KRAS^G12/13 mutation modulates CRC outcomes via disrupting positive feedback between macrophage and CD4⁺ T cell

Summary

Kirsten rat sarcoma viral oncogene (KRAS) mutation influences colorectal cancer (CRC) progression, but their specific impact on the tumor immune microenvironment remains poorly defined. This study establishes that KRAS^G12/13 mutation disrupts critical positive feedback between macrophages and CD4⁺ T cells. Mechanistically, KRAS^G12/13-mutant tumor cells secrete elevated interleukin-10 (IL-10), which suppresses macrophage production of chemokine ligand 9/10 (CXCL9/10) and major histocompatibility complex II (MHC-II). This impairs the infiltration and activation of CXCR3⁺CD4⁺ T cells, fostering an immunosuppressive niche. By integrating multi-omics data from clinical cohorts and validating findings in vivo, we show that combining KRAS inhibitors with CXCL9/10 restoration effectively overcomes this immune suppression and controls tumor growth. Our work delineates a targetable immune evasion mechanism and provides a cohesive prognostic and therapeutic framework for KRAS^G12/13-mutant CRC.

Graphical abstract

Highlights

• •KRAS^G12/13 mutation plays an important role in CRC prognosis
• •KRAS^G12/13 mutation downregulates CXCL9/10 and MHC-II in macrophages
• •KRAS^G12/13 mutation disrupts positive feedback between macrophages and CD4⁺ T cells
• •Combined KRAS inhibition and CXCL9/10 restoration suppress CRC growth

Health sciences

Introduction

CRC (colorectal cancer) is the third most common cancer worldwide according to cancer statistics in 2022¹ and accounts for approximately 10% of all annually diagnosed cancers and cancer-related deaths.² Many clinical features of CRC, including age, gender, and TNM stage, are tightly associated with prognosis of patients.² TNM stage is still the gold standard for evaluating various tumor types, including CRC, and remains the most used method in clinical practice. However, extensive clinical data have shown that TNM stage has limitations in prognostic prediction, particularly as KRAS (Kirsten rat sarcoma viral oncogene‌) mutation can drive malignant progression and therapy resistance even in early-stage tumors.³ Therefore, incorporating KRAS mutation into prognostic analysis is crucial, as it may enable the development of a more efficient prognostic model for CRC and help identify potential therapeutic targets.

The infiltration of immune cells is crucial in TME (tumor microenvironment).⁴ Macrophages, which are abundant in tumors, express chemokines such as CXCL9 and CXCL10 (chemokine ligand 9/10), attracting CXCR3⁺ T cells (chemokine receptor) to the tumor site.^5,6 Besides, they can also recognize and engulf tumor cells, thus presenting tumor antigens to MHC-II (major histocompatibility complex II) to activate CD4⁺ T cells via TCR (T cell receptor) binding.⁷ Conversely, activated CD4⁺ T cells secrete IFN-γ (interferon gamma),⁸ activating the JAK-1/STAT-1 (Janus kinase/signal transducer and activator of transcription) pathway in macrophages to upregulate CXCL9/10 and MHC-II expression.⁹ This establishes positive feedback that sustains an immunologically active TME and mediates anti-tumor immunity. However, previous research indicates that KRAS^G12D mutation in pancreatic cancer drives tumor cells to hypersecrete IL-10 (interleukin-10), a key immunosuppressive cytokine that directly inhibits the capacity of CD4⁺ T cells to secrete IFN-γ.¹⁰ These findings lead to the hypothesis that KRAS mutation similarly disrupts this positive feedback in CRC via IL-10 hypersecretion, ultimately driving tumor progression, which has not yet been formally investigated.

Thus, we analyzed a large-scale CRC dataset from our center and an external dataset from cBioPortal to develop a more efficient model containing KRAS mutation to predict the prognosis of CRC. Moreover, by using transcriptomic data from external datasets and samples from our center, we may uncover a link between KRAS G12/13 mutation and the positive feedback between macrophages and CXCR3⁺CD4⁺ T cells in TME, which could offer more precise therapy in KRAS G12/13 mutation-related CRC.

Results

General description of the internal cohort

The internal cohort was randomly split into a discovery cohort (n = 1,276) and an internal validation cohort (n = 544) in a 7:3 ratio, with data from both cohorts being quite similar (all p values >0.05), as shown in Table 1.

Table 1.

Characteristics of the discovery and internal validation cohort

Characteristic	Discovery cohort (n = 1,276)	Internal validation cohort (n = 544)	p value
Age, mean ± SD(years)	66.08 ± 10.54	66.25 ± 11.22	0.758
Gender, n (%)			0.916
Male	790 (61.9)	339 (62.3)	–
Female	486 (38.1)	205 (37.7)	–
Primary location, n (%)			0.959
Colon	734 (57.5)	314 (57.7)	–
Rectum	542 (42.5)	230 (42.3)	–
T stage, n (%)			0.658
Tis & T1	41 (3.2)	12 (2.2)	–
T2	137 (10.7)	63 (11.6)	–
T3	106 (8.3)	44 (8.1)	–
T4	992 (77.8)	425 (78.1)	–
N stage, n (%)			0.469
N0	671 (52.6)	276 (50.7)	–
N1-x	605 (47.4)	268 (49.3)	–
M stage, n (%)			0.795
M0	1,198 (93.9)	509 (93.6)	–
M1	78 (6.1)	35 (6.4)	–
TNM stage, n (%)			0.927
Ⅰ stage	139 (10.9)	59 (10.8)	–
Ⅱ stage	520 (40.8)	213 (39.2)	–
Ⅲ stage	539 (42.2)	237 (43.6)	–
Ⅳ stage	78 (6.1)	35 (6.4)	–
MLH-1, n (%)			0.658
Negative	43 (3.4)	16 (2.9)	–
Positive	1,018 (79.8)	444 (81.6)	–
Missing data	215 (16.8)	84 (15.4)	–
MSH-2, n (%)			0.790
Negative	9 (0.7)	4 (0.7)	–
Positive	1,051 (82.4)	455 (83.6)	–
Missing data	216 (16.9)	85 (15.6)	–
MSH-6, n (%)			0.773
Negative	8 (0.6)	3 (0.6)	–
Positive	1,052 (82.4)	456 (83.8)	–
Missing data	216 (16.9)	85 (15.6)	–
MMR∗, n (%)			0.756
dMMR∗	45 (3.5)	19 (3.5)	–
pMMR∗	1,016 (79.6)	441 (81.1)	–
Inevaluable	215 (16.8)	84 (15.4)	–
TP53 mutation, n (%)			0.403
Negative	196 (15.4)	96 (17.6)	–
Positive	637 (49.9)	257 (47.2)	–
Undetected	443 (34.7)	191 (35.1)	–
KRAS mutation type, n (%)			0.115
Non-mutation	804 (63.0)	348 (64.0)	–
G12/13 mutation	420 (32.9)	163 (30.0)	–
Else mutation	52 (4.1)	33 (6.1)	–

Continuous variables were tested using a Student’s t test, while categorical variables were tested using a chi-square test. All statistical tests were two-tailed. MLH-1, MutL homolog 1; MSH-2, MutS homolog 2; MSH-6, MutS homolog 6; MMR∗, mismatch repair∗; dMMR∗, deficient mismatch repair∗; pMMR∗, proficient mismatch repair∗ (∗ represents adaption from the standard due to practical constraints in clinical test).

The average age in the discovery cohort was 66.08 ± 10.54, and in the internal validation cohort, it was 66.25 ± 11.22. The gender ratio in both groups was approximately 60% male to 40% female, and the distribution of tumor location was about 60% colon to 40% rectum. Regarding T stage, around 80% of patients were in T4 stage, indicating that most patients with CRC had tumors that had already grown to the serosal layer at the time of detection, which underscores the importance of early screening. Lymph node metastasis was about evenly split between non-metastatic and metastatic cases, with only around 6% of patients showing distant metastasis. According to the AJCC 8th edition, the stage distribution was roughly 1:4:4:1 for stage I to IV, suggesting that most patients were in stages II and III. In terms of pathology, among patients with available data, only 3% were MLH-1 (MutL homolog 1) negative, 0.7% were MSH-2 (MutS homolog 2) negative, and 0.6% were MSH-6 negative, resulting in just 3.5% of patients being dMMR∗ (deficient mismatch repair∗), while the majority (80%) were pMMR∗ (proficient mismatch repair∗). About 15% of patients had TP53 mutation, and around one-third had KRAS G12/13 mutation, which is consistent with the prior report.¹¹

KRAS mutation as an independent factor in model construction

We first applied Lasso regression for variable selection, using Log(λ) = −4 to determine the regularization strength. This process resulted in the retention of seven variables: age, gender, T stage, N stage, M stage, TNM stage, and KRAS mutation (Figure S1A). Subsequently, univariate Cox regression analysis was conducted to identify independent risk factors among these variables. Variables with a p value <0.05 were then included in a multivariate Cox regression analysis to evaluate their combined effect. Finally, five variables were included in the final multivariate analysis.

We then visualized the results using a forest plot to illustrate the hazard ratios and confidence intervals for each variable (Figure 1A). The results for age (hazard ratio [HR]: 1.024, 95% confidence interval [CI]: 1.016–1.033, p < 0.001), T stage (T2 vs. Tis & T1, HR: 4.626, 95% CI: 1.108–19.312, p = 0.036; T3 vs. Tis & T1, HR: 6.542, 95% CI: 1.573–27.214, p = 0.010; T4 vs. Tis & T1, HR: 9.583, 95% CI: 2.384–38.516, p = 0.001), N stage (N1-x vs. N0, HR: 2.357, 95% CI: 1.969–2.820, p < 0.001), and M stage (M1 vs. M0, HR: 2.298, 95% CI: 1.759–3.001, p < 0.001) were in line with common knowledge.

Figure 1.

Model construction, visualization, and validation for 5-year CRC outcome prediction

(A) Visualized results of multivariate Cox regression by a forest plot.

(B) Visualized predictive nomogram of discovery cohort.

(C) ROC curves of three cohorts used in machine learning.

(D) DCA curves of three cohorts used in machine learning.

(E) ROC curves of cohorts with three different therapies.

(F) DCA curves of cohorts with three different therapies.

While in KRAS mutation, G12/13 mutation has a higher HR (G12/13 mutation vs. non-mutation, HR: 1.231, 95% CI: 1.037–1.463, p = 0.018), while else mutation (else mutation vs. non-mutation, HR: 0.915, 95% CI: 0.583–1.438, p = 0.702) does not show statistical significance. However, given that else mutation comprised only 52 cases (4.07% of the discovery cohort), and considering the need for classification completeness and the minimal impact of the small sample size on overall results, it was still included in the nomogram construction.

Development, visualization, and validation of the predictive model containing KRAS mutation

Using the remaining five variables, we constructed a nomogram (Figure 1B). The formula is Score = (1.065 × age − 21.1) + T stage + N stage + M stage + KRAS mutation. In this equation, age refers to the patient’s age. The scoring for T stage is as follows: 0 for T1, 67.5 for T2, 83 for T3, and 100 for T4. For N stage, the scoring system is 0 for N0 and 38 for N1-x. The M stage scores are 0 for M0 and 37 for M1. As for KRAS mutation, the scores are 0 for any other mutation, 4 for non-mutation, and 13 for G12/13 mutation. Using this formula, we can calculate the score for each sample and determine the 5-year survival probability. Overall, the 5-year C-index for this nomogram is 0.690 (95% CI: 0.678–0.701), indicating a good performance for assessment.

Next, we applied the model to assign scores to each sample across the three cohorts. The relevant data of the external cohort were presented in Table S1. To investigate whether our model retains its predictive power across various treatments, the external dataset was stratified by therapeutic modality, which included 5-Fu (20.6%), oxaliplatin (24.5%), and bevacizumab (5.9%) (Table S1).

Subsequently, we evaluated each cohort using ROC (receiver operating characteristic) curves, DCA (decision curve analysis) curves, and calibration curves, with all time points set at 5 years (Figure 1). ROC and DCA curves for TNM stage were also plotted as a reference. In the discovery cohort, the model demonstrated a significantly higher AUC (area under curve) of 0.739, surpassing 0.687 for TNM stage. For the internal validation cohort, the model achieved 0.740, whereas TNM stage resulted in 0.676. In the external validation cohort, the model’s AUC reached 0.746, in contrast to 0.671 for TNM stage (Figure 1C). The DCA curves showed that in all three cohorts, the model’s net benefit was consistently higher than that of TNM stage (Figure 1D). The calibration curves for all three cohorts were near the diagonal, indicating that the model’s predicted probabilities closely matched the observed outcomes (Figure S1B). Overall, our model outperforms traditional TNM stage, making it a viable alternative.

Given the profound influence of therapies in prognosis, we further validated our model in treatment-specific subgroups (5-Fu, oxaliplatin, and bevacizumab) from the external cohort. Its predictive accuracy still surpassed TNM stage system in all scenarios, proving its independence of specific therapeutic interventions.

Worse prognosis in KRAS G12/13 mutation group

Given that KRAS mutation was a critical tumor driver mutation with significant relevance in our model, we categorized the study population into two groups based on the KRAS genotype: non-mutation and G12/13 mutation. Then, we analyzed the survival outcomes of patients in these two categories.

In the internal cohort, we found that the non-mutation group exhibited a significantly better overall survival probability than the G12/13 mutation group (p = 0.004 for the discovery cohort and p = 0.019 for the internal validation cohort). While in the external cohort, although there was no significant difference in overall survival probability between these two groups (p = 0.902) (Figure S1C), the 3-year survival data indicated that the mutation group had poorer prognostic outcomes than the non-mutation group (p = 0.033) (Figure 2A). Overall, patients with KRAS G12/13 mutation experienced lower 3-year survival probability compared to those without such mutation.

Figure 2.

Survival curves and analysis of differences between KRAS non-mutation and G12/13 mutation groups

(A) Kaplan-Meier survival curves for KRAS non-mutation and G12/13 mutation in three cohorts (overall survival for discovery cohort and internal validation cohort and 3-year survival for external validation cohort).

(B) Volcano plot for all genes with 9 genes associated with chemokine and antigen presentation pathways labeled.

(C and D) KEGG and BP analysis of differentially expressed genes (DEGs) between two groups.

(E) GSEA analysis between two groups.

(F and G) Visualized GSEA pathways associated with chemokine and antigen presentation pathways.

Significant downregulation of chemokine and antigen presentation in the KRAS G12/13 mutation group

Since our model highlighted the prognostic value of KRAS among several detected genomic features, and patients with KRAS G12/13 mutation experienced poorer outcomes in the first 3 years, we aimed to investigate the mechanisms contributing to this phenomenon.

We compared the transcriptome data between two groups, identified 570 differentially expressed genes, and visualized them in a volcano plot (Figure 2B). GO (Gene Ontology) and KEGG (Kyoto Encyclopedia of Genes and Genomes) enrichment analyses revealed that these genes were primarily associated with chemokines that induced cell chemotaxis and MHC-II antigen presentation pathways (Figures 2C, 2D, and S1D). We found that 2 of 8 (CXCL9 and CXCL10) related genes were correlated with T cell infiltration while rest 6 (HLA-DMB, HLA-DOA, HLA-DPA1, HLA-DPB2, HLA-DQA2, and HLA-DQB2) contributed to MHC-II presentation. All these genes showed a significant downregulation trend in the KRAS G12/13 mutation group (Figure 2B), indicating that patients with KRAS G12/13 mutation may have poorer prognosis due to reduced infiltration and activation of CD4⁺ T cells in the tumor. We then uploaded data to the GSEA (gene set enrichment analysis) dataset for thorough analysis and selected pathways containing these genes for bubble chart visualization (Figure 2E). Our analysis indicated that all these enriched pathways were downregulated (normalized enrichment score [NES] < 0) (Figures 2E–2G). Pathways associated with chemokines included GPCR ligand binding (Reactome), chemokine signaling pathway (KEGG), and chemokine receptors binding chemokines (Reactome), while those related to antigen presentation included TCR pathway (PID) and antigen processing and presentation (KEGG).

Additionally, we identified some other pathways, including Th1Th2 pathway, cancer immunotherapy by PD-1 blockade, costimulation by the CD28 family, and CTLA4 pathway and PD1 signaling, which were relevant to T cell activation and exhaustion, and their GSEA results were illustrated in the supplemental information (Figure S1E).

Downregulation of macrophages CXCL9/10 and MHC-II expression in the KRAS G12/13 mutation group through inhibited IFN-γ signaling pathway

Considering the key role of myeloid cells in chemotaxis and antigen presentation within the tumor immune microenvironment, we performed a clustering analysis of myeloid cells in scRNA-seq (single-cell RNA sequencing) from GSE132465.

Myeloid cells can be broadly categorized into macrophages, monocytes, and dendritic cells (Figure 3A), where we presented the relative expression level of CD68, which is highly expressed in macrophages (Figure S2A). The annotated genes for each subpopulation and their sources are detailed in Figures S2B and S2C. Our analysis revealed that CXCL9 and CXCL10, as well as HLA-DPA1 and HLA-DQB2, were predominantly expressed in the macrophage subpopulation (Figure 3B). Moreover, in both the non-mutation and KRAS G12/13 mutation group, macrophages represented the highest proportion; however, this proportion decreased in the mutation group (p < 0.0001) (Figure 3C). Additionally, CXCL9, CXCL10, HLA-DPA1, and HLA-DQB2 showed a downregulation trend in the mutation sample (Figure 3D). These single-cell results indicated that in the KRAS G12/13 mutation group, there was a reduction in the proportion of macrophages in the TME, along with a significant decrease in the expression levels of genes associated with chemotaxis and antigen presentation.

Figure 3.

Single-cell analysis of myeloid cells and macrophages

(A) Uniform manifold approximation and projection (UMAP) clustering plot of myeloid cells.

(B) Relative expression levels of CXCL9/10 and HLA-DPA1, HLA-DQB2.

(C) Comparison of the proportion of myeloid cell subpopulations between non-mutation and KRAS G12/13 mutation groups, with differential analysis conducted using the chi-square test.

(D) Volcano plot distribution of gene expressions in macrophages between KRAS G12/13 mutation and non-mutation groups.

(E) GSEA analysis between macrophages from two groups.

(F) Visualized GSEA pathways associated with interferon gamma pathways.

(G and H) UMAP clustering plot of macrophages and relative expression levels of CXCL9/10.

(I) Bubble plot of CXCL9/10, HLA-DRA, HLA-DRB1, and HLA-DRB5 in different macrophage subtypes.

To investigate the underlying mechanism, we performed GSEA on differential genes. The analysis revealed a significant downregulation of interferon signaling pathways, particularly type II interferon (IFN-γ), in patients with KRAS G12/13 mutation (Figures 3E and 3F). This finding suggested that alterations observed in macrophages may result from suppressed IFN-γ signaling. Consistently, the expression of IFN-γ was also found reduced in macrophages in the G12/13 mutation group (Figure 3D).

Further clustering analysis of macrophages (subpopulation annotations in Figure S2D) revealed that the expression of CXCL9/10, HLA-DRA, HLA-DRB1, and HLA-DRB5 was localized predominantly within the immunoregulatory subpopulation (Figures 3G–3I). This pattern suggested that this specific macrophage subset was primarily responsible for the chemotaxis and activation of CD4⁺ T cells.

Macrophages reprogrammed by IL-10 from KRAS G12/13 mutation CRC cells

Based on the proposed hypothesis of mechanism in background, we established a triple-cell two-dimensional model for experimental validation (Figure 4A) as described in STAR Methods.

Figure 4.

In vitro co-culture model and in vivo combined therapy validation

(A) Triple-cell two-dimensional model.

(B and C) Flow cytometry and comparison of expression levels of CXCL9/10 and MHC-II in THP-1 from different groups.

(D) Expression levels of proteins in MEK pathway in different HT-29 were shown by western blot.

(E) Relative expression of IL-10 mRNA in different HT-29.

(F) Measurement of IL-10 and IFN-γ concentrations in different group supernatants.

(G) Relative expression of JAK-1/STAT-1 mRNA in THP-1 from different groups.

(H) Tumor volume measurement of KRAS^G12C MC38 subcutaneous tumors.

(I) Comparison of percentage of Ki67+ cells between monotherapy and combined therapy.

(J) Representative IHC images of Ki67 across treatment groups under the same scale (200 μm).

n = 5 for each group and differences were compared using unpaired two-tailed Student’s t tests. Data are represented as mean ± SEM.

When co-cultured with HT-29 cells overexpressing either KRAS^G12C or KRAS^G13D, THP-1-derived macrophages exhibited decreased expression of CXCL9 (both p < 0.0001), CXCL10 (p = 0.0024 for G12C and p = 0.0014 for G13D), and MHC-II (both p < 0.0001) (Figures 4B and 4C). Further mechanistic investigation revealed that following the exogenous introduction of equivalent KRAS expression, HT-29 cells harboring KRAS G12C or G13D mutations demonstrated a marked increase in ERK1/2 phosphorylation compared to wild type (Figures 4D and S3) and a corresponding increase in IL-10 transcription (p < 0.0001 for G12C and p = 0.0003 for G13D) (Figure 4E). Analysis of the co-culture supernatant confirmed higher protein levels of IL-10 (p = 0.0003 for G12C and p = 0.0018 for G13D) and reduction in IFN-γ (p = 0.0109 for G12C and p = 0.0211 for G13D) in KRAS mutation groups (Figure 4F). Accordingly, co-cultured THP-1 cells showed reduced transcriptomic levels of JAK-1 (both p < 0.0001) and STAT-1 (both p < 0.0001) (Figure 4G).

Collectively, these results validated our proposed pathway: KRAS G12/13 mutation promoted expression of IL-10 in CRC cells via ERK pathway activation and then suppressed CD4⁺ T cells to secrete IFN-γ. This led to inhibition of JAK-1/STAT-1 signaling pathway in macrophages, a classical cascade downstream of IFN-γ, resulting in downregulation of CXCL9/10 and MHC-II expression.

Superior anti-tumor efficacy of KRAS^G12C inhibitor combined with CXCL9/10

To determine if restoring CXCL9/10 could enhance therapy, we tested the combination of KRAS^G12C inhibitors (sotorasib and adagrasib) with exogenous CXCL9/10 in vivo (experimental design, Figure S3B). Across two trials, the combination therapy resulted in superior suppression of subcutaneous tumor growth from KRAS^G12C-overexpression MC38 compared to monotherapy (Figure 4H), as showed by photographs of tumors (Figure S3C). This anti-tumor effect was correlated with a significant decrease in Ki67+ cells (p = 0.0001 and p = 0.0003) (Figure 4I), with representative IHC (immunohistochemistry) shown in Figure 4J.

Positive feedback between macrophages and CD4⁺ T cells disrupted by KRAS G12/13 mutation

To further confirm our findings, we performed flow cytometry and IHC on tumor tissues from patients without mutation compared to those with KRAS G12/13 mutation.

In the G12/13 mutation group, flow cytometry revealed a significant downregulation of CXCL9 and CXCL10 expression on macrophages within the tumor (both p < 0.0001) (Figures 5A and 5B). Moreover, CD4⁺ T cells showed a decreased frequency of CXCR3⁺ subset in this group (p < 0.0001) (Figures 5C and 5D). Through Pearson’s analysis, we found that the expression levels of CXCL9/10 on macrophages were positively correlated with the infiltration ratio of CXCR3⁺CD4⁺ T cells (CXCL9: R = 0.91, p = 4.1e−08, CXCL10: R = 0.91, p = 2.1e−08) (Figures 5E and 5F). In addition, the population of MHC-II⁺ macrophages was reduced in patients with KRAS G12/13 mutation (p = 0.0003) (Figures 5G and 5H). IHC revealed that tumors from the KRAS G12/13 mutation group exhibited higher expression of IL-10 (p = 0.0014) alongside lower levels of IFN-γ (p = 0.0120) (Figures 5I and 5J). At the same time, mIHC (multiplex immunohistochemistry) revealed stronger fluorescence intensities of CXCL9/10, MHC-II, and CD4 together in the tumor tissues of patients without KRAS mutation (Figure 5K).

Figure 5.

Validation of clinical specimens

(A and B) Flow cytometry and comparison of the expression levels of CXCL9/10 in macrophages.

(C and D) Flow cytometry of CXCR3⁺CD4⁺ T cells and comparison between two groups.

(E and F) Correlation analysis between the ratio of CXCR3⁺CD4⁺ T cells and the expression of CXCL9/10 in macrophages by Pearson’s analysis.

(G and H) Flow cytometry of MHC-II⁺CD68⁺ macrophages and comparison between two groups.

(I) Representative IHC of IL-10 and IFN-γ from different groups under the same scale (100 μm).

(J) Comparison of IL-10 and IFN-γ IHC score between two groups.

(K) Multiplex immunohistochemistry of two representative samples from non-mutation and KRAS G12/13 mutation patient with CXCL9/10, MHC-II, and CD4 stained with the same scale (50 μm).

n = 10 for each group, and differences were compared using unpaired two-tailed Student’s t tests. Data are represented as mean ± SEM.

These results suggested that KRAS G12/13 mutation disrupted positive feedback between macrophages and CD4⁺ T cells, leading to the downregulation of CXCL9/10 and MHC-II in macrophages as well as the infiltration and activation of CD4⁺ T cells.

Discussion

CRC has been a major focus of global research according to its high incidence and mortality.¹ Consequently, developing effective prognostic models for CRC is essential. In this study, we developed a novel prognostic model using clinical information and KRAS mutation from a cohort in our hospital. The model incorporated key independent risk factors including age, T stage, N stage, M stage, and KRAS mutation. The first four factors are consistent with findings from majority of studies. Although KRAS mutation is recognized as an independent risk factor for progression, metastasis, and drug resistance in CRC, its direct prognostic effect about OS (overall survival) remains controversial in different trails. Our study identified that G12/13 mutation was associated with a significantly higher HR compared to non-mutation (HR: 1.231, 95% CI: 1.037–1.463, p = 0.018), while else mutation did not show difference (HR: 0.915, 95% CI: 0.583–1.438, p = 0.702). However, due to limitations in detection methods, we were unable to separate G12 and G13 mutations or classify their post-mutation status, suggesting the need for further research in this area. Our model demonstrated robust performance in predicting 5-year survival, with an AUC of approximately 0.74 across cohorts, all outperforming the traditional TNM stage. According to machine learning performance thresholds, the model consistently achieved an AUC greater than 0.7, whereas the AUC for the TNM stage system remained below 0.7, highlighting a notable advancement in prognostic prediction. Furthermore, DCAs revealed that the model’s net benefit consistently surpassed that of TNM stage. These results underscored the reliability and superiority of our model compared to traditional TNM stage systems. Furthermore, using the external dataset, we validated that our model maintains its superior predictive performance over TNM stage across different therapies, demonstrating its therapy-independent nature. It should be noted that our study did not account for dynamic clinical variables, including treatment response and post-operative biomarkers such as CEA. The inclusion of these factors in future studies could enhance the model’s prognostic accuracy.

Due to the significant impact of KRAS mutation as an independent risk factor on the prognosis of CRC, we used KRAS mutation type as a grouping method to evaluate patient prognosis. Our analysis revealed that patients with KRAS G12/13 mutation exhibited poorer prognosis than their non-mutation counterparts across three cohorts, with statistically significant differences. However, the lack of difference in 5-year survival rate in the external cohort requires elucidation through future research.

KRAS is thought to be associated with tumor immune microenvironment remodel.¹² To investigate the mechanism, we analyzed bulk RNA-seq and scRNA-seq from the external cohort and identified two key differential pathways associated with chemokines and antigen presentation in macrophages, especially those immunoregulatory ones, downregulated. CXCL9 and CXCL10, both chemokines, stimulate T cell migration by binding to the G protein-coupled receptor CXCR3,⁶ while MHC-II primarily presents various tumor antigens to CD4⁺ T cells after engulfing them, thereby activating adaptive immunity.^7,13 CXCR3⁺CD4⁺ T cells represent an activated and functional T cell population, and their enrichment contributes to enhanced immune function within the TME. Specifically, they secrete IFN-γ, which activates the JAK-1/STAT-1 pathway in macrophages, thereby upregulating the expression of CXCL9/10 and MHC-II.^9,14 This interplay constitutes a positive feedback loop, where mutual activation between macrophages and CD4⁺ T cells maintains an immunologically active state in TME. However, this loop is disrupted in patients with KRAS G12/13 mutation, as evidenced by low expression of both pathways, ultimately leading to an immunologically quiescent TME. Building on prior literature,¹⁰ we experimentally confirmed that KRAS G12/13 mutation acts by activating the downstream MEK pathway, which upregulates IL-10. This cytokine, in turn, suppresses IFN-γ secretion in CD4⁺ T cells, thereby breaking the critical positive feedback. This phenomenon was corroborated by our analysis of clinical tumor specimens.

Our therapeutic investigations revealed that co-administration of KRAS^G12C inhibitors (sotorasib or adagrasib) and CXCL9&10 represents a promising strategy, demonstrating enhanced anti-tumor activity over monotherapy. This synergy highlights the potential of CXCL9/10 for prospective drug development in KRAS-mutant CRC.

As an important tumor driver gene, KRAS mutation has always been a hot point in therapy.¹⁵ This study developed a prognostic prediction model for CRC incorporating KRAS mutation through a retrospective analysis, yielding excellent results. Besides, we demonstrated that KRAS G12/13 mutation disrupts the positive feedback between macrophages and CD4⁺ T cells. Finally, the efficacy of combining KRAS inhibitors with an immunotherapeutic strategy was validated in vivo, providing a rationale for future clinical trials targeting KRAS G12/13 mutation.

Limitations of the study

While our study delineates the role of KRAS^G12/13 mutation in disrupting the positive feedback between macrophages and CD4⁺ T cells, several limitations should be considered. Firstly, the clinical associations were primarily derived from retrospective analyses of public cohorts. Although we integrated multiple datasets to enhance robustness, MMR (mismatch repair) status information was incomplete, and dynamic clinical biomarkers such as CEA and CA-199 were not included, which constrains the comprehensiveness of our prognostic model. Secondly, as illustrated in Figure S1, KRAS^G12/13 mutation is also associated with the downregulation of immune checkpoints in the TME. This indicates that additional distinct pathways contribute to the immune landscape remodeling in CRC, necessitating further research to identify potential therapeutic targets beyond the mechanism identified here. Finally, our in vivo therapeutic findings are based on a single mouse model. The generalizability and efficacy of the combined strategy (KRAS inhibitors plus CXCL9/10 restoration) should be evaluated in additional preclinical models and future clinical trials to firmly establish its therapeutic potential.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Zheng Wang (wangzh1972@126.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •This study did not generate any new sequencing datasets. All transcriptomic data used in this work are publicly available from established repositories. Bulk RNA-seq data are available from cBioPortal for Cancer Genomics (https://www.cbioportal.org/) while single-cell RNA-seq data are available via GEO GSE132465 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE132465). All dataset identifiers and accession codes are listed in the key resources table and cited herein.
• •This study did not involve the development or use of custom code. All analyses were performed using open-source software R (version 4.2.1) packages.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This study was funded by the National Natural Science Foundation of China (82270552).

Author contributions

Conceptualization, X.Y. and J.S.; data curation, X.Y. and J.S.; formal analysis, X.Y. and H.L.; funding acquisition, Z.W.; investigation, X.Y. and J.S.; methodology, X.Y. and H.L.; project administration, Z.W.; resources, J.S., T.Y., Y.Z., and Z.W.; software, X.Y. and H.L.; supervision, Z.W.; validation, T.Y., Y.Z., and T.X.; visualization, X.Y. and H.L.; writing – original draft, X.Y.; writing – review and editing, X.Y., J.S., and Z.W.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
APC anti-human CD68 Antibody	Bioledgend	Cat#333809; RRID:AB_10567107
PerCP/Cyanine5.5 anti-human CD11b (activated) Antibody	Bioledgend	Cat#301418; RRID:AB_2715897
PE anti-human CXCL9 (MIG) Antibody	Bioledgend	Cat#357904; RRID:AB_2562009
PE/Cyanine7 anti-human CXCL10 (IP-10) Antibody	Bioledgend	Cat#519508
Brilliant Violet 421™ anti-human HLA-DR Antibody	Bioledgend	Cat#307635; RRID:AB_10897449
Brilliant Violet 510™ anti-human CD3 Antibody	Bioledgend	Cat#317332; RRID:AB_2561943
Brilliant Violet 785™ anti-human CD4 Antibody	Bioledgend	Cat#317442; RRID:AB_2563242
FITC anti-human CD183 (CXCR3) Antibody	Bioledgend	Cat#353703; RRID:AB_10962910
K-Ras (E2M9G) Rabbit mAb	Cell Signaling Technology	Cat#71835; RRID:AB_3705482
p44/42 MAPK (Erk1/2) (137F5) Rabbit mAb	Cell Signaling Technology	Cat#4695; RRID:AB_390779
Phospho-p44/42 MAPK (Erk1/2) (Thr202/Tyr204) (D13.14.4E) XP® Rabbit mAb	Cell Signaling Technology	Cat#4370; RRID:AB_2315112
β-Actin Antibody	Cell Signaling Technology	Cat#4967; RRID:AB_330288
Goat Anti-Rabbit IgG H&L (HRP)	Abcam	Cat#ab205718; RRID:AB_2819160
Anti-Ki67 antibody [SP6]	Abcam	Cat#ab16667; RRID:AB_302459
Anti-IL-10 antibody	Abcam	Cat#ab217941
Anti-Interferon gamma antibody	Abcam	Cat#ab25101; RRID:AB_448613
Human CXCL9/MIG Antibody	R&D Systems	Cat#MAB392; RRID:AB_2261214
Human CXCL10/IP-10/CRG-2 Antibody	R&D Systems	Cat#MAB266; RRID:AB_2261309
Anti-MHC Class II antibody [6C6]	Abcam	Cat#ab55152; RRID:AB_944199
Recombinant Anti-CD4 antibody (Rabbit mAb)	Servicebio	Cat#GB15064; RRID:AB_3095557
Biological samples
Human colorectal cancer tissue	This paper	N/A
Mouse subcutaneous tumor tissue	This paper	N/A
Chemicals, peptides, and recombinant proteins
Hexadimethrine bromide	MCE	Cat#HY-112735
Puromycine	MCE	Cat#HY-B1743
Phorbol myristate acetate	MCE	Cat#HY-18739
Sotorasib (AMG-510)	MCE	Cat#HY-114277
Adagrasib (MRTX849)	MCE	Cat#HY-130149
MIG/CXCL9 protein, Mouse (HEK293, His)	MCE	Cat#HY-P70008
IP-10/CRG-2/CXCL10 protrien, Mouse	MCE	Cat#HY-P7227
Critical commercial assays
Human KRAS Gene 7 Mutations Fluorescence Polymerase Chain Reaction (PCR) Diagnostic Kit	AmoyDx	Cat#ADx-KR0X
RNAsimple Total RNA Extraction Kit	Tiangen Biotech	Cat#DP419
Hifair® II 1st Strand cDNA Synthesis Kit	YEASEN	Cat#111119ES60
Hifair® III One Step RT-qPCR SYBR Green Kit	YEASEN	Cat#11143ES70
Human IL-10 ELISA Kit (Interleukin-10)	Abcam	Cat#ab185986
Human IFN gamma ELISA Kit	Abcam	Cat#ab174443
TSA 5-color Kit	Servicebio	Cat#G1255
Deposited data
Internal Cohort	This paper	N/A
External Validation Cohort	cBioPortal for Cancer Genomics	https://www.cbioportal.org/
Raw and analyzed scRNA-seq data	GSE132465	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE132465
Experimental models: Cell lines
Human: HEK293T	ATCC	Cat#CRL-3216;RRID:CVCL_0063
Human: Jurkat, Clone E6-1	ATCC	Cat#TIB-152
Human: HT-29	ATCC	Cat#HTB-38
Human: THP-1	ATCC	Cat#TIB-202
Mouse: MC38	Procellsystem	Cat#CL-0972
Experimental models: Organisms/strains
C57BL/6 Mouse	Cyagen Biosciences	Cat#C001089
Oligonucleotides
Human IL10 qPCR Primer Pair	Beyotime	Cat#QH03717S
Human JAK1 qPCR Primer Pair	Beyotime	Cat#QH03905S
Human STAT1 qPCR Primer Pair	Beyotime	Cat#QH05753S
Human ACTB qPCR Primer Pair	Beyotime	Cat#QH00001S
Recombinant DNA
TK-PCDH-human KRASWT-copGFP-T2A-Puro	This paper	N/A
TK-PCDH-human KRASG12C-copGFP-T2A-Puro	This paper	N/A
TK-PCDH-human KRASG13D-copGFP-T2A-Puro	This paper	N/A
TK-PCDH-mouse KRASG12C-copGFP-T2A-Puro	This paper	N/A
Software and algorithms
Flowjo	This paper	https://www.flowjo.com/
Case Viewer	This paper	https://www.caseviewer.app
ImageJ	This paper	https://imagej.nih.gov/ij/
R (version 4.2.1)	This paper	https://www.r-project.org
Graphpad Prism 10	This paper	https://www.graphpad.com
Other
Human CD3/CD28 T cell Activation Magnetic Beads	MCE	Cat#HY-K0353
EZ Cell Transfection Reagent	Life-iLab	Cat#AC04L092
Omni-Easy™ One-Step PAGE Gel Fast Preparation Kit	Epizyme Biomedical Technology	Cat#PG223

Experimental model and study participant details

Patients

Male and female patients from our center were diagnosed with CRC at the Department of Gastrointestinal Surgery between Jan. 2015 and Dec. 2020, underwent colorectal cancer surgery, and subsequently received long-term follow-up. We excluded deaths that were not attributable to CRC. All patients were over 18 and had signed informed consent prior to the study. Authors had omitted details that could reveal the identity of patients. The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board (IRB) of Renji Hospital Ethical Committee (LY2025-042-B).

Human cells lines and culture condition

HT-29, THP-1, Jurkat, MC38 and HEK-293T were obtained from and authenticated by the American Type Culture Collection (ATCC) without any mycoplasma contamination and maintained in RPMI-1640 (Roswell Park Memorial Institute-1640) or DMEM (‌Dulbecco’s Modified Eagle Medium) supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin at 37°C in a 5% CO₂ humidified incubator.

Animals

C57BL/6 male mice (6-8 weeks old, 18–22 g) were used for in vivo subcutaneous tumor experiments from Cyagen Biosciences. Mice were housed in a controlled environment under a 12 h light/12 h dark cycle. The environment was temperature-controlled at 22 ± 2°C with a relative humidity of 50%–70%. Ethical approval for this study was obtained from the Animal Ethics Committee of Renji Hospital (RJ2021-0630) and Cyagen Biosciences (TACU25-MS001-3041).

Method details

Diagnosis criteria

Diagnosis of colorectal cancer was confirmed through clinical symptoms and auxiliary tests, including colonoscopy and CT scans, etc. Pathological examination was conducted following surgical resection, with immunohistochemical staining used for diagnosis. Gender and age, two key indicators in our analysis, were obtained from the patient’s ID card. Primary location was classified as either colon or rectum. The colon includes the ascending, transverse, descending, and sigmoid regions while the rectum ranges from the distal sigmoid colon to the anal verge. T, N, M, and TNM stage were determined by the 8th edition of the AJCC colorectal cancer stage system. The pathological indicators such as MLH1, MSH2, and MSH6 were classified as negative, intermediate, or positive based on immunohistochemical score: “-” for negative, “−/+” for intermediate, and “+” to “++++” for positive according to their pathology reports. MMR∗ (mismatch repair∗) represents adaption from the standard due to practical constraints in clinical test. Loss of expression in one or more MMR proteins were defined as dMMR∗, whereas those with retained expression of all three MMR proteins were defined as pMMR∗. The mutation type of P53 and KRAS were identified from the genetic test report by the Department of Pathology. P53 was classified as either mutated or non-mutated, while KRAS mutation is categorized as non-mutation, G12/13 mutation, or other mutation type, which was detected by Human KRAS Gene 7 Mutations Fluorescence Polymerase Chain Reaction (PCR) Diagnostic Kit and multiple mutation was excluded.

Development and validation of predictive model for prognosis

The internal cohort was randomly divided into a discovery cohort (n=1,276) and an internal validation cohort (n=544) in a 7:3 ratio. The cohort from cBioportal was used as an external validation cohort (n=564) to examine the accuracy of our model, where deaths unrelated to CRC were excluded from the analysis. LASSO (Least Absolute Shrinkage and Selection Operator) regression was first carried out and then univariate Cox regression was used. Those variables meet the threshold of P value < 0.05 were selected for multivariable Cox regression to build the prognosis predictive model in the discovery cohort. To visualize this model, we established a nomogram. Then we used ROC (receiver operating characteristic) curve and calibration curve to estimate the prediction accuracy of this model while DCA (decision curve analysis) for its clinical utility in both the internal and the external validation cohort.

BulkRNA-Seq and ScRNA-Seq Analysis

Based on KRAS mutation of the patient, we divided them into two groups: non-mutation and G12/13 mutation and plotted survival curves for both internal and external cohorts. After excluding samples with missing data, we performed an analysis of the bulk RNA-seq data in the external cohort. DEGs (differentially expressed gene) between 2 groups were identified based on |Log₂FC| ≥ 0.585 with P value < 0.05 and visualized using a volcano plot. GO (gene ontology) and KEGG (Kyoto encyclopedia of genes and genomes) enrichment analyses were then conducted. Pathways with P value < 0.05 were selected for further investigation and visualized in bubble plots. We then intersected DEGs from those selected GO and KEGG pathways and identified genes associated with chemokine and antigen presentation. The entire gene list was then uploaded into GSEA (Gene Set Enrichment Analysis) for further validation. GSEA pathways containing these genes were identified and further visualized by GSEA enrichment plots. ScRNA-seq data came from GEO accession GSE132465. To investigate the associated genes after dimensionality reduction and annotation, we conducted differential expression analysis, the results of which were displayed in a volcano plot, followed by pathway analysis visualized using GSEA.

Stable KRAS wild type or point mutation CRC cell lines construction

To establish KRAS non-mutation or G12/13 mutation cell lines, the lentiviral system was employed. The target plasmids (shown in key resource table) were co-transfected with packaging plasmids psPAX2 and pMD2.G (2:1:1) into HEK293T cells using EZ Cell Transfection Reagent (3μL/1μg plasmids). The viral supernatant was harvested at 48 and 72hours after transfection, filtered through a 0.45μm filter, and subsequently used to infect HT-29 or MC38 in the presence of 5μg/mL polybrene. Following a 48-hour incubation, stable polyclonal populations were selected and maintained in complete medium supplemented with 5μg/mL puromycin for 2 days. Transfection and expression efficiency were evaluated by quantifying the percentage of GFP (green fluorescent protein) positive cells under a fluorescence microscope.

Triple-cell two-dimensional model

As previously reported, a triple-cell two-dimensional model was employed to investigate cell-cell interactions (Figure 4A).¹⁶ A transwell insert with 0.2μm pore membrane, which prevents cell migration, was used. Following the established protocol, transfected HT-29 were seeded on the bottom of the transwell insert. Jurkat cells were pre-activated with CD3/CD28 beads prior to co-culture and placed inside the insert. Meanwhile, suspended THP-1 cells were differentiated into macrophages using PMA and allowed to adhere to the bottom of the 6-well plate. After 48 hours of co-culture in RPMI-1640 + 10% FBS + 1% penicillin-streptomycin, the supernatant and individual cell fractions were collected for subsequent experiments.

Flowcytometry

A single-cell suspension was obtained from adherent cells and tumor tissue through enzymatic digestion, followed by staining of surface markers with antibodies listed in key resources table. The cluster characterized by CD68⁺CD11b⁺ was classified as macrophages, where the expression of CXCL9, CXCL10, and MHC-II were evaluated. In contrast, the group identified by CD3⁺CD4⁺ was recognized as CD4⁺ T cells, and the expression of CXCR3 was assessed.

IHC (immunohistochemistry) & mIHC (multiplex immunohistochemistry)

Tissue sections were deparaffinized, rehydrated, and subjected to heat-induced antigen retrieval in citrate buffer. After blocking endogenous peroxidase with H₂O₂ and non-specific sites with normal serum, slides were incubated with primary antibody overnight at 4°C. Then sections were probed with an HRP-conjugated secondary antibody, developed with DAB substrate, and counterstained with hematoxylin. Finally, sections were dehydrated, cleared, and mounted for microscopic examination. Multiplex staining of CXCL9, CXCL10, MHC-II, and CD4 co-expression on human cancer tissues was conducted using a 5-color kit following the manufacturer’s instructions. All primary antibodies used were listed in key resources table.

Western Blot

HT-29 were lysed using RIPA buffer supplemented with protease and phosphatase inhibitor cocktail. Equal amounts of protein were separated by SDS-PAGE on 12.5% gradient gels and subsequently transferred onto PVDF membranes. The membranes were then blocked with 5% non-fat milk in TBST for 1 hour at room temperature. Following blocking, the membranes were incubated overnight at 4°C with specific primary antibodies. After extensive washing with TBST, the membranes were incubated with appropriate HRP-conjugated secondary antibodies for 1 hour at room temperature. The protein bands were visualized using an enhanced chemiluminescence (ECL) substrate and captured on a chemiluminescence imaging system.

Real-time RT-PCR(Reverse transcription-polymerase chain reaction)

Following total RNA extraction from HT-29 and THP-1, cDNA was synthesized using Hifair® II 1st Strand cDNA Synthesis Kit. Gene expression was quantified by qPCR with Hifair® III One Step RT-qPCR SYBR Green Kit and normalized to β-actin. The primer sequences are provided in the key resources table.

ELISA(Enzyme-linked immunosorbent assay)

IL-10 and IFN-γ were quantified using correspongding ELISA kits according to the manufacturer’s protocol.

Animal model

Mice (n=5 per group) were subcutaneously inoculated with 1∗10⁶ KRAS^G12C MC38 cells on Day 0. Tumor formation was confirmed on Day 10. Subsequently, mice were treated orally with Adagrasib or Sotorasib (30mg/kg) every three days, with or without intratumoral injection of recombinant mouse CXCL9 and CXCL10 protein (500ng each). Tumor dimensions were measured regularly throughout the treatment period. All mice were euthanized by cervical dislocation on Day 28, and subcutaneous tumors were harvested for subsequent analysis.

Quantification and statistical analysis

The results in this study were analyzed using GraphPad Prism 10. Continuous variables were compared using unpaired two-tailed student’s t-tests, while categorical variables were analyzed with Chi-square tests. Correlation analyses were performed using Spearman’s rank correlation. Statistical significance was marked in each figure.

Published: December 17, 2025