KRAS-G12D mutation drives immune suppression and the primary resistance of anti-PD-1/PD-L1 immunotherapy in non-small cell lung cancer

2.1 Clinical cohorts

To analyze the association between the KRAS variant status and efficacy of anti-PD-1/PD-L1 immunotherapy in NSCLC, the clinical data of 240 patients treated with anti-PD-1/PD-L1 alone or in combination with anti-cytotoxic T-lymphocyte antigen 4 (anti-CTLA-4) immunotherapy between April 2011 and January 2017 at the Memorial Sloan Kettering Cancer Center (MSKCC, New York, NY, USA) was downloaded from a previously published clinical cohort study [31]. According to the inclusion criteria (patients had KRAS mutations and received anti-PD-1/PD-L1 monotherapy) and exclusion criteria (patients with EGFR mutation), the clinical data of 74 patients who had KRAS mutations and received anti-PD-1/PD-L1 monotherapy was included in the analysis.

To investigate the clinical outcome of patients with NSCLC harboring KRAS-G12D mutations after initiation of PD-1/PD-L1 blockade monotherapy or chemo-immunotherapy, the clinical information of 11 NSCLC patients who harbored KRAS-G12D mutations and received immunotherapy, with or without chemotherapy, at the Cancer Hospital and Institute, Chinese Academy of Medical Sciences (CICAMS, Beijing, China) was retrospectively collected between September 2017 and April 2021. The inclusion criteria were that patients had KRAS-G12D mutations and received ICIs; the exclusion criteria were that patients had EGFR mutation and received other treatments in addition to ICIs and chemotherapy. Of the 11 patients enrolled, 3 received anti-PD-1 agent monotherapy, and 8 received chemo-immunotherapy. The last follow-up time was June 15^th, 2021.

A total of 112 KRAS-mutant NSCLC specimens surgically resected between January 2008 and December 2013 were collected from the biobank of CICAMS for immunohistochemistry (IHC) analysis. The inclusion criteria were that the tumor tissues were pathologically diagnosed as NSCLC with KRAS mutations; the exclusion criteria were that patients had EGFR mutation and received chemotherapy or radiotherapy before surgery. This study was approved by the Ethics Committee of CICAMS (approval number: #20/242-2438). The tissue samples were obtained with written informed consent from each patient.

2.2 Cell lines and cell culture

Immortalized human bronchial epithelial cells (Beas-2B) were cultured in bronchial epithelial cell growth medium (BEGM) media (#CC-3170; Lonza, Basel, Switzerland) with 10% fetal bovine serum (FBS, #35-081-CV; Corning, New York, NY, USA). Human lung cancer cell lines (Calu3, H1703, H2030, H358, SK-LU-1, A427, H441, H292, A549, and H460) were cultured in RPMI-1640 media (#10-040-CV; Corning) containing 10% FBS and antibiotic mixture (100 U/mL penicillin and 100 μg/mL streptomycin, #15140-122; Gibco, Billings, MT, USA). H2009 and 293T cells were cultured in Dulbecco's modified eagle medium (DMEM) media (#10-013-CVR; Corning) containing 10% FBS and antibiotic mixture. LA795 cells (clones from the lungs of a 615 mouse with LUAD) were cultured in RPMI-1640 media containing 10% FBS and an antibiotic mixture.

2.3 Reverse transcription quantitative PCR (RT-qPCR), Western blotting, flow cytometry, IHC, and immunofluorescence

RT-qPCR, Western blotting, flow cytometry and IHC were performed as previously described [17, 32, 33]. Total RNA was extracted with TRIzol reagent (#15596; Life Technologies, Carlsbad, CA, USA). TransScript^® II All-in-One First-Strand cDNA Synthesis SuperMix kit (#AH341-01; TransGen, Beijing, China) was used for reverse transcription. RT-qPCR was performed with TransStart^® Top Green qPCR SuperMix (#AQ132; TransGen) on ABI 7900HT Real-Time PCR thermocycler (Life Technologies). The human and mouse PCR primers used in this study are shown in Supplementary Tables S1 and Supplementary Tables S2, respectively.

The primary antibodies for Western blotting used in this study are listed as follows: PD-L1 (#13684T; Cell Signaling Technology, Danvers, MA, USA), KRAS (#3339T; Cell Signaling Technology), KRAS G12D (#14429S; Cell Signaling Technology), β-Tubulin (#2128T; Cell Signaling Technology), Flag (#F3165; Sigma-Aldrich, Saint Louis, MO, USA), GAPDH (#ab8245; Abcam, Cambridge, UK), extracellular signal-regulated kinase 1 and 2 (Erk1/2, #4695; Cell Signaling Technology), Phospho-Erk1/2 (Thr202/Tyr204) (#4370; Cell Signaling Technology), AKT (#4691; Cell Signaling Technology), Phospho-AKT (Ser473) (#4060; Cell Signaling Technology), Phospho-AKT (Thr308) (#13038; Cell Signaling Technology), p70S6K (#2708; Cell Signaling Technology), phospho-p70S6K (Thr389) (#9234; Cell Signaling Technology), p70S6K (#66638-1-Ig; Proteintech, Rosemont, IL, USA), high mobility group protein A2 (HMGA2; #8179S; Cell Signaling Technology), chemokine (C-X-C motif) ligand 10 (CXCL10, #ab137018; Abcam), CXCL11 (#MAB672-SP; R&D, Minneapolis, MN, USA) and Actin (#66009-1-Ig; Proteintech). Briefly, cell protein lysates were electrophoresed using PAGE Gel Quick Preparation Kit (#8012011; Dakewe, Shenzhen, Guangdong, China) and transferred to PVDF membranes (#P2120-2; APPLYGEN, Beijing, China). After incubation with the primary antibodies at 4°C overnight, the membranes with HRP (horseradish peroxidase)-conjugated secondary antibodies at room temperature for 1 h. Finally, we visualized the protein bands using enhanced chemiluminescence.

The antibodies for flow cytometry used in this study included PD-L1 (APC, #329707; BioLegend, San Diego, CA, USA) and its isotype control (APC, #401210; BioLegend). Briefly, the antibodies were used to incubate cells at 4°C for 30 min. Phosphate buffer saline (PBS, #D8537; Sigma-Aldrich) was then used to wash and suspend cells for flow cytometric analysis.

The primary antibodies for IHC used in this study are listed as follows: PD-L1 (SP263 clone, #740-4907; Ventana Medical Systems, Oro Valley, AZ, USA), CD8 (#ZA-0508; Zsbio Tech, Beijing, China), HMGA2 (#8179S; Cell Signaling Technology), CXCL10 (#MAB2662-SP; R&D), CXCL11 (#MAB672-SP; R&D), PD-L1 (#ab238697; Abcam), CD8 (#98941; Cell Signaling Technology), CXCL10 (#10937-1-AP; Proteintech) and CXCL11 (#MAB572-SP; R&D). Briefly, the tumor tissue sections were incubated with these primary antibodies at 4°C overnight and then incubated with HRP-conjugated secondary antibodies at 25°C for 1 h after antigen retrieval. The PD-L1 tumor proportion score (TPS) and the proportion of CD8⁺ T cells were assessed according to the evaluation criteria of the previously published approach [17]. The IHC score was calculated using the following formula: IHC score = staining intensity × percentage of positive tumor cells × 100. There are four grades of staining intensity: no color staining was recorded as 0, pale yellow staining as 1, yellow staining as 2, and brown-yellow staining as 3 [34].

The primary antibodies for immunofluorescence used in this study included CD86 (#GB13585; Servicebio, Wuhan, Hubei, China) and CD11c (#GB11059; Servicebio). Briefly, the tumor tissue sections were incubated with these primary antibodies at 4°C overnight, and then stained with 4',6-diamidino-2-phenylindole (DAPI, #28718-90-3; Beyotime, Shanghai, China) to label nuclei for 15 min. Next, the sections were probed with secondary antibodies with fluorescence according to the primary antibodies used.

2.4 Cell supernatant extraction and enzyme-linked immunosorbent assay (ELISA)

Cell supernatant extraction and ELISA were performed as previously described [32]. The concentration of free CXCL10 and CXCL11 protein in extracted cell supernatants was detected by the Human CXCL10/IP-10 Quantikine ELISA Kit (#DIP100; R&D) and the Human CXCL11/I-TAC Quantikine ELISA Kit (#DCX110; R&D) according to the manufacturer's instructions.

2.5 In vitro tumoricidal activity assays

Human peripheral blood mononuclear cells (PBMCs) were obtained from peripheral blood provided by healthy donators using Ficoll Paque Plus density centrifugation (#17-1440-02; GE Healthcare, Cleveland, OH, USA). We then used the Human CD8⁺ T Cell Isolation kit (#130-096-495; Miltenyi Biotec, Bergisch Gladbach, Germany) to isolate CD8⁺ T cells from the PBMCs, and cultured the isolated CD8⁺ T cells in RPMI-1640 media containing 10% FBS, 200 U/mL interleukin-2 (IL-2, #200-02-50; PeproTech, Rocky Hill, NJ, USA) and an antibiotic mixture. Meanwhile, we added human anti-CD3/CD28 Dynabeads (#40203D; Thermo Fisher Scientific, Waltham, MA, USA) in the media to activate CD8⁺ T cells for 3 days based on the manufacturer's protocol. After A427, SK-LU-1, H2009, H358 and H441 cells adhered to the plate overnight, we co-cultured tumor cells and activated CD8⁺ T cells in a ratio of 1:5 for 48 h in the presence of isotype control or nivolumab (200 μg/mL, #A2002; Selleck, Shanghai, China). Next, we used the Cell Counting Kit-8 (#CK04; Dojindo, Kumamoto, Japan) to quantify living tumor cells after removing T cells and cellular debris.

2.6 Construction of KRAS-knockout cells

We first designed the single-guide RNA (3’-CTGAATTAGCTGTATCGTCA-5’) for mice and single-guide RNA (5’-CAATGAGGGACCAGTACATG-3’) for human using DeepCRISPR [35]. For single-guide RNA cloning, the SpCas9 targeting vector (#H11761; OBiO, Shanghai, China) was then digested with BsmBI (#ER0451; Thermo Fisher Scientific) and ligated with BsmBI-compatible annealed oligos (OBiO). We harvested lentiviruses using 293T cells as previously described [36]. Briefly, 293T cells were co-transfected with the SpCas9 targeting vector and packaging plasmid (pLP1, pLP2, and pLP/VSVG; OBiO) using Lipofectamine 3000 (#L3000015; Invitrogen, Carlsbad, CA, USA). We harvested the infectious lentiviruses at 48 h after transfection. LA795 and Beas-2B cells were then infected with harvested lentiviruses by the addition of 6 μg/mL polybrene (#P4505; Sigma-Aldrich), and single-cell clones were selected with mCherry using flow cytometry sorting. The knockout effect of KRAS in the infected cells was confirmed by Western blotting and Sanger sequencing after selection.

2.7 Construction of stable KRAS-overexpressing cells

For stable overexpression of exogenous KRAS plasmids, coding for wild-type KRAS (WT), mutant KRAS-Gly12Ala (G12A), mutant KRAS-Gly12Cys (G12C), mutant KRAS-Gly12Asp (G12D), or mutant KRAS-Gly12Val (G12V), cDNA was ligated into the lentiviral vector (mouse: #H11219; human: #GL122; OBiO). It is worth noting that the protospacer adjacent motif (PAM) region of single-guide RNA needs synonymous mutation in the cDNA. According to the above-mentioned methods of lentivirus production and infection, we gained the infected cells. Infected single-cell clones were then selected with EGFP using flow cytometry sorting. Western blotting was performed to determine the efficiency of the overexpression.

2.8 In vivo mouse experiments

Female mice (specific pathogen-free, 5-6 weeks old) were used for the animal experiments. The mice were purchased from the Institute of Hematology, Chinese Academy of Medical School (Tianjin, China) and housed at the Center of Experimental Animals of CICAMS. The Animal Care and Use Committee of CICAMS approved all procedures concerning in vivo mouse experiments (approval number: #NCC2020A163).

A total of 615 mice were subcutaneously implanted with murine LA795-derived xenografts that harbored different KRAS oncogene substitutions, including G12A, G12C, G12D and G12V, and treated with anti-PD-L1 monoclonal antibody (10 mg/kg three times weekly; Bio X Cell, West Lebanon, NH, USA) or paclitaxel (20 mg/kg twice weekly; CSPC, Shijiazhuang, Hebei, China) when the tumor volume reached ∼100 mm³. After the treatment intervention, all mice were euthanized by carbon dioxide asphyxiation to dissect the subcutaneous tumor for IHC analysis.

2.9 Transcriptome sequencing data, immune cell infiltration analysis and TMB calculation

HTseq-Counts data of 139 LUAD samples with KRAS mutations were extracted from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/) and divided into two groups with or without KRAS-G12D mutations. We previously identified 1,584 genes that were differentially expressed between the two groups at the thresholds of P < 0.05 and log2(fold-change) > 1 using the “DESeq2” R package [37]. Finally, 39 out of 730 immune-related genes extracted from the nCounter PanCancer Immune Profiling Panel (NanoString) [38] were identified. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed to determine the biological function and pathway correlated with the 39 differentially expressed genes in the DAVID 6.8 (http://david.abcc.ncifcrf.gov).

The Tumor Immune Estimation Resource (TIMER; https://timer.cistrome.org ) was used to calculate the fractions of six intra-tumoral immune cells in each sample, according to the gene expression profile data. Notably, quantile-normalization method was used to standardize the gene expression profile data to eliminate the influence of confounding variables [39].

Somatic mutation data of LUAD were downloaded from the MSKCC dataset (https://www.cbioportal.org/) and TCGA database. By counting the total count of somatic mutations detected in the coding region, except silent mutations, TMB was calculated for each tumor sample.

2.10 Organoid culture

LUAD samples with KRAS-G12D mutations were obtained from the CICAMS and were transported directly to the laboratory after the surgery was performed. The patients provided informed consent. The study was approved by the Ethics Committee of the CICAMS (approval number: #20/242-2438). The tumor samples were washed twice with cold PBS and minced into smaller pieces using scissors. The protocol for establishing organoids was previously described [40]. Organoids were cultured in OrganoPro^TM Tumor Organoids Culture Media (#K2O-M-NSCLC; Ketu Tech, Beijing, China), and passaged at a 1:3 ratio every 2-3 weeks.

2.11 Statistical analysis

SPSS v21.0 (SPSS, Chicago, IL, USA) and Prism v8.0 (GraphPad, San Diego, CA, USA) was used for data analysis. The experimental data are expressed as the mean ± standard deviation, and the data analysis between groups was conducted with Mann-Whitney U tests, one-way analysis of variance, and Student's t-test. Pearson correlation analysis was used to assess the association between samples. We used the Kaplan-Meier method to generate survival curves, and performed comparisons between groups by log-rank tests. P values less than 0.05 were considered statistically significant. The graphical abstract was created with BioRender.com. Based on the ARRIVE1 guidelines, we have completed the reporting checklist for study.

3.1 KRAS-G12D mutation was correlated with the primary resistance of ICI monotherapy in NSCLC

A clinical cohort of 74 NSCLC patients that harbored KRAS mutations and received anti-PD-1/PD-L1 monotherapy was enrolled to explore the correlation between KRAS variant status and the clinical benefit of ICIs (Supplementary Table S3). Depending on the variant status of the KRAS oncogene, these patients were assigned to five subgroups: G12A (n = 7), G12C (n = 30), G12D (n = 11), G12V (n = 12), and others (n = 14). As shown in Figure 1A, there were significant differences among the Kaplan-Meier survival curves of these five subgroups (P = 0.044). Notably, the NSCLC patients with KRAS-G12D mutations had significantly shorter progression-free survival (PFS) than those with other point mutations (P = 0.006; Figure 1B). Although the effect of immunotherapy is slow when compared with conventional anti-tumor therapy, a durable clinical benefit (DCB, PFS > 6 months) can often be obtained [15]. Differences were found among groups when analyzing the DCB of patients with different substitutions (Figure 1C). Of note, the proportion of patients achieved DCB was lower in the G12D group than in the non-G12D group (P = 0.028; Figure 1D). By investigating the relationships between KRAS oncogene substitution and TMB in these patients, the results showed that patients with KRAS-G12D mutations exerted the lowest TMB (Supplementary Figure S1A-B). This finding was also verified using mutation data from the TCGA database (Supplementary Figure S1C).

To further confirm the clinical findings, a co-cultured system consisting of cytotoxic CD8⁺ T lymphocytes and tumor cells was established to evaluate the tumoricidal activity of CD8⁺ T cells cocultured with tumor cells harboring different KRAS oncogene substitutions (Supplementary Table S4). Western blotting and flow cytometry were firstly performed to detect the protein level of PD-L1 in Beas-2B cells and 11 NSCLC cell lines. The results showed that cell lines with KRAS-G12D mutation (A427 and SK-LU-1) had a lower level of PD-L1 than the other cell lines without KRAS-G12D mutation (Figure 1E-F). Of note, the tumoricidal activity of CD8⁺ T cells cocultured with the cell lines harboring non-G12D mutations increased significantly after nivolumab intervention; yet the cells carrying G12D mutations did not (Figure 1G).

In addition to the co-culture model, a mouse model was established to evaluate the efficacy of anti-PD-L1 drugs. We first knocked out the KRAS gene in LA795 cells without KRAS mutation, and then transferred lentivirus vectors coding four common KRAS oncogene substitutions (G12A, G12C, G12D and G12V) in KRAS-knockout LA795 cells (Supplementary Figure S2A-C). A previous study reported that KRAS mutations at different sites did not affect cell growth [27]. To further explore whether KRAS mutations at different sites affect the growth of subcutaneous tumors in mice, we subcutaneously transplanted LA795 cells carrying different KRAS substitutions into nude mice. It was found that there was no significant difference in the growth of subcutaneous tumors in these groups (Supplementary Figure S2D-E). Next, mice loaded with different KRAS point mutations were established by subcutaneously transplanting tumor xenografts derived from mouse LA795 cell lines carrying different KRAS substitutions into 615 mice (Figure 2A). Consistent with our findings in the clinical cohorts, anti-PD-L1 monoclonal antibodies (mAb) in mice with G12D mutations were less effective against tumor growth compared to other groups (P < 0.001; Figure 2B). Furthermore, from IHC analysis, there was a lower protein level of PD-L1 and less CD8⁺ T cell infiltration in mouse tumor specimens carrying the KRAS-G12D point mutation before the intervention of anti-PD-L1 antibodies compared to the other groups. In contrast, after anti-PD-L1 antibody intervention, the mice in the G12D group did not show a markedly decreased PD-L1 protein level and increased CD8⁺ T cell infiltration as did the other groups (Figure 2C-D). In addition, results of immunofluorescence revealed that there was no significant difference in the number of activated CD86⁺ CD11c⁺ dendritic cells among these groups before the intervention of anti-PD-L1 antibodies, whereas the number of these cells in each group increased after anti-PD-L1 antibody intervention (Supplementary Figure S3).

3.2 KRAS-G12D mutation contributed to an immune-suppressive TIME

Based upon 112 resected NSCLC samples with KRAS mutations (Supplementary Table S5), lower proportions of cells positive for PD-L1 protein expression and CD8⁺ T cell infiltration was observed in the G12D group compared with the non-G12D group (PD-L1: P = 0.003; CD8⁺ T cell: P = 0.043; Figure 3A-B). When the TIME was classified according to the PD-L1 status and the presence or absence of TILs [41, 42], significantly higher proportions of the TIME state with PD-L1-negative and no TILs was found in the G12D group (P = 0.009; Figure 3C), suggesting an immune-ignorance TIME in NSCLC samples with KRAS-G12D mutations.

To explore the underlying mechanism of immunosuppressive TIME induced by the KRAS-G12D mutation, we analyzed the gene expression profile of 139 LUAD samples with KRAS mutations from the TCGA database. Among 1584 genes differentially expressed in LUAD specimens with or without G12D mutations, 39 immune genes were determined (Figure 3D-E). We then performed GO and KEGG enrichment analysis to determine the biological function and pathway correlating with the 39 differentially expressed genes. The results showed that these genes were enriched in the leukocyte chemotaxis pathway, leukocyte migration pathway, and cytokine-cytokine receptor interaction pathway (Figure 3F-G). Looking for differential genes related to adaptive immune response, significantly lower expression of genes related to effector T cell activation and killing tumor cells, including CD274, CXCL10, CXCL11 and Granzyme B (GZMB), were observed in the G12D group than in the non-G12D group (Figure 3H). CXCL10 and CXCL11, as chemokines, are involved in the recruitment of CD8⁺ T cells into tumor tissues, thereby resulting in the killing of tumor cells [43]. IHC analysis of the protein levels of CXCL10 and CXCL11 in the 112 resected NSCLC samples with KRAS mutations confirmed that the G12D group showed lower levels of CXCL10 and CXCL11 than the non-G12D group (CXCL10: P < 0.001; CXCL11: P = 0.001; Figure 3I).

3.3 KRAS-G12D mutation suppressed the infiltration of CD8⁺ T cells by regulating the HMGA2-CXCL10/CXCL11 axis

To explore the molecular mechanism by which the KRAS mutation regulates differential expression of CXCL10 and CXCL11 in the tumor microenvironment, we first planned to knock out the KRAS gene in Beas-2B cells which express wild-type KRAS gene (Figure 4A). However, only single-cell clones with heterozygous KRAS gene knockout survived (Figure 4B). Beas-2B cells with heterozygous KRAS gene knockout were then used to stably overexpress four common KRAS oncogene substitutions (Figure 4C). Of note, KEGG analysis of the biological processes and pathways enriched for the differential genes between the G12D-mutant and non-G12D-mutant cells suggested the involvement of cytokine-cytokine receptor interaction (Figure 4D). Next, we looked for the same significantly dysregulated genes between the LUAD samples and cell lines when comparing LUAD with KRAS-G12D mutations with other types of KRAS mutations, and identified 4 upregulated genes and 10 downregulated genes (Figure 4E-F). Then, a cross-correlogram was generated based on Pearson's correlation coefficient values among levels of PD-L1, CXCL10, CXCL11 and 14 significantly dysregulated genes in the LUAD samples with KRAS mutations. A markedly positive relationship between the HMGA2 level and that of PD-L1, CXCL10 and CXCL11 was found (Figure 4G). Considering that immune response is closely associated with the immune cell landscape, a TIMER algorithm was applied to analyze the relationship of HMGA2 level with the intra-tumoral immune cell composition. HMGA2 expression was positively related to the infiltration level of CD8⁺ T cells, neutrophils, myeloid dendritic cells and macrophages (Figure 4H). Furthermore, a positive correlation was confirmed through a combined analysis of the levels of HMGA2, PD-L1, CXCL10 and CXCL11 and the infiltration of CD8⁺ T cells in the IHC-detected group consisting of 112 resected NSCLC samples (Figure 4I). The IHC results of tumor tissues also demonstrated that the KRAS-G12D mutation induced low level of HMGA2 compared to the other mutations (P < 0.001; Figure 4J). Consistent with the IHC results obtained from tumor tissues, Western blotting and RT-qPCR in cells with four common KRAS oncogene substitutions further confirmed that the KRAS-G12D mutation led to low levels of HMGA2, CXCL10, CXCL11 and PD-L1 compared to other mutations (Figure 4K, Supplementary Figure S4A-D). The ELISA results also showed that KRAS-G12D mutation resulted in low secretion of CXCL10 and CXCL11 (Supplementary Figure S4E-F).

To verify the regulation mechanism between HMGA2, PD-L1, CXCL10 and CXCL11 levels, HMGA2-overexpressing Beas-2B-G12D, SK-LU-1, and A427 cell lines were analyzed. As speculated, HMGA2 expression, whether at the protein or messenger RNA (mRNA) level, upregulated levels of CXCL10 and CXCL11. HMGA2 expression also increased levels of secreted CXCL10 and CXCL11. Curiously, the level of PD-L1 was not regulated by HMGA2 expression (Figure 4L, Supplementary Figure S5).

3.4 KRAS-G12D mutation suppressed PD-L1 expression level via the P70S6K/PI3K/AKT axis

Previous studies have demonstrated that KRAS mutation stabilizes PD-L1 mRNA via the MEK-ERK signaling pathway [44]. However, a study reported that in addition to the MEK-ERK pathway, NSCLC cell lines with KRAS-G12D mutations had the activated PI3K-AKT pathway, whereas NSCLC cell lines with KRAS-G12V and KRAS-G12C mutations had the activated RalGDS-Ral pathway and the weakened activation of the PI3K-AKT pathway [27]. Hence, we speculated that different substitutions of the KRAS oncogene might differentially regulate PD-L1 level via the activation of different downstream signaling pathways.

Consistent with the previous study, KRAS mutation, regardless of the type of variation, promoted the phosphorylation of Erk1/2. However, compared to G12A, G12C and G12V point mutations, G12D point mutation activated the phosphorylation of AKT and reduced the phosphorylation of p70S6K (Figure 5A). These findings were also confirmed in LA795 cells with different substitutions of the KRAS oncogene (Supplementary Figure S6). Given that mechanistic target of rapamycin (mTOR) activation leads to feedback repression of the PI3K-AKT pathway through the effector p70S6K [45], KRAS-transfected Beas-2B cell lines were treated with the mTOR inhibitor (rapamycin, 0.5 μmol/L) for 24h, and the impact on the levels of PD-L1, HMGA2, CXCL10 and CXCL11, as well as the downstream signaling, were investigated. Rapamycin induced activation of the AKT pathway in Beas-2B cells expressing G12A, G12C and G12V point mutations. However, in Beas-2B cells expressing the G12D point mutation, the AKT pathway was constitutively activated, and rapamycin did not further increase the phosphorylation of AKT (Figure 5B). In addition, regardless of the type of KRAS variation, PD-L1 levels were reduced to a consistent level under the intervention of rapamycin (Figure 5B, Supplementary Figure S7A). However, after the intervention of rapamycin, the protein or mRNA levels of HMGA2, CXCL10 and CXCL11 as well as levels of secreted CXCL10 and CXCL11 in Beas-2B cells expressing the G12D point mutation remained lowest among the KRAS-transfected Beas-2B cell lines (Figure 5B, Supplementary Figure S7B-F).

3.5 Paclitaxel-based chemotherapy recruited CD8⁺ T cells by upregulating CXCL10 and CXCL11 via HMGA2

A recent study showed that chemotherapy can induce tumor cells to secrete CXCL11 via the upregulation of HMGB1 level, thereby promoting the infiltration of CD8⁺ T cells in NSCLC [46]. Given that HMGA2 and HMGB1 are HMG family proteins, and paclitaxel is approved for the treatment of NSCLC, including LUAD and LUSC [47, 48], three cell lines harboring KRAS-G12D mutations (Beas-2B-G12D, SK-LU-1, A427) were treated with paclitaxel-based chemotherapy, and the effect on the levels of HMGA2, CXCL10 and CXCL11 were evaluated. The results revealed that paclitaxel increased the expression levels of HMGA2, CXCL10 and CXCL11 in a concentration-dependent manner (Figure 6A-C, Supplementary Figure S8). In addition, a patient-derived organoid (PDO) model with KRAS-G12D mutation was constructed and treated with paclitaxel (Supplementary Figure S9). Consistent with the results of the cells, the levels of HMGA2, CXCL10, and CXCL11 in PDO model with KRAS-G12D mutation were also upregulated in a concentration-dependent manner (Figure 6D, Supplementary Figure S10).

3.6 Chemo-immunotherapy as a treatment for KRAS-G12D-mutant NSCLC with primary resistance to ICI monotherapy

As chemotherapy could ameliorate the immunosuppressive TIME induced by the KRAS-G12D mutation by recruiting CD8⁺ T cells via upregulation of HMGA2, CXCL10 and CXCL11, we hypothesized that combined treatment with ICIs and chemotherapy would be a better option for NSCLC with KRAS-G12D mutation. A mouse model with KRAS-G12D mutation was established to receive anti-PD-L1 mAb or paclitaxel as a monotherapy or combined therapy (Figure 7A). Compared to the other groups, anti-PD-L1 mAb with paclitaxel remarkably suppressed tumor growth in the mice (P < 0.001; Figure 7B-C). IHC analysis showed that combined treatment significantly reduced PD-L1 protein levels, increased the protein levels of CXCL10 and CXCL11, and promoted the infiltration of CD8⁺ T cells (Figure 7D-G). Although the infiltration of CD8⁺ T cells and the expression of CXCL10 and CXCL11 increased after the intervention of paclitaxel, more abundant CD8⁺ T cells and more levels of CXCL10 and CXCL11 were observed in the combined treatment group (Figure 7E-G).

The above-mentioned mouse data paved the way for combination treatment with ICIs and chemotherapy as a rational, promising therapy to improve the immunosuppressive state and enhance the infiltration of CD8⁺ T cells in patients with KRAS-G12D-mutant NSCLC. Therefore, we retrospectively collected the clinical information of patients with NSCLC harboring KRAS-G12D mutation after ICI monotherapy or chemo-immunotherapy (Supplementary Table S6). In total, 11 patients were included to undertake treatment efficacy evaluations. Three (27.3%) patients received anti-PD-1 agent monotherapy, and 8 (72.7%) received chemo-immunotherapy. Three (27.3%) patients were identified with partial response, 3 (27.3%) patients with stable disease, and 5 (45.5%) with progressive disease (PD). Because the follow-up time of 2 (18.2%) patients was less than 6 months, 5 (45.5%) patients were determined to have DCB, and 4 (36.4%) to have non-DCB (PFS ≤ 6 months). Although there was no statistical difference, the proportion of PD and non-DCB was lower in the chemo-immunotherapy group compared to the ICI monotherapy group (Figure 8A-B). At the time of survival analysis, patients with NSCLC could obtain more benefit from chemo-immunotherapy with respect to PFS; however, there were no significant difference because of the small sample size (P = 0.362; Figure 8C). Of note, patients with KRAS-G12D-mutant NSCLC receiving chemo-immunotherapy had better overall survival (OS) than those with ICI monotherapy (P = 0.002; Figure 8D).