Impact of Aneuploidy and Chromosome 9p Loss on Tumor Immune Microenvironment and Immune Checkpoint Inhibitor Efficacy in NSCLC

Abstract

Introduction:

Although gene-level copy number alterations have been studied as a potential biomarker of immunotherapy efficacy in NSCLC, the impact of aneuploidy burden and chromosomal arm-level events on immune checkpoint inhibitor (ICI) efficacy in NSCLC is uncertain.

Methods:

Patients who received programmed cell death protein 1 or programmed death-ligand 1 (PD-L1) inhibitor at two academic centers were included. Across all 22 chromosomes analyzed, an arm was considered altered if at least 70% of its territory was either gained or deleted. Among nonsquamous NSCLCs which underwent targeted next-generation sequencing, we retrospectively quantified aneuploidy using the adjusted fraction of chromosomal arm alterations (FAA), defined as the number of altered chromosome arms divided by the number of chromosome arms assessed, adjusted for tumor purity.

Results:

Among 2293 nonsquamous NSCLCs identified, the median FAA increased with more advanced cancer stage and decreased with higher PD-L1 tumor proportion score (TPS) levels (median FAA in TPS < 1%: 0.09, TPS 1%–49%: 0.08, TPS ≥ 50%: 0.05, p < 0.0001). There was a very weak correlation between FAA and tumor mutational burden when taken as continuous variables (R: 0.07, p = 0.0005). A total of 765 advanced nonsquamous NSCLCs with available FAA values were treated with ICIs. With decreasing FAA tertiles, there was a progressive improvement in objective response rate (ORR 15.1% in upper tertile versus 23.2% in middle tertile versus 28.4% in lowest tertile, p = 0.001), median progression-free survival (mPFS 2.5 versus 3.3 versus 4.1 mo, p < 0.0001), and median overall survival (mOS 12.5 versus 13.9 versus 16.4 mo, p = 0.006), respectively. In the arm-level enrichment analysis, chromosome 9p loss (OR = 0.22, Q = 0.0002) and chromosome 1q gain (OR = 0.43, Q = 0.002) were significantly enriched in ICI nonresponders after false discovery rate adjustment. Compared with NSCLCs without chromosome 9p loss (n = 452), those with 9p loss (n = 154) had a lower ORR (28.1% versus 7.8%, p < 0.0001), a shorter mPFS (4.1 versus 2.3 mo, p < 0.0001), and a shorter mOS (18.0 versus 9.6 mo, p < 0.0001) to immunotherapy. In addition, among NSCLCs with high PD-L1 expression (TPS ≥ 50%), chromosome 9p loss was associated with lower ORR (43% versus 6%, p < 0.0001), shorter mPFS (6.4 versus 2.6 mo, p = 0.0006), and shorter mOS (30.2 versus 14.3 mo, p = 0.0008) to immunotherapy compared with NSCLCs without 9p loss. In multivariable analysis, adjusting for key variables including FAA, chromosome 9p loss, but not 1q gain, retained a significant impact on ORR (hazard ratio [HR] = 0.25, p < 0.001), mPFS (HR = 1.49, p = 0.001), and mOS (HR = 1.47, p = 0.003). Multiplexed immunofluorescence and computational deconvolution of RNA sequencing data revealed that tumors with either high FAA levels or chromosome 9p loss had significantly fewer tumor-associated cytotoxic immune cells.

Conclusions:

Nonsquamous NSCLCs with high aneuploidy and chromosome 9p loss have a distinct tumor immune microenvironment and less favorable outcomes to ICIs.

Introduction

Although immune checkpoint inhibitors (ICIs) have resulted in meaningful improvements in overall survival (OS) in several tumor types, only a fraction of patients respond to treatment with programmed cell death protein 1 (PD-1) pathway inhibitors. Cancer aneuploidy, an unbalanced number of chromosomes, is a widespread characteristic of human cancer and is associated with somatic mutation rate, expression of proliferative genes, and altered immune signaling.^1–3 Aneuploidy, therefore, is a central feature of tumor evolution and immune evasion.

In NSCLC, programmed death-ligand 1 (PD-L1) tumor proportion score (TPS) is the primary clinically available biomarker of response to ICIs,^4–9 and tumor mutational burden (TMB) may also identify subsets of NSCLC more likely to benefit from treatment with PD-(L)1 inhibitors.^10,11 Whereas cancer aneuploidy positively correlates with TMB,¹² and different arm-level gains or losses affect PD-L1 expression on tumor cells,¹³ little is known about the efficacy of immunotherapy according to aneuploidy burden in advanced NSCLCs.

Recent retrospective analyses have revealed that high levels of focal copy number alterations (CNAs) may correlate with worse clinical outcomes to ICI in solid tumors, including NSCLC.^12,14,15 Moreover, prior work has revealed that low CNA combined with high TMB level identifies patients with advanced NSCLCs more likely to benefit from ICIs.¹⁴ Although these studies suggested an association of focal CNA with outcomes to ICIs, whether chromosomal arm level aneuploidy also correlates with immunotherapy efficacy in NSCLC is still unknown. Compared with focal CNAs, chromosomal arm-level aneuploidy may affect larger regions of the cancer genome, and this impact on treatment outcomes is unclear.¹⁶ Therefore, using two large cohorts of non-squamous NSCLCs which underwent comprehensive genomic profiling, we sought to better define the clinicopathologic, genomic, and immunophenotypic correlates of chromosomal arm-level aneuploidy in NSCLC and evaluate the impact of aneuploidy on clinical outcomes to PD-(L)1 inhibition. To determine the potential mechanisms by which aneuploidy affects outcomes to PD-(L)1 inhibition in NSCLC, we also investigated whether different aneuploidy levels correlate with distinct immunophenotypes in nonsquamous NSCLCs.

Methods

Aneuploidy Assessment

Samples with more than 20% tumor purity and chromosomal arms with adequate sequencing coverage were included to determine arm-level events. A chromosomal arm was considered altered if at least 70% of its territory was either gained or deleted. The number of altered chromosome arms for each tumor was calculated as the sum total of altered arms, ranging from 0 (no arm alterations) to 39 (all arms altered); the chromosome arms included in this calculation included the long and short arms for chromosomes 1 to 12 and 16 to 20, including the long arms for the acrocentric chromosomes 13 to 15 and 21 to 22.¹ The aneuploidy score is defined as the sum of altered arms (0–39). The fraction of chromosomal arm alterations is defined as the number of altered chromosome arms divided by the number of chromosome arms assessed for each sample. Because of the relationship between fraction of chromosomal arm alterations and tumor content,¹⁷ we multiplied the fraction of chromosomal arm alterations by [1 – (tumor content / 100)]^18,19 to calculate the adjusted fraction of chromosomal arm alterations, hereafter referred to as FAA.

Statistical Analysis

The FAA distributions were normalized within different platforms by applying a natural logarithmic transformation followed by standardization toZ scores, as previously described for TMB harmonization across different platforms.^20,21 FAA as a continuous variable was assessed in terms of clinical outcomes in the individual Memorial Sloan Kettering Cancer Center (MSKCC) and Dana-Farber Cancer Institute (DFCI) cohorts. Baseline covariates including age at diagnosis, sex, smoking status, Eastern Cooperative Oncology Group performance status (ECOG PS), line of treatment, and biomarkers of PD-L1 TPS and TMB were adjusted for potential confounding effects. Inverse probability weighting (IPW) was applied to address for potential selection bias owing to PD-L1 TPS missingness. Arm-level deletion and gain events were first assessed with respect to objective response rate (ORR) in the univariable analysis, and those with false discovery rate–adjusted p value less than 0.05 were further evaluated in the association with progression-free survival (PFS) and OS in the multivariable analysis. Receiver operating characteristic curve (ROC) analysis incorporating different biomarkers on ORR was constructed and evaluated in patients with complete information (n = 617) using fivefold cross-validation. Comparisons of the FAA and chromosomal arm-level events were computed using the Mann-Whitney U test or the Kruskal-Wallis test, when appropriate. Linear correlations were evaluated using Pearson’s test, and categorical variables were evaluated using Fisher’s exact test. All statistical analyses were performed using R statistical analysis version 3.6.3.

Detailed methods, including additional analysis, methods used for FAA, PD-L1 and TMB assessment, clinicopathologic, genomic, and immunophenotypic analyses, are reported in the Supplementary Methods.

Results

Clinicopathologic and Genomic Correlates of Aneuploidy in Nonsquamous NSCLC

A total of 2293 patients with nonsquamous NSCLC and comprehensive genomic profiling were identified to explore clinicopathologic correlates of aneuploidy (Table 1). The median age was 66 (range: 22–99) years, 76.0% were current or former smokers, 89.9% had adenocarcinoma, and 56.7% had stage IV disease at the time of next-generation sequencing (NGS). We observed that gains of chromosome arms 5p, 1q, 7p, and 8q and deletion of 8p were the most frequent arm-level events, consistent with prior reports^1,22 (Supplementary Fig. 1). A total of 1998 NSCLCs (87.1%) had at least one arm-level gain or loss per tumor, and of these, 949 (47.5%) had more gains than losses, 816 (40.8%) had more losses than gains, and 233 (11.7%) had an equal number of gains and losses (Supplementary Fig. 2 A). Consistent with other solid tumors,²³ we found that NSCLCs with low aneuploidy were likely to have more chromosome arm gains than losses, whereas tumors with high aneuploidy had more arm losses than gains (Supplementary Fig. 2B).

Table 1.

Clinical and Pathologic Characteristics of the 2293 Patients

Clinical Characteristics	Overall cohort N = 2293
Age, median (range)	66 (22–99)
Sex
Male	899 (39.2)
Female	1394 (60.8)
Smoking status
Current	434 (18.9)
Former	1310 (57.1)
Never	549 (24.0)
Stage at NGS
I	546 (23.8)
II	165 (7.2)
III	281 (12.3)
IV	1301 (56.7)
Histology
Adenocarcinoma	2062 (89.9)
NSCLC NOS	231 (10.1
PD-L1 expression
≥50%	393 (31.1)
1%–49%	446 (35.2)
<1%	427 (33.7)
N.A.	1027
Oncogene driver
KRAS	824 (36.0)
EGFR	482 (21.0)
BRAF	91 (4.0)
MET	76 (3.3)
ALK	53 (2.3)
HER2	40 (1.7)
RET	25 (1.1)
ROS1	16 (0.7)
None identified	686 (29.9)
TMB level, median (range)	9.1 mut/Mb (0–95.6)
Aneuploidy score, median (range)	6 (0–35)
Fraction of chromosomal arm alterations unadjusted for tumor content, median (range)	0.15 (0–0.90)
Tumor content (%), median (range)	50 (25–100)
Fraction of chromosomal arm alterations adjusted for tumor content (FAA), median (range)	0.07 (0–0.57)

PD-L1, programmed cell death -ligand 1; mut/Mb, mutations per megabase; N.A., not available; NGS, next-generation sequencing; NSCLC NOS, NSCLC not otherwise specified; TMB, tumor mutational burden.

In the entire cohort of patients, the median fraction of chromosomal arm alterations was 0.15 (range 0–0.90), unadjusted for tumor content. There was a moderate positive correlation between tumor content and the fraction of chromosomal arm alterations (Pearson R: 0.40, p < 0.0001) (Supplementary Fig. 3). Because tumor content itself is associated with survival in lung cancer,²⁴ to reduce the potential interaction between aneuploidy, tumor content, and clinical outcomes, we adjusted the fraction of chromosomal arm alterations (“adjusted FAA”) by tumor content by multiplying the unadjusted fraction of chromosomal arm alterations by [1 – (tumor content/100)],^18,19 hereafter referred to as FAA (Supplementary Fig. 4). Supplementary Table 1 provides each sample chromosome arm call, number of altered chromosome arms, fraction of chromosomal arm alterations, and tumor content.

To determine clinicopathologic correlates of aneuploidy, we next evaluated the association of smoking status, tumor histology, cancer stage, genotype, and site of biopsy with FAA. There was no difference in the median FAA among current or former smokers compared with never smokers (median 0.07 versus 0.08, respectively; Fig. 1A). The median FAA increased with more advanced cancer stage (stage I: 0.04, stage II: 0.06, stage III: 0.07, stage IV: 0.08, p < 0.0001) (Fig. 1B) and decreased with increasing PD-L1 TPS expression categories: 0.09 for PD-L1 TPS less than 1%, 0.08 for PD-L1 TPS 1% to 49%, 0.05 for PD-L1 TPS of more than or equal to 50% (p < 0.0001; Fig. 1C). When taken as continuous variables, there was a weak negative correlation between FAA and PD-L1 (Pearson R: −0.16, p < 0.0001, Supplementary Fig. 5) and a very weak positive correlation between FAA and TMB (Pearson R: 0.07, p = 0.0005; Fig. 1D). When assessed by site of biopsy (Fig. 1E), lesions from the liver and brain had the highest FAA (median [range] 0.11 [0–0.43] and 0.10 [0–0.35], respectively), whereas tumor samples from the lung and pleura had the lowest FAA (median [range] 0.05 [0–0.57] and 0.08 [0–0.40], respectively). By oncogenic driver status, NSCLCs with RET rearrangements and EGFR mutations and those with no known driver mutation tended to have a higher FAA, whereas those with BRAF, MET, and KRAS mutations had a lower FAA (Fig. 1F). Pairwise comparisons between genomic subsets of NSCLC in terms of median FAA are found in Supplementary Figure 6.

Figure 1.

The median FAA is illustrated according to (A) smoking status, (B) stage, (C) PD-L1 TPS, (D) TMB, and (E) site of disease involvement to the lung, pleura, LN, *others (adrenal, kidney, bowel, bone, and soft tissue), liver, and brain. (F) FAA by oncogenic driver mutation. FAA, adjusted fraction of chromosomal arm alterations; LN, lymph node; PD-L1, programmed death-ligand 1; TMB, tumor mutational burden; TPS, tumor proportion score.

FAA and Outcomes to PD-(L)1 Inhibition in Advanced Nonsquamous NSCLC

We next investigated the association of FAA with clinical outcomes among patients who received a PD-(L) 1 inhibitor alone or in combination with a CTLA-4 inhibitor at DFCI (N = 323) and MSKCC (N = 442). Because FAA was estimated using different platforms at each institution (OncoPanel at DFCI and MSK-IMPACT at MSKCC), we first harmonized the FAA distribution across the two platforms by applying a normal transformation followed by standardization to Z-scores (Supplementary Fig. 7A–D), to merge data for analysis in a larger cohort of immunotherapy-treated patients. Clinical, pathologic, and genomic characteristics of patients in the DFCI and MSKCC immunotherapy-treated cohorts are found in Supplementary Table 2.

The median FAA Z-score was significantly lower among patients with a partial or complete response to ICI compared with those with stable or progressive disease (−0.305 versus 0.089, respectively, p = 0.0001; Fig. 2A) in the combined FAA cohort (DFCI + MSKCC). In univariable analysis, a decreasing FAA, taken as a continuous variable (per unit change), was associated with significantly increased ORR (OR = 1.39, 95% CI: 1.17–1.67, p < 0.0001), durable clinical benefit (DCB) (OR = 1.28, 95% CI: 1.10–1.49, p = 0.002), median progression-free survival (mPFS) (hazard ratio [HR] = 0.87, 95% CI: 0.80–0.93, p = 0.0003), and mOS (HR = 0.89, 95% CI: 0.82–0.97, p = 0.007) in the combined cohort (Supplementary Table 3). Lower FAA Z-scores, when evaluated by tertiles, were also associated with significant improvements in ORR (15.1% in upper tertile versus 23.2% in middle tertile versus 28.4% in lowest tertile, p = 0.001), mPFS (2.5 versus 3.3 versus 4.1 mo, p < 0.0001), and mOS (12.5 versus 13.9 versus 16.4 mo, p = 0.006) to immunotherapy in the combined FAA cohort (DFCI + MSKCC) (Fig. 2B–D) and in the individual DFCI (Supplementary Fig. 8A–D) and MSKCC (Supplementary Fig. 9) cohorts.

Figure 2.

(A) Median FAA Z-score in NSCLCs from patients who experienced a CR or PR or SD or PD as best response to immunotherapy combined cohort (DFCI + MSKCC). (B) ORR in each FAA Z-score tertile in the combined cohort. (C) PFS and (D) OS to immunotherapy according to FAA tertiles. CI, confidence interval; CR, complete response; DFCI, Dana-Farber Cancer Institute; FAA, adjusted fraction of chromosomal arm alterations; MSKCC, Memorial Sloan Kettering Cancer Center; ORR, objective response rate; OS, overall survival; PD, progressive disease; PFS, progression-free survival; PR, partial response; SD, stable disease.

Aneuploidy and Outcomes to Platinum-Doublet Chemotherapy

To dissect the predictive versus prognostic role of aneuploidy, we evaluated the effect of FAA as a continuous variable and by tertiles on outcomes to cytotoxic chemotherapy. Of the 323 immunotherapy-treated patients in our DFCI cohort, 85 patients received platinum-doublet chemotherapy without immunotherapy in the first-line setting, and there was no difference in ORR (OR = 0.96, 95% CI: 0.89–1.04, p = 0.30) or mPFS (HR = 1.01, 95% CI: 0.97–1.05, p = 0.69) to first-line chemotherapy according to FAA as a continuous variable and when examined by tertiles (Supplementary Fig. 10A and B). In a separate nonoverlapping cohort of 916 patients with nonsquamous NSCLC at DFCI who had stage IV disease and never received immunotherapy at any point in their disease course, there was also no difference in OS from the date of advanced disease according to FAA as a continuous variable (HR = 1.50, 95% CI: 0.60–3.79, p = 0.38) or when FAA was examined by tertiles (Supplementary Fig. 11).

Multivariable Analysis

Multivariable Cox proportional hazard regression to identify factors that retained a significant association with immunotherapy efficacy, after adjusting for potential confounders, confirmed that decreasing FAA was independently associated with improved ORR, DCB, mPFS, and mOS (Supplementary Table 4). In addition, increasing PD-L1 expression levels and TMB were both independently associated with improved ORR, DCB, mPFS, and mOS (Supplementary Table 4). These results were confirmed in a sensitivity analysis conducted using IPW to account for PD-L1 missing data (Supplementary Table 5). Multivariable Cox regression and sensitivity analyses using IPW for the individual DFCI and MSKCC cohorts are found in Supplementary Tables 6 and 7, respectively.

Integration of FAA, TMB, and PD-L1 Expression With Clinical Outcomes to Immunotherapy

Because TMB and PD-L1 expression exhibited poor correlation with FAA in our cohort and were each independent predictive factors for immunotherapy response, we explored the potential impact of FAA together with TMB and PD-L1 on efficacy of ICIs. To investigate the relative contribution of different biomarkers in predicting ORR of ICIs in the immunotherapy-treated cohort, an ROC analysis was performed including FAA, PD-L1, and TMB. As individual biomarkers, PD-L1 had the highest area under the ROC curve (AUC = 0.67), followed by TMB (AUC = 0.60) and FAA (AUC = 0.57), for differentiating responders from nonresponders to immunotherapy (Supplementary Fig. 12). Combining FAA with TMB or PD-L1 had an improved prediction of response to ICIs, and use of all three biomarkers (FAA, TMB, and PD-L1) resulted in an AUC of 0.72 (Supplementary Fig. 12).

To find how FAA can be used as an independent predictive biomarker of immunotherapy efficacy in addition to TMB and PD-L1, we performed additional survival analyses with FAA, TMB, and PD-L1. As we have found that decreasing FAA tertiles are associated with improved clinical outcomes, NSCLCs were classified into high and low FAA grouping using the upper tertile and lower tertile, respectively. When dividing TMB groupings into high (≥50th percentile) and low (<50th percentile) categories, the TMB^high FAA^low group had the highest ORR (32.7%), mPFS (5.3 mo), and mOS (24.5 mo), whereas the TMB^low FAA^high group had the lowest ORR (7.7%), PFS (1.8 mo), and OS (7.8 mo) to ICIs (Fig. 3A–C). Pairwise comparison between FAA groups within TMB high and low is found in Supplementary Figure 13A–F. A similar pattern of survival benefit was observed combining FAA and PD-L1 expression when PD-L1 was classified into TPS more than or equal to 50% and TPS less than 50% (Fig. 3D–F). The deleterious effect of high FAA on immunotherapy outcomes was most pronounced in the PD-L1 TPS 1% to 49% subset, with a numerically lower ORR (19.4% versus 9.8%, p = 0.22), a significantly shorter PFS (4.3 versus 2.6 mo; HR = 0.58, 95% CI: 0.37–0.92, p = 0.02), and a numerically shorter mOS (16.3 versus 14.3 mo; HR = 0.70, 95% CI: 0.43–1.14, p = 0.15) compared with NSCLCs with an FAA low and a PD-L1 TPS 1% to 49% (Supplementary Fig. 14A–I).

Figure 3.

Patients were classified into high and low TMB groups if their cancer had a TMB greater than or equal to or less than of the 50th percentile, respectively. Similarly, patients were also classified into high and low FAA groups using the higher tertile and lower tertile, respectively. (A) ORR, (B) PFS, and (C) OS into four groups of the combined FAA level and TMB percentile. (D) ORR, (E) PFS, and (F) OS into four groups of combined FAA level and PD-L1 groups (≥ or <50%). CI, confidence interval; FAA, adjusted fraction of chromosomal arm alterations; ORR, objective response rate; OS, overall survival; PD-L1, programmed death-ligand 1; PFS, progression-free survival; TMB, tumor mutational burden.

Chromosomal Arm-Level Events and Impact on Outcomes to Immunotherapy

In the combined cohort of 765 patients with non-squamous NSCLC treated with ICIs, to determine which chromosomal arm events might be affecting the FAA findings, we also analyzed the association of chromosomal arm-level events and response to immunotherapy. Compared with immunotherapy responders, lack of response to immunotherapy was significantly associated with chromosome 9p loss (OR = 0.22; Q = 0.0002) and chromosome 1q gain (OR = 0.43; Q = 0.002), after false discovery rate adjustment (Fig. 4A).

Figure 4.

(A) Volcano plot illustrating chromosomal arm-level events associated with response to immunotherapy by multivariate logistic regression (by Q-value after correction for multiple comparisons). (B) ORR, (C) PFS, and (D) OS to immunotherapy in NSCLCs by 9p status. (E) ORR, (F) PFS, and (G) OS to immunotherapy in NSCLCs by 1q status. *Actual p values for (B) (p = 8.5e-09), (C) (p = 4e-08), and (D) (p = 1e-07). CI, confidence interval; ORR, objective response rate; OS, overall survival; PFS, progression-free survival.

We then explored the impact of 9p status on clinical outcomes to immunotherapy. NSCLCs with 9p loss were enriched for higher TMB levels, KRAS wild-type status, lower PD-L1 expression levels, and use of ICI in the greater than or equal to second-line setting compared with NSCLCs with no 9p loss (Supplementary Table 8). Chromosome 9p loss was also more prevalent with increasing disease stage (frequency was 12.7% versus 19.1% versus 20.6% versus 28.2% in stage I, II, III, and IV NSCLC, respectively; p < 0.0001; Supplementary Fig. 15). Other baseline clinical characteristics were balanced in terms of age, sex, ECOG PS, and smoking status (Supplementary Table 8). Patients whose tumors had 9p loss compared with those with no loss had significantly lower ORR (7.8% versus 28.1%, p < 0.0001) and a significantly shorter mPFS (2.3 versus 4.1 mo, HR = 1.70, 95% CI: 1.41–2.06, p < 0.0001) and mOS (9.6 versus 18.0 mo, HR = 1.72, 95% CI: 1.40–2.10, p < 0.0001) to immunotherapy, respectively (Fig. 4B and C). Similar results were observed in the individual DFCI and MSKCC cohorts (Supplementary Fig. 16A–F).

We next integrated 9p status with other predictors of immunotherapy efficacy, such as TMB and PD-L1 expression, to determine whether using multiple biomarkers together could strengthen the ability to predict response to immunotherapy. First, when dividing TMB into high (≥50th percentile) and low (<50th percentile) groupings, the TMB^high/no 9p loss group had the highest ORR (33.6%), mPFS (5.4 mo), and mOS (26.5 mo), whereas the TMB^low/9p loss group had the lowest ORR (3.3%), PFS (2.1 mo), and OS (7.8 mo) to ICIs (Supplementary Fig. 17A–C). Second, because cancers with chromosome 9p loss have lower levels of PD-L1 expression, to determine whether 9p loss was predictive of lower immunotherapy efficacy independent of PD-L1 expression, we next evaluated the impact of 9p status in tumors with PD-L1 TPS categories of less than 1%, 1% to 49%, and more than or equal to 50%. Although we observed a negative impact of 9p loss across each PD-L1 TPS category, the deleterious effect of 9p loss on immunotherapy outcomes was most pronounced in the PD-L1 TPS more than or equal to 50% subset. A concurrent 9p loss in tumors with PD-L1 TPS more than or equal to 50% resulted in a significantly lower ORR (6% versus 43%, p < 0.0001), a significantly shorter mPFS (2.6 versus 6.4 mo, HR = 1.94, 95% CI: 1.32–2.86, p = 0.0006), and a significantly shorter mOS (14.3 versus 30.2 mo, HR = 1.98, 95% CI: 1.32–2.96, p = 0.0008) compared with NSCLCs with intact 9p arm and a high PD-L1 expression (Supplementary Fig. 18A–H). An ROC analysis of 9p status, PD-L1 expression, and TMB revealed that the integration of all three of these biomarkers together improved prediction of response to immunotherapy in NSCLC (AUC = 0.74), as found in Supplementary Figure 19.

When looking at 9p status among lung cancers with specific genotypes, among KRAS-mutant NSCLCs, 9p loss resulted in lower ORR (7% versus 25%, p = 0.01), shorter mPFS (HR = 1.82, 95% CI: 1.29–2.52, p = 0.0005), and shorter mOS (HR = 1.94, 95% CI: 1.36–2.76, p = 0.0002) compared with KRAS-mutant NSCLCs with no 9p loss. Supplementary Figure 20A–C provides outcomes across oncogenic drivers, including KRAS, EGFR, and other oncogenic drivers (BRAF, MET, ALK, RET) according to 9p status. In addition, Supplementary Figure 21A–F illustrates response rate to ICI in NSCLCs without targetable alterations by 9p status.

Our chromosomal arm analysis also revealed that chromosome 1q gain was associated with decreased outcomes to immunotherapy. Baseline clinicopathologic characteristics of the 1q gain and no 1q gain groups were balanced between the two cohorts in terms of age, sex, smoking status, ECOG PS, and TMB level. We observed higher PD-L1 expression levels, a higher proportion of patients receiving ICI in the first-line setting, and a lower prevalence of KRAS mutations in tumors with no 1q gain versus with 1q gain (Supplementary Table 9). For the 1q clinical outcomes analysis, there were significantly lower ORR (13.8% versus 27.8%, p < 0.0001) and significantly shorter mPFS (2.5 versus 3.7 mo, HR = 1.39, 95% CI: 1.18–1.63, p < 0.0001) and mOS (12.5 versus 16.4 mo, HR = 1.39, 95% CI: 1.14–1.63, p = 0.0005) in tumors with 1q gain versus those with no gain, respectively (Fig. 4D and E), and this observation was also noted when looking at the DFCI and MSKCC cohorts individually (Supplementary Fig. 22). When assessing the impact of 1q gain across clinically relevant PD-L1 TPS levels (<1%, 1%–49%, and ≥50%), we saw no differences in clinical outcomes to immunotherapy in the combined cohort (Supplementary Fig. 23A–H).

We next evaluated the effect of 9p loss and 1q gain on outcomes to cytotoxic chemotherapy. Neither chromosomal arm 9p loss nor 1q gain affected ORR or mPFS to first-line chemotherapy (Supplementary Fig. 24A–D). In addition, among patients with metastatic nonsquamous NSCLC at DFCI who never received immunotherapy at any point in their disease course, there was also no difference in OS from the date of advanced disease according to chromosomal 9p (HR = 1.13, 95% CI: 0.92–1.39, p = 0.24) or 1q status (HR = 1.10, 95% CI: 0.92–1.33, p = 0.29) (Supplementary Fig. 25A and B).

In multivariate analysis adjusting for key variables (such as TMB, PD-L1, and FAA), chromosome 9p loss retained significant impact on ORR (HR = 0.25, 95% CI: 0.11–0.50, p < 0.001), DCB (HR = 0.43, 95% CI: 0.24–0.74, p = 0.003), mPFS (HR = 1.49, 95% CI: 1.18–1.89, p = 0.001), and mOS (HR = 1.47, 95% CI: 1.14–1.89, p = 0.003) compared with no 9p loss (Supplementary Table 10A), and sensitivity analysis using IPW also confirmed our findings (Supplementary Table 10B). By contrast, as chromosome 1q gain is associated with lower PD-L1 expression, NSCLCs with 1q gain did not retain a significant association with immunotherapy efficacy after adjustments in multivariable analysis (Supplementary Table 11).

Impact of Aneuploidy and Chromosome 9p Status on the Tumor Immune Microenvironment

To evaluate the potential mechanism by which NSCLCs with high FAA and or chromosome 9p loss are less likely to respond to immunotherapy, we interrogated immune cell subsets of nonsquamous NSCLCs using a multiplexed immunofluorescence (mIF) platform (ImmunoProfile) which quantifies CD8, FOXP3, PD-1, and PD-L1, in a separate cohort of 404 nonsquamous NSCLC samples (early stage, n = 231; advanced stage, n = 173). With decreasing FAA levels, moving from upper to middle to lower tertiles, there was a significant enrichment in intratumoral and total CD8+ cells, PD-1+ cells, double-positive PD-1+ CD8+ T cells, and FOXP3+ T cells (Fig. 5). There were no significant differences in CD8+ to FOXP3+ T-cell ratio or PD-L1 positivity on immune cells according to FAA tertiles, and there were significant differences in PD-L1 positivity on tumor or total cells according to FAA tertiles (Fig. 5). In addition, there was a significant negative correlation on a continuous scale between increasing FAA levels and immune cell subsets, including intratumoral and total CD8+ T cells, PD-1+ immune cells, and FOXP3+ T cells (Supplementary Fig. 26). No correlation between tumor and immune cells PD-L1+ with FAA as continuous variable was observed (Supplementary Fig. 27A–C).

Figure 5.

(A) CD8+, (B) PD-1+, (C) CD8+ to FOXP3+ ratio, (D) PD-1+ CD8+, and (E) FOXP3+ cells/mm² by FAA tertiles. (F) PD-L1 distribution in nonsquamous NSCLCs by FAA tertiles. FAA, adjusted fraction of chromosomal arm alterations; NS, not significant; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1.

We next evaluated whether nonsquamous NSCLCs harboring 9p loss or 1q gain also had a distinct tumor immunophenotype. Tumors with 9p loss had significantly lower tumor-associated and total PD-1+ immune cells, double-positive PD-1+ CD8+ T cells, and FOXP3+ T compared with those without loss (Supplementary Fig. 28A–E). There was no difference in PD-L1 positivity on tumor or immune cells according to 9p status (Supplementary Fig. 28F). According to chromosome 1q status, although there were significantly lower PD-L1 levels on tumor and immune cells in NSCLCs with 1q gain versus no 1q gain, we did not observe a difference in immune cell subsets (Supplementary Fig. 29A–F).

Because the small sample size limited the analyses previously discussed for chromosomal arm-level analyses, we leveraged data from The Cancer Genome Atlas lung adenocarcinoma data set to perform cell-type enrichment analysis by computational deconvolution of gene expression data into tumor-associated cell populations using xCell.²⁵ We identified a significant negative correlation between increasing FAA as a continuous variable and tumor-associated CD4+, CD8+, and dendritic cells (Supplementary Fig. 30A–C). In addition, tumors with 9p loss also had lower proportions of tumor-associated lymphocyte B+, CD8+, CD4+, and dendritic cells compared with tumors without 9p loss (Fig. 6A). Last, tumors with 1q gain had lower levels of tumor-associated lymphocyte B+ and dendritic cells compared with NSCLCs without 1q gain (Fig. 6B).

Figure 6.

Deconvolution of RNA sequencing data from the NSCLC TCGA data set into tumor-associated immune cells, revealing cell types that are differentially enriched in NSCLCs by (A) chromosomal arm 9p status (no loss versus loss) and (B) chromosomal arm 1q status (no gain versus gain). TCGA, The Cancer Genome Atlas.

Discussion

Although PD-(L)1 inhibitors have improved clinical outcomes for patients with advanced NSCLC, PD-L1 expression alone is not sufficient to fully discriminate responders from nonresponders to immunotherapy, highlighting the need for additional biomarkers of immunotherapy efficacy. In this study, we report that tumor aneuploidy and chromosome 9p loss are associated with unique clinicopathologic, genomic, and immunophenotypic characteristics among patients with nonsquamous NSCLC and reveal that lower aneuploidy is an independent predictor for improved clinical outcomes in patients with metastatic disease receiving PD-(L)1 blockade. We also found that loss of chromosomal arm 9p is associated with diminished immunotherapy efficacy independent of aneuploidy and PD-L1 expression level. Multivariable analysis confirmed that 9p loss was the strongest arm-level alteration associated with ICI response and survival outcomes, independent of overall aneuploidy level, which was also significantly associated with ICI outcomes. These findings provide important insights to guide treatment decision-making, clinical trial design, and interpretation.

Consistent with previous studies, we also found that increasing FAA level is associated with more advanced cancer stage and metastatic sites such as the liver and brain, which might contribute to explain the decreased likelihood of response to ICIs in these sites of disease in NSCLC.²⁶ Importantly, we also noted aneuploidy differences among genomically defined subsets of non-squamous NSCLC with the lowest aneuploidy among tumors harboring activating mutations in BRAF, KRAS, or MET exon 14 skipping alterations. By contrast, cancers with RET and ALK rearrangements and mutations in EGFR were found to have the highest aneuploidy burden. Although oncogenic drivers are associated with genomic instability,²⁷ it seems that specific pathways such as the epidermal growth factor family and chromosomal rearrangements are more likely to generate aneuploidy than RAS/RAF (BRAF, KRAS) and might explain in part the decreased activity of immunotherapy in such cases.^28,29

Whereas our analysis and previous work¹ have revealed that aneuploidy in NSCLC positively correlates with TMB, whether PD-L1 expression in NSCLC also associates with aneuploidy burden was unknown. In this study, we observed that aneuploidy levels decreased with the increase of clinically relevant PD-L1 TPS cutoffs of less than 1%, 1% to 49% and more than or equal to 50%. Although a recent analysis of tumor cell lines has revealed that induction of chromosomal instability activates the STING pathway as a result of cytosolic DNA fragment accumulation and leads to expression of PD-L1 on tumor cells,²³ it has been suggested that increased aneuploidy suppresses the innate immune signaling machinery, antigen presentation, and PD-L1 expression in cancer, leading to a more immunologically cold tumor microenvironment,³⁰ which is in line with our findings.

Our study has implications for treatment decisions and clinical trial design and interpretation. Currently, multiple first-line options have been approved for patients with NSCLC lacking actionable genomic alterations, including PD-(L)1 monotherapy and the combination of PD-(L)1 inhibition with chemotherapy. As aneuploidy seems to be an independent predictor of immunotherapy efficacy, it may help to identify patients who can benefit from PD-(L)1 blockade alone, sparing the side effects of cytotoxic chemotherapy. In addition, as there are several ongoing and planned randomized immunotherapy clinical trials, our results indicate that aneuploidy should be considered as stratification factor to ensure that differences in outcomes are not influenced by imbalances between groups in such factor.

Along with PD-L1 expression, higher TMB is also associated with improved clinical outcomes to PD-1 blockade in patients with advanced NSCLC.^10,11,21 In our analysis, aneuploidy and TMB can be combined to improve prediction to immunotherapy response for patients diagnosed with having metastatic NSCLC. A high mutational load and low aneuploidy may be associated with improved benefit to ICIs because TMB serves as a proxy of neoantigen load from somatic nonsynonymous mutations with a higher likelihood of triggering a T-cell response, whereas increased cancer aneuploidy strongly associates with reduced expression of markers of cytotoxic infiltrating immune cells, especially CD8⁺ T cells, and reduced or inefficient antigen presentation.^2,27 In addition, although high PD-L1 expression correlates to immunotherapy efficacy in NSCLC, patients with negative or low PD-L1 expression levels may also respond to ICIs, highlighting the need for identifying additional biomarkers. Here, we found that low FAA level correlated with improved ICI efficacy in tumors with higher TMB level. The possibility that combination assessment of PD-L1, TMB, and FAA levels is a superior predictive biomarker for immunotherapy response in NSCLC warrants further study.

When analyzing correlates of response at the chromosome level, loss of 9p in our study was associated with impaired efficacy of immunotherapy in NSCLC. Among genes of interest on chromosome 9p are CD274 (encoding PD-L1), located at the 9p24.1 locus, and CDKN2A, located at 9p21.3. Gene copy loss of CD274 has been associated with reduced PD-L1 expression and impaired outcomes to immunotherapy in NSCLC.^13,31 Similarly, lower transcript levels of CDKN2A correlate with shorter PFS and OS to immunotherapy in urothelial carcinoma and NSCLC,³² and loss of 9p21.3 is associated with a cold tumor immune microenvironment and primary resistance to PD-(L)1 inhibitor monotherapy,^15,33 but not to chemoimmunotherapy, in NSCLC.¹⁵ Mechanistically, both gene-level and complete chromosome 9p loss down-regulate factors regulating immune cell recruitment and T-cell activation and up-regulate immune suppressive pathways to enable primary resistance to immunotherapy.^15,34 A recent publication in patients with advanced HPV-negative head and neck cancers also observed that tumors harboring 9p21.3 loss were associated with diminished CD8+ T cell infiltration, largely driven by 9p arm-level loss effects on decreased expression of immune-regulatory genes, and lower efficacy of anti–PD-1 therapy compared with tumors without chromosome 9p loss.³⁴ Using multiplexed immunofluorescence and computational cell type deconvolution, we also found that NSCLCs display distinct immunophenotypes according to chromosome 9p status.

Our findings reveal that arm-level loss of chromosome 9p significantly impairs outcomes to immunotherapy in NSCLC even after adjusting for other predictive biomarkers, such as PD-L1 and TMB. Therefore, assessment of 9p loss might serve as a genomic biomarker to identify patients who are less likely to benefit from PD-1 inhibition, particularly among patients with PD-L1 TPS more than or equal to 50%. An ongoing question is whether to use single-agent PD-(L)1 inhibition or a PD-(L)1 inhibitor plus chemotherapy in patients with NSCLC and a PD-L1 level of more than or equal to 50% because there has been no direct comparison between the two regimens in this population. Although our observation suggests that a concurrent 9p loss in tumors with high PD-L1 expression might have unfavorable outcomes to anti–PD-(L)1 blockade alone, additional studies are needed to investigate whether this population may benefit from chemoimmunotherapy strategies.

Limitations of this study include the retrospective nature of the analysis of cases from two academic medical centers. Furthermore, factors such as sample tumor purity limited analysis for chromosomal arms with low coverage, and lack of a standard method to assess aneuploidy from NGS panels is challenging. Nonetheless, to ensure adequate power for the analysis, we used a tool highly concordant with calls from exome-based methods and excluded tumors with low purity (≤20%).¹⁷ In addition, as aneuploidy is not an established biomarker for immunotherapy patient selection in lung cancer, further data from prospective and multicenter clinical trials are needed to validate the association with ICIs and identify an optimal cutoff to discriminate cases with the greatest likelihood of responding to PD-(L)1 inhibition.

In conclusion, our report of 2293 cases advances our understanding of how aneuploidy burden and chromosome 9p loss correlate with clinicopathologic, genomic, and immunophenotypic factors in lung cancer and strengthen the importance of these biomarkers in predicting immunotherapy efficacy among nonsquamous NSCLC.

Data Availability Statement

Data are available on reasonable request. The data that support the finding of our study are available on request from the corresponding author.