Sample collection

23 MSI-H/dMMR mCRC patients were treated with anti-PD-1 monotherapy at Yunnan Cancer Hospital (The Third Affiliated Hospital of Kunming Medical University) between August 1, 2020, and May 31, 2022. A PD-1 blocker (200 mg, tislelizumab, Bei Gene Ltd., China) was injected intravenously on the 1st day of each 21-day cycle. Efficacy was evaluated radiologically after every third cycle of treatment. Patients were considered sensitive to anti-PD-1 treatment if they showed a complete response (CR) or partial response (PR), and patients were considered resistant to anti-PD-1 treatment if they had progressive disease (PD) or stable disease (SD). Fresh intestinal tumor tissues (2–4 mm) were taken during colonoscopy. Tissue samples from a total of 23 patients (10 sensitive and 13 resistant) were analyzed by immunohistochemistry (IHC) and immunofluorescence (IF). Six patients (3 PR and 3 PD) were randomly selected for scRNA-seq. All participants provided written informed consent before beginning the study. Additionally, the ethics committee of Yunnan Cancer Hospital approved all study protocols that involved human subjects according to the ethical principles described in the Helsinki Declaration. The inclusion criteria were as follows: primary colorectal adenocarcinoma confirmed by colonoscopy biopsy and distant metastasis confirmed by CT/MR; MSI-H/dMMR status confirmed by multiplex qPCR or IHC; undergoing first-line anti-PD-1 monotherapy. Patients were excluded if we were unable to obtain a sample through colonoscopy due to contraindications or we were unable to perform single-cell sequencing due to insufficient cell activity or quantity.

Radiology and colonoscopy

Siemens (SOMATOM Definition AS +) 128 slice spiral CT was used for plain and enhanced scanning. Patients were fasting and had performed bowel preparation as directed. The scanning range covered the entire abdominal cavity. The scanning layer was 1.0 mm thick with an interval of 0.6 mm. Iohexol (300 mg/ml, 100 ml) was used as the contrast agent. The delay time of arterial phase scanning was 35–40 s, and that of venous phase scanning was 70–80 s. MRI was performed using a Philips Elision 3.0 T, and the scanning sequence included transverse T1WI_Tse, sagittal and coronal T2WI_Tse, high-resolution T2WI_Tse, diffusion-weighted imaging, and multiphase dynamic enhancement. Gadolinium diamine was used as the contrast agent. Colonoscopy was performed using the Olympus CV-290, and tissues were collected by doctors with at least five years of experience.

Single-cell preparation

Fresh tumor tissues obtained from CRC patients were immediately transferred to MACS C-tubes (Miltenyi Biotec) with digestive enzymes. Then, digestion was performed by a gentleMACS Octo Dissociator (Miltenyi Biotec) (30 min at 37 °C). Single cells were processed by Chromium Controller (10X Genomics) according to the manufacturer's protocol. Briefly, cells were washed with 20 mL of RPMI 1640 (Gibco), filtered through a 70-μm nylon strainer (BD Falcon), collected by centrifugation (330 × g, 10 min, 4 °C), and resuspended in a basic solution containing 0.2% fetal bovine serum (FBS; Gibco).

Single-cell capture and library preparation

A 10 × Chromium system (10 × Genomics) and library preparation by LC Sciences were utilized to run the single cells according to the recommended protocol for the Chromium Single Cell 30 Reagent Kit (v2 Chemistry). The Illumina HiSeq4000 was used for sequencing, and a 10 × Cell Ranger package (v1.2.0; 10 × Genomics) was used for the postprocessing and quality control of libraries.

Quantitative analysis of the single-cell sequencing data

All single-cell sequencing data were analyzed by Cell Ranger V6.1.1. The results are shown in Additional file 1: Table S1. CRC samples from 3 resistant patients (named R1, R2, R3) and 3 sensitive patients (named S1, S2, S3) contained a total of 56,092 cells; the number of genes detected in each sample ranged from 17,564 to 18,093; the median number of cellular unique molecular identifiers ranged from 3,403 to 6,915; and the sequencing saturation ranged from 47 to 63%. These results indicate that, overall, the sequencing quality was sufficient for use in subsequent correlation analysis.

Processing of CRC single-cell sequencing data

In total, this analysis included 56,092 cells from the sensitive group and resistant group. The Seurat package in R (version 4.0.5) was used for analysis and quality control [9]. Low-quality cells (n = 3,071) were removed if genes were detected only in < 3 cells or if there were < 200 total genes detected by gene-cell matrixes. Next, the global-scaling method LogNormalize was performed to normalize the gene expression values for the remaining 53,021 cells. Then, the FindVariableFeatures function combined with the vst method in R studio was employed to identify the most variable genes, which were used for dimensionality reduction. After principal component analysis, 2000 highly variable genes were identified. Jackstraw and ScoreJackStraw functions were applied to determine the most significant principal components. Finally, graph-based unsupervised clustering was conducted and visualized using a nonlinear t-distributed stochastic neighbor embedding (t-SNE) plot, defined by the FindNeighbors and FindClusters functions.

Identification and characterization of cell subtypes

The identities of cell types were characterized using the SingleR (V1.6.1) package based on the Celldex database. The FindMarkers function in the R package Seurat was used to list the markers of each cell cluster with min.pct = 0.5, logfc.threshold = 1, min.diff.pct = 0.3, and P < 0.05. The markers used in this pipeline are listed in Additional file 2: Table S2.

To investigate the molecular mechanisms involved in each cell subtype, biological process (BP), cell composition (CC), and molecular function (MF) GO enrichment analysis and KEGG pathway analysis were performed using the R package clusterProfiler (version 3.14.3), with the threshold of significance set to adjusted (adj.) P < 0.05.

Screening of differentially expressed genes (DEGs)

The first 2000 highly variable genes were screened using the FindVariableFeatures function combined with the vst method. The FindMarkers function in the R package Seurat was used not only to find the marker genes for different cell subtypes (with screening parameter thresholds of min.pct = 0.5, logfc.threshold = 1, min.diff.pct = 0.3, P < 0.05) but also to identify DEGs of each cell subtype between anti-PD-1-sensitive and anti-PD-1-resistant samples (with the threshold set to |avg_log2FC|≥ 0.5, P ≤ 0.05). The overlapping genes between pseudotime-related genes and PD-1 resistance-related DEGs were considered to be candidate key genes.

Pseudotime analysis

To reveal differences in immune cells in the sensitive and resistant groups, Monocle software (version 2.20.0) [10] was used to analyze sample trajectories and explore the differentiation process. First, a more sophisticated method (dpFeature) was created based on the foundation of cluster and custom developmental marker genes of Monocle. The signature genes with a high degree of dispersion (q < 0.01) were identified among cell subtypes that were selected by dpFeature. Next, DDRTree was applied for dimensionality reduction and pseudotemporal alignment of cells along the trajectory, and finally, the trajectories were visualized as 2D t-SNE maps. Pseudotime-related genes were identified based on q < 0.05 of the abovementioned signature genes.

Protein‒protein interaction analysis of candidate key genes

The Search Tool for the Retrieval of Interacting Genes (STRING; http://string.embl.de/) was used to perform protein‒protein interaction (PPI) analysis on candidate key genes. The STRING database can be used to assess the direct (physical) and indirect (functional) associations of proteins [11]. Cytoscape 3.6.1 was used to establish a network model of PPI analysis results. Based on the STRING online tool, the PPI network of the candidate key genes was constructed with medium confidence = 0.4. The top four genes were selected as the key genes in this study based on the connectivity (degree) of each node in the PPI network.

Cell culture and transfection

Colorectal cancer cell line CT26 was selected for this study. The plasmid used for cell transfection was synthesized by Ono Company. Eighteen to twenty-four hours prior to lentivirus infection, 3 × 105/well adherent cells were spread in 6-well plates. The number of cells transfected with lentivirus was approximately 6 × 105/well. When cells adhered to the wells and were 70% confluent, the original culture medium was replaced by 2 ml fresh culture medium containing 8 μg/ml polyzoan and a proper amount of virus suspension. Cells were then incubated at 37 °C for 8 h, after which the virus-containing medium was replaced with fresh medium. If transfection was successful, fluorescent protein was visible after 48–72 h. If no fluorescence was observed, the infection protocol was repeated. Puromycin was added to screen for lentivirus overexpression for one month.

Animal experiments

All animal experiments were approved by the Animal Ethics Committee of Kunming Medical University. In the process of experimental operation, we strictly follow the ARRIVE guidelines. Throughout the experiment, researchers did not know which group the animals taken out of the cage would be assigned to, animal managers and researchers doing the experiment did not know the assignment sequence, and researchers evaluating, testing or quantifying the results of the experiment did not know the means of intervention.

BALB/c mice (male, aged 6–8 weeks old, 20-30 g) were purchased from Beijing Sipeifu Biotechnology Co., Ltd. All mice were maintained in an SPF room in the animal-housing facilities at the Kunming Medical University with food and water provided at will. The experiment was carried out after 1 week of adaptive feeding. Based on the degree of freedom (E) of variance analysis proposed by Mead, we estimated the sample size of the experimental animals we need [12]. The total sample size of mice in this study was 25, and they were randomly divided into 5 groups with 5 mice in each group. All mice were randomly divided into 5 groups using a simple random sampling method, defined as OE-NC, OE-IL-1β, OE-IL-1β + Diacerein, OE-IL-1β + Nivolumab, and OE-IL-1β + Diacerein + Nivolumab groups, respectively. For the control group model (OE-NC group), 1 × 106 empty vector stabilized cells in 100ul PBS were subcutaneously injected into the flank of mice. For the treatment model, 1 × 106 over-expressed IL-1β stably transfected cells in 100ul PBS were subcutaneously injected into the same site of mice. When the tumor volume reached 40mm3 (i.e. the 7th day), the OE-IL-1β + Diacerein group began intraperitoneal (i.p.) injection of the IL-1β antagonist (Diacerein, 0.07 mg/kg), OE-IL-1β + Nivolumab group received i.p. injection of anti-PD-1 antibody (Nivolumab, 2 mg/kg); The OE-IL-1β + Diacerein + Nivolumab group was treated with a combination of diacerein and Nivolumab intraperitoneally. The mice in the OE-NC group and OE-IL-1β group were administered the same volume of PBS. Diacerein and Nivolumab were intraperitoneally injected every 3 days from the 7th day of tumor cell inoculation.

The weight of mice and the volume of tumor were measured in 0d, 7d, 10d, 12d, 14d and 16d in each group, respectively. Tumor volume was calculated as ½ (Length × Width2). On the 16th day after inoculation of tumor cells, all mice were sacrificed with an intraperitoneal injection of sodium pentobarbital (200 mg/kg). Determine mice death based on the disappearance of corneal reflex and the emission of pupils. Tumor tissues were stripped and weighed, and subsequent molecular biological experiment were performed. The selection of 16d as the experimental endpoint is based on the pre- experiment results. The selection of 16d as the experimental endpoint is based on the pre- experiment results. Throughout the entire experiment, the length of tumor should not exceed 20 mm, and the tumor weight should not exceed 10% of the mice weight [13].

Quantitative real-time PCR

Total RNA was extracted from the cultured cell line CT26 or tissues using TRIzol reagent (Ambion) and reverse transcribed into cDNA by the SureScript first strand cDNA synthesis kit (Servicebio) according to the manufacturer’s instructions. 2xUniversal Blue SYBR Green qPCR Master Mix (Servicebio) and CFX96 sequence detection system (Bio-Rad, Hercules, CA, USA) were used for qPCR, and the following primers were used: IL-1β (human), forward: 5'- AATCTCCGACCACCACTACA-3' and reverse: 5'-GACAAATCGCTTTTCCATCT-3'; MMP9 (human), forward: 5'-ATGAGCCTCTGGCAGCCCCTGGTCC-3' and reverse: 5'- GGACCAGGGGCTGCCAGAGGCTCAT-3'; GAPDH (human), forward: 5'-CCCATCACCATCTTCCAGG-3' and reverse: 5'-CATCACGCCACAGTTTCCC-3'; IL-1β (mouse), forward: 5'- CCTATGTCTTGCCCGTGG-3' and reverse: 5'- GTGGGTGTGCCGTCTTTC-3'; MMP9 (mouse), forward: 5'-GTGTGTTCCCGTTCATCTTT-3' and reverse: 5'- GCCGTCTATGTCGTCTTTAT-3'; GAPDH (mouse), forward: 5 '- CCTTCCGTGTTCCTACCCC-3' and reverse: 5'-GCCCAAGATGCCCTTCAGT-3'. GAPDH was used as a standardized endogenous control, and 2−△△CT was used to calculate the relative mRNA expression.

Western blotting

Protein samples were isolated from tissues or cells using RIPA lysis buffer (Servicebio, Wuhan, China) containing 1% protease and phosphatase inhibitors (PMSF; Servicebio). A BCA protein assay kit (Biyuntian Biotechnology) was used for protein quantification. Sodium dodecyl sulfate‒polyacrylamide gel electrophoresis was applied to separate proteins of different molecular weights, and proteins were transferred to a polyvinylidene fluoride (Servicebio) membrane. The membrane was blocked with 5% skim milk for 90 min at room temperature and subsequently incubated with primary antibodies (Proteintech; IL-1β 1:1000; MMP9 1:1000; β-actin 1:25,000) at 4 °C overnight, followed by incubation with HRP-conjugated secondary antibodies for 2 h at room temperature.

Immunohistochemistry and immunofluorescence

Paraffin sections of tissues were deparaffinized and rehydrated. Then, sodium citrate buffer was used to extract the antigens to be detected, and antigen retrieval was completed by heat induction. Sectioned tissues were incubated with 3% H2O2 for 15 min to block endogenous peroxidase activity (IF proceeded without this step) and then blocked with PBS containing 5% fetal bovine serum for 30 min. Tissues were incubated with primary antibodies overnight at 4 degrees C, followed by incubation in the dark with the conjugated secondary antibodies at room temperature for another 2 h. DAB was used as nuclear markers for IHC. DAPI (EX:330-380 nm, Em:420 nm) was used to stain the cell nuclei (blue), Alexa Fluor 488 (EX:495 nm, Em:519 nm) was used to stain CD11b (green), Alexa Fluor 555(EX:555 nm, Em:565 nm) was used to stain CD14, CD15 and CD8 (red), Alexa Fluor 594 (EX:590 nm, Em:617 nm) was used to stain CD33 (orange).

Flow cytometry

Polymorphonuclear (PMN)-MDSCs/monocytic (M)-MDSCs were stained with CD11b-FITC (Biolegend, 101,205, USA), Ly-6G-PE (Biolegend, 127,607, USA), and Ly-6C-APC (Biolegend, 128,016, USA), and CD8+ T cells were stained with CD3-FITC (Biolegend, 100,203, USA) and CD8-PE (Biolegend, 100,707, USA) according to the manufacturer’s instructions. Samples were run on a Guava easyCyte 8HT flow cytometer (Millipore). Forward and side scatter gating were performed using FlowJo_V10.

Statistical analysis

Bioinformatics analyses were performed using R software. The varElect online tool was applied to analyze the correlation between genes in the PPI network and CRC. A high score indicated a strong correlation, with a significance threshold of P < 0.05 unless otherwise stated. All data are presented as the mean ± standard error (SE) of independent experiments. Two-tailed one-way analysis of variance (ANOVA) with multiple comparison post hoc analysis was used, and P values < 0.05 (*), P < 0.01 (**), P < 0.001 (***), and P < 0.0001 (****) are indicated as significant. Statistical analysis was performed using GraphPad Prism 9.0.