2.1 Patient Selection

We collected data from nine whole exome sequencing (WES) datasets of patients who underwent ICI therapy for biomarker discovery (Table S1). The Miao2019 cohort [9], the Hugo cohorts [10], the Riaz cohorts [11], the Miao2018 cohort [12], the Rizvi cohort [13], the Snyder cohorts [14], and the Van Allen cohorts [15], these seven datasets constituted the discovery cohort. The Hellmann cohort [16], and the Liu cohort [17], these two datasets were designated as the validation cohorts. We downloaded eight WES datasets and corresponding clinical information from the cBioPortal database (https://www.cbioportal.org). The Riaz cohort was obtained from the original literature [11].

In order to combine this data from multiple sources, we utilized the processing method described by Zhang et al. [18]. Initially, we excluded three tumor types with a sample size of less than 10. We also removed 33 samples that had a non-evaluable response (NE), 7 samples that were not profiled and 7 samples classified as “OTHER CONCURRENT THERAPY”. Furthermore, we eliminated 151 duplicate samples in the Miao2018 cohort. This cohort had 27 overlapping samples with the Rizvi cohort, 37 with the Snyder cohort, and 87 with the Van Allen cohort. After filtering, only 10 patients with head and neck cancer were left, which were also deleted. Table 1 summarizes the characteristics of the filtered data. The ICI discovery cohort includes four tumor types and the following types of drug treatment: anti-PD-1, anti-CTLA-4 and anti-CTLA-4 + anti-PD-1. The validation cohort included melanoma and NSCLC, the two tumor types with the highest proportion in the discovery cohort (70% and 15%).

Furthermore, we collected data from 32 types of solid cancer data from The Cancer Genome Atlas (TCGA) to conduct further analysis on the CDC42 gene set mutations as a biomarker. This data includes WES data and RNA-seq data, which were obtained using TCGAbiolinks [19]. Additionally, we obtained Cibersort immune infiltration values and TCR Shannon for each TCGA cancer sample from Thorsson et al. [20].

2.2 CDC42 Gene Set Mutation Definition

We identified the CDC42 gene set, comprising key genes like CDC42, CDC42BPA, CDC42BPB, CDC42BPG (CDC42 binding protein kinase alpha, beta, gamma), CDC42EP1–5 (CDC42 effector protein 1–5), CDC42SE1–2 (CDC42 small effector 1–2). These genes are integral to CDC42 function, and any non-synonymous mutation in this set signifies a mutated CDC42 gene set status. Such mutations could influence CDC42 signaling, potentially compromising its function.

2.3 Clinical Endpoint Analysis

The objective response rate (ORR) was defined as the proportion of patients who received ICI therapy and achieved a complete response (CR) or partial response (PR) [21]. Durable clinical benefit (DCB) was defined as a CR, PR, or stable diseases (SD) that lasted for more than 6 months [22].

2.4 Immune Cell Fraction Analysis

We obtained the leukocyte fraction from Thorsson et al. [20]. The lymphocyte fractions were aggregated by using the cibersort estimate, including B cells naïve, B cells memory, T cells CD4 naïve, T cells CD4 memory resting, T cells CD4 memory activated, T cells follicular helper, Tregs, T cells gamma delta, T cells CD8, NK cells resting, NK cells activated, and Plasma cells [20]. The molecular estimate for tumor-infiltrating lymphocyte (TIL) fraction was obtained by multiplying the aggregated lymphocyte fraction from the cibersort estimate with the leukocyte fraction obtained from Thorsson et al. The image estimate for TIL fraction was obtained from Saltz et al. [23].

2.5 Immune Signatures Analysis

We obtained 29 immune signatures from He et al. [24] and performed single-sample gene set enrichment analysis (ssGSEA) using the “GSVA” R package [25] based on these signatures.

2.6 Gene Set Enrichment Analysis (GSEA) Analysis

We used TCGA RNA-seq data and the DESeq2 package [26] to identify differentially expressed genes. Subsequently, we conducted GSEA analysis on the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway using the clusterProfiler package [27].

2.7 Estimation of Cytolytic Activity

We estimated cytolytic activity (CYT) based on the method described by Rooney et al. [28]. This involves calculating the geometric mean of granzyme A (GZMA) and perforin 1 (PRF1) expression.

2.8 Mutation and Neoantigens Analysis

For this study, we assessed tumor mutational burden (TMB) using the number of non-synonymous mutations. The data for nonsilent mutation, silent mutation, single nucleotide variation (SNV) neoantigens and indel neoantigens were obtained from Thorsson et al. [20].

2.9 Statistical Analysis

We used the two-sided Fisher's exact test to explore the clinical benefit difference between CDC42 gene set status. The two-sided Wilcoxon rank sum test was used to compare the TMB and neoantigen load (NAL) of ICI therapy data. We also used the two-sided Wilcoxon rank sum test to compare TCGA gene expression levels, mutation rate, neoantigens, cell fraction, immune signatures, TCR Shannon and CYT between the CDC42 gene set mutation group and the CDC42 gene set wild type group. Additionally, we plotted the KM curve of PFS and OS using the logrank test based on CDC42 gene set status, which used the $$$ {\chi}^2 $$$ test statistic to calculate P values [29]. Fisher's exact test was implemented using the python package scipy [30]. The logrank test, Cox Proportional-Hazards analysis, and Wilcoxon rank sum test were implemented using the survminer package and ggsignif package in R version 4.2.3 (https://www.r-project.org). In in vivo pharmacodynamic study, statistical analysis was done by GraphPad software, version 9, and one-way ANOVA followed by Bonferroni post hoc test was used to analyze multiple groups.

2.10 Animal Experiment

All procedures performed on animals were conducted in accordance with the Institutional Animal Care and Use Committee at the Shanghai Institute of Materia Medica, Chinese Academy of Sciences (IACUC Issue NO. 2023-10-ZMY-03). For the pharmacodynamics experiment, BALB/c mice (6–8 weeks old) were purchased and inoculated subcutaneously with 1 × 10⁶ 4 T1 tumor cells into the right side of the mice's axilla. The animals were divided into four groups irregularly once the tumor volume reached approximately to 100 mm³. ML141 (#HY-12755, MedChemExpress) was administered in a solution containing PEG300, dimethyl sulfoxide, and PBS [40/5/55 (v/v/v)]. The mice were then treated intraperitoneally with or without ML141 (30 mg/kg once a day) and/or 150 μg/mouse of anti-PD-1 antibody (#BE0273, BioXCell) every other day for one injection. Tumor volumes were calculated using the formula: V = (length×width²)/2. Body weights and tumor volumes of the mice were measured daily. The tumor growth inhibition (TGI) value was calculated using the formula: TGI = [1-Relative Tumor Volume (Treatment)/Relative Tumor Volume (Vehicle)] × 100%.

2.11 Flow Cytometry Analysis

The tumor tissues were firstly digested into single cells using a digestion solution containing 0.001% hyaluronidase, 0.1% collagenase, 0.002% DNase, 120 μM MgCl₂, and 120 μM CaCl₂ in RPMI 1640 medium. Subsequently, red blood cells were lysed using ammonium chloride for 3 min and then the cell samples were stained with Fixable Viability Stain 510 (#564406, BD). The Fc receptors were blocked with TruStain FcX (anti-mouse CD16/32) antibody (#101320, Biolegend) and stained with the following antibodies: APC-Cy7 rat anti-mouse CD45 (#557659, BD), FITC CD3 monoclonal antibody (17A2) (#11–0032-82, Invitrogen), APC anti-mouse CD25 (#17-0251-82, Invitrogen), BV786 rat-anti-mouse CD4 (#563331, BD), PE rat anti-mouse FOXP3 (#563101, BD), and Brilliant Violet 421 anti-mouse CD8a antibody (#100738, Biolegend). The stained cells were analyzed using the Agilent Novocyte 3000 instrument, and all data were analyzed with FlowJo software. Flow cytometry gating strategies were illustrated in Figure S1.

2.12 RT-PCR Analysis

Total RNA was extracted from samples using the Vazyme FastPure Cell/Tissue Total RNA Isolation Kit (RC112-01). RNA was reverse transcribed into cDNA using the HiScript III qRT SuperMix (Vazyme, R323-01-AC). RT-qPCR was performed on the QuanStudio 5 PCR Detection System (Thermo Fisher) using the ChamQ SYBR qPCR Master Mix (Vazyme, Q331-AA). All experiments were conducted according to the manufacturer's instructions (Vazyme Biotech). The primer sequences used are as follows: mouse Actb forward: tgagctgcgttttacaccct, mouse Actb reverse: gccttcaccgttccagtttt; mouse Arg-1 forward: acattggcttgcgagacgta, mouse Arg-1 reverse: atcaccttgccaatccccag; mouse CD206 forward: gactgctgctgagtccagtt, mouse CD206 reverse: agggatcgcctgttttccag; mouse IL-10 forward: gctccaagaccaaggtgtct, mouse IL-10 reverse: cggagagaggtacaaacgagg; mouse Nos2 forward: gttctcagcccaacaatacaaga, mouse Nos2 reverse: gtggacgggtcgatgtcac, mouse TNFα forward: aggctgccccgactacgt, mouse TNFα reverse: gactttctcctggtatgagatagcaaa; mouse IL-1β forward: tcgctcagggtcacaagaaa, mouse IL-1β reverse: catcagaggcaaggaggaaaa.

2.13 Histological Analyses

The tumor tissue samples from mice in each group were fixed in 4% formaldehyde solution and then sent to Wuhan Servicebio Technology Co. Ltd. for immunohistochemical staining. The images of these stained samples were viewed and captured using SlideViewer software, and the immunohistochemistry (IHC) results were statistically processed using ImageJ software.

3.1 Mutation in CDC42 Gene Set Was Associated With Improved Clinical Outcomes for ICI Therapy

The results of discovery cohort were shown in Figure 1, the ORR of patients in the CDC42 gene set mutation was significantly higher (p = 1.26E-3) compared to the CDC42 gene set wild type. Additionally, the DCB of patients in the CDC42 gene set mutation was also significantly higher (p = 2.66E-2) than the wild type group. Moreover, CDC42 gene set mutation patients had significantly longer overall survival time (p = 0.01, HR = 0.59, 95% CI = 0.39–0.89) and PFS time (p = 3.9E-3, HR = 0.46, 95% CI = 0.27–0.79) compared to the CDC42 gene set wild type group.

The biomarker function of CDC42 gene set mutations was validated on two validation cohorts. As shown in Figure 2a, the PFS of patients with CDC42 mutations in NSCLC patients was significantly longer than that of wild type patients (p = 3.8E-2, HR = 0.49, 95% CI = 0.25–0.97). As shown in Figure 2b,c, in the melanoma validation cohort, patients with CDC42 gene set mutations had significantly longer PFS and OS (OS: p = 1.16E-2, HR = 0.38, 95% CI = 0.17–0.83; PFS: p = 1.18E-2, HR = 0.49, 95% CI = 0.28–0.86). Figure 2d shows the prediction of patient survival by CDC42 gene set in the TCGA dataset (p = 0.18, HR = 0.91, 95% CI = 0.8–1.04).

3.2 Assessment of Intrinsic Immune Response Landscapes in CDC42 Gene Set Wild Type and Mutation Tumors

We initially examined the relationship between CDC42 gene set status and immunogenicity in the ICI therapy cohort. As shown in Figure 3a,b, the levels of TMB and NAL in the CDC42 gene set mutation group were significantly higher than those in the CDC42 gene set wild type group (TMB: p < 2.22E-16, NAL: p = 2.9E-8). These findings suggest increased immunogenicity in tumors with CDC42 gene set mutations. We also explored the relationship between CDC42 gene set status and immunogenicity in the TCGA cohort. Compared to CDC42 gene set wild type tumors, both the nonsilent mutation rate and the silent mutation rate were significantly higher in CDC42 gene set mutation tumors (p < 2.22E-16, Figure 3c,d). Additionally, both SNV neoantigens and indel neoantigens were significantly more abundant in in CDC42 gene set mutation tumors compared to CDC42 gene set wild type tumors (p < 2.22E-16, Figure 3e,f). These results in the TCGA cohort further support the notion that the CDC42 gene set mutations are associated with enhanced tumor immunogenicity.

Then, we investigated the relationship between CDC42 gene set status and the expression of immune-related molecules, including two class MHC molecules, immune checkpoints, and co-stimulators. We found that immune checkpoint genes PDCD1, CD274 and CTLA-4 were upregulated in the CDC42 gene set mutation group, as shown in Figure 3h–j. Additionally, we observed significantly higher expression of MHC1, MHC2, other immune checkpoints (ICPs), and co-stimulators in CDC42 gene set mutation tumors compared to CDC42 gene set wild type, as shown in Figure 3g.

3.3 Assessment Extrinsic Immune Response Landscapes in CDC42 Gene Set Wild Type and Mutation Tumors

The different situations of immune cell infiltration result in different clinical outcomes of ICI therapy [31]. Therefore, we investigated the difference in tumor microenvironment (TME) between CDC42 gene set wild type and mutation tumors. This included analyzing immune cell score, signatures representing immune cell function, and differential gene expression related to immune cell and ICI therapy efficiency.

As shown in Figure 4a,b, the leukocyte fraction and lymphocyte fraction in CDC42 gene set mutation tumors were significantly higher than those in CDC42 wild type gene set tumors (leukocyte fraction, p = 5.7E-06; lymphocyte fraction, p = 6.3E-06). As TIL is crucial for killing tumors [32], TIL fractions estimated at both molecular and image levels were compared. Figure 4c shows that TIL fractions (molecular estimate) in CDC42 gene set mutation tumors are significantly higher than those in CDC42 gene set wild type tumors (p = 9.5E-3). Figure 4d shows that TIL fractions (images estimate) in CDC42 gene set mutation tumors are significantly higher than in those CDC42 gene set wild type tumors (p = 3E-06).

Cibersort scores based on the TCGA cohort were also compared between CDC42 gene set wild type and mutation tumors. As shown in Figure 4e, the percentages of immune cell types were compared in detail. We found significant differences in most of the immune cell scores between CDC42 gene set wild type and mutation tumors. For example, the CD8 T cell and macrophage M1 cell scores in the CDC42 gene set mutation type were significantly higher than that in the CDC42 gene set wild type tumors. Figure 4f shows ssGSEA results based on 29 immune signatures. We found that CD8 T cell and checkpoint signatures in CDC42 gene set mutation type tumors were significantly higher than those in CDC42 gene set wild type tumors. Figure 5a further shows that immune signatures were significantly enriched in CDC42 gene set mutation tumors compared to CDC42 gene set wild type tumors.

We further conducted GSEA analysis, as well as expression analysis of chemokines and chemokines receptors, interleukins and interleukins receptors, TCR, and cytolytic activity score based on CDC42 gene set status. In Figure 5b, we observed enrichment of base excision repair, homologous recombination, mismatch repair pathway enriched in CDC42 gene set mutation tumors, while fatty acid degradation and Ras signaling pathway were enriched in CDC42 gene set wild type tumors. These results related to ICI therapy response. For example, Jiang et al. reported that DDR pathways are associated with the response to ICIs treatment [33]. Ward et al. reported that activation of the Ras signaling pathway leads to an immunosuppressive tumor microenvironment, hindering T cells activation and infiltration, thus affecting the therapeutic efficacy of ICI [34]. Li et al. reported that increased lipid content is correlated with a favorable ICI therapy response [35], suggesting that higher lipid accumulation may indicate a higher likelihood of a positive response to ICI therapy. In Figure 5c, we found a higher TCR Shannon score in CDC42 gene set mutation tumors compared to that in CDC42 gene set wild tumors (p = 2.9E-3). The cytolytic activity score in CDC42 gene set mutation patients was significantly higher than in CDC42 gene set wild type patients (Figure 5d,p < 2.22E-16). As shown in Figure 5e, lots of chemokines in CDC42 gene set mutation patients were significantly higher than in CDC42 gene set wild type patients, such as CXCL9, CXCL10, CXCL11 and CXCL13. There were also significant differences in interleukins and interleukins receptors expression between CDC42 gene set statuses, such as IL-10 and IL21. All these results could help prove CDC42 gene set mutations as effective biomarkers, see discussion.

3.4 Exploring CDC42 Inhibitor Combined With ICI to Enhance the Efficacy of Immunotherapy

We investigated the antitumor effect of combining a CDC42 inhibitor (ML141) with anti-PD-1 antibody. Treatment with anti-PD-1 antibody alone demonstrated ordinary therapeutic capacity, with a TGI of 38% (Figure 6a). Treatment with ML141 alone resulted in a TGI of 63% (Figure 6a). Notably, the combination of ML141 and anti-PD-1 antibody significantly reduced tumor growth (TGI = 86%) and prolonged survival time compared to the anti-PD-1 antibody alone group (Figure 6a,b). This suggests that the combination therapy can effectively exert potent antitumor immune activity. Furthermore, these treatments did not lead to weight loss in the mice (Figure 6c), indicating that this dosing regimen is safe. To determine the role of the immune response in the antitumor activity of this dosage regimen, we analyzed the immune cells' infiltration in the TME using flow cytometry experiments. The results showed that a combination of ML141 and an anti-PD-1 antibody significantly increased the frequency of CD45⁺ lymphocytes, CD3⁺ T cells, and CD8⁺ cytotoxic T cells in the TME compared to the anti-PD-1 antibody alone (Figure 6d–f). However, it does not affect the frequency of CD4⁺CD25⁺FOXP3⁺ regulatory T cells (Tregs) in the TME (Figure S2). To further analyze the distribution and quantity of CD3⁺ T cells and CD8⁺ cytotoxic T cells in tumor tissue, we performed IHC staining. The results showed that the distribution ratio of CD3⁺ and CD8⁺ cells was significantly higher in the combination therapy group compared to the monotherapy group (Figure S3). The use of ML141 alone also raised the frequency of CD3⁺ T cells in the TME, aligning with earlier discoveries that the pharmacological inhibition of CDC42 triggers antitumor immune activity [4]. We also analyzed the expression of M1 markers (IL-1β, Nos2, TNFα) and M2 markers (Arg-1, CD206, IL-10) in tumor-associated macrophages. The results showed that, compared to the anti-PD-1 antibody alone, the combination of ML141 and the anti-PD-1 antibody significantly increased the expression of M1 markers and significantly reduced the expression of M2 markers in tumor-associated macrophages (Figure S4). This suggests that the combination therapy can significantly promote the transformation of tumor-associated macrophages into the M1 type, compared to the monotherapy. In conclusion, these findings suggest that the use of a CDC42 inhibitor in conjunction with an ICI has an additive therapeutic effect, enhancing the efficacy of immunotherapy.