Background

Cervical cancer imposes a significant burden on global health, necessitating the exploration of innovative treatment approaches to improve patient outcomes.1 2 Despite advancements in conventional chemotherapy, treatment for cervical cancer still lacks effective and targeted interventions.3 In response to this problem, the immune checkpoint blockade has emerged as an encouraging treatment strategy. Immune checkpoint inhibitors for cervical cancer have been approved with clinical trials testing immunotherapies using different targets such as cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), TNF receptor superfamily member 9 (TNFRSF9), T cell immunoglobulin and mucin domain 3 (TIM-3), and programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1).4 5 PD-1 is the most specific target for immunotherapy of cervical cancer. Drugs inhibiting PD-1 include pembrolizumab, nivolumab, balstilimab, cemiplimab, and cadonilimab, all currently second-line treatments for recurrent/metastatic cervical cancer.6 However, over 50% of patients do not respond to PD-1 inhibitors.7 Therefore, new therapies or targets are urgently needed to improve outcomes.

Neoadjuvant chemotherapy (NACT) improves operability and subsequent treatment outcomes.8 Platinum-based NACT has a response rate of ∼60% in patients with locally advanced cervical cancer.9 The NACT drug cisplatin (CDDP) is the most widely used chemotherapeutic agent for gynecology malignancies and acts through cross-linking to interfere with DNA replication. An increasing body of evidence shows that combining chemotherapy and immunotherapy synergistically enhances anticancer effects in multiple solid tumors, including gastric and ovarian cancers. The mechanism is likely to involve increased lymphocyte infiltration and PD-L1 expression with the combination of chemotherapeutics and the enhanced recognition and elimination of tumors by the immune system.10 11 The efficacy of PD-L1 blockade combined with chemotherapy was tested in several clinical trials. For instance, a phase 3 trial of patients with recurrent cervical cancer revealed that survival was significantly longer under PD-1 antibody (cemiplimab) treatment than under single-agent chemotherapy.12 Therefore, National Comprehensive Cancer Network guidelines recommend PD-1 monoclonal antibody pembrolizumab plus chemotherapy as the first-line therapy for PD-L1-positive advanced cervical cancer. However, the effectiveness and necessity of the combination of PD-1 blockade therapy before surgical treatment has not been adequately demonstrated and is not widely used in cervical cancer. More evidence from preclinical studies and clinical trials is needed to support the use of PD-1 in neoadjuvant combination therapies.

Although this combination therapy is promising, the underlying molecular alterations remain largely unknown. One candidate for research on immune regulation of cancer is CD74, a non-polymorphic type II transmembrane glycoprotein responsible for antigen presentation, endocytic maturation, and cell migration.13 14 Furthermore, CD74 is a receptor for macrophage migration inhibitory factors (MIF) that induce changes in macrophage function and polarization,15 and its activation helps trigger the M2 shift in macrophages.16 Several recent studies have provided evidence of CD74 as a cancer prognostic factor and therapeutic target.1 17–19 However, its precise role in cervical cancer progression remains unclear.

In this study, we collected biopsy and surgical specimens from patients with locally advanced cervical cancer receiving neoadjuvant PD-1 blockade plus chemotherapy and who participated in a multicenter, prospective phase II trial (NCT04516616).20 Single-cell RNA sequencing (scRNA-seq) analysis was performed using specimens from patients before chemotherapy, after NACT, and after receiving combination therapy (NACT plus PD-1 blockade). We aimed to describe the transcriptomic evolution throughout the sequential stages of combination therapy and to compare tumor microenvironment (TME) at different stages. Through in vitro and in vivo experiments, we also intended to clarify the role of CD74 in NACT combination immunotherapy for cervical cancer. Our findings should provide a comprehensive, single-cell-resolution picture of cervical TME under combined NACT and PD-1 blockade.

MethodsPatients and sample collection

Nine specimens were collected from stage IB3 and IIA2 (FIGO 2018)21 female patients with locally advanced cervical cancer, who were treated at the Department of Gynecology and Obstetrics, Qilu Hospital, Shandong University. The patients had not undergone prior treatment before receiving platinum-based NACT combined with PD-1 blockade therapy and radical hysterectomy. The pretreatment (T1, T2, T3) and post-NACT (T1a, T2a, T3a) specimens were collected by puncture biopsy. The post-NACT combined with PD-1 blockade specimens (T2b, T2c, T3b) were collected during surgical resection. Pathologic type of tumor tissue determined to be moderately or poorly differentiated squamous cell carcinoma of the cervix. Patient T1 was HPV 16 (+), P16 (+), and Ki-67 proliferation index 80%. Patient T2 was HPV 16 (+), P16 (+), D2-40 (+), and CD31 (+). Patient T3 was P16 (+), P63 (+), P40 (+), CK7 (partially +), and Ki-67 proliferation index 50%. All patients provided written informed consent.

Tissue dissociation and preparation

The newly collected tissue was preserved on ice in sCelLiVE Tissue Preservation Solution (Singleron). The specimens were washed using Hanks’ Balanced Salt Solution and finely fragmented into small pieces. Singleron PythoN Tissue Dissociation System performed tissue digestion with sCelLiVE Tissue Dissociation Solution (Singleron). The cell suspension was filtered through a 40 µm sterile strainer. Mix two volumes of GEXSCOPE erythrocyte lysate (Singleron) and incubate for 5–8 min at room temperature. The mixture was then centrifuged and suspended in PBS (HyClone) for downstream library construction.

Single-cell library construction

Single-cell library construction was performed by Singleron Matrix single cell processing system according to the instructions of the GEXSCOPE Single Cell RNA Library Kits. Single-cell suspensions with a concentration of 2×105 cells/mL were loaded onto a microwell chip. Barcoding beads from the microwell chip were collected and used for reverse transcription and polymerase chain reaction (PCR) amplification. The complementary DNA (cDNA) was fragmented and ligated with sequencing adapters. Libraries were sequenced on the Illumina NovaSeq 6000 platform (PE150).

Primary analysis of the sequencing data

Gene expression profiles were generated from the raw reads utilizing CeleSCOPE (V.1.5.2, Singleron) with default parameters. The process involved extracting barcodes and unique molecular identifiers (UMIs) from R1 reads, followed by their correction. Adapter sequences and poly-A tails were trimmed from R2 reads, and the clean R2 reads were aligned to the GRCh38 (hg38) transcriptome using STAR (V.2.6.1a).22 Uniquely mapped reads were assigned to genes using FeatureCounts (V.2.0.1).23 Reads with the same cell barcode, UMI, and gene were grouped to generate the gene expression matrix for subsequent analysis.

Quality control, dimension-reduction, and clustering

Seurat (V.4.3.0) was used for quality control, dimensionality reduction, and clustering under R (V.4.3.1). For each sample dataset, we filtered the expression matrix by the following criteria: (1) cells with a gene count less than 200 or with a top 2.5% gene count were excluded; (2) cells with a top 2.5% UMI count were excluded; (3) cells with over 20% of UMIs derived from the mitochondrial genome were excluded; (4) genes expressed in less than five cells were excluded. After filtering, 46,950 cells were retained for the downstream analyses. SCTransfrom function in the Seurat R package was used to minimize the batch effects and normalization. 3000 highly variable genes were identified in each sample based on a variance-stabilizing transformation to generate an integrated expression matrix. Principal component analysis was performed on the scaled variable gene matrix, and the top 15 principal components were used for clustering and dimensional reduction. Cells were separated into 24 clusters using the Louvain algorithm, setting the resolution parameter at 0.8. Cell clusters were visualized by using Uniform Manifold Approximation and Projection (UMAP).

Cell type annotation and subtyping of epithelium and macrophages

The differential expressed genes of each cell subcluster were identified by the FindAllMarkers function of Seurat (V.4.3.0) with the parameters logfc.threshold = 0.25, test.use = “wilcox”, min.pct = 0.1, only.pos = T. Cell types were annotated by SingleR with manual correction and identification. Canonical cell type markers for single-cell sequencing data were achieved from CellMakerDB, PanglaoDB, and recently published studies. To obtain a high-resolution map of epithelial cancer cells and macrophages, cells from the specific cluster were extracted and reclustered for more detailed analysis following the same procedures described above and by setting the clustering resolution as 0.3. Cell double detection was performed using a cluster-level approach.24 Cells expressed both monocyte signature genes (CD14, FCGR3A) and cytotoxic signature genes (GZMA, CD3G, CD8B) were identified as potential monocyte-natural killer cell doublets.

Cell–cell interaction analysis

Cell–cell interaction analysis was performed with CellChat (V.1.5.0).25 Based on the previous study, we added 42 extra receptor-ligand pairs of immune checkpoint pathways into the CellChatDB.26 The gene expression matrix was projected to proteome–proteome interaction networks to reduce the dropout effect of signaling genes. Due to the completeness of the series of samples and the stability of the data quality, the T3 sample was selected for preliminary analysis and other samples for validation. The number and strength of the inferred interactions were analyzed between seven main cell types. The differential interaction strength was compared between the pretreatment and NACT groups, and between the NACT and the NACT+anti-PD-1 Ab groups. The relative interaction strength and information flow of the detailed receptor-ligand signal pathways were analyzed. Cell communication was also analyzed between six epithelial subgroups and seven macrophage subgroups. The epithelial and macrophage subgroups were merged to create a CellChat object and analyzed following the same process.

Pseudotime trajectory analysis

The differentiation trajectory of epithelial cancer cells and macrophage subtypes was reconstructed with Monocle 2 (V.2.24.1) and Monocle 3 (V.1.2.9).27 28 For constructing the trajectory, highly variable genes (q value<0.01) were selected by the differentialGeneTest function, and dimension-reduction was performed by DDRTree. The trajectory visualization was conducted using the plot_cell_trajectory function in Monocle 2. Monocle 3 was also used in pseudotime trajectory analysis. UMAP projection was imported from Seurat Object. The trajectory was created with the learn_graph function. The subpopulation of cells in which the pretreatment group predominates was chosen as the starting point for the pseudotime trajectory. The trajectory corresponding to UMAP projections was visualized by the plot_cells function in Monocle 3.

Functional annotation and enrichment analysis

Pathway enrichment in each epithelial cancer cell subgroup was performed with the irGSEA package (V.1.1.3) (https://github.com/chuiqin/irGSEA). The predefined sets of genes in the MSigDB database were used for analysis. The robust rank aggregation algorithm (RRA) integrated the enrichment scores from AUCell, UCell, singscore, and single-sample gene set enrichment analysis (ssGSEA) and was used for visualization.29 Pathway enrichment in each macrophage subgroup was performed with the clusterProfiler package (V.4.4.4).30 The top 100 marker genes in each subgroup were identified with the COSG package (V.0.9.0).31 Significant Reactome pathways (p value<0.05) were visualized with the dot plot function. Gene signatures of M1-like and M2-like macrophages used for ssGSEA enrichment were achieved from the previous study.32 UMAP projection of pathway enrichment and relative expression was visualized by Nebulosa.33

In vivo studies

C57bl-6j and BALB/cJGpt-Cd74em1Cin(hCD74)/Gpt female mice (6–8 weeks old) were purchased from GemPharmatech. A total of 2×105 cells were injected into the right flanks of mice. Anti-mouse PD-1 (CD279)-InVivo (10 mg/kg, every other day, Selleck) and CDDP (7 mg/kg, every 4 days, MedChemExpress) were injected into the abdominal cavity of mice to build an NACT combined with PD-1 therapy model. Milatuzumab (anti-CD74 Ab) (15 mg/kg, every other day, MedChemExpress) was used to block the expression of CD74. Mice in the control group were intraperitoneally injected with InVivoMAb polyclonal Armenian hamster IgG (10 mg/kg, every other day, Bio X Cell) and PBS. Then, after 12 days, the mice were euthanized, the tumors were dissected, and tumor weights were measured. Tumor sizes were measured using a Vernier caliper, and the tumor volume was calculated using the following formula: volume = 1/2 × length × width.2

Immunofluorescence staining

Multiple Immunofluorescence kits (ImmunoWay) are used for multicolor immunofluorescence staining. Deparaffinization, antigen retrieval, endogenous peroxidase removal, and blocking for non-specific binding were performed on tissue sections according to the instructions. The histological section and cells were incubated with primary antibody (online supplemental table 1). Subsequently, the samples were treated with an horseradish peroxidase (HRP) polymer secondary antibody (Anti-Rabbit/Mouse) and fluorescent chromogenic solution. Fluorescence images were taken using fluorescence microscopy and confocal microscopy. At least three fields of view per slide were quantified using ImageJ, and the average intensity density was compared with another group in the clinical specimen validation.

Cell culture and establishment of a coculture system

CaSki and THP-1 cells were cultured in 1640 medium (Gibco), and HeLa, SiHa, and TC-1 cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) (Gibco). The medium contains 10% fetal bovine serum (Gibco) and 1% Penicillin-Streptomycin (100×) (Solarbio), and cells were cultured at 37°C with a 5% CO2 atmosphere. SiHa was isolated from the primary uterine tissue of a patient with HPV-16-positive squamous cell carcinoma. CaSki is an epithelial cell isolated from the cervix of an HPV-16 and HPV-18 positive patient. HeLa was isolated from an HPV-18 positive patient with cervical cancer. TC-1 cells are mouse lung epithelial cells stably expressing HPV-16 E6/E7 and are commonly used to construct animal models of HPV-positive cancers.34

THP-1-derived macrophages were obtained after phorbol ester (Sigma) (50 ng/mL) treatment for 48 hours and validated CD11b and CD14 expression. Peripheral blood from patients with cervical cancer was collected, and human mononuclear cells were selected using gradient density centrifugation using Percoll gradients (Solarbio). EasySep Human CD14 Positive Selection Kit (STEMCELL) was used to select CD14+ monocytes. CD14+ monocytes were cultured in Iscove's Modified Dulbecco's Medium (IMDM) containing 50 ng/mL M-CSF for 7 days.

For the experiments in which anti-CD74 Ab (Milatuzumab, anti-CD74 Ab, 5 µg/mL, 12 hours, MedChemExpress) was combined with CDDP, adherent monolayers of 3×104 macrophages were directly overlaid with 1×105 cervical cancer cells. For the experiments in which anti-PD-1 antibody was combined with CDDP, both direct and indirect cocultures were performed. Using Transwell inserts (0.4 µm, Corning), 1×104 cervical cancer cells were cultured in the lower part and 5×103 macrophages in the upper part of each well. CDDP (30 µM, 4 hours, MedChemExpress) and pembrolizumab (anti-PD-1 Ab) (20 µg/mL, 24 hours, MedChemExpress) were used to act on a coculture system to build an NACT combined with PD-1 therapy model.

Flow cytometry

THP-1-derived macrophages, primary monocyte-derived macrophages, and single-cell suspension from the mouse xenograft tumor model were collected and evaluated by flow cytometry. Single-cell suspension from tumors or spleens was stained with fluorochrome-labeled antibodies. In the CD45-positive cell population, the macrophages expressing F4/80 and CD11b were gated to detect PD-1 and CD74 expression. THP-1-derived macrophages were stained with fluorochrome-labeled antibodies detecting CD206, CD163, CD74, and CD279. Detailed antibody information is presented in online supplemental table 1.

The apoptotic cells were detected using the Annexin V-FITC Apoptosis Detection Kit (Beyotime). Cells were harvested and resuspended in the Annexin V-FITC and PI binding buffer. The total percentages of the groups containing early and late apoptosis were compared.

The stained samples were detected using CytoFLEX, and the data were analyzed using CytExpert software. All staining was performed according to the manufacturer’s protocols. The control group without antibodies and staining was used for gating, and single-color stain controls were used to enable correct compensation.

Small interfering RNA and transfection

Small interfering RNA (siRNA) targeting CD74 or scramble sequences were purchased from Research Cloud Biology. siRNA and plasmids were transfected into macrophages derived from THP-1 with Lipofectamine 3000 (Invitrogen) according to the manufacturer’s instructions. 80 pmol of siRNA or 2 µg of plasmid was transfected in each well of a 6-well plate. The siRNA sequences are presented in online supplemental table 2.

RNA extraction and real-time quantitative reverse transcription PCR (qRT-PCR)

TRIzol (LIFE Ambion) was used to extract the total RNA of tissues and cells. Total RNA was reverse transcribed into cDNA using the HiScript III RT SuperMix for qPCR (Vazyme, R223-01). Then, we amplified the aimed gene fragment and detected the relative expression with the SYBR Green qPCR kit (TOYOBO). Quantitative real-time PCR was performed for 40 cycles. All experiments were conducted at least three times. Relative gene expression levels were calculated using the 2–∆∆Ct method. Primer sequences are presented in online supplemental table 2.

Phagocytosis assay

Macrophages derived from THP-1 and CaSki cells were digested using Trypsin (Macgene) and stained using the Cell Plasma Membrane Staining Kit with Dil (Red Fluorescence) or DiO (Green Fluorescence) (Beyotime), respectively. The macrophages were plated at a density of 5×104 cells per well in a 24-well tissue-culture plate, and 2×105 cancer cells were used for staining per experiment. After 3 hours coculturation in serum-free medium, the cocultured cells were analyzed by flow cytometer and imaged by a fluorescence microscope. The control group without antibodies and staining was used for gating, and single-color stain controls were used to enable correct compensation.

Cell migration and proliferation assays

Standard transwell inserts (0.8 µm, Corning) were used to detect cell migration. Cells were plated into the upper compartment with 500 µL of serum-free medium, and the lower compartment was filled with medium containing 10% fetal bovine serum. After incubation for 24–48 hours at 37°C, cotton swabs were used to remove the non-invaded cells on the filter’s upper surface, and the cells were fixed for 2 min in 100% methanol. Invaded cells on the lower side of the filter were stained with 0.5% crystal violet for 20 min. Cells were counted at ×100 magnification from three random microscopic fields for each sample in three independent experiments. The counting process was performed in an observer-blind manner.

The CCK-8 (Cell Counting Kit-8) assay kit (Beyotime) was used to evaluate the level of cell proliferation. Briefly, the 10 µL Cell Counting Kit solution was added to the culture medium and incubated for 3 hours. The absorbance was determined at a wavelength of 450 nm.

ResultsCellular dynamics and TME heterogeneity in cervical cancer receiving NACT and PD-1 blockade therapy

Using puncture biopsy or surgery, we collected nine tissue samples from three patients with cervical cancer. T1, T2, and T3 patients underwent neoadjuvant therapy before surgery. We collected samples before (T1, T2, T3) and after (T1a, T2a, T3a) NACT treatment for all three patients, as well as samples after receiving PD-1 blockade therapy (T2b, T2c, T3b) from the patients T2 and T3 (figure 1A). We normalized transcriptome expression profiles from 46,950 cells. After annotation, 24 cell groups were identified and combined into seven clusters: cancer cells, mesenchymal stem cells, fibroblast, endothelial cells, T cells, myeloid cells, and B cells (figure 1B and online supplemental figure S1). Their relative cell ratios and UMAP projection of cells from the nine samples revealed that the proportion of cancer cells decreased after combined NACT and PD-1 antibody treatment (figure 1C and D). Additionally, immune cell proportions were heterogeneous and dynamic in the TME (figure 1E and F). Finally, B cell proportions increased significantly when looking across different treatment stages. Heatmap visualization showed that the top marker genes of each cell type were clustered together (figure 1G). These included EPCAM, CDH1, and KRT8 for epithelial cells,35 as well as HSPB6, SERPINI1, and THY1 for mesenchymal stem cells.36 We also observed specific expression of marker genes for fibroblast (DCN), endothelial cells (VWF), T cells (CD3D), myeloid cells (CD163 and CD68), and B cells (CD79A) (figure 1H–J).

Figure 1

Study design and single-cell transcriptomic analysis identifying diverse cell types in cervical cancer under neoadjuvant chemotherapy (NACT) and programmed cell death protein 1 (PD-1) antibody treatment. (A) Overall flowchart of study (created with BioRender.com): sample collection, single-cell RNA sequencing library construction, bioinformatics analysis, and experimental validation. (B) The Uniform Manifold Approximation and Projection (UMAP) plot illustrates 7 cell types in cervical cancer tissues (n=9). (C) UMAP projection with each sample colored. (D) Stack bar plot summarizing proportions of assigned cell types per sample. (E) UMAP projection with colors delineating different stages of NACT combined with anti-PD-1 Ab (n=3 per group). (F) Stack bar plot summarizes cell type proportion in samples from different stages of combination therapy (n=3 per group). (G) Heatmap showing marker gene expression in each cell type. (H) Dot plot depicting average expression and expression percentage of marker genes for 7 cell types. (I) Violin plots display representative marker expression across the cell types identified in cervical cancer. The p value was obtained by ordinary one-way analysis of variance. (J) UMAP plots of marker gene expression for cell type identification. The legend shows a color gradient of the normalized read count. Ab, antibody; CDDP, cisplatin; cDNA, complementary DNA; MSC, mesenchymal stem cell; PBS, phosphate buffered saline.

Transcriptional evolution of cervical cancer cells during combination therapy

Epithelial cells were grouped into six subclusters to assess the impact after combination therapy (figure 2A). In post-NACT groups, particularly T3 patients, epithelial subgroup (Epi) 1 proportion decreased while Epi3 proportion increased (figure 2B and online supplemental figure S2A). Evaluation of activated pathways using the RRA algorithm revealed that Epi4 was enriched in the mitotic spindle, E2F targets, and adipogenesis pathways. The immunoreactive Epi3 exhibited upregulation of interferon (IFN)-gamma, IFN-alpha, and complement pathways. Lastly, Epi2 was associated with hypoxia, angiogenesis, apoptosis, and inflammatory response (figure 2C). Pathway enrichment results were consistent with the UMAP distribution of corresponding subclusters (figure 2D and online supplemental figure S2B).

Figure 2

Pseudotime analysis of cervical cancer cells receiving combination therapy. (A) UMAP projection of 6 subgroups generated from clustering epithelial cancer cells. (B) Stack bar plot summarizing proportions of epithelial subgroups in each sample. (C) Pathway enrichment using the RRA algorithm showing the top 100 genes specific to each epithelial subgroup. The statistical test used in pathway enrichment analysis is the hypergeometric test. (D) UMAP projection of mitotic spindle and IFN-alpha response-pathway density. (E) Monocle analysis of unsupervised transcriptional trajectory in epithelial cells, color-coded based on pseudotime. (F) Monocle inference of epithelial cell pseudotime. Colors reflect different stages of combination therapy (n=3 per group). The arrows indicate the direction of the pseudotime. (G) Heatmap showing differential gene expression in each state along the pseudotime of combination therapy, clustered into four groups according to expression pattern. Marker gene-associated pathways are labeled on the right side with font colors consistent with (F). (H) Unsupervised transcriptional trajectory in epithelial cells, color-coded based on pseudotime stages. (I) IFN-alpha response pathway enrichment score of three pseudotime stages. The p value was obtained using a two-tailed unpaired Student’s t-test, and the results are presented in a boxplot (10th–90th percentile). ***p<0.001. Ab, antibody; Epi, epithelial subgroup; IFN, interferon; NACT, neoadjuvant chemotherapy; PD-1, programmed cell death protein 1; RRA, robust rank aggregation; UMAP, Uniform Manifold Approximation and Projection.

Pseudotime analysis revealed the timing of combination therapy correlated with pseudotime, with pretreatment dominant at the start, while NACT and anti-PD-1 Ab groups were dominant at the end (figure 2E–F). Mitosis-related gene expression (CDKN3, HMGB2, TOP2A, and CCNB1) decreased over the pseudotime, while immune-related genes (CD274, CD36, TNFSF10, CD69, TGFBI, and IFI27) were upregulated at later stages of the pseudotime time series. Likewise, ESR1, AGR3, APX8, VIM, MMP7, IGFBP8, XBP1, CD55, and MUC16 were upregulated at the end (figure 2G). Epithelial cells were divided into three states based on pseudotime, and the IFN-alpha response pathway was enriched in post-nonadjacent therapy-specific state 2 (figure 2H–I and online supplemental figure S2C). Monocle 3 analysis was also used and showed the evolutionary processes start from Epi4, which was related to the mitotic spindle; the final destination of cell fate is the immunoreactive Epi3 (online supplemental figure S2D–E). Overall, NACT and anti-PD-1 Ab increased immune response in epithelial cells.

Tumor cell-immune cell communication dynamically responds to combination therapy

Cell communication analysis of seven cell clusters demonstrated that the total number and strength of inferred interactions between tumor and immune cells increased after NACT treatment, suggesting that NACT activates the immune response. However, PD-1 blockade combination treatment decreased these interactions (figure 3A–D). Cell type level communication analysis revealed cancer cell interactions with T cells were weakened after NACT. However, the interaction strength between cancer and myeloid cells increased (figure 3E–G and online supplemental figure S3A). Anti-PD-1 antibody combination treatment then silenced cancer cell-myeloid cell communication induced by NACT (figure 3H).

Figure 3

Cell communication analysis of tumor microenvironment after receiving NACT and PD-1 antibody treatment. (A) Number of significant ligand-receptor pairs between any two cell populations. Edge width is proportional to the indicated interaction number of ligand-receptor pairs. (B, C) Number and strength of inferred interactions among seven cell types in the pretreatment, NACT, combination therapy (NACT+anti-PD-1 Ab) group. (D) Quantification results of the summed intensity of interactions for each cell type in the pretreatment, NACT, and combination therapy (NACT+anti-PD-1 Ab) groups. The p value was obtained using multiple comparisons in an ordinary one-way ANOVA with cancer cells. The solid and dashed lines in the violin plot represent the median and IQR. (E, F) Quantification results of the strength of interaction between cancer (F) or myeloid cells (E) and other cells in different stages of combination therapy. The p value was obtained using a two-tailed unpaired Mann-Whitney test, the results are presented as the mean±SEM. (G, H) Differential interaction strength between any two cell populations before and after NACT only (G) and combination therapy (H). Edge width and heatmap shade are proportional to normalized interaction strength. (I) Heatmap shows the relative interaction strength of 38 significant ligand-receptor signaling pathways belonging to each cell type. Left, pretreatment group. Right, NACT group. (J) Significant signaling pathways were ranked based on differences in overall information flow within inferred networks between NACT and combination therapy groups. (K) Strength of myeloid cell interactions with other cells during different stages of combination therapy within the immune checkpoint pathway. The p value was obtained by a two-tailed paired Wilcoxon test. The bounds of the box were the upper and lower quartile, with the median value in the center. The whiskers indicated the minima and maxima. *p<0.05, **p<0.01, ***p<0.001. Ab, antibody; ANOVA, analysis of variance; MSC, mesenchymal stem cell; NACT, neoadjuvant chemotherapy; PD-1, programmed cell death protein 1.

We subsequently determined the pathways and gene families responsible for these global communication changes. First, we found that major histocompatibility complex (MHC)-I and immune-checkpoint signaling pathways for myeloid cell communication were significantly upregulated (figure 3I). However, immune-checkpoint interaction strength diminished after anti-PD-1 antibody combination therapy (figure 3J–K). As immune checkpoint regulators, both PD-1 and PD-L1 bind ligands to participate in immune escape of cervical cancer. Although PD-1 was not detected in scRNA-seq data, we found PD-1 ligand PD-L1 (CD274) expression was upregulated in immunoreactive Epi2 and Epi3 cancer subgroups after NACT treatment (online supplemental figure S3B–E). Pseudotime analysis confirmed that PD-L1 was expressed during the mid-to-late temporal stage (online supplemental figure S3F). In summary, NACT treatment activated the immune checkpoint pathway between tumor cells and immune cells.

The therapeutic effect of anti-PD-1 antibody in combination with platinum-based agent in cervical cancer partly depends on macrophages

We verified the combined effect of the anti-PD-1 antibody and platinum-based agent on macrophage-tumor cell communication in vitro. We first treated cervical cancer cells (SiHa and HeLa) alone with IgG, platinum-based agent CDDP, anti-PD-1 antibody, or combined therapy (CDDP plus anti-PD-1 antibody) in the absence of macrophages. The percentage of apoptotic cells included late (PI+/Annexin-V+) and early (PI−/Annexin-V+) apoptotic cells. CDDP treatment increased the percentage of apoptotic cells (online supplemental figure S4A–B), and cell viability and migration ability decreased (online supplemental figure S4C–E). However, anti-PD-1 antibody did not directly affect cervical cancer cells, and the combined treatment did not show any difference.

When cervical cancer cells were directly or indirectly cocultured with THP-1-derived and primary monocyte-derived macrophages, the combination therapy increased the number of apoptotic cells compared with CDDP alone (figure 4A–B, online supplemental figure S5A–B, online supplemental figure S5G–H), while also decreasing cell viability and migration (figure 4C–E, online supplemental figure S5C–E). Treatment of cocultured THP-1-derived or primary monocyte-derived macrophages with cervical cancer cells indicated that CDDP inhibited macrophage phagocytosis (figure 4F–G and online supplemental figure S5F). We also constructed a subcutaneous xenograft model with mice TC-1 cells in immunocompetent mice. The results showed that anti-PD-1 antibody increased the inhibitory effect of CDDP on tumor growth (figure 4H–I). Besides, both in vitro and in vivo data demonstrated that CDDP did not increase PD-1 expression, whereas anti-PD-1 Ab treatment decreased PD-1 expression (figure 4J, online supplemental figure S5I–K).

Figure 4

The therapeutic effect of combining anti-PD-1 antibody and CDDP in cervical cancer is dependent on macrophages. (A, C) Percentage of apoptotic cells (A) and cell viability (C) of direct coculture of HeLa cells and primary monocyte-derived macrophages treated with anti-PD-1 Ab and/or CDDP. The upper right and lower right quadrants indicate the percentage of late and early apoptotic cells in the measured cell population. Apoptotic cells percentage included both early and late apoptotic cells. (B, D) Percentage of apoptotic cells (B) and cell viability (D) of indirect coculture of SiHa cells and THP-1-derived macrophages treated with anti-PD-1 Ab and/or CDDP. (E) Transwell migration assay of cervical cancer cells cocultured with THP-1-derived macrophages. The number of purple-stained cells indicates migration capacity. (F) Phagocytosis assay on macrophages derived from THP-1 cocultured with SiHa. Phagocytosis efficiency was quantified as the percentage of Dil and DiO double fluorophore-positive THP-1-derived macrophages. (G) Phagocytosis assay on primary monocyte-derived macrophages cocultured with CaSki cells. (H) Representative image of the subcutaneous tumors. TC-1 cells were subcutaneously inoculated into C57BL/6J mice. Each group (n=5) was treated with CDDP and/or anti-PD-1 Ab separately. (I) Growth curve of subcutaneous tumors in mice (n=5). (J) The PD-1 expression of macrophages isolated from the subcutaneous tumor microenvironment of mice was detected by flow cytometry (n=5). All functional experiments were performed with n=3 biological replicates unless otherwise stated. The p value was obtained using a two-tailed unpaired Student’s t-test (A–G, I–J), and the results are presented as the mean±SD. *p<0.05, **p<0.01, ***p<0.001. Ab, antibody; CDDP, cisplatin; PD-1, programmed cell death protein 1.

Distinct subgroups of myeloid cells and their communication with epithelial cancer cell subgroups in cervical cancer

Myeloid cell-cancer cell communication changed across the duration of our treatments. We determined how combination therapy affected the seven myeloid cell subgroups (M1–M7) (figure 5A). Based on specific marker genes, seven myeloid cell subgroups were categorized as macrophages (M1, M2, M4, M5), monocytes (M3), DC cells (M4), and a potentially doublet subgroup (M7) (figure 5B and online supplemental figure S6A). The proportions of M3 and M5 subgroups increased slightly after NACT treatment (figure 5C). M1 was associated with complement-associated immune responses, neutrophil degranulation, and M2 with interleukin signaling, whereas M5 was related to mitosis (online supplemental figure S6B).