study of peripheral blood single-cell expression reported that patients with lung cancer with high CH burden exhibited activated TNF signaling via the Nuclear factor kappa-light-chain-enhancer of activated B (NF-kB) cells pathway in the myeloid compartment.³¹ Previous in vivo clustered regularly interspaced short palindromic repeats (CRISPR) experiments showed that deletion of gene members of this signaling pathway can sensitize tumors to ICI.³² Together, these findings suggest several plausible mechanisms that could link CH to a reduced ICI clinical response, including decreased abundance of ICI-sensitizing immune cells and elevated levels of immunosuppressive circulatory cytokines in CH-harboring HSCs.

Beyond the potential adverse impact of CH on systemic inflammation, evidence also points to the presence of CH-derived immune cells infiltrating into the tumor microenvironment (TME) in up to 42% of CH carriers,^{33 34} suggesting a direct impact of CH on the local inflammation within the tumor niche. A recent study reported an enrichment of CH-harboring myeloid populations in the TME, leading to adverse progression of NSCLC.³⁴ This finding is further supported by tumor transcriptomic analyses that revealed a neutrophil-activated inflammatory landscape in the TME of CH carriers, implicating the pathogenicity of CH.²⁶ In line with this observation, a CRISPR-mediated in vitro experiment demonstrated that CH with DNMT3A loss resulted in an increased CXCL2 expression, a cytokine that promotes neutrophil migration into the tumors.^{26 29 35} Collectively, this body of evidence indicates that the reduced ICI efficacy in CH carriers, as observed in our study, may be driven by direct tumor-infiltrating CH-associated inflammation, and its contribution to protumor immunosuppressive myeloid remodeling within the TME.

Interestingly, in contrast to our observations, several mechanistic studies, conducted in vitro and in vivo, have demonstrated that T cells engineered with deletions of CH-driver genes exhibit reduced T-cell exhaustion and enhanced stem-like progenitor states, resulting in improved efficacy of CAR-T and ICI therapies.^{3 36} Our study, which focused on human subjects with NSCLC, suggests that CH, with mutated HSCs of various lineages, including myeloid cells, is associated with poorer ICI objective response. These observations underscore the potential variable effects of CH-driver gene mutations in different cellular contexts—such as T cells versus myeloid cells—and across different study models, emphasizing the need for further mechanistic exploration in future research efforts.

A key strength of this report is the large sample size, focused on a single tumor type, providing the first statistically significant evidence linking CH to ICI objective response in NSCLC. Our data stem from US Food and Drug Administration-approved clinical trials with well-standardized sequencing and phenotyping information, harmonized across trials. The inclusion of the non-ICI treatment arms allowed us to postulate that CH is an ICI predictive biomarker, not just a prognostic marker of tumor regression.

However, there are several limitations in this report that present opportunities for additional research. We employed a VAF cut-off at 2% to ensure reliable CH phenotyping. While this cut-off is commonly used in CH research and supported by compelling evidence of clinical relevance, future studies with deeper sequencing coverage and larger sample size could offer valuable insights into the role of CH with smaller clones and variable effects at different VAF strata. Another potential limitation relates to between-trial heterogeneity in study protocols. To address this, we employed random-effects meta-analyses approaches to account for such heterogeneity, and to allow us to benefit from the increase in sample size achieved by combining multiple trials. Future research should also expand to other cancer types of different disease stages, different sequential ICI treatments, ICI therapeutic subtypes (anti-cytotoxic T-lymphocyte associated protein 4 (CTLA4), or combinations of anti-CTLA4 with anti-PD-1/L1), while accounting for potential confounders not available in our study sample, including ICI-associated tumor-centric biomarkers like TMB. Additionally, future larger studies should investigate the potential interaction between CH and PD-L1 status. Such efforts, together, could comprehensively characterize the potential role of CH as a novel ICI predictive biomarker.

In conclusion, this report demonstrates the role of CH as a predictive biomarker of anti-PD-L1 checkpoint inhibitor treatment in NSCLC, thereby broadening the clinical significance of the increasing scientific interest in CH to complement the incomplete understanding of ICI resistance mechanisms. With further validation and a deeper mechanistic understanding, CH-associated ICI-resistant inflammatory pathways could be leveraged as novel targets to enhance antitumor immune responses. Moreover, CH, which can be conveniently sampled from peripheral blood and profiled using a cost-effective gene panel targeting CH-driver mutations, as recently proposed,³⁷ offers a significant advantage as an accessible and scalable biomarker. Its integration with other ICI-associated tumor-centric factors has the potential to refine the currently limited predictive models,7,9 leading to optimal patient selection for anti-PD-L1 treatment in NSCLC.

Data availability statement

Data are available upon reasonable request.