We acquired bulk RNA-seq data from TCGA (https://portal.gdc.cancer.gov) database, including transcriptomic data and clinical information regarding 33 cancer as well as non-carcinoma samples. Additionally,we obtained scRNA-seq data of six types of cancer, containing breast cancer (GSE176078), CRC (GSE166555), renal cell carcinoma (GSE159115), glioma (GSE135045), gastric cancer (GSE167297) and head and neck squamous cell carcinoma (HNSC) (GSE139324) from Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), in which three samples were selected from each dataset and integrated for subsequent data analysis.

CD44 mRNA levels in major human and tumor samples, as well as its subcellular localization were explored using Human Protein Atlas (HPA) (https://www.proteinatlas.org/) database,through searching CD44 in Tissue module or Subcell module. GeneeDE module of TIMER2.0 (http://timer.cistrome.org/) database was adopted for analyzing CD44 transcription level within different cancer and non-carcinoma samples. Differential CD44 mRNA expression within normal and cancer samples and at different pathological stages, was performed using Single Gene Analysis module of the Gene Expression Profiling Interactive Analysis (GEPIA) (http://gepia.cancer-pku.cn/) database.

Based on median CD44 gene expression level, we classified patients into low and high CD44 expression groups. Besides, survival package of R software was utilized for survival analysis, followed by plotting of Kaplan-Meier survival curve. Furthermore, univariate Cox regression analysis was utilized for assessing whether CD44 was significant for predicting overall survival (OS) and disease-specific survival (DSS) of patients with cancers.

We utilized cBioPortal platform (https://www.cbioportal.org/) for analyzing CD44 mutation frequency and characteristics within pan-cancer from TCGA database. The gene alternations and mutation sites of CD44 were viewed with the OncoPrint module. The impact of CD44 gene alternations on survival was evaluated using tumor sample data from TCGA database with the Comparison/Survival module. We acquired the somatic mutations and somatic copy number alterations of tumor samples from cutaneous melanoma based on TCGA database. The patients were allocated to high or low CD44 expression group in line with a cutoff value of 25%. The somatic mutation data were calculated using the Maftools package and visualized using a waterfall diagram.

The Correlation Analysis module of GEPIA was adopted for calculating correlation with target gene level using Pearson correlation method. The TOP 100 genes associated with CD44 in 33 types of cancer were obtained from GEPIA with the similar genes detection function, and Gene Ontology (GO) analysis was performed using these 100 genes. Corresponding genetic characteristics were constructed through CancerSEA (http://biocc.hrbmu.edu.cn/CancerSEA) database, besides, correlation analysis of target genes with 14 cancer functional states in different cancers was performed using gene set variation analysis (GSVA) algorithm. In addition, we obtained the HALLMARK gene set in Molecular Signatures Database (MSigDB) (http://software.broadinstitute.org/gsea/msigdb/index.jsp) database, and conducted gene set enrichment analysis (GSEA) of cancers through clusterProfiler package of R. The false discovery rate (FDR) and normalized enrichment score (NES) were determined in diverse biological processes of each cancer type.

We adopted CIBERSORT method for calculating relative scores of 22 kinds of immune cells within 33 types of cancer, and the Spearman’s correlation of CD44 level with immune infiltration was further measured. Spearman’s correlation analysis was performed using TIMER2.0 database for analyzing the correlation between CD44 and immune checkpoints, chemokines, and receptor genes. In line with somatic data obtained based on TCGA database, we applied Maftools package in calculating TMB value and MSI score for each tumor. Then, the correlation of CD44 level with TMB or MSI was examined through Spearman’s correlation analysis.

TIDE (http://tide.dfci.harvard.edu/) has been developed as the computational framework for evaluating tumor immune escape potential according to gene levels in tumor samples. By comparing target genes with recognized immune response evaluation biomarkers [including TMB, MSI, CD274, CD8, interferon gamma (INFγ), and TIDE], the area under receiver operating characteristic (ROC) curve (AUC) was determined for assessing the ICB response prediction performance. Furthermore, we predicted the relation of gene level with clinical outcomes of ICB, and the changes in CD44 gene expression induced by ICB therapy were identified on the basis of immunotherapy cohorts of homologous in vivo mouse models and in vitro cell models obtained from Tumor Immune Syngeneic MOuse (TISMO) database (http://tismo.cistrome.org). Finally, the role of CD44 in clinical response of anti-PD-L1 immunotherapy was predicted using the GSE91061 dataset.

10x Genomics data of three patients were read from single-cell datasets of six types of cancer using R Seurat package (4.2.3) to process data to construct Seurat objects, integrated with Harmony to eliminated batch effects. The low-quality cells were filtered according to the quality control criteria: cells with gene <200 & gene >5000, more than 15% derived from the mitochondrial genome UMI. The FindVariableFeatures function was used to select 2000 highly variable genes for PCA analysis and dimensionality reduction. The average CD44 expression of each cancer in each cell subpopulation was calculated and displayed by a heatmap.

Monocle2 was used to explore the developmental trajectories of CD44 gene expression during differentiation in T cells, B cells, and myeloid lineages. The UMI matrix was read from the seurat object, and then the monocle object was created using the newCellDataSe function with the parameter expressionFamily=negbinomial.size. Further, the DDRTree method of reduceDimension function was used to reduce dimension, and the cells sequentially ordered by orderCells function were visualized.

Monocytes with CD44 gene expression > 0 was defined as CD44⁺ monocytes. Conversely, equal to 0 is defined as CD44^-monocyte. The weights of interactions of CD44⁺ monocytes or CD44^- cells with other cells were calculated by CellChat (1.1.3) computeCommunProbe function. Further, the netAnalysis_signalingRole_scatter function was used to identify the signals that contribute most to the communication patterns and outgoing or incoming signals of certain cell populations. The function netVisual_bubble was used to identify key receptor-ligand pairs mediating intercellular communication.

In this work, we purchased human THP-1 monocytes and human HCT-116 CRC cell in Procell Life Science & Technology Co., Ltd. (China) Cells with STR identification and mycoplasma detection and cultivated them within RPMI-1640 medium that contained 10% fetal bovine serum (FBS) in an incubator with 5% CO2 under 37°C. 0.05 mM β-mercaptoethanol was further added into the THP-1 cell culture medium.

THP-1 cells (1×105/well) were inoculated in the 24-well plate, followed by 24-h exposure to 100 ng/ml phorbol 12-myristate 13-acetate (PMA) (Abcam, ab120297) for differentiation induction to M0 macrophages. Then, lipopolysaccharide (100 ng/ml, LPS) (Sigma Aldrich, L2630) as well as interferon-gamma (20 ng/ml, IFN-γ) (Novoprotein, GMP-CI57) was added to induce differentiation into M1 macrophages. M0 macrophages were also exposed to interleukin (IL)-4 (20 ng/ml, Sigma Aldrich, SRP3093-20UG) together with IL-13 (20 ng/ml, Sigma Aldrich, SRP3274) for 24 h to induce differentiation into M2 macrophages.

siRNAs specifically targeting CD44 were prepared in Tsingke Biotechnology (China). Cells (3 × 10⁵/well) were seeded into 6-well plates till reaching reach 30–40% density, followed by transfection using TSnanofect V2 transfection reagent in line with specific protocols (Tsingke Biotechnology, China). Supplementary Table S1 displays siRNA sequences. The transfection concentration of siRNA was 50 nM, that is, if the total volume of each well was 1 mL, the system consisted of 800 μl antibiotic-free cell culture medium + 2.5 μl siRNA diluted with 100 μl Opti-MEM +2 μl transfection reagent diluted with 100 μl Opti-MEM. After transfection, the cells were cultured in an incubator with 5% CO2 at 37°C for 24 hours, and then replaced with fresh medium. The transfection efficiency of FAM negative control was observed by fluorescence microscope. When the fluorescence positive cells exceeded 80%, qRT-PCR was used to detect the target gene expression.

TRIzol reagent (Gene, China) was used for extracting total RNA from cultured cells. The concentration and purity of RNA were evaluated using Nanodrop (Thermo,USA). The CDNA was synthesized using the Takara PCR Thermal cycler (Takara,Japanese), according to the instructions of HiScript III 1st Strand cDNA Synthesis Kit (Vazyme Biotech, China) with the reaction condition at 37°C for 15 min and 85°C for 5 sec. The total amount of RNA was 1μg.QRT-PCR were performed using Bio-rad T100 Real-Time PCR system (Bio-rad, USA) according to the manufacturer’s instructions of HiScript II Q RT SuperMix for qPCR (Vazyme Biotech, China). The reaction system was: 10 μ SYBR qPCR Master Mix, 0.5 μl forward Primer (10 µM) and 0.5 μl reverse Primer (10 µM) and 1μl CDNA template. The reaction conditions used was: stage 1 pre-denaturation, 95°C for 30s; stage 2 cycling reaction, performed with 40 cycles for 95° C 10 s,60° C,20s. The primer sequences were listed in Supplementary Table S1. The relative mRNA expression levels of target genes was calculated by 2^-ΔΔCt approach.

HCT116 cells (5×10³/well) were inoculated into the 96-well plate. At 0, 24, 48, 72, and 96 h after treatment, CCK-8 reagent (10 μl, Dojindo, Japan) and RPMI 1640 medium (90 μl) were introduced into every well. After 1-h incubation under 37°C and 5% CO2, absorbance values were measured at 450 nm.

We inoculated control or transfected HCT-116 cells (800 cells/well) into a 6-well plate prior to another 14-day incubation within 2 mL medium under 37°C with 5% CO2. Thereafter, those colonies formed were immersed in 4% paraformaldehyde for a 20-min period and washed with PBS before 10-min staining using 0.1% crystal violet. Finally, colony counts of different groups were determined.

Three parallel lines were pre-marked on the back of 6 well cell culture plates before cell inoculation. HCT 116 or cells transfected with siRNA were seeded in 6-well cell culture plates with 1×10⁶ cells/well till reaching 90% density. Cells were then scraped with the 200-μl pipette tip perpendicular to the marker line, and cellular fragments was washed with PBS. After that, cultures were changed to serum-free medium for 24 hours. At 0 and 24 h, the degree of scratch healing at the same scratch was observed under an inverted microscope. Image J was then adopted for measuring and analyzing wound width, and migration rate was calculated. Wound healing rate = (0 h-24 h wound width)/0 h wound width ×100%.

We re-suspended HCT-116 cells (5×10⁴/well) into serum-free RPMI-1640 medium (200 μl) and seeded them into top Transwell chambers (8μm), which were coated with matrix gel (BD Biosciences, USA), and contained RPMI 1640 medium (600 μl) with 10% FBS in the bottom chambers. In addition, in the co-cultured transwell system, macrophages under different treatment conditions were incubated in the bottom chambers with 10% FBS RPMI 1640 medium. After 48h co-culture, we immersed cells on the membrane bottom in the Transwell chamber in 4% formaldehyde prior to staining using 0.1% crystal violet. While cotton balls were used to remove cells on top of the membrane. A microscope was utilized for image acquisition.

Proteins from cells were extracted using RIPA lysis buffer (Solarbio,R0020). BCA Protein Assay Kit (Beyotime, P0012S) was used to quantify protein concentrations. The protein was separated by 10% or 15% SDS-PAGE at 80V voltage. After the markers was separated, the voltage was changed to 120V.Afterwards the protein on PAGE gel were electrotransferred to a PVDF membrane at 200 mA for 1 h. The PVDF membrane was blocked with 5% milk blocking solution for 2 hours. Then, PVDF membrane were incubated with the diluted primary antibodies overnight at 4 ° C. The antibodies used in this study were listed in Supplementary Table S2. After washing with PBST, the PVDF membrane was incubated with a 1:1000 diluted second antibody conjugated with HRP for 1 hour. Exposure was performed under the Tanon5200 automatic chemiluminescence imaging system. The density of bands were analyzed using image J,normalized with GAPDH levels.

R 4.2.3, SPSS v26, and GraphPad Prism 8.0 were employed for statistical analysis. Two group data in normal distribution and homogeneity of variance were compared by student’s t-test (two-tailed). Variables in non-normal distribution were analyzed by Wilcoxon rank sum test. Multiple groups were compared by Kruskal-Wallis test and one-way analysis of variance (ANOVA).

To determine CD44 levels within normal and cancer samples, we investigated CD44 gene expression pattern in pan-cancer based on publicly available gene expression data. Firstly, the CD44 mRNA expression within different non-carcinoma samples was analyzed through HPA database. As a result, CD44 showed wide expression within various tissues, and was highly expressed in salivary gland, skin, bone marrow, pancreas, and bladder tissues (Figure 1A). According to the immunohistochemical analysis of the expression level of CD44 in cancer from the HPA database, the study found that CD44 exhibited moderate to strong membranous immunoreactivity, often accompanied with weak cytoplasmic staining, in a few cases of most cancer types. Most melanoma and cervical cancer cases were strongly stained. (Figure 1B). Further, TIMER2.0 platform was adopted for analyzing differential CD44 expression within cancer and non-carcinoma samples from TCGA dataset. Compared with normal tissues, CD44 was markedly up-regulated within many cancer tissues, including cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), glioblastoma multiforme (GBM), HNSC, kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), pheochromocytoma and paraganglioma (PIPG), rectum adenocarcinoma (READ), stomach adenocarcinoma (STAD), and thyroid carcinoma (THCA) cohort tissues. However, CD44 expression apparently decreased within bladder urothelial carcinoma (BLCA), lung adenocarcinoma (LUAD), prostate adenocarcinoma (PRAD), and uterine corpus endometrial carcinoma (UCEC) tissues (Figure 1C). Additionally, the CD44 expression in diverse cancer pathological stages was analyzed based on GEPIA database, as a result, CD44 expression related to the clinical pathological stages of breast invasive carcinoma (BRCA), pancreatic adenocarcinoma (PAAD), skin cutaneous melanoma (SKCM), and STAD (Figure 1D). Based on HPA immunofluorescence staining of subcellular localization, CD44 protein was strongly stained on the plasma membrane of A-431, U2OS, and U-251MG cells (Figure 1E), and CD44 protein was also detected on the Golgi apparatus and extracellularly secreted (Figure 1F).

Correlation between CD44 expression and OS as well as DSS was evaluated through univariable COX regression and Kaplan-Meier survival analyses. As suggested by the forest plot for univariable COX regression, high CD44 expression was related to dismal OS in thymic epithelial neoplasms(THYM), STAD, PAAD, and MESO brain lower grade glioma (LGG) [hazard ratio (HR)>1, P<0.05), whereas CD44 served as the good prognostic factor for OS in THCA (HR<1, P<0.05) (Figure 2B). Additionally, CD44 level was related to adverse DSS of BLCA, KIRC, HNSC, PAAD, STAD and LGG (HR>1, P<0.05), but CD44 was a good prognostic factor for DSS of BRCA cohort (HR<1, P<0.05) (Supplementary Figure S2B). As shown by the Kaplan-Meier survival analysis, CD44 up-regulation was indicative of shortened OS and DSS of GBM, BLCA, STAD, liver hepatocellular carcinoma (LIHC), PAAD, and KIRC, while it was associated with prolonged OS or DSS in UCEC, THCA and BRCA (Figure 2A; Supplementary Figure S2A).

Next, the cBioPortal database was used to analyze the overall genome alternations of CD44 in 10,967 samples from 32 types of cancer in TCGA. The types of CD44 mutations mainly included amplification, mutation, and structural variation. The mutation rate of CD44 in STAD and ESCA was the highest (>6%), and the main mutation type was amplification. In addition, CD44 had the highest proportion of structural variations in UCEC (Figure 3A). In the mass, the main mutation type of CD44 in pan-cancer was amplification, and the overall genome alternation rate of CD44 in pan-cancer was 2.2% (Figure 3B). The mutation types, quantities, and sites of CD44 gene alternations were explored using the cBioPortal tool. CD44 presented 114 mutations with 0~742 amino acids, and the frequency of somatic mutation was 0.9%. Among them, missense mutation was predominant, with a total of 83 mutation sites and 9 fusion mutation sites (Figure 3C). By examining the effect of different types of gene copy number alterations (CNAs) on CD44 gene expression, amplification was the most common and was associated with an increase in CD44 expression (Figure 3D). To investigate the association between CD44 gene mutations and alternations and clinical outcomes in, we found that CD44 gene mutations were associated with shortened OS (log-rank P=0.0192),as well as shortened DFS (log-rank P=0.0340) in esophageal adenocarcinoma and shortened DSS (log-rank P=0.0474) in sarcoma (Figure 3E). Furthermore, in TCGA-SKCM cohort, diversity of mutation profile in the CD44 high expression group and the CD44 low expression group were observed.GPR98, DSCAM, FAT3, PTPRT, FLG, MGAM, USH2A, and SPHKAP had higher mutation rates in the CD44 ^highgroup, while DNAH7, DNAH8, HYDIN, XIRP2, NRAS, ZFHX4, and MUC17 had higher mutation rates in the CD44 ^low group. Moreover, the CD44^high group had decreased BRAF somatic mutation frequency but increased CSMD1 and ANK3 somatic mutation frequencies compared to the CD44 ^low group (Figures 3F, G).

For exploring the CD44 molecular mechanisms affecting cancer occurrence biological processes, we performed pathway enrichment analysis on CD44 and its associated genes. GO enrichment analysis was conducted on 100 genes screened by GEPIA2 with the highest correlation with CD44 in pan-cancer. These genes were involved in the following biological processes: cell adhesion, nutrient binding, organelle formation, and tissue organ formation (Figure 4A). The relation of CD44 with 14 functional states within pan-cancer was analyzed based on CancerSEA database. As a result, CD44 was closely related to tumor-promoting biological processes, such as metastasis, angiogenesis and epithelial-mesenchymal transition (EMT) of AML, LUAD, and RCC (Figure 4B). To clarify the specific signaling pathways regulated by CD44, we performed GSEA and confirmed that CD44 mainly participated in inflammatory responses, interferon, interleukin, EMT, and KRAS signaling pathways in pan-cancer (Figure 4C). Further enrichment analysis of CD44 with Kyoto Encyclopedia of Genes and Genomes (KEEG) and Hallmark gene sets in breast cancer and melanoma cohorts from TCGA showed that CD44 was involved in the following immune-related pathways in melanoma: antigen processing and presentation; B cell receptor signaling pathway; T cell receptor signaling pathway; and IFNγ response. CD44 was involved in the chemokine signaling pathway, Fc receptor-mediated phagocytosis, natural killer (NK) cell-mediated cytotoxicity, and T cell receptor signaling pathway in breast cancer (Figure 4D). Therefore, these findings indicated that CD44 is crucial for tumor progression and immune regulation.

For exploring the effect of CD44 on TME, CIBERSORT deconvolution algorithm was adopted for calculating overall correlation landscape of CD44 with immune cell infiltration in 33 cancer types. The results indicated that CD44 exhibited significant heterogeneity in infiltrating degrees of immune cells within diverse cancer types. In general, however, CD44 was positively related to infiltration of macrophages, neutrophils, and CD4⁺ memory T cells, but negatively related to T follicular helper cells, B cells, NK cells, and regulatory T cells (Tregs) in most cancers. From the perspective of cancer types, CD44 was most significantly positively related to immune cell infiltration within testicular germ cell tumors (TGCTs) (Figure 5A). Spearman’s correlation analysis revealed that CD44 was correlated with 14 major immune checkpoints and co-stimulating factors within diverse cancer types. CD44 showed positive relation to the sub-immune checkpoint in most cancers, including TGCTs, adrenocortical carcinoma (ACC), KICH, LGG, LIHC, LUAD, and ovarian serous cystadenocarcinoma (OV) (Figure 5B).

Because chemokines have important effects on immune cell migration into cancers, we investigated the effect of CD44 on chemokines and chemokine receptors. The correlation of CD44 with the main chemokines of the CXC subfamily and CC subfamily was analyzed, and a heatmap was used to visualize chemokines with a Pearson correlation coefficient of P<0.05. In most cancers, CD44 was positively correlated with chemokines of the CXC subfamily, while CD44 was negatively correlated with most chemokines in ESCA. In addition, CD44 was negatively correlated with multiple chemokines of the CC subfamily in the HNSC-HPV⁺ and ESCA cohorts (Figure 5C). CD44 was positively correlated with most chemokine receptors in KICH, KIRC, KIRP, LGG, LIHC, LUAD, STAD, TGCT, and THYM. On the contrary, CD44 showed negative relation to most chemokine receptors within ESCA and LUSC (Figure 5D).

TMB and MSI represent two important biomarkers used to predict the ICIs therapeutic effect. CD44 expression showed positive relation to the TMB values within COAD, LGG, and UCEC, but negative relation to the TMB levels within BLCA, BRCA, ESCA, LIHC, LUAD, LUSC, PRAD, and THCA (Figure 5E). Additionally, CD44 showed positive relation to the MSI values in COAD, READ, and UCEC, but negative relation to the MSI values of DLBC, ESCA, KIRC, LGG, LUAD, PCPG, PRAD, and TGCT (Figure 5F). On the whole, CD44 exerts an essential impact on TME, including immune checkpoints, immune cell infiltration, chemokines, MSI and TMB, suggesting that CD44 may be novel key target for immunotherapy.

Six were downloaded from GEO database containing 18 patients of six types of cancer. A total of 47,023 cells were obtained after performing the quality control process (Supplementary Figure S4). After reclustering, a total of ten major cell clusters were obtained as follows: three stromal cell types [endothelial cells (PECAM1), epithelial cells (EPCAM and KRT18), and cancer-associated fibroblasts (CAFs; COL1A1)]; seven immune cell types [CD4+ T cells (CD4), CD8+ T cells (CD8A), B cells (CD79A and MS4A1), plasma cells (JCHAIN), macrophages (CD68), monocytes (CD14 and FCGR3A), and mast cells (TPSB2)] (Figure 6A; Supplementary Figure S4). The cells derived from different tumor tissues and different datasets were evenly distributed and did not show obvious disease specificity (Figure 6B). This work examined CD44 expression and distribution in stromal cells and immune cells at the single-cell level. CD44 had a higher expression in immune cells compared to stromal cells (Figure 6C). Although CD44 was widely distributed in B cells and T cells, it had higher expression within monocytes and mast cells. Comparison of the expression differences of CD44 among different cell clusters in the six cancer types demonstrated that CD44 had the highest expression in mast cells of KIRC, STAD, HNSC and CRC. In addition, monocytes also had high expression of CD44 in HNSC and CRC. Intertumoral heterogeneity analysis indicated that CD44 had the broadest distribution of cell subsets in STAD (Figure 6D). According to Monocle2 analysis, CD44 gene expression altered depending on cell trajectory with the differentiation of various immune cells. In the T lineage development trajectory, CD44^high T cells were located at the end of each branch point of the development tree, while CD44^low T cells were located in the early CD4⁺ T cell cluster development and late CD8⁺T cell differentiation. However, expression of CD44 in the B lineage was evenly distributed throughout the cell development trajectory. With the differentiation trajectory of myeloid lineage cells, the expression of CD44 showed an increasing trend, especially in monocyte subsets (Figure 6E).

To clarify the regulatory role of CD44 in cell-cell communication, we selected monocytes which have a high CD44 expression as the research object. CellChat was used to analyze the differences in signaling interaction patterns between CD44⁺ monocytes and CD44^- monocytes. When CD44⁺ monocytes acted as the receiver of incoming signals, the strength of communication increased when interacting with CAFs, epithelial cells, and macrophages (as signaling senders) compared to CD44^- monocytes (Figure 7A). When monocytes acted as the sender of outgoing signals, the interaction of CD44⁺ monocytes with CAFs or endothelial cells was more intense than that of CD44^- monocytes. Similarly, in the incoming and outgoing information flows mediated by different signaling pathways, CD44⁺ monocytes had a higher overall information flow intensity than CD44^- monocytes, including incoming signaling patterns mediated by MHC-II, CD22, CD23, CD45, and ICAM, as well as outgoing signaling patterns mediated by ICAM, ITGB2, CD45, CD86, and ADGRE5 (Figure 7B). In addition, ligand-receptor (L-R) pairs mediated by CD44 exhibited high activity in the interaction between endothelial cells and CAFs with CD44⁺ monocytes. In particular, L-R pairs formed by the combination of CD44 and the collagen family genes (COL1A1, COL1A2, COL4A1, COL4A2, COL6A1, and COL6A2), which were highly expressed in CAFs, were observed in the interaction between CAFs and CD44⁺ monocytes. In the signaling communication between endothelial cells and monocytes, the L-R pairs formed by CD44 and the adhesion protein gene family (LAMA4, LAMA5, LAMB1, LAMB2, and LAMC1) were also highly active in the CD44⁺ monocyte subgroup (Figure 7C).

The TIDE database was used to evaluate the possibility of CD44 as a new predictive marker for ICB. Among the 25 immunotherapy cohorts, CD44 had an AUC of >0.5 in 10 cohorts and >0.7 in 3 cohorts (Uppaluri2020_PD1_HNSC_Pre, Uppaluri2020_PD1_HNSC_Pos, and Nathanson 2017_CTLA4_Melanoma_Pre). In the melanoma anti-CTLA4 cohort, the AUC of CD44 was 0.8, indicating a higher predictive ability than any other marker, except MSI (AUC = 0.9) (Figure 8A).

For the in vitro models, we analyzed the B16 (mouse melanoma), EMT6 (mouse breast cancer), KPC (mouse pancreatic cancer), and MC38 (mouse colon cancer cells) models treated with IFNβ, IFNγ, and TNF-α. Compared to baseline, there was in increase in CD44 expression with the following treatments in the specific models: IFNβ in the B16 and MC38 models; IFNγ in B16 and EMT6 models; and TNF-α in MC38 model. The in vivo mouse model allowed analysis of ICB treatment, in which the responders of the CT26 anti-PD1 (GSE139475) cohort had significantly decreased CD44 expression compared to baseline but had no significant difference in CD44 expression compared to non-responders (Figure 8B). Moreover, we investigated the impact of CD44 on immunotherapy effect on immunotherapy cohorts in TIDE database. In Zhao2019_PD1-Glioblastoma and Gide2019_PD1+CTLA4-Melanoma cohorts, high-CD44-expression patients exhibited the unfavorable OS relative to low-CD44-expression counterparts, representing a worse immunotherapy outcome. However, high-CD44-expression patients exhibited an extended PFS compared with low-CD44-expression patients in Gide2019_PD1+CTLA4-Melanoma and Prat2017_PD1-NSCLC-HNSC-Melanoma cohorts (Figure 8C). Further, we verified the value of CD44 as a predictor for ICB efficacy, as a result, CD44 down-regulation was related to better immunotherapy response in a melanoma cohort undergoing anti-PD-1 therapy (GSE91061), and CD44 expression of non-response group markedly elevated (Figure 8D). As for CR (complete response) group, the proportion of low-CD44-expression patients was 100%, while those in the PR (partial Response), PD (progressive disease), and SD (stable disease) groups increased to 28.6%, 44.1%, and 47.1%, respectively (Figure 8E).

Two independent small interfering RNAs (siRNAs) were prepared for silencing CD44 expression. CD44 siRNA was transfected into HCT-116 cells for a 24-h period. As demonstrated by qRT-PCR analysis, CD44 expression of CD44 siRNA2 group markedly decreased in comparison with negative control (NC) group (Supplementary Figure S3A). According to CCK-8 results, CD44 silencing markedly suppressed the proliferation of HCT-116 cells (Figure 9A). Similarly, colony formation assay revealed that knockdown of CD44 decreased colony formation of HCT-116 cells (Figure 9B). For exploring how CD44 affected HCT-116 cell migration and invasion, we conducted Transwell and scratch assays. Compared with NC group, CD44 knockdown inhibited HCT-116 cell migration and invasion (Figures 9C, D). Based on these results, CD44 down-regulation markedly inhibited HCT-116 cell growth, migration and invasion.

An in vitro model of macrophage polarization was constructed by first inducing THP-1 monocytes to differentiate into M0 macrophages using PMA and then incubating them with IFN-γ and LPS to differentiate them into classical M1-like macrophages or with IL-4 and IL-13 for differentiating in M2-like macrophages (Figure 10A). We conducted qRT-PCR for detecting macrophage markers levels in different phenotypes [M1 markers: CD86, tumor necrosis factor alpha (TNFα), and IL-6; M2 markers: CD163, CD206, transforming growth factor beta (TGFβ), and IL-10], confirming that macrophages with different phenotypes were successfully induced (Figures 10B–D). CD44 levels within diverse phenotype macrophages were analyzed, and CD44 expression significantly increased in M2 macrophages (Supplementary Figure S3B). For confirming the key effect of CD44 on maintaining M2 phenotype of macrophages, CD44 in M2-like macrophages was silenced (Supplementary Figures S3C, S5A). CD44 silencing significantly decreased the levels of CD163, CD206, TGFβ, and IL-10 in M2-like macrophages (Figure 10G). Knockdown of CD44 increased the expression of TNF-α in M2-like macrophages, however, the change of CD86 and IL-6 was not statistically significant (Figure 10F). Western blot assay comes to the similar conclusion (Supplementary Figure S5B).These results indicated that CD44 is essential for maintaining M2-like macrophage polarization. In the macrophage and CRC cell co-culture system, M2 macrophages significantly enhanced the migration of CRC cells compared to M0 and M1 macrophages, but this migration was significantly inhibited after knockdown of CD44.