Approval for this study, including waiver of informed consent, was obtained from the Western Institutional Review Board Copernicus Group (WCG protocol # 1340120).

We retrospectively analyzed clinical CGP testing data from NSCLC FFPE tumor biopsy specimens submitted for CGP testing at a reference laboratory (OmniSeq/Labcorp, Buffalo, NY) during standard clinical care from June 2021 - June 2024. Specimens were collected from 647 provider facilities across the United States and Alaska. Any cases that were ultra-hypermutated (TMB > 200 mutations/Mb) and did not have adenocarcinoma or squamous cell carcinoma histology were excluded from the study. The total number of cases included in the study was 7,606 (5,523 with adenocarcinoma and 2,083 with squamous cell carcinoma).

DNA and RNA were co-extracted from FFPE tissue specimens and submitted for library preparation and sequencing using the hybrid-capture-based TSO 500 assay (Illumina, San Diego, CA, USA) as part of OmniSeq^® INSIGHT (OmniSeq/Labcorp, Buffalo, NY, USA). OmniSeq^® INSIGHT is a comprehensive genomic and immune profiling assay performed in a laboratory accredited by the College of American Pathologists (CAP) and certified by the Clinical Laboratory Improvement Amendments (CLIA) (26). OmniSeq^® INSIGHT is an NGS-based in vitro diagnostic device for detecting genomic variants, signatures, and immune gene expression in FFPE tumor tissue. Within the OmniSeq^® INSIGHT framework, DNA sequencing with hybrid capture (via TSO 500) detects small nucleotide variants (SNVs) in exonic regions of 523 genes (single and multi-nucleotide substitutions, insertions, and deletions), copy number variants (CNVs) in 59 genes (gains and losses), as well as analysis of microsatellite instability (MSI) and TMB genomic signatures. RNA sequencing with hybrid capture (via TSO 500) detects fusions and splice variants in 55 genes. Variant annotation is performed using the GenomeOncology Precision Oncology Platform (GenomeOncology, Cleveland, OH, USA). Only genomic alterations annotated as known pathogenic were analyzed in the current study.

For all tumor types, PD-L1 expression on the surface of tumor cells was measured by Dako PD-L1 IHC 22C3 pharmDx (Agilent, Santa Clara, CA). A board-certified anatomical pathologist scored expression according to published guidelines (27) as tumor proportion score (TPS), the percentage of tumor cells with positive linear membranous staining.

Statistical analysis and plot generation were performed in R v 4.4.1 (https://www.r-project.org/). All plotting was performed using the ggplot2 v 3.5.1 package (https://ggplot2.tidyverse.org/) and various packages to extend the ggplot2 functionality (ggpubr, ggtext, GGally, ggraph).

To assess differences between adenocarcinoma and squamous cell carcinoma NSCLC in patient and tumor characteristics and genomic alteration prevalence, Fisher’s exact test (via `fisher.test` function) or linear regression (via `lm` function) was performed for categorical or quantitative variables, respectively.

To assess differences in the distributions of immunotherapy markers in tumors with guideline-indicated alterations, non-guideline-indicated alterations, and tumors without alterations in any currently known driver genes, we tested for differences in immunotherapy markers between the different alteration groups in known NSCLC driver genes (ALK, EGFR, MET, BRAF, ROS1, KRAS, ERBB2). TMB was treated as a quantitative variable (mutations/Mb), and PD-L1 22C3 IHC TPS was treated as a categorical variable with levels of “Negative (<1%)”, “Low (1 - 49%)”, and “High (≥50%)”. Tests for differences in TMB and PD-L1 were performed using linear regression (via the `lm` function) and penalized likelihood ratio tests (via `logistf` and `logistftest` functions from the logistf v 1.26.0 package), respectively. Tests were adjusted for NSCLC histology by adding this variable as a covariate in the analysis. P-values were corrected for multiple testing within each driver gene using the `p.adjust` function with `method=“ bonferroni”`. Distributions of TMB and PD-L1 levels were plotted for alteration groups within each gene using ggplot2.

To detect co-occurrence or mutual exclusivity of genomic alterations, pairwise Fisher’s exact tests were performed between each pair of gene-level summarized genomic alterations using the `fisher.test` function. P-values were multiple testing corrected using the Benjamini-Hochberg false discovery rate (FDR) method as implemented in the `p.adjust` function when `method=“BH”`. Multiple testing corrected FDR q-values < 0.05 were considered significant. The proportion of co-occurrence between genomic alterations was also calculated to discern which pairs had complete exclusivity (i.e., the proportion of co-occurrence = 0). Only genomic alterations detected in at least 1% of patient tumors were included in the analysis.

To model higher-order co-occurrences or mutual exclusivities and detect gene modules of tightly co-occurring genomic alterations, network analysis was applied to co-occurrence results to generate a network of pairwise interactions between gene-level summarized genomic alterations. Co-occurrence analysis results were filtered for significant associations and then imported into the igraph v 1.3.5 package (https://igraph.org/) to create an igraph network. The degree and weighted degree for each node of the network (i.e., the genomic alterations and immunotherapy markers) were calculated using the `degree` and `strength` functions, respectively, from the graph specifying `weights` of the weighted degree to be the absolute odds ratio (OR) from Fisher’s exact test. The absolute OR was taken to be the OR if the OR was > 1 or the inverse of the OR if the OR < 1. Gene modules were defined within the network by how tightly their genomic alterations associated with one another using the community detection algorithm implemented in the `cluster_infomap` function in igraph, specifying the `e.weights` to be the ORs from Fisher’s exact test and the `v.weights` to be the weighted degree of each node. The R implementation of the Force Atlas 2 algorithm (from the ForceAtlas2 v 0.1 package, https://github.com/analyxcompany/ForceAtlas2) was used to position nodes in the network. The network was plotted using the graph v 2.1.0 package (https://bioconductor.org/packages/release/bioc/html/graph.html) specifying the Force Atlas 2 coordinates as the positions for the nodes.

To assess if a dose-dependent relationship existed between histology, immunotherapy markers (TMB, PD-L1), and an increasing number of genomic alterations within a gene module, differences in each variable were tested between NSCLC tumors with none, one, or ≥2 genomic alterations within genes co-occurring in each network-derived gene module.

We retrospectively analyzed real-world CGP data from 7,606 patients with NSCLC who received CGP testing via TSO 500 and PD-L1 IHC at a reference laboratory during standard clinical care. Testing was performed on FFPE tumor biopsy specimens between 2021 and 2024. Specimens were collected from 647 provider facilities across the United States and Alaska. Patient and patient tumor characteristics in the full cohort and when stratified by histology subtype are provided in Table 1 . Most of the cohort was lung adenocarcinomas (73%), with squamous cell carcinomas comprising the remaining third (27%). The mean age of patients was 70.7 ± 9.7 years, with 49% males and 51% females. A significantly higher frequency of females had tumors with adenocarcinoma, while males had significantly more tumors with squamous cell histology (P=4E-35; Table 1 ). Tissue specimens for this study came from the primary tumor site in 70.4% of cases and the advanced or metastatic site in 18.9% and 10.8% of cases, respectively, with adenocarcinoma cases being significantly enriched in the latter (P<3E-21; Table 1 ). For patients with known staging information (N=2,631), 16% were stage III and 66.2% were stage IV. The mean number of detected pathogenic alterations for all NSCLC cases was 3.9 ± 2.4 with adenocarcinoma cases having significantly less on average than squamous cell carcinoma (3.8 ± 2.3 vs 4.2 ± 2.5, P=7E-8; Table 1 ). The mean TMB level for all NSCLC cases was 11 ± 11.1 mutations/Mb, with 12.5% of cases having a TMB of ≥20 mutations/Mb, 30.2% having a TMB of 10-19 mutations/Mb, and 57.2% having a TMB of <10 mutations/Mb. The mean PD-L1 TPS score among tumors was 22.2 ± 31.1, with 25.1% of cases having a TPS of ≥50%, 37.7% having a TPS of 1 - 49%, and 37.2% having a TPS of <1%. Most cases were microsatellite stable, with only 23 MSI high cases (0.3%). Tumors with adenocarcinoma histology tended to have more cases with TMB of <10 mutations/Mb (P=1E-22) and negative PD-L1 (P=5E-8) compared to squamous cell carcinoma cases, however, showed a slight, but significant, enrichment of tumors with high/very high levels of these markers (P<0.02; Table 1 ).

Of the 7,606 samples tested, 44% were positive for therapy-associated genomic alterations, and 77% were positive for clinical trial-associated genomic alterations or genomic alterations with an accepted therapy in another tumor type ( Figure 1A ; Table 1 ). Adenocarcinomas harbored significantly more therapy- (57.5%) and clinical trial- (79.4%) associated variants than squamous cell carcinomas (7.3% and 72.1%, respectively) (P<3E-11; Table 1 ). The most prevalent genomic alterations detected in NSCLC tumors were substitutions (79%) followed by small insertions/deletions (Indels; 45%), splice site variants (26%), gene fusions (14%), larger CNVs (12%), and other complex types of SNVs (6%) ( Figure 1A ).

Pathogenic alterations in 296 genes were detected in ≥0.1% of tested tumors with alterations in TP53, KRAS, or EGFR being detected in >10% of tested tumors ( Figure 1B ; Supplementary Table S1 ). Adenocarcinoma cases had a higher prevalence of SNVs, CNVs, and/or fusions/rearrangements in 28 genes including therapy-associated genes such as KRAS (37.5% vs 4.6%), EGFR (17.1% vs 1.3%), BRAF (5.2% vs 1%), ERBB2 (1.8% vs 0.8%), RET (0.9% vs 0.3%), and ROS1 (1.2% vs 0.1%) ( Figure 1B ; Supplementary Table S1 ). Squamous cell carcinomas had a significantly higher prevalence of genomic alterations in 43 genes including TP53 (81.9% vs 48.7%), CDKN2A (16.3% vs 4.4%), PIK3CA (11.6% vs 4.2%), RB1 (6.6% vs 4.1%), and NFE2L2 (13% vs 1.2%) ( Figure 1B ; Supplementary Table S1 ). FDA-emerging biomarkers were present in both histologies, with genes like STK11, KEAP1, and others more frequently altered in lung adenocarcinoma ( Figure 1B ; Supplementary Table S1 ). The most prevalent alterations identified by RNA sequencing included fusions or exon skipping mutations involving therapy-associated genes MET (2.9%), ALK (1.7%), ROS1 (0.9%), and RET (0.7%), all of which were detected at a significantly higher frequency in adenocarcinoma cases ( Figure 1C ; Supplementary Table S1 ). Of note, squamous cell carcinoma cases had enrichment of fusions involving FGFR1 (0.9% vs 0.3%), FGFR3 (0.8% vs 0.2%), and PIK3CA (0.5% vs <0.01%) ( Figure 1C ; Supplementary Table S1 ).

In total, 399 specific mutations were detected at a prevalence of ≥0.1% among tested NSCLC tumors ( Supplementary Table S2 ). The most prevalent single variants detected were therapy-associated substitutions in KRAS including G12C (11.3%), G12V (5.8%), G12D (4.2%), and G12A (2%). First-line therapy-associated variants in EGFR and BRAF were also among the topmost prevalent single variants including the EGFR L858R substitution (4.6%) and E746-A750 deletion (3.3%) and the BRAF V600E substitution (1.5%). These and 28 other mutations were detected at a significantly higher prevalence in adenocarcinoma cases including therapy-associated MET exon 14 skipping mutations (3.8% vs 0.8%), EML4-ALK fusions (2% vs 0%), and KIF5B-RET fusions (0.5% vs 0.1%) ( Supplementary Table S2 ). Squamous cell carcinoma cases were enriched for 54 specific mutations compared to adenocarcinoma cases including several substitutions in PIK3CA, TP53, and NFE2L2 that had current clinical trials in NSCLC or had an approved therapy in another tumor type ( Supplementary Table S2 ). The most significantly enriched mutations, however, were amplifications of PIK3CA (2.9% vs <0.01%) and fibroblast growth factor (FGF)-associated genes including FGF4 (5% vs 0.7%), FGF3 (5.2% vs 0.8%), CCND1 (5.2% vs 0.8%), FGF19 (4.7% vs 0.8%), and FGFR1 (3.8% vs 0.2%), some of which (PIK3CA, FGFR1) had current clinical trials in NSCLC ( Supplementary Table S2 ).

We next examined the association between TMB and PD-L1 expression and oncogenic driver mutations in known driver genes of NSCLC including ALK, EGFR¸ MET, BRAF, ROS1, ERBB2, and KRAS. To detect associations, we tested for differences in TMB and PD-L1 between tumors with known driver mutations compared to tumors with non-driver mutations in the same gene and tumors without any alterations in any known driver genes of NSCLC (driver gene negative) ( Figure 2 ; Supplementary Figure S1 ; Supplementary Tables S3 , S4 ).

Overall, significantly lower TMB levels were observed across all tumors with therapy-associated driver mutations and significantly higher PD-L1 levels were observed in tumors with ALK, MET, BRAF, or ROS1 driver mutations ( Figure 2 ; Supplementary Table S4 ). Tumors with EML4-ALK and other ALK fusions had a significantly lower median TMB (3.1 mut/Mb; 3.9 mut/Mb) compared to those with non-fusion ALK alterations (11.7 mut/Mb) and driver gene negative tumors (10.2 mut/Mb) yet had a significantly higher frequency of high PD-L1 TPS scores (scores ≥50%) than the other tumor groups (36% vs ≤23%) ( Figure 2A ). Tumors harboring therapy-associated EGFR alterations had significantly lower TMB (median 4.6 mut/Mb) and frequency of high PD-L1 scores (15%) compared to tumors with other EGFR alterations and driver-gene negative tumors (≥9 mut/Mb and ≥19%) ( Figure 2B ). Tumors with MET exon 14 skipping mutations had a significantly lower TMB (5.5 mut/Mb) compared to tumors with MET amplifications, other MET alterations, and driver gene negative tumors (≥8.5 mut/Mb) ( Figure 2C ). Tumors with MET exon 14 skipping and MET amplifications had significantly higher frequencies of high PD-L1 scores (46% and 70%) compared to tumors with other MET alterations and driver gene-negative tumors (≤29%) ( Figure 2C ). Of note, MET-amplified tumors had the highest frequency of high PD-L1 scores out of all tested tumor groups, with most of these tumors being detected as PD-L1 high ( Figure 2C ). Tumors with BRAF V600E mutations had significantly lower TMB scores (4.6 mut/Mb) compared to other BRAF alterations and driver gene negative tumors (≥10.2 mut/Mb) yet had a significantly higher frequency of high PD-L1 scores (46%) than the other tumor groups (≤28%) ( Figure 2D ). Similarly, tumors with ROS1 fusions had significantly lower TMB (4.6 mut/Mb) than driver gene-negative tumors, but significantly higher frequency of high PD-L1 scores (29%) ( Figure 2E ). Tumors with ERBB2 exon 20 insertions (Y772-A775dup) had significantly lower TMB (5 mut/Mb) than driver gene-negative tumors and tumors with ERBB2 amplifications or other therapy-associated ERBB2 alterations (≥7 mut/Mb) ( Figure 2F ). Tumors with ERBB2 amplifications or other therapy-associated ERBB2 alterations had significantly higher frequencies of PD-L1 negative scores (≥49%) compared to driver gene negative tumors (38%) ( Figure 2F ).

Comparing different KRAS alterations, tumors with KRAS G12D (6.2 mut/Mb), G12A (8.6 mut/Mb), G12V (8.6 mut/Mb), and Q61H (7 mut/Mb) alterations had the lowest TMB among tumors with KRAS alterations (<10 mut/Mb) and significantly lower TMB than all other tested KRAS tumor groups (≥10 mut/Mb) ( Supplementary Figure S1A ; Supplementary Table S4 ). Tumors with KRAS G13C and G13D had the highest frequency of PD-L1 scores ≥50% (≥41%) among tumors with KRAS alterations and had a significantly higher frequency than tumors positive for Q61H and driver gene negative tumors (≤23%) ( Supplementary Figure S1B ; Supplementary Table S4 ). Tumors with KRAS Q61H had the lowest frequency of high PD-L1 scores (18%) and was the only group of KRAS-altered tumors that did not exhibit significantly different PD-L1 levels from driver gene-negative tumors ( Supplementary Figure S1B ; Supplementary Table S4 ).

Evaluation of co-occurring genomic alterations in NSCLC revealed co-mutational patterns including alterations with and without associated targeted therapies and clinical trials ( Figure 3A ; Supplementary Table S5 ). Genes with therapy-associated alterations were mainly mutually exclusive, with KRAS and EGFR having the most significant mutual exclusivities with other genes (23 each) ( Figure 3A ; Supplementary Table S5 ). In contrast, genes with clinical trial-associated alterations or alterations with therapy in another tumor type had more significant co-occurrences, with TP53 having the most of all tested genes (23 co-occurring genes) ( Figure 3A ; Supplementary Table S5 ). The highest proportion of co-occurrence (i.e., percent of tumors that harbored genomic alterations from a pair of genes) was observed between genes involved in the FGF pathway (FGF3/4/19, CCND1; 80% - 87%) followed by MDM2 and CDK4 (27%) and STK11 and KEAP1 (15%) ( Supplementary Table S5 ).

To further investigate significantly co-occurring and mutually exclusive genomic alterations (FDR q-value < 0.05), we performed network analysis followed by gene module detection to model higher-order interactions between genomic alterations and detect modules of tightly associated genes ( Figure 3B ). Network analysis highlighted the mutual exclusivity of essential driver genes in NSCLC including EGFR, KRAS, BRAF, ALK, and MET ( Figure 3B ). A tight cluster was formed in the network between CCND1, FGF19, FGF4, and FGF3, which mirrors our previous observation of them having the highest proportions of co-occurrence than any other groups of genes ( Figure 3B ). Genes in the network were grouped into 13 modules of tightly associating genes (including co-occurring and mutually exclusive genes) ( Figure 3B ). Most gene modules included a mix of genes with clinical trial-associated alterations or alterations with therapy in another tumor type and genes with pathogenic alterations of unknown clinical significance. Interestingly, if a gene module included a gene with therapy-associated alterations, it typically was the only therapy-associated gene in the module, reinforcing their mutual exclusivity to other genes with therapy-associated alterations ( Figure 3B ).

To further characterize the relationships between co-occurring genomic alterations in detected gene modules and clinically important factors of NSCLC including histology and biomarkers of ICI response (TMB, PD-L1), we tested for dose-dependent associations between these factors and an increasing number of genomic alterations within a gene module ( Figure 4 ; Supplementary Figure S2 ). For each detected gene module, we calculated the frequencies of each unique combination of genes with co-occurring alterations (including single genes) and labeled each gene module by the gene(s) with the highest frequencies ( Figure 4 ; Supplementary Figure S2 , first column). This resulted in gene modules being designated the following: ATRX/GRIN2A, LRP1B/BRCA, DNMT3A/TET2, TP53, NF1, KRAS, BRAF, STK11/KEAP1, PIK3CA, EGFR, MUTYH/RET, RBM10/MET, and AXIN1 – associated gene modules. We then tested differences in histology, TMB, and PD-L1 between tumors with none, one, or ≥2 genomic alterations within genes co-occurring in each network-derived gene module ( Figure 4 ; Supplementary Figure S2 , columns 2-4).

Out of the 11 gene modules, 8 showed a clear dose-dependent association between an increasing number of genomic alterations and TMB or PD-L1 ( Figure 4 ). Results for the remaining gene modules can be found in Supplementary Figure S2 .

Significant associations between increasing numbers of genomic alterations and TMB were observed for ATRX/GRIN2A, LRP1B/BRCA, DNMT3A/TET2, TP53, NF1, and KRAS-associated gene modules (P ≤ 0.02), starting at a median TMB below 10 mut/Mb and climbing to medians above 10 mut/Mb ( Figures 4A–F ). The most dramatic increases were observed in the ATRX/GRIN2A, LRP1B/BRCA, and DNMT3A/TET2-associated gene modules, reaching upwards of 50 mut/Mb for tumors with ≥2 genomic alterations ( Figures 4A–C ). Of note, the DNMT3A/TET2-associated gene module consists of genes involved in clonal hematopoiesis, so this association may not be directly related to changes in the tumor itself.

For PD-L1, increasing number of genomic alterations in the BRAF-associated gene module was associated with higher frequencies of PD-L1 positive tumors (P ≤ 0.02; Figure 4G ), while increasing number of genomic alterations in the STK11/KEAP1-associated gene module was associated with lower frequencies of PD-L1 positive tumors (P ≤ 0.02; Figure 4H ). The KRAS-associated gene module also showed a significant association with PD-L1 levels; however, no difference was found between tumors with one or ≥2 genomic alterations within this gene module suggesting that the more prevalent KRAS alterations were driving this signal as additional genomic alterations within this module did not affect PD-L1 levels ( Figure 4F ).

In addition to the TMB and PD-L1 associations, we observed significant associations between increasing numbers of genomic alterations and higher frequencies of squamous cell carcinoma within the TP53 and PIK3CA-associated gene modules (P<0.0001; Figure 4D ; Supplementary Figure S2A ). KRAS, BRAF, STK11/KEAP1, MUTYH/RET, and RBM10/MET-associated modules showed the opposite trend, having significantly higher frequencies of adenocarcinoma cases as genomic alterations increased (P ≤ 0.02; Figures 4F–H ; Supplementary Figures S2C, D ).

Mutual exclusivity of strong oncogenic driver gene variants (i.e. KRAS, EGFR, ALK) in NSCLC tumors is well described (39) and was observed in the current study. Identification of this driver oncogene in a patient’s tumor is critical to ensuring they receive the correct therapy. In a recent study, overall survival was significantly compromised in NSCLC patients who received genomic test results after starting first-line therapy compared to those who received results before starting on the appropriate matched therapy (40). Overall survival was also impacted in patients whose treatment was initiated before receiving results and then switched to the matched targeted therapy once the corresponding driver gene variant was identified. This further enforces the importance of obtaining CGP results before first-line therapy to identify the true driver mutation and matched targeted therapy.

We confirmed gene alterations that have significant co-occurrence with key therapy-associated genes and created a network of genes across the entire NSCLC cohort to model higher-order interactions in a novel way ( Figure 3B ). Several closely associated genes were identified in the current study that have been previously described across multiple individual analyses with varying levels of evidence (9). However, less is known about the clinical significance of other gene-to-gene associations we observed including co-occurring alterations in MDM2, CDK4, and MET (41) or the impact of co-occurring CTNNB1 and EGFR alterations that may contribute to resistance to tyrosine kinase inhibitors (TKIs) (42, 43) ( Figure 3 ; Supplementary Table S5 ).

Broad coverage CGP has become increasingly important for patients with advanced NSCLC tumors with targetable oncogenic drivers considering evidence supporting lack of efficacy and even harm to NSCLC patients receiving ICI. Response to ICI or ICI plus chemotherapy is often limited or still under investigation in the presence of a therapy-associated oncogenic driver, regardless of PD-L1 expression or TMB status (8). CGP is even more important in advanced NSCLC patients who are non-smokers, as they are more likely to have tumors with targetable driver alterations.

ICI, with or without chemotherapy, is now routinely used as first-line therapy in advanced NSCLC patients with PD-L1 tumor expression ≥1%. Our analysis reveals that most tumors with therapy-associated driver mutations have low/intermediate (1 - 49%) PD-L1 expression, with 8% to 70% having high (≥50%) PD-L1 expression depending on the underlying driver mutations ( Figure 2 ; Supplementary Figure S1 ). If CGP is not ordered, or test results have not been received before first-line treatment decision making, there is an increased risk of advanced NSCLC patients with tumors containing driver mutations receiving ICI therapy that would be ineffective or potentially dangerous. This is particularly concerning in advanced NSCLC patients with EGFR tumor mutations receiving sequential ICI therapy followed by osimertinib, as they experienced severe immune-related adverse events requiring hospitalization (65). Timing of osimertinib or ICI therapy for those with EGFR exon 19 deletions or exon 21 L858R mutations and high (≥50%) PD-L1 expression is complex, as these patients have worse progression-free and overall survival when receiving osimertinib as first-line therapy (66). Further studies are indicated in this cohort of patients to determine if the combination of osimertinib plus chemotherapy can combat potential resistance. Finally, a small subset of NSCLC patients can experience rapid accelerations of tumor growth or hyper-progressive disease upon receipt of ICI therapy (67). Though evidence of predictive biomarkers for this phenomenon is limited, some studies suggest co-occurrence between MDM2 amplification or EGFR mutations, which we also saw evidence of ( Figure 3 ; Supplementary Table S5 ), could be contributing to the progression.

Additionally, as shown by our final analysis, more complex relationships exist between the increasing number of tightly co-occurring alterations and increasing or decreasing levels of PD-L1 and TMB ( Figure 4 ), which may have certain implications for guiding clinical decisions regarding ICI treatment. Although some of the TMB changes were found to be subtle (e.g., Figures 4F, H ), the addition of one ( Figure 4H ) or two ( Figure 4F ) variants in these gene modules seems to push TMB levels from below 10 (i.e., low TMB) to ≥10 (i.e., high TMB). This subtle change may or may not on its own have a significant influence on the immunogenicity of the tumor itself. Still, it places patients in the high TMB category, making them eligible for certain ICI treatments (68). However, TMB levels do not tell the full story and a better clinical guide for ICI response involves looking at changes in TMB and PD-L1 together, which was a goal of this analysis. For the KRAS–associated gene module ( Figure 4F ), we detected a subtle change in increasing TMB in conjunction with a 10-12% increase in high PD-L1 IHC cases, which together may signify that patients with more variants in this gene module might benefit from ICI treatment. Alternatively, for the STK11/KEAP1–associated gene module ( Figure 4H ), although we see subtle changes in increasing TMB, we see the opposite trend for PD-L1 with a 15-22% reduction in high PD-L1 cases, suggesting that more variants in this gene cluster might influence TMB levels, but may potentially be reducing the immunogenicity of the tumor, making the patient less likely to benefit from ICI treatment. These are just a couple of examples of the complex relationships we observed between tightly co-occurring alterations and increasing or decreasing levels of PD-L1 and TMB, and further investigation is warranted to tease apart the mechanisms of these interactions.