The Combination of R848 with Sorafenib Enhances Antitumor Effects by Reprogramming the Tumor Immune Microenvironment and Facilitating Vascular Normalization in Hepatocellular Carcinoma

Abstract Novel promising strategies for combination with sorafenib are urgently needed to enhance its clinical benefit and overcome toxicity in hepatocellular carcinoma (HCC). the molecular and immunomodulatory antitumor effects of sorafenib alone and in combination with the new immunotherapeutic agent R848 are presented. Syngeneic HCC mouse model is presented to explore the antitumor effect and safety of three sorafenib doses alone, R848 alone, or their combination in vivo. R848 significantly enhances the sorafenib antitumor activity at a low subclinical dose with no obvious toxic side effects. Furthermore, the combination therapy reprograms the tumor immune microenvironment by increasing antitumor macrophages and neutrophils and preventing immunosuppressive signaling. Combination treatment promotes classical M1 macrophage‐to‐FTH1high M1 macrophage transition. The close interaction between neutrophils/classical M1 macrophages and dendritic cells promotes tumor antigen presentation to T cells, inducing cytotoxic CD8+ T cell‐mediated antitumor immunity. Additionally, low‐dose sorafenib, alone or combined with R848, normalizes the tumor vasculature, generating a positive feedback loop to support the antitumor immune environment. Therefore, the combination therapy reprograms the HCC immune microenvironment and normalizes the vasculature, improving the therapeutic benefit of low‐dose sorafenib and minimizing toxicity, suggesting a promising novel immunotherapy (R848) and targeted therapy (tyrosine kinase inhibitors) combination strategy for HCC treatment.

loaded into a tube containing anticoagulant. All the red blood cells of above tissues were lysed by ACK lysis buffer on ice for 5 minutes. The 1×10 5 cells were stained in 1.5-mL tubes with antibody for 15 minutes at room temperature (25°C) in the dark.

Toxicity study
For toxicity assessment, mice were weighed before and after treatment (on day 21 for R848 and sorafenib). Blood samples were collected from the orbital sinus using a microhematocrit tube after each treatment and subjected to biochemical analysis for kidney function markers blood urea nitrogen (BUN) and creatinine and liver marker enzymes alanine transaminase (ALT) and aspartate transaminase (AST) to evaluate treatment toxicity by COBAS ® 8000 analyzer series (Roche Diagnostics, Rotkreuz, Switzerland).

RNA extraction, cDNA synthesis, and quantitative real-time PCR
Mouse tissue total RNA was isolated from the digested tumor cells using TRIzol reagent (Ambion, USA). cDNA was synthesized by RNA reverse transcription using a quantitative RT-PCR kit (Takara, Japan). The amplification reaction was performed according to the manufacturer's instructions (Takara, Japan) using predesigned primers. Primer sequences are listed in Table S1, Supporting Information.

scRNA-seq generation (Single cell isolation/library construction/sequencing)
Following with 10× Genomics® (Novogene, China) Cell Preparation Guide described, mice were euthanized by cervical dislocation, and the tumor tissues were peeled off and made into a single cell suspension after splitting red blood cells by Tumor Dissociation Kit (Miltenyi Biotec, Germany) and ACK lysis buffer. Cell viability should exceed 80% as determined by Taipan Blue staining. Calculate the appropriate volume of cell suspension so that each sample contains approximately 12,000 cells. Raw reads were transformed into FASTQ files using Illumina sequencers, then checked by FastQC. Cell Ranger pipeline (v.5.0.1, 10× Genomics (https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/5. 0)) was used to perform basic statistics.

Raw data processing, data filtering and cell clustering of scRNA-seq
The 10× Genomics Cell Ranger pipeline was used to demultiplex raw reads, read alignment and generation of the gene-cell matrix (10× Genomics, v5.0.1, GRCh38).
Genes detected in less than 3 cells and cells in which detected transcripts were either fewer than 200 genes or >6000 genes and >50% of transcripts derived from mitochondrial genes were filtered out and excluded from the subsequent analysis. And Doublets were removed by DoubletFinder. Seurat R package (version 4.1.0) were used to normalized and scaled the gene count matrix to identify highly variable genes for unsupervised cell clustering via FindVariableFeatures function with default parameter.
Principal component analysis (PCA) was performed on top 2000 highly variable genes.

The cells were clustered on the basis of the first 30 PCs using FindNeighbors and
FindClusters with a resolution set to 0.5. Finally, we manually annotated the cell types using canonical marker genes. The second round of clustering T cells and macrophage cells was the same as above: starting from normalized and scaled the expression matrix, identifying highly variable genes with FindVariableFeatures method, and clustering with FindNeighbors and FindClusters function.

Cell-cell communications analysis
Mouse genes were mapped to their human genes orthologs before analysis.
CellPhoneDB (https://www.cellphonedb.org/) was used to anticipate enriched ligand-receptor interactions between two distinct cell types based on single-cell transcriptomics data.
Briefly, on the basis of literature and public databases, a vetted database of ligand-receptor interactions were constructed. The average receptor expression level and ligand expression level for each pair of cell types were calculated using a random permutation of the cell type labels on all cells. The process was repeated 1000 times to create a null distribution for each ligand-receptor pair in each pair of cell types. P value was calculated by computing the percentage of the means that are equal to or higher than the null distribution for a certain ligand-receptor pair. Only receptors and ligands that were generated by more than 30% of the cells in the specific cluster were accounted for.
To further verify the results of CellPhoneDB analysis, we also applied the CellChat algrithms to infer cell-cell communications between immune cell subclusters. The

Pathway enrichment analysis and ssGSEA analyses
To identify differential expression genes (DEGs), FindMarkers and FindAllMarkers functions in Seurat were performed. By GO and Reactome pathway enrichment analysis, DEGs related to the differentiation from M1 macrophages to FTH1 high macrophages was assessed to find enriched pathways. Gene Ontology (GO) enrichment analysis of target genes was developed by the clusterProfiler R package. Reactome pathway-based analysis was performed using the ReactomePA R package. To evaluate the function of DEGs associated with T cells, we used ssGSEA (run by R package GSVA) to evaluate the activity levels of gene sets associated with T cell cytotoxicity and exhaustion for each sample. Only genes were significant (p<0.05) and with an average logFC higher than log (2) were considered.

Trajectory and pseudotime analysis for Single-cell RNA-seq
The R package CytoTRACE v.0.3.3 was used to predict the differentiation state of cells from the single-cell RNA-seq (scRNA-seq) data. R package Monocle 3 was used to discover the differentiation trajectory of M1 macrophage converting into FTH1 high macrophage. Utilizing M1 macrophages as the root state, each cell was given a pseudotime value using the order_cells function. We also applied the CytoTRACE algorithm to predict the differentiation state of macrophage. The identified path was mapped to UMAP projection for visualization.

CIBERSORT deconvolution for surgical tissues bulk RNA-seq profiles
CIBERSORT (Cell-type Identification by Estimating Relative Subsets of RNA Transcripts, http://cibersort.stanford.edu) is a computational tool that can estimate the relative fractions of various cell clusters in the gene expression profiles. We imported 37 RNA-seq dates of HCC tissues, which are classified as nonresponse and response to sorafenib, respectively, into CIBERSORT in order to assess the varying ratios of infiltrating immune cells in various response groups to sorafenib. The average mean of 14 different infiltrating-immune cells has been recognized by  and was computed individually in the two groups.

Western blot assay
To prepare the tissue proteins, the tumor tissues were first homogenized in a lysis buffer containing protease and phosphatase inhibitors to prevent protein degradation.
The lysates were then centrifuged to remove any debris, and the supernatants were collected as the protein samples. The protein samples were loaded onto an SDS-PAGE gel and separated by size using an electric field. The separated proteins were then transferred onto PVDF membranes (Roche, Basel, Switzerland). After blocking with 3% albumin from bovine serum (BSA) for 1 h, the membrane was incubated with various primary antibodies overnight at 4°C, followed by incubation with corresponding secondary antibodies at a 1:4000 dilution for 1 h at room temperature. . Error bars represent means ± SEMs; ns: p > 0.05, * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001, one-way ANOVA (B and C), t-test (D), log-rank test (E).