Immune Surveillance in Clinical Regression of Preinvasive Squamous Cell Lung Cancer

Before the development of lung squamous cell carcinoma (LUSC), preinvasive lesions can be observed in the airways. These evolve stepwise, progressing through mild and moderate dysplasia (low-grade lesions) to severe dysplasia and carcinoma in situ (CIS; high-grade lesions), before the development of invasive cancer (1). In cross-sectional studies, markers of immune sensing and escape have been associated with increasing grade (2). However, longitudinal bronchoscopic surveillance of such lesions has shown that progression of preinvasive lesions to cancer is not inevitable; only half of high-grade CIS lesions will progress to cancer within 2 years, whereas a third will spontaneously regress (3). Our previous work defined the genomic, transcriptomic, and epigenetic landscape of carefully phenotyped airway CIS lesions (4). Here, we combine these data with IHC, imaging, and transcriptomic analysis of adjacent stroma (Supplementary Table S1; Supplementary Fig. S1) to assess the role of immune surveillance in lesion regression. We identify key immune escape mechanisms enriched in preinvasive lesions that later progressed to cancer. Understanding these mechanisms may offer new therapeutic strategies to induce regression and prevent the development of invasive disease.

To assess our hypothesis that lesion regression is driven by immune surveillance, we used a deep-learning approach (5) to quantify lymphocytes from hematoxylin and eosin (H&E)–stained slides in a large dataset of 112 samples from 62 patients, which contained more infiltrating lymphocytes in regressive lesions than progressive (Fig. 1A; P = 0.049). We next performed IHC on 28 progressive and 16 regressive CIS lesions from 29 patients (Fig. 1B and C). Regressive lesions showed higher concentrations of intralesional CD8⁺ cytotoxic T cells (Fig. 1A; P = 0.055) but no significant difference in CD4⁺ Th cells (P = 0.26) or FOXP3⁺ regulatory T cells (P = 0.42). We then quantified immune cells in stromal regions adjacent to CIS lesions, but found no significant differences between progressive and regressive lesions for CD8⁺ (P = 0.50), CD4⁺ (P = 0.43), or FOXP3⁺ (P = 0.64) cells. We confirmed these findings in an independent dataset of 19 progressive and 9 regressive samples subjected to multiplex IHC (mIHC; refs. 6, 7) using a wider antibody panel of lymphoid biomarkers (Supplementary Table S2), in which we again observed that regressive lesions had an increased proportion of infiltrating lymphocytes (Fig. 2A; P = 0.032). Specifically, regressive lesions showed significantly more infiltrating CD3⁺CD8⁺ cytotoxic T cells (P = 0.017), but no significant difference in CD3⁺CD4⁺ Th cells (P = 0.18), T regulatory cells (P = 0.12), B cells (P = 0.12), macrophages (P = 0.79), or neutrophils (P = 0.53). In the mIHC cohort, the proportion of CD3⁺CD8⁺ cells positive for granzyme B and EOMES was similar between progressive and regressive lesions (P = 0.63 and P = 0.18, respectively), which may indicate that disruption of T-cell infiltration into lesions has a greater impact on their capacity for immune evasion than impairment of cytotoxic function or differentiation. Again, stromal regions in this cohort showed no significant differences between progressive and regressive lesions.

For a broader assessment of transcriptomic differences between CIS lesions and their adjacent stroma, we isolated epithelial tissue and paired stroma separately using laser capture microdissection for 10 progressive and 8 regressive CIS lesions. Similarly to IHC data, cell-type deconvolution analysis using the Danaher method (8) demonstrated higher infiltrating lymphocytes within regressive lesions (Fig. 2B; P = 0.0046), as did deconvolution of methylation data from 36 progressive and 18 regressive CIS lesions using methylCIBERSORT (ref. 9; Fig. 2C; P = 0.0081). Comparing predictions for individual cell types across gene expression and methylation data revealed an increase in most immune cell types in regressive lesions compared with progressive (Supplementary Table S3).

Analysis of cytokines classically considered to be pro- or anti-inflammatory within the epithelial compartment (Supplementary Table S4) demonstrated an increase in proinflammatory (P = 3.7 × 10⁻⁴), but not anti-inflammatory (P = 0.32) response in regressive lesions compared with progressive (Supplementary Fig. S2A–S2F). IL2, TNF, IL12A, and IL23A were all increased in regressive lesions (Supplementary Fig. S3A and S3B; FDR = 0.0081, FDR = 0.00051, FDR = 0.00078, FDR = 0.011, respectively). Only CXCL8 was upregulated in progressive samples compared with regressive (FDR = 0.0063); produced by macrophages, the expression of CXCL8 correlated strongly with macrophage quantification from deconvoluted gene expression data (r² = 0.62, P = 0.007). Taken together, these data are in keeping with a model in which inflammation via pathways including IL2 and TNF fosters effective immune surveillance, while lesion-associated macrophages, similar to tumor-associated macrophages in advanced cancers, have an immunosuppressive effect.

Given the well-known immunosuppressive effects of smoking, we hypothesized that patients who were current smokers were more likely to show reduced immune infiltrate and therefore a higher chance of progression. Smoking status was available for 132 CIS lesions from 59 patients (24 lesions from 13 current smokers; 104 from 43 former smokers; 4 from 3 never-smokers; Supplementary Fig. S4A–S4J). Using a Cochrane–Armitage test to look for a trend from current to former to never-smokers, we found a trend toward higher chance of regression (P = 0.002) and more infiltrating lymphocytes (P = 0.095). This trend is still observed, yet no longer statistically significant, using a bootstrapping method to account for samples from the same patient (P = 0.069 for regression; P = 0.12 for infiltrating lymphocytes). Interestingly, within the former smoker group, we did not observe increasing lymphocytes or chance of regression with increasing time since quitting smoking, suggesting that the observed differences in outcome are driven by the active process of smoking and its direct effects on the immune response, rather than by chronic processes of airway remodeling and repair (10).

Recent advances have demonstrated heterogeneity of lung cancer immune infiltration, with patients whose tumors have predominantly infiltrated, “immune hot” regions having improved survival as compared with those with abundant poorly infiltrated, “immune cold” regions (11, 12). Hierarchical clustering of immune cell quantification by mIHC and by deconvolution of both transcriptomic and epigenetic data demonstrated clear clusters of “cold” lesions, almost all of which progressed to cancer (Fig. 2D–F). However, we also observed some “hot” progressive lesions, suggesting the presence of other immune evasion mechanisms in these lesions. We therefore sought to address two questions: First, could deficits in antigen presentation and immune recruitment in progressive lesions be identified, which could explain the observed “cold” lesions? Second, could disordered immune cell function explain the existence of progressive immune “hot” lesions?

The acquisition of mutations that result in clonal neoantigens drives T-cell immunoreactivity in cancer (13). We hypothesized that immune-active regressive lesions may contain more neoantigens than progressive lesions; however, this was not supported by whole-genome sequencing data (ref. 4; n = 39). Predicted neoantigens correlated very closely with mutational burden (r² = 0.94), and progressive lesions have been shown to have significantly higher mutational burden than regressive lesions (4); therefore, more neoantigens were identified in progressive than regressive lesions (Supplementary Fig. S5A and S5B; P = 0.088). This remained true when the analysis was limited to clonal neoantigens (Supplementary Fig. S5C; P = 0.023) and there was no difference in the proportion of neoantigens that were clonal (Supplementary Fig. S5D; P = 0.76). Furthermore, there were no significant differences in binding affinity (P = 0.46) or differential agretopicity index (ref. 14; P = 0.58), and the ratio of observed to expected neoantigens (“depletion score”; ref. 15) was not significantly different (Supplementary Fig. S5E–S5H; P = 0.94); therefore, the putative neoantigens themselves were not qualitatively different in the regressive group. The increased number of neoantigens identified in progressive lesions suggests that immune escape mechanisms must be active in these lesions; indeed, the presence of these antigens may act as a selective pressure to promote the development of immune escape (16). Importantly, no overlap in putative tumor neoantigens was observed between different patients, suggesting that vaccine-based approaches aiming to prevent progression will most likely need to be designed on a personalized basis.

Given that neoantigens are present in progressive lesions, we assessed the ability of these lesions to present antigens to the immune system. In cancer, genomic alterations have previously been associated with modulation of immune response (17, 18). We studied mutations in and copy-number burden of 62 genes expressed by cancer cells that are involved in the following pathways: antigen presentation by MHC mechanisms, antigen processing, and immunomodulation (stimulation and inhibition of T-cell responses; Fig. 3). Mutations and copy-number aberrations (CNA) in these genes were more prevalent in progressive than regressive samples (P = 0.003). Four of these genes—B2M, CHUK, KDR, and CD80—had a significantly elevated dN/dS ratio (19), a comparison of the rates of nonsynonymous to synonymous mutations, indicating positive selection for acquisition of mutations in these genes. We observe that expression of immunostimulatory genes predominantly positively correlates with infiltrating lymphocytes in CIS, and these genes are mostly downregulated in progressive compared with regressive CIS. Conversely, inhibitory genes predominantly correlate negatively with infiltrating lymphocytes and are upregulated in progressive lesions.

Loss of heterozygosity (LOH) in the HLA region, which is found in 61% of patients with LUSC (20), was identified in 34% of patients with CIS lesions. Interestingly, a similar proportion of patients with LUSC (28%) demonstrated clonal HLA LOH (20), suggesting that such clonal events may often occur prior to tumor invasion; future longitudinal studies will be required to confirm this. We did not find a statistically significant difference in the prevalence of HLA LOH between progressive and regressive lesions (P = 0.25), although sample numbers were small. Expression of HLA-A was significantly reduced in progressive compared with regressive lesions (P = 1.9 × 10⁻¹⁰).

In addition, hypermethylation of the HLA region, which is well described in invasive cancers (21, 22), was commonly observed, suggesting that epigenetic HLA silencing may be an important immune escape mechanism in preinvasive disease. Genome-wide differential methylation analysis between progressive and regressive lesions identified differentially methylated regions (DMR) including a striking cluster of hypermethylation in chromosome 6 (ref. 4; Supplementary Fig. S6A and S6B), covering a region containing all of the major HLA genes. This cluster was also identified in analysis of 370 LUSC versus 42 control samples published by The Cancer Genome Atlas (TCGA; ref. 23). Further analysis of TCGA data demonstrates strong evidence for epigenetic silencing of multiple genes in the antigen presentation pathway: mean methylation beta value over the gene is inversely correlated with expression for HLA-A (r² = −0.32, P = 2.5 × 10⁻¹⁰), HLA-B (r² = −0.42, P < 2.2 × 10⁻¹⁶), HLA-C (r² = −0.18, P = 3.6 × 10⁻⁴), TAP1 (r² = −0.53, P < 2.2 × 10⁻¹⁶), and B2M (r² = −0.38, P = 1.1 × 10⁻¹⁴). Similar trends were observed in CIS data (Supplementary Fig. S7A and S7B). The methylation pattern affecting these genes is predominantly promoter hypermethylation (Supplementary Fig. S8).

Demethylating agents have been shown to promote immune activation through improved antigen presentation, immune migration, and T-cell activity (24–26). These data support the case for moving ongoing trials of demethylating agents in combination with immunotherapy from advanced lung cancer into early disease (examples of such trials include NCT01928576 and NCT03220477, registered at https://clinicaltrials.gov/). In addition, several other cancer-associated pathways are known to be affected by methylation changes (4); therefore, the benefits of these drugs may extend beyond immune activation. Nevertheless, we note with caution that some key immune genes demonstrate positive correlations in TCGA data between gene expression and methylation, including the immune costimulating ligand gene TNFSF9 (coding for 4-1BBL; r² = 0.32, P = 1.7 × 10⁻¹⁰) and the MHC class II transcriptional activator gene CIITA (r² = 0.39, P = 2.5 × 10⁻¹⁵; Supplementary Fig. S7). Further studies will be required to demonstrate that immunologic benefits of demethylating agents are not outweighed by effects on these important pathways.

Despite this evidence for impairment of antigen presentation mechanisms in CIS, we do observe “immune hot” CIS lesions which progress to cancer. We therefore next considered functional and microenvironment-related mechanisms to explain how these lesions were able to evade immune predation.

To study microenvironment effects on the immune response, we performed gene expression profiling on laser-captured stromal tissue taken from regions adjacent to CIS lesions. In contrast to data from gastrointestinal preinvasive lesions (27), no genes were significantly differentially expressed on comparing stromal expression between progressive (n = 10) and regressive (n = 8) lesions when a FDR of <0.1 was applied. This result holds true with restricted hypothesis testing considering only genes that are related to immunity and inflammation (Fig. 4A and B; Supplementary Table S4).

Targeting immunomodulatory molecules such as PD-1 now forms part of first-line lung cancer management (28). PD-L1 expression is common in invasive LUSC with estimates of positivity ranging from 34% to 52%, depending on criteria (29). Although we did not identify transcriptional upregulation of the gene encoding PD-L1 (CD274; Fig. 4C and D), IHC data identified 3 samples with >25% of epithelial cells (PanCK⁺) also positive for PD-L1 (Fig. 4E), all of which progressed to cancer, suggesting that targeting this pathway early in the clinical course may have therapeutic benefit in selected patients.

To investigate the role of immunomodulatory molecules more broadly in preinvasive immune escape, we performed differential expression analysis between progressive and regressive lesions, focusing on 28 known immunomodulatory genes (Supplementary Table S4). TNFSF9 (encoding 4-1BBL, also known as CD137L) was significantly downregulated in progressive lesions (FDR = 4.34 × 10⁻⁵; Fig. 4C and D) with no corresponding change identified in its receptor TNFRSF9 (FDR = 0.6). These findings were corroborated by IHC (Fig. 4E and F). TNFSF9 promotes activation of T cells and natural killer (NK) cells (30); in CIS lesions, TNFSF9 expression correlates with cytotoxic cell (r² = 0.77, P = 0.0002) and NK-cell infiltration (r² = 0.54, P = 0.02), as predicted from gene expression data. Agonists of 4-1BB have been shown to be clinically efficacious in several cancers (31–33), and these data support their investigation in targeted early lung cancer cohorts. Furthermore, individual lesions showed notably high or low expression of other immunomodulatory genes, raising the possibility that other immunomodulators may be targets for therapy in individual cases (Supplementary Fig. S9).

To identify differences in cytokine responses between progressive and regressive lesions, we calculated the ligand:receptor mRNA expression ratio for 52 known cytokine–receptor pairs (34). Only one, CCL27–CCR10, was significant with FDR < 0.01 (fold change 1.55, FDR 0.003); progressive samples expressed more CCL27 (P = 2.6 × 10⁻⁶) and less CCR10 (P = 0.1 × 10⁻⁴) than regressive samples (Fig. 4C and D). Although sample numbers were small, these findings were broadly supported by IHC (Fig. 4E–G). CCL27–CCR10 signaling has been associated with immune escape in melanoma through PI3K–AKT activation in a mouse model (35); in CIS, CCL27 expression correlates with expression of both PIK3CA (r² = 0.61; P = 0.008) and AKT1 (r² = 0.68, P = 0.002; Supplementary Fig. S10A and S10B). CCL27 is minimally expressed in both normal lung tissue and invasive squamous cell lung cancer (23, 36), suggesting that this effect is specific to early carcinogenesis and therefore warrants further investigation as a target for preventative therapy.

Our previous research highlighted occasional cases of “late progressive” lesions, which met a clinical endpoint of regression (defined by the subsequent biopsy at the same site showing resolution to normal epithelium or low-grade dysplasia), but the index CIS biopsy had the molecular appearance of a progressive lesion, and it indeed subsequently developed into cancer months or years later. Clinical review identified 11 lesions across the 53 regressive lesions in our current cohort (20.7%) that at later clinical follow-up subsequently progressed to cancer, and hence are termed “late progressive.” These included 4 previously published lesions subjected to whole-genome sequencing and/or methylation and shown to display the genomically unstable appearance of progressive lesions, as well as 7 with IHC data and 10 with lymphocyte quantification performed from H&E slides (Supplementary Table S1; Supplementary Fig. S1). Interestingly, based on these data, late progressive lesions appear immunologically similar to regressive lesions, showing increased infiltration with lymphocytes and CD8⁺ T cells compared with progressive lesions (Supplementary Fig. S11).

Although we acknowledge that sample numbers are small when examining subgroups of regressive lesions in this way, our data support a model in which lesions can be considered on two axes: genomic stability and immune competence. Our previous work predicts that chromosomally unstable lesions will usually progress, implying that they have escaped immune predation. Yet some may regress if they remain immune competent only to later progress, potentially due to their genomic instability making them more likely to evolve immune escape mechanisms during regression and hence become “late progressors.” Of 11 late progressors in this cohort, median time from regressive index biopsy to progression was 3.2 years (range 0.8–4.6 years). This time period represents a change from a point of known immune competence to demonstrated immune escape. Hence, we might estimate that a successful therapeutic strategy to block a particular immune escape mechanism might delay the onset of cancer by around 3 years. Of the remaining 42 regressive samples in this cohort, median follow-up time was 4.73 years (range 0.42–13.5 years), suggesting that genomically stable samples are likely to regress and remain regressed long-term. Given their immunologic competence, late progressors are included in the regressive cohort when analyzing immune escape mechanisms in this study.

In summary, we present evidence that immune surveillance may play a critical role in spontaneous regression of precancerous lesions of the airways. Although recent cross-sectional studies have greatly furthered our understanding of immune signals prior to cancer invasion, and indeed at earlier disease stages than CIS (2, 12), we have for the first time shown an association with lesion regression. Including such outcome data offers insight into the dynamics of immune surveillance and evasion; assuming that lesion regression is driven by immune surveillance, which is likely based on our data, we are able to directly compare preinvasive lesions that are immune competent (regressed) with those that are able to evade immune predation (progressed). Analysis of “late progressive” samples furthers this model by providing estimates of timescales over which immune evasion evolves. Hence, we provide a road map for manipulation of the immune system as a cancer intervention strategy, by identifying and targeting differences between these two immune states.

To this end, we identify mechanisms of immune escape present before the point of cancer invasion, many of which offer potential therapeutic targets. These data present an opportunity to induce regression and prevent cancer development. Demethylating agents, 4-1BB agonists, and CCL27 blockade are therapeutic candidates that warrant further research, as well as targeting the PD-1–PD-L1 axis in highly selected patients. As a result of field carcinogenesis, patients with preinvasive lesions are at risk of synchronous cancers at other sites, which are likely to be clonally related (4, 37) and therefore may benefit from systemic immunomodulatory treatment. The data presented here support a new paradigm of personalized immune-based systemic therapy in early disease.

Additional methods are provided in a Supplementary File accompanying this article.

All tissue and bronchial brushing samples were obtained under written informed patient consent and were fully anonymized. Study approval was provided by the UCL/UCLH Local Ethics Committee (REC references 06/Q0505/12 and 01/0148). All relevant ethical regulations were followed.

For more than 20 years, patients presenting with preinvasive lesions, which are precursors of LUSC, have been referred to the UCLH Surveillance Study. As described previously (3), patients undergo repeat bronchoscopy every 4 months, with definitive treatment performed only on detection of invasive cancer. Autofluorescence bronchoscopy is used to ensure the same anatomic site is biopsied at each time point. Gene expression, methylation, and whole-genome sequencing analyses of CIS samples have been performed on this cohort, and data have been published (4). These data are used in this study.

All patients enrolled in the UCLH Surveillance Study who met a clinical end point of progression or regression were included; by definition they underwent an “index” CIS biopsy followed by a diagnostic cancer biopsy (progression) or a normal/low-grade biopsy (regression) 4 months later. Index lesions were identified between 1999 and 2017. Cases meeting an endpoint of regression underwent clinical review to identify those which subsequently progressed; 11 samples (20.7%) were identified, which are described as “late progressors” in the main text. Of these 11, median time from “regressive” index biopsy to progression was 3.2 years (range 0.8–4.6 years) whereas the remaining 42 samples had a median follow-up time of 4.73 years (range 0.42–13.5 years). Although we cannot fully exclude that any regressive sample may later develop cancer, the fact that median follow-up in the study group was longer than the maximum follow-up in the late progression group suggests that late progression in included samples is unlikely.

All samples underwent laser capture microdissection (LCM) to ensure only CIS cells underwent molecular profiling. Methods for sample acquisition, quality control, and mutation calling are as described previously (4), as are full details regarding patient clinical characteristics.

Briefly, gene expression profiling was performed using both Illumina and Affymetrix microarray platforms. Normalization was performed using proprietary Illumina software and the RMA method of the affy (38) Bioconductor package, respectively. This study includes 18 previously unpublished gene expression arrays from stromal tissue. These samples were collected using LCM to identify stromal regions adjacent to 18 already published CIS samples (corresponding to the 18 samples undergoing Affymetrix microarray profiling described above). These new stromal samples underwent Affymetrix profiling using the exact same methodology as described previously (4) for CIS tissue samples. To avoid issues related to batch effects between platforms, the analyses in this article utilize only samples profiled on Affymetrix microarrays, which include both CIS and matched stromal samples (see Supplementary Methods; Supplementary Table S5).

Methylation profiling was performed using the Illumina HumanMethylation450K microarray platform. All data processing was performed using the ChAMP Bioconductor package (39).

Whole-genome sequencing data was obtained using the Illumina HiSeq X Ten system. A minimum sequencing depth of 40x was required. BWA-MEM was used to align data to the human genome (NCBI build 37). Unmapped reads and PCR duplicates were removed. Substitutions, insertions/deletions, CNAs, and structural rearrangements were called using CaVEMan (40), Pindel (41, 42), ASCAT (43), and Brass (44), respectively.

As described previously, all patients enrolled in the surveillance program discussed above were considered for this study. For a given CIS lesion under surveillance, when a biopsy from the same site in the lung showed evidence of progression to invasive cancer or regression to normal epithelium or low-grade dysplasia, we defined the preceding CIS biopsy as a progressive or regressive “index” lesion, respectively. Because of the small size of bronchoscopic biopsy samples, not all profiling techniques were applied to all samples. Patients with fresh-frozen (FF) samples underwent whole-genome sequencing and/or methylation analysis depending on sample quality. Patients with formalin-fixed paraffin-embedded (FFPE) samples underwent gene expression analysis. Further detail is available in our previous article (4). In addition, any patient with an available FFPE block underwent image analysis as described below, and all patients with Affymetrix-based gene expression profiling underwent further profiling of laser-captured adjacent stroma.

Unless otherwise specified, all analyses were performed in an R statistical environment (v3.5.0; www.r-project.org/) using Bioconductor (45) version 3.7. Code to reproduce a specific statistical test is publicly available at the Github repository above.

Unless otherwise stated, comparisons of means between two independent groups were performed using a two-sided Wilcoxon test. In some cases, multiple samples have been profiled from the same patient, although always from distinct sites within the lung. In such cases we used mixed-effects models to compare means between groups, treating the patient ID as a random effect, as implemented in the Bioconductor lme4 library (46), with P values calculated using the ANOVA method from the Bioconductor ImerTest library (47). Differential expression was performed using the limma (48) Bioconductor package to compare microarray data between two groups. When adjustment for multiple correction is required, we quote an FDR, which is calculated using the Benjamini–Hochberg method (49). Cluster analysis and visualization was performed using the pheatmap Bioconductor package (available from https://cran.r-project.org/web/packages/pheatmap/).

All raw data used in this study are publicly available. Previously published CIS gene expression and methylation data are stored in the Gene Expression Omnibus (GEO) under accession number GSE108124; matched stromal gene expression data are stored under accession number GSE133690. Previously published CIS whole-genome sequencingdata are available from the European Genome–Phenome Archive (https://www.ebi.ac.uk/ega/) under accession number EGAD00001003883.Annotated H&E images of all samples used for lymphocyte quantification were deposited to the Image Data Resource (https://idr.openmicroscopy.org) under accession number idr0082.

All code used in our analysis will be made available at http://github.com/ucl-respiratory/cis_immunology on publication. All software dependencies, full version information, and parameters used in our analysis can be found here.

A. Pennycuick reports grants from Wellcome Trust (salary is funded by the Wellcome Trust clinical PhD programme; Wellcome Trust had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript) during the conduct of the study; in addition, A. Pennycuick has a patent for United Kingdom Patent application no. 1819452.2 pending (covers gene expression and methylation markers of cancer progression in squamous disease). V.H. Teixeira reports a patent for United Kingdom patent application no. 1819452.2 pending (covers gene expression and methylation markers of cancer progression in squamous disease). R. Rosenthal reports personal fees from Achilles Therapeutics outside the submitted work. C.P. Pipinikas reports a patent for United Kingdom patent application no. 1819452.2 pending (covers gene expression and methylation markers of cancer progression in squamous disease). D.A. Moore reports personal fees from AstraZeneca (speaker's fees) outside the submitted work. A.J.S. Furness reports personal fees from Bristol-Myers Squibb (speaker fees) and personal fees from Ipsen (speaker's fees) outside the submitted work. C. Marceaux reports grants from NHMRC and other from Viertel Foundation (charitable foundation) during the conduct of the study. M.-L. Asselin-Labat reports grants from NHMRC, other from Viertel Foundation (charitable foundation), and other from Cancer Early Detection and Advanced Research Center at OHSU (philanthropy) during the conduct of the study. L.M. Coussens reports grants from Brenden-Colsson Center for Pancreatic Care; AACR-SU2C during the conduct of the study; Syndax Pharmaceuticals (sponsored research), Innate Pharma (sponsored research), Prospect Creek Foundation (sponsored research), Lustgarten Foundation for Pancreatic Cancer Research (sponsored research); nonfinancial support from Cell Signaling Technology (reagent support), Syndax Pharmaceuticals, Inc. (reagent support), Deciphera Pharmaceuticals, LLC (reagent support), Pharmacyclics, Inc [steering committee for PCYC-1137-CA (NCT02436668); advisory board (unpaid)]; personal fees and nonfinancial support from Syndax Pharmaceuticals Inc. (advisory board), Carisma Therapeutics, Inc. (scientific advisory board), Verseau Therapeutics, Inc. (scientific advisory board), Zymeworks, Inc (scientific advisory board), CytomX Therapeutics, Inc. (scientific advisory board), Kineta, Inc. (scientific advisory board), (P30) Koch Institute for Integrated Cancer Research, Massachusetts Institute of Technology [advisory board (academic)], Bloomberg-Kimmel Institute for Cancer Immunotherapy, Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins [advisory board (academic)], (P30) Salk Institute Cancer Center [advisory board (academic)], Dana Farber Cancer Center Breast SPORE [advisory board (academic)], (P30) Dana-Farber/Harvard Cancer Center [advisory board (academic)], (P30) University of California, San Diego Moores Cancer Center [advisory board (academic)]; nonfinancial support from Cancer Research Institute (CRI) [advisory board (philanthropic)], The V Foundation for Cancer Research [advisory board (philanthropic)]; Starr Cancer Consortium [advisory board (philanthropic)], Lustgarten Foundation for Pancreatic Cancer Research, Therapeutics Working Group [advisory board (philanthropic)], NIH/NCI-Frederick National Laboratory Advisory Committee (FNLAC) [advisory board (federal government)], Cell Signaling Technology (consultant), Susan G. Komen Foundation, Komen Scholar (consultant); personal fees from AbbVie Inc (consultant), Shasqi, Inc. (consultant), other from AACR: Senior Editor, Cancer Immunology Research (journal); other from AACR: Scientific Editor, Cancer Discovery (journal), and Editorial Board member, Cancer Cell (2014 – present; journal) outside the submitted work. C. Swanton acknowledges grant support from Pfizer, AstraZeneca, Bristol-Myers Squibb, Roche-Ventana, Boehringer-Ingelheim, Archer Dx Inc (collaboration in minimal residual disease sequencing technologies), and Ono Pharmaceutical, is an AstraZeneca Advisory Board member and Chief Investigator for the MeRmaiD1 clinical trial, has consulted for Pfizer, Novartis, GlaxoSmithKline, MSD, Bristol-Myers Squibb, Celgene, AstraZeneca, Illumina, Genentech, Roche-Ventana, GRAIL, Medicxi, and the Sarah Cannon Research Institute, has stock options in Apogen Biotechnologies, Epic Bioscience, GRAIL, and has stock options and is co-founder of Achilles Therapeutics. C. Swanton holds European patents relating to assay technology to detect tumor recurrence (PCT/GB2017/053289); to targeting neoantigens (PCT/EP2016/059401), identifying patent response to immune checkpoint blockade (PCT/EP2016/071471), determining HLA LOH (PCT/GB2018/052004), predicting survival rates of patients with cancer (PCT/GB2020/050221), identifying patients who respond to cancer treatment (PCT/GB2018/051912), a U.S. patent relating to detecting tumor mutations (PCT/US2017/28013) and both a European and U.S. patent related to identifying insertion/deletion mutation targets (PCT/GB2018/051892). C. Thirlwell reports personal fees from Ipsen (honoraria), Boehringer-Ingelheim (consulting), and Novartis (conference travel support) outside the submitted work. P.J. Campbell reports grants from Wellcome Trust during the conduct of the study. N. McGranahan reports personal fees from Achilles Therapeutics outside the submitted work; in addition, N. McGranahan has patents 20200000904, 20200000903, and 20180251553 issued. S.M. Janes reports grants from Wellcome Trust, Rosetrees Trust, Welton Trust, Garfield Weston Trust, Stoneygate Trust, UCLH Charitable Foundation, and Stand Up To Cancer during the conduct of the study; grants and personal fees from AstraZeneca [paid advisory board, assistance for travel to meetings (ATS 2018 Takeda WCLC 2019)]; personal fees from Bard1 Bioscience (paid advisory board); personal fees from Achilles Therapeutics (paid advisory board); grants and personal fees from Janssen (paid advisory board, grant income investigator lead); and grants from GRAIL Inc. (grant income investigator lead) and Owlstone (grant income investigator lead) outside the submitted work; in addition, S.M. Janes has a patent for United Kingdom Patent application no. 1819452.2 pending (covers gene expression and methylation markers of cancer progression in squamous disease); and his wife works as a physician for AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

A. Pennycuick: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. V.H. Teixeira: Conceptualization, supervision, investigation, methodology, writing-original draft, project administration, writing-review and editing. K. AbdulJabbar: Resources, software, formal analysis, methodology, writing-review and editing. S.E.A. Raza:Resources, software, formal analysis, methodology, writing-review and editing. T. Lund: Conceptualization, investigation, methodology, writing-review and editing. A. Akarca: Resources, formal analysis, validation, investigation, methodology. R. Rosenthal: Resources, software, methodology. L. Kalinke: Investigation. D.P. Chandrasekharan: Investigation. C.P. Pipinikas: Resources. H. Lee-Six: Resources, investigation. R. Hynds: Investigation, writing-review and editing. K.H.C. Gowers: Investigation, project administration, writing-review and editing. J.Y. Henry: Investigation. F.R. Millar: Investigation. Y.B. Hagos: Formal analysis, investigation. C. Denais:Conceptualization. M. Falzon: Validation, investigation. D.A. Moore: Validation. S. Antoniou: Investigation. P.F. Durrenberger: Investigation. A.J. Furness: Investigation. B. Carroll: Resources, project administration. C. Marceaux: Investigation, writing-review and editing. M.-L. Asselin-Labat: Conceptualization, resources, supervision, investigation, methodology, writing-review and editing. W. Larson: Formal analysis. C. Betts: Formal analysis, investigation. L. Coussens: Resources, supervision, methodology, writing-review and editing. R. Thakrar: Resources. J. George: Resources. C. Swanton: Writing-review and editing. C. Thirlwell: Writing-review and editing. P.J. Campbell: Resources, supervision, writing-review and editing. T. Marafioti: Resources, supervision, methodology. Y. Yuan: Resources, supervision, writing-review and editing. S.A. Quezada: Conceptualization, resources, supervision, methodology, writing-review and editing. N. McGranahan: Conceptualization, resources, data curation, software, supervision, methodology, writing-review and editing. S.M. Janes: Conceptualization, resources, supervision, funding acquisition, methodology, writing-original draft, project administration, writing-review and editing.

We thank all of the patients who participated in this study. We thank P. Rabbitts, A. Banerjee, and C. Read for their early development of the study. The results published here are in part based on data generated by a TCGA pilot project established by the NCI and National Human Genome Research Institute. Information about TCGA and the investigators and institutions that constitute the TCGA research network can be found at http://cancergenome.nih.gov. R.E. Hynds, N. McGranahan, P.J. Campbell, and S.M. Janes are supported by Wellcome Trust fellowships. S.M. Janes is also supported by the Rosetrees Trust, the Welton Trust, the Garfield Weston Trust, the Stoneygate Trust and UCLH Charitable Foundation, as well as Stand Up To Cancer-LUNGevity-American Lung Association Lung Cancer Interception Dream Team Translational Cancer Research Grant (grant number: SU2C-AACR-DT23-17). Stand Up To Cancer (SU2C) is a division of the Entertainment Industry Foundation. The research grant is administered by the American Association for Cancer Research, the scientific partner of SU2C. V.H. Teixeira, C.P. Pipinikas, R.E. Hynds, and S.M. Janes have been funded by the Roy Castle Lung Cancer Foundation. A. Pennycuick and D.P. Chandrasekharan are funded by Wellcome Trust clinical PhD training fellowships. H. Lee-Six is funded by the Wellcome Trust Sanger Institute nonclinical PhD studentship. C. Thirlwell was a CRUK Clinician Scientist. This work was partially undertaken at UCLH/UCL, who received a proportion of funding from the Department of Health's NIHR Biomedical Research Centre's funding scheme (to S.M. Janes). R.E. Hynds, D.A. Moore, N. McGranahan, C. Swanton, and S.M. Janes are part of the CRUK Lung Cancer Centre of Excellence. C. Swanton is Royal Society Napier Research Professor. His work is supported by the Francis Crick Institute, which receives its core funding from Cancer Research UK (FC001169), the UK Medical Research Council (FC001169), and the Wellcome Trust (FC001169). C. Swanton is funded by Cancer Research UK (TRACERx, PEACE and CRUK Cancer Immunotherapy Catalyst Network), Cancer Research UK Lung Cancer Centre of Excellence, the Rosetrees Trust, Butterfield and Stoneygate Trusts, NovoNordisk Foundation (ID16584), Royal Society Research Professorship Enhancement Award (RP/EA/180007), the NIHR BRC at University College London Hospitals, the CRUK-UCL Centre, Experimental Cancer Medicine Centre and the Breast Cancer Research Foundation (BCRF). This research is supported by a Stand Up To Cancer-LUNGevity-American Lung Association Lung Cancer Interception Dream Team Translational Research Grant (SU2C-AACR-DT23-17). Stand Up To Cancer is a program of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the Scientific Partner of SU2C. C. Swanton also receives funding from the European Research Council (ERC) under the European Union's Seventh Framework Programme (FP7/2007-2013) Consolidator Grant (FP7-THESEUS-617844), European Commission ITN (FP7-PloidyNet 607722), an ERC Advanced Grant (PROTEUS) from the European Research Council under the European Union's Horizon 2020 research and innovation programme (835297), and Chromavision from the European Union's Horizon 2020 research and innovation programme (665233). Y. Yuan acknowledges funding from Cancer Research UK Career Establishment Award, Breast Cancer, Children's Cancer and Leukaemia Group, NIH U54 CA217376 and R01 CA185138, CDMRP Breast Cancer Research Program Award, CRUK Brain Cancer Award (TARGET-GBM), European Commission ITN, Wellcome Trust, and The Royal Marsden/ICR National Institute of Health Research Biomedical Research Centre. S.A. Quezada is funded by a CRUK Senior Cancer Research Fellowship, a CRUK Biotherapeutic Program Grant, the Cancer Immunotherapy Accelerator Award (CITA-CRUK), and the Rosetrees Trust. L.M. Coussens acknowledges funding from the NIH (1U01 CA224012, U2C CA233280, R01 CA223150, R01, R01 CA226909, R21 HD099367), the Knight Cancer Institute, and the Brenden-Colson Center for Pancreatic Care at OHSU. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.