Single-cell RNA sequencing reveals neutrophil differentiation-related genes for immunotherapy response prediction in colorectal cancer

Abstract

Background

Immunotherapy has demonstrated promising results in treating colorectal cancer (CRC), yet not all patients benefit equally, highlighting the need for predictive biomarkers and additional adjuvant. Neutrophils, pivotal in immune regulation and inflammation, remain underexplored in the context of CRC immunotherapy, holding the potential targets for improving immunotherapy efficacy.

Methods

Single-cell RNA sequencing data from 19 CRC patients, including immunotherapy-treated individuals and controls, were analyzed. Genes associated with neutrophil differentiation, termed Neutrophil Differentiation-Related Genes (NDRGs), were identified through trajectory analysis and refined to nine genes (TMBIM6, CTSS, CYCS, DDX3X, DYNLL1, LGALS1, GANI2, RPS29, and TUBA1A) using Elastic Net. Drug screening was performed to identify potential therapeutic agents targeting NDRGs.

Results

Immunotherapy induced a shift in neutrophil composition, marked by a reduction in inflammatory neutrophils and an increase in immune neutrophils. The nine NDRGs were successfully validated and utilized to develop a predictive model for immunotherapy response. Drug screening identified Ivermectin as a potential therapeutic agent targeting NDRGs.

Conclusions

This study highlights the dynamic role of neutrophils in the CRC immune microenvironment, their potential as biomarkers for immunotherapy response, and their utility as therapeutic targets to improve immunotherapy outcomes.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-025-15355-7.

Introduction

Colorectal cancer (CRC) remains one of the most prevalent and deadly malignancies worldwide, with immunotherapy emerging as a promising treatment strategy [1–3]. Immune checkpoint inhibitors (ICIs) have demonstrated remarkable efficacy in a subset of CRC patients, leading to durable responses and improved survival. However, a significant proportion of patients fail to benefit from immunotherapy, underscoring the urgent need for reliable biomarkers to predict treatment response, optimizing patient selection, and novel agents boostering immunotherapy [4–7].

The tumor immune microenvironment (TIME) plays a critical role in shaping the response to immunotherapy [8–10]. Among the diverse immune cell populations within the TIME, neutrophils have garnered increasing attention due to their dual role in both promoting and suppressing tumor progression [11–13]. Neutrophils are highly plastic cells that can be divided into different phenotypes depending on the context [14, 15]. Despite their abundance in CRC and their involvement in immune regulation, the specific contributions of neutrophils to immunotherapy response remain poorly understood.

In this study, we analyzed scRNA-seq data from 19 CRC patients, comparing those treated with ICIs to untreated controls. Using pseudotime trajectory analysis, we observed significant shifts in neutrophil subsets, with a decline in inflammatory neutrophils and a rise in immune neutrophils following immunotherapy. Building on these findings, we identified 9 Neutrophil Differentiation-Related Genes (NDRGs) that were strongly associated with patient immunotherapy responses. These genes were integrated into a predictive model using Elastic Net, validated for their ability to forecast immunotherapy response and overall survival. Furthermore, we investigate potential therapeutic agents that may enhance immunotherapy efficacy by targeting NDRGs.

Methods

Patient samples and data collection

The scRNA-seq data with 19 CRC patients were obtained from the Gene Expression Omnibus (GEO) database (GSE205506). This dataset comprised 30 immunotherapy-treated samples and 10 untreated controls from 19 CRC patients. Clinical information and expression data for each patient were additionally collected from The Cancer Genome Atlas Colon Adenocarcinoma (TCGA-COAD) dataset.

For validation, single-cell RNA-seq data 169 CRC patients receiving ICIs treatment (GSE236581), bulk RNA-seq data from 62 CRC patients (GSE12945), 226 CRC patients (GSE14333), 177 CRC patients (GSE17536), 55 CRC patients (GSE17537), 65 CRC patients (GSE29621), 122 CRC patients (GSE38832), 562 CRC patients (GSE39582) were retrieved from the GEO database.

Furthermore, an in-house validation cohort comprising five CRC patients was used to corroborate the single-cell findings. For each patient, paired tumor and adjacent non-tumorous tissues were collected. Both qPCR and bulk RNA-seq data were generated for validation analyses.

Human colorectal tissue samples were obtained from the Gastrointestinal Surgery Department of Dongguan People’s Hospital under an approved protocol. Informed consent for participation in the study was obtained from all patients prior to sample collection. Samples were collected during surgical resection, immediately stored at −80℃ until analysis.

RNA sequencing and preprocessing

For bulk RNA-seq data processing, raw reads were subjected to quality control checks. Cleaned reads were aligned to the reference genome using STAR [16], and gene expression levels were quantified using RSEM [17].

For scRNA-seq preprocessing, raw data was conducted using the Seurat [18] package (v5.2) in R. Cells were normalized by NormalizeData function, and low-quality cells were filtered based on the following criteria: (1) nFeature_ < 100 or nFeature > 7500; (2) nCount_ < 200; (3) pMT < 25%. The analysis followed the standard Seurat workflow. Batch effects were corrected using the Harmony package (v1.2) with the top 30 principal components and default parameters, while the parameter max.iter.harmony was set to 20 to ensure convergence.

Quantitative real-time PCR (qPCR)

Total RNA was extracted from bulk colorectal tumor and adjacent normal tissues using the TRIzol reagent (Invitrogen, USA) according to the manufacturer’s protocol. RNA concentration and purity were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA). Complementary DNA (cDNA) was synthesized from 1 μg of total RNA using the PrimeScript RT reagent kit (Takara, Japan). Quantitative PCR was performed using TB Green Premix Ex Taq II (Takara, Japan) on a QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems, USA). Each reaction was conducted in triplicate. Relative mRNA expression levels were calculated using the 2^ − ΔΔCt method, normalized to GAPDH as an internal control. Each patient had one replicate. Primers were followed:

Gene	Forward primer (5′–3′)	Reverse primer (5′–3′)
DYNLL1	AGAGATGCAACAGGACTCGGT	CCAGGTGGGATTGTACTTCTTG
LGALS1	CTGTGCCTGCACTTCAACC	CATCTGGCAGCTTGACGGT
GNAI2	ACGACTCAGCCGCCTAC	GTGCGTCTCCACGATCC
TMBIM6	CATATAACCCCGTCAACGCAG	GCAGCCGCCACAAACATAC
CTSS	TGACAACGGCTTTCCAGTACA	GGCAGCACGATATTTTGAGTCAT
TUBA1A	TCGATATTGAGCGTCCAACCT	CAAAGGCACGTTTGGCATACA
CYCS	CTTTGGGCGGAAGACAGGTC	TTATTGGCGGCTGTGTAAGAG
RPS29	CGCTCTTGTCGTGTCTGTTCA	CCTTCGCGTACTGACGGAAA
DDX3X	AGCAGTTTTGGATCTCGTAGTG	ACTGTTTCCACCACGTTCAAAT

Pseudotime trajectory analysis

To explore the dynamic changes in neutrophil composition, pseudotime trajectory analysis was conducted using Monocle2 [19]. Cells were ordered along a pseudotime axis according to their transcriptional profiles. Differential gene expression analysis was performed to identify key genes driving these transitions. Genes with significant changes along the pseudotime axis were further analyzed to elucidate their roles in neutrophil differentiation and functional states.

Cell annotation

For initial annotation, SingleR was applied using the Human Primary Cell Atlas as reference datasets. To refine the identification of neutrophil subpopulations, we further collected neutrophil subset marker genes from ten pan-cancer neutrophil clusters reported by Wu et al., 2024 [20] and used these markers to manually validate and annotate the neutrophil clusters in our dataset. Marker expression was visualized with FeaturePlot and DotPlot functions in Seurat to confirm cell-type identity. Canonical marker genes: CD3E, CD3D (lymphocyte), CD4 (CD4 + T cells), CD8A (CD8 + T cells), NKG7, KLRD1 (NK cells), CD79A, MS4A1 (B cells), CD38 (plasma cells), LYZ, CD68 (MonoΦ), CSF3R, S100A8/9 (Neutrophils), CD1D, (cDC), TCF4, LILRA4 (pDC), EPCAM, KRT18 (Epithelial cells), PECAM1, VWF (Endothelial cells), ACTA2 (Smooth muscle cells).

Cell–cell communication analysis

To analyze cell–cell communication, we utilized the CellChat [21] package on scRNA-seq data. Following data preprocessing and cell type annotation, a CellChat object was constructed to model intercellular signaling networks. Communication networks were inferred by analyzing ligand-receptor interactions, leveraging a curated database of known ligand-receptor pairs. The strength and specificity of these interactions were quantified, enabling the identification of key signaling pathways and communication patterns between distinct cell types.

Tumor Immune Dysfunction and Exclusion (TIDE) prediction

To assess the immune response to immunotherapy in CRC, we utilized the TIDE algorithm (http://tide.dfci.harvard.edu/). TIDE predicts immune evasion mechanisms, such as immune dysfunction and exclusion, which could affect the response to ICIs. TIDE was applied to predict the immunotherapy response for each CRC patient based on their tumor transcriptome data. The TIDE score was calculated for each sample to assess the likelihood of response to immunotherapy.

Construction and validation of the predictive model

Elastic Net regression was employed to identify genes predictive of immunotherapy response. Using the glmnet package, the analysis was performed on the TCGA-COAD cohort, with ICI response scores from TIDE serving as the response indicator. The optimal lambda value was determined through cross-validation using the cv.glmnet function, and the corresponding coefficients for each gene were calculated. A final set of 9 genes with non-zero coefficients, designated as Neutrophil Differentiation-Related Genes (NDRGs), was selected to construct a predictive model for immunotherapy response. The model was subsequently validated using the GSE12945 dataset, and its performance was assessed using receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics.

Functional research high and low response score groups

To conduct functional enrichment analysis, the GSVA [22] package was utilized to perform single-sample gene set enrichment analysis (ssGSEA) on TCGA-COAD data stratified into high and low response score groups. Curated gene sets were obtained from the MSigDB database [23]. Pathways with significant enrichment were identified based on an adjusted p-value threshold of < 0.05. The results were visualized using violin plots to emphasize key biological processes and pathway activity differences between the response groups.

Drug screening

Potential small-molecule compounds targeting the NDRGs were identified in silico using the Comparative Toxicogenomics Database (CTD) [24], which curates experimentally observed effects of chemicals on gene expression. For each NDRG, we retrieved small molecules reported to upregulate or downregulate its expression. To prioritize candidate compounds, we aligned the reported regulatory direction of each small molecule with the coefficient direction in our Elastic Net NDRG model: 1) Genes with positive coefficients were paired with compounds reported to upregulate their expression; 2) Genes with negative coefficients were paired with compounds reported to downregulate their expression. Compounds satisfying these criteria across multiple NDRGs were considered potential therapeutic candidates.

Statistical analysis

All statistical analyses were conducted using R (version 4.4.1). Differential expression analysis was performed using the Wilcoxon rank-sum test, while survival analysis was carried out using the Kaplan–Meier method and assessed with the log-rank test. A p-value < 0.05 was considered statistically significant. Correlations between variables were evaluated using Spearman Rank Correlation Analysis.

Figures

The illusion figures were collected from Vecteezy.com.

Results

Cell type identification and immune cell landscape

Analysis of single-cell RNA sequencing data from 19 CRC patients including 30 immunotherapy-treated and 10 untreated samples unveiled a heterogeneous immune cell landscape within the tumor microenvironment (Fig. 1a, Fig. S1a, Table S1, 2). Following data preprocessing (111,158 cells after QC) and clustering, distinct immune cell populations were identified, including T cells, B cells, macrophages, dendritic cells, and other subsets (Fig. 1b, c). To refine the analysis, lymphoid and myeloid immune cell clusters were further subclustered, enabling a detailed characterization of their compositional and functional diversity within the tumor microenvironment.

Fig. 1.

Single-cell analysis in colorectal cancer. a Schematic of workflow; b UMAP visualization of Seurat clusters of CRC samples; c UMAP visualization of cell annotation of CRC samples; d UMAP visualization of Seurat clusters of lymphatic cells; e UMAP visualization of cell annotation of lymphatic cells; f UMAP visualization of Seurat clusters of myeloid cells; g UMAP visualization of cell annotation of myeloid cells

The lymphoid immune cell clusters were further subdivided into 29 distinct clusters, which were categorized into six main subtypes: NK cells, CD4 + T cells, CD8 + T cells, B cells, and plasma cells (Fig. 1d, e). Similarly, myeloid immune cells were partitioned into 12 clusters and annotated as monoΦ, neutrophils (4442 cells after QC), conventional dendritic cells (cDCs), and plasmacytoid dendritic cells (pDCs) (Fig. 1f, g). The expression of marker genes for these cell types was visualized in a dot plot, confirming the identity and purity of each annotated cluster (Fig. 2a).

Fig. 2.

Cell–cell communication. a Dot plot showing the cell marker genes; b Boxplot of immune cell types in TCGA by CIBERSORT; c Neutrophils cell–cell communication strength between other cell types; d GLAECTIN signaling cell–cell communication strength

When comparing samples across treatment conditions, the majority of immune cell types were evenly distributed (Fig. S1b, c). Notably, epithelial cells were enriched in untreated samples, whereas endothelial cells and B cells exhibited reduced proportions in the Anti–PD-1 plus celecoxib treatment group.

Cell–cell interaction and immune network dynamics in the TIME

Analysis of immune cell subtypes in the TCGA cohort revealed significant differences in cell type proportions between tumor and normal samples (Fig. 2b). Neutrophils’ proportion was significantly higher in tumor, suggesting an importance role in tumor microenvironment.

To investigate intercellular communication within the TIME, we performed cell–cell interaction analysis using CellChat. The results demonstrated extensive interactions between neutrophils and other immune cells, highlighting their central role in CRC regulation (Fig. 2c, Fig. S1e). Among these interactions, the GALECTIN signaling pathway, mainly mediated by galectins such as LGALS1 [25] and LGALS9 [26], is known to modulate ICI responses, showed high communication strength across multiple cell types, including neutrophils and CD8 + T cells (Fig. 4d). This finding further underscores the regulatory role of neutrophils in modulating immune responses within the CRC microenvironment.

Fig. 4.

Construction and verification of the predictive response model. a, b Elastic Net Regression algorithm predictive charactieristics; c The ROC curve of nine NDRGs model validated in TCGA-COAD cohort; d Heatmap of qPCR expression of the nine NDRGs in five independent CRC validation patients; e Barplot of NDRG response scores for the five validation patients, used to classify responders, non-responders, and unclassified cases; f–h Representative H&E staining images of tumor tissues from unclassified (f), responder (g), and non-responder (h) patients, showing differences in lymphocyte infiltration within the tumor microenvironment

Differentiation patterns of neutrophils in response to immunotherapy

To investigate the transition of neutrophils during therapy, we performed pseudotime trajectory analysis using Monocle2 (Fig. 3a). The trajectory revealed that neutrophils were partitioned into five distinct states (Fig. 3b). The root of trajectory was defined by the expression of mature neutrophil markers S100A8 and S100A9, corresponding to State 1, which exhibited the lowest expression levels of these markers (Fig. 3d). From this origin, neutrophils predominantly progressed along two developmental branches, terminating in State 4 and State 5. Analysis of treatment-associated changes demonstrated a clear shift in neutrophil composition over the course of therapy (Fig. 3c, h). In untreated patients, inflammatory neutrophils (State 4), characterized by high expression of CXCL8 and IL1B (Fig. 3g, j), were predominant. In contrast, immunotherapy-treated patients showed a significant decrease in these inflammatory neutrophils and a concomitant increase in immune neutrophils (State 5), marked by elevated CD74 and HLA-DRA expression (Fig. 3g, k). Furthermore, although not reaching statistical significance, immune neutrophils tended to be more abundant in the pCR group (Fig. 3l). This transition was associated with changes in the expression of key neutrophil differentiation markers, suggesting that immunotherapy may drive neutrophils toward a more immune phenotype.

Fig. 3.

Pseudotime analysis of neutrophils. a Pseudotime trajectory depicting the continuous developmental progression of neutrophils, with color gradients representing pseudotime; b Identification of five distinct differentiation states along the trajectory; c Pseudotime trajectories comparing neutrophils under different treatment conditions; d Expression dynamics of neutrophil maturation-related genes along pseudotime; e Expression patterns of inflammatory neutrophil marker genes along pseudotime; f Expression patterns of immune neutrophil marker genes along pseudotime; g Dot plot showing the expression of inflammatory and immune neutrophil marker genes across the five states; h Bar plot showing the proportions of the five differentiation states across different treatment groups; i Proportions of inflammatory and immune neutrophil subtypes among patients with different clinical responses (CR, complete response; PR, partial response; SD, stable disease); j Box plot comparing the proportions of inflammatory neutrophils between treated and untreated samples; k Box plot comparing the proportions of immune neutrophils between treated and untreated samples; l Box plot comparing the proportions of immune neutrophils between patients with pCR (pathological complete response) and non-pCR

In another single-cell CRC patients receiving ICI treatment cohort GSE236581, according to similar process, annotated neutrophils were clustered into inflammatory and immune neutrophils. The complete response (CR) group had higher proportion of immune neutrophils, confirming the role of increasing immune neutrophils we observed in neutrophil differentiation during ICI treatment (Fig. 3i). Further investigation into the differentiation dynamics of neutrophils highlighted their potential role in modulating the tumor immune microenvironment in response to immunotherapy.

Development of a prognostic risk model based on neutrophil differentiation

To establish a prognostic tool for immunotherapy response, we identified 498 differentially expressed genes in neutrophils across five pseudotime states, termed Neutrophil Differentiation-Related Genes (NDRGs). Using the TCGA cohort with predicted immunotherapy response (based on TIDE scores), we trained the model (Fig. 4a, b). The predictive model for immunotherapy response was constructed using Elastic Net to select the most relevant genes. Nine predictive genes were identified, and the final prognostic model was: Response score = −0.528 * DYNLL1 Exp −0.723 * LGALS1 Exp −0.526 * GNAI2 Exp + 0.616 * TMBIM6 Exp + 1.010 * CTSS Exp −0.536* TUBA1A Exp + 0.728 * CYCS Exp −1.058 * RPS29 Exp + 0.548 * DDX3X Exp. These nine NDRGs expressed across different cell types, mainly in myeloid cells (Fig. S1d, Table S3). The predictive performance of the model was evaluated using ROC curves, with the AUC demonstrating its robust predictive ability (0.78, Fig. 4c). This model provides a reliable tool for stratifying patients based on their likelihood of responding to immunotherapy.

Clinical relevance and validation of the prognostic model

The nine-gene NDRGs prediction model was further validated using in-house cohort from five CRC patients. qPCR results revealed distinct expression patterns of the nine NDRGs across patients (Fig. 4d), and response scores were calculated by multiplying gene expression levels by their corresponding model coefficients (Fig. 4e). According to the model, two patients (P3, P5) were predicted as responders, exhibiting positive response scores, while one patient (P4) were predicted as non-responders with negative scores. The remaining two patients had response scores near baseline and were therefore classified as unclassified. Histopathological evaluation using H&E staining supported these predictions: the predicted responder group displayed a lymphocyte-infiltrated tumor microenvironment (Fig. 4g), whereas the non-responder group showed a lymphocyte-depleted phenotype (Fig. 4h). The unclassified group exhibited a heterogeneous microenvironment, with tumors showing both high and low levels of lymphocyte infiltration (Fig. 4f).

To assess the broader applicability of our model, we evaluated its prediction power in additional CRC cohorts. We found that response scores were significantly negative correlated with TIDE scores (Fig. 5a-g). Given that the high TIDE scores are indicative of ICI non-response, whereas high response scores correspond to ICI responding, our model demonstrated robust predictive efficacy (Fig. 5h-n).

Fig. 5.

Validation of NDRGs response prediction model in public available cohorts. a-g Scatter plots showing the Pearson correlation between the NDRG response score and TIDE score in seven independent CRC cohorts; h-n The ROC curve for the NDRG response model in the same seven cohorts, indicating the predictive performance and discriminative power of the model across multiple datasets

Next, we explored the clinical relevance of the prediction model in the TCGA-COAD cohort. By assigning response scores to TCGA samples, we observed that female patients had significantly higher response scores compared to male patients (Fig. 6a). Additionally, early-stage tumors exhibited higher response scores than advanced-stage tumors (Fig. 6b-d), suggesting that immunotherapy may be more effective in earlier disease stages.

Fig. 6.

Clinical predictive power of the response score and immune subtype cells correlation. a-d Violin plots comparing clinical characteristics (gender, N stage, stage, and T stage) response score difference. Significant differences are indicated in the plots; e Heatmap depicting the correlation between immune cell subtypes and response score. Correlations are color-coded from positive (red) to negative (blue)

In immune subtype correlation analysis, neutrophils showed a positive correlation with mast cells, a recently identified cell type implicated in influencing ICI responses (Fig. 6e). This finding highlights the potential interplay between neutrophils and mast cells in shaping the tumor immune microenvironment and modulating immunotherapy outcomes.

Functional analysis of key predictive genes

Further analysis of the nine NDRGs revealed distinct expression patterns, survival correlations, and pathway alterations. Genes positively associated with immune ICI responses—CTSS, CYCS, TMBIM6, and DDX3X—were upregulated in the response group. In contrast, genes negatively associated with ICI responses—TUBA1A, LGALS1, GNAI2, RPS29, and DYNLL1—were downregulated (Fig. 7a, b).

Fig. 7.

Correlation of gene signatures with overall survival and pathway enrichment. a Heatmap showing gene expression patterns in TCGA as training group; b Heatmap showing gene expression patterns in GSE12945 as testing g group; c-g Kaplan–Meier survival curves for genes associated with survival outcomes, including response score and prediction genes; h Violin plots illustrating differential expression of key genes across different clinical groups, with enrichment for pathways related to immune response and cancer progression. i Boxplot of immune cell types in validation patients by CIBERSORT

To assess the prognostic value of the nine NDRGs and the associated response scores, we performed survival analyses using TCGA-COAD data. Survival analysis demonstrated that patients with higher response scores and elevated expression of CTSS and CYCS exhibited better overall survival (Fig. 7c- e). Conversely, patients with high expression of LGALS1, and RPS29 had worse survival outcomes (Fig. 7f, g). These findings underscore the prognostic value of the 9 NDRGs and their potential roles in modulating immunotherapy response and patient survival.

Using the response score, we stratified the TCGA cohort into response (high) and nonresponse (low) groups. Functional enrichment analysis revealed that inflammatory signaling pathways were downregulated in the response group (Fig. 7h), suggesting a more immune-suppressed TIME that could potentially be reversed by ICIs.

Immune subtype analysis showed that CD4 + resting memory T cells were significantly upregulated in the response group (Fig. 7i), aligning with our in-house data where activated CD4 + memory T cells were downregulated in the response group (Fig. 4f). This finding highlights a potential role for the transition between resting and activated CD4 + memory T cell states in mediating immunotherapy outcomes related to neutrophils.

Small molecule compounds of predictive genes

To explore potential therapeutic strategies, we conducted a drug screen targeting the nine NDRGs using small molecule compounds from the CTD. Compounds were selected based on their ability to modulate the expression of these genes. Screening results were categorized into two groups: small molecules that increased the expression of all positive coefficient NDRGs (Table S4) and those that decreased the expression of all negative coefficient NDRGs (Table S5).

Among the candidates, Ivermectin, a well-known antiparasitic drug, emerged as a promising candidate. It was found to be able to decrease the expression of all six negative coefficient NDRGs (LGALS1, DYNLL1, GNAI2, RPS29, and TUBA1A) (Table S5). This finding is especially noteworthy given recent studies demonstrating Ivermectin’s ability to modulate the tumor immune microenvironment. Specifically, Ivermectin has been reported to convert “cold” tumors into “hot” tumors. Additionally, preclinical studies have shown that Ivermectin inhibits CRC cell growth, further supporting its potential as an adjunctive therapy in CRC immunotherapy. Other compounds were listed in the tables.

Discussion

The TIME of CRC is a dynamic ecosystem shaped by complex interactions between malignant cells and diverse immune populations. Among these, neutrophils have historically been overlooked despite their abundance and functional plasticity [27–30]. Our study provides critical insights into the role of neutrophils in CRC immunotherapy, revealing their capacity to transition between inflammatory and immune states in response to treatment. This plasticity, captured through pseudotime trajectory analysis, underscores neutrophils as active modulators of the TIME rather than passive bystanders. The observed shift from CXCL8 +/IL1B + inflammatory neutrophils to immune-regulatory subsets in immunotherapy-treated patients suggests that therapeutic interventions may reprogram neutrophil phenotypes to favor antitumor immunity. Importantly, these findings align with emerging evidence in other cancers, such as pancreatic cancer [31] and non-small cell lung cancer [32], where neutrophil subtypes has been linked to immunotherapy resistance. Our work extends this paradigm to CRC, positioning neutrophil heterogeneity as a central determinant of therapeutic outcomes.

Central to this study is the development of a nine-gene Neutrophil Differentiation-Related Gene (NDRG) model, which integrates transcriptional dynamics and clinical outcomes to predict immunotherapy response. The model’s robustness was validated across multiple cohorts, including an external dataset and an in-house cohort with histopathological confirmation. Notably, the NDRG model highlights genes with direct mechanistic relevance to immune regulation. For example, CTSS encodes a protease critical for antigen processing [33] and MHC class II presentation [34], while DDX3X is involved in apoptosis [35] and interferon signaling. Conversely, negative-coefficient genes such as LGALS1 [25] are known to promote immunosuppression by inhibiting T cell activation and fostering regulatory T cell infiltration. These molecular links not only validate the model’s biological plausibility but also suggest actionable targets for therapeutic intervention.

Clinically, the NDRG model revealed intriguing associations between immunotherapy response and patient demographics. The higher response scores observed in female patients resonate with broader trends in oncology, where females often exhibit stronger immune activation and better outcomes in immune checkpoint inhibitor (ICI) trials. Similarly, the correlation between early-stage tumors and favorable response scores underscores the importance of tumor burden and immune exhaustion in shaping therapeutic efficacy. These findings advocate for personalized treatment strategies that consider sex- and stage-specific immune profiles [36, 37]. Furthermore, the positive correlation between neutrophils and mast cells adds a novel dimension to our understanding of myeloid crosstalk in the TIME. Mast cells, recently implicated in ICI through their associations with activation of immune subtype cells [38, 39]. This interplay warrants further investigation, particularly in the context of combination therapies targeting multiple myeloid subsets.

Functional enrichment analysis revealed that responders exhibited downregulation of inflammatory pathways, a finding initially counterintuitive given the pro-inflammatory nature of antitumor immunity. However, this may reflect a resolution of chronic, non-productive inflammation—a hallmark of advanced tumors—and a shift toward coordinated immune activation [40]. This balance between inflammation and immune memory echoes recent studies in melanoma, where prolonged ICI efficacy correlates with the maintenance of stem-like T cell populations [41].

The drug repurposing screen identified Ivermectin through in silico analyses, a widely used antiparasitic agent, as a promising candidate to augment immunotherapy. By suppressing negative-coefficient NDRGs (e.g., LGALS1), Ivermectin may reverse neutrophil-driven immunosuppression. Preclinical studies support this hypothesis: Ivermectin has been shown to inhibit NF-κB signaling [42], modulate purinergic signaling and the ATP/P2X4/P2 × 7/Pannexin-1 axis, and enhance immune cell infiltration in breast cancer models. Its ability to convert immunologically “cold” tumors to “hot” aligns with our goal of remodeling the TIME [43]. In is also shown directly that Ivermectin can inhibit CRC growth [44]. Building upon these insights, we propose a comprehensive series of future experiments to evaluate Ivermectin’s therapeutic potential in CRC: (1) In vitro Studies: We plan to conduct dose–response analyses and cytotoxicity assays on CRC cell lines to determine the optimal concentration of Ivermectin that modulates NDRG expression without inducing significant cytotoxic effects.; (2) In Vivo Studies: Utilizing CRC mouse models under ICIs treatment, we intend to investigate the effects of Ivermectin on tumor growth, immune cell infiltration, and the tumor microenvironment. Key endpoints will include evaluating changes in neutrophils populations, CD8 + T cell activation within the tumor milieu.

By undertaking these studies, we seek to provide robust empirical evidence supporting the repurposing of Ivermectin as an adjunctive therapy in CRC. We emphasize that our current findings are exploratory and serve as a foundation for future research endeavors aimed at validating Ivermectin’s role in modulating the tumor immune microenvironment and enhancing the efficacy of immunotherapeutic strategies.

Conclusion

In conclusion, our work positions neutrophils as central players in CRC immunotherapy, offering a dual opportunity for biomarker discovery and therapeutic targeting. The NDRG model provides a framework for patient stratification, while the exploration of agents like Ivermectin opens new avenues for combination therapies. Future research should focus on unraveling the mechanistic links between neutrophil states, TIME remodeling, and therapeutic resistance, ultimately advancing toward precision immunotherapy in CRC.

Abbreviations

CRC

Colorectal cancer

NDRG

Neutrophil differentiation-related gene

ICI

Immune checkpoint inhibitor

TIME

Tumor immune microenvironment

scRNA-seq

Single-cell RNA sequencing

GEO

Gene expression omnibus

TCGA-COAD

The cancer genome atlas colon adenocarcinoma

TIDE

Tumor immune dysfunction and exclusion

ROC

Receiver operating characteristic

AUC

Area under the curve

ssGSEA

Single-sample gene set enrichment analysi

CTD

Comparative toxicogenomics database

MHC

Major histocompatibility complex

Authors’ contributions

Conceptualization, L.W. and H.W.; methodology, G.W.; software, J.Z.; validation, H.W.; formal analysis, Y.C.; investigation, S.R.; resources, Y.Z.; data curation, H.W.; writing—original draft preparation, L.W.; writing—review and editing, X.Z.; visualization, J.C.; supervision, J.H.; project administration, X.Z.; funding acquisition, X.Z.. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Dongguan Science and technology of social development Program, grant number 20231800935742, Oncology Clinical Research Foster Program, grant number Z202405, Guangdong Basic and Applied Basic Research Foundation, grant number 2023A1515140182, and Dongguan Science and technology of social development Program, grant number 20221800902092.

Data availability

Published data can be accessed from the GEO under the accession number GSE205506, GSE236581, GSE12945, GSE14333, GSE17536, GSE17537, GSE29621, GSE38832, GSE39582, and TCGA-COAD.

Declarations

Ethics approval and consent to participate

All procedures involving human participants were performed and approved by ethical standards of the Ethics Committee of The Affiliated Dongguan People’s Hospital of Southern Medical University (approval no. KYKT2021‑018). The study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from all individual participants included in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.