Genomic Landscape of Atypical Adenomatous Hyperplasia Reveals Divergent Modes to Lung Adenocarcinoma

Non–small cell lung cancer (NSCLC) constitutes the majority (∼85%) of lung malignancies (1). NSCLC is also the leading cause of cancer mortality (2). This is largely attributed to late diagnosis after locoregional or distant spread (1). Advances (e.g., molecular-based) in early detection and prevention of NSCLC have been limited by a poor understanding of early changes in the pathogenesis of the disease (3).

NSCLC is mainly composed of two histologic subtypes, adenocarcinomas (LUAD) and squamous cell carcinomas (LUSC), with LUAD the most common of these (1). Studies have demonstrated that LUADs and LUSCs carry different genomic landscapes of driver mutations as well as disparate clinicopathologic features (3). While LUSCs are diagnosed almost exclusively in smokers, LUADs are known to develop in both smokers and nonsmokers (3). In contrast to the well-characterized histopathologic sequence of various lesions (bronchial hyperplasias, metaplasias, and dysplasias, as well as carcinomas in situ) associated with LUSC pathogenesis, atypical adenomatous hyperplasias (AAH) are the only known precursor (premalignant) lesions in the development of LUAD (3, 4).

Very few molecular alterations have been described in AAHs. Mutations in EGFR and KRAS have been identified previously in AAH, showing mutual exclusivity (5). Whereas EGFR mutations were shown to present at similar frequencies in AAH and LUAD, KRAS mutations (e.g., codon 12) were demonstrated to occur more frequently in AAH compared with LUAD (5). Further, it has also been shown that KRAS-mutant AAHs are more prevalent in smokers while EGFR-mutant AAHs do not show association with smoking status (5). Loss-of-heterozygosity of chromosome arms 3p, 9p, 16p, 17q, and 17p has been shown in AAH, and in common with corresponding primary LUADs (3). Other molecular aberrations that have been identified in AAH include overexpression of cyclin D1, survivin, ERBB2 oncoproteins, and NKX2-1 (3). Epigenetic modifications such as increased promoter hypermethylation of genes implicated in lung cancer (e.g., p16) with histologic progression from normal to AAH and finally LUAD (6), as well as DNA methylation of CDKN2A and PTPRN2, have also been reported in AAHs (7). A recent multiregion sequencing study identified several clonal events predicted to be involved in the transformation of these precursor lesions and that were further shown to be present in paired circulating DNA (8). These studies point to the molecular complexity of AAH, which is still poorly understood, and its role as a precursor to LUAD.

We interrogated, by deep sequencing, somatic mutations and gene expression changes in normal tissues, AAHs and malignant tumors from patients with LUAD. We identify subgroups of AAHs with distinct driver mutations, expression signatures and aberrant markers of the immune response that inform the molecular pathogenesis of AAH. These alterations offer a window to explore new strategies for early clinical management (e.g., detection and targeted chemoprevention) of LUAD.

Normal lung parenchyma tissues (NL), AAHs, and lung adenocarcinomas (LUAD; n = 67 samples) were acquired from 22 patients with early-stage LUAD who were evaluated at the Aichi Cancer Center (Nagoya, Japan) and Nagasaki University (Nagasaki, Japan). Specimens were approved for study by Institutional Review Boards and according to the international ethical guidelines for biomedical research involving human subjects (Council for International Organizations of Medical Sciences). Informed written consents were received from all subjects wherever necessary. Cases with available AAHs were included in the analysis to study mutational and expression signatures involved in the pathogenesis of AAH (Fig. 1). Clinicopathologic features of these patients are summarized in Table 1. The diagnosis, specimen collection and slide preparation were carried out between 2011 and 2015 for all patients. Specimens were obtained formalin-fixed and paraffin-embedded (FFPE) and stained by hematoxylin and eosin (H&E). A tabulation of cases and samples analyzed by DNA- (22 cases) and RNA- (17 cases) sequencing is provided in Supplementary Table S1.

DNA/RNA was extracted from all samples using the AllPrep DNA/RNA FFPE Kit from Qiagen and suspended in nuclease free water (RNA) or AE buffer (DNA). Sample concentrations were measured on a NanoDrop 1000 (Thermo Fisher Scientific), and RNA integrity numbers indicative of overall quality were obtained on the 2100 Bioanalyzer (Agilent Technologies) using the RNA 6000 Nano or Pico kit based on RNA concentrations and according to the manufacturer's protocol. DNA was quantified using the Quant-iT PicoGreen double-stranded DNA (dsDNA) kit (Thermo Fisher Scientific) according to the manufacturer's instructions.

The Ion AmpliSeq Comprehensive Cancer Panel (Thermo Fisher Scientific) comprising primers for 409 canonical cancer-associated genes and the AmpliSeq Library Kit 2.0 (Thermo Fisher Scientific) were used to prepare barcoded libraries from the FFPE DNA samples. Sequencing was performed on the Ion Torrent Proton platform according to the manufacturer's instructions. The targeted DNA sequencing data files have been deposited in the sequence read archive (SRA) under Bioproject accession PRJNA398260.

Using BAM files generated in the Ion Torrent server, somatic variants were rigorously identified based on four different variant calling methods: the Ion Torrent proprietary software Ion Reporter, marginal VCFs generated from Torrent Variant Caller (TVC), MuTect (9) and Varscan2 (10). Instead of using all variants detected by any of the four callers, for each sample, the stringency was increased to include variants that were detected by at least two different callers, with the exception of those identified by Ion Reporter software and TVC. Variants in exonic, splicing, and untranslated regions (UTR) were assessed, focusing on exonic single-nucleotide variants within the targeted 409 cancer gene panel. Nonsynonymous mutations in genes considered to be bona fide drivers of cancer according to the previous report by Vogelstein and colleagues (11) were analyzed in AAHs and LUADs. Mutations in genes previously determined by The Cancer Genome Atlas (TCGA) to be significantly mutated in LUAD (12) were also assessed in both AAHs and LUADs.

A subset of the cases (15 NLs, 17 AAHs, and 16 LUADs from 17 different cases) was selected for transcriptome sequencing based on specimen availability as well as on transcriptome sequencing quality indicated by percentage of mapped reads and valid on-target reads. All samples were reverse-transcribed to generate cDNA libraries using the Ion AmpliSeq Transcriptome Human Gene Expression Kit (Thermo Fisher Scientific) adhering to the manufacturer's protocol. For sequencing, specimens from two cases were processed together in one chip. All samples were sequenced on an Ion Proton sequencer. Raw transcriptome sequence data files have been deposited in the gene expression omnibus under data series GSE102511.

Transcriptomes were quantified from BAM alignment files generated in the Ion Torrent server using an expectation-maximization (E/M) algorithm based procedure (13). Resultant gene-based counts were normalized, log (base 2) transformed, and corrected for batch using the R limma package (14). Hierarchical clustering using Pearson correlation was performed in R. Pathways, gene-network identification and gene set enrichment analyses were performed using the commercially available software Ingenuity Pathways Analysis.

Additional details are found in the Supplementary Methods.

There is a limited understanding of molecular alterations in AAH, the only known precursor lesion for the major lung cancer subtype adenocarcinoma (3, 4). Here, we sought to survey mutation and expression profiles in the pathogenesis of AAHs by studying these premalignant lesions along with normal tissues and primary LUADs (Fig. 1) from 22 patients with early-stage LUAD (Table 1). Diagnosis and histopathologic determination of specimens (normal lung, AAH, LUAD) were conducted by analysis of H&E staining of 5-μm sections (Fig. 1) and according to the WHO report on lung tumor classification by Travis and colleagues (15). Among the 22 patients, 12 were ever-smokers and 10 were nonsmokers (Table 1). The cohort comprised 12 females and 10 males with a median age of 64. The majority (n = 18) of LUADs were stage IA with the remaining four cases determined to be IB (n = 3) and IIIA (n = 1; Table 1). We studied matched samples (n = 67; 1 patient comprised two tumor specimens; Supplementary Table S1) from the 22 early-stage LUAD patients by deep (mean = 961×) targeted sequencing of a 409 pan-cancer gene panel (Supplementary Table S2). A subset of the cases (n = 17) and samples (n = 48), based on availability and sequencing quality of AAHs, were also surveyed by transcriptome sequencing (Supplementary Table S1). Details of the quality metrics for each sequencing platform are listed in Supplementary Table S2.

All 45 AAHs and LUADs exhibited at least one somatic variant (exonic, splicing or in UTRs) with a mean of 6.1 variants (min = 1, max = 19) in AAHs and 10.6 (min = 1, max = 60) in LUADs (Supplementary Fig. S1). Nonsmokers displayed an expected lower somatic mutation burden than ever-smokers in the LUADs (5.4 vs. 14.5); however, they exhibited a similar burden in AAHs (6.4 vs. 5.9; Supplementary Fig. S2). We then specifically interrogated in AAH, nonsynonymous mutations in genes considered to be bona fide drivers of cancer (Fig. 2A; ref. 11). We also examined mutations in genes previously determined by TCGA to be significantly recurrently altered in LUAD (12). AAHs from 5 patients (23%) exhibited somatic activating mutations in the BRAF oncogene (Fig. 2A). Interestingly, the BRAF mutations were not detected in the paired LUAD specimens (Fig. 2A; Fig. 2B; Supplementary Tables S3 and S4). Four of the five AAHs exhibited a BRAF p.K601E mutation, the other AAH contained a BRAF p.N581S variant (Fig. 2C; Supplementary Table S3). KRAS was the second most recurrently mutated gene in AAHs (four cases, 18%; Fig. 2A). All four KRAS-mutant premalignant tissues were from ever-smokers, in contrast to BRAF-mutant AAHs (from three nonsmokers and two ever-smokers; Fig. 2A). Also, AAH KRAS and BRAF mutations showed mutual exclusivity (Fig. 2A). An interesting observation was that for four of the five cases (80%) with BRAF-mutant AAHs, their paired LUADs harbored activating mutations (three p.L858R in exon 21 and one p.S752F in exon 19; Supplementary Table S3) of the EGFR oncogene (Fig. 2A). The other LUAD exhibited inactivating mutations in KEAP1 and STK11 tumor suppressors (Supplementary Table S3). LUADs of cases with the KRAS-mutant AAHs exhibited mutations in other drivers besides KRAS such as TP53 (Fig. 2A). We further validated by digital PCR, the presence of all sequencing derived BRAF and KRAS mutations in AAHs, and the EGFR p.L858R mutation in both AAHs and LUADs that exhibited these patterns (Supplementary Table S5; Supplementary Fig. S3). Variant allele frequencies (VAF) based on digital PCR were consistent with sequencing-based VAFs for these loci (Supplementary Table S5). TSC1 was the most frequently mutated tumor suppressor in AAHs (13.6%; two nonsense and one missense mutation; Fig. 2A; Supplementary Table S3). We also noted other mutated oncogenes (EGFR and JAK3) and tumor suppressors (CDKN2A and TP53) in AAHs (Fig. 2A). Additionally, in this cohort, 28 genes were mutated in both AAHs and LUADs (e.g., KRAS, TP53, KEAP1, and CDKN2A), 84 were found only in LUADs (STK11 and PIK3CA) and 29 were found only in the preneoplastic lesions (AMER1, BRAF, KDM5C, and ERBB2; Fig. 2B; Supplementary Table S4). Even for genes that were shared between AAH and LUAD tissues, there were notable examples of differential frequency (e.g., KRAS more common in AAH; EGFR and TP53 more common in LUAD; Supplementary Fig. S4). On further examination of these 28 shared genes, we found that EGFR and KAT6B exhibited the same mutations in both tissues (Supplementary Table S3). There was also an enrichment of different mutations on codon 12 of KRAS in both the tissues (Supplementary Table S3). Our findings underscore subgroups of AAH with different mutated driver genes (BRAF vs. KRAS) suggestive of potentially various mechanisms in the pathogenesis of these premalignant lesions.

Next, we sought to characterize expression profiles signifying the development of AAH from NL, and its progression to LUAD. We performed RNA sequencing of a subset of the cases and samples (15 NLs, 17 AAHs, and 16 LUADs—from 17 patients), based on availability and quality of AAHs, using a capture method targeting over 20,000 Refseq genes. Using an ANOVA model with tissue type (NL, AAH, and LUAD) as a factor and patient as a random effect, with a test of no expression difference (P < 0.001) and an observed 2-fold change minimum among pairwise tissue comparisons, we identified 1,008 genes differentially expressed in at least one of three tissue types (Fig. 3; Supplementary Table S6). Using one-sided t tests to interrogate the two-step (NL to AAH and AAH to LUAD) modes of differential expression, we identified eight patterns or clusters of expression among the identified profiles (Fig. 3). These consisted of the following: decrease (n = 214) from NL to AAH and from AAH to LUAD; increase (n = 204) from NL to AAH and from AAH to LUAD; decrease (n = 116) and increase (n = 146) from NL to AAH alone (no change from AAH to LUAD); decrease (n = 85) and increase (n = 126) from AAH to LUAD alone (no change from NL to AAH) and less prevalent forms with no net change in expression such as an increase (n = 33) or decrease (n = 84) in AAH alone relative to other tissues (Supplementary Table S6). A pathway-based enrichment analysis was performed for genes in each cluster to infer potentially altered signaling (Fig. 3). This analysis pinpointed decreased antitumor T-helper (Th1) immunity, and conversely, increased protumor Th2-based immune response and signaling in both phases, the development of AAH from NL and their progression to LUAD. Inhibition of IFN-γ and TGFB1 signaling occurred early in AAH, when compared with NLs, and reduced surfactant protein signaling occurred thereafter in LUAD only. Pathways and gene set enrichment analysis also revealed an activation of B-cell receptor, CSF2 (indicative of protumor immune response), MYC, and ERBB2 signaling in AAH and LUAD (Fig. 3). Activation of WNT and β-catenin signaling as well as modulation of gene sets associated with increased immune cell (phagocytes) migration were activated in AAH relative to NL (Fig. 3). Gene sets associated with enhanced cell cycle and proliferation as well as reduced apoptosis were modulated in LUAD relative to AAH or NL.

We compared and contrasted gene expression among three groups of AAHs based on driver gene mutation status identified above (Fig. 2): BRAF mutant, KRAS mutant, and BRAF/KRAS wild-type (WT). Using a similar model as described above for global gene expression analysis but with a P < 0.01 threshold and a 1.5 fold-change cutoff, we identified 327 differentially modulated genes between the three different groups of AAHs (Fig. 4; Supplementary Table S7). Accordingly, these gene features indeed clustered the three groups separately based on the driver mutation status but with BRAF- and KRAS-mutant AAHs grouped closer together than with BRAF/KRAS WT AAHs (Fig. 4; Supplementary Table S7). Among the genes that were enriched in the BRAF-mutant AAHs is the cytokinesis promoting gene KIF5C and the cell proliferation promoting transcription factor (MYC Associated Factor X) MAX, typically associated with MYC oncoprotein (Fig. 4; ref. 16). On the other hand, KRAS-mutant AAHs displayed upregulated expression of tumor necrosis factor receptor superfamily members 9 and 10B (TNFRSF9 and TNFRSF10B), the NF-kB subunit RELB and the proliferation promoting ubiquitin ligase UBE2C (Fig. 4). Of note, both BRAF-mutant and KRAS-mutant AAHs exhibited suppressed expression of the epithelial mesenchymal transition-promoting tyrosine kinase receptor AXL relative to BRAF/KRAS WT AAHs (Fig. 4). These findings suggest shared and disparate expression programs among AAHs with activating mutations in the oncogenic GTPases BRAF and KRAS.

Accumulating evidence suggests a pivotal role for the host immune response in the evolution of cancer as well as dynamic interplay between emerging tumor cells and immune-based expression programs (17, 18). We sought to begin to understand contextual immune marker profiles in the development and progression of AAHs. Among these profiles we identified genes with known roles in immune signaling based on an annotated and a priori list from the nCounter PanCancer Immune Profiling Panel (nanoString Technologies). Using a similar random effects model described above, a significance threshold P < 0.001 and a 1.5-fold change, we identified 131 markers of immune response that were differentially modulated among NLs, AAHs, and LUADs (Fig. 5; Supplementary Table S8). Overall, the immune markers followed similar patterns or clusters of expression described above. This analysis revealed that IL12A, a cytokine most notably associated with an antitumor immune response (19), was decreased in AAHs and LUADs relative to NLs (Fig. 5; Supplementary Table S8). Conversely, the cytokines CXCL13 and CXCL14, indicative of activated B-cell chemotaxis and signaling (20, 21), were upregulated in AAHs and LUADs (Fig. 5). Moreover, we found aberrant immune marker expression occurring early in AAHs, relative to NLs (Fig. 5). We found early and significantly decreased expression of prototypical markers of the antitumor immune response (e.g., GZMB) in AAHs relative to NLs (Fig. 5). On the other hand, AAHs exhibited increased expression of the tumor-supportive chemokine receptor CCR2 (Fig. 5; ref. 22). Of note, we found that the major immune checkpoint cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4; ref. 23) was significantly upregulated in LUADs relative to AAHs but not so in the premalignant lesions relative to NLs (Fig. 5; Supplementary Table S8), suggesting that aberrant immune checkpoint function by CTLA-4 may be implicated in progression of AAH to LUAD. These findings accentuate the role of aberrant immune function and signaling early on in the development and progression of AAH.

There is a lack of understanding of the molecular aberrations leading to the initiation as well as the progression of AAH, the only known precursor lesion to LUAD (3). Here, we probed the mutation and gene expression landscapes of AAHs in comparison with normal tissues and early-stage LUADs from the same (matched) patients. We delineated subgroups of AAHs with mutually exclusive and distinct driver gene mutation status; namely BRAF-mutant (both nonsmokers and ever-smokers), KRAS-mutant (ever-smokers only) and KRAS/BRAF WT AAHs. By agnostic transcriptome sequencing analysis, we also identified various patterns of expression profiles and pathways in the molecular pathogenesis of AAH. Further analysis underscored markers of immune function that are significantly differentially modulated, early on, in AAH (downregulation of GZMB) relative to NL as well as those deregulated in LUADs relative to AAHs (e.g., CTLA-4). Our study highlights early recurrent driver mutations, expression profiles, and markers of immune response in AAH that offer a window to understand the molecular pathogenesis of these premalignant lesions.

In this cohort, we found that the BRAF oncogene was the most commonly mutated gene in AAH (four patients with p.K601E and one with p.N581S; Supplementary Table S3) followed by KRAS (predominantly in codon 12). No BRAF variants were found in the paired LUADs. Of note, among AAHs, mutations in BRAF and KRAS were mutually exclusive. Whereas KRAS-mutant AAHs were from ever-smokers, BRAF mutations in AAHs occurred in both nonsmokers and ever-smokers. The BRAF p.K601E variant has been previously noted in preneoplastic melanocytic lesions and melanomas in situ as well as in thyroid adenomas (24–26), thus pointing to the probable role of BRAF in early stages of oncogenesis (i.e., development of preneoplasia such as AAH). The BRAF p.K601E mutation was also found in small proportions of cancers of the thyroid, colon, and skin (27). This may suggest that an enrichment for this hotspot driver mutation highlights a crucial mechanism for AAH and LUAD pathogenesis. Indeed, studies by the TCGA (12) and our group (28) showed relatively infrequent (∼3%) BRAF mutations in LUADs. Yet, its absence in our sample set of LUADs, including in tumor specimens from patients with BRAF-mutant AAH is intriguing. It cannot be neglected that this may, in part, be due to our relatively modest set of samples (further discussed below).

An intriguing finding from our study is the pattern of BRAF and EGFR mutations in the paired LUADs. In four of the five BRAF-mutant AAHs, the paired LUADs exhibited driver EGFR mutations (e.g., p.L858R). Conversely, LUADs of KRAS-mutant AAHs displayed several other driver mutations (TP53, KRAS) that are typically associated with smoking (3). Our study also underscores previously uncharacterized properties of these AAH BRAF mutations, namely mutual exclusivity with KRAS and correlation with smoking patterns. Based on our findings on mutual exclusivity of BRAF and KRAS in AAHs along with the disparate patterns of mutations in the paired LUADs, it is plausible to suggest that there are divergent pathways in pathogenesis of these preneoplastic lesions. A schematic of this paradigm is represented in Fig. 6. A similar divergent model to malignancy has also been recently described in the evolution of different melanoma subtypes from their precursor lesions (26).

The multiregion sequencing study by Izumchenko and colleagues (8) supports several of our findings. First, their study noted BRAF as the most commonly mutated gene in AAH, albeit in a smaller cohort of six patients. Further, they identified the BRAF p.K601E variant in a single AAH sample at a very low variant allele frequency (2%). Second, the mutual exclusivity of BRAF and KRAS mutations that we observed in AAHs can also be inferred from their data. Third, in both studies, EGFR and KRAS were the only driver genes sharing the exact mutation in both AAHs and LUADs. Fourth, among the 28 genes we found to be mutated in both AAHs and LUADs (Fig. 2B), they observed three (TP53, EGFR, and KRAS) to be mutated. Additionally, APC, CDKN2A, CREBBP, and NF1, which we observed as shared mutated genes were only reported in either AAHs or malignant lesions in their study. These disparities are not absolute as they may result from imperfect detection due to technological limitations (e.g., sequencing depths) and differences in study design. In comparison with their study, we surveyed a greater number of patients and genes as well as achieved a higher sequencing depth to detect potentially rare (within-sample) variants. Further, we identified recurrent BRAF mutations (most commonly p.K601E) that exhibited molecular (mutual exclusivity with KRAS) and clinicopathologic (also found in nonsmokers) features and that were present at a higher allele frequency (10%–37%; Supplementary Table S3) than observed previously. Nonetheless, given the complexity of AAHs and the malignant lesions leading to invasive LUAD, it is interesting, and indeed confirmatory, to identify similar gene mutation patterns across studies.

Complementing our DNA analysis is our agnostic transcriptome sequencing analysis that revealed differential gene expression programs that occur in different stages of LUAD pathogenesis—early in development of AAH from normal tissue, in LUADs, or in both lesion types. Gene set enrichment and pathway analysis pinpointed elevated immune cell trafficking and WNT/β-catenin signaling as well as an inhibition of the antitumor inflammatory response (Th1) and transforming growth factor beta 1 (TGFB1) signaling in development of AAH from normal lung. WNT/β-catenin signaling has been previously shown to be activated in progression of oral leukoplakia, a precancerous lesion of head and neck squamous cell carcinoma (29). Conversely, gene sets associated with increased cell cycle and proliferation, decreased apoptosis, and reduced function of pulmonary surfactant proteins (e.g., SFTPs A1, A2, C, and D) were enriched in profiles differentially expressed in LUAD relative to AAH. Several key signaling pathways (e.g., those mediated by EGFR, MYC, and CSF2) were elevated/enriched in AAH and furthermore in LUAD, consistent with their role in tumor progression (3). Increased EGFR was shown to promote cellular proliferation, inhibit apoptosis, and drive development and progression of bronchial dysplasia (30). Similarly, MYC overexpression has been previously reported in colorectal polyps with a level of expression proportional to the polyp size as well as dysplastic histology (31). By transcriptome sequencing, we also identified differentially expressed profiles between AAHs with mutations in BRAF and KRAS. BRAF-mutant AAHs showed significant downregulation of adenosine deaminase, an enzyme involved in nucleotide metabolism, the deficit of which may lead to impaired DNA synthesis and repair (32). In contrast, SOS2, an oncogene known to confer increased growth potential of tumor cells exhibiting an oncogenic KRAS (33), and ubiquitin conjugating enzyme E2C (UBE2C), previously shown to be elevated in lung cancer lesions, particularly in ever-smokers (34), was enriched in the KRAS-mutant AAHs compared with BRAF-mutant AAHs. Taken together, these data point to early changes in the development and progression of AAH and that would thus comprise ideal targets for chemoprevention of LUAD.

Our findings on aberrant immune-regulated pathways from the agnostic gene expression analysis prompted us to more closely probe the modulation of known markers of immune function. Our immune marker profiling overall suggested an activation of protumor immune pathways (i.e., Th2) and B-cell receptor signaling as well as an inhibition of antitumor immune response (e.g., Th1-derived IFN-γ signaling). Similar findings have been reported in previous studies of Barrett esophageal tissues, a premalignant condition with a high risk of progression to esophageal adenocarcinomas (35). IL12A, known for its proinflammatory antitumor response, along with antitumor immune chemokines (e.g. CCL3, CCL4, TLR4) and apoptosis-inducing proteases (GZMB) were decreased in AAH relative to normal lung. On the other hand, we found elevated expression of the CCL2 chemokine receptor CCR2 in AAH relative to normal lung. CCR2 has been shown to enhance tumor growth, angiogenesis, and tumor progression and was demonstrated to be overexpressed in several tumor tissues (22). Recently, CCL2/CCR2-based immune prevention models were shown to attenuate tumor development and metastasis (22, 36). Of note, we identified an increasing expression in chemokines CXCL13 and CXCL14, both known for their role in inflammatory processes and immune response (20), and SPP1, previously shown to be overexpressed in premalignant lesions of the oral epithelium as well as actinic keratosis, the premalignant lesion to skin squamous cell carcinomas (37). We also found that CD27, which in combination with its ligand CD70 is known to generate a potent costimulatory signal, was increased in AAH relative to normal lung. Notably, our analysis pointed to significantly increased expression of the major immune checkpoint CTLA-4 in LUAD relative to AAH (23). We also observed disparate patterns of immune marker expression among AAHs with different recurrent mutated driver genes (BRAF-mutant, KRAS-mutant, and BRAF/KRAS WT). We noted decreased expression of TNFRSF9, also called CD137, known to regulate the activation of T cells and a promising target for enhancing antitumor immune responses (38), in BRAF-mutant AAHs, suggesting a relatively dampened immune response in this subgroup of AAH. Conversely, we found decreased expression of the receptor tyrosine kinase AXL, previously shown to promote protumor immune responses (39), in AAHs with mutations in either BRAF or KRAS. These findings, albeit based on a limited cohort of AAHs analyzed by both DNA and RNA sequencing, point to differential aberrant immune signaling among AAH based on driver mutation status. Future studies surveying a larger number of AAH have the potential to corroborate these observations. Nonetheless, we posit here that aberrant immune signaling (e.g., attenuated antitumor immune response) is a common, perhaps critical, feature of AAH and LUAD development, as illustrated in Fig. 6. Indeed, ongoing whole-exome sequencing studies have begun to shed light on neoantigen profiles in AAH (40), analogous to observations recently made for LUADs (41). Also, further studies examining protein (e.g., by multiplexed immunohistochemistry) levels of markers of various immune cell infiltrates will shed more light on the role of the immune response in the pathogenesis of AAH. It is important to mention that immune-based therapy has come to the forefront of targeted therapeutic strategies for various malignancies including those of the lung (42). For instance, monoclonal antibodies targeting genes such as CD27 and CTLA-4 are already being tested for treatment of various cancers, including lung adenocarcinomas (43). In this context and as alluded to previously (44, 45), our findings suggest that targeting immune responses and signaling (e.g., immune checkpoint blockade) may be a viable strategy to prevent progression of preneoplasia such as AAH.

It is important to note that our study is not without limitations. While we comprehensively studied paired AAHs and LUADs, our cross sectional study design is not best positioned to thoroughly characterize the “progression” of AAH to LUAD. Naturally, the AAHs that already progressed to LUADs are no longer available for analysis. Our present report lends to the need for further longitudinal studies with a larger number of samples in which expression data and markers of the immune response can be aligned with time and space. Additionally, future longitudinal studies surveying AAHs in patients without overt lung malignancy may also help pinpoint drivers of AAH progression. Further, based on our findings on the absence of BRAF mutations in the LUADs studied in our cohort, one cannot rule out the possibility that the BRAF-mutant AAHs are benign and may not be the preneoplastic lesions that eventually progress to LUADs. Earlier studies have insinuated that BRAF mutations are important for initiation of premalignancy rather than their expansion or progression (8, 46). Lungs of mice genetically engineered to express a mutant form of Braf (p.V600E) were shown to develop hyperplasias that progressed to adenoma (47). Of note, only after mutations in other genes (e.g., Tp53) did the Braf-mutant lesions progress to LUAD (47). Yet, the strong pairing of BRAF-mutant AAHs with EGFR-mutant LUADs is nonetheless an interesting observation that is worth investigating in future studies comprising a larger number of patients with both AAHs and LUADs. Further, that these patterns hold across lesions arising independently, although potentially from the same cell lineage, reflect the patient-specific nature of their development and highlight the potential for personalized prevention strategies. It is also worthwhile to mention that our cohort was mainly comprised of East Asian patients. Earlier studies have demonstrated that LUADs of East Asians exhibit disparate mutational spectra (e.g., more prevalent activating mutations in EGFR) relative to LUADs from Western (or Caucasian) patients (48, 49). It is reasonable to surmise that mutational differences in AAHs, across patients of different ethnicities, will roughly reflect those we observe in LUADs. In this context, our results and proposed paradigm could be more relevant to the East Asian population-based on our cohort and necessitate future work in larger populations comprising diverse ethnicities. Also, recent pathological classification guidelines for LUAD have underscored subgroups with pure lepidic growth (adenocarcinoma in situ; AIS) and those that exhibit predominant lepidic growth and with less than 5-mm invasion (minimally invasive adenocarcinoma; MIA; ref. 3). Earlier work in East Asian LUAD patients suggested that LUADs of the “terminal respiratory unit” progress from AAH to AIS and then to invasive lesions (3). It is plausible that AIS may have distinct profiles that suggest an intermediate stage (8, 26) in the progression of AAH to LUAD. Our cohort largely comprised LUADs with very few AIS or MIA, too limited in size to further delineate profiles along this progression. Future studies are warranted to align with space mutational profiles, gene expression, and markers of the immune response, particularly those shared between AAH and LUAD, and determine their contextual role in development of AIS and their progression to LUAD. Also, indeed, recent and ongoing efforts have begun to distinguish potentially distinct profiles in the multistep progression of AAH to AIS and subsequently to LUAD (3, 8, 40).

Findings from our study shed light on some of the earliest mutation events, expression changes as well as altered immune pathways in the pathogenesis of AAH, the only known precursor lesion to LUAD. The different mechanisms in the pathogenesis of AAH we explore here may further identify novel biomarkers and potentially offer immune-based intervention or other personalized prevention in patients with early-stage LUAD. This further accentuates the need for a greater depth and understanding of immunotherapeutic strategies early on, in potentially less hostile environments such as those typified by premalignant lung lesions.

J. Zhang has received honoraria from speakers bureau from BMS. J. Fukuoka is CEO of Pathology Institute Corp. A.E. Spira is a consultant/advisory board member for Janssen Pharma. E.T. Hawk has received honoraria from speakers bureau of National Cancer Institute, Huntsman Cancer Institute, American Cancer Society, University of New Mexico ECHO Advisory Group, Rice University, Mayo Clinic Cancer Center, Kansas University Medical Center, Ohio State University Comprehensive Cancer Center, Iowa Cancer Consortium, USA Mitchell Cancer Institute, Roswell Park Cancer Institute, Simmons Comprehensive Cancer Center, and University of Nebraska Medical Center, and is a consultant/advisory board member for Cancer Prevention Pharmaceuticals, PLx Pharma, and POZEN. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Sivakumar, J. Fujimoto, J. Fukuoka, S.M. Dubinett, I.I. Wistuba, P. Scheet, H. Kadara

Development of methodology: S. Sivakumar, T.L. McDowell, L. Xu, J. Fujimoto, J. Fukuoka, J. Fowler, I.I. Wistuba, H. Kadara

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Xu, J. Fujimoto, J. Fukuoka, Y. Yatabe, I.I. Wistuba

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Sivakumar, F.A. San Lucas, L. Xu, J. Zhang, J. Fowler, E.T. Hawk, P. Scheet, H. Kadara

Writing, review, and/or revision of the manuscript: S. Sivakumar, T.L. McDowell, J. Fujimoto, J. Zhang, P.A. Futreal, J. Fukuoka, S.M. Dubinett, A.E. Spira, J. Fowler, E.T. Hawk, I.I. Wistuba, P. Scheet, H. Kadara

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.L. McDowell, W. Lang, J. Fujimoto, J. Fukuoka

Study supervision: P. Scheet, H. Kadara

We thank Dr. Eva Szabo from the Division of Cancer Prevention at the National Cancer Institute for her insightful comments.

This work is supported in part by Cancer Prevention and Research Institute of Texas (CPRIT) grant RP150079 (P. Scheet and H. Kadara), NIH grant R01HG005859 (P. Scheet), and The University of Texas MD Anderson Cancer Center Core Support Grant.

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.