Integrative proteomic characterization of human lung adenocarcinoma with KRAS G12 mutations reveals molecular pathogenesis

Graphical abstract

Highlights

• •Genomic mutation characteristics of 9,479 patients with solid cancers.
• •Proteomics data from 96 LUAD patients with KRAS G12 mutations, and the correlation with immune profiling.
• •Protein interactome of G12-mutant KRAS proteins.

Abstract

Introduction

Integrated proteogenomic studies of lung adenocarcinoma (LUAD) have provided unique insights with potential clinical effects. Comprehensive proteomic analyses are needed to better understand the molecular landscape of LUAD with KRAS G12 mutations.

Objectives

This study aims to characterize the proteogenomic profile of LUAD with KRAS G12 mutation variants.

Methods

We performed next-generation sequencing (NGS) on 9,479 solid tumors, including 3,523 lung cancers. Proteomic profiling was conducted on 96 LUAD patients with KRAS G12 mutations or wild-type status using liquid chromatography-tandem mass spectrometry (LC-MS/MS). The LUAD tumor immune microenvironment (TIME) was characterized by multiplex immunohistochemistry (mIHC), and protein–protein interactions (PPIs) in KRAS G12-mutant lung cancer cells were mapped using biotin-based proximity labeling.

Results

Genomic analysis revealed heterogeneous KRAS-informed mutational profiles across pan-cancer and lung cancer, with allele-specific resolution for KRAS G12 mutations. Proteomic profiling of KRAS G12-mutant LUAD delineated distinct molecular features and tumor progression hallmarks. Unsupervised clustering identified three molecular subtypes, with the immune modulation subtype (S2) characterized by KRAS G12C enrichment, aggressive clinical features, and the greatest potential benefit from immunotherapy. The immune landscape of LUAD showed increased immune cell infiltration in KRAS G12C-mutant tumors. Additionally, proximal proteomics mapped a landscape of gain-of-interactions driven by KRAS G12 mutations, with ion transporter SLC4A7 emerged as a potential effector in immune modulation in the G12C variant.

Conclusions

This comprehensive proteogenomic study provides insights into the molecular features of LUAD harboring KRAS G12 mutations, which may inform patient stratification and potential therapeutic approaches, pending further clinical validation.

Introduction

Lung cancer ranks first among malignant tumors in the world (Siegel et al, 2024) [1], and is also the leading cause of cancer in China (Han et al, 2024) [2]. Approximately 85 % of patients have a group of histological subtypes collectively known as non-Small Cell Lung Cancer (NSCLC), of which lung adenocarcinoma (LUAD) is one of the most common subtypes, with its incidence steadily increasing over time [3]. Traditional treatment approaches include surgery, chemotherapy, and radiation therapy. However, lung cancer is often asymptomatic in early stage, leading many patients to present with advanced stage, for which curative treatment is often not possible [4]. Recent advances in science have led to the emergence of novel and more efficient methods. Immunotherapy and targeted therapy, along with other treatments, demonstrate successful outcomes in treating advanced lung cancer, especially NSCLC [5].

Multi-omics studies provide a crucial framework for unraveling the molecular mechanisms of LUAD and developing novel therapeutic strategies. Integrating genomics, proteomics, and phosphoproteomics in 103 Chinese patients revealed three distinct subtypes with unique clinical and molecular characteristics, highlighting potential prognostic biomarkers and drug targets. The deep proteogenomic landscape of LUAD has provided valuable insights into oncogenesis, therapeutic vulnerabilities, and actionable targets [6]. Furthermore, studies focusing on non-smoking lung cancer patients in East Asia have revealed distinctive molecular traits and key drivers of disease pathogenesis and progression [7]. These efforts have not only uncovered fundamental biological mechanisms but also highlighted promising avenues for targeted therapeutic interventions.

Although proteogenomic studies in LUAD have provided valuable molecular insights, most analyses to date have focused on heterogeneous patient populations without specific stratification by driver mutations, such as Kirsten rats sarcoma viral oncogene (KRAS) which is one of the most frequently mutated oncogenes in LUAD. KRAS mutations, particularly those involving codon 12, occur in approximately 30 % of LUAD cases and are strongly associated with poor prognosis [8]. Despite advances in targeted therapies, therapeutic resistance remains a major clinical challenge for KRAS-mutant LUAD. These issues underscore the need for comprehensive multi-omics studies that dissect the unique molecular landscape of LUAD harboring KRAS mutations.

The tumor immune microenvironment (TIME) is a complex and multifaceted cellular environment, composed of a diverse array of cell types, including tumor cells, immune cells like T cells, B cells, and macrophages, as well as stromal cells like endothelial cells and fibroblasts [9]. These cells interact to establish an immunosuppressive environment that facilitates tumor growth while maintaining potential for anti-tumor immune responses. KRAS mutations mediate immune escape by regulating the intrinsic characteristics of tumor cells [10,11]. For example, mutant KRAS is able to upregulate PD-L1 expression through p-ERK signaling [12] to drive tumor cells to secrete IL10 and TGFβ1 by activating the MEK-ERK-AP1 pathway [13], and to modulate the infiltration of regulatory T cells (Tregs) [14]. NSCLC patients with KRAS G12C mutation have higher tumor mutation burden (TMB) and PD-L1 expression. Co-mutation of TP53 and KRAS results in elevated PD-L1 levels, increased CD8+ TIL recruitment, and higher TMB [15]. In NSCLC, TP53/KRAS mutations are associated with better clinical benefit to immune checkpoint blockade [16].

In this study, we constructed a genomic map of KRAS mutations and their associated genetic interactions in a pan-cancer cohort of 9,479 cases, as well a lung cancer sub-cohort of 3,523 cases. Proteomic profiling further elucidated the molecular attributes associated with KRAS G12-mutant LUAD. Notably, we identified an immune modulation subtype (S2) strongly linked to KRAS G12C. LUAD with KRAS G12 mutation displayed distinct TIME, characterized by increased immune cell infiltration. Additionally, proximal proteome revealed a network of gain-of-interactions driven by KRAS G12 mutations, with the ion transporter SLC4A7 acting as a potential regulator of immune suppression in the G12C variant. These findings deepen our understanding of LUAD with KRAS G12 mutations and provide potential targets for therapeutic interventions.

Methods

Study population

The next-generation sequencing technology (NGS) data of this study were provided by the Department of Molecular Diagnosis, Sun Yat-sen University Cancer Hospital. NGS data of 9479 patients were collected for genomic analysis. NGS detection was performed using a Gene + 2000 sequencer (Gene+, Suzhou) within the MGI sequencing platform. Target region capture high-depth sequencing was performed on all exon regions of 312 genes, introns of 38 genes, promoter or fusion breakpoint regions, and coding regions of 709 genes. Four types of tumor gene mutations (including point mutations, small fragment insertions and deletions, copy number variations, and currently known fusion genes) were detected. The original sequencing data was processed by the standard process of the molecular diagnostics department, including the quality filtering criteria for using the BOX analyzer (Gene+, Beijing) to detect the original data, including quality standards and quantitative standards. For single nucleotide variants (SNVs) and Indel mutations, the number of mutant reads must be greater than eight and five, respectively. For hotspot mutations, the requirement is greater than four mutant reads for SNVs and greater than two mutant reads for Indels.

The discovery cohort comprising 96 human LUAD tissues samples was collected at Sun Yat-sen University Cancer Center. Patients were consecutively enrolled between 2019 and 2021 and were pathologically and clinically confirmed to have primary lung adenocarcinoma. Only patients who underwent curative surgery, had next-generation sequencing (NGS) results were included. Patients who had received any neoadjuvant treatment, such as chemotherapy or radiation therapy prior to surgery, or had a history of other malignant tumors were excluded. Among those who met these criteria—including KRAS WT, G12C, G12D, and G12V mutation cases—a total of 96 LUAD patients were randomly selected to form the final study cohort (Supplementary Fig.10). Written informed consent was obtained from all participants, and the study was approved by the Institutional Research Ethics Committee of Sun Yat-sen University Cancer Center (SL-G2022-126–01).

External cohorts

To compare the hotspot gene mutation frequencies in the SYSUCC matrix with those in other datasets, we additionally analyzed the MSKCC [17] and Chinese [18] cohorts, with data downloaded from cBioPortal. Furthermore, we incorporated externally published data on lung adenocarcinoma to characterize protein expression in different KRAS mutation states and to verify the consistency and universality of the expression profile.

Genomic data analysis

SELECT analysis

The identification of co-occurrence or mutual exclusivity of genetic alterations was performed using the SELECT algorithm, implemented in the select R package (version 1.6). The analysis was conducted with the GAM (genetic alteration matrix) as the input, allowing unbiased evaluation of all alteration pairs without any a priori assumptions. To estimate the expected background signal, 1000 random matrices were generated to establish a null model, while other parameters were kept at their default settings. To improve stability and reduce variance arising from the seed used for random matrix generation, SELECT was run 110 times, and the median SELECT score was computed for each co-occurrence and mutual exclusivity. Statistical significance was determined using the p-value of weighted mutual information (wMI) [19], [54].

Oncoplot visualization of gene mutation landscape

The oncoplot is generated using the ComplexHeatmap package in R, with the oncoPrint function serving as the main tool for visualization. The process starts by loading and preprocessing mutation data, ranking genes based on mutation frequencies, and selecting the top mutated genes. Each mutation type is assigned a specific color, and the data is formatted accordingly. The oncoPrint function is then used to create the plot, displaying mutation patterns across samples in a heatmap-like format.

Visualization of mutation sites

The mutation sites were visualized using the trackViewer package in R. First, the mutation data for the gene were processed, and the mutation frequencies were calculated. The mutation type and frequency at the specific locus were displayed at the gene locus. Finally, the mutation sites were depicted using the lolliplot function, providing a clear, visual representation of the mutation types and their frequencies.

Proteomic analysis

Protein extraction and sample preparation for lc-ms/ms analysis

Proteins were extracted from FFPE tissue sections (larger than 50 μm) or cells. For FFPE tissues, the following procedures were performed after deparaffinization with xylene and rehydration in a gradient ethanol series: high-concentration SDS lysis buffer (0.3 mmol/L Tris-HCl, 4 % SDS, pH 8.0) with protease inhibitors was added to the tissue. Sonication was performed at 20 % amplitude for 5 min (4 s on, 4 s off). The sample was then centrifuged at 12,000 g for 10 min to remove insoluble debris. The supernatant was transferred to a new tube and heated at 99 °C for 60 min to reverse protein crosslinking. Protein precipitation was carried out using pre-chilled acetone to remove contaminants. For both FFPE-derived and cell-derived proteins, 8 M urea was added to dissolve the proteins, followed by trypsin digestion using the Filter-Aided Sample Preparation (FASP) method. Peptides were desalted with C18 Tips (Thermo Fisher Scientific, Cat. No. 87782) and reconstituted in 0.1 % formic acid (Thermo Fisher, USA) with iRT-Standard (Biognosys, USA) added for alignment. LC-MS/MS analysis was conducted on an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) in data-independent acquisition mode. Protein identification and quantification were performed using Spectronaut software (Biognosys, Zurich).

Differential analysis in LUAD proteome

Proteins were required to be expressed in at least 50 % of the experimental group. Data were log₂-transformed and analyzed to identify differentially expressed proteins between characteristic group and non-characteristic group. For each protein, we use self-made R scripts, a two-step statistical approach was applied: normality was assessed using the Shapiro-Wilk test, followed by either a t-test (for normally distributed data) or a Wilcoxon rank-sum test (for non-normal data). Log₂ fold changes (log₂|FC|) were calculated. Proteins with p-value < 0.05 and absolute FC > 1.5 were classified as significantly upregulated or downregulated.

Analysis of characteristic proteins

To address the overlap of differentially expressed proteins across the four groups, which results from comparing each group against the other three, the following screening criteria were applied: (1) proteins that were consistently up-regulated or down-regulated in multiple groups were excluded; (2) for proteins that were classified as both up-regulated and down-regulated across different groups, only the up-regulated status was retained in the group where the protein was up-regulated. After applying these criteria, we identified a total of 78 characteristic proteins with prognostic value, based on univariate Cox regression (p < 0.05) and mutation classification (ROC, AUC > 0.65). Druggability information was obtained from the Drug Gene Interaction Database (http://www.dgidb.org/) and is provided alongside information about the availability of FDA-approved drugs targeting these proteins. The log-transformed druggability score reflects the number of PubMed journal articles that support the drug-gene interaction. Proteins with specific functions or types, including those related to the druggable genome, transcription factors, enzymes, transporters, plasma membranes, and growth factors, were identified [55].

Model construct

To stratify LUAD communities with KRAS G12 mutations, stringent cutoffs were applied for differential quantification (log₂|FC| > 0.58, p < 0.05), prognosis assessment (univariate Cox regression, p < 0.05), and mutation classification (ROC, AUC > 0.7). This process yielded 31 intersecting candidate biomarkers. These features were then subjected to LASSO regression, and those with coefficients greater than 0.01 were selected. Subsequently, an 11-protein panel was constructed using a logistic regression model. Model performance was evaluated using 5-fold cross-validation (K = 5).

Functional enrichment analysis

To identify significantly regulated pathways enriched in differentially expressed proteins (DEPs) in LUAD tissues, we performed Gene Ontology (GO) biological process (BP) enrichment analysis using the clusterProfiler R package. Subsequently, similarity clustering was conducted on the enriched signaling pathways using the GOSemSim R package.

Functional module enrichment analysis

We employed the enrichment method [24] based on ranking to assess the immunity and cell cycle progression (CCP) score [56] underlying our LUAD subtypes. Firstly, we ranked proteins of each sample in descending order of abundance, then determined for each immune pathway protein the probability that it would by random chance occupy its observed or a higher position in that list. An individual protein would therefore have a smaller probability (be enriched toward the top of the list) the higher it was on the list. The process was then repeated, this time combining individual probabilities that a protein was enriched toward the bottom of the abundance list to assess for significant underrepresentation of the immune pathway in a sample. The inferred pathway activation score was defined as the negative log of the ratio of these two probabilities. This score is positive when pathway proteins occur in the top half of the abundance list, and negative when confined to the bottom.

Additionally, for the hypoxia expression score, we calculated it by subtracting the median expression of the genes with the lowest correlation from the median expression of the genes with the highest correlation. According to the gene set of the reference [56], the calculation is based on the gene existing in the LUAD proteome.

Proteomics subtyping analysis

Molecular subtype identification and integrative analysis were performed on proteomic data from 96 LUAD tumor samples using the R package MOVICS [57], with the aim of identifying features with distinct expression patterns for each cluster. The raw data were log₂ transformed, and missing values were imputed using the k-Nearest Neighbors (kNN) algorithm. To determine the optimal number of clusters, we evaluated a range of cluster numbers from k = 2 to k = 8. Based on the cophenetic coefficient, k = 3 was selected as the optimal clustering solution. Subsequently, molecular subtyping was conducted using the built-in function getMOIC, based on three algorithms: SNF (Similarity Network Fusion), ConsensusClustering, and iClusterBayes. The results from these methods were integrated to derive the final subtype classification.

Network construction

We used STRING (https://cn.string-db.org/) to obtain the interaction between proteins, which are characteristic proteins of immune-related pathways in KRAS G12C. Protein-protein interaction (PPI) was generated based on Cytoscape 3.10. The color depth represents the size of log2 (FC). The size of the circle corresponds to the degree of connectivity.

Tumor immune microenvironment

Fluorescent multiplex immunohistochemistry (mIHC)

Fluorescent multiplex immunohistochemistry (mIHC) was conducted using Opal 6-Colour Manual IHC Kits (PhenoVision Bio, Beijing) according to the manufacturer’s protocol. In this study, two panels were used to determine the immune pattern of the case. Of which, the order of staining of the slides with antibodies and fluorescent dyes, the first panel is anti-CD4/Opal-520, anti-Foxp3/Opal-690, anti-PD1/Opal-570, anti-Cytokeratin/Opal-480 and anti-CD8/Opal-780. The second panel is anti-CD19/Opal-690, anti-CD163/Opal-570, anti-Cytokeratin/Opal-480, and anti-CD68/Opal-780. The mIHC staining was visualized by Akoya Vectra Polaris (Perkin Elmer, Waltham) and analysed using digital pathology analysis platform HALO (Indica Labs, Corrales). We applied HALO® Highplex FL v4.1.3 algorithm to identify different kinds of immune cells according to the cell morphological features and the pattern of fluorophore expression.

Immunohistochemistry (IHC)

Immunohistochemistry (IHC) was used to evaluate the expression of SLC4A7 and SLC16A3 in lung adenocarcinoma tissues from patients with different KRAS mutation statuses, employing manual IHC methods. Formalin-fixed paraffin-embedded tissue sections were cut into 4 μm thick slices and baked at 60 °C for 3 h, followed by deparaffinization with xylene and rehydration through a gradient of ethanol. Antigen retrieval was performed using an antigen retrieval solution (Zhongshan Golden Bridge, Beijing) in a pressure cooker, boiling for 2.5 min. The sections were then treated with 3 % hydrogen peroxide to block endogenous peroxidase activity and incubated with primary antibodies overnight at 4 °C. After three washes with PBS buffer, the sections were incubated with anti-rabbit or anti-mouse HRP-conjugated secondary antibodies (Zhongshan Golden Bridge, Beijing) for 30 min, followed by three additional washes with PBS. Finally, DAB chromogen substrate was added for staining, and the staining intensity was observed under a microscope. The sections scanned by an automatic slice scanning system (Kfbio), and the analysis was performed using the digital pathology analysis platform HALO (Indica Labs, Corrales). The antibodies used in the experiments were as follows: anti-SLC4A7 (Abmart, TD9927) and anti-SLC16A3 (Proteintech, 22787–1-AP).

Networks of immune microenvironment and proteome

The Spearman correlation coefficient was employed to analyze the correlation between the infiltration density of 10 immune cells and the expression of differential proteins in 96 lung adenocarcinoma (LUAD) samples. Proteins with an absolute correlation coefficient value greater than 0.25 and a p-value < 0.05 were included in the mapping of immune cell networks. Blue edges indicated a negative correlation, and red edges indicated a positive correlation. The width of the edge indicates the magnitude of the correlation coefficient. The size of the circles corresponds to the degree of connectivity. The network was constructed, visualized, and generated using Cytoscape (version 3.10.2).

Proximal proteome analysis

Protein extraction and sample preparation for AP-MS analysis

Cell samples were collected and lysed on ice for 30 min using RIPA lysis (Cell Signaling Technology) buffer containing 1 % Cocktail. The samples were then centrifuged at 12,000 g for 10 min at 4 °C. 950 μL of the supernatant was collected and incubated with streptavidin magnetic beads (Thermo Fisher, USA) at 4 °C with three-dimensional rotation for 2 h. The eluted bead samples were washed twice with 1 mL of 2 % SDS, 500 μL of TBST (Tris-buffered saline with Tween-20), and 500 μL of TBS, respectively. After discarding the supernatant, the beads were resuspended in 50 μL of 50 mM TEAB containing 0.1 % SDS, and 1 μL of 1 M DTT (final concentration 20 mM) was added. The mixture was incubated at 30 °C with shaking at 300 rpm for 60 min. The temperature was then reduced to 25 °C, and 2 μL of 1 M IAA (final concentration 40 mM) was added, followed by incubation at 25 °C with shaking at 300 rpm for 30 min. The remaining supernatant was discarded, and the beads were washed three times with 500 μL of 50 mM TEAB. The beads were then resuspended in 120 μL of 50 mM TEAB, followed by the addition of 1.5 μL of 1 mg/mL trypsin. Incubation was performed at 37 °C with shaking at 900 rpm for 16 h. The samples were desalted using C18 spin columns according to the manufacturer’s protocol. After desalting, the samples were freeze-dried and reconstituted in 0.1 % formic acid. All raw MS data were searched using MaxQuant (V1.6.5.0), with peptides, proteins, and post-translational modifications (PTMs) filtered at a 1 % false discovery rate (FDR). In addition to the default parameters, with a matching between runs time window set to 1 min. A variable mass addition of biotin was applied, and the iBAQ value was used as the quantitative basis. The data were screened to remove reverse, potential contaminants, and proteins only identified by site, and the list of candidate proteins was finalized. The common contaminant proteins include keratin, ACACA, ACACB, MCCA, PCCA, and PYC. Subsequently, the FOT value was calculated by dividing the iBAQ value of each protein in each replicate by the total iBAQ of the corresponding replicate. The formula can be expressed as:

$${\mathit{FOT}}_{ni}=\frac{{\mathit{iBAQ}}_{ni}}{\sum {\mathit{iBAQ}}_n}$$

The identification of interacting proteins meets the requirements: 1) unqiue peptide in three repeated samples is not less than 2; 2) identified in at least two replicates; 3) the FOT value of the experimental group was 3 times that of the control group, and p < 0.05. Additionally, to classify proteins as preferential interactors, a protein must meet the following criteria, fold change > 1.5 compared to the control group and p < 0.05.

Cell culture

A549, HEK293 and HEK293T cell lines were cultured in Dulbecco's Modified Eagle Medium (DMEM, Gibco) supplemented with 1 % penicillin–streptomycin and 10 % fetal bovine serum (FBS, Gibco). The cells were incubated at 37 °C under 5 % CO₂. All cell lines were authenticated by short tandem repeat profiling and tested negative for mycoplasma contamination.

Plasmid construction and Lentivirus production

KRAS WT, KRAS G12C, KRAS G12D, KRAS G12V were generated by PCR amplification and cloned into the pLVX- blasticidin vector for overexpression. Four KRAS sequences with different mutation states with HA tag and Pyrococcus horikoshii biotin protein ligase (PhBPL)-assisted biotin identification (PhastID) tag at the N-terminus were cloned into the pLVX- blasticidin vector. The pLentiCRISPR v2 vector, which encodes the sgRNA was procured from Synbio-technoloies (Suzhou). The sgRNA sequences for KRAS was CACCGGGACCAGTACATGAGGACTGGTTT. Stable cell lines were established by with Lipofectamine 3000 (Invitrogen, Boston), according to the manufacturer's instructions. For biotin-labeled cells, the transfected plasmid was stably expressed in HEK 293 cells. At this time, exogenous biotin (Sigma, # B4639) was added for biotin labeling, so that the biotin concentration in the system was 50 μM, and then expressed for 3 h. Western blotting was used to verify the efficiency of sgRNA-mediated knockout and gene overexpression.

Western blotting

The cells washed in PBS and lysed in cell lysis buffer (Cell Signaling Technology,USA). Protein quantification was performed using a BCA Protein Assay Kit (Thermo Fisher, USA). Equal amounts of protein were then separated by SDS-PAGE and subsequently transferred onto PVDF membranes (Bio-Rad, Hercules, USA). The membranes were first blocked with 5 % skim milk at room temperature for 1 h, followed by overnight incubation with primary antibodies at 4 °C. After three washes with TBST, the membranes were incubated with secondary antibodies for 2  h. Immunoreactive bands were then visualized using an ECL chemiluminescence kit (Bio-Rad, USA) and captured with a chemiluminescence imaging system (Tanon, China).

The primary antibodies used in the experiments were as follows: anti-KRAS (Proteintech, 12063–1-AP), anti-GAPDH (Proteintech,10494–1-AP), anti-HA (Cell Signaling Technology, #3724), and anti-SLC4A7 (Abmart,TD9927). HRP-conjugated goat anti-rabbit/mouse IgG was used as the secondary antibody.

CCK-8 assay

Cell proliferation assays were performed using the Cell Counting Kit-8 (CCK8) (TargetMol, Shanghai) following the manufacturer's instructions. The A549 cells, transfected with the plasmid were seeded into 96-well plates in triplicate at a density of 1000 cells/well in 100  µl of DMEM medium and cultured at 37 °C. At the end of each experiment, cells were incubated with the CCK-8 reagent at a 10 % concentration in DMEM medium for 1.5 h. Subsequently, the absorbance was measured at 450 nm with the microplate reader. We performed this assay at 0  h, 24  h, 48  h, 72 h, and 96  h.

Colony-formation assay

To detect cell colony formation, the cells were seeded onto the six-well plates for 2 weeks, which were then fixed with methanol and stained with 0.5 % crystal violet. The cells were rinsed with tap water and dried at room temperature. Finally, the visible colonies were counted using the Image J software in a blinded manner with the colonies > 0.05 mm included.

Immunofluorescence

Cells were cultured in confocal dishes and fixed using 4 % paraform-aldehyde. Permeabilization was achieved with 0.1 % Triton X-100 in PBS, followed by blocking with 5 % BSA. Subsequently, the confocal dishes were incubated overnight at 4 °C with a primary antibody of anti-HA (Cell Signaling Technology, #3724). The following day, immunofluorescent antibodies were applied and incubated for 2 h at room temperature. Nuclei were counterstained using DAPI. Finally, the cells were visualized and imaged using a fluorescence microscope (Zeiss LSM900).

Co-immunoprecipitation

Immunoprecipitation between KRAS and SLC4A7, cells were lysed in RIPA buffer (1 % Triton X-100, 100  mM Tris-HCl pH 8.8, 100  mM NaCl, 0.5  mM EDTA) supplemented with a complete protease inhibitor cocktail (Bimake, added fresh), and mixed with antibodies at 4 °C for 12–16 h; Protein A/G PLUS-Agarose beads (SANTA CRUZ, SC-2003) were added and incubated at 4 °C for 4 h. Beads were washed three times with Dilution/Wash buffer (50 mM Tris-HCl pH 7.5,150 mM NaCl,1mM EDTA) and subjected to western blotting.

RNA extraction and quantitative PCR

Total RNA was extracted using the Trizol Reagent (Invitrogen, Boston) following the manufacturer’s protocol. cDNA synthesis was conducted using the Hifair II 1st strand cDNA synthesis kit (YEASEN, Shanghai). Quantitative PCR was performed on Bio-Rad CFX 96 (Bio-Rad, Hercules) using Hieff® qPCR SYBR® Green Master Mix (YEASEN, Shanghai) and specific primers. Actin served as an internal control for normalizing the relative gene expression. The sequences of all primers used for qRT-PCR are listed in Supplementary Table 15.

Transwell migration assay

LUAD cells A549 were transfected with the indicated constructs and cultivated for 36 h. For migration assay, 1.8 × 10⁵ transfected cells in 200 µL serum-free medium were seeded into the upper chamber of 8-µm pore size transwell inserts (BD Biosciences, Franklin Lakes). The lower chamber was filled with 600 µL DMEM medium with 20 % FBS. After 14 h of incubation, cells were fixed with 4 % paraformaldehyde and stained with 0.5 % crystal violet. Five random fields were imaged using a microscope (Nikon, Tokyo) and quantified via ImageJ software.

Univariable and multivariable cox regression analysis

After log transformation or standardization, univariable Cox regression analysis was performed to identify prognostic proteins with a p-value < 0.05 based on Wald test. Kaplan–Meier plots were drawn for representative proteins to show their significant relationship between protein expression and the optimal cut point for each protein, determined by surv_cutpoint under the premise that no less than 25 % of each group was guaranteed.

The univariate COX proportional hazards model was also used to test the association of clinical and molecular features with DFS independently first. Based on the univariate analysis result, clinically and biologically relevant features with statistical significance (p-value < 0.05) were selected for the multivariate COX proportional hazards model analysis.

Statistical analysis

Computational and statistical analyses were executed utilizing SPSS software (version 25, IBM, New York, NY, USA), GraphPad Prism 8.0 along with R programming environment (version 4.1.2). The Wilcoxon rank sum test and Kruskal-Wallis test were used to examine differences between continuous variables, and Fisher’s exact test and chi-square test were used to compare associations between categorical variables, when appropriate. Data were analyzed by the Mann-Whitney U test, Kruskal-Wallis test with Bonferroni-adjusted pairwise comparisons. Significance levels were denoted as follows: *p < 0.05, ^**p < 0.01, ^***p < 0.001, ****p < 0.0001.

Results

KRAS-informed mutational profiles

To investigate the KRAS mutation-informed genomic profile of Chinese patients with solid tumors, tumor specimens were collected from 9,479 patients across more than 10 tumor types (Fig. 1A). The major types included lung cancer (Lung; n = 3,523, 37.2 %), colorectal carcinoma (COADREAD; n = 2,607, 27.5 %), ovarian carcinoma (OV; n = 481, 5.1 %), uterine corpus endometrial carcinoma (UCEC, n = 233, 2.5 %), gastric cancer (STAD; n = 222, 2.3 %), cervical carcinoma (CESC; n = 197, 2.1 %), liver cancer (Liver; n = 184, 1.9 %), pancreatic adenocarcinoma (PDAC; n = 156, 1.6 %), breast carcinoma (BRCA; n = 155, 1.6 %) and melanoma (Melanoma; n = 106, 1.1 %) (Fig. 1A). Next-generation sequencing (NGS) data revealed a total of 110,754 mutation events across 1,017 genes (Supplementary Table 1). The most frequently mutated gene in the overall cohort was TP53 (62.2 %), followed by KRAS (23.9 %), APC (23.8 %), and EGFR (22.0 %) (Fig. 1B). TP53 mutations were identified in 59.0 % of individuals with lung cancer. EGFR (51.0 %), LRP1B (16.1 %) and KRAS (10.6 %) were the next most frequently mutated genes (Fig. 1C). Notably, mutation frequencies of the cancer-related genes exhibited ethnic, geographic and anatomical variations across the MSKCC [17], Chinese [18] and SYSUCC cohorts. Specifically, TP53 showed a markedly elevated mutation prevalence in the Chinese population (41.5 % in MSKCC vs. 57.3 % in Chinese vs. 62.2 % in SYSUCC, p = 0.009) (Fig. 1D and Supplementary Table 2). The top-ranking mutations, TP53 and KRAS, also varied in incidence across different anatomical tumor types, mirroring observations from another large-scale Chinese project (Fig. 1E and Supplementary Fig. 1A and Supplementary Table 3) [18]. Furthermore, the two most frequently occurring TP53 mutations were identified as R273 and R248 across pan-cancer and lung cancer (Supplementary Fig. 1B). For KRAS mutations, G12, G13, and Q61 mutations represent hotspots (Supplementary Fig. 1C). Interestingly, KRAS G12C mutation, which is notably prevalent in lung cancer, was surpassed by G12D, G12V and G13D mutations across pan-cancer (Fig. 1F).

Fig. 1.

Genomic landscape of patients with solid cancers. A. The overview of genomic cohort. B. Genomic profile of 9,479 patients with solid tumors in SYSUCC cohort. C. Genomic profile of 3,523 patients with lung cancer in SYSUCC cohort. D. Bar plots depicting mutation frequencies of cancer-related genes across MSKCC, China and SYSUCC cohorts. E. Bar plots depicting mutation frequencies of KRAS and TP53 across multiple cancer types in SYSUCC cohort. F. Donut charts of mutation frequencies of specific KRAS-alleles mutation in pan-cancer and lung cancer. G-H. Co-occurrence and mutually exclusivity of KRAS-associated genomic alterations in pan-cancer (G) and lung cancer (H). The horizontal axis represents the deviation between the observed and expected frequencies of genomic interaction pairs co-occurring within the same sample. The vertical axis indicates statistical significance.

We sought to identify pairwise interactions among mutations using the SELECT algorithm [19]. For KRAS, we detected 32 mutual exclusion events in the pan-cancer cohort and 41 genetic interaction events in lung cancer, with 8 co-occurring and 33 mutually exclusive events (Fig. 1G-Hand Supplementary Table 4). KRAS and EGFR represented clear mutual exclusivity, while no interaction was observed between KRAS and TP53 mutations [7].

Next, to characterized genomic heterogeneity in global context, we conducted a similarity analysis using mutation frequency between KRAS mutation status, as well across mutant alleles (Supplementary Fig. 2A–B). The results demonstrated that the mutational patterns of the KRAS mutation (MUT) group exhibited a greater propensity for deviation compared to the KRAS wild-type (WT) counterpart, which was more pronounced in lung cancer (r = 0.95 vs. r = 0.83 in pan-cancer; r = 0.99 vs. r = 0.58 in lung cancer) (Fig. 2A–B). The genomic correlation coefficients between the KRAS WT and MUT groups were 0.62 in pan-cancer and 0.46 in lung cancer, respectively (Fig. 2A–B). Concurrently, several genes exhibited markedly elevated mutation rates depending on KRAS mutation status. Examples include APC, SAMD4, FBXW7 and TCF7L2 in pan-cancer and LRP1B, STK11, KEAP1 and ATM in lung cancer (Fig. 2A–B and Supplementary Table 5). To further delineate the KRAS mutation at allele-specific resolution, we found a comparable pattern of KRAS-mutant allele-specific atlas (Supplementary Fig. 2C–E). In lung cancer, KRAS G12D exhibited the most pronounced difference, while there was a high degree of similarity between specific mutations (Fig. 2C–D). Only a limited number of genes showed significantly disparate mutation frequencies (Supplementary Table 5). For instance, LRP1B, RBM10, KEAP1 and STK11, were significantly enriched in KRAS G12C in both pan-cancer and lung cancer (Fig. 2D). Additionally, certain oncogenes displayed a higher proportion of mutations in KRAS G12D and G12V variants, such as PIK3CA, PIK3CG, NKX2-1, TERT and SETBP1, as well as tumor suppressors APC, CDKN1A, BRCA1, KDM6A and AXIN2 (Supplementary Fig. 2F–G). A greater proportion of individuals exhibited combined mutations in one or more tumor suppressors (Supplementary Fig. 2H–I). KRAS mutant tumors differed in the frequency of individual or combined mutations of TP53, LRP1B and STK11, with alterations in one or more tumor suppressors being more common than in the KRAS G12C variant (Fig. 2E). These results indicated the presence of distinct mutational subtypes and increased inactivation of tumor suppressors associated with mutated KRAS.

Fig. 2.

KRAS-informed mutational profiles. A. Correlation heatmap (left) and scatter plot (right) illustrating mutation frequencies across different KRAS statuses in pan-cancer. Significant differences in mutation frequencies are highlighted in red. B. Similar analysis as in A, specifically to lung cancer. C. Correlation heatmap illustrating mutation frequencies across specific KRAS-mutant alleles, presented for both pan-cancer (left) and lung cancer (right). D. Clustergrams highlighting mutations enriched significantly in specific KRAS variants. E. Pie charts showing the proportions of individual and combined mutations in TP53, LRP1B, and STK11 for each KRAS variant (WT, G12C, G12D, and G12V). F-G. Violin plots depicting tumor mutation burden (TMB) across KRAS statuses. H. Bar chart showing microsatellite stability across different KRAS statuses. I-J. Kaplan-Meier survival curves for patients stratified by different KRAS statuses. K. Correlation heatmap showing mutation frequencies across combinations of KRAS and TP53 mutations. L. Heatmap representing the normalized average mutation frequency of pathways/biological processes influenced by KRAS and TP53 statuses. M. Heatmap representing the significant fold changes in mutation frequency associated with KRAS and TP53 statuses, referenced to the overall mutation frequency observed across all cases. N. Violin plot of TMB stratified by KRAS and TP53 statuses. O. Kaplan-Meier survival curves for patients grouped by combinations of KRAS and TP53 statuses. P. Bar chart of microsatellite stability stratified by KRAS and TP53 statuses.

In comparing the clinicopathologic features of KRAS status, significantly higher median TMB was observed in the KRAS MUT cases, particularly in the G12C variant (Fig. 2F–G). KRAS MUT cases showed a slightly elevated fraction of high-microsatellite instability (MSI-H), mainly observed in the G12D variant (Fig. 2H and Supplementary Fig. 2J). Disease-free survival (DFS) was better in patients with KRAS WT, compared to those with KRAS MUT (median DFS: 30.3 months vs. 16.7 months in pan-cancer; 58.7 months vs. 40.1 months in lung cancer) (Fig. 2I–J). In addition, lung cancer patients with KRAS G12C were significantly more likely to have an older age of onset, be predominantly male, and have a smoking history. Other baseline clinicopathologic characteristics of patients with KRAS allele-specific mutations are presented (Supplementary Table 6).

The genomic subgroup landscapes in this cohort were shaped by the presence or absence of TP53 mutations, as well as distinct clinicopathological characteristics (Supplementary Fig. 3A–E and Supplementary Table 5). When comparing combinations of KRAS and TP53, the greatest genomic heterogeneity in global context was identified between co-mutated and co-wildtype groups, as well as between single mutations (r = 0.26 and r = 0.23 in pan-cancer; r = 0.17 and r = 0.14 in lung cancer) (Fig. 2K). Additionally, the distribution of changes in individual genes or gene sets associated with tumor-related biological processes varied among the KRAS-TP53 cooperation subgroups (Fig. 2L–M and Supplementary Fig. 3F–I). A comparison of clinical features revealed that both individual and combined mutations were characterized by high TMB and poorer disease progression (Fig. 2N–O). A higher proportion of patients harboring only KRAS mutations in the pan-cancer cohort exhibited MSI-H, whereas lung cancer patients as a whole demonstrated MSS, irrespective of mutation status (Fig. 2P).

The proteomic characteristics of LUAD patients with KRAS G12 mutation

To reveal the proteomic features associated with KRAS G12mutation, a deep-scale proteomic analysis was conducted on 96 treatment-naive and surgically resected LUAD samples. The prospective cohort consisted of 31 patients with KRAS WT, 22 with KRAS G12C, 21 with KRAS G12D, and 22 with KRAS G12V (Fig. 3A). The clinicopathological characteristics of the patients are summarized in Table 1. The postoperative recurrence rates were significantly higher in patients with KRAS G12 mutations, whereas no clear separation was noted across mutant allele resolution (Fig. 3B). An indexed retention time (iRT)-based label-free quantification approach was employed to the precise relative quantification of protein. A total of 11,529 proteins were quantified, and 10,601 proteins were common across all subgroups (Fig. 3A). The number of proteins identified per sample ranged from 7,073 to 9,261 (Supplementary Fig. 4A). Principal-component analysis (PCA) showed no discrete clustering between KRAS WT and G12 mutations (Supplementary Fig. 4B). Overall, 8,610 proteins were identified with a sample coverage rate exceeding 50 %, and these were consequently included in the subsequent analysis (Supplementary Table 7).

Fig. 3.

Proteomics of LUAD patients with KRAS G12 mutations. A. Overview of the proteomic cohort and the quantified protein numbers for four subgroups. An upset plot displays the common and unique protein sets across subgroups. B. Disease-free survivals for patients with different KRAS G12 mutations. C. Heatmap of differentially expressed proteins and clinical information. D. Rank-based enrichment analysis using fold changes in protein abundance. Red indicates enrichment of upregulated proteins, blue for downregulated. Dot size and color intensity reflect enrichment strength and direction. E. Representative differentially expressed proteins within specific pathways, showing well-correlation with the clinical outcomes. F. Area under the curve (AUC) values calculated using individual features and an eleven-feature panel. G. Disease-free survival curves for patients with combined KRAS and TP53 mutation statuses. H. Gene set enrichment analysis illustrating dysregulated pathways in LUAD patients with co-occurring KRAS and TP53 mutations. Yellow nodes indicate enrichment of upregulated proteins, purple for downregulated.

Table 1.

Baseline patient characteristics (n = 96).

Variables	KRAS WT (n = 31)	KRAS G12C (n = 22)	KRAS G12D (n = 21)	KRAS G12V (n = 22)	P
Age					0.453
≥60	18(58.1 %)	13(59.1 %)	13(61.9 %)	12(54.5 %)
<60	13(41.9 %)	9(40.9  %)	8(38.1 %)	10(45.5 %)
Mean ± SD	58.03 ± 12.55	64.23 ± 7.30	58.48 ± 13.70	61.27 ± 9.13
Range	21–78	54–80	24–73	45–77
Sex					0.001
Male	18(58.1 %)	22(100.0 %)	13 (61.9 %)	16 (72.7 %)
Female	13 (41.9 %)	0 (0.0 %)	8 (38.1 %)	6 (27.3 %)
Smoke					0.019
Yes	19 (61.3 %)	4 (18.2 %)	9 (42.9 %)	9 (40.9 %)
No	12 (38.7 %)	18 (81.8 %)	12 (57.1 %)	13 (59.1 %)
Stage					0.271
I	21 (67.7 %)	11 (50.0 %)	14 (66.7 %)	13 (59.1 %)
II	2(6.5 %)	4 (18.2 %)	6 (28.6 %)	4 (18.2 %)
III	6(19.4 %)	6(27.3 %)	1 (4.8 %)	5(22.7 %)
IV	1(3.2 %)	1(4.5 %)	0(0.0 %)	0(0.0 %)
Tumor					0.146
I	21(67.7 %)	15 (26.4 %)	12(19.0 %)	10(23.6 %)
II	7(22.6 %)	7 (19.4 %)	4 (11.9 %)	8 (18.0 %)
III	2(6.5 %)	0(0.0 %)	5 (19.0 %)	3 (16.7 %)
IV	0(0.0 %)	0(0.0 %)	0(0.0 %)	1 (15.3 %)
Node					0.081
Yes	9(29 %)	9 (40.9 %)	2 (9.5 %)	4 (18.2 %)
No	22(71.0 %)	12 (54.5 %)	19 (90.5 %)	18 (81.8 %)
TMB					0.024
TMB-H	0(0.0 %)	11(50.0 %)	2 (9.5 %)	3 (13.6 %)
TMB-L	4(12.9 %)	9 (40.9 %)	11 (52.4 %)	14 (63.6 %)
TP53 status					0.0004
WT	20(35.5 %)	18 (81.8 %)	5 (23.8 %)	7 (31.8 %)
Mut	41 (64.5 %)	4 (18.2 %)	16 (76.2 %)	15 (68.2 %)

Rank-based enrichment analysis using the fold change of protein abundance revealed the overall perturbations of biology pathway. The results indicated downregulated protein signatures chiefly involving RNA-associated function and metabolic process. In addition, the dominant signatures of molecular transport, receptor tyrosine kinase (RTK) signaling, stress response, immune response, cell survival and cell adhesion were primarily enriched in KRAS G12-mutant LUAD (Fig. 3C–D and Supplementary Table 8). Distinctive proteins in these relevant pathways, such as KRAS-associated transcription factor NKX2-1 [20], kinase CDK1 [21,22] and transporter SLC7A5 [23], were well-correlated with the clinical outcomes (Fig. 3E).

The deep proteomics profile also provided a resource for stratifying LUAD communities with KRAS G12MUT. By applying stringent cutoffs for differential quantification, prognosis assessment and classification (details in the Supplementary Fig. 4C), we nominated 31 candidate biomarkers. The features were subjected to lasso regression to construct an 11-protein panel. This multi-marker panel demonstrated superior performance in terms of the area under the curve (AUC) when distinguishing KRAS mutation status, compared to any single candidate marker, including the pathological diagnostic marker NKX2-1 for LUAD (Fig. 3F and Supplementary Fig. 4D–F).

To assess the generalizability and robustness of the 11-protein panel, we performed transcriptomic-level validation using an external cohort (Gilette MA et al., 2020) [24], which included 106 LUAD samples with KRAS WT or G12MUT. The panel demonstrated an AUC of 0.78 for predicting G12MUT tumors in the validation dataset (Supplementary Fig. 4G). Notably, both RNA and protein levels of the top two key proteins (SCD5, CYP4X1) were significantly higher in KRAS WT compared to G12MUT patients (Supplementary Fig. 4H–I), and these elevated levels were associated with favorable overall survival (Supplementary Fig. 4J–K).

Compared to KRAS or TP53 mutations alone, their concurrence was clearly associated with reduced disease-free survival of LUAD patients (Fig. 3G). Moreover, our data provides evidence for protein-level trans effects and mechanistic hypotheses of genetic co-mutation. These include perturbations of cell cycle progression, genome integrity and fatty acid metabolism, as well as synergistic trans-effects on the expression of specific functional proteins (Fig. 3H and Supplementary Fig. 4L–M).

The characteristic proteins in patients with KRAS G12 mutations

We next intended to identify the distinctive proteins associated with KRAS G12 mutations. Respectively, 334, 220, 160 and 85 proteins were identified to have differential expressions in the WT, G12C, G12D and G12V groups when compared with the remaining ones (Supplementary Fig. 5A and Supplementary Table 9). After removing overlapping differentially expressed proteins, we identified a total of 609 characteristic proteins, comprising 260 in the WT group, 154 in the G12C group, 138 in the G12D group, and 57 in the G12V group (Supplementary Fig. 5B). Patients with KRAS G12C mutation exhibited distinct characteristics compared to the other groups (Fig. 4A). Most of them were older male smokers, with a higher prevalence of TMB-L and regional lymph node metastasis (Fig. 4A). In the WT group, the distinctive proteins were predominantly enriched in metabolism-related functions. Meanwhile, in the G12C group, immune-related pathways were remarkably activated (Fig. 4B). Based on the G12C immune-related protein–protein interactions (PPIs) network, TYROBP was revealed as the highly linked hub protein while MMP12 was the most representative protein (Fig. 4C). Furthermore, CRYM and GLS involved in amino acid catabolic process were expressed higher in the WT group, and were associated with better prognosis. Increased expression of CTSD, CPS1 and NOP2 were respectively found in G12C, G12D and G12V groups. Patients with high expression of these proteins were likely with worse DFS (Fig. 4D).

Fig. 4.

The characteristic proteins in patients with KRAS G12 mutations. A. Heatmap showing scaled expression profiling of 609 differentially expressed proteins across the KRAS wild-type and mutant groups. Clinical features are annotated at the top. B. Heatmap of significantly enriched Gene Ontology (GO) biological processes identified by Gene Set Enrichment Analysis (GSEA). C. Interaction network of immune-related proteins upregulated in KRAS G12C group. Nodes size and color represent log₂(fold change). D. Kaplan-Meier survival curves for representative proteins in specific biological pathways, showing their association with clinical prognosis. E. Heatmap depicting the expression profile of characteristic proteins across KRAS wild-type and mutant groups. Left-side annotations show log₂-transformed hazard ratio (HR) and AUC, representing the prognostic significance and stratification capacity for patients with KRAS G12 mutations. Druggability and clinical actionability based on the Drug Gene Interaction Database (http://www.dgidb.org/) are also indicated. The druggability score reflects the sum of PubMed journal articles supporting the drug-gene relationship. F. The hazard rations of characteristic proteins in the characteristic group and non-characteristic group, with solid dots indicating proteins that are significantly associated with prognosis only in the characteristic group. G. The expression profiles of the characteristic proteins were validated using proteomics data from paraffin-embedded tissues, cellular proteomics, and external proteomics datasets.

To further uncover the prognostic proteins associated with KRAS mutation, using the aforementioned 609 characteristic proteins as a basis, we identified the prognostic proteins that possess favorable classification capabilities. The up-regulated proteins in the WT group exhibited a protective impact, and the down-regulated proteins were associated with increased risk (Fig. 4E). Combined with the druggability and clinical actionability [25] of these proteins, we found that the prognostic proteins in the G12C group demonstrated more uniformity and higher druggability (Fig. 4E). The risk ratios of nine characteristic proteins, mainly distinctive in the G12D group, were significantly affected in the characteristic group rather than other groups (Fig. 4F and Supplementary Fig. 5C).

Stable A549 cell lines were constructed, KRAS knockout strains and cell strains with KRAS WT, G12C, G12D and G12V mutations complemented on the knockout background (Supplementary Fig. 5D–E). CCK-8 assays demonstrated that KRAS knockout significantly suppressed cell viability, whereas cell proliferation was markedly enhanced in the KRAS mutation groups (Supplementary Fig. 5D–E). In parallel, external mass spectrometry data were used for validation, and the results were largely consistent with the clinical proteomic data. For example, CAMK1D was highly expressed in the wild-type group, while ISG20 exhibited low expression in the wild-type group but was highly expressed in the G12C and G12D groups. Additionally, PLAUR expression was higher in the G12C and G12V groups, compared to the wild-type group (Fig. 4G).

Proteomic subtypes of Chinese LUAD with KRAS G12 mutation

To investigate the intra-tumor heterogeneity and intrinsic structure of proteomics data, a comprehensive protein-driven clustering analysis was performed. The 96 LUAD samples were classified into three subtypes, designated as S1, S2 and S3, comprising 48, 34 and 14 cases, respectively (Fig. 5A). Notably, the proteomic subtypes were found to significantly distinguish clinical features and prognosis outcome (Fig. 5B–C). S2 was enriched in a cohort of older individuals, predominantly male and smokers, and exhibited the highest recurrence rate with a median disease-free survival of less than two years, representing the most aggressive subtype. The multivariate Cox regression analysis demonstrated that, even after adjusting for multiple prognostic factors, including stage, age, sex, and smoking history, the S2 subtype was still significantly associated with poorer disease-free survival compared to the S1 subtype (Supplementary Fig. 6A). In contrast, S1 showed the highest proportion of stage I class and the lowest rate of disease progression. Upon separately decomposing the mutation spectra of the three subtypes, it was found that the core samples in S1 manifested wild-type KRAS or TP53 status. Conversely, all cases in S3 harbored KRAS mutations, primarily G12D and G12V, while 84.6 % retained wild-type TP53. The majority of KRAS G12C cases (n = 16, 72.7 %) were present in the S2 subtype. Additionally, 61.0 % of TP53-mutant tumors were classified into S2 through unsupervised clustering. As expected, S2 also exhibited significantly elevated TMB.

Fig. 5.

Proteomic subtypes of Chinese LUAD with KRAS G12 mutations. A. Heatmap showing scaled expression profiling of differentially upregulated proteins across the three proteomic subtypes. Clinical features are annotated at the top. B. Pie charts display the distribution of clinical metadata across samples. Labels on the left denote the categories of metadata. Dark green and light green used in pie charts represent samples that meet or do not meet the respective attributes, respectively. A chi-square test was conducted to assess the sample distribution across the three subtypes, with the corresponding p-value annotated on the right side. Pie charts outlined with a black border highlight attributes that are significantly overrepresented in a specific subtype. Unavailable data were excluded from the statistical analysis. C. Disease-free survival curves for patients with three proteomic subtypes. Log-rank test and univariate Cox regression analysis. D. Pathways enriched in the three subtypes, with green representing S1, orange for S2, and blue for S3. E. Comparison of activation score for cell cycle progression (CCP) and hypoxia status among the proteomic subtypes. F. Comparison of enrichment scores for immune cells and markers across the proteomic subtypes. G-K. Boxplot comparing protein abundance across the LUAD proteomic subtypes. PLOD2 (G), CTSL (H), MSR1 (I), MMP12 (J), TREM1 (K).

We next sought to elucidate the underlying biology to address unmet clinical needs. A total of 752 proteins exhibited subtype-specific upregulation: 342 proteins in S1, 233 proteins in S2, and 177 proteins in S3 (Supplementary Table 10A). Enrichment analysis revealed that S2, the clinically high-risk category, was distinguished from S1 and S3 by deregulation of multiple immune-related and oncogenic signaling (Fig. 5D). Rapid cell cycle progression (CCP) and localized hypoxia were significantly enriched exclusively in the S2 (Fig. 5E). Although comparable levels of eosinophils, neutrophils, B cells, NK cells and macrophages were observed across the three subtypes, it is noteworthy that the balance between Th1 and Th2 responses was disrupted, with the presence of stronger STAT1 activity and elevated cytokine abundances like IFN-γ, TGFβ, IL2, IL4 and IL12 observed in S2 (Fig. 5F). S1, on the other hand, showed deregulation of lipid metabolism and adhesion molecules that might provide a mechanical barrier against immune cell infiltration (Fig. 5D) [26,27]. S3 tumors exhibited disorder of hormone metabolism (e.g. estrogen and corticosteroid) and prominent protein glycosylation (Fig. 5D). These findings suggest that S2 patients are more likely to benefit from immunotherapy, thereby providing a foundation for tailored therapeutic strategies.

To further explore the clinical relevance, we examined the relationship between subtype-specific protein signatures and immune checkpoint blockade (ICB) therapy response. We identified 860 and 472 differentially expressed protein-coding genes between the Response and Non-Response groups for anti-PD1 (SRP183455, GSE136961) and anti-PDL1(SRP217040) ICB therapy in non-small cell lung carcinoma (NSCLC) (|fold change|>2, FDR < 0.05, Supplementary Table 10B).[28]. The results indicated that protein signatures in S2 subtype were more strongly associated with response to both ICB therapies. Specifically, proteins associated with anti-PD1 therapy response were identified in 8, 48, and 1 cases in S1, S2, and S3, respectively. While anti-PDL1 therapy-responsive proteins were exclusively found in S2, with 8 proteins identified (Supplementary Fig. 6B–D and Supplementary Table 10C). Furthermore, the upregulated markers in S1 and S3 exhibited a stronger preference for the non-response group, whether for anti-PD1 (S1: n = 18; S2: n = 2; S3: n = 6) or anti-PDL1 (S1: n = 12; S2: n = 4; S3: n = 8) (Supplementary Fig. 6B–D and Supplementary Table 10B). Among these, PLOD2, CTSL, MSR1, MMP12 and TREM1 were shared by both anti-PD1 and anti-PDL1 responders and were uniquely upregulated in S2 (Fig. 5G–K). A combined expression panel of these five proteins demonstrated high diagnostic accuracy for distinguishing S2 tumors (AUC = 0.96, Supplementary Fig. 6E). These findings highlight the potential of these five proteins as biomarkers for patient stratification and prediction of immunotherapy response.

The tumor immune microenvironment landscape of LUAD patients

To investigate the impact of KRAS G12 mutations on TIME, we performed multiplex immunohistochemistry (mIHC) on 54 LUAD samples, including 15 wt, 12 G12C, 11 G12D and 16 G12V cases (Supplementary Fig. 7A and Supplementary Table 11). Representative images were presented to show the immune cell infiltration in LUAD tissues (Fig. 6A). Distinct immune cell types displayed significant differences in infiltration level depending on KRAS mutation status, molecular subtypes, and co-mutation with TP53, both general (KRAS&TP53.1) and G12C-specific (KRAS&TP53.2) co-mutations (Fig. 6B). Specially, macrophage-associated markers (CD68+, CD68+CD163+, and CD68+CD163-) demonstrated significant variations across all groups. Additionally, CD8+PD1+ T cells were significantly associated with KRAS variants and molecular subtypes (Fig. 6B).

Fig. 6.

The immune profiling of LUAD with KRAS G12 mutations. A. Multiplex immunohistochemical (mIHC) was conducted to assess the tumor immune microenvironment (TIME) in LUAD tissues with two panels of immune markers. B. Impacts of KRAS statuses, proteomics subtypes, KRAS/TP53 co-mutation statuses, as well as the co-mutation statuses of KRAS G12C/TP53 on immune cell density. Kruskal-Wallis test was used to assess significant differences among groups. C-F. The infiltration densities of immune cell markers that indicate significant overall differences among KRAS mutation statuses (C), proteomics subtypes (D), KRAS/TP53 co-mutation status (E), and KRAS G12C/TP53 co-mutation status (F). The y-axis represents log₂(density + 1), with the unit of measurement as cells/cm² G-I. The network diagram depicts the correlations between the abundance of prognostic characteristic proteins and immune cell densities in: overall cohort (G), KRAS WT group (H), and KRAS G12C group (I). Nodes represent biomarkers (colored by category) and immune cells. Edges represent significant correlations (red: positive, blue: negative; p < 0.05). Node size reflects connectivity degree.

In patients harboring KRAS G12C mutation, CD8+, PD1+ and CD68+ cells were more frequently found in intratumor areas, compared to other groups (Fig. 6C). Among the molecular subtypes, S2 exhibited the highest levels of immune cell infiltration, likely due to its greater proportion of G12C mutations (Fig. 6D). Independently considering the TP53 mutation status, no significant differences in immune cell infiltration were observed between the mutated and wild-type (WT) groups, suggesting that TP53 alone does not substantially impact the immune microenvironment (Supplementary Fig. 7B). However, within the groups stratified by diverse KRAS and TP53 co-mutation states, both macrophages and CD8+ T cells manifested substantial variations in infiltration level. The peak infiltration level was detected in the group with concurrent KRAS and TP53 mutations, whereas the nadir was observed in the group with only TP53 mutations (Fig. 6E). Among all mutations, patients with KRAS G12C exhibited the highest immune cell infiltration levels (Fig. 6F), while those with G12D or G12V mutations exhibited peak infiltration only in the presence of KRAS and TP53 co-mutations (Supplementary Fig. 7C–D).

Correlation network analyses between prognostic proteins and immune cell infiltration revealed the strongest correlation with CD8+PD1+ T cells and the weakest with CD19+ B cells (Fig. 6G). Further comparison between KRAS WT and G12C mutation groups disclosed a more intricate and abundant network in the WT group (Fig. 6H-I). Specifically, ISG20 in the WT group was significantly positively correlated with CD19+ B cells and CD8+ T cells, whereas in the G12C group, CTSD showed a notably correlation with CD4+ T cells and PD1+ T cells (Supplementary Fig. 7E). Collectively, our data highlight the intricate relationship between the proteomic landscape and the immune microenvironment in LUAD.

Interactome of G12-mutant KRAS protein

Protein interplay represents another compelling structural framework for understanding KRAS. We developed a proteomic workflow based on rapid proximity labelling, known as Pyrococcus horikoshii biotin protein ligase (phBPL)-assisted biotin identification (PhastID) [29], to identify the interactome of the mutated KRAS proteins, including direct clients and bystanders [30]. The site-directed mutagenesis of plasmids with phBPL were genomically integrated into HEK293 cells, with sustained expression at a comparable level (Supplementary Fig. 8A–B). Subcellular compartment localization and functional phenotyping demonstrated that the expression of the fusion phBPL did not affect the original localization and carcinogenic properties of the KRAS proteins (Supplementary Fig. 8C–D). The proximal proteome of KRAS were then generated from three replicates of each variant. In total, we uncovered 245–436 proximity interactions per bait, representing 528 non-redundant preys (Supplementary Fig. 8E and Supplementary Table 12). The numbers and overlap of identified interactions varied slightly across variants, with a notable decrease in G12V mutant (Fig. 7A and Supplementary Table 13). Approximately half of the interactions (44.3 %–45.7 %) were previously documented [31], providing evidence for the reliability of our data. Correlation analyses disclosed a noticeable interactome in G12V group, compared to other counterparts (Fig. 7B).

Fig. 7.

Profiling the proximal proteome of KRAS. A. Bar plot showing the number of KRAS preys identified in the proximal proteome. Known interactions are indicated in dark green, while novel interactions are shown in light green. B. Heatmap showing the qualitative similarity of KRAS preys across different KRAS mutations, using the Jaccard index. C. Heatmap of uniquely enriched binding partners across various KRAS mutation groups. FOT (fraction of total) represents the normalized abundance of each enriched protein in the corresponding experiment. D. Dot plot illustrating pathways significantly enriched among the gain of interactions specific to KRAS status. E. Interaction network of ion transport proteins, with peripheral pie charts indicating the bait proteins. White nodes represent proteins identified in LUAD proteomics, while grey nodes correspond to those not identified. The heatmap in the upper-right corner displays the log₂-transformed average fold change of preys in both the proximal proteome and LUAD proteome, comparing KRAS-mutant to wild-type samples. F. Scatter plot showing the relationship between the gain of interaction in KRAS G12C variant (y axis) and gene essentiality in KRAS G12C-mutant LUAD cell lines (x axis) based on dependency screening database. G. Bar plot of enrichment significance for glycolysis-associated pathways. H. Correlation between SLC4A7 protein expression levels and lactate abundance across 37 LUAD cell lines from the CCLE database. I. Proposed model depicting intra- and extracellular pH imbalances mediated by KRAS/SLC4A7 axis. J. Disease-free survival of LUAD patients with high versus low SLC4A7 expression. K-L. Comparison of infiltration density for CD8+PD1+ T cells (K) and CD68+CD163+ M2 macrophages (L) in relation to protein levels of SLC4A7 (left), SLC16A3 (middle), and combined SLC4A7/SLC16A3 (right). The y-axis represents log₂(density+1), with the unit of measurement as cells/cm². M. Disease-free survival of LUAD patients with high versus low SLC16A3 expression. N. Disease-free survival of LUAD patients with different combined SLC4A7/SLC16A3 levels.

Allele-enriched partners were examined to gain insight into allele-driven oncogenic programs (Fig. 7C and Supplement Table 14). Over-representation analysis suggested shared oncogenic pathways across different KRAS variants (Fig. 7D). For G12C-specific interactions, we found that a set of proteins interacting with KRAS G12C were ion transporters, such as SLC12A2 and SLC4A7 (Fig. 7E). To nominate functionally essential proteins within the G12C-specific proximal proteome, we compared labeling outliers relative to KRAS wild- type variant (gain-of-interaction) and a CRISPR-mediated loss-of-function screen from the Cancer Dependency Map Project (DepMap) [32]. Notably, the gain-of-interaction correlated with gene essentiality profiles for four proteins, including mTOR and SLC4A7, in KRAS G12C-mutant LUAD cells (Fig. 7F and Supplementary Fig. 8F). The activation of mTOR signaling, a central regulator of glycolysis, was driven by KRAS mutations has been well documented [33]. Consistently, glycolytic activity was significantly elevated in KRAS G12C-mutant cells (Fig. 7G). SLC4A7, a critical ion transporter involved in nutrient and metabolic product transport, plays a key role in maintaining intracellular pH homeostasis [34]. Co-immunoprecipitation (Co-IP) assay confirmed a direct interaction between SLC4A7 and the KRAS G12C variant (Supplementary Fig. 8G). Both proteomics and Immunohistochemistry (IHC) analyses demonstrated elevated SLC4A7 expression in KRAS G12C-mutant tumors (Supplementary Fig. 8H–J). Based on these findings, we propose that SLC4A7 contributes to metabolic reprogramming in KRAS G12C-mutant LUAD, potentially linking its activity to enhanced glycolytic output.

The lactate overproduction in tumor cells, driven by hyperactivated glycolysis, necessitates efficient transporter-mediated efflux to maintain intracellular pH homeostasis. We explored the correlation between SLC4A7 expression and lactate abundance across 37 LUAD cell lines from Cancer Cell Line Encyclopedia (CCLE) (Fig. 7H) [35]. A significant positive correlation was observed. Moreover, SLC4A7 knockdown resulted in intracellular lactate accumulation in both KRAS wild-type and G12C-mutant A549 cell lines (Supplementary Fig. 8K–L). Previous studies in LUAD have suggested that SLC4A7 regulates lactate efflux through modulation of SLC16A3 [36]. In our dataset, SLC16A3 was significantly upregulated in KRAS G12C-mutant samples (Supplementary Fig. 8 M−N), and IHC data revealed a strong positive correlation between SLC4A7 and SLC16A3 expression in LUAD tissues (Supplementary Fig. 8O). These results suggest that SLC4A7 functions as the regulator of intracellular lactate efflux and contributes to the acidic tumor microenvironment in KRAS G12C-mutant LUAD.

Elevated glycolysis activity, coupled with increased SLC4A7 expression, led to intracellular alkalization and extracellular acidification, thereby shaping acidic TIME (Fig. 7I). The mIHC analysis reflect remarkable disparity in infiltration density of CD8+PD1+ T cells and CD68+CD163+ M2 macrophages in the LUAD tissues in response to SLC4A7 protein level, as a surrogate for TIME acidity (Fig. 7K). Through analysis of non-small cell lung carcinoma (NSCLC) cohorts (SRP183455, GSE126044) [37], we consistently identified significantly higher SLC4A7expression in anti-PD1 therapy responders compared to non-responders (Supplementary Fig. 8P). Further investigations revealed that pharmacological inhibition of SLC4A7 with S0859 (Medchemexpress,45692) (20 μM) induces deregulation of immune-related genes and cytokines (Supplementary Fig. 8Q) [38]. High SLC4A7 expression was associated with poorer patient outcomes (Fig. 7J). In contrast, SLC16A3 expression alone did not correlate with prognosis nor immune cell infiltration (Fig. 7L–M). However, the combined upregulation of SLC4A7 and SLC16A3 was associated with significantly worse prognosis (Fig. 7N), suggesting a synergistic effect in promoting an immunosuppressive and metabolically favorable tumor microenvironment.

Discussion

In-depth and quantitative proteomics investigation, integrated with genomic-driven biological understanding, provide unique insights with potential clinical relevance. Comprehensive integrated proteogenomic data of lung cancer has been documented in European and Chinese populations. However, significant clinical challenges remain unresolved in NSCLC patients harboring KRAS G12 mutations. In this study, we comprehensively interpreted the genomic features of 9,479 patients with solid tumors and further illustrated the mutation landscape of 3,523 patients with lung cancer. We also depicted proteomic profiles from 96 LUAD patients with KRAS G12 mutations, revealing distinct molecular features and tumor progression hallmarks. Molecular subtypes driven by KRAS mutation-induced proteomic alterations were identified and linked to the immune profiling, uncovering a specific subtype with dysregulated immune responses, which showed the highest potential benefit from immunotherapy. Furthermore, we profiled TIME of KRAS G12-mutated LUAD and mapped the PPIs landscape of these variants in live cells. Collectively, our study provides a multilayered molecular perspective of LUAD in the context of KRAS G12 mutations, highlighting new avenues for biomarker discovery and therapeutic strategy.

Genomic landscape of nearly 10,000 cancer cases revealed that the KRAS mutations were most prevalent in pancreatic ductal adenocarcinoma, followed by colorectal cancer and lung cancer, consistent with data from multiple large tumor genome repositories [8,39]. In pan-cancer cases, the G12D mutation was the most frequently occurring KRAS mutation, while in lung cancer, the G12C mutation was the most common. Furthermore, KRAS mutation often occurred concurrently with mutations in tumor suppressor genes such as TP53, LRP1B, and STK11, in line with the previous study in patients with KRAS-mutant advanced NSCLC [40] The co-occurrence of KRAS and TP53 mutations is associated with poor prognosis and potential benefits from immunotherapy [16], [41]. Our proteogenomic data revealed that the coexistence of KRAS and TP53 mutations leads to significant perturbations in cell cycle progression and increase in somatic mutations. These alterations suggest a synergistic interaction between oncogenic KRAS and the loss of TP53, driving uncontrolled proliferation and genomic instability. Furthermore, we observed the deregulation of lipid metabolism pathways in KRAS/TP53 co-mutant tumors, which influence immune cell recruitment [42]. Previous studies have demonstrated that KRAS/TP53 co-mutation in NSCLC significantly affect the immune microenvironment and response to immune checkpoint blockade, particularly in G12C mutant patients [41], [43]. Therefore, KRAS-mutant lung cancer represents a highly heterogeneous disease, and co-occurring mutations such as TP53 may influence therapeutic strategies and clinical outcomes.

Our proteomics data elucidated the molecular attributes associated with KRAS G12-mutant LUAD. In comparison with the WT group, the G12MUT group exhibited substantial alterations in several key biological pathways, particularly those related to cell cycle regulation, immune response, molecular transport, and metabolism, highlighting the oncogenic phenotype driven by the KRAS G12 mutation. Notably, LUAD harboring G12C mutation exhibits significant alterations in immune-related pathways, including T cell-mediated immune responses and interferon signaling. KRAS G12C mutations are also associated with high TMB and CD8+PD1+ T cells infiltration, both favorable indicators for immunotherapy response. Our current exploration has revealed that in genetically edited KRAS G12C-mutant lung cancer cell lines, there is a significantly higher transcriptional expression of immune-related genes and cytokines (e.g., PD-L1, IL-10, TGF-β, TNF-α, IL-1β, CCXL-9) compared to the wild-type counterpart (Supplementary Fig.9). These findings suggest the potential immunomodulatory effects of G12C mutation, offering the opportunities to expand therapeutic options for KRAS G12C-mutant LUAD.

Clinically, there is a growing demand for precision treatment strategies targeting other KRAS G12 mutations. MRTX1133, a highly selective and potent KRAS G12D inhibitor, has being presently in clinical trials [44]. Recent advancements have identified highly specific T cell receptors (TCRs) that selectively recognize KRAS G12D-derived neoantigens, enabling TCR-engineered adoptive T cell therapy as a promising immunotherapeutic approach [45]. Additionally, another KRAS G12D-specific inhibitor, HRS-4642, has demonstrated potent preclinical and clinical anti-tumor efficacy, along with a promising combinatorial strategy with carfilzomib to enhance sensitization [46]. Collectively, due to their high frequency and oncogenic potential, KRAS G12 mutations have emerged as critical clinical therapeutic targets. Accumulating omics data further highlight the urgent need for diversified therapeutic strategies to effectively target these specific mutations.

Our study identified three distinct LUAD subtypes associated with KRAS G12 mutations, revealing significant differences in molecular characteristics, clinical features, and prognosis. The Immune-Modulation subtype (S2) is characterized by aggressive clinical features, including a high recurrence rate, the shortest median DFS, and a higher proportion of smokers and older individuals. Notably, patients in the S2 subtype exhibit pronounced immune regulation disruptions, with upregulated cytokine profiles (e.g., IFN-γ, TGFβ, IL-2, IL-4, and IL-12), along with enhanced STAT1 activity. The mIHC assay revealed high infiltration of CD8 + PD1 + T cells and macrophages in the S2 subtype. Additionally, S2 is strongly associated with KRAS G12C mutations, TP53 mutations, and high TMB, suggesting the potential clinical utility of ICB for this subtype. In contrast, the Metabolism subtype (S1) is marked by the deregulation of lipid metabolism. This subtype is predominantly enriched in early-stage disease and shows the lowest rate of disease progression. S1 samples also predominantly harbor wild-type KRAS or TP53. The Hormone-Disorder subtype (S3) exhibits significant deregulation of hormone metabolism, including estrogen and corticosteroids. S3 tumors are predominantly characterized by KRAS mutations, specifically G12D and G12V.

The landscape of KRAS G12 mutation-driven gain-of-interactions, as revealed by the PhastID proximity labeling approach, identifies SLC4A7 as a previously underappreciated interactor in KRAS G12C-mutant LUAD. SLC4A7 is significantly upregulated in KRAS G12C-mutant LUAD and directly interacts with mutant KRAS, acting as a pivotal factor in intracellular lactate regulation and the shaping of an immunosuppressive TIME. Previous reports also showing elevated SLC4A7 expression in RAS-mutant PDAC [47]. Oncogenic KRAS activation is known to stimulate mTOR signaling and promote aerobic glycolysis [48]. Notably, such metabolic reprogramming leads to excessive lactate production and acidification of the TIME [49]. Elevated lactate concentrations and low pH environments have been shown to compromise the functional T cells and NK cells, while facilitating the expansion of immunosuppressive myeloid cells and Treg cells [50], [51]. Moreover, extracellular acidification has been reported to diminish the therapeutic efficacy of ICB [52]. Our findings demonstrate that SLC4A7 reduces intracellular lactate accumulation, thereby alleviating cellular metabolic acid stress. Supporting this, previous investigation in LUAD has indicated that SLC4A7 modulates lactate flux in part by downregulating the monocarboxylate transporter SLC16A3 [36]. This aligns with our observation of a significant positive correlation between SLC4A7 and SLC16A3 expression levels in LUAD tissues.

Integrated proteomic and immune profiling further reveals that elevated SLC4A7 expression-either alone or in combination with high SLC16A3-correlates with increased infiltration of CD8+PD1+ T cells and CD68+CD163+ macrophages, both indicative of an immunosuppressive TIME. In PDAC, SLC4A4, another member of the SLC4 family, has similarly been implicated in tumor acidification [53]. Pharmacological or genetic inhibition of SLC4A4 in PDAC mouse models reduces lactate production and promotes extracellular bicarbonate accumulation, leading to enhanced T cell-mediated immunity and suppression of macrophage-driven immune tolerance [53]. Collectively, these findings point to SLC4A7 as a key mediator linking KRAS G12C-driven metabolic reprogramming with immune suppression, via lactate homeostasis and pH balance. However, further functional exploration and clinical validation are needed to confirm these findings and translate them into clinically actionable strategies.

This study has several limitations. First, the proteomics data lack validation in another large-scale cohort. Second, although the proteomic data of 96 patients with LUAD were fully interpreted, in vitro experiments are required to demonstrate the essential findings for molecular classification of KRAS G12-mutant lung cancer. Third, tumor microenvironment composes many types of non-tumor cells. More immune cells, as well as fibroblast cells, should be included in the determination of immune landscape in KRAS G12-mutant tissues. Fourth, the absence of direct assessment of therapeutic responses to KRAS inhibitors or immunotherapy presents a significant gap, as this is critical for informing treatment decisions. In conclusion, the findings of this study provide a deep understanding of the molecular characteristics of KRAS G12 mutant-LUAD.

Data availability Statement

The proteomics data of LUAD samples have been deposited to the ProteomeXchange Consortium (https://pdc.cancer.gov/pdc/) under the accession number PXD058891. Access credential is as follows: Username: reviewer_pxd058891@ebi.ac.uk, Password: xs5fSJZ7Qk7E.

Additionally, the following public repositories have been used in this study: ICBatlas (https://guolab.wchscu.cn/ICBatlas#!/) provides differential expression resources for NSCLC patients undergoing anti-PD1 (SRP183455, GSE136961) and anti-PDL1 (SRP217040) immune checkpoint blockade (ICB) therapy. IID (https://iid.ophid.utoronto.ca/) provides human context-specific protein-protein interaction database. DGIdb (https://dgidb.org/) provides druggability information. DepMap (https://depmap.org/portal/) offers data from large-scale loss-of-function screen using CRISPRi (CRISPRGeneDependency, 24Q2). CCLE (https://sites.broadinstitute.org/ccle) provides cell metabolomics dataset.

The data that support the findings of this study are available under restricted access. Access can be obtained by contacting zhangzy@jnu.edu.cn. A signed data access agreement is required before data sharing.

Compliance with Ethics Requirements

All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1975, as revised in 2008 (5). Informed consent was obtained from all patients for being included in the study.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

The following are the Supplementary data to this article: