Cytokine mRNA-based therapy alleviates dendritic cell and T cell paucity to eliminate aggressive pancreatic cancer in preclinical mouse models

Summary

Background

Pancreatic ductal adenocarcinoma (PDAC) with peritoneal dissemination is highly refractory to chemotherapy and immunotherapy, leading to poor prognosis. We aimed to develop an innovative therapeutic approach for advanced PDAC.

Methods

We performed comprehensive analyses of 498 bulk and 99 single-cell RNA-sequencing datasets. We established a syngeneic mouse model for subcutaneous and intraperitoneal metastatic tumours using mouse Kras^G12D; Trp53^R172H PDAC cells. A multimodal immunotherapy with mRNA-induced cytokines (MIMIC), that is, oxaliplatin, anti-PD-1 and anti-CTLA-4 antibodies, and intratumoural administration of mRNA therapeutics encoding interferon-α and interleukin-12, was evaluated in this preclinical model.

Findings

The aggressive PDAC subtype exhibited a paucity of dendritic cells (DCs) and T cells, causing an immunosuppressive tumour microenvironment. The syngeneic mouse model recapitulated this immunological phenotype with resistance to conventional systemic therapies. The MIMIC therapy not only significantly reduced the local tumour burden but also elicited a robust abscopal effect, suppressing distant peritoneal metastases and prolonging survival (P < 0.001). The omission of any single agent from the MIMIC regimen substantially abrogated the therapeutic efficacy. Flow cytometry and immunohistochemical analyses revealed that the MIMIC treatment enhanced immunogenic cell death, increased peripheral CD44+ CD62L− effector memory T cells, induced intratumoural infiltration of CD11c+ DCs and CD8+ T cells, and expanded TCR repertoire diversity.

Interpretation

Combining cytokine mRNA immunotherapy with cytotoxic killing and immune checkpoint blockade can reactivate antitumour immunity, offering a promising strategy for treating advanced PDAC.

Funding

This work was supported by Ministry of Education, Culture, Sports, Science and Technology of Japan (MEXT), Japan Agency for Medical Research and Development (AMED), and the Princess Takamatsu Cancer Research Fund.

Research in context.

Evidence before this study

Pancreatic cancer is an extremely aggressive disease with unfavourable prognosis, particularly in cases with peritoneal dissemination, which severely impairs quality of life. Standard chemotherapeutic agents and immune checkpoint inhibitors have shown minimal efficacy, and emerging immunotherapies have yet to achieve clinical success, likely due to the immunosuppressive phenotype of pancreatic cancer. Yet, its detailed immune landscape remains poorly understood.

Added value of this study

This study elucidates that the aggressive subtype of pancreatic cancer exhibits a marked paucity of dendritic cells and T cells, contributing to its immunosuppressive milieu and resistance to conventional immunotherapies. Using a syngeneic mouse model bearing both subcutaneous and peritoneal tumours that reflect this immune phenotype, we reveal that a multimodal immunotherapy with mRNA-induced cytokines, termed as MIMIC, comprising oxaliplatin, anti-PD-1 and anti-CTLA-4 antibodies, and intratumoural administration of mRNA therapeutics encoding interferon-α and interleukin-12, effectively restores antitumour immune responses. This therapeutic approach not only reduces primary tumour growth but also represses peritoneal dissemination via abscopal effects, ultimately prolonging overall survival.

Implications of all the available evidence

This study highlights the therapeutic potential of combining cytokine mRNA-based therapy with standard treatments to overcome immune resistance in pancreatic cancer. The MIMIC combination therapy demonstrates significant antitumour and abscopal effects in a relevant preclinical model, providing a strong rationale for translating this multimodal strategy into early-phase clinical trials. These findings may guide future research, clinical practice, and health policy toward the implementation of mRNA-based immunomodulation in the treatment paradigm for aggressive, immune-cold malignancies.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is an exceptionally lethal malignancy with extremely poor prognosis, and the five-year survival rate remains dismally low at approximately 10%,¹ despite progress in multimodal therapies including surgery, chemotherapy, radiotherapy, and immunotherapy. Moreover, 25%–50% of patients with PDAC suffer from peritoneal metastases irrespective of surgical intervention or systemic treatment,² which causes treatment resistance and produces severe symptoms such as abdominal distension and pain, critically impacting patients’ quality of life. Currently, there are no robust therapeutic options for PDAC with peritoneal dissemination, highlighting the urgent need for novel treatment approaches.

While conventional chemotherapies have demonstrated only limited efficacy in improving patient survival, their combination with immune checkpoint inhibitors (ICIs) has also failed to offer additional clinical benefits in patient outcomes.^3,4 Although next-generation immunotherapies for PDAC have emerged, including the whole-cell immunomodulator composed of allogeneic pancreatic cancer cells, algenpantucel-L,⁵ the telomerase peptide vaccine GV1001,⁶ and the CD40 agonistic antibody sotigalimab,⁷ their potential remains unproven.

To develop effective treatment strategies for refractory neoplasms, it is crucial to stratify tumours into subtypes, clarify the molecular mechanisms and immune profiles of each subtype, and identify precision medicine accordingly.^8,9 Collisson et al. categorised PDAC into three molecular subtypes: quasi-mesenchymal (QM), Classical, and Exocrine-like.¹⁰ Subsequently, Bailey et al. proposed an additional classification by incorporating immunological features, which was composed of the Squamous, Progenitor, aberrantly differentiated endocrine exocrine (ADEX), and Immunogenic subtypes.¹¹ However, the immunological characteristics of these subtypes remain incompletely understood.¹² Recent advances in single-cell RNA sequencing (scRNA-seq) technologies have enabled detailed characterisation of the cellular composition within tumour tissues, and several laboratories have performed scRNA-seq analyses of PDAC samples, providing insights into the immunological classification of PDAC.¹³ However, the relationship between molecular subtypes and immune microenvironments has yet to be fully elucidated.

In this study, we integrated 498 bulk RNA-seq and 99 scRNA-seq datasets of PDAC specimens to define the immune profiles of distinct molecular subtypes. Our findings suggested that the aggressive subtype exhibited defects in dendritic cell (DC) activation and T cell priming, resulting in a “cold” tumour phenotype. Furthermore, we established a syngeneic mouse model of peritoneal dissemination, which recapitulated the immunosuppressive trait of this PDAC subtype with resistant to the combination of chemotherapy and immune checkpoint blockade. Instead, we discovered mRNA-based cytokine therapy as a promising therapeutic approach to directly invigorate DCs and T lymphocytes, thereby reversing immune dysfunction in PDAC.

Methods

Ethics

This study was conducted ethically in accordance with the World Medical Association Declaration of Helsinki and the ARRIVE guidelines. All experimental protocols were reviewed and approved by Institutional Review Board (G2017-018, Medical Research Ethics Committee for Life Science of the Institute of Science Tokyo; G2023-084A, Medical Research Ethics Committee for Genetic Research of the Institute of Science Tokyo; A2023-003C3, Institutional Animal Care and Use Committee of the Institute of Science Tokyo). Written informed consent was obtained from participants.

Bulk transcriptome analysis

Raw count data of bulk RNA-seq analyses from the PAAD-US, PACA-AU, and PACA-CA cohorts were previously downloaded via the ICGC Data Portal. The gene expression data were read, normalised and log-transformed using Scanpy. After a total of 4000 highly variable genes were selected using “seurat_v3” algorithm and scaled, principal component analysis and Harmony integration were conducted. We utilised uniform manifold approximation and projection (UMAP) for visualisation and the Leiden algorithm with a resolution of 0.1 for clustering analysis. Single-sample gene set enrichment analysis (ssGSEA) was performed with the MSigDB gene sets using the scGSVA package in R.

Single-cell transcriptome analysis

Raw count data of scRNA-seq analyses from CRA001160, GSE154778, GSE155698, GSE156405, GSE205013, GSE205049, GSE212966, GSE214295, GSE229413, and GSE242230 were downloaded via the Genome Sequence Archive and the Gene Expression Omnibus. However, GSE205013 and GSE229413 were excluded, because the former was derived from biopsy samples and contained an excess of immune cells, and the latter had a high number of dead cells. The gene expression data were read, filtered with min_genes = 200 and min_cells = 5, normalised with target_sum = 1e4 and then log-transformed using Scanpy. Cell types were predicted using the CellTypist program with the “Immune_ALL_Low.pkl” model. For quality control, cells were excluded if any of the three metrics, that is, log1p_n_genes_by_counts (the log-transformed number of genes expressed in the count matrix), log1p_total_counts (the log-transformed total counts per cell) and pct_counts_mt (the percentage of counts of mitochondrial genes) exceeded 5 mean absolute deviations. After a total of 4000 highly variable genes were selected using “seurat_v3” algorithm and scaled, principal component analysis and Harmony integration were conducted. We utilised UMAP for visualisation and the Leiden algorithm with a resolution of 0.4 for clustering analysis. Cell components were scaled and clustered using the Euclidean distance and the Ward linkage prior to the generation of a heatmap with dendrograms.

Pseudo-bulk transcriptome analysis

Pseudo-bulk datasets were generated from single-cell datasets by calculating the mean expression level of each of 4000 genes used for Harmony integration across all cells in each case. The pseudo-bulk datasets were integrated with the bulk datasets based on 663 shared genes using the Harmony program. We utilised UMAP for visualisation and the Leiden algorithm with a resolution of 0.2 for clustering analysis.

Human tissue samples

A total of 112 Japanese patients underwent curative resection for PDAC at Tokyo Medical and Dental University Hospital (now known as Institute of Science Tokyo Hospital) between 2013 and 2016. All the patients provided informed consent before enrolment and were anonymously coded in accordance with ethical guidelines. The clinicopathological classification of PDAC was determined following the TNM classification system defined by the Union for International Cancer Control.

Cell lines

The FC1245 and FC1242 cell lines were originally isolated from PDAC tissues of Pdx1-Cre⁺;Kras^LSL−G12D/+; Trp53^{LSL−R172H/+} mice on a C57BL/6 genetic background¹⁴ and generously provided by Prof. David Tuveson, with the collaboration of Prof. Yoshiya Kawaguchi. Cells were cultured in DMEM medium (Wako, Osaka, Japan) supplemented with 5% foetal bovine serum and maintained in a humidified incubator at 37 °C with 5% CO₂ and harvested using 0.05% trypsin/0.03% EDTA solution (Wako). All cell lines used in this study were routinely tested for mycoplasma contamination and confirmed to be negative.

Preparation of mRNA therapeutics

Sequences of the T7 promoter and protein coding regions of mouse IFNα11 (RefSeq: NM_008333.2), IL12α (RefSeq: NM_001159424.2), and IL12β (RefSeq: NM_001303244.1) were inserted into the pUC57 plasmid DNA vector (Genscript, Tokyo, Japan). The resulting plasmid DNA was subjected to polymerase chain reaction to prepare template DNA using a primer set, with a reverse primer containing an 80-nt poly T sequence. Subsequently, in vitro transcription was then performed using the mMessage mMachine T7 kit (Thermo Fisher Scientific, Waltham, MA, USA). Lipid nanoparticles based on ALC-0315 were prepared as previously described.¹⁵

Mouse tumour models

Only male C57BL/6 mice, aged 4–6 weeks, were used to minimise the potential influence of oestrogen on oncogenic and immunological processes. Mice were housed in a temperature-controlled environment with ad libitum access to sterile water and food, under a 12-h light/dark cycle. Before tumour cell implantation, all mice were acclimated for at least 3 days. Mice were anesthetised with 2.5% isoflurane during the procedure. A total of 1 × 10⁵ cells of FC1245 and 3 × 10⁵ cells of FC1242 were suspended in 100 μL of phosphate-buffered saline and 100 μL of Matrigel (BD Biosciences, San Jose, CA, USA) and subcutaneously transplanted into C57BL/6 mice on Day 0. Three days later (Day 3), 1 × 10⁵ cells of FC1245 and 3 × 10⁵ cells of FC1242 were suspended in 200 μL of phosphate-buffered saline and intraperitoneally inoculated to establish a peritoneal dissemination model. Cytokine mRNA-loaded lipid nanoparticle (LNP) or luciferase mRNA-LNP as a negative control was prepared in a volume of 50 μL and directly injected into the subcutaneous tumour on Day 5, 8, 11, and 14. Oxaliplatin (Tokyo Chemical Industry, Tokyo, Japan) was intraperitoneally administered as a single dose of 2 mg/kg on Day 5. Both of anti-PD-1 antibody (αPD-1) (J43; Bio X Cell, West Lebanon, NH, USA; RRID: AB_1107747) and anti-CTLA-1 antibody (αCTLA-4) (9D9; Bio X Cell; RRID: AB_10949609) were intraperitoneally administered at a dose of 100 μg/head on Day 5 and 11. Anti-CD3ε (200 μg/head; 145-2C11; Bio X cell; RRID: AB_1107634) and anti-NK1.1 (100 μg/head; PK136; Bio X Cell; RRID: AB_1107737) antibodies were intraperitoneally administered on Day −1, followed by additional dose every 3 days. Oral treatment of FTY720 (Tokyo Chemical Industry) was performed at 25 μg/head on Day 3, followed by 5 μg/head daily, as previously described.¹⁶ For chemotherapeutic treatments, mice received intraperitoneal injections on Day 5 after intraperitoneal tumour implantation: gemcitabine at 25 mg/kg, followed by additional doses every 7 days, or irinotecan as a single dose of 50 mg/kg. Body weight and tumour volume were measured at least twice weekly. Tumour volume was calculated using the formula: A × B × (A + B)/2, where A and B represent the longitudinal and transverse diameters, respectively. All intraperitoneal metastatic tumours with a major axis greater than 2 mm were collected, and the tumour count, volume, and total weights were measured. A total of 313 mice were used in this study; 306 completed the planned procedures and were included in the analyses. Seven mice (2.2%) died perioperatively due to procedure-related complications (e.g., anaesthesia or haemorrhage) and were excluded from subsequent analyses according to pre-specified criteria.

Flow cytometric analysis

Peripheral blood was obtained 17 days after subcutaneous tumour injection into C57BL/6 mice. Red blood cells were lysed using Red Blood Cell Lysis Buffer (Puliselect GmbH, Leipzig, Germany). The remaining cells were then neutralised in a medium containing foetal bovine serum, filtered through a 35-μm cell strainer (Corning, Corning, NY, USA) into 5 mL tubes. The cells were incubated with fluorophore-conjugated antibodies listed in Supplementary Table S1 at 4 °C in the dark for 30 min, washed, and subsequently analysed. Fluorescence intensity was measured using a FACSLyric flow cytometer (BD Biosciences).

Immunohistochemical analysis

Pancreatic cancer tissues resected from human patients and subcutaneous and intraperitoneal metastatic tumour tissues harvested from C57BL/6 mice were fixed overnight in Mildform 20N (Wako), embedded in paraffin, and sectioned at a thickness of 4 μm. The sections were immersed in sodium citrate buffer (pH 6.0) or 1 mM EDTA buffer (pH 8.0) for antigen retrieval and then incubated overnight at 4 °C with primary antibodies as listed in Supplementary Table S2. Sections were probed with horseradish peroxidase-labelled anti-mouse or anti-rabbit IgG antibody (Histofine Simple Stain MAX-PO, Nichirei Bioscience, Tokyo, Japan; RRID: AB_2819094) and visualised using diaminobenzidine (Wako), followed by nuclear counterstaining with haematoxylin. The number of stained cells and the area of stained regions were evaluated by randomly capturing images at 200× or 100× magnification in three different fields for human tissues or six different fields for mouse tissues, excluding necrotic areas, and automatically performing quantification with Fiji software.

Immunofluorescent analysis

Multiplex immunofluorescent staining was carried out utilising FlexAble CoraLite Plus 488, 555, and 647 Antibody Labelling Kits for rabbit IgG (Proteintech, Rosemont, IL, USA; RRID: AB_3095334 [488], AB_3095335 [555], and N/A [647]), according to the manufacturer's protocols. In brief, rabbit IgG antibodies were labelled with the fluorophores CoraLite 488, 555, and 647 using the labelling kits. The tissue sections were submerged in sodium citrate buffer (pH 6.0) for antigen retrieval, and then incubated at 4 °C for 4 h with the fluorophore-conjugated antibodies. The sections were then counterstained and mounted using ProLong Gold Antifade Mountant (Thermo Fisher Scientific), and the slides were examined using an Axio Observer Z1 fluorescent microscope (Carl Zeiss, Oberkochen, Germany) and a BZ-X700 fluorescence microscope (Keyence, Osaka, Japan).

Enzyme-linked immunosorbent assay (ELISA)

Four hours after intratumoural administration of cytokine mRNA therapeutics, subcutaneous tumours and serum were collected. Tumour specimens were minced into 2-mm fragments and homogenised in 100 μL of RIPA buffer (Thermo Fisher Scientific). Levels of IFNα and IL12 in tumour lysates and serum were quantified using Mouse IFN-Alpha All Subtype ELISA Kit (PBL Assay Science, Piscataway, NJ, USA) and Mouse IL-12 ELISA Kit (Invitrogen, Carlsbad, CA, USA), respectively, according to the manufacturer's instructions. Absorbance was measured at 450 nm using an iMark microplate reader (Bio-Rad, Hercules, CA, USA), and cytokine concentrations were calculated from standard curves.

On day 17, the peritoneal cavity of tumour-bearing mice was lavaged with 5 mL of PBS, and the lavage fluid was collected. The HMGB1 concentration in the lavage fluid was quantified using HMGB1 ELISA kit (Arigo Biolaboratories, Hsinchu, Taiwan), and the total amount of HMGB1 in the peritoneal cavity was calculated based on the measured concentration and the lavage volume.

TCR repertoire analysis

Subcutaneously transplanted tumours were excised from mice on Day 17, and total RNA was isolated from the tumour tissue. Amplification of the V and J regions, which are the variable regions of the TCR gene, was performed, followed by next-generation sequencing to assess the TCR repertoire (Azenta Life Science, Burlington, MA, USA). Using the Trim Galore program, quality control of raw sequencing data was conducted, and adaptor sequences and low-quality bases were removed. TRA and TRB repertoires were extracted from FASTQ files using the MiXCR program. To visualise V-J gene pairing, Manhattan plots, heatmaps, and circos plots were generated by the ggplot 2 and plotly packages in R.

Statistics

Statistical analyses were performed using Python (Python Software Foundation, Beaverton, OR, USA), R (R Foundation for Statistical Computing, Vienna, Austria), EZR (Saitama Medical Center, Jichi Medical University, Saitama, Japan) and GraphPad Prism (GraphPad Software, La Jolla, CA, USA). P < 0.05 was considered statistically significant. P-values were calculated by the log-rank test to compare recurrence-free survival (RFS) and overall survival (OS). The Cox proportional hazards model was employed for univariate and multivariate analyses. Variables with P < 0.05 in the univariate analysis were included in the multivariate analysis with stepwise selection based on the Akaike Information Criterion. For categorical data analysed by contingency tables, P-values were calculated using the Chi-square test. For correlation analyses, R-values and P-values were calculated using the Pearson correlation test. Data normality was assessed using the Shapiro–Wilk test. For parametric data, P-values were calculated by the Welch's t-test or ANOVA with the Tukey–Kramer post hoc test, and data are the mean ± SE (tumour volumes) or the mean ± SD (others) in the figures. For non-parametric data, P-values were calculated by the Mann–Whitney U-test or the Kruskal–Wallis test with the Steel-Dwass post hoc test, and boxes represent 25th, 50th and 75th percentiles in the figures. When the statistical test yields only non-significant results, “NS” is noted. In other cases, only statistically significant results are labelled in the figures. For the immunohistochemical and survival analyses of human specimens, required sample sizes were determined using Schoenfeld's formula (α = 0.05, power = 0.8, P = 0.5, HR = 2.0), yielding 66 estimated events, although 112 samples were analysed in this study. For animal experiments, we allocated 6–12 mice per group and confirmed sufficiency using the resource equation approach, with E values exceeding the commonly accepted thresholds (>20). As detailed in the following sections, all experiments were conducted using multiple animals and were replicated independently. Mice were randomly assigned to experimental groups. Blinding was not implemented during outcome assessment or data analysis due to practical constraints.

Role of funders

The funders had no role in study design, data collection, experimental analysis, writing of the report, or decision to publish

Results

Bulk transcriptomics of human PDAC samples

To evaluate the molecular and immunological features of PDAC tissues, we initially performed comprehensive bulk analysis of transcriptome data collected from the PAAD-US, PACA-AU, and PACA-CA cohorts (n = 142, 92, and 264, respectively) of the International Cancer Genome Consortium project. Statistical analyses of clinical factors demonstrated no significant differences in age or sex among the cohorts, although the PACA-AU cohort contained a higher proportion of patients with American Joint Committee on Cancer (AJCC) stage III/IV PDAC (Supplementary Table S3). After removing batch effects using Harmony integration (Silhouette score: 0.190 to 0.041), we conducted dimensionality reduction with UMAP for visualisation. Clustering analysis with the Leiden algorithm identified five groups, termed C1–C5 (Fig. 1A), and gene expression profiling clarified that clusters C1 and C2 had elevated expression of mesenchymal markers CTGF, SPARC, and vimentin (VIM) and lineage transcription factors HNF4A and GATA6, respectively (Fig. 1B and Supplementary Table S4). Notably, the expression level of KDM6A was increased in cluster C2, which was decreased in the Squamous subtype of the Bailey classification,¹⁷ while the expression levels of carboxyl ester lipase (CEL), REG3A, and RBPJL were upregulated in cluster C3 (Fig. 1B), previously reported as hallmark genes of the Exocrine-like subtype of the Collisson classification¹⁰ and the ADEX subtype of the Bailey classification.¹¹ Single-sample gene set enrichment analysis (ssGSEA) revealed that gene sets related to epithelial–mesenchymal transition (EMT) and TGFβ signalling were enriched in cluster C1, whereas cell proliferation-related gene sets such as MITOTIC_SPINDLE and G2M_CHECKPOINT were enriched in cluster C2 (Fig. 1C and Supplementary Table S5). Taken together, these findings suggested that clusters C1, C2, and C3 corresponded to the QM, Classical, and Exocrine-like subtypes of the Collisson classification and the Squamous, Progenitor, and ADEX subtypes of the Bailey classification, respectively.

Fig. 1.

Comprehensive molecular and immunological analysis of human PDAC samples. (A) UMAP plot of 498 PDAC samples from three cohorts. (B) UMAP plots showing expression patterns of representative differentially expressed genes in molecular subtypes. P-values were calculated by the Kruskal–Wallis test with the Steel-Dwass post hoc test. (C) UMAP plots (left) and dot plot (right) showing enrichment scores estimated by the ssGSEA program for hallmark pathways. Only pathways with absolute differences of ≥1.5 in median enrichment scores between the C1 and C2 subtypes are displayed. P-values were calculated by the Kruskal–Wallis test with the Steel-Dwass post hoc test. ES: enrichment score, ROS: reactive oxygen species, OXPHOS: oxidative phosphorylation. EMT: epithelial–mesenchymal transition. (D) UMAP plots of 498 bulk samples (left) and 99 pseudo-bulk samples (right) with spatial regions of each molecular subtype defined by kernel density estimation. P-value was calculated by the Chi-square test. (E) UMAP plot of 228,366 live cells in eight scRNA-seq datasets annotated by the CellTypist program with the immune model. Treg: regulatory T cell, Th; helper T cell, CTL: cytotoxic T cell, Mono: monocyte, MΦ: macrophage, DC: dendritic cell. (F) Heatmap of cellular composition. P-value was calculated by the Chi-square test. (G, H) Proportions of immune cells in immune groups (G) and molecular subtypes (H). P-values were calculated by the Mann–Whitney U-test (G) or the Kruskal–Wallis test with the Steel-Dwass post hoc test (H). (I, J) Representative immunohistochemical images of CD11c+ DCs (I) and CD8+ T cells (J) in human PDAC samples. The scale bar represents 100 μm. (K, L) Kaplan–Meier curves of recurrence-free survival (RFS) and overall survival (OS) between the CD11c-high and -low groups (K) and the CD8-high and -low groups (L). P-values were calculated by the log-rank test. (M) Forest plots showing the hazard ratios and 95% confidence intervals calculated from multivariate analyses for RFS and OS. The Cox proportional hazards model was employed for univariate and multivariate analyses. Variables with P < 0.05 in the univariate analysis were included in the multivariate analysis with stepwise selection based on the Akaike Information Criterion. PD: pancreatoduodenectomy, TP: total pancreatectomy. (N) Overview of the molecular and immunological classification of PDAC through bulk and single-cell transcriptomic integration. QM: quasi-mesenchymal subtype. SQ: squamous subtype. ADEX: aberrantly differentiated endocrine exocrine subtype. Epi: epithelial cell. FB: fibroblast. EC: endothelial cell. NS: not significant. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

Single-cell immune profiling of PDAC subtypes

We next investigated the immune profiles of each subtype using publicly available scRNA-seq datasets (Supplementary Table S6). We selected eight of ten scRNA-seq datasets deposited in the Genome Sequence Archive and the Gene Expression Omnibus and generated pseudo-bulk data from 99 PDAC samples. We then integrated the pseudo-bulk transcriptome datasets with the abovementioned bulk datasets (Fig. 1A) using the Harmony program, and elucidated that the bulk and pseudo-bulk PDAC samples were divided into four groups which were significantly associated with the bulk subtype classification (P = 6.07 × 10⁻²⁴⁹), namely C1/3, C2, and C4 (Fig. 1D and Supplementary Table S7). To further examine the immune landscape, we estimated cell types of the 228,366 live cells in the scRNA-seq datasets using the CellTypist program with the immune model (Fig. 1E and Supplementary Table S8), followed by cellular composition analysis (Fig. 1F and Supplementary Table S9). Clustering analysis of cell types accounting for 95% of all cells stratified the 99 PDAC tissues into two distinct groups: one dominated by tumour and stromal cells such as fibroblasts and endothelial cells and the other enriched with immune cells including DCs and T cells, designated as stromal-dominant and immune-active, respectively (Fig. 1F and G). Comparative analysis of the two classifications discovered that clusters C1/3 and C4 were closely connected to the stromal-dominant and immune-active groups, respectively (P = 1.36 × 10⁻⁸; Supplementary Table S10), suggesting that cluster C4 matched the Immunogenic subtype of the Bailey classification. Further cellular composition analysis across molecular subtypes demonstrated that cluster C1/3 was characterised by the accumulation of fibroblasts, endothelial cells, and macrophages and the depletion of monocytes, DCs, helper T cells, and cytotoxic T cells (Fig. 1H).

Prognostic impact of immune-cell components in human PDAC tissues

To assess the prognostic impact of these immune-cell components, immunohistochemical staining of CD11c and CD8, the specific markers for DCs and cytotoxic T cells, respectively, was performed on 112 PDAC tissues surgically resected at our institution (Fig. 1I and J). Quantitative analysis of CD11c+ DCs and CD8+ T cells indicated a significantly positive correlation between them (P = 9.92 × 10⁻⁹; Supplementary Figure S1A–C), and clarified that the CD11c-low group exhibited worse RFS (P < 0.001) and OS (P < 0.001), while the CD8-low group only showed a trend towards unfavourable RFS and OS (Fig. 1K and L). Multivariate Cox regression analysis identified serum CA19-9, tumour size, lymph node metastasis, and CD11c-low status in RFS and sex, serum CA19-9, surgical procedure, lymph node metastasis, and CD11c-low status in OS (P < 0.05) as independent prognostic factors, respectively (Fig. 1M, Supplementary Figure S1D, Supplementary Table S11, and Supplementary Table S12). In summary, PDAC is classified into the QM/Squamous/C1, Classical/Progenitor/C2, Exocrine-like/ADEX/C3, and Immunogenic/C4 subtypes, and the aggressive QM/Squamous/C1 subtype displays the immunosuppressive profile with a marked paucity of DCs and T cells (Fig. 1N).

Establishment of subcutaneous and peritoneal tumour models exhibiting a cold tumour immune microenvironment with high therapeutic resistance

To develop preclinical models of aggressive PDAC with peritoneal dissemination, we established syngeneic subcutaneous and intraperitoneal co-implantation models using the FC1242 and FC1245 Kras^G12D; Trp53^R172H PDAC cell lines of C57BL/6 mouse origin, which are characterised by low immunogenicity and high tumourigenic ability compared with other PDAC cell lines^14,17 (Supplementary Figure S2A). Histological analysis revealed that both cancerous lesions exhibited poor infiltration of CD11c+ DCs and CD8+ T cells (Supplementary Figure S2A), indicating that these tumours recapitulated a cold tumour immune microenvironment (TIME) consistent with aggressive PDAC (Fig. 1). The peritoneal tumours derived from FC1245 cells showed therapeutic resistance not only to treatments with gemcitabine, αPD-1, and αCTLA-4 but also to combination therapy of gemcitabine and αPD-1, which failed to improve prognosis or reduce tumour volume (Supplementary Figure S2B).

Direct and abscopal antitumour effects of combination mRNA therapy on subcutaneous and peritoneal tumours

Given the substantial tolerance to chemotherapy and ICIs, we hypothesised that the activation of DCs and T lymphocytes might be insufficient (Fig. 1), potentially disrupting the stages of antigen presentation and T cell priming in the cancer–immunity cycle.¹⁸ To drive the cancer–immunity cycle, we explored antitumour effects of the combination of chemotherapeutics inducing immunogenic cell death (ICD), cytokine mRNA drugs that directly and potently invigorate DCs and T cells, and ICIs to overcome immune exhaustion.

Oxaliplatin is well known to robustly elicit ICD¹⁹ and in our study induced ICD more effectively than gemcitabine and irinotecan among standard-of-care agents for PDAC (Supplementary Figure S3A). Moreover, given that interferon-α (IFNα) and interleukin-12 (IL12) each stimulate DCs and T cells,¹⁸ oxaliplatin, αPD-1, and αCTLA-4 were administered along with the direct injection of IFNα/IL12-mRNA into subcutaneous tumour tissues in the syngeneic transplantation model, as shown in Fig. 2A. We performed ELISA analyses of IFNα and IL12 in tumour tissues 4 h after direct intratumoural injection of cytokine mRNA therapeutics and confirmed local upregulation of both cytokines by ∼10-fold relative to controls (Supplementary Figure S3B and C; P < 0.05). ELISA analysis of peripheral blood revealed that serum concentrations of IFNα and IL12 were also elevated compared with controls but approximately 10-fold lower than in tumour tissues (Supplementary Figure S3D and E; P < 0.001), consistent with earlier data.²⁰ No apparent adverse effects on body weight or pathological features of organs were seen following the treatment (Fig. 2B and Supplementary Figure S3F), and no metastatic lesions were detected in the lung or liver of the treatment group, similar to the control group (Supplementary Figure S3F).

Fig. 2.

Direct and abscopal antitumour effects of cytokine mRNA combination therapy on subcutaneous and peritoneal tumours. (A) Protocol for transplantation of FC1245 cells and administration of therapeutic agents. Ox: oxaliplatin. ICI: immune checkpoint inhibitors consisting of anti-PD-1 antibody (αPD-1) and anti-CTLA-4 antibody (αCTLA-4). mRNA: IFNα/IL12-mRNA. s.c.: subcutaneous injection. i.p.: intraperitoneal injection. (B) Body weight changes. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (C) Kaplan–Meier curves of overall survival in treatment groups receiving Ox, Ox + ICI + Luc-mRNA, and Ox + ICI + IFNα/IL12-mRNA (n = 12, respectively). P-values were calculated by the log-rank test. (D) Subcutaneous tumour volumes. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (E) Spider plots showing tumour volumes. (F) Representative photo images of peritoneal dissemination. (G–I) Numbers (G), volumes (H), and total weights (I) of peritoneal tumours. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (J) Kaplan–Meier curves of overall survival in treatment groups lacking individual agents (n = 12, respectively). “All” represents the combination of oxaliplatin, αPD-1, αCTLA-4, and IFNα/IL12-mRNA. “Minus” indicates the omission of a single agent from the combination. P-values were calculated by the log-rank test. (K) Subcutaneous tumour volumes. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (L) Spider plots showing tumour volumes. (M) Representative photo images of peritoneal dissemination. (N–P) Numbers (N), volumes (O), and total weights (P) of peritoneal tumours. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

While neither oxaliplatin monotherapy nor the combination of oxaliplatin with αPD-1 and αCTLA-4 showed significant survival benefits or tumour growth inhibition compared with the control group, the multimodal immunotherapy with mRNA-induced cytokines (MIMIC), that is, oxaliplatin + αPD-1 + αCTLA-4 + IFNα/IL12-mRNA, amazingly extended OS and attenuated subcutaneous tumour progression (Fig. 2C–E; P < 0.001). Macroscopically, the MIMIC combination markedly suppressed peritoneal dissemination induced by intraperitoneal tumour cell seeding, whereas oxaliplatin alone or in combination with ICIs failed to prevent peritoneal tumour formation, resulting in intestinal ischaemia and necrosis (Fig. 2F). Notably, despite the local administration of cytokine mRNA into subcutaneous tumours, distant metastatic lesions were significantly impaired, indicating an abscopal effect. Quantitative analysis of intraperitoneal tumours measuring ≥2 mm in diameter revealed that only the MIMIC combination therapy reduced their number, size, and total weight of them (Fig. 2G–I; P < 0.01).

Similar therapeutic effects were observed in the FC1242 Kras^G12D; Trp53^R172H PDAC cell line (Supplementary Figure S4A). Only the MIMIC treatment group dramatically prolonged OS (Supplementary Figure S4B; P < 0.01), inhibited subcutaneous tumour growth (Supplementary Figure S4C; P < 0.001), and reduced intraperitoneal metastasis (Supplementary Figure S4D–G; P < 0.05).

Requirement of all five agents for effective treatment

To check the indispensability of all five agents (oxaliplatin, IFNα-mRNA, IL12-mRNA, αPD-1, and αCTLA-4) for therapeutic efficacy, we investigated the impact of omitting individual agents from the combination therapy, using the same protocol as shown in Fig. 2A. In all the suboptimal treatment regimens, neither improvement in outcomes (Fig. 2J; P < 0.001) nor shrinkage of subcutaneous and intraperitoneal tumours (Fig. 2K–P; P < 0.01) was detected, suggesting that the lack of any single agent significantly diminished therapeutic efficacy.

Increased intratumoural infiltration of DCs and T lymphocytes by the MIMIC combination therapy

Infiltration of CD11c+ DCs into the subcutaneous tumours and intraperitoneal metastases was limited in FC1245 tumour specimens (Supplementary Figure S2A). At the endpoint of day 17, treatment with oxaliplatin or combination therapy of oxaliplatin and ICIs produced no changes in the number of DCs within the two tumour sites, whereas the MIMIC therapy greatly recruited DCs (Fig. 3A–C; P < 0.001). Further immunohistochemical evaluation demonstrated that the cDC1 (CD103+) subset was enriched in the MIMIC group compared with other treatment groups (Fig. 3D–F; P < 0.001). Although CD11b staining alone showed no clear difference (Supplementary Figure S5A–C), the CD11b+ CD11c+ cDC2 subset is likely elevated following the MIMIC treatment, given the increase of CD11c+ DCs (Fig. 3A–C). We performed immunohistochemical staining of MHC class II, a marker for enhanced antigen presenting capacity, and frequently observed MHC class II-positive cells not only in both lesions but also in the draining lymph nodes of mice treated with the MIMIC therapy (Fig. 3G–I and Supplementary Figure S5D; P < 0.001). Flow cytometric analysis revealed an increase in CD11c+ MHC class II+ cells in subcutaneous tumours specifically in the MIMIC treatment group (Fig. 3J and K; P < 0.01), suggesting DC activation.

Fig. 3.

Increased intratumoural infiltration of DCs and T lymphocytes by the MIMIC combination therapy. (A) Representative immunohistochemical images of CD11c+ DCs in subcutaneous and peritoneal tumour tissues of treatment groups receiving Ox, Ox + ICI + Luc-mRNA, and Ox + ICI + IFNα/IL12-mRNA in the FC1245 model. The scale bar represents 100 μm. Ox: oxaliplatin. ICI: immune checkpoint inhibitors consisting of anti-PD-1 antibody (αPD-1) and anti-CTLA-4 antibody (αCTLA-4). (B, C) Numbers per field of CD11c-positive cells in subcutaneous (B) and peritoneal (C) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (D) Representative immunohistochemical images of CD103-positive cells in subcutaneous and peritoneal tumour tissues. The scale bar represents 100 μm. (E, F) Numbers per field of CD103-positive cells in subcutaneous (E) and peritoneal (F) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (G) Representative immunohistochemical images of MHC class II-positive cells in subcutaneous and peritoneal tumour tissues. The scale bar represents 100 μm. (H, I) Numbers per field of MHC class II-positive cells in subcutaneous (H) and peritoneal (I) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (J, K) Representative flow cytometry plots (J) and quantification (K) of CD11c+ MHC class II+ cells in subcutaneous tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (L) Representative immunohistochemical images of CD8+ T cells in subcutaneous and peritoneal tumour tissues. The scale bar represents 100 μm. (M, N) Numbers per field of CD8-positive cells in subcutaneous (M) and peritoneal (N) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (O) Protocol for treatment with MIMIC and anti-CD3ε (αCD3ε) or anti-NK1.1 (αNK1.1) antibodies in the FC1245 model. MIMIC: the combination of oxaliplatin, anti-PD-1 antibody, anti-CTLA-4 antibody, and IFNα/IL12-mRNA. (P) Subcutaneous tumour volumes. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (Q–S) Numbers (Q), volumes (R), and total weights (S) of peritoneal tumours. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

Similarly to CD11c+ DCs, the number of CD8+ T cells was extremely low in both subcutaneous and intraperitoneal metastatic tumour tissues derived from FC1245 cells (Fig. 3L). While no significant increase in CD8+ T cells was observed even after treatment with oxaliplatin or oxaliplatin combined with ICIs, the MIMIC administration attracted CD8+ T cells to both tumour lesions (Fig. 3L–N; P < 0.001). The numbers of CD11c+ DCs and CD8+ T cells were not elevated in treatment regimens where any of the five agents was omitted (Supplementary Figure S6A and B). Additional immunohistochemical staining of F4/80 (macrophages), iNOS (M1 macrophages), arginase-1 (M2 macrophages), Ly6C (monocytes), NK1.1 (NK cells), and FoxP3 (regulatory T cells) indicated no significant differences in the presence of these immune-cell types between the control and MIMIC treatment groups in either subcutaneous or intraperitoneal tumours (Supplementary Figure S7).

We also noticed that the MIMIC therapy suppressed tumour growth already a few days after treatment initiation. To clarify the mechanism underlying this early effect of the MIMIC therapy, we collected transplanted tumour tissues three days after the first administration and performed immunohistochemical analyses of innate immune-cell markers. While there were no significant differences in the numbers of macrophages or monocytes among the four groups, NK cells were significantly enriched in the MIMIC group, particularly at the invasive front, compared with the remaining groups (Supplementary Figure S8; P < 0.001). We next conducted immune-cell depletion experiments (Fig. 3O). Administration of anti-CD3ε or anti-NK1.1 antibodies during the MIMIC therapy attenuated its antitumour effect in subcutaneous and peritoneal sites (Fig. 3P–S), indicating that both T cells and NK cells contribute to MIMIC-mediated tumour control. Taken together, these observations suggested that only the MIMIC combination therapy could convert a cold TIME to a hot TIME through the activation of DCs and T cells, with the early innate response mediated by NK cells.

Enhancement of immunogenic cell death by the MIMIC combination therapy

To further determine whether the MIMIC combination therapy could drive the cancer–immunity cycle, we performed two analyses on the localisation of damage-associated molecular patterns (DAMPs) and investigated its effect on ICD, the initial stage of the cancer–immunity cycle. First, we assessed calreticulin presentation on the cell surface by conducting double immunofluorescence staining of calreticulin and E-cadherin, which served as a cell membrane marker (Fig. 4A). The membrane-to-cytoplasm ratio of calreticulin fluorescence intensity in subcutaneous tumours and peritoneal metastases was higher in the MIMIC treatment group than in the other groups (Fig. 4B and C; P < 0.05). Second, to evaluate the extracellular release of HMGB1, we measured HMGB1 levels in peritoneal lavage fluid. ELISA analysis demonstrated that HMGB1 concentrations were increased in the MIMIC treatment group compared with the other groups (Fig. 4D; P < 0.001). Collectively, these findings provide supportive evidence that MIMIC treatment induces the relocalisation of DAMP molecules through ICD. Additionally, we examined Ki67 expression, and elucidated that the MIMIC combination therapy significantly attenuated tumour cell proliferation (Supplementary Figure S10; P < 0.05).

Fig. 4.

Enhancement of immunogenic cell death and increase in effector memory T cells by the MIMIC combination therapy. (A) Representative immunofluorescent staining of calreticulin (green) and E-cadherin (red) in subcutaneous and peritoneal tumour tissues of treatment groups receiving Ox, Ox + ICI + Luc-mRNA, and Ox + ICI + IFNα/IL12-mRNA in the FC1245 model. Ox: oxaliplatin. ICI: immune checkpoint inhibitors consisting of anti-PD-1 antibody and anti-CTLA-4 antibody. (B, C) Membrane-to-cytoplasm ratio of calreticulin (CRT) fluorescence intensity in subcutaneous (B) and peritoneal (C) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (D) HMGB1 level in peritoneal lavage fluid. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (E, F) Representative flow cytometry plots (E) and quantification (F) of CD8+ CD44+ CD62L-effector memory T (Tem) cells in peripheral blood. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

Increase in effector memory T cells by the MIMIC combination therapy

Immunohistochemical analysis showed an increase in DCs following the MIMIC combination therapy (Fig. 3A–C), prompting an investigation into the impact of this improved TIME on the activation state of effector memory T cells (Tem). Seventeen days after subcutaneous tumour transplantation, peripheral blood was collected from the tumour-bearing mice to examine the characteristics of CD8+ T cells. Flow cytometry analysis demonstrated that the proportion of CD8+ CD44+ CD62L− Tem cells was importantly augmented in mice treated with the combination of oxaliplatin and ICIs compared with oxaliplatin monotherapy. Furthermore, the MIMIC administration led to a substantial rise in the number of Tem cells, surpassing both oxaliplatin monotherapy and oxaliplatin + ICIs treatment (Fig. 4E and F), consistent with the findings that the number of Tem cells was increased in tumour specimens from the MIMIC group compared with the other groups (Supplementary Figure S11A–C). Similarly, in peritoneal lavage fluid obtained 17 days after subcutaneous tumour implantation, the number of intraperitoneal CD8+ T cells, but not Tem cells, was elevated following oxaliplatin treatment and its combination with ICIs. In contrast, the MIMIC therapy not only increased CD8+ T cells but also enhanced the proportion of Tem cells (Supplementary Figure S11D–F; P < 0.001).

Increase in granzyme B and interferon-γ expression in tumour-infiltrating CD8+ T cells by the MIMIC combination therapy

We subsequently assessed the cytotoxic ability of intratumoural CD8+ T cells. Immunohistochemical and immunofluorescent staining for granzyme B (GZMB) and interferon-γ (IFNγ) in both subcutaneous tumours and peritoneal metastases showed an increase in CD8+ GZMB+ and CD8+ IFNγ+ T cells in the group treated with the MIMIC combination therapy (Fig. 5A, D, E, H and Supplementary Figure S12). Quantitative immunohistochemical examination indicated a significant increase in intratumoural GZMB+ and IFNγ+ T cells within subcutaneous and intraperitoneal cancerous lesions in the MIMIC treatment group compared with the control, oxaliplatin alone or oxaliplatin + ICIs groups (Fig. 5B, C, F, G; P < 0.001).

Fig. 5.

Increase in granzyme B and interferon-γ expression in infiltrated CD8+ T cells by the MIMIC combination therapy. (A) Representative immunohistochemical images of granzyme B (GZMB)+ cells in subcutaneous and peritoneal tumour tissues of treatment groups receiving Ox, Ox + ICI + Luc-mRNA, and Ox + ICI + IFNα/IL12-mRNA in the FC1245 model. The scale bar represents 50 μm. Ox: oxaliplatin. ICI: immune checkpoint inhibitors consisting of anti-PD-1 antibody and anti-CTLA-4 antibody. (B, C) Numbers per field of GZMB-positive cells in subcutaneous (B) and peritoneal (C) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (D) Representative immunofluorescent images of CD8+GZMB+ T cells. The scale bar represents 50 μm. (E) Representative immunohistochemical images of interferon-γ (IFNγ)+ cells in subcutaneous and peritoneal tumour tissues. The scale bar represents 50 μm. (F, G) Numbers per field of IFNγ-positive cells in subcutaneous (F) and peritoneal (G) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (H) Representative immunofluorescent images of CD8+ IFNγ+ T cells. The scale bar represents 50 μm ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

Inhibition of the abscopal effect on peritoneal dissemination by FTY720 treatment

We hypothesised that the abscopal effect on peritoneal metastatic lesions was mediated through the trafficking of immune cells which were activated by the MIMIC combination therapy within subcutaneous tumour tissues. To test this hypothesis, we performed the MIMIC treatment with and without FTY720, a sphingosine-1-phosphate receptor agonist that inhibits lymphocyte egress from lymph nodes,²¹ and evaluated changes in therapeutic efficacy (Fig. 6A and B). We preliminarily compared tumour volumes and immune-cell profiles, including CD11c+ DCs and CD8+ T cells, between the control and FTY720 treatment groups, but observed no significant differences (Supplementary Figure S13), suggesting that FTY720 exerted no direct effects on tumour cells or the tumour microenvironment. As expected from its mechanism, FTY720 administration had minimal impact on the growth of subcutaneous tumours directly injected with IFNα/IL12-mRNA, as the local immune responses induced by the MIMIC therapy were not impaired (Fig. 6C; P < 0.001). In contrast, intraperitoneal tumour regression was achieved by MIMIC but markedly diminished by FTY720, likely due to inhibition of activated T cell trafficking (Fig. 6D). Quantification confirmed that the number, volume, and total weight of the peritoneal tumours were higher in the MIMIC + FTY720 treatment group than in the MIMIC treatment group (Fig. 6E–G). Flow cytometry analysis of peripheral blood from the tumour-bearing mice identified a significant reduction in lymphocytes, CD8+ T cells, and CD44+ CD62L− Tem cells in the MIMIC + FTY720 treatment group (Fig. 6H–K). Immunohistochemical analyses of subcutaneous and intraperitoneal tumour specimens revealed that CD11c+ DCs were comparable between the MIMIC and MIMIC + FTY720 groups, whereas CD8+ T cells showed no significant difference in subcutaneous tumours but were markedly decreased in peritoneal metastases of mice treated with MIMIC + FTY720 (Fig. 6L–Q). These findings supported our hypothesis on the abscopal effect of the MIMIC therapy.

Fig. 6.

Inhibition of the abscopal effect of the MIMIC combination therapy on peritoneal dissemination by FTY720 treatment. (A, B) Overview (A) and protocol (B) for treatment with MIMIC and FTY720 in the FC1245 model. MIMIC: the combination of oxaliplatin, anti-PD-1 antibody, anti-CTLA-4 antibody, and IFNα/IL12-mRNA. Ox: oxaliplatin. ICI: immune checkpoint inhibitors consisting of anti-PD-1 antibody and anti-CTLA-4 antibody. mRNA: IFNα/IL12-mRNA. s.c.: subcutaneous injection. i.p.: intraperitoneal injection. (C) Subcutaneous tumour volumes (Control, n = 4; MIMIC, n = 6; MIMIC + FTY720, n = 6). P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (D) Representative photo images of peritoneal dissemination. (E–G) Numbers (E), volumes (F), and total weights (G) of peritoneal tumours. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (H) Representative flow cytometry plots of CD8+ CD44+ CD62L− effector memory T (Tem) cells. (I–K) Quantification of lymphocytes (I), CD8+ T cells (J), and CD8+ CD44+ CD62L− Tem cells (K) in peripheral blood. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (L) Representative immunohistochemical images of CD11c+ DCs in subcutaneous and peritoneal tumour tissues. The scale bar represents 100 μm. (M, N) Numbers per field of CD11c-positive cells in subcutaneous (M) and peritoneal (N) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. (O) Representative immunohistochemical images of CD8+ T cells in subcutaneous and peritoneal tumour tissues. The scale bar represents 100 μm. (P, Q) Numbers per field of CD8-positive cells in subcutaneous (P) and peritoneal (Q) tumour tissues. P-values were calculated by ANOVA with the Tukey–Kramer post hoc test. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

Expansion of TCR repertoire diversity induced by the MIMIC combination therapy

We next performed TCR repertoire analysis on tumour-infiltrating T cells in subcutaneous tumour specimens harvested on Day 17. Manhattan plots, heatmaps, and circos plots displayed that the MIMIC treatment group exhibited greater V-J gene pairing diversity than the control group (Fig. 7A and B). This expansion of TCR repertoire diversity indicated that the MIMIC therapy successfully induced T cell priming, resulting in the enhancement of antigen-agnostic activity and the improvement of immune surveillance.

Fig. 7.

Expansion of TCR repertoire diversity induced by the MIMIC combination therapy. (A, B) Manhattan plots (top), heatmaps (bottom left) and circos plots (bottom right) of TCR repertoires in the control (A) and MIMIC treatment (B) groups in the FC1245 model. MIMIC: the combination of oxaliplatin, anti-PD-1 antibody (αPD-1), anti-CTLA-4 antibody (αCTLA-4), and IFNα/IL12-mRNA. (C) Overview of the cancer immunity cycle driven by the MIMIC combination therapy. DC: dendritic cell; TIME: tumour immune microenvironment.

Discussion

While the “two-class” model of PDAC classification, consisting of the Basal-like and Classical subtypes,^9,22 is widely accepted, Bailey et al. expanded the Collisson classification into the Squamous, Progenitor, ADEX and Immunogenic subtypes.^10,11 Wartenberg et al. pioneered an immunological stratification of PDAC into the Immune-escape, Immune-exhausted, and Immune-rich subtypes based on immunohistochemical analysis,²³ and subsequent studies using immune-cell deconvolution analysis have established a consensus classification, categorising PDAC into Immune-low and Immune-high groups.¹² Although single-cell analyses have been employed for comprehensive characterisation of individual cell types, most research teams have analysed a small number of tumour samples and focused on the role of cancer-associated fibroblasts in the Basal-like and Classical subtypes.¹³ In this work, by integrating eight scRNA-seq datasets comprising 99 tumour specimens and 228,366 cells, pseudo-bulk analysis identified three molecular subtypes (C1/3, C2, and C4) corresponding to the Collisson and Bailey classifications, whereas cellular composition analysis stratified PDAC samples into a stromal-dominant group enriched with fibroblasts and endothelial cells and an immune-active group characterised by DC and T cell infiltration, consistent with the consensus immunological classification (Fig. 1N). The aggressive QM/Squamous/C1 plus Exocrine-like/ADEX/C3 subtype exhibited a strong correlation with the stromal-dominant/DC-low group (Fig. 1F), and further bioinformatics investigation and immunohistochemical validation suggested that the paucity of DCs and T cells within the tumour microenvironment may contribute to the “cold” tumour phenotype of aggressive PDAC, linked to immunotherapy resistance and unfavourable prognosis (Fig. 1G–M).

In order to convert a cold TIME to a hot TIME, novel immunotherapies have recently been developed. Cancer vaccine therapy aimed to promote antigen presentation and stimulating the immune system, but no clinical benefit was observed in patients with advanced PDAC as documented in the TeloVac and PEGASUS-PC trials.^6,24 Cancer vaccine therapy combined with ICIs²⁵ and mRNA-based neoantigen vaccination²⁶ also yielded discouraging outcomes in pancreatic cancer. While the PRINCE and OPTIMISE-1 trials evaluated CD40 agonistic antibodies in patients with PDAC, DC activation was inadequate, leading to limited therapeutic efficacy.^7,27 Although intratumoural administration of mRNA-LNPs encoding cytokines drives tumour regression and abscopal effects in preclinical models of immune-hot tumours such as the B16F10 melanoma cell line and the MC38 microsatellite-instable colon cancer cell line,^20,28, 29, 30 this approach failed clinical translation for advanced PDAC in the MEDI1191 trial (conference abstract; non–peer-reviewed).³¹ One plausible explanation is that optimal management of this disease requires the full functionality of all stages of the cancer–immunity cycle,¹⁸ namely, release of cancer cell antigens, cancer antigen presentation by DCs, T cell priming and activation, and killing of cancer cells, and if any of these stages are disturbed, immunotherapy may be disrupted. Specifically, conventional chemotherapy with ICIs may lack DC and T cell invigoration, cancer vaccines induce insufficient T cell priming, and CD40 agonistic antibodies or mRNA induced cytokines alone achieve minimal ICD initiation. In designing our therapeutic strategy, we highlighted IFNα and IL12 to target the stages of antigen presentation and T cell priming in the cancer–immunity cycle,¹⁸ which were disrupted in aggressive PDAC (Fig. 1), and combined these cytokines with oxaliplatin that elicits ICD and ICIs that alleviate DC suppression and T cell exhaustion to MIMIC immunostimulatory pathways.

IFNα and IL12 are potent immunomodulatory cytokines known to enhance both innate and adaptive antitumour immunity. IFNα directly suppresses tumour growth by inducing cell cycle arrest and apoptosis, while also promoting DC activation and maturation, thereby facilitating antigen presentation and T cell priming.³² IL12 plays a critical role in bridging innate and adaptive immunity by activating T cells and NK cells and accelerating IFNγ production, which in turn amplifies antigen presentation and pro-inflammatory cytokine release.³³ Clinical use of both cytokines has provided proof-of-concept for their immunotherapeutic potential, as evidenced by recombinant IFNα improving survival in melanoma and renal cell carcinoma and recombinant IL12 inducing tumour regression in subsets of patients with melanoma and head and neck cancer. However, their systemic administration is limited by severe toxicities such as neuropsychiatric disorders and hepatotoxicity with IFNα.³⁴ and cytokine-release syndrome with IL12.³⁵ To address this challenge, recent advances in mRNA therapeutics have enabled spatially restricted cytokine production within the TIME, reducing systemic diffusion and thereby minimising adverse effects.³⁶ In our study, intratumoural administration of mRNA therapeutics for IFNα and IL12 resulted in robust immune activation without systemic toxicity, weight loss, excessive mortality, or signs of inflammation in major organs (Fig. 2B and C and Supplementary Figure S3F), supporting the safety and feasibility of this innovative approach.

Recently, several mRNA-induced cytokines for cancer immunotherapy have entered clinical evaluation.^37,38 MEDI1191, an mRNA therapeutic encoding single-chain IL12, demonstrated a favourable safety profile when administered intratumourally, even in combination with immune checkpoint blockade (NCT03946800).³¹ Clinical trials for other IL12 mRNA agents, including JCXH-211 (NCT05727839), ABO2011 (NCT06088004), and STX-001 (NCT06249048), are also ongoing. In particular, BNT131/SAR441000 (NCT03871348) is a cocktail of mRNAs encoding single-chain IL12, IL15 sushi, GM-CSF, and IFNα, developed on the basis of promising preclinical results.²⁰ Beyond IL12 mRNA, IL2 mRNA monotherapy (BNT151) and the combination of IL7 mRNA (BNT152) with IL2 mRNA (BNT153) are under clinical investigation (NCT04455620 and NCT04710043, respectively), as well as an mRNA construct encoding the co-stimulatory molecule OX40L (mRNA-2416; NCT03323398) and an mRNA mixture designed to produce IL23, IL36γ, and OX40L (mRNA-2752; NCT03739931). Thus, several early-phase clinical trials of mRNA therapeutics, both as monotherapies and in combination with ICIs, are underway across a range of advanced solid tumours, including pancreatic cancer. While these initiatives primarily explore the feasibility of mRNA-based immunotherapy in various tumour settings, our study first clarified the immunological mechanisms underlying poor prognosis and marked resistance to conventional therapies in aggressive PDAC, and then demonstrated the efficacy of multimodal strategies incorporating mRNA-induced cytokines that target distinct stages of the cancer–immunity cycle, representing “bottom-up” and “top-down” approaches, respectively.

Limitations of this study include the following five points. First, while not completely comprehensive in patient representation, our datasets were pooled from multiple cohorts with large sample sizes (bulk: 3 cohorts, 498 samples; single-cell: 8 cohorts, 99 samples) and can be considered broadly representative and minimally biased. We performed immunohistochemical analyses using clinical specimens from our institution to validate findings from scRNA-seq analyses, but this single-centre study, consisting exclusively of Japanese patients, was enriched with patients aged >70 years and male patients. Although no significant difference was detected in AJCC stage (Supplementary Table S3), further examination is required. Second, given that orthotopic and autochthonous models raise serious concerns about reproducibility and quantitative assessment, we selected subcutaneous and intraperitoneal co-implantation models for the current study, which aimed to quantitatively and statistically evaluate the efficacy of our mRNA-induced cytokine therapy in suppressing peritoneal dissemination and improving survival duration. Our models recapitulate key immunological features of aggressive PDAC (Fig. 2) but do not fully reflect the complexity of the human tumour microenvironment. Additionally, although the combination of five agents demonstrated strong therapeutic efficacy in preclinical models (Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 7), it may face logistical and safety challenges in clinical implementation. Third, while the MIMIC therapy successfully expanded TCR repertoire diversity and enhanced antigen-agnostic activity (Fig. 7A and B), the long-term durability of the antitumour response remains unclear. Fourth, no apparent adverse events were observed on pathological examination (Supplementary Figure S3F), but comprehensive toxicity assessments are required. We observed leakage of cytokines into the circulation following mRNA-LNP administration (Supplementary Figure S3D and E), which is well-known to result from their preferential accumulation and off-target expression in the liver.³⁹ To suppress the hepatic expression, several approaches have been developed, including (1) introduction of the target sequence of miR-122, a liver-specific microRNA, into the 3′ UTR of cytokine mRNA constructs to induce their degradation and inhibit their translation in hepatocytes²⁸ and (2) transient coating of the liver sinusoidal walls with polyethylene glycol to prevent nanoparticle delivery, as we have previously reported.⁴⁰ Finally, the therapeutic regimen involves intratumoural administration of cytokine mRNA agents, which may limit clinical applicability. However, we plan to perform this procedure during staging laparoscopy upon incidental detection of peritoneal dissemination or via endoscopic ultrasound-guided fine needle injection, followed by intravenous administration of oxaliplatin, anti-PD-1 antibody, and anti-CTLA-4 antibody. In clinical practice, our surgical team conducts direct intratumoural injections of ICG for marking pancreatic tumours under laparoscopic guidance (Supplementary Video). Several endoscopic ultrasound-guided treatments for solid pancreatic tumours, including fine-needle injection of biologic and chemotherapeutic agents, radiofrequency ablation, and photodynamic therapy, have been reported (Supplementary Figure S13).⁴¹ Thus, we consider intratumoural administration of cytokine-inducing mRNA therapeutics technically feasible with a favourable safety profile.

In conclusion, comprehensive bulk and single-cell analyses demonstrated a paucity of DCs and T cells in the aggressive PDAC subtype. We established subcutaneous and intraperitoneal tumour models using the FC1245 pancreatic cancer cells, which harbour Kras and Trp53 mutations commonly detected in human PDAC and exhibit a “cold” tumour phenotype with extremely poor responsiveness to chemotherapy and immunotherapy. Based on the cancer–immunity cycle theory, the MIMIC combination therapy including IFNα/IL12-mRNA for the direct activation of DCs and T cells in subcutaneous tumours not only exerted inhibitory effects on tumour growth but also significantly suppressed intraperitoneal metastatic tumours, leading to pivotal survival prolongation (Fig. 7C). In clinical practice, patients could be stratified into stromal-dominant/DC-low and immune-active/DC-high groups through pathological examination of endoscopic ultrasound-guided biopsy samples. The immune-active/DC-high group, predicted to have a favourable prognosis (Fig. 1I–M), could receive standard treatment options such as surgical resection, chemotherapy, and radiotherapy, whereas patients in the stromal-dominant/DC-low group might be optimal candidates for the MIMIC therapy (Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 7). Such classification and subtype-specific therapeutic approaches may help optimise both patient burden and healthcare resources.

Contributors

All authors read and approved the final version of the manuscript. YT, SS, S. Tanaka have accessed and verified the underlying data.

Conceptualisation: YT, SS, S. Tanaka.

Methodology: YT, SS, MK, S. Tsukihara, Y. Igarashi, KK, KO, KY, EM, YM, SU.

Investigation: YT, SS, MK, S. Tsukihara, Y. Igarashi, KK, KO, KY, KU, AK, AN, MY.

Data collection: YT, MK, AN, MY, Y. Ishikawa, DB, S. Tanaka.

Writing: YT, SS, SU, S. Tanaka.

Manuscript review and editing: YA, MH, TI, DB.

Visualisation: YT, SS, S. Tsukihara, Y. Igarashi, KK, KO, KY.

Funding acquisition: SS, SU, S. Tanaka.

Data sharing statement

The raw data, study materials, analytic methods, and source codes used in this study will be made available to other researchers for purposes of reproducing the results or replicating the procedure. Data and materials can be obtained from the corresponding authors (shimada.monc@tmd.ac.jp and tanaka.monc@tmd.ac.jp) upon reasonable request.

Declaration of interests

The authors declare no potential conflicts of interest. The intellectual property related to the findings in this paper has been filed as a patent application (Japanese Patent Application No. 2025-16064).