Inactivation of AMPK Leads to Attenuation of Antigen Presentation and Immune Evasion in Lung Adenocarcinoma

STK11/LKB1 mutations represent a challenge in non–small cell lung cancer (NSCLC) occurring in 17% to 23% of NSCLC adenocarcinomas (1) often together with KRAS-activating mutations. Importantly, the presence of LKB1 mutations or reduction of LKB1 expression in patients with lung cancer associates with a more aggressive clinical phenotype (2) and resistance to conventional chemotherapy (3) as well as target therapies inhibiting either downstream effectors of KRAS signaling (4), suggesting novel therapeutic strategies are needed to target this subset of lung cancers. PD-1/PD-L1 checkpoint blockade immunotherapy elicits durable antitumor effects in multiple cancers including NSCLC, yet not all patients respond. Retrospective studies in four independent cohorts identified genomic alterations of STK11/LKB1 as a major driver of immune escape and tumor cell–intrinsic determinant of primary resistance to PD-1/PD-L1 blockade in NSCLC, despite the presence of intermediate or high tumor mutation burden (5). Consistently, a pan-tumor analysis of genomic markers associated with anti–PD-1 response demonstrated that tumor mutation burden and a T-cell–inflamed gene-expression profile (GEP) can jointly stratify patients with cancer into groups with different responses to PD-1 blockade, with LKB1 mutation in LUAD identified to have the most significant negative associations with T-cell–inflamed GEP across 20 cancer types (6). These findings are recapitulated in a genetically engineered mouse model (GEMM) of lung cancer driven by oncogenic Kras mutation (Kras^G12D), where Lkb1 inactivation leads to accelerated tumor growth (7), metastatic potential (7), and suppressed T-cell activity in the tumor immune microenvironment (TIME; ref. 8).

STK11/LKB1 encodes a serine/threonine kinase activating at least 14 related kinases including AMPKs, NUAKs, MARKs, SIKs, and BRSKs and thereby regulates several cellular processes including cell metabolism, cell polarity, and growth (9). Using CRISPR/Cas9-mediated deletion of these substrates, SIK1-3 deletion was noted to recapitulate the accelerated growth noted after Lkb1 deletion in the Kras^G12D GEMM (10, 11). The substrate(s) mediating effects on the tumor microenvironment and checkpoint inhibitor responses, however, remain to be identified.

The widely used antidiabetic drug metformin can reprogram the tumor microenvironment in lung cancer (12), esophageal squamous cell carcinoma (13), and colorectal cancer (14). As metformin also activates the LKB1 substrate AMPK (15), we hypothesized that inactivation of AMPK may contribute to the suppressive TIME in LKB1-deficient tumors. To address this, we investigated the association with AMPK activity and immune suppression in human lung cancer and generated GEMMs, enabling inactivation of AMPK simultaneously with activation of Kras to investigate the causal role of AMPK in immune evasion.

All the experimental mouse strains (below) were generated from Kras^LSL-G12D/+ (16), Lkb1 (17), and AMPKa1^fl/fl and AMPKa2^fl/fl (a kind gift from Benoit Violet) mice. The mice are in mixed genetic background of C57BL/6J and CD-1. All experiments utilized a mixture of female and male mice. Mice aged about 8 to 10 weeks were intranasally instilled with Ad5-CMV-Cre (5 × 10⁷ PFUs, University of Kupio, Finland) or Ad5-SPC-Cre (2.5 × 10⁹ PFUs) in 50-μL MEM per mouse following the protocol previously described (18). Animals were housed in a pathogen-free facility at the Laboratory Animal Center of the University of Helsinki. Animal studies have been approved by the Finnish National Experiment Board (ESAVI/2327/04.10.07/2017).

Surgically resected tumor specimens were received from NSCLC patients at the Hospital District of Helsinki and Uusimaa (HUS), as approved by the ethical board of the Joint Authority of the HUS, Finland (Dnro: 85/13/03/00/15). We obtained written informed consent from the patients and the studies were conducted in accordance with the Declaration of Helsinki.

The Cancer Genome Atlas (TCGA) human lung adenocarcinoma (LUAD) bulk RNA-seq data sets: PanCancer Atlas, TCGA (n = 566; ref. 19), and Collisson et al. Nature, TCGA (n = 230; ref. 1) with RNA-sequencing by expectation maximization (RSEM) estimated batch-normalized median expression values were downloaded from the cBioPortal (https://www.cbioportal.org/). From these, STK11/LKB1-mutant (K^Mut;LKB1^Mut = KL) and KRAS control (K^Mut;LKB1^Wt = K) samples were selected based on loss-of-function mutations in LKB1. Mutation data of the samples were accessed from the cBioPortal (https://www.cbioportal.org/). The fold-change difference in gene expression between KL and K samples was calculated by computing Log2 ratio of the average expression values in LKB1-mutant and KRAS control samples. To generate pAMPK transcriptomes, PanCancer Atlas LUAD cohort was accessed through cBioPortal. Patients were queried for pAMPK (PROT: PRKAA1_PT172), as determined by reverse phase protein array (RPPA). Whole transcriptome, correlated with pAMPK, was uploaded using the co-expression tool. To obtain pAMPK transcriptome specifically in LKB1 wild-type patients, we excluded patients harboring genomic alterations in the LKB1 locus predicted to have oncogenic driver functions (query parameter DRIVER). For the Venn diagram, the overlap between LKB1 mutations and pAMPKα patients were divided into high (Z-score ≥ 0) and low (Z-score < 0) pAMPK groups based on RPPA values obtained through cBioPortal. For the analysis of infiltration levels of different immune cells in PanCancer Atlas patients, precalculated xCell deconvolution results were uploaded from xCell web application (https://xcell.ucsf.edu/). Respective immune subpopulations were aggregated as follows:

• CD8⁺ T cells: CD8⁺ T cells + CD8⁺ naïve T cells + CD8⁺ Tcm + CD8⁺ Tem.

• CD4⁺ T cells: CD4⁺ T cells + CD4⁺ naïve T cells + CD4⁺ Tcm + CD4⁺ Tem + CD4⁺ memory T cells.

• Denditric cells: DC + aDC + cDC + iDC + pDC

• Macrophages: Macrophages + Macrophages.M1 + Macrophages.M2

• B cells: B cells + Memory B cells + Class-switched memory B cells + naïve B cells + Pro B cells

Human and murine lung samples were fixed (formalin for human samples; paraformaldehyde for murine samples), paraffin-embedded, sectioned, and immunohistochemistry (IHC) stained as described earlier (20). The following primary antibodies were used: TTF-1 (1:1,000, ab133638, abcam), p63 (1:5,000, ab124762, abcam), phospho-AMPK-alpha (1:100, No. 2535, Cell Signaling Technology), phospho-ACC (1:800, No. 3661, Cell Signaling Technology), LKB1 (1:500, No. 13031, Cell Signaling Technology), CD8 (1:200, No. 98941, Cell Signaling Technology), CD4 (1:50, No. 25229, Cell Signaling Technology), CD3 (1:150, ab16669, abcam), B2M (1:1,000, ab218230, abcam), and HLA class I ABC (1:150, ab70328, abcam). Stained slides were scanned with a digital slide scanner (3D Histech Pannoramic 250 FLASH II).

About 10-mg tumor tissue was dissected from RNAlater (Thermo Fisher Scientific, AM7020) preserved lung tumor sample, and homogenized in Lysing Matrix Tubes (MP, 6913–100) using Precellys 24 homogenizer (Bertin Instruments, P000669-PR240-A) prior to RNA extraction with RNeasy Mini Kit (QIAGEN, 74104). mRNA-seq libraries were prepared with the SureSelect Strand Specific RNA library prep kit (Agilent Technologies) after PolyA capture, and then sequenced by single-end 75 base pairs using the Illumina NextSeq 500 platform at Biomedicum Functional Genomics Unit. Sequenced reads were aligned as described previously (20). Gene-wise aligned read counts were analyzed for differential gene-expression analysis using limma-voom R package (http://bioconductor.org/packages/release/bioc/html/limma.html). For immune deconvolution of mouse lung tumors, CIBERSORTx algorithm (https://cibersortx.stanford.edu/) was utilized using default parameters using counts per million (cpm)-transformed expression matrix for mouse tumor sequencing results. Custom gene sets for mouse CD8⁺ T cells, CD4⁺ T cells, dendritic cells, monocytes, macrophages, and B cells were generated using published RNA-sequencing data obtained from purified immune populations (GSE109125).

Human KL LUAD immune signature gene set was created by pooling unique genes from the hallmark gene sets (1,383 genes in total) listed in Supplementary Table S1. This provided a list of 973 unique genes (Supplementary Table S2), which were assigned as immune signature genes. The list of immune signature genes was used as gene set to perform gene-expression enrichment analysis (GSEA) on fold-change ranked expressed genes in mouse KL and KAA data sets. The leading-edge genes from the output of this analysis in KL and KAA data sets were selected, and log2 fold-change values of the top 75 genes were visualized using pheatmap package in R (v3.6.2).

Western blotting analysis was performed as described earlier (20). The primary antibodies and their dilutions were as follows: LKB1 (1:2,500, ab15095, abcam), phospho-AMPKa (1:1,000, No. 2535, CST), B2M (1:5,000, ab75853, abcam), H3K27m3 (1:1,000, 07-449, Merck Millipore), HLA class I ABC (1:1,000, ab70328, abcam), and GAPDH (1:15,000, No. 2118, CST). Quantification of the B2M protein level was performed using ImageJ: B2M band sizes were normalized to that of GAPDH, and finally to corresponding empty vector controls.

Statistical analyses were performed using GraphPad Prism 7.0c. Unpaired two-tailed Student t-test or Mann–Whitney test was used to determine the significance between two groups, and ordinary one-way ANOVA with Tukey multiple comparisons test was used to determine the significance among three groups. Log-rank (Mantel–Cox) test was used to determine the significance of Kaplan–Meier plot. Data are presented as mean ± standard error (SEM). Statistical significance was set at P value of NS > 0.05; *, <0.05; **, <0.01; ***, <0.001; and ****, <0.0001.

RNA-sequencing data are available at Gene-Expression Omnibus under the accession number GSE175479.

To interrogate how LKB1 mutation affects the TIME, we made use of two large LUAD patient data sets (PanCancer Atlas, TCGA, 566 samples with RNA-seq data from 510; Collisson et al. Nature, TCGA, 230 samples, named as Collisson data set thereafter; ref. 1) and compared the transcriptome between STK11/LKB1 wild-type and mutant LUAD with KRAS-mutant background. We then probed for potentially relevant expression modules using GSEA performed with the Molecular Signatures Database (MSigDB) hallmark gene set collection (Fig. 1A; Supplementary Fig. S1A–S1C). By using false discovery rate (FDR) < 0.05 as a cutoff, the analysis yielded 22 and 21 negatively enriched hallmark gene sets in PanCancer Atlas and Collisson data sets, respectively (Fig. 1A; Supplementary Fig. S1A). By ranking these hallmark gene sets according to their enrichment scores (normalized enrichment scores (NES) values), eight out of top 10 negative enriched hallmark gene sets are immune-related in both data sets (Fig. 1A; Supplementary Fig. S1A). These include “Interferon gamma response,” “TNFα signaling,” and “IL2–STAT5 signaling” hallmarks representing genes with key roles in stimulating antitumor immune response and explored as targets for cancer immunotherapy (21). Taken together, these analyses demonstrated that LKB1-mutant LUAD exhibit suppressive TIME, consistent with previous studies (5).

Given the availability of RPPA-derived data in a subset of the PanCancer Atlas data set (Fig. 1C), we sought to examine the possible association between TIME and phospho-AMPKα (pAMPKα; Thr183 in AMPKα1 and Thr172 in AMPKα2) phosphorylated by LKB1. Consistent with the notion that LKB1 is required for AMPK activation (9), we noted that LKB1 mutations were enriched in the lo-pAMPK group (76%; green) as compared with the hi-pAMPK group (23%; blue; Fig. 1C). We then constructed a transcriptomic profile of LUAD with low pAMPKα (lo-pAMPK) by ranking (reversed) genes with respect to their correlation coefficients between gene expression and AMPKα phosphorylation level (low to high pAMPKα). Interestingly, the GSEA indicated that lo-pAMPK was also associated with a suppressive TIME as evidenced by the same hallmarks identified (Fig. 1B), suggesting inactivation of AMPK could mediate the suppressive TIME in LKB1-mutant LUAD. Importantly, digital deconvolution (xCell) of RNA-seq data from PanCancer Atlas (TCGA) showed decreased infiltration of CD8⁺ and CD4⁺ T cells in lo-pAMPK LUAD (Supplementary Fig. S2A and S2B), further demonstrating that LUAD with attenuated AMPK activity is associated with suppressed TIME.

To establish whether the suppressive TIME observed in LUAD patient with lo-pAMPK is causally linked with AMPK inactivation, we crossed Kras^LSL-G12D/+ mice with mice carrying AMPKα1^fl/fl and AMPKα2^fl/fl alleles or mice carrying Lkb1^fl/fl allele as a control leading to the generation of the following genotypes: (i) Kras^LSL-G12D/+ (K); (ii) Kras^LSL-G12D/+; Lkb1^fl/fl (KL); (iii) Kras^LSL-G12D/+;AMPKα1^fl/fl;α2^fl/fl (KAA); (iv) AMPKα1^fl/fl;α2^fl/fl (AA). Intranasal instillation of adenovirus expressing Cre-recombinase (Ad-Cre) in these mice activated KrasG12D in the lung epithelium and homozygous deletion of Lkb1 in KL and AMPKα1 and AMPKα2 in KAA and AA mice (Supplementary Fig. S3A). Recombination of both AMPKα1 and AMPKα2 flox alleles from the dissected KAA tumors confirmed the deletion of AMPKα1 and AMPKα2 (Supplementary Fig. S3C). Consistently, IHC staining with phospho-Acetyl-CoA Carboxylase (pACC), phospho-AMPKα (p-AMPKα), and LKB1 antibodies on the serial sections from the lung tumors confirmed the reduced expression of Lkb1 in KL tumors and reduced pACC and pAMPK in both KL and KAA tumors (Supplementary Fig. S3D).

As expected (7), KL mice had a shortened median survival (45 days after Ad-Cre) as compared with K mice (170 days after Ad-Cre; Fig. 2A; Supplementary Fig. S3B). Notably, KAA mice also had a significantly shortened survival (127 days after Ad-Cre) compared with K mice (Supplementary Fig. S3B), becoming progressively moribund with respiratory distress starting at 77 days (after Ad-Cre) as compared with 98 days of K mice (Fig. 2A). Consistently, at 90 days after Ad-Cre KAA lungs showed 1.8-fold increase in tumor burden compared with K mice (Fig. 2B and C). The acceleration of tumorigenesis in KAA mice is in line with a recent study using in vivo CRISPR/Cas9-mediated genome editing, demonstrating that inactivation of AMPKα1 and α2 accelerates Kras-driven lung tumorigenesis albeit to lesser extent compared with inactivation of SIK family proteins (10). An earlier study (22) did not note accelerated tumorigenesis following AMPK, which may be due to differences in background, approach, or efficiency of recombination.

LKB1 loss has been shown to be associated with the appearance of tumors with squamous differentiation (7) in addition to adenocarcinomas. Based on staining with TTF-1 and p63 antibodies, all tumors in the KAA mice represent adenocarcinomas (Fig. 2D), suggesting that other LKB1 substrates are involved in the squamous differentiation noted in KL tumors.

To investigate whether AMPKα1/α2 inactivation has an impact on the TIME, we analyzed T-cell infiltration in the tumors by staining and quantification (Fig. 2E and F; Supplementary Fig. S3E). In the control KL tumors, a significant reduction of total T cells (CD3⁺, 3.91% of total area in K vs. 1.49% in KL), cytotoxic T cells (CD8⁺, 0.83% of total area in K vs. 0.54% in KL) as well as helper T cells (CD4⁺, 0.49% of total area in K vs. 0.41% in KL) was noted as expected (5, 8, 23). Interestingly, the KAA tumors also exhibited a reduction of infiltrated CD3⁺ (3.91% of total area in K vs. 2.10% in KAA), CD8⁺ (0.83% of total area in K vs. 0.50% in KAA) as well as CD4⁺ T cells (0.49% of total area in K vs. 0.35% in KAA), indicating AMPK loss fosters a non-T-cell–inflamed TIME in Kras-driven LUAD similarly to Lkb1 loss.

To characterize the tumor microenvironment of KL and KAA mouse tumors in greater detail, we measured global gene-expression changes by RNA-seq from primary tumors micro-dissected from K, KL, and KAA lungs. Unbiased GSEA on the fold-change ranked expressed genes in KAA lung tumors showed strong negative enrichment of immune signature gene sets—six of the top 10 hallmark gene sets were immune-related, including interferon and inflammatory responses similarly to KL tumors (Fig. 3A). Consistent with the TCGA LUAD study (1) as well as a study of syngeneic tumor model with Lkb1 deficiency (5), our results did not provide evidence for neutrophil enrichment as noted in one study (8). Next, we compared the KAA mouse model with KL human LUAD (KL1, PanCancer Atlas, TCGA; KL2 (1)), lo-pAMPK human LUAD (low pAMPK; LKB1 wild-type) as well as the KL LUAD mouse model by comparing GSEA hallmarks of each (Fig. 3B; Supplementary Fig. S4A). Remarkably, the sets enriched in KAA shared a striking overall similarity with KL and lo-pAMPK patient sets with several immune-related hallmarks. Interestingly, KAA GSEA demonstrated a higher similarity to human sets than the KL mouse model, which, as expected (5, 8), also displayed negative enrichment of several immune-related hallmarks (Fig. 3B). Furthermore, deconvolution analysis using CIBERSORTx (https://cibersortx.stanford.edu/), a computational method for quantifying cell fractions from bulk RNA-seq profiles, validated the decreased infiltration of CD8⁺ T cells in both KL and KAA as well as CD4⁺ T cells in KAA LUAD (Supplementary Fig. S4B). Therefore, these results are in line with our IHC analysis (Fig. 2E and F; Supplementary Fig. S3E) and provide evidence, at the molecular level, that AMPK inactivation drives a suppressive TIME with high similarity to human LKB1-mutant LUAD, suggesting a major role for AMPK in mediating LKB1 effects on immune evasion.

To further delineate clinically relevant mechanisms and genes that could contribute to the immune evasion observed in both KL and KAA mouse LUAD, we derived an immune signature gene set consisting of 973 genes (Supplementary Table S1) by pooling unique genes from eight immune function–related hallmark gene sets (Supplementary Table S1) in the KL human LUAD data sets (see Materials and Methods; Fig. 3B) and performed GSEA in both KL and KAA mouse LUAD. The analysis revealed significant enrichment of the human KL LUAD immune signature (immune signature) gene set in both KL and KAA mouse LUAD, and notably the significance was much more profound in KAA than in KL mouse LUAD (Fig. 3C). To gain further mechanistic insight into downstream pathways potentially mediating the suppressive immune signature in both KL and KAA LUAD, we performed leading-edge gene analysis from the GSEA (immune signature) of both KL and KAA mouse LUAD (ranked by GSEA of KAA LUAD; Fig. 3D) and focused on genes intrinsic to the tumor cells. We found both KL and KAA tumors have suppressed expression of antigen processing and presentation related genes, including Nlrc5, B2m (Fig. 3D and E; Supplementary Fig. S4C). Notably, both NLRC5 and B2M are frequently downregulated in many different cancers by enabling immune evasion (24, 25), and a recent study identified an evolutionarily conserved function of polycomb repressive complex 2 (PRC2) in epigenetically modulating the regulation of NLRC5 and antigen presentation pathway including B2M (26). Both murine (KAA and KL) and human (KL and lo-pAMPK) LUAD display significantly suppressed expression of PRC2 target genes (Supplementary Fig. S4E), corroborating with previous reports on the regulation of PRC2/EZH2 activity by either AMPK (27) or LKB1 (28). Notably, treatment of the LKB1-mutant NSCLC cell line A549 with the EZH2 inhibitor GSK126 led to restoration of B2M and HLA class Ia expression with concomitant downregulation of H3K27me3 (Supplementary Fig. S4D), suggesting that PRC2/EZH2 activation could be a potential mechanism to silence antigen presentation pathway and elicit the evasive TIME in LKB1-mutant LUAD. Taken together, these analyses demonstrate that the immune signature of KL human LUAD is recapitulated by AMPK inactivation and therefore provide further evidence for the contribution of AMPK inactivation to the suppressive TIME in KL LUAD.

Antitumor immune response depends on the recognition of tumor-specific neoantigens presented through human leukocyte antigen class I (HLA-I)–encoded major histocompatibility complex (MHC) class I molecules by cytotoxic T cells (CD8⁺; ref. 29). Recently, loss of heterozygosity at the HLA-I loci (30, 31) and homozygous loss of ß2-microglobulin (B2M) or downregulation of B2M expression (25) have been identified as a mechanism of immune evasion and resistance to immune-checkpoint inhibitors in multiple cancer types including NSCLC. We noted that B2M was significantly reduced in both KL and KAA mouse LUAD at both transcriptional level (Fig. 3E; Supplementary Fig. S4C) and by IHC (Fig. 4A and B). Importantly, B2M was also reduced in human KL LUAD cohorts (Fig. 4C) and accompanied by reduction of several MHC class I genes (HLA-B, HLA-C, HLA-E, and HLA-F; Supplementary Fig. S5A) as well as immunoproteasome-related genes (PSMB8, PSMB9, PSMB10, and UBE2L6; Supplementary Fig. S5B) as noted by a recent study in human KL LUAD (32). The reduction of B2M expression was also noted in human lung cancer cell lines (Cancer Cell Line Encyclopedia; Supplementary Fig. S5C), demonstrating the reduction in a tumor cell-intrinsic mechanism. To investigate the relationship between AMPK activity and B2M levels more directly, we reconstituted LKB1 null A549 human LUAD cells (Kras^G12S;LKB1^−/−) with a constitutively active AMPKα1 (AMPKα1-T183E). Interestingly, reconstitution of AMPKα1 activity by AMPKα1-T183E increased B2M levels significantly and comparably to reconstitution of LKB1. As expected, expression of wild-type AMPKα1 was not sufficient to reconstitute AMPK activity in a cell line without LKB1 (Fig. 4D; Supplementary Fig. S5D).

To investigate whether low pAMPKα was associated with expression of antigen presentation genes, we compared the mRNA expression of B2M and MHC class I chain encoding genes between pAMPK high and low LUAD with wild-type LKB1. The results indicated that B2M (Fig. 5A) as well as all the MHC class I chain encoding genes (HLA-A, HLA-B, HLA-C, HLA-E, HLA-F, and HLA-G) were significantly downregulated in human LUAD with reduced pAMPK (lo-pAMPK, LKB1 wild-type; Supplementary Fig. S6A). Similar to the observation in mouse KAA and KL LUAD, the MHC class I transactivator NLRC5 was downregulated in lo-pAMPK (LKB1 wild-type, Fig. 5B) and KL human LUAD (Supplementary Fig. S6B). Consistent with KL patients (Supplementary Fig. S5B), we also noted suppressed expression of immunoproteasome components in LKB1 wild-type patients with lo-pAMPK (Supplementary Fig. S6C). Previous analysis of large collection of NSCLC tissue microarray samples (more than 400) revealed that the surface expression of B2M and HLA class I genes correlates with infiltration of CD8⁺ cytotoxic T-cell in the tumors (33). Thus, we compared the composition of T cells between pAMPK high (hi-pAMPK) and low (lo-pAMPK) LUAD by digital deconvolution (xCell) of RNA-seq data from PanCancer Atlas (TCGA) in cases where LKB1 mutations were not identified. Interestingly, lo-pAMPK was associated with decreased infiltration of CD8⁺ (Fig. 5C) and CD4⁺ (Fig. 5D) T cells, similar to LUAD with identified LKB1 mutations as well as with KL and KAA mouse LUAD. Strikingly, the LUAD cases with low pAMPK (lo-pAMPK; Fig. 5E) extended significantly beyond cases with identified LKB1 mutations, suggesting that AMPK inactivation is sufficient to attenuate antigen presentation and immune evasion in human LUAD.

Our study provides evidence that the clinically relevant immune evasion noted in LKB1-mutant lung cancer is due to subsequent inactivation of AMPK and attenuation of antigen presentation.

The identification of AMPK as a mediator of immune evasion provides important mechanistic insight and potential target for therapeutic intervention. The observation complements the earlier reports that the LKB1 substrates MARK1 and MARK4 are required for LKB1 to suppress the EMT transcription factor Snail and metastasis (34) and that the LKB1 substrates SIK1 and SIK3 are required for growth suppression by LKB1 in lung cells (10, 11). Whether the immune evasion noted in this study following AMPK loss—and not studied by others—is causally associated with accelerated progression remains to be resolved thoroughly. Furthermore, our study (using Cre/LoxP-mediated AMPK knockout) is consistent with Murray and colleagues 2019 (using CRISPR/Cas9-mediated gene editing; ref. 10) as we both observed increased tumor area/burden (Fig. 2B and C in this study and Fig. 1D and F in Murray and colleagues; ref. 10) albeit at variable degrees, whereas Eichner and colleagues (22) did not observe increased tumorigenesis using either Ad-Cre or Lenti-Cre-induced Cre/LoxP-mediated AMPK knockout. Although one trivial explanation might be due to the lower Ad-Cre dose used in Eichner and colleagues (1 × 10⁶ PFU vs. 5 × 10⁷ PFU in this study), it is notable that in both murine KL mice—which displays enhanced tumor growth compared with K mice (7)—and LKB1-mutant NSCLC cell lines (35) as well as primary tumors (this study), both AMPK phosphorylation and AMPK-dependent phosphorylation of ACC are nearly completely abrogated, suggesting that inactivation of AMPK is associated with tumor growth in LKB1-mutant NSCLC. Together, these results suggest that LKB1 substrates mediate specific downstream effects of LKB1 contributing to tumor suppression and provides a rationale for not finding the substrates inactivated alone in human cancers.

Consistent with increasing evidence that cancer cells can evade immune surveillance by attenuating antigen presentation machinery (APM; ref. 36), we identified reduction of several APM components including B2M, the scaffolding component of MHC-I complex, in both human and mouse KL LUAD, mouse KAA LUAD as well as in human LUAD with reduced pAMPKα (lo-pAMPK). This is consistent with the observed coregulation of APM components (29) to meet with local requirements for an adequate immune response (29) by master regulators such as NLRC5 (29). Interestingly, NLRC5 is downregulated in both KL and KAA mouse LUAD (Fig. 3D and E) as well as in human KL LUAD (Supplementary Fig. S6B) and lo-pAMPK LUAD (Fig. 5B), providing evidence for NLRC5 as a potential mediator for the regulation of APM genes by LKB1 via AMPK. A potential link between AMPK, NLRC5, and APM components is provided by the observations that AMPK suppresses the PRC2 histone methylase complex through phosphorylation of that PRC2 suppresses NLRC5 and APM components to enable immune evasion in multiple cancer types (26). This is consistent with our observation that other PRC2 targets are also downregulated in both human and mouse KL and KAA (lo-pAMPK) LUAD (Supplementary Fig. S4E) and that EZH2 inhibition led to upregulation of B2m and HLA class Ia expression in the LKB1-mutant NSCLC cell line (Supplementary Fig. S4D). Taken together, these observations provide a possible mechanism for the decreased antigen presentation following defects in LKB1 whereby the resulting inactivation of AMPK enables PRC2 to suppress NLRC5 and APM components such as B2M.

The study identified attenuated phospho-AMPKα (lo-pAMPK) as a new biomarker associating with a suppressive TIME. This is interesting from a clinical point of view as: (i) a suppressive TIME has been associated with resistance to PD-1 immunotherapy (6); (ii) LKB1 mutations represent the most prevalent identified mechanism of primary resistance to PD-1/PD-L1 axis inhibitors in LUAD (5); and (iii) the remarkable similarity of the TIME in LKB1 mutant and lo-pAMPK LUAD (this study). As the landscape of primary resistance to PD-1 immunotherapy in LUAD is not fully understood and the current approaches to investigate loss of LKB1 by mutation screens (37) or LKB1 staining (2) may not be comprehensive due to nongenomic alterations (38) or defects in cofactors required for LKB1 function (39), there is a clear unmet need for new biomarkers to predict checkpoint therapy resistance.

The results here indicate lo-pAMPK represents not only a potential surrogate biomarker for LKB1 inactivation (Fig. 5E) but more importantly a biomarker identifying a larger LUAD set with attenuation of the pathway critical for maintenance of antigen presentation. In PanCancer Atlas, patients for which both genomic and phosphorylation data were available, lo-pAMPK was noted not only in 39 patients with identified LKB1 mutations, but importantly in 114 additional patients with a similar suppressed TIME (Fig. 3B and Fig. 5C and D) and reduced APM genes (Fig. 5A and B; Supplementary Fig. S6A and S6B) as those with LKB1 mutations. This suggests that the primary resistance to PD-1 immunotherapy noted in LKB1-mutant LUAD (5) could extend to the additional patients with lo-pAMPK but without an identified LKB1 mutation. It will be interesting to investigate whether in the additional lo-pAMPK patients LKB1 inactivation had gone undetected (2) or whether there might be alternative ways to attenuate AMPK (40–42). However, from a clinical standpoint the discovery of lo-pAMPK identifying an enlarged set of patients with a suppressive TIME suggests that investigating the potential of lo-pAMPK in predicting primary resistance to PD-1/PD-L1 immunotherapy in patients with LUAD is a priority. Based on the our initial attempt with limited amount of NSCLC patient samples, it is notable that pAMPK and B2M are both absent in LKB1-mutant tumor (Supplementary Fig. S6D), prompting future efforts with extended patient samples to validate the utility of pAMPK IHC as a biomarker.

No disclosures were reported.

Y. Gao: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. P. Päivinen: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Tripathi: Data curation, formal analysis, investigation, methodology, writing–original draft. E. Domènech-Moreno: Data curation, formal analysis, and methodology. I.P.L. Wong: Investigation. K. Vaahtomeri: Conceptualization and investigation. A.S. Nagaraj: Investigation. S.S. Talwelkar: Investigation. M. Foretz: Investigation. E.W. Verschuren: Resources, investigation. B. Viollet: Investigation. Y. Yan: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. T.P. Mäkelä: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We thank Saara Ollila and Elina Niemelä for insightful discussions and Saana Ruusulampi and Melissa Montrose for technical assistance. We thank Gonghong Wei and Qin Zhang for constructive suggestions and help with LUAD patient data analysis. We thank Kaisa Salmenkivi from the Department of Radiology and Pathology at Uusimaa Hospital District (HUS) and Jere Linden from Finnish Center for Laboratory Animal Pathology for histology and pathology examination. We thank Laboratory Animal Center, Genome Biology Unit and Biomedicum Functional Genomics Unit for excellent service.

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.