Tobacco smoke alters the trajectory of lung adenocarcinoma evolution via effects on somatic selection and epistasis

Tobacco smoke alters the trajectory of lung adenocarcinoma evolution via effects on somatic selection and epistasis | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Tobacco smoke alters the trajectory of lung adenocarcinoma evolution via effects on somatic selection and epistasis View ORCID Profile Krishna Dasari , View ORCID Profile Jorge A. Alfaro-Murillo , View ORCID Profile Jeffrey P. Townsend doi: https://doi.org/10.1101/2024.11.27.625765 Krishna Dasari 1 Department of Biostatistics, Yale School of Public Health , New Haven, CT 2 Medical Scientist Training Program, Perelman School of Medicine, University of Pennsylvania , Philadelphia, PA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Krishna Dasari Jorge A. Alfaro-Murillo 1 Department of Biostatistics, Yale School of Public Health , New Haven, CT Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jorge A. Alfaro-Murillo Jeffrey P. Townsend 1 Department of Biostatistics, Yale School of Public Health , New Haven, CT 3 Yale Cancer Center, Yale University , New Haven, CT 4 Department of Ecology and Evolutionary Biology, Yale University , New Haven, CT 5 Program in Computational Biology and Bioinformatics, Yale University , New Haven, CT Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jeffrey P. Townsend For correspondence: jeffrey.townsend{at}yale.edu Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract The mutagenic effects of tobacco smoke are well-reported. However, tobacco smoke also likely alters the somatic selective pressures that drive lung cancer evolution, and these altered pressures have never been quantified nor compared to the oncogenic effects of tobacco-induced mutagenesis. This comparison is necessary for prediction of targeted therapy efficacy given exposure history. Here, we estimated background mutation rates, somatic selection, and selective epistasis for 21 driver genes in ever- and never-smoker lung adenocarcinoma (ES-and NS-LUAD). As expected, mutation rates were gene-specifically elevated in ES-LUAD. However, variant prevalence differences between ES- and NS-LUAD could not be explained by differential mutagenesis. KRAS , KEAP1 , and STK11 mutations were more strongly selected in ES-LUAD. EGFR , PIK3CA , and SMAD4 mutations were more strongly selected in NS-LUAD, wherein EGFR mutations were associated with upregulation of epithelial-mesenchymal transition—a functional consequence specific to exposure history. Epistasis was pervasive and context-dependent: ES-LUAD exhibited more synergy and less antagonism, implying a broader, more navigable adaptive landscape than NS-LUAD. We also performed the first systematic quantification of higher-order-than-pairwise epistasis, revealing sub-additive and emergent synergies that shape cancer evolution. Our results disambiguate the mutagenic and selective effects of tobacco smoke, demonstrating how environmental insults reshape cancer evolution and enabling prediction of therapeutic efficacy based on smoking history and somatic genotype. Author Summary Smoking is a well-known cause of lung cancer because it increases the number of mutations in lung cells. But tobacco smoke also causes inflammation and other physiological changes that might influence how tumors evolve. Using DNA from lung tumors, we asked whether these changes make certain mutations more helpful to cancer growth and thereby better targets for therapy. We also asked whether some mutations were more helpful to cancer growth in those who haven’t smoked. To do so, we estimated not only how often mutations occurred, but also how strongly they were favored during cancer development. Focusing on 21 key lung cancer genes, including KRAS , EGFR, and TP53 , we found that mutations in some, such as KRAS and KEAP1 were more strongly favored in smokers. Mutations in genes such as EGFR and PIK3CA were more strongly favored in non-smokers, where they were also linked to non-smoker-specific changes in gene activity. We also found that mutations in tumors interact differently depending on smoking history: smokers’ tumors showed more cooperation between mutations, which may help them evolve in more flexible ways. This study shows how smoking reshapes the genetic paths lung cancers take—and why a patient’s smoking history matters when choosing targeted treatments. Understanding these evolutionary patterns can help improve treatment decisions and lead to more personalized care for lung cancer patients. Introduction Tobacco smoke is well known to cause cancer, in part through its mutagenic effects [ 1 ]. Carcinogen exposure from tobacco generates distinct patterns of mutations [ 2 ], a subset of which then contribute to oncogenesis [ 3 – 9 ]. In addition to this mutagenic effect, tobacco smoke induces physiological changes in the lung—inflammation, oxidative stress, immune dysregulation, and tissue damage [ 1 , 10 – 12 ]—that can alter the lung microenvironment and thereby the selective pressures acting on specific somatic mutations. These microenvironmental disruptions could modulate the fitness advantages conferred by specific mutations, thereby influencing their likelihood of contributing to cancer development. Indeed, several studies have suggested that, in principle, endogenous or exogenous alterations to the microenvironment caused by factors such as aging or tobacco exposure likely shape the trajectories of cancer evolution [ 13 – 17 ]. However, the extent to which such physiological effects alter the adaptive landscape of cancer has not previously been quantified. Therefore, quantification of the adaptive impacts of a highly disruptive exogenous insult such as tobacco smoke is crucial to understanding how environmental exposures shape the evolution of individual cancer evolutionary trajectories. Among lung cancer subtypes, LUAD is the most common and accounts for the highest proportion of cases in never-smokers [ 18 ]. As of 2002, nearly 23% of LUADs in the United States occurred in never-smokers [ 19 ], a proportion that has been rising as cigarette usage declines [ 20 ]. Numerous studies have reported that LUAD exhibits distinct somatic variant profiles in never-smokers compared to ever-smokers, which has been taken as evidence of their divergent genetic vulnerabilities [ 21 – 25 ]. However, differences in variant prevalence alone are confounded as to whether they are a consequence of effects of tobacco-induced mutagenesis or a consequence of tobacco-altered selection dynamics. Therefore, quantification of the relative oncogenic roles of the mutagenic and physiological effects of smoking in LUAD will be increasingly crucial to determination of optimal use of targeted therapies for patients with and without a prior smoking history. Computational models now enable inference of baseline somatic mutation rates from synonymous mutations observed in tumor genomes—implicitly [ 6 ] or explicitly [ 26 ] accounting for carcinogen-induced mutagenesis, such as that caused by tobacco smoke. Comparing these baseline mutation rates to the observed prevalence of nonsynonymous mutations enables estimation of somatic selection acting on specific substitutions [ 8 , 27 , 28 ]. This framework provides a foundation for disentangling the oncogenic contributions of the mutagenic and selective effects of smoking. Moreover, selection on each mutation is likely influenced not only by the microenvironmental effects of tobacco smoke, but also by pairwise [ 28 – 30 ] and higher-order [ 28 ] selective epistasis between mutations. In selective epistasis, mutations affect the selection operating on other mutations. These interactions between the selective effects of mutations shape the adaptive landscape of cancer evolution [ 5 , 28 , 31 , 32 ]. Understanding how tobacco smoke influences not just mutation and selection, but also epistasis, is essential for a complete picture of LUAD evolution. Because knowledge of epistasis is crucial to inference of cancer-driving mutation modules and synthetic vulnerabilities, a number of algorithms have been developed that attempt to detect epistasis between mutations or sets of mutations by quantifying over- and under-representation [ 33 , 34 ] in the context of heterogeneous mutation rates [ 30 , 35 – 37 ]. However, with the exception of Coselens [ 30 ], these methods do not enable meaningful quantitative comparisons of the effects of mutations in distinct somatic genotypes. This quantification is a crucial step for prioritization and translation of knowledge of epistasis into effective therapeutic strategies. Moreover, all of these approaches are restricted to pairwise inferences [ 30 , 34 ] or to inference of discrete modules [ 33 , 35 – 37 ], and are not well suited to the quantitative inference of higher-order epistasis (involving more than two mutations). When present, higher-order epistasis substantially influences the adaptive landscape and shapes evolutionary trajectories [ 38 – 43 ], including the degree to which selective epistasis diminishes along the adaptive trajectory [ 44 – 47 ]. Systematic characterization of the presence and biological significance of higher-order epistasis to cancer evolution has been challenging. Recent theory facilitates understanding of higher-order epistasis in cancer by providing the selection strength of mutations for each somatic genotype [ 28 ]. This novel approach has not yet been applied in the context of environmental risk factors that could also affect selection, such as tobacco smoke. Accordingly, analysis of the effect of smoke on the adaptive landscape of LUAD will benefit from consideration of pairwise and higher-order epistatic effects and potential differences in epistasis between microenvironments. To address these gaps, we quantified the mutagenic, selective, and epistatic effects of tobacco smoke on single-nucleotide variants (SNVs) in LUAD. We assembled a large dataset of lung adenocarcinoma tumor genomes and classified them by smoking status into ever-smoker (ES) and never-smoker (NS) groups. We estimated background mutation rates in each group and inferred the strength of somatic selection acting on driver mutations in each gene. We then investigated how selection pressures differed between ES- and NS-LUAD and associated these differences with differential gene expression. Finally, we extended our analysis to inference of selection in specific somatic genotypes, detecting and quantifying the extent of pairwise and higher-order epistasis. The degree of divergence in the mutational, selective, epistatic, and transcriptional landscapes of LUAD between ever-smokers and never-smokers enables novel insight into how environmental exposure shapes cancer evolution—identifying genotype- and environment-specific vulnerabilities that can inform personalized therapeutic strategies. Results Mutation rates and somatic selection pressures jointly shape the prevalence of driver variants in lung adenocarcinoma To disentangle the relative contributions of mutation and selection to observed mutation frequencies in LUAD, we deconvolved the prevalence of somatic variants in our curated set of 21 well-supported driver genes into estimates of the background oncogenic mutation rates and scaled selection coefficients. Across all smoking-status-classified LUAD samples in our dataset ( n = 1,722), TP53 and KRAS were the most frequently mutated genes, followed distantly by EGFR , KEAP1 , and STK11 ( Fig. 1A ). Background oncogenic mutation rates for the majority of these genes spanned slightly more than an order of magnitude—between 7.6 × 10 −8 and 8.7 × 10 −7 ( Fig. 1B ). When compared to the observed variant prevalences, this variation in background oncogenic mutation rate frequently implies substantial and variable impacts of somatic selection on the prevalence of each variant. Download figure Open in new tab Figure 1. Fixation rates and scaled somatic selection coefficients for 21 known and potential driver mutations among 1,722 lung adenocarcinomas. (A) Poisson-corrected fixation rates (rates at which each mutation occurs and becomes fixed within the population). (B) Background rate at which oncogenic gene mutations occur. (C) Strength of somatic selection (log-scaled) experienced by oncogenic gene mutations. Inset: Correlation between the prevalence of an oncogenic mutation and the strength of selection acting on it ( P < 0.001). Scaled selection coefficients for oncogenic mutations in each gene exhibited a long-tailed distribution ( Fig. 1C ), with mutations in KRAS experiencing the strongest selection, followed distantly by mutations of TP53 and CTNNB1 , then mutations in ATM , KEAP1 , and PIK3CA , and then other genes that experienced weaker but still substantial selection ( Fig. 1C ). Notably, CTNNB1 mutations were present in only 3.7% of tumors. Nevertheless, knowledge of the low underlying oncogenic mutation rate of CTNNB1 indicates that these infrequent mutations experienced stronger selection than did the more frequent mutations in STK11, RBM10 , and SMARCA4 . Generally, selection and mutation prevalence were correlated ( r = 0.68, P < 0.001), but several discrepancies were present between the prevalence of a mutation and the relative intensity of selection it experienced ( Fig. 1C , inset). These discrepancies highlight the importance of reporting selection as the effect size of mutations, rather than reporting solely prevalence, which is only moderately associated with strength of selection. The oncogenic mutation rates and selection coefficients presented above were quantified based on all tumors, regardless of smoking status. However, both the mutagenic processes and selective advantages of mutations were plausibly influenced by the genotoxic and physiological effects of tobacco smoke. To assess these distinct effects, we stratified tumors by smoking status and performed separate analyses of somatic selection in ES- and NS-LUAD tumors. Mutation and selection differ substantially between ever-smoker and never-smoker LUAD Stratification of LUAD tumors by smoking status based on mutational signature attributions and clinical annotation yielded 1,066 ES-LUADs and 656 NS-LUADs. In each group, we independently estimated oncogenic mutation rates and selection coefficients for the 21 well-established driver genes ( Fig. 2 ). Both mutation rates and selection pressures differed markedly between ES- and NS-LUAD. Oncogenic mutation rates were elevated in ES-LUAD across all 21 genes. Increases ranged from modest (39%–43% for RB1 and CTNNB1 ) to extreme (571–1474% for EGFR , SMAD4 , GNAS , and ALK ; Fig. 2A inset ). These elevated mutation rates demonstrate the intense mutagenic effect of tobacco smoking on the peripheral tissue of the lung. Download figure Open in new tab Figure 2. Mutational and adaptive landscapes of ever-smoker lung adenocarcinoma (ES-LUAD) and never-smoker lung adenocarcinoma (NS-LUAD). ( A ) Strengths of somatic selection ( y axis is square-root scaled; *: non-overlapping 95% confidence intervals) for mutations of 21 known and potential driver genes in NS-LUAD (teal bars; n = 656) and ES-LUAD (salmon bars; n = 1,066), averaged across somatic genotypes. Inset: Background oncogenic mutation rates for these genes in ES-LUAD and NS-LUAD (dashed line: equality). ( B ) Differential expression of genes between EGFR wild-type (WT) and EGFR mutant samples in ES-LUAD ( n = 273) and NS-LUAD ( n = 59; log2FC: log 2 of fold change in expression) for genes deemed to have a significant gene expression interaction between EGFR allelic status and smoking status. Color of each gene name denotes whether the change in expression was more positive (red) or more negative (blue) in NS-LUAD compared to ES-LUAD (*: P < 0.05, •: P < 0.1, SOX21 : P = 0.116; background color: change in gene expression from normal ( n = 33) to LUAD ( n = 332) is increased [pink], decreased [light blue], or insignificant [gray]). Leftward squares designate whether the gene produces a transmembrane (TM) or extracellular matrix (ECM) protein as well as whether it facilitates or interacts with the Wnt, and TGF-β and the epithelial-mesenchymal transition (EMT) signaling pathways. ( C ) Interactions between proteins (red: implicated in EMT) whose EGFR -mutation-associated expression changes differ between ES-LUAD and NS-LUAD. ( D ) Sample-specific DNA replication expression scores for KEAP1 WT and mutant ES- and NS-LUAD (**: P < 0.01). We found that reported differences between ES- and NS-LUAD in the prevalences of key driver mutations [ 21 , 22 , 48 ] were often driven by differential selection, rather than solely by differential mutagenesis ( Fig. 2A ). Among the 21 driver genes analyzed, TP53 , EGFR , PIK3CA , SMAD4 , MET , MGA , and GNAS mutations experienced stronger positive selection in NS-LUAD. STK11 , KEAP1 , and KRAS mutations experienced stronger positive selection in ES-LUAD. RB1 , BRAF , and BRCA2 mutations experienced similar selection in both cancer types. Most mutations that experienced significantly different selection between ES- and NS-LUAD were strongly selected in one context and weakly selected in the other, demonstrating pronounced context-dependence in LUAD evolution. In ES-LUAD, KEAP1 mutations experienced the third strongest selection—after TP53 and KRAS —and experienced over three times stronger selection than in NS-LUAD. Likewise, STK11 mutations experienced over twice as much selection in ES-LUAD. Indeed, both KEAP1 and STK11 mutations are enriched in ever-smoker populations [ 23 , 49 , 50 ]. EGFR mutations exhibited the most dramatic difference: they were more than twenty times more strongly selected in NS- than ES-LUAD, and were also the third most strongly selected mutations in NS-LUAD after TP53 and KRAS . EGFR mutations are much more common in NS-LUAD than in ES-LUAD [ 21 , 51 ]. Similarly, PIK3CA , SMAD4 , and MET mutations experienced over three times stronger selection in NS-relative to ES-LUAD. To investigate mechanisms underlying the differential selection of EGFR mutations, we assessed their differential transcriptional effects in ES- and NS-LUAD using TCGA expression data. In fewer than one percent of genes did the expression fold change from EGFR wild type samples compared to EGFR mutant samples differ significantly between ES- and NS-LUAD ( Fig. 2B ). However, several of these genes exhibited markedly stronger upregulation or downregulation in EGFR-mutant NS-LUAD compared to ES-LUAD. Genes that were upregulated were frequently components of Wnt, TGF-β, and Hedgehog signaling pathways or of the extracellular matrix ( Fig. 2B–C ). Indeed, in an analysis of gene sets upregulated in NS-versus ES-LUAD, the Hallmark gene set associated with epithelial-mesenchymal transition was the most enriched. On the other hand, KEAP1 mutations were associated with a substantially higher DNA replication enrichment score in ES-LUAD but not in NS-LUAD ( Fig. 2D ). These examples relate microenvironmental context to modifications of the downstream effects of driver mutations and their accompanying selective advantage. Extending our analysis from 21 to 1,200 LUAD-associated genes revealed a striking asymmetry in selection by smoking status: only six genes— KRAS , KEAP1 , STK11 , PBRM1 , DNMT3B, and CDH1— were significantly more strongly selected in ES- than NS-LUAD, whereas 87 experienced stronger selection in NS-LUAD ( Fig. S1 ). Among the mutations most strongly favored in NS-LUAD were those in EGFR , SMAD4 , SMAD2 , PTEN , PIK3R1, PAK1 , FGF3 , GATA3 , JUN , CTNNA2 , and TLR4 . Mutations often experienced weaker selection in ES-LUAD, which is more heavily mutated and frequently possesses multiple known drivers, than in NS-LUAD, which often features a single known driver ( Fig. S1 ). Overall, somatic mutations often experienced substantially altered selection in the tobacco-smoke-altered ES-LUAD compared to NS-LUAD, contributing to differential evolution of the two cancers. Along with the somatic environment, the somatic genotype can also induce differential selection on specific mutations: effects of each mutation can be analyzed across somatic genotypes to reveal selective epistasis that shapes the evolutionary trajectories of tumors in ever-smokers and never-smokers. Therefore, we assessed the influence of pairwise selective epistasis. Pairwise epistasis shapes the adaptive landscape of ES-LUAD and NS-LUAD To assess how frequently and how strongly the evolutionary trajectory of LUAD is affected by pairwise epistasis, we quantified the effects of driver mutations both in isolation (in tumors that were wild type at a paired gene) and in the presence of each paired gene mutation. For ES- and NS-LUAD, we evaluated all 420 possible gene pairs among the 21 driver genes: pairwise interactions were identified that modulate the selective advantage of mutations, and did so in a smoking-dependent manner. Mutation of TP53 followed by RB1 is the most likely trajectory for ES-LUAD and NS-LUAD To investigate directional epistasis between two common LUAD drivers, TP53 and RB1 , we first stratified tumors by smoking status and somatic genotype at these two loci, and then estimated the fixation rates of each mutation from each somatic genotype ( Fig. 3A ). In ES-LUAD, TP53 mutations exhibited a high fixation rate when RB1 was wild-type but were inferred to exhibit a negligible fixation rate when RB1 was mutated ( λ , Fig. 3B ). Conversely, RB1 mutations exhibited a low fixation rate when TP53 was wild-type but exhibited a high fixation rate when TP53 was mutated. These asymmetries indicate the most likely evolutionary sequence in ES-LUAD: mutation of TP53 followed by mutation of RB1 . In wild-type tumors, the mutation rate of TP53 was over double the mutation rate of RB1 ( µ WT→TP53 = 7.4 × 10 −7 , µ WT→RB1 = 2.0 × 10 −7 ); the same was true of mutant tumors ( µ RB1→RB1+TP53 = 8.9 × 10 −7 , µ TP53→RB1+TP53 = 3.6 × 10 −7 ; Fig. 3B ). Accordingly, RB1 variants in a TP53 -mutant genotype must have experienced comparable selection to TP53 -variants in a wild-type genotype ( γ , Fig. 3B ), despite the substantially lower fixation rate of the RB1 variants. This pattern reflects synergistic epistasis: prior mutation of TP53 increases selection for RB1 . In contrast, the reverse order exhibited antagonistic epistasis— TP53 was less strongly selected in the presence of RB1 mutation—though this last effect did not achieve statistical significance. Download figure Open in new tab Figure 3. Pairwise selective epistasis inferred within the evolutionary trajectories of lung adenocarcinoma (LUAD), depicted for TP53 and RB1 . ( A ) Genomic sequence data from lung adenocarcinomas can be stratified by clinically- and genetically-determined patient smoking status (blue: never-smokers; red: ever-smokers) and somatic genotype. Analysis of ( B ) ever-smoker LUAD and ( C ) never-smoker LUAD deconvolves somatic genotype prevalences (gray circle radius) into genotype-specific fixation rates (λ; arrow width), oncogenic mutation rates (μ; arrow width), and strengths of somatic selection (γ; arrow width) on mutations of TP53 (dark blue arrows) and RB1 (brown arrows). In NS-LUAD, the fixation rate for TP53 mutations from a wild-type genotype was lower than in ES-LUAD, and the fixation rate for RB1 mutations in a TP53 -mutant background was higher ( λ , Fig. 3C ). In single-mutant tumors, the mutation rate of TP53 is nearly double that of RB1 ( µ RB1→RB1+TP53 = 4.0 × 10 −7 , µ TP53→RB1+TP53 = 2.2 × 10 −7 ), but in wild-type NS-LUAD tumors, the mutation rate of TP53 was similar to the mutation rate of RB1 ( µ WT→TP53 = 1.48 × 10 −7 , µ WT→RB1 = 1.8 × 10 −7 ). This similarity of mutation rates contrasts with the disparity in ES-LUAD. Despite these differences, the inferred evolutionary trajectory mirrored that of ES-LUAD: the most likely ( λ ) and most adaptive ( γ ) path involved mutation of TP53 followed by mutation of RB1 . As in ES-LUAD, prior TP53 mutation significantly synergistically enhanced selection for RB1 mutation, and prior RB1 mutation antagonized selection for TP53 mutation, without reaching statistical significance ( γ , Fig. 3C ). Synergistic epistasis is widespread and often substantial in ES-LUAD Analysis of all pairwise combinations of the 21 reputed LUAD driver genes revealed frequent synergistic and antagonistic epistasis. In ES-LUAD, synergistic interactions were especially pronounced: for genes whose mutations were under strong positive selection (with the notable exceptions of TP53 and KRAS ) synergistic epistasis led to two- to eight-fold increases in selection. For example, selection on mutations in KEAP1, ATM , BRCA2 , PIK3CA , and ARID1A was markedly enhanced in the presence of prior TP53 or KRAS mutations ( Fig. 4A ). Additional strong synergistic interactions included those affecting mutants of ARID1A in the context of somatic genotypes with ATM or RBM10 mutations; mutants of STK11 in the context of mutated KEAP1 ; and mutants of ATM in the context of mutated STK11 . Mutations in the non- TP53 tumor suppressors BRCA2 , SETD2 , RBM10 , RB1 , MGA , and APC , as well as in the chromatin-remodeling genes ARID1A and SMARCA4 , experienced moderate baseline selection that was frequently enhanced by the presence of diverse prior mutations. These genes never experienced antagonistic epistasis with each other and rarely with other mutations. Download figure Open in new tab Figure 4. Pairwise epistasis in ever-smoker (ES) and never-smoker (NS) lung adenocarcinoma (LUAD). Strengths of somatic selection for mutations in wild-type tissue (green) and in previously mutated tissue (gray, labeled with previous mutation in brackets) in ( A ) ES-LUAD and ( B ) NS-LUAD. Significant pairwise epistatic interactions depicted in heatmap form for ( C ) ES-LUAD and ( D ) NS-LUAD based on non-overlapping confidence intervals for epistatic strengths of selection. Heatmap values are ratios of selection coefficients for mutations (in each row) from a somatic genotype containing a mutation in another gene (in each column) to the selection coefficient from a somatic genotype lacking the mutation. ( E ) Strengths of somatic selection for mutation of KRAS in TP53 -mutant and TP53 -wild-type somatic genotypes in ES-LUAD (red) and NS-LUAD (blue). Patterns of epistasis involving oncogenes were heterogeneous. Notably, KRAS and CTNNB1 mutations showed no elevation of selection in the presence of any prior mutations in the 21-gene set. Conversely, selection for mutations of GNAS and ALK was enhanced by three prior mutations ( Fig. 4A ). For many oncogenes, selection was enhanced by prior TP53 mutation, but was antagonized by mutations in other oncogenes such as MET . Across all tested genes, the median effect size of statistically significant synergistic epistasis was a 3.6× increase in selection, and the magnitude of this synergy did not correlate with the baseline strength of selection for either the first or the second mutation. Antagonistic effects can reduce positive selection or induce negative selection on drivers In ES-LUAD, we identified eight gene pairs exhibiting statistically significant antagonistic sign epistasis, in which selection for a driver mutation shifted from positive to negative in the presence of another mutated gene. Mutations in CTNNB1 and GNAS became selectively disadvantageous in the presence of MET and APC mutations. Similarly, selection for SMAD4 mutations was antagonized by prior BRCA2 and PIK3CA mutations, and MET mutations were antagonized by mutations of RB1 , EGFR , CTNNB1 , and GNAS . In each of these eight cases, the reciprocal mutation order also induced negative selection (e.g., mutated CTNNB1 induced negative selection for MET mutations). For over a hundred other ordered pairs, antagonistic sign epistasis was estimated, without reaching statistical significance. Antagonistic shifts in the magnitude of selection, in which selection for a mutation remained positive but was substantially reduced, were indicated in 33 cases, of which three were significant: the presence of KRAS mutations caused a 64% reduction in selection for TP53 mutations and an 81% reduction in selection for EGFR mutations, and TP53 mutations caused a 58% reduction in selection for KRAS mutations ( Fig. 4A ). Across all analyses of ES- and NS-LUAD, new mutations in TP53 and KRAS were never inferred to experience significantly stronger selection in the context of another driver mutation. Antagonistic epistasis is more frequent in NS-LUAD Compared to ES-LUAD, NS-LUAD exhibited a higher frequency of antagonistic epistasis and fewer instances of synergistic epistasis ( Fig. 4B ). Of the 65 gene pairs exhibiting significant antagonistic epistasis in ES-LUAD, 25 also reached significance in NS-LUAD, and an additional 36 exhibited antagonistic trends without achieving significance ( Fig. 4C–D ). Thus, instances of selective antagonism inferred in ES-LUAD were nearly always present in NS-LUAD. This consistency contrasted with a substantial inconsistency in selective synergies between ES- and NS-LUAD. Of the 42 gene pairs whose mutations exhibited significant synergistic epistasis in ES-LUAD, only five were significantly synergistic in NS-LUAD. An additional 25 trended synergistic without reaching significance. Strikingly, 16 of these 42 instances were inferred to be significantly antagonistic in NS-LUAD, including many involving BRCA2 . For example, KEAP1 mutation enhanced selection for BRCA2 mutation in ES-LUAD but antagonized it in NS-LUAD ( Fig. 4C–D ). These reversals reveal distinct adaptive constraints imposed by the smoking-altered tumor microenvironment. Overall, antagonistic interactions were relatively stable across smoking status, whereas synergistic epistasis was highly contingent on smoking status. Mutations in these 21 driver genes were subject to a broad range of epistatic and environmental interactions, as well as combinations thereof ( Fig. S2 ). For example, TP53 mutations antagonized KRAS mutations in ES-LUAD but had no substantial effect on KRAS mutations in NS-LUAD ( Fig. 4E ). These context-dependent epistatic effects substantially contribute to the divergent trajectories of ES- and NS-LUAD. Fewer examples of synergistic epistasis were observed in NS-LUAD compared to ES-LUAD. Nevertheless, synergistic epistasis contributed to a median twelve-fold increase in selection on secondary mutations. Notable examples include enhanced selection for CTNNB1 , RB1 , and RBM10 mutations after EGFR mutation, for PIK3CA , BRCA2 , SETD2 , and ARID1A mutations after ATM mutation, and for SETD2 mutation after ARID1A mutation ( Fig. 4B ). Even in relatively less mutated NS-LUAD, cooperative genetic interactions dramatically reshape the fitness landscape of evolving tumors. Pairwise epistasis can be asymmetric in both sign and magnitude In both ES- and NS-LUAD, several gene pairs exhibited asymmetric epistasis, in which the direction or strength of interaction depended on the order of mutation acquisition. For instance, in ES-LUAD, mutation of either KEAP1 or STK11 enhanced selection on mutations of the other, but with markedly different magnitudes: primary STK11 mutation was estimated to increase selection for a secondary KEAP1 mutation by 140%, whereas primary KEAP1 mutation significantly increased selection for a secondary mutation of STK11 by 360%. In other cases, the sign of epistasis reversed depending on mutational order. Most notably in ES-LUAD, primary KRAS mutation synergistically increased selection for a secondary KEAP1 mutation, but primary KEAP1 mutation strongly antagonized selection for a secondary KRAS mutation ( Fig. 4A ). This asymmetry means that a mutation in KRAS followed by a mutation in KEAP1 was advantageous, but a mutation in KRAS was disadvantageous or lethal when preceded by a mutation in KEAP1 . This dramatic directional asymmetry and eighteen other such instances exemplify how somatic evolutionary trajectories can be constrained or potentiated by the specific sequence in which mutations occur. Higher-order epistasis shapes evolutionary trajectories of ES-and NS-LUAD KRAS , KEAP1 , and STK11 constitute an epistatically interacting triad in ES-LUAD Having identified KRAS , KEAP1 , and STK11 as experiencing strong, ever-smoker-specific selection, we next investigated whether their interactions extended beyond pairwise epistasis to form a higher-order epistatic network. Mutations in all three genes exhibited substantial fixation rates from nearly any initial somatic genotype, particularly en route to the triple-mutant genotype. Fixation rates for KRAS mutations were estimated to be very low when arising after KEAP1 or STK11 mutations alone compared to when arising as the initial mutation or in the KEAP1 - STK11 double-mutant ( Fig. 5A ). Nevertheless, these three genes form an epistatic triad in which mutations in any one tend to increase selection for the others. Download figure Open in new tab Figure 5. Evolutionary trajectory for KRAS , KEAP1 , and STK11 mutations in ever-smoker lung adenocarcinoma. ( A ) Fixation rates, ( B ) oncogenic mutation rates, and ( C ) scaled selection coefficients for mutations of KRAS (orange arrows; arrows width proportional to rate or coefficient), KEAP1 (light blue arrows), and STK11 (yellow arrows) conditioned on somatic genotype (gray circles; radius proportional to genotype prevalence). Among the three genes, STK11 manifested the highest oncogenic mutation rate at each progressive step, followed by KEAP1 , and then KRAS ( Fig. 5B ). Given these differential underlying rates of mutation, we computed the consequent scaled selection coefficients for mutations in each genotype in the three-gene model, and compared selection between genotypes to quantify pairwise and higher-order selective epistasis. Many tumors possessed mutations in at least one of these genes ( Fig. 5C , second column of nodes), providing high power to detect pairwise epistasis. Indeed, our three-gene model recapitulated the pairwise epistatic relationships inferred from the two-gene models ( Fig. 4C ). However, concurrent mutations of two or more of these genes were relatively infrequent ( Fig. 5C , third and fourth columns of nodes), limiting power to detect higher-order epistasis with statistical confidence. For example, substantial uncertainty in the selection estimate for KRAS mutations in tumors harboring STK11 and KEAP1 mutations precluded a definitive inference of higher-order interactions. In contrast, because of the sufficiently high prevalence of KRAS - KEAP1 - and KRAS - KEAP1 - STK11 -mutant genotypes, we were able to infer higher-order epistatic effects on STK11 with greater confidence. STK11 mutations experienced significantly stronger selection in the presence of either KRAS or KEAP1 mutations. If the synergistic effects of KRAS and KEAP1 mutations on STK11 compounded additively, then STK11 mutations would experience selection equivalent to the three-way product of the individual synergistic effects and the baseline selection for STK11 mutations. However, the upper bound for selection for STK11 mutations in a KRAS - KEAP1 double-mutant genotype was less than this product ( Fig. 5C ). Accordingly, the compound synergistic effect of concurrent KEAP1 and KRAS mutations represents an example of sub-additive higher-order epistasis. Most pairwise epistatic effects compound without strong higher-order interaction The sub-additive epistasis affecting mutations of STK11 demonstrates that assessments of higher-order epistasis will be necessary to thoroughly understand cancer evolution. However, most effects of driver mutations compounded in a manner consistent with minimal higher-order epistasis. Examining all 1330 triads out of the 21 genes, we compared selection for a new mutation in single-versus double-mutant backgrounds. In the vast majority of cases, a combination of pairwise epistatic effects was sufficient to explain significant differences in selection for a new mutation between single- and double-mutant genotypes. Generally, when selection for a secondary mutation in one gene was epistatically altered by presence of a primary mutation in a second gene and was unaffected by a primary mutation in a third gene, selection on the secondary mutation remained commensurately epistatically altered when both other variants of the triad were concurrently present. For example, in ES-LUAD, selection for SMARCA4 mutations was synergistically enhanced by STK11 mutations but unaffected by TP53 mutations, and correspondingly, selection for mutations in SMARCA4 was synergistically increased in a STK11 - TP53 -mutant genotype ( Fig. 6A,B ). Similarly, in NS-LUAD, selection for RBM10 mutations was synergistically increased in both a TP53 -mutant and TP53 - EGFR -mutant genotype ( Fig. 6C ). Overall, pairwise interactions typically dominate the adaptive landscape of LUAD, but higher-order epistasis can have substantial influence on specific evolutionary trajectories. Download figure Open in new tab Figure 6. Higher-order selective epistasis in lung adenocarcinoma (LUAD). ( A ) Selection within gene triads ( y axis is square-root scaled), for a mutation (bold font) in a context of no co-occurring mutations (circle), one co-occurring mutation (square; labeled with first letter(s) of the gene(s) indicated in the tick label that are mutated), and two co-occurring mutations (diamond) in ever-smoker (ES) LUAD [pink: visualized in panels B and F; purple: visualized in panel D; green: visualized in panel E]. ( B ) In ES-LUAD, the synergistic effect of SMARCA4 mutations (light blue arrows) on selection for APC mutations (pink arrows; arrow width proportional to scaled selection coefficient; circle radius proportional to somatic genotype prevalence) does not diminish in the presence of a neutrally-interacting KRAS mutation (orange arrows). ( C ) Selection within gene triads in never-smoker (NS) LUAD; y axis is square-root scaled, symbols are as in panel A]. ( D ) Pairwise and higher-order epistasis in ES-LUAD: co-mutation of TP53 (blue) and KRAS (orange), each of which independently increase selection for ARID1A mutations (purple), produces an even stronger synergy for ARID1A mutations; and ( E ) antagonistic effect of KRAS mutations (orange) on selection for EGFR mutations (green) outweighs the synergistic effect of TP53 mutations (blue). ( F ) In ES-LUAD, co-mutation of KEAP1 (light blue) and STK11 (yellow) synergistically increases selection for APC mutations (pink) despite neutral pairwise effects. ( G ) Comparison of the epistatic effect of double-mutant-genotype (label: bottom row, two-letter abbreviation) on selection for new mutation (label: top row) to the “expected” product of pairwise epistatic effects (blue points: both pairwise epistatic effects are statistically significant, but not the higher order effect; green: the higher-order effect is significant, but neither pairwise effect is; red: all effects are significant) When two co-occurring mutations each independently enhanced selection for a third mutation, the synergistic effects frequently compounded. For instance, ARID1A mutations experienced increased selection in both KRAS- and TP53 -mutant genotype. These synergistic effects compounded, resulting in mutations of ARID1A experiencing three times stronger selection in a KRAS - TP53 -mutant genotype than in a TP53 -mutant genotype and nine times stronger selection than in a wild-type genotype ( Fig. 6D ). Similar patterns of compounded synergy were evident for ATM mutations in KRAS - STK11 -mutant genotypes, as well as for SMARCA4 mutations in a KEAP1 - STK11 -mutant background. Across cases, compounded synergy frequently drove selection to levels up to an order of magnitude higher than in genotypes lacking either partner mutation. This substantial compounding constitutes a crucial element of how driver mutations accumulate in cancer genomes, positively reinforcing progression of cancer toward more aggressive forms. Higher-order epistasis often manifests as sub-additive or emergent synergistic interactions In the case of compounding synergistic effects, it is also possible to observe super-additive higher-order epistasis, in which selection for a mutation in the presence of two independently synergistic mutations is even larger than what would be expected from the product of their independent synergistic effects. Indeed, some cases are consistent with either additive synergy or super-additive epistasis, such as mutations of SMARCA4 in a KRAS - KEAP1 -mutant genotype in ES-LUAD ( Fig. 6A ). More frequently, cases are consistent with either additive synergy or sub-additive epistasis, in which selection for a mutation in the presence of two independently synergistic mutations is smaller than expected from the sum of synergistic effects. Indeed, in 43 of 45 cases wherein a new mutation arose in a genotype of two significantly synergistic mutations, estimates of selection for the new mutation were lower than would be expected from purely additive synergistic epistasis. Such widespread sub-additivity pervades the adaptive landscape, tempering rather than amplifying selective pressures, even in the context of strong pairwise synergy. We also identified cases of higher-order epistasis involving the co-occurrence of synergistic and antagonistic partner mutations, as well as the emergence of synergy in the co-occurrence of neutrally-interacting partner mutations. In ES-LUAD, for example, selection for EGFR mutations was synergistically enhanced by prior TP53 mutations but antagonized by KRAS mutations. In a TP53 - KRAS -mutant genotype, EGFR mutations experienced significantly less selection than in a TP53 -mutant KRAS -wildtype genotype ( Fig. 6A , E ). In fact, these EGFR mutations experienced levels of selection similar to those they would experience in a genotype lacking either mutation, indicating a neutralizing result of the combination of synergistic and antagonistic epistasis. Another form of higher-order epistasis is the emergence of synergy or antagonism from a combination of mutations, each of which does not individually exhibit pairwise epistasis with the affected gene. We observed several instances in which selection for a mutation was significantly increased by the co-occurrence of two mutations that did not individually have significant synergistic effects: in ES-LUAD, selection for mutations of APC increased by an order of magnitude in a KEAP1 - STK11 -mutant genotype despite not being significantly altered when only one of KEAP1 or STK11 was mutated ( Fig. 6A , F ). In this instance, significance was reached for synergy when compared to wild-type or a KEAP1 -mutant genotype, but not when compared to the STK11 -mutant genotypes, which was itself epistatically neutral. We detected multiple similar instances of emergent synergy, such as for ARID1A mutation in an EGFR - PIK3CA -mutant genotype in NS-LUAD ( Fig. 6G ). We also observed one instance of emergent antagonism, in which STK11 mutations were antagonized in a TP53 - EGFR -mutant genotype but not in either single-mutant genotype alone. Overall, we observed two major classes of higher-order epistasis: sub-additive synergy and emergent—and consequently super-additive—synergy. In particular, when both component epistatic effects were strongly synergistic, their combination was almost always sub-additively synergistic ( Fig. 6G ). On the other hand, when both component epistatic effects were neutral or insignificant, a super-additive synergistic effect occasionally emerged from their combination ( Fig. 6G ). Interestingly, as the strength of the component epistatic effects increases, the inferred epistatic effect of the double-mutant combination rapidly plateaus at an approximately ten to twenty-fold increase in selection. In this survey of the adaptive landscape of LUAD, we discovered several cases of higher-order epistasis that substantially influence the selective advantage of oncogenic mutations. Over a third of LUAD samples in ever-smokers possessed mutations in three or more of 21 tested genes—a proportion that would likely rise in an analysis of a greater number of driver genes. In these patients, higher-order epistasis may play a substantial role in the evolutionary trajectories of their tumors and their responses to targeted therapies. Indeed, we found that a wide variety of adaptive trajectories ensued from the combined epistatic effects of multiple mutations. Some of these trajectories cannot be predicted from pairwise analyses. However, our large-scale analysis of the adaptive landscape of 21 mutations has significantly supported only a subset of the many potential instances of higher-order epistasis between oncogenic mutations in LUAD. Indeed, multiple other instances of higher-order epistasis were estimated to be present. Their effect would be clarified with greater sample sizes. Discussion Here we have characterized the distinct adaptive landscapes of lung adenocarcinoma in ever-smokers and never-smokers by first estimating gene-specific background mutation rates in ES- and NS-LUAD, then comparing the intensity of somatic selection and selective epistasis acting on those mutations between the two groups. Consistent with the established mutagenic effects of tobacco carcinogens [ 4 ], we detected elevated oncogenic mutation rates in all tested genes in ES-compared to NS-LUAD. Accounting for these background rates revealed that mutations in genes such as EGFR , PIK3CA , and MET provide a stronger selective advantage in NS-LUAD, while mutations in KRAS , KEAP1 , and STK11 provide a stronger selective advantage in ES-LUAD. Focusing on EGFR and KEAP1 , we found that these adaptive differences were at least partially explained by the differential effects of these mutations on epithelial-mesenchymal transition and DNA replication transcriptional program activity in ES-and NS-LUAD. By quantifying the selective advantage of these mutations within specific somatic genotypes, our analysis recovered known epistatic interactions—such as the enhancement of selection on RB1 and EGFR mutations by prior TP53 mutations—and revealed context-specific interactions, such as the synergistic effect of EGFR mutations on CTNNB1 mutations uniquely in NS-LUAD. We also detected substantial asymmetries: CTNNB1 mutations did not impact EGFR mutations in ES- or NS-LUAD. Notably, we found more frequent synergy in ES-LUAD and substantially more pervasive antagonism in NS-LUAD, constituting fundamental differences in the evolutionary navigability of their adaptive landscapes. Finally, we showed that higher-order epistasis—particularly sub-additive synergistic and emergent synergistic epistasis—is frequent in and substantially guides the evolutionary trajectories of cancer. These findings of differential selection and epistasis reveal that exposure to tobacco smoke reshapes the adaptive landscape of LUAD not only by elevating mutation burden, but also by altering the selective value of mutations and their epistatic interactions—highlighting opportunities to leverage these adaptive differences for biomarker development and therapeutic stratification. The prevalences of driver mutations have been reported to substantially differ between smokers and never-smokers in LUAD [ 21 , 22 ] as well as in non-small cell lung cancer [NSCLC; 23,24]. However, previous studies have either refrained from explaining these differences or attributed them primarily to tobacco-induced mutagenesis [ 22 , 23 ]. Our results show that they often cannot be explained by differential mutagenesis alone. For example, KRAS variants are over five times more frequent in ES- than NS-LUAD tumors, but the background, neutral rate at which KRAS mutations occur is only about three times higher in ES-LUAD. To resolve this discrepancy, KRAS mutations must be nearly twice as oncogenic in ES-compared to NS-LUAD. In general, the differential prevalence of these driver mutations is often not solely attributable to tobacco-induced mutagenesis, but also due to their altered selective advantage in the tobacco-altered lung environment. This physiological-adaptive effect of smoking is crucial to account for: changes in the selective advantage of a mutation suggest altered functional impact within a given tissue environment, with direct implications for the efficacy of targeted therapies and the interpretation of biomarkers in smokers versus non-smokers. Multiple perspectives have proposed that the adaptive landscape of oncogenesis is altered by physiological insults from endogenous and exogenous factors [ 14 , 17 , 52 , 53 ]. However, data-driven evidence for this hypothesis has remained limited [cf. 54]. Here, we provide compelling evidence that canonical driver mutations confer altered selective advantages in the smoking-affected lung microenvironment of ever-smokers with LUAD. These shifts in selection are linked to context-dependent phenotypic effects of mutations. For example, EGFR mutations are associated with greater activity of the epithelial-mesenchymal transition (EMT) transcriptional program in NS-LUAD, but not in ES-LUAD—likely explaining their stronger selective advantage in never-smokers. Such divergent phenotypic consequences can be attributed to smoking-induced alterations to transcriptional program activity [ 55 – 57 ], tumor microenvironment composition [ 58 ], and cell fitness. These findings provide a mechanistic basis for how exogenous exposures can reshape the evolutionary trajectory of cancer by altering the adaptive value of oncogenic mutations. In the case of TP53 , Rodin and Rodin [ 13 ] reported comparable background mutation rates between smokers and nonsmokers with lung cancer, and attributed the higher frequency of TP53 mutations in smokers to increased selection in the smoking-altered microenvironment. By contrast, our analysis—leveraging a substantially larger dataset, incorporating tissue- and context-specific covariates, and applying advanced statistical methodologies to estimate background mutation rates [ 6 ]—identified a markedly elevated TP53 mutation rate in ES-LUAD. With this substantial differential mutagenesis accounted for, our analysis revealed that TP53 mutations are actually under weaker positive selection in ever-smokers than in never-smokers. This finding challenges earlier conclusions and underscores the importance of accurately resolving both mutational processes and selection when interpreting variant prevalence across environmental exposures. Despite mounting evidence of the substantial biological impact of tobacco smoke, clinical trials for lung cancer therapies often do not assess the influence of smoking status—and frequently do not even collect usage history [ 59 ], perhaps due to a conception that tobacco smoke has a negligible effect on outcomes [ 60 ]. However, our findings reveal that tobacco smoke profoundly reshapes the evolutionary trajectory of lung adenocarcinoma, necessitating more systematic incorporation of smoking history into trial design and treatment decision-making, particularly for targeted therapies. For example, the substantially stronger selection for EGFR mutations in NS-LUAD implies that EGFR tyrosine kinase inhibitors (TKIs) should yield greater clinical benefit in never-smokers. Indeed, a retrospective analysis of patients with EGFR -mutant LUAD found that the objective response rate (ORR) of patients receiving EGFR TKIs declined from 73% among never-smokers to 46% among individuals with over 10 pack-years of smoking history [ 61 ]. Similar patterns have been reported in additional retrospective studies of LUAD [ 62 ] and across NSCLC more broadly [ 63 , 64 ]. This concordance between selection-based predictions and observed outcomes highlights the translational value of selection coefficients as predictors of targeted therapy efficacy. By identifying likely responders based on environmental exposures that shape the tumor microenvironment, exposure-informed selection coefficients can improve clinical trial stratification, enhance statistical power, and refine treatment personalization in precision oncology. Prior studies have questioned whether computational analyses can reliably detect epistasis, citing concerns about confounding by subtype-specific mutation profiles, variation in mutation burden, and limited statistical power [ 31 , 65 ]. To address these concerns, we analyzed ES- and NS-LUAD independently and comparatively, modeled subtype- and gene-specific background mutation rates and effect sizes of mutations in specific somatic genotypes instead of suboptimal measures of co-occurrence or mutual exclusivity, and integrated data from multiple cohorts to ensure sufficient statistical power. Consequently, we were able to make robust inferences of both pairwise and higher-order selective epistasis, many of which align with experimental and clinical evidence. For example, consistent with our finding that KRAS mutations synergistically enhance selection for KEAP1 and STK11 mutations, prior studies show that KEAP1 loss accelerates LUAD progression in KRAS -mutant mouse tumors [ 66 , 67 ] and compensates for the deleterious effects of KRAS - STK11 -co-mutation [ 68 ], and mutations of KEAP1 and/or STK11 worsen survival and treatment response in KRAS -mutant LUAD [ 69 , 70 ] and NSCLC [ 71 – 76 ]. Furthermore, in accordance with the inferred mutually synergistic effects of mutations of KEAP1 and STK11 in ES-LUAD, co-mutation of KEAP1 and STK11 promote cell proliferation [ 77 ] and worsen survival in NSCLC independently of KRAS [ 78 , 79 ]. Our analyses also identified synergies between EGFR and other drivers: RBM10 and CTNNB1 in NS-LUAD, and RB1 in both NS- and ES-LUAD. These interactions are corroborated by experimental and clinical studies showing that RBM10 mutations promote EGFR -mutant LUAD [ 80 ] and confer resistance to EGFR TKIs [ 81 ], that CTNNB1 mutations are enriched in advanced EGFR -mutant NSCLC and associated with metastasis and TKI resistance [ 82 ], and that RB1 mutations in EGFR -mutant LUAD mediate histologic transformation to small-cell lung cancer [ 83 , 84 ]. We also revealed that mutations of TP53 and KRAS were widely antagonized by other driver mutations, consistent with their early, clonal emergence during LUAD evolution [ 85 , 86 ] and supporting their frequently primary role in tumor initiation. This panoply of alignments between our results and experimental and clinical evidence encourages deeper investigation into the many epistatic relationships we have revealed that remain untested in the laboratory or clinic. They represent fertile ground for mechanistic investigation and promise new therapeutic targets and biomarker candidates. Previous studies have shown that epistatic interaction networks vary across cancer types in humans [ 87 ] and across environmental contexts in other organisms [ 88 – 93 ]. However, extensive differences in selective epistasis within a single cancer type exposed to distinct tissue environments have not previously been demonstrated. Here, we have shown that tobacco smoke exposure markedly reshapes the selective epistatic interactions underlying LUAD evolution. Broadly, synergistic epistasis was more frequent in ES-LUAD, whereas antagonistic epistasis was more frequent in NS-LUAD. Indeed, in some cases, epistasis between driver genes was synergistic in ES-LUAD, but antagonistic in NS-LUAD. In other cases, the magnitude of synergy differed across exposures: synergy for mutation of TP53 in EGFR -mutant tumors was stronger in ES- than in NS-LUAD. The higher frequency of synergy and lower frequency of antagonism in ES-LUAD suggests a more navigable adaptive landscape—characterized by a higher frequency of beneficial mutational combinations and fewer evolutionary barriers—than in NS-LUAD. This increased accessibility enables ES-LUADs to develop along a greater diversity of evolutionary trajectories, contributing to increased interpatient heterogeneity. Moreover, it may also help to explain why tobacco smoke elevates the risk of lung cancer to such an exceptional degree. In NS-LUAD, widespread antagonistic epistasis may reflect a greater fitness cost of mutation accumulation, possibly due to a more immunologically reactive tumor microenvironment [ 12 , 48 , 58 , 94 ] or due to a lower tolerance for mutation accumulation in the non-dysregulated lung tissue cells of never-smokers. Taken together, these findings reveal that environmental exposures modulate selective epistasis within a cancer type, suggesting the importance of exposure history to predicting tumor evolutionary trajectories and to guiding epistasis-informed treatment strategies. It has often been assumed that the adaptive landscape of cancer can be adequately described without epistasis [ 6 , 95 – 99 ] or with only pairwise epistasis [ 31 , 32 , 100 ]. However, our results demonstrate that not only pairwise but also higher-order epistasis—particularly sub-additive and emergent epistasis—play a crucial role in modulating the adaptive value of cancer drivers. For instance, in ES-LUAD, selection for APC mutations increased by an order of magnitude in an KEAP1 - STK11 -mutant genotype, despite showing no significant pairwise synergy with either mutation alone. This emergent epistasis, arising only in the triadic context, would be missed by analyses restricted to pairwise comparisons, necessitating consideration of multi-mutant genotypes in both computational and experimental studies. Moreover, almost all significant pairwise synergistic effects combined sub-additively in triads, rapidly reaching a plateau of synergistic benefit. This pattern aligns with the diminishing-returns epistasis observed in asexual populations [ 44 – 47 ]. Overall, higher-order synergistic epistasis biases tumors toward some evolutionary trajectories, while higher-order antagonism prunes most trajectories. Both make tumor evolution more predictable. Thus, incorporation of higher-order epistasis into models of tumorigenesis is essential to understanding and guiding the evolutionary paths that tumors take under therapy. Current NSCLC treatment guidelines for EGFR TKIs do not account for co-occurring mutations. However, our findings suggest that concurrent mutations could serve as biomarkers of treatment response. Antagonistic epistasis between EGFR mutations and mutations in KRAS or KEAP1 implies that the efficacy of EGFR TKIs may be reduced in tumors harboring these concurrent alterations. This prediction aligns with clinical observations: KRAS mutations decrease survival in patients with EGFR -mutant LUAD patients treated with EGFR TKIs [ 101 ], and KEAP1 mutations reduced EGFR TKI efficacy in EGFR -mutant lung cancer [ 102 , 103 ]. More broadly, antagonistic epistasis is a signal of synthetic lethality [ 32 ]. Our results also demonstrate mutual negative selection between BRCA2 and SMAD4 loss-of-function mutations—suggesting that pharmacological inhibition of SMAD4 could be therapeutically beneficial in BRCA2 -mutant tumors, in analogy to usage of PARP inhibitors in BRCA1 / 2 deficient tumors. In general, selective epistasis can clarify the frequent heterogeneity of response to targeted therapies [ 37 , 104 ], and can improve stratification of responders and nonresponders to targeted therapy in clinical trials [ 105 , 106 ], enhancing trial power and therapeutic precision. Because TP53 mutations synergize with mutations of EGFR , we would expect that in tumors with concurrent EGFR and TP53 mutations, EGFR TKIs would be especially effective because they would both eliminate the baseline effect of the EGFR mutation and diminish the effect of the concurrent TP53 mutation that is mediated through the EGFR mutation. However, TP53 mutations are consistently associated with resistance to EGFR TKIs in EGFR -mutant LUAD [ 107 – 109 ] and NSCLC [ 110 – 114 ]. This contradiction may be explained by the pleiotropic effects of TP53 loss, including increased somatic nucleotide mutation rates [ 109 ], widespread copy-number alterations [ 107 , 115 ], and dysregulation of gene expression programs linked to resistance [ 116 ] and progression [ 115 ]. These pleiotropic effects provide greater genetic variation that facilitates the evolution of resistance independently of the selective synergy between TP53 loss and EGFR mutation. Previous computational studies of pairwise interactions have demonstrated asymmetries in the magnitude of pairwise epistasis [cf. 30]. Our analysis has revealed not only epistasis of asymmetric magnitudes, but also asymmetric signs of selection, in which one ordering of the mutations is advantageous while the reverse is disadvantageous. These directional asymmetries likely arise from cellular or stable transcriptional transitions initiated by the primary mutation that unidirectionally alter the selective context for subsequent alterations—an effect not captured by conventional analyses of mutual exclusivity or co-occurrence. Indeed, the presence of such asymmetries of epistasis are evident in established genetic models of tumor evolution, such as the APC → KRAS → TP53 mutation sequence in colorectal cancer [ 117 , 118 ], whose strongly preferential ordering requires directional epistasis. Asymmetric epistasis has also been supported by genetic and computational genomic studies of TP53 and CCNE1 mutations [ 32 ], TP53 and RB1 mutations [ 30 ], and JAK2 and TET2 mutations [ 119 ]. The presence of directional epistasis implies preferential orders of mutation acquisition that offer new opportunities to predict and potentially redirect cancer evolution. This study was subject to several limitations. First, its retrospective nature introduces the possibility that covariates of smoking behavior—such as comorbid health states, environmental exposures, inherited genetics, and sex biases—may contribute to the observed differences in the adaptive landscapes of ES-LUAD and NS-LUAD. However, none of these factors are known to exert a physiological impact on the lung that is comparable to smoking tobacco. Accordingly, the selective differences we report are most parsimoniously attributed to the direct and downstream effects of smoking. Second, smoking status in our study was primarily inferred from mutational signatures in tumor sequences, which can be susceptible to misclassification. However, this method has been validated for classification of smoking status in NSCLC and is believed to outperform self-reported smoking histories [ 21 , 25 , 120 ]; self-reported status has been found to be unreliable [ 121 , 122 ] and often does not account for the risk conferred by secondhand smoke [ 123 ]. Furthermore, because our approach explicitly quantifies the smoking-associated mutational burden in the lung periphery—where LUAD develops—it also implicitly identifies patients who have experienced significant injury from smoking in the regions where LUAD originates. Finally, our analysis was restricted to somatic SNVs and did not consider indels, copy number alterations (CNAs), or other structural variants (SVs). CNAs are common in cancer [ 124 ] and pairwise selective epistasis has been indicated between SNVs and CNAs and other SVs [ 125 , 126 ] or between SVs [ 37 , 87 , 127 ]. Unfortunately, accurate background mutation rates are currently unavailable for SVs. As such rates become calculable, integrated analysis of SNVs and SVs will reveal new epistatic relationships and clarify the somatic adaptive landscape of cancer. Overall, we have demonstrated that the adaptive landscapes of ES- and NS-LUAD diverge substantially, implicating tobacco smoking not only as a mutagen but also as a modulator of somatic selection. Our analysis also shows that symmetric and asymmetric selective epistasis among driver mutations is both substantial and pervasive in lung adenocarcinoma and generates the distinct evolutionary trajectories of ES-LUAD and NS-LUAD. Furthermore, we provide a systematic characterization of higher-order selective epistasis in LUAD, revealing its underrecognized role in shaping tumor evolution. Together, these results show that the contribution of mutations to tumor growth is profoundly context-dependent—modulated by both tumor somatic genetics and the environmental exposures it has experienced. This insight illuminates new opportunities for precision oncology through the integration of somatic genotype and exposure history into treatment strategies. Continued investigation into epistatic, environmental, and even gene-by-gene-by-environment interactions in lung adenocarcinoma and other cancers will be essential for the advancement of our understanding of cancer evolution and the improvement of patient-specific therapeutic strategies. Methods Design overview The aim of this study was to test for differences in background mutation rates, subsequent somatic selection, and gene-by-gene epistatic effects between ES- and NS-LUAD. To do so, we aggregated mutation data of 2,097 LUAD tumor samples from multiple institutional sources. We classified tumors as either ES- or NS-LUAD either by detection of the smoking-associated mutational signature SBS4 in exome or genome sequences or, in the case of targeted panel sequence data, by relying on the patient’s self-reported smoking history. We applied established approaches to estimate background mutation rates, scaled selection coefficients, and pairwise and higher-order selective epistasis within the ES- and NS-LUAD cohorts. To assess the differential transcriptional effects of driver mutations, we compared ES- and NS-LUAD tumors with and without specific mutations. Differences in selection were conservatively evaluated based on non-overlapping 95% confidence intervals. Data sources We sourced clinicogenomic data from (i) eight published datasets hosted by cBioPortal [ 85 , 128 – 135 ], (ii) The Cancer Genome Atlas dataset hosted on Genomic Data Commons [ 136 ], and (iii) 108 samples sequenced at Yale University [ 137 ]. All datasets reported sample sizes, sequencing methods, and clinical covariates, and were incorporated into our analysis accordingly ( Supplementary Table 1 ). We only included samples from primary LUAD tumors, and we did not include any datasets from studies that preselected patients based on their mutational profiles. When multiple samples from a single patient were available, one was selected at random. In cases where patients appeared in more than one dataset, we retained the sample with the highest-quality sequencing and most complete clinical annotations. For consistency across datasets, we standardized all variant genomic coordinates to the hg19 reference genome using the liftOver function of rtracklayer v1.62.0 [ 138 ]. We used cancereffectsizeR v2.10.1 to remove variants that failed liftover, were incorrectly classified as single-nucleotide variants (SNVs), or did not meet quality criteria for downstream analysis. For panel-sequenced samples, we inferred the genomic regions covered by each panel from the gene list provided for each panel using the corresponding hg19 coordinates. Mutational signature attributions and smoking-status classification We performed mutational signature deconvolution for tumor samples with whole-exome or whole-genome sequence data, but not for panel-sequenced samples, due to the low reliability of signature attribution when few silent somatic variants are present. We conducted signature attribution using cancereffectsizeR with MutationalPatterns v3.12.0 [ 139 ], applying COSMIC v3.2 single-base substitution signatures [ 26 , 140 ]. We excluded COSMIC signatures that were unlikely to have been observed in lung adenocarcinomas [ 140 ], and we excluded treatment-associated signatures when deconvolving samples known to be treatment-naive. Samples in which the smoking-associated SBS4 signature contributed more than the MutationalPatterns default threshold of 5% of the total signature burden were classified as ever-smoker ES-LUAD, and otherwise were classified as never-smoker NS-LUAD. Zhang et al. detected no signature of SBS4 in the samples used in their study; therefore we directly classified these samples as NS-LUAD. As in Rosenthal et al. [ 141 ], we excluded samples with ≤50 single-nucleotide variants in whole-exome data from classification due to insufficient mutation burden for reliable signature deconvolution. We compared signature-based smoking assignments to available clinical annotation of smoking status across datasets, validating that there was 67–90% concordance. Discrepancies were likely due to underreporting of smoking, second-hand smoke exposure, minimal mutational consequences in light smokers, or misclassification in clinical records. The selection and epistasis results produced from either classification method were 99.4% consistent with each other. We classified panel-sequenced samples according to clinical annotations of smoking status when available; we excluded samples annotated as exposed to “former light” smoking from all smoking-stratified analyses due to ambiguity in exposure and effect. Estimation of oncogenic mutation rates We estimated gene-specific oncogenic mutation rates separately for the ES- and NS-LUAD cohorts following the approach of Alfaro-Murillo and Townsend [ 28 ]. The oncogenic mutation rate for a gene was defined as the neutral rate at which oncogenic variants occur within the gene. We defined oncogenic variants as variants that were observed at least once in any tumor sample in the dataset. The background oncogenic mutation rate of each gene was computed as the sum of the background rates of all oncogenic variants within the gene. For a specific variant derived from a trinucleotide mutation type m ( e.g. CCT → CAT) at a site i within gene g in a tumor sample cohort c , the somatic background mutation rate where μ g,c is the background mutation rate of the entire gene g (the rate at which variants occur in g ) within cohort c , T m is the event that an arising variant is of trinucleotide mutation type m , and is the event that the variant occurs at site i among all other trinucleotide sequences that match the wild-type (WT) allele of m ( m WT , e.g. CCT), within g . The probability of the variant being of type m given a cohort c , Prob{ T m | c }, is computed as the proportional frequency of m within cohort c , equivalent to the ratio of the number of variants of type m in cohort to the total number of variants in c , N c . Assuming an equal probability of such a variant occurring at all possible sites within g , the probability of such a variant occurring at site is the inverse of the number of sites within g with the same trinucleotide sequence as Therefore The background oncogenic mutation rate of gene g is the sum of the background rates of all oncogenic variants within g : where S g denotes the set of pairs of site indices and mutation types identifying the oncogenic variants in g . This approach enables estimation of mutation rates that are both gene-specific and context-dependent, accounting for trinucleotide mutational biases across the genome. The oncogenic mutation rates of the genes were consistently lower than the mutation rates of the genes. Finally, the oncogenic mutation rates per somatic genotypes were computed by rescaling the per-gene background rates using the tumor mutation burden (TMB) observed in each genotype. Specifically, for a given cohort c , the mutation rate for a jump between the somatic genotype x the somatic genotype y is given by where TMB {x,y}, c is the average TMB in tumors with the somatic genotypes x or y , TMB all is average TMB for all tumors, and g is the gene that is mutated in the somatic genotype y but not in x (the mutation gained by the jump). In cases where one genotype has no samples, the average is taken only from the observed partner. If both endpoints have zero samples, the rate defaults to the baseline per-gene rate genotype classes. We considered TMB only from our largest dataset (TCGA) because after comparing the datasets with the Kolmogorov–Smirnov test, all of them proved to come from different distributions. Estimation of scaled selection coefficients and selective epistasis To quantify the growth and survival advantages conferred by somatic SNVs in LUAD, we calculated scaled selection coefficients for recurrent mutations in a curated set of 1,200 genes [ 136 , 142 – 145 ]. These coefficients roughly equate to the ratio of the fixation rate for the oncogenic mutant (the rate at which the mutant is generated and subsequently selected to a high enough frequency to be observed as a somatic variant within the cancer cell population) to its background mutation rate. Precisely, we computed the fixation rate by analytical maximization of the binomial likelihood of observing at least one non-silent SNV in the gene given the number of samples with and without a non-silent SNV in the gene. Scaled selection coefficients for each mutated gene were estimated separately for ES- and NS-LUAD, as well as for a combined cohort. Within each group, we included samples with whole-exome, whole-genome, or targeted gene panel sequences (TGS). Results were consistent when excluding TGS samples, indicating robustness to sequencing strategy. Inclusion of TGS samples provided greater statistical power to detect significant differences in selection between cohorts. Results were also robust to cohort-specific normalization: applying median-scaling of coefficients prior to between-cohort comparison yielded consistent outcomes. From the curated set of 1,200 genes, we selected 21 for detailed analysis of selective epistasis based on strong prior evidence as LUAD drivers [ 136 , 143 ] and their elevated average strength of selection across somatic genotypes. We quantified epistasis by calculating scaled selection coefficients conditional on the somatic genotype as in Alfaro-Murillo and Townsend [ 28 ]: for each pair or triplet of genes, we calculated the fixation rate of mutations in one gene conditional on the presence or absence of somatic variants in the others. We performed this calculation by maximizing the multinomial likelihood for the number of samples with each somatic genotype using PyMC [ 146 ]. The probability of a tumor being in a specific somatic genotype during the somatic evolutionary trajectory from organogenesis to biopsy was computed via a continuous-time Markov chain model [ 28 ]. We computed asymmetric 95% confidence intervals using Wilks’ theorem [ 147 ]. To obtain scaled selection coefficients for each mutation occurring within each somatic genotype, we divided the respective Poisson-corrected fixation rate [ 148 ] by the background oncogenic mutation rate of the gene. To systematically assess selective epistasis in both the ever-smoker and the never-smoker groups, we evaluated 3,080 models representing all possible two- and three-gene combinations from the 21 LUAD driver genes. For each model, somatic genotypes were defined exclusively by the variant status (mutant or wildtype) of the genes included in each combination. To ensure complete genotype classification, we only included panel-sequenced samples in a given model if the sequencing panel covered all of the genes under analysis. Classification of synergistic and antagonistic selective epistasis To classify a mutant somatic genotype as interacting synergistically or antagonistically with a certain mutation, we compared the scaled selection coefficients for a mutation when it occurred in the mutant somatic genotype to the scaled selection coefficients obtained for the same mutation in a somatic genotype lacking mutations in the genes included in the model. If the scaled selection coefficient for a mutation in gene A when it occurred in a wild-type genotype (∅ → A ) was less than the scaled selection coefficient for a mutation in gene A when it occurred in a somatic genotype with a mutation in gene B ( B → A + B ), then we deemed the epistatic effect of mutant gene B on mutant gene A to be synergistic. If the scaled selection coefficient for ∅ → A was greater than the scaled selection coefficient for B → A + B , then we deemed the epistatic effect of mutant gene B on mutant gene A to be antagonistic. We further distinguished between two scales of antagonism: antagonistic sign epistasis when there was a reversal of the direction of selection—from positive (coefficient γ > 1) to negative selection (coefficient γ < 1)—in the antagonistic genotype, and antagonistic magnitude epistasis when there was a reduction in strength of selection but selection remained positive in the antagonistic genotype. Differential expression analysis To evaluate the transcriptional impact of key driver mutations across smoking-stratified LUAD tumors, we performed differential expression analysis using RNA-seq data from the TCGA-LUAD project. We obtained unstranded raw read counts for 59 normal lung and 539 primary LUAD tumor samples using TCGAbiolinks v2.30.0. We excluded formalin-fixed paraffin-embedded samples and tumors which received neoadjuvant treatment to minimize technical and biological confounders. We retained solid-tissue normal samples as controls. We excluded genes with fewer than 10 total counts across all samples. For each sample, we assigned smoking status based on SBS4 signature weight from the matched MAF. The genes EGFR , KRAS , and KEAP1 were considered mutated if at least one non-silent SNV, insertion, or deletion was present in the gene (excluding intronic regions). After this filtering and annotation, we validated that a principal-component analysis revealed no evidence of batch effects, confirming the suitability of the dataset for differential expression analysis. We performed differential expression analysis with DESeq2 v1.42.1 [ 149 ] using two designs. Differential transcriptional effects of EGFR mutations in ES- and NS-LUAD were quantified as where K i is the count for gene i , and x t , x s , x e , x r are indicator variables for the sample type, smoking status, EGFR mutation status, and KRAS mutation status. The differential effect in question was measured by the interaction term between smoking status and EGFR allele status. The EGFR -focused-design accounts for the fact that most EGFR -wildtype LUADs are KRAS -mutant. Differential transcriptional effects of KEAP1 mutations in ES- and NS-LUAD, were quantified as where x k is the KEAP1 mutation status. We performed gene-set enrichment analysis with fgsea v1.28.0 using the MSigDB Hallmark gene sets. We ranked transcripts by signed t -statistics to capture both degree of significance and magnitude of fold change. We performed gene set variation analysis with GSVA v1.50.5 using the MSigDB canonical gene sets. Statistical Analysis Differences in selection coefficients were assessed conservatively by non-overlapping 95% confidence intervals for each estimate. Differential gene expression was analyzed using two-tailed Wald tests, and statistical significance was determined based on Benjamini-Hochberg–adjusted P < 0.05. Gene set enrichment analyses were likewise considered significant at a Benjamini-Hochberg–adjusted P < 0.05. Analysis was performed in Python (3.9.5) and R (4.3.0). Data and code availability All code, data, and instructions necessary to reproduce the results and figures in the current study is available in a public Zenodo repository, https://zenodo.org/records/16380037 [ 150 ] with no access restrictions. Authors’ contributions JPT, JAA-M, and KD conceptualized the project. KD curated the data and performed the formal analysis. KD and JAA-M wrote the software. KD, JAA-M, and JPT contributed to the methodology and investigation. KD drafted the original manuscript and JPT and JAA-M reviewed and edited the manuscript. All authors read and approved the final manuscripts. Competing interests The authors declare that they have no competing interests. Materials and correspondence Correspondence and material requests should be addressed to Jeffrey Townsend. Funding Support for this research was supplied by a Yale College First-Year Summer Research Fellowship in the Sciences and Engineering to KD as well as by developmental program funds from NIH P50CA196530 and the Elihu Endowment at Yale to JPT. Supporting information Download figure Open in new tab Supplementary Figure 1. Strengths of somatic selection on 1,200 oncogenic gene mutations between lung adenocarcinoma (LUAD) in ever-smokers ( n = 1,066) and in never-smokers ( n = 656; significant: yellow circle, insignificant: no circle, red gene name: stronger selection in ES-LUAD, blue gene name: stronger selection in NS-LUAD). Download figure Open in new tab Supplementary Figure 2. Four patterns of the influences of genetic and environmental interactions in ever-smoker and never-smoker lung adenocarcinoma. ( A ) RB1 mutations with wild-type (WT) or mutant KRAS , ( B ) RB1 mutations with wild-type or mutant EGFR , © MET mutations with wild-type or mutant TP53 , and ( D ) STk11 mutations with wild-type or mutant KEAP1 . View this table: View inline View popup Table S1. Characteristics of datasets included in study. Acknowledgements We thank Jeff Mandell for his advice on mutational signature attribution and mutation rate estimation, and Elizabeth Perry for assistance with gene expression analysis. Funder Information Declared Yale College , First Year Summer Research Fellowship in the Sciences and Engineering Yale School of Public Health , Elihu Endowment National Institutes of Health, https://ror.org/01cwqze88 , P50CA196530 Footnotes Mutation rate calculation methodology updated; Figures 3-6 and corresponding results revised accordingly. Author summary section added. https://github.com/Townsend-Lab-Yale/lung-smoking References 1. ↵ HHS . How Tobacco Smoke Causes Disease: The Biology and Behavioral Basis for Smoking-attributable Disease . U.S. Department of Health and Human Services ; 2010 . 2. ↵ Alexandrov LB , Ju YS , Haase K , Van Loo P , Martincorena I , Nik-Zainal S , et al. Mutational signatures associated with tobacco smoking in human cancer . Science . 2016 ; 354 : 618 – 622 . OpenUrl Abstract / FREE Full Text 3. ↵ Pfeifer GP , Denissenko MF , Olivier M , Tretyakova N , Hecht SS , Hainaut P. Tobacco smoke carcinogens, DNA damage and p53 mutations in smoking-associated cancers . Oncogene . 2002 ; 21 : 7435 – 7451 . OpenUrl CrossRef PubMed Web of Science 4. ↵ Yoshida K , Gowers KHC , Lee-Six H , Chandrasekharan DP , Coorens T , Maughan EF , et al. Tobacco exposure and somatic mutations in normal human bronchial epithelium . Nature . 2020 ; 578 : 266 . OpenUrl CrossRef PubMed 5. ↵ Dasari K , Somarelli JA , Kumar S , Townsend JP . The somatic molecular evolution of cancer: Mutation, selection, and epistasis . Prog Biophys Mol Biol . 2021 ; 165 : 56 – 65 . OpenUrl CrossRef PubMed 6. ↵ Martincorena I , Raine KM , Gerstung M , Dawson KJ , Haase K , Van Loo P , et al. Universal patterns of selection in cancer and somatic tissues . Cell . 2017 ; 171 : 1029 – 1041 .e21. OpenUrl CrossRef PubMed 7. Cannataro VL , Kudalkar S , Dasari K , Gaffney SG , Lazowski HM , Jackson LK , et al. APOBEC mutagenesis and selection for NFE2L2 contribute to the origin of lung squamous-cell carcinoma . Lung Cancer . 2022 ; 171 : 34 – 41 . OpenUrl CrossRef PubMed 8. ↵ Tan C , Mandell JD , Dasari K , Cannataro VL , Alfaro-Murillo JA , Townsend JP . Heavy mutagenesis by tobacco leads to lung adenocarcinoma tumors with KRAS G12 mutations other than G12D, leading KRAS G12D tumors-on average-to exhibit a lower mutation burden . Lung Cancer . 2022 ; 166 : 265 – 269 . OpenUrl CrossRef PubMed 9. ↵ Mochizuki A , Shiraishi K , Honda T , Higashiyama RI , Sunami K , Matsuda M , et al. Passive smoking-induced Mutagenesis as a promoter of lung carcinogenesis . J Thorac Oncol . 2024 ; 19 : 984 – 994 . OpenUrl CrossRef PubMed 10. ↵ Lee J , Taneja V , Vassallo R . Cigarette Smoking and Inflammation: Cellular and Molecular Mechanisms . J Dent Res . 2012 ; 91 : 142 . OpenUrl CrossRef PubMed 11. van der Vaart H , Timens W , Ten Hacken NHT . Acute effects of cigarette smoke on inflammation and oxidative stress: a review . Thorax . 2004 ; 59 : 713 – 721 . OpenUrl Abstract / FREE Full Text 12. ↵ Li X , Li J , Wu P , Zhou L , Lu B , Ying K , et al. Smoker and non-smoker lung adenocarcinoma is characterized by distinct tumor immune microenvironments . Oncoimmunology . 2018 ; 7 : e1494677 . OpenUrl CrossRef PubMed 13. ↵ Rodin SN , Rodin AS . Human lung cancer and p53: the interplay between mutagenesis and selection . Proc Natl Acad Sci U S A . 2000 ; 97 : 12244 – 12249 . OpenUrl Abstract / FREE Full Text 14. ↵ Laconi E , Marongiu F , DeGregori J . Cancer as a disease of old age: changing mutational and microenvironmental landscapes . Br J Cancer . 2020 ; 122 : 943 – 952 . OpenUrl CrossRef PubMed 15. Marusyk A , DeGregori J . Declining cellular fitness with age promotes cancer initiation by selecting for adaptive oncogenic mutations . Biochim Biophys Acta . 2008 ; 1785 : 1 – 11 . OpenUrl CrossRef PubMed 16. Lopez-Bigas N , Gonzalez-Perez A . Are carcinogens direct mutagens? Nat Genet . 2020 ; 52 : 1137 – 1138 . OpenUrl CrossRef PubMed 17. ↵ Stratton MR , Humphreys L , Alexandrov LB , Balmain A , Brennan P , Campbell PJ , et al. Implementing mutational epidemiology on a global scale: Lessons from mutographs . Cancer Discov . 2025 ; 15 : 22 – 27 . OpenUrl PubMed 18. ↵ Okamoto T , Suzuki Y , Fujishita T , Kitahara H , Shimamatsu S , Kohno M , et al. The prognostic impact of the amount of tobacco smoking in non-small cell lung cancer--differences between adenocarcinoma and squamous cell carcinoma . Lung Cancer . 2014 ; 85 : 125 – 130 . OpenUrl CrossRef PubMed 19. ↵ Kenfield SA , Wei EK , Stampfer MJ , Rosner BA , Colditz GA . Comparison of Aspects of Smoking Among Four Histologic Types of Lung Cancer . Tob Control . 2008 ; 17 : 198 . OpenUrl Abstract / FREE Full Text 20. ↵ National Center for Health Statistics . Table 17, Current cigarette smoking among adults aged 18 and over, by sex, race, and age: United States, selected years 1965–2018 . National Center for Health Statistics (US) ; 2021 . 21. ↵ Devarakonda S , Li Y , Rodrigues FM , Sankararaman S , Kadara H , Goparaju C , et al. Genomic Profiling of Lung Adenocarcinoma in Never-Smokers . J Clin Oncol . 2021 [cited 25 Apr 2024]. doi: 10.1200/JCO.21.01691 OpenUrl CrossRef PubMed 22. ↵ Moorthi S , Paguirigan A , Itagi P , Ko M , Pettinger M , Hoge AC , et al. The genomic landscape of lung cancer in never-smokers from the Women’s Health Initiative . JCI Insight . 2024 ; 9 . doi: 10.1172/jci.insight.174643 OpenUrl CrossRef 23. ↵ Wang X , Ricciuti B , Nguyen T , Li X , Rabin MS , Awad MM , et al. Association between smoking history and tumor mutation burden in advanced non-small cell lung cancer . Cancer Res . 2021 ; 81 : 2566 – 2573 . OpenUrl Abstract / FREE Full Text 24. Govindan R , Ding L , Griffith M , Subramanian J , Dees ND , Kanchi KL , et al. Genomic landscape of non-small cell lung cancer in smokers and never-smokers . Cell . 2012 ; 150 : 1121 – 1134 . OpenUrl CrossRef PubMed Web of Science 25. ↵ Díaz-Gay M , Zhang T , Hoang PH , Leduc C , Baine MK , Travis WD , et al. The mutagenic forces shaping the genomes of lung cancer in never smokers . Nature . 2025 ; 1 – 12 . 26. ↵ Blokzijl F , Janssen R , van Boxtel R , Cuppen E . MutationalPatterns: comprehensive genome-wide analysis of mutational processes . Genome Med . 2018 ; 10 : 1 – 11 . OpenUrl CrossRef PubMed 27. ↵ Mandell JD , Cannataro VL , Townsend JP . Estimation of Neutral Mutation Rates and Quantification of Somatic Variant Selection Using cancereffectsizeR . Cancer Res . 2023 ; 83 : 500 – 505 . OpenUrl CrossRef PubMed 28. ↵ Alfaro-Murillo JA , Townsend JP . Pairwise and higher-order epistatic effects among somatic cancer mutations across oncogenesis . Math Biosci . 2023 ; 366 : 109091 . OpenUrl CrossRef PubMed 29. Wang X , Fu AQ , McNerney ME , White KP . Widespread genetic epistasis among cancer genes . Nat Commun . 2014 ; 5 : 1 – 10 . OpenUrl CrossRef PubMed 30. ↵ Iranzo J , Gruenhagen G , Calle-Espinosa J , Koonin EV . Pervasive conditional selection of driver mutations and modular epistasis networks in cancer . Cell Rep . 2022 ; 40 : 111272 . OpenUrl CrossRef PubMed 31. ↵ Blair LM , Juan JM , Sebastian L , Tran VB , Nie W , Wall GD , et al. Oncogenic context shapes the fitness landscape of tumor suppression . Nat Commun . 2023 ; 14 : 1 – 19 . OpenUrl CrossRef PubMed 32. ↵ Mina M , Iyer A , Ciriello G . Epistasis and evolutionary dependencies in human cancers . Curr Opin Genet Dev . 2022 ; 77 : 101989 . 33. ↵ Zhang Z , Yang Y , Zhou Y , Fang H , Yuan M , Sasser K , et al. A forward selection algorithm to identify mutually exclusive alterations in cancer studies . J Hum Genet . 2021 ; 66 : 509 – 518 . OpenUrl CrossRef PubMed 34. ↵ Fedrizzi T , Ciani Y , Lorenzin F , Cantore T , Gasperini P , Demichelis F . Fast mutual exclusivity algorithm nominates potential synthetic lethal gene pairs through brute force matrix product computations . Comput Struct Biotechnol J . 2021 ; 19 : 4394 – 4403 . OpenUrl CrossRef PubMed 35. ↵ Liu S , Liu J , Xie Y , Zhai T , Hinderer EW , Stromberg AJ , et al. MEScan: a powerful statistical framework for genome-scale mutual exclusivity analysis of cancer mutations . Bioinformatics . 2021 ; 37 : 1189 – 1197 . OpenUrl CrossRef PubMed 36. Klein MI , Cannataro VL , Townsend JP , Newman S , Stern DF , Zhao H . Identifying modules of cooperating cancer drivers . Mol Syst Biol . 2021 [cited 2 Jul 2024]. doi: 10.15252/msb.20209810 OpenUrl CrossRef 37. ↵ Mina M , Raynaud F , Tavernari D , Battistello E , Sungalee S , Saghafinia S , et al. Conditional Selection of Genomic Alterations Dictates Cancer Evolution and Oncogenic Dependencies . Cancer Cell . 2017 ; 32 : 155 – 168 .e6. OpenUrl CrossRef PubMed 38. ↵ Kryazhimskiy S , Rice DP , Jerison ER , Desai MM . Global epistasis makes adaptation predictable despite sequence-level stochasticity . Science . 2014 [cited 24 Apr 2024]. doi: 10.1126/science.1250939 OpenUrl Abstract / FREE Full Text 39. Weinreich DM , Lan Y , Wylie CS , Heckendorn RB . Should evolutionary geneticists worry about higher-order epistasis? Curr Opin Genet Dev . 2013 ; 23 : 700 – 707 . OpenUrl CrossRef PubMed 40. Sailer ZR , Harms MJ . High-order epistasis shapes evolutionary trajectories . PLoS Comput Biol . 2017 ; 13 : e1005541 . OpenUrl CrossRef PubMed 41. Weinreich DM , Lan Y , Jaffe J , Heckendorn RB . The influence of higher-order epistasis on biological fitness landscape topography . J Stat Phys . 2018 ; 172 : 208 – 225 . OpenUrl CrossRef PubMed 42. Domingo J , Diss G , Lehner B . Pairwise and higher-order genetic interactions during the evolution of a tRNA . Nature . 2018 ; 558 : 117 – 121 . OpenUrl CrossRef PubMed 43. ↵ Papkou A , Garcia-Pastor L , Escudero JA , Wagner A . A rugged yet easily navigable fitness landscape . Science . 2023 ; 382 : eadh3860. 44. ↵ Wei X , Zhang J . Patterns and Mechanisms of Diminishing Returns from Beneficial Mutations . Mol Biol Evol . 2019 ; 36 : 1008 – 1021 . OpenUrl CrossRef PubMed 45. Aggeli D , Li Y , Sherlock G . Changes in the distribution of fitness effects and adaptive mutational spectra following a single first step towards adaptation . Nat Commun . 2021 ; 12 : 1 – 14 . OpenUrl CrossRef PubMed 46. Chou H-H , Chiu H-C , Delaney NF , Segrè D , Marx CJ . Diminishing returns epistasis among beneficial mutations decelerates adaptation . Science . 2011 ; 332 : 1190 – 1192 . OpenUrl Abstract / FREE Full Text 47. ↵ Wünsche A , Dinh DM , Satterwhite RS , Arenas CD , Stoebel DM , Cooper TF . Diminishing-returns epistasis decreases adaptability along an evolutionary trajectory . Nature Ecology & Evolution . 2017 ; 1 : 1 – 6 . OpenUrl PubMed 48. ↵ Sui Q , Liang J , Hu Z , Chen Z , Bi G , Huang Y , et al. Genetic and microenvironmental differences in non-smoking lung adenocarcinoma patients compared with smoking patients . Transl Lung Cancer Res . 2020 ; 9 : 1407 – 1421 . OpenUrl PubMed 49. ↵ Frank R , Scheffler M , Merkelbach-Bruse S , Ihle MA , Kron A , Rauer M , et al. Clinical and Pathological Characteristics of KEAP1- and NFE2L2-Mutated Non-Small Cell Lung Carcinoma (NSCLC) . Clin Cancer Res . 2018 ; 24 . doi: 10.1158/1078-0432.CCR-17-3416 OpenUrl Abstract / FREE Full Text 50. ↵ Nadal E , Palmero R , Muñoz-Pinedo C . Mutations in the Antioxidant KEAP1/NRF2 Pathway Define an Aggressive Subset of NSCLC Resistant to Conventional Treatments . J Thorac Oncol . 2019 ; 14 . doi: 10.1016/j.jtho.2019.08.005 OpenUrl CrossRef 51. ↵ Subramanian J , Govindan R . Lung Cancer in Never Smokers: A Review . J Clin Oncol . 2016 [cited 25 Apr 2024]. doi: 10.1200/JCO.2006.06.8015 OpenUrl Abstract / FREE Full Text 52. ↵ Scott J , Marusyk A . Somatic clonal evolution: A selection-centric perspective . Biochim Biophys Acta . 2017 ; 1867 : 139 – 150 . OpenUrl 53. ↵ Fleenor CJ , Marusyk A , DeGregori J . Ionizing radiation and hematopoietic malignancies: altering the adaptive landscape . Cell Cycle . 2010 ; 9 : 3005 – 3011 . OpenUrl CrossRef PubMed 54. Hill W , Lim EL , Weeden CE , Lee C , Augustine M , Chen K , et al. Lung adenocarcinoma promotion by air pollutants . Nature . 2023 ; 616 : 159 – 167 . OpenUrl CrossRef PubMed 55. ↵ Bossé Y , Postma DS , Sin DD , Lamontagne M , Couture C , Gaudreault N , et al. Molecular signature of smoking in human lung tissues . Cancer Res . 2012 ; 72 : 3753 – 3763 . OpenUrl CrossRef PubMed 56. Pintarelli G , Noci S , Maspero D , Pettinicchio A , Dugo M , De Cecco L , et al. Cigarette smoke alters the transcriptome of non-involved lung tissue in lung adenocarcinoma patients . Sci Rep . 2019 ; 9 : 13039 . OpenUrl CrossRef PubMed 57. ↵ Ramirez JM , Ribeiro R , Soldatkina O , Moraes A , García-Pérez R , Oliveros W , et al. The molecular impact of cigarette smoking resembles aging across tissues . Genome Medicine . 2025 ; 17 : 1 – 23 . OpenUrl PubMed 58. ↵ Luo W , Zeng Z , Jin Y , Yang L , Fan T , Wang Z , et al. Distinct immune microenvironment of lung adenocarcinoma in never-smokers from smokers . Cell Rep Med . 2023 ; 4 : 101078 . OpenUrl PubMed 59. ↵ Price SN , Land SR , Pebley K , Fahey MC , Palmer AM , McCall MH , et al. Tobacco assessment in actively accruing National Cancer Institute clinical trials network trials . Nicotine Tob Res . 2025 ; ntaf071. 60. ↵ Peters EN , Warren GW , Sloan JA , Marshall JR . Tobacco assessment in completed lung cancer treatment trials: Pooling Project: Smoking and Lung Cancer . Cancer . 2016 ; 122 : 3260 – 3262 . OpenUrl PubMed 61. ↵ Kim IA , Lee JS , Kim HJ , Kim WS , Lee KY . Cumulative smoking dose affects the clinical outcomes of EGFR-mutated lung adenocarcinoma patients treated with EGFR-TKIs: a retrospective study . BMC Cancer . 2018 ; 18 : 768 . OpenUrl CrossRef PubMed 62. ↵ Kim MH , Kim HR , Cho BC , Bae MK , Kim EY , Lee CY , et al. Impact of cigarette smoking on response to epidermal growth factor receptor (EGFR)-tyrosine kinase inhibitors in lung adenocarcinoma with activating EGFR mutations . Lung Cancer . 2014 ; 84 : 196 – 202 . OpenUrl CrossRef PubMed 63. ↵ Zhang P , Nie X , Bie Z , Li L . Impact of heavy smoking on the benefits from first-line EGFR-TKI therapy in patients with advanced lung adenocarcinoma . Medicine . 2018 ; 97 : e0006 . OpenUrl CrossRef PubMed 64. ↵ Igawa S , Sasaki J , Otani S , Ishihara M , Takakura A , Katagiri M , et al. Impact of smoking history on the efficacy of gefitinib in patients with non-small cell lung cancer harboring activating epidermal growth factor receptor mutations . Oncology . 2015 ; 89 : 275 – 280 . OpenUrl CrossRef PubMed 65. ↵ van de Haar J , Canisius S , Yu MK , Voest EE , Wessels LFA , Ideker T . Identifying epistasis in cancer genomes: A delicate affair . Cell . 2019 ; 177 : 1375 – 1383 . OpenUrl CrossRef PubMed 66. ↵ Romero R , Sayin VI , Davidson SM , Bauer MR , Singh SX , LeBoeuf SE , et al. Keap1 loss promotes Kras-driven lung cancer and results in dependence on glutaminolysis . Nat Med . 2017 ; 23 : 1362 – 1368 . OpenUrl CrossRef PubMed 67. ↵ Best SA , Ding S , Kersbergen A , Dong X , Song J-Y , Xie Y , et al. Distinct initiating events underpin the immune and metabolic heterogeneity of KRAS-mutant lung adenocarcinoma . Nat Commun . 2019 ; 10 : 4190 . OpenUrl CrossRef PubMed 68. ↵ Galan-Cobo A , Sitthideatphaiboon P , Qu X , Poteete A , Pisegna MA , Tong P , et al. LKB1 and KEAP1/NRF2 pathways cooperatively promote metabolic reprogramming with enhanced glutamine dependence in KRAS-mutant lung adenocarcinoma . Cancer Res . 2019 ; 79 : 3251 – 3267 . OpenUrl CrossRef PubMed 69. ↵ Ricciuti B , Arbour KC , Lin JJ , Vajdi A , Vokes N , Hong L , et al. Diminished Efficacy of Programmed Death-(Ligand)1 Inhibition in STK11- and KEAP1-Mutant Lung Adenocarcinoma Is Affected by KRAS Mutation Status . J Thorac Oncol . 2022 ; 17 : 399 – 410 . OpenUrl CrossRef PubMed 70. ↵ Kwack WG , Shin SY , Lee SH . Primary resistance to immune checkpoint blockade in an STK11/TP53/KRAS-mutant lung adenocarcinoma with high PD-L1 expression . Onco Targets Ther . 2020 ; 13 : 8901 – 8905 . OpenUrl CrossRef PubMed 71. ↵ Arbour KC , Jordan E , Kim HR , Dienstag J , Yu HA , Sanchez-Vega F , et al. Effects of Co-occurring Genomic Alterations on Outcomes in Patients with KRAS-Mutant Non–Small Cell Lung Cancer . Clin Cancer Res . 2018 ; 24 : 334 – 340 . OpenUrl Abstract / FREE Full Text 72. Proulx-Rocray F , Routy B , Nassabein R , Belkaid W , Tran-Thanh D , Malo J , et al. The prognostic impact of KRAS, TP53, STK11 and KEAP1 mutations and their influence on the NLR in NSCLC patients treated with immunotherapy . Cancer Treat Res Commun . 2023 ; 37 : 100767 . OpenUrl PubMed 73. Baptiste Oudart J , Garinet S , Leger C , Barlesi F , Mazières J , Jeannin G , et al. STK11/LKB1 alterations worsen the poor prognosis of KRAS mutated early-stage non-squamous non-small cell lung carcinoma, results based on the phase 2 IFCT TASTE trial . Lung Cancer . 2024 ; 190 : 107508 . OpenUrl CrossRef PubMed 74. Bange E , Marmarelis ME , Hwang W-T , Yang Y-X , Thompson JC , Rosenbaum J , et al. Impact of KRAS and TP53 co-mutations on outcomes after first-line systemic therapy among patients with STK11-mutated advanced non-small-cell lung cancer . JCO Precis Oncol . 2019 ; 3 : 1 – 11 . OpenUrl 75. West HJ , McCleland M , Cappuzzo F , Reck M , Mok TS , Jotte RM , et al. Clinical efficacy of atezolizumab plus bevacizumab and chemotherapy in KRAS-mutated non-small cell lung cancer with STK11, KEAP1, or TP53 comutations: subgroup results from the phase III IMpower150 trial . J Immunother Cancer . 2022 ; 10 : e003027 . OpenUrl Abstract / FREE Full Text 76. ↵ Aredo JV , Padda SK , Kunder CA , Han SS , Neal JW , Shrager JB , et al. Impact of KRAS mutation subtype and concurrent pathogenic mutations on non-small cell lung cancer outcomes . Lung Cancer . 2019 ; 133 : 144 – 150 . OpenUrl CrossRef PubMed 77. ↵ Wohlhieter CA , Richards AL , Uddin F , Hulton CH , Quintanal-Villalonga À , Martin A , et al. Concurrent mutations in STK11 and KEAP1 promote ferroptosis protection and SCD1 dependence in lung cancer . Cell Rep . 2020 ; 33 : 108444 . OpenUrl CrossRef PubMed 78. ↵ Julian C , Pal N , Gershon A , Evangelista M , Purkey H , Lambert P , et al. Overall survival in patients with advanced non-small cell lung cancer with KRAS G12C mutation with or without STK11 and/or KEAP1 mutations in a real-world setting . BMC Cancer . 2023 ; 23 . doi: 10.1186/s12885-023-10778-6 OpenUrl CrossRef 79. ↵ Boeschen M , Kuhn CK , Wirtz H , Seyfarth H-J , Frille A , Lordick F , et al. Comparative bioinformatic analysis of KRAS, STK11 and KEAP1 (co-)mutations in non-small cell lung cancer with a special focus on KRAS G12C . Lung Cancer . 2023 ; 184 : 107361 . OpenUrl CrossRef PubMed 80. ↵ Bao Y , Zhang S , Zhang X , Pan Y , Yan Y , Wang N , et al. RBM10 Loss Promotes EGFR-Driven Lung Cancer and Confers Sensitivity to Spliceosome Inhibition . Cancer Res . 2023 ; 83 : 1490 – 1502 . OpenUrl CrossRef PubMed 81. ↵ Nanjo S , Wu W , Karachaliou N , Blakely CM , Suzuki J , Chou YT , et al. Deficiency of the splicing factor RBM10 limits EGFR inhibitor response in EGFR-mutant lung cancer . The Journal of clinical investigation . 2022 ; 132 . doi: 10.1172/JCI145099 OpenUrl CrossRef 82. ↵ Gini B , Thomas N , Blakely CM . Impact of concurrent genomic alterations in epidermal growth factor receptor (EGFR)-mutated lung cancer . J Thorac Dis . 2020 ; 12 : 2883 – 2895 . OpenUrl CrossRef PubMed 83. ↵ Offin M , Chan JM , Tenet M , Rizvi HA , Shen R , Riely GJ , et al. Concurrent RB1 and TP53 alterations define a subset of EGFR-mutant lung cancers at risk for histologic transformation and inferior clinical outcomes . J Thorac Oncol . 2019 ; 14 : 1784 – 1793 . OpenUrl CrossRef PubMed 84. ↵ Pros E , Saigi M , Alameda D , Gomez-Mariano G , Martinez-Delgado B , Alburquerque-Bejar JJ , et al. Genome-wide profiling of non-smoking-related lung cancer cells reveals common RB1 rearrangements associated with histopathologic transformation in EGFR-mutant tumors . Ann Oncol . 2020 ; 31 : 274 – 282 . OpenUrl CrossRef PubMed 85. ↵ Jamal-Hanjani M , Wilson GA , McGranahan N , Birkbak NJ , Watkins TBK , Veeriah S , et al. Tracking the Evolution of Non-Small-Cell Lung Cancer . N Engl J Med . 2017 ; 376 : 2109 – 2121 . OpenUrl CrossRef PubMed 86. ↵ Frankell AM , Dietzen M , Al Bakir M , Lim EL , Karasaki T , Ward S , et al. The evolution of lung cancer and impact of subclonal selection in TRACERx . Nature . 2023 ; 616 : 525 – 533 . OpenUrl CrossRef PubMed 87. ↵ Park S , Lehner B . Cancer type-dependent genetic interactions between cancer driver alterations indicate plasticity of epistasis across cell types . Mol Syst Biol . 2015 ; 11 : 824 . OpenUrl Abstract / FREE Full Text 88. ↵ Guénolé A , Srivas R , Vreeken K , Wang ZZ , Wang S , Krogan NJ , et al. Dissection of DNA damage responses using multiconditional genetic interaction maps . Mol Cell . 2013 ; 49 : 346 – 358 . OpenUrl CrossRef PubMed Web of Science 89. Harrison R , Papp B , Pál C , Oliver SG , Delneri D . Plasticity of genetic interactions in metabolic networks of yeast . Proc Natl Acad Sci U S A . 2007 ; 104 : 2307 – 2312 . OpenUrl Abstract / FREE Full Text 90. Zhu C-T , Ingelmo P , Rand DM . G×G×E for lifespan in Drosophila: mitochondrial, nuclear, and dietary interactions that modify longevity . PLoS Genet . 2014 ; 10 : e1004354 . OpenUrl CrossRef PubMed 91. Shi X , Reinstadler B , Shah H , To T-L , Byrne K , Summer L , et al. Combinatorial GxGxE CRISPR screen identifies SLC25A39 in mitochondrial glutathione transport linking iron homeostasis to OXPHOS . Nat Commun . 2022 ; 13 : 2483 . OpenUrl CrossRef PubMed 92. Flynn KM , Cooper TF , Moore FB-G , Cooper VS . The environment affects epistatic interactions to alter the topology of an empirical fitness landscape . PLoS Genet . 2013 ; 9 : e1003426 . OpenUrl CrossRef PubMed 93. ↵ Costanzo M , Hou J , Messier V , Nelson J , Rahman M , VanderSluis B , et al. Environmental robustness of the global yeast genetic interaction network . Science . 2021 ; 372 . doi: 10.1126/science.abf8424 OpenUrl Abstract / FREE Full Text 94. ↵ Sun Y , Yang Q , Shen J , Wei T , Shen W , Zhang N , et al. The Effect of Smoking on the Immune Microenvironment and Immunogenicity and Its Relationship With the Prognosis of Immune Checkpoint Inhibitors in Non-small Cell Lung Cancer . Front Cell Dev Biol . 2021 ; 9 : 745859 . OpenUrl PubMed 95. ↵ Boström M , Larsson E . Somatic mutation distribution across tumour cohorts provides a signal for positive selection in cancer . Nat Commun . 2022 ; 13 : 7023 . OpenUrl CrossRef PubMed 96. Bolton KL , Ptashkin RN , Gao T , Braunstein L , Devlin SM , Kelly D , et al. Cancer therapy shapes the fitness landscape of clonal hematopoiesis . Nat Genet . 2020 ; 52 : 1219 – 1226 . OpenUrl CrossRef PubMed 97. Salehi S , Kabeer F , Ceglia N , Andronescu M , Williams MJ , Campbell KR , et al. Clonal fitness inferred from time-series modelling of single-cell cancer genomes . Nature . 2021 ; 595 : 585 – 590 . OpenUrl CrossRef PubMed 98. Liggett LA , DeGregori J . Changing mutational and adaptive landscapes and the genesis of cancer . Biochim Biophys Acta Rev Cancer . 2017 ; 1867 : 84 – 94 . OpenUrl CrossRef 99. ↵ Hsu T-K , Asmussen J , Koire A , Choi B-K , Gadhikar MA , Huh E , et al. A general calculus of fitness landscapes finds genes under selection in cancers . Genome Res . 2022 ; 32 : 916 – 929 . OpenUrl Abstract / FREE Full Text 100. ↵ Misra N , Szczurek E , Vingron M . Inferring the paths of somatic evolution in cancer . Bioinformatics . 2014 ; 30 : 2456 – 2463 . OpenUrl CrossRef PubMed 101. ↵ Marchetti A , Milella M , Felicioni L , Cappuzzo F , Irtelli L , Del Grammastro M , et al. Clinical implications of KRAS mutations in lung cancer patients treated with tyrosine kinase inhibitors: an important role for mutations in minor clones . Neoplasia . 2009 ; 11 : 1084 – 1092 . OpenUrl CrossRef PubMed Web of Science 102. ↵ Hellyer JA , Stehr H , Das M , Padda SK , Ramchandran K , Neal JW , et al. Impact of KEAP1/NFE2L2/CUL3 mutations on duration of response to EGFR tyrosine kinase inhibitors in EGFR mutated non-small cell lung cancer . Lung Cancer . 2019 ; 134 : 42 – 45 . OpenUrl CrossRef PubMed 103. ↵ Krall EB , Wang B , Munoz DM , Ilic N , Raghavan S , Niederst MJ , et al. KEAP1 loss modulates sensitivity to kinase targeted therapy in lung cancer . Elife . 2017 ; 6 . doi: 10.7554/eLife.18970 OpenUrl CrossRef PubMed 104. ↵ Iyer G , Hanrahan AJ , Milowsky MI , Al-Ahmadie H , Scott SN , Janakiraman M , et al. Genome sequencing identifies a basis for everolimus sensitivity . Science . 2012 ; 338 : 221 . OpenUrl Abstract / FREE Full Text 105. ↵ Weigelt B , Reis-Filho JS . Epistatic interactions and drug response . J Pathol . 2014 ; 232 : 255 – 263 . OpenUrl CrossRef PubMed Web of Science 106. ↵ Wilkins JF , Cannataro VL , Shuch B , Townsend JP . Analysis of mutation, selection, and epistasis: an informed approach to cancer clinical trials . Oncotarget . 2018 ; 9 : 22243 – 22253 . OpenUrl CrossRef PubMed 107. ↵ Hobor S , Al Bakir M , Hiley CT , Skrzypski M , Frankell AM , Bakker B , et al. Mixed responses to targeted therapy driven by chromosomal instability through p53 dysfunction and genome doubling . Nat Commun . 2024 ; 15 : 4871 . OpenUrl CrossRef PubMed 108. Bria E , Pilotto S , Amato E , Fassan M , Novello S , Peretti U , et al. Molecular heterogeneity assessment by next-generation sequencing and response to gefitinib of EGFR mutant advanced lung adenocarcinoma . Oncotarget . 2015 ; 6 : 12783 – 12795 . OpenUrl CrossRef PubMed 109. ↵ Vokes NI , Chambers E , Nguyen T , Coolidge A , Lydon CA , Le X , et al. Concurrent TP53 Mutations Facilitate Resistance Evolution in EGFR-Mutant Lung Adenocarcinoma . J Thorac Oncol . 2022 ; 17 : 779 – 792 . OpenUrl CrossRef PubMed 110. ↵ Canale M , Petracci E , Delmonte A , Chiadini E , Dazzi C , Papi M , et al. Impact of TP53 Mutations on Outcome in EGFR-Mutated Patients Treated with First-Line Tyrosine Kinase Inhibitors . Clin Cancer Res . 2017 ; 23 : 2195 – 2202 . OpenUrl Abstract / FREE Full Text 111. Ibusuki R , Iwama E , Shimauchi A , Tsutsumi H , Yoneshima Y , Tanaka K , et al. TP53 gain-of-function mutations promote osimertinib resistance via TNF-α-NF-κB signaling in EGFR-mutated lung cancer . NPJ Precis Oncol . 2024 ; 8 : 60 . OpenUrl PubMed 112. Canale M , Petracci E , Delmonte A , Bronte G , Chiadini E , Ludovini V , et al. Concomitant TP53 Mutation Confers Worse Prognosis in EGFR-Mutated Non-Small Cell Lung Cancer Patients Treated with TKIs . J Clin Med Res . 2020 ; 9 . doi: 10.3390/jcm9041047 OpenUrl CrossRef 113. Choudhury NJ , Lavery JA , Brown S , de Bruijn I , Jee J , Tran TN , et al. The GENIE BPC NSCLC cohort: A real-world repository integrating standardized clinical and genomic data for 1,846 patients with non-small cell lung cancer . Clin Cancer Res . 2023 ; 29 : 3418 – 3428 . OpenUrl CrossRef PubMed 114. ↵ Fu J , Tong Y , Xu Z , Li Y , Zhao Y , Wang T , et al. Impact of TP53 mutations on EGFR-tyrosine kinase inhibitor efficacy and potential treatment strategy . Clin Lung Cancer . 2023 ; 24 : 29 – 39 . OpenUrl PubMed 115. ↵ Donehower LA , Soussi T , Korkut A , Liu Y , Schultz A , Cardenas M , et al. Integrated analysis of TP53 gene and pathway alterations in The Cancer Genome Atlas . Cell Rep . 2019 ; 28 : 1370 – 1384 .e5. OpenUrl CrossRef PubMed 116. ↵ Rho JK , Choi YJ , Ryoo B-Y , Na III , Yang SH , Kim CH , et al. p53 enhances gefitinib-induced growth inhibition and apoptosis by regulation of Fas in non-small cell lung cancer . Cancer Res . 2007 ; 67 : 1163 – 1169 . OpenUrl Abstract / FREE Full Text 117. ↵ Fearon ER , Vogelstein B . A genetic model for colorectal tumorigenesis . Cell . 1990 ; 61 : 759 – 767 . OpenUrl CrossRef PubMed Web of Science 118. ↵ Paterson C , Clevers H , Bozic I . Mathematical model of colorectal cancer initiation . Proc Natl Acad Sci U S A . 2020 ; 117 : 20681 – 20688 . OpenUrl Abstract / FREE Full Text 119. ↵ Ortmann CA , Kent DG , Nangalia J , Silber Y , Wedge DC , Grinfeld J , et al. Effect of mutation order on myeloproliferative neoplasms . N Engl J Med . 2015 ; 372 : 601 – 612 . OpenUrl CrossRef PubMed Web of Science 120. ↵ Ernst SM , Mankor JM , van Riet J , von der Thüsen JH , Dubbink HJ , Jgjv A , et al. Tobacco Smoking-Related Mutational Signatures in Classifying Smoking-Associated and Nonsmoking-Associated NSCLC . J Thorac Oncol . 2023 ; 18 . doi: 10.1016/j.jtho.2022.11.030 OpenUrl CrossRef PubMed 121. ↵ Hobbs SD , Wilmink AB , Adam DJ , Bradbury AW . Assessment of smoking status in patients with peripheral arterial disease . J Vasc Surg . 2005 ; 41 . doi: 10.1016/j.jvs.2004.12.039 OpenUrl CrossRef PubMed Web of Science 122. ↵ Hald J , Overgaard J , Grau C . Evaluation of objective measures of smoking status--a prospective clinical study in a group of head and neck cancer patients treated with radiotherapy . Acta Oncol . 2003 ; 42 . doi: 10.1080/02841860310005020 OpenUrl CrossRef PubMed Web of Science 123. ↵ Brennan P , Buffler PA , Reynolds P , Wu AH , Wichmann HE , Agudo A , et al. Secondhand smoke exposure in adulthood and risk of lung cancer among never smokers: a pooled analysis of two large studies . International journal of cancer . 2004 ; 109 . doi: 10.1002/ijc.11682 OpenUrl CrossRef PubMed Web of Science 124. ↵ Beroukhim R , Mermel CH , Porter D , Wei G , Raychaudhuri S , Donovan J , et al. The landscape of somatic copy-number alteration across human cancers . Nature . 2010 ; 463 : 899 – 905 . OpenUrl CrossRef PubMed Web of Science 125. ↵ Besedina E , Supek F . Copy number losses of oncogenes and gains of tumor suppressor genes generate common driver mutations . Nat Commun . 2024 ; 15 : 6139 . OpenUrl CrossRef PubMed 126. ↵ Tomanek I , Guet CC . Adaptation dynamics between copy-number and point mutations . Elife . 2022 ; 11 . doi: 10.7554/eLife.82240 OpenUrl CrossRef PubMed 127. ↵ Ciriello G , Cerami E , Sander C , Schultz N . Mutual exclusivity analysis identifies oncogenic network modules . Genome Res . 2012 ; 22 : 398 – 406 . OpenUrl Abstract / FREE Full Text 128. ↵ Chen J , Yang H , Teo ASM , Amer LB , Sherbaf FG , Tan CQ , et al. Genomic landscape of lung adenocarcinoma in East Asians . Nat Genet . 2020 ; 52 : 177 – 186 . OpenUrl CrossRef PubMed 129. Imielinski M , Berger AH , Hammerman PS , Hernandez B , Pugh TJ , Hodis E , et al. Mapping the hallmarks of lung adenocarcinoma with massively parallel sequencing . Cell . 2012 ; 150 : 1107 – 1120 . OpenUrl CrossRef PubMed Web of Science 130. Jordan EJ , Kim HR , Arcila ME , Barron D , Chakravarty D , Gao J , et al. Prospective Comprehensive Molecular Characterization of Lung Adenocarcinomas for Efficient Patient Matching to Approved and Emerging Therapies . Cancer Discov . 2017 ; 7 : 596 – 609 . OpenUrl Abstract / FREE Full Text 131. Rizvi NA , Hellmann MD , Snyder A , Kvistborg P , Makarov V , Havel JJ , et al. Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer . Science . 2015 ; 348 : 124 – 128 . OpenUrl Abstract / FREE Full Text 132. Rizvi H , Sanchez-Vega F , La K , Chatila W , Jonsson P , Halpenny D , et al. Molecular Determinants of Response to Anti-Programmed Cell Death (PD)-1 and Anti-Programmed Death-Ligand 1 (PD-L1) Blockade in Patients With Non-Small-Cell Lung Cancer Profiled With Targeted Next-Generation Sequencing . J Clin Oncol . 2018 ; 36 : 633 – 641 . OpenUrl CrossRef PubMed 133. Zhang T , Joubert P , Ansari-Pour N , Zhao W , Hoang PH , Lokanga R , et al. Genomic and evolutionary classification of lung cancer in never smokers . Nat Genet . 2021 ; 53 : 1348 . OpenUrl CrossRef PubMed 134. Gillette MA , Satpathy S , Cao S , Dhanasekaran SM , Vasaikar SV , Krug K , et al. Proteogenomic characterization reveals therapeutic vulnerabilities in lung adenocarcinoma . Cell . 2020 ; 182 : 200 – 225 .e35. OpenUrl CrossRef PubMed 135. ↵ Gao J , Aksoy BA , Dogrusoz U , Dresdner G , Gross B , Sumer SO , et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal . Sci Signal . 2013 ; 6 : l1 . 136. ↵ Cancer Genome Atlas Research Network . Comprehensive molecular profiling of lung adenocarcinoma . Nature . 2014 ; 511 : 543 – 550 . OpenUrl CrossRef PubMed Web of Science 137. ↵ Kadara H , Choi M , Zhang J , Parra ER , Rodriguez-Canales J , Gaffney SG , et al. Whole-exome sequencing and immune profiling of early-stage lung adenocarcinoma with fully annotated clinical follow-up . Ann Oncol . 2017 ; 28 : 75 – 82 . OpenUrl CrossRef PubMed 138. ↵ Lawrence M , Gentleman R , Carey V. rtracklayer: an R package for interfacing with genome browsers . Bioinformatics . 2009 ; 25 : 1841 – 1842 . OpenUrl CrossRef PubMed Web of Science 139. ↵ Manders F , Brandsma AM , de Kanter J , Verheul M , Oka R , van Roosmalen MJ , et al. MutationalPatterns: the one stop shop for the analysis of mutational processes . BMC Genomics . 2022 ; 23 : 134 . OpenUrl CrossRef PubMed 140. ↵ Alexandrov LB , Kim J , Haradhvala NJ , Huang MN , Tian Ng AW , Wu Y , et al. The repertoire of mutational signatures in human cancer . Nature . 2020 ; 578 : 94 – 101 . OpenUrl CrossRef PubMed 141. ↵ Rosenthal R , McGranahan N , Herrero J , Taylor BS , Swanton C . deconstructSigs: delineating mutational processes in single tumors distinguishes DNA repair deficiencies and patterns of carcinoma evolution . Genome Biol . 2016 ; 17 : 1 – 11 . OpenUrl CrossRef PubMed 142. ↵ Devarakonda S , Morgensztern D , Govindan R . Genomic alterations in lung adenocarcinoma . Lancet Oncol . 2015 ; 16 . doi: 10.1016/S1470-2045(15)00077-7 OpenUrl CrossRef PubMed 143. ↵ Greulich H . The Genomics of Lung Adenocarcinoma: Opportunities for Targeted Therapies . Genes Cancer . 2010 ; 1 : 1200 . OpenUrl PubMed 144. Seo JS , Ju YS , Lee WC , Shin JY , Lee JK , Bleazard T , et al. The transcriptional landscape and mutational profile of lung adenocarcinoma . Genome Res . 2012 ; 22 . doi: 10.1101/gr.145144.112 OpenUrl Abstract / FREE Full Text 145. ↵ Li S , Choi YL , Gong Z , Liu X , Lira M , Kan Z , et al. Comprehensive Characterization of Oncogenic Drivers in Asian Lung Adenocarcinoma . J Thorac Oncol . 2016 ; 11 . doi: 10.1016/j.jtho.2016.08.142 OpenUrl CrossRef 146. ↵ Abril-Pla O , Andreani V , Carroll C , Dong L , Fonnesbeck CJ , Kochurov M , et al. PyMC: a modern, and comprehensive probabilistic programming framework in Python . PeerJ Comput Sci . 2023 ; 9 : e1516 . OpenUrl CrossRef 147. ↵ Wilks SS. The Large-Sample Distribution of the Likelihood Ratio for Testing Composite Hypotheses . aoms . 1938 ; 9 : 60 – 62 . OpenUrl 148. ↵ Cannataro VL , Gaffney SG , Townsend JP . Effect sizes of somatic mutations in cancer . J Natl Cancer Inst . 2018 ; 110 : 1171 – 1177 . OpenUrl CrossRef PubMed 149. ↵ Love MI , Huber W , Anders S . Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2 . Genome Biol . 2014 ; 15 : 550 . OpenUrl CrossRef PubMed 150. ↵ Dasari K . Townsend-Lab-Yale/lung-smoking: Software for “Tobacco smoke alters the trajectory of lung adenocarcinoma evolution via effects on somatic selection and epistasis.” [cited 23 Jul 2025]. doi: 10.5281/zenodo.16379015 View the discussion thread. Back to top Previous Next Posted December 12, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Tobacco smoke alters the trajectory of lung adenocarcinoma evolution via effects on somatic selection and epistasis Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Tobacco smoke alters the trajectory of lung adenocarcinoma evolution via effects on somatic selection and epistasis Krishna Dasari , Jorge A. Alfaro-Murillo , Jeffrey P. Townsend bioRxiv 2024.11.27.625765; doi: https://doi.org/10.1101/2024.11.27.625765 Share This Article: Copy Citation Tools Tobacco smoke alters the trajectory of lung adenocarcinoma evolution via effects on somatic selection and epistasis Krishna Dasari , Jorge A. Alfaro-Murillo , Jeffrey P. Townsend bioRxiv 2024.11.27.625765; doi: https://doi.org/10.1101/2024.11.27.625765 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Genomics Subject Areas All Articles Animal Behavior and Cognition (7644) Biochemistry (17725) Bioengineering (13915) Bioinformatics (42031) Biophysics (21486) Cancer Biology (18635) Cell Biology (25548) Clinical Trials (138) Developmental Biology (13397) Ecology (19937) Epidemiology (2067) Evolutionary Biology (24361) Genetics (15619) Genomics (22538) Immunology (17763) Microbiology (40468) Molecular Biology (17206) Neuroscience (88733) Paleontology (667) Pathology (2842) Pharmacology and Toxicology (4834) Physiology (7657) Plant Biology (15172) Scientific Communication and Education (2046) Synthetic Biology (4304) Systems Biology (9831) Zoology (2272)