Abstract

   Metabolic reprogramming of tumor cells dynamically reshapes the
   distribution of nutrients and signals in the tumor microenvironment
   (TME), affecting intercellular interactions and resulting in metabolic
   immune suppression. Increased glucose uptake and metabolism are
   characteristic of many tumors. Meanwhile, the progression of colorectal
   carcinoma (CRC) relies on lipid metabolism. Therefore, investigating
   the role of glucolipid metabolic reprogramming on tumor immunity
   contributes to identifying new targets for immune suppression
   intervention in CRC. Our previous work demonstrated that SIRT1 is the
   hub gene involved in glucolipid metabolic conversion in CRC. Here, it
   is found that upregulated SIRT1 in CRC cells increases Treg
   functionality by promoting the secretion of CX3CL1. The CX3CL1‐CX3CR1
   signaling activated transcription factors SATB1 and BTG2, promoting the
   differentiation of TCF7^+ Treg cells into functionally enhanced
   TNFRSF9^+ Treg cells. Multiplex immunofluorescence (mIHC) analysis of a
   CRC tissue microarray confirmed the promoting effect of CX3CL1 on Treg
   infiltration. Additionally, the therapeutic efficacy of CX3CR1
   inhibitor monotherapy and combination therapy is validated with the
   PD‐1 antibody in the humanized subcutaneous CRC mouse model. This study
   elucidates a potential mechanism that metabolic reprogramming of cancer
   cells collaborates with subsequent immunosuppression to promote CRC
   progression.

   Keywords: Colorectal carcinoma, CX3CL1, Immunosuppression, Regulatory T
   cell, SIRT1
     __________________________________________________________________

   This study discovers that SIRT1, a hub gene involved in glucolipid
   metabolic conversion in colorectal carcinoma (CRC), stimulates CX3CL1
   secretion in CRC cells by activating FOXO1. The CX3CL1‐CX3CR1 signaling
   promotes the differentiation of TCF7^+ regulatory T cells (Tregs) into
   an enhanced immunosuppressive TNFRSF9^+ Treg phenotype. CRC cells exert
   metabolic immunosuppression on the tumor microenvironment via the
   SIRT1‐CX3CL1 axis.

   graphic file with name ADVS-12-2404734-g006.jpg

1. Introduction

   Metabolic reprogramming and immune escape are two fundamental hallmarks
   of cancer. Nevertheless, they are not mutually exclusive. The dynamic
   variation of nutrients and signals within the TME drives the metabolic
   plasticity of cancer cells and facilitates the immune suppression of
   effector cells.^[ [60]^1 ^] In contrast to the traditional Warburg
   effect, the progression of CRC relies on lipid metabolism.^[ [61]^2 ^]
   However, whether glucolipid metabolic reprogramming affects the
   crosstalk between tumor and immune cells is elusive. Our previous
   studies revealed that the deacetylase SIRT1, which is widely
   distributed in the cytoplasm and nucleus, is a metabolic switch between
   glycolysis and fatty acid oxidation, its upregulation enables tumor
   cells to adapt to glucose deprivation and promotes tumor progression.^[
   [62]^3 ^] Given that high SIRT1 expression in tumor cells can affect
   the expression and secretion of cytokines, thus influencing tumor
   growth and metastasis, further exploration of whether SIRT1 inhibits
   tumor immunity is warranted.^[ [63]^4 ^]

   Chemokines and their receptors have been a hot topic in tumor immunity
   research. The exocrine cytokine CX3CL1 (C‐X3‐C Motif Chemokine Ligand
   1, also known as Fractalkine) has been shown to play an important role
   in the immune response. It is traditionally believed that CX3CL1 exerts
   an antitumor effect by recruiting CD8^+ T cells. However, CX3CL1
   participates in tumor invasion and metastasis in some tumors, leading
   to a poor prognosis.^[ [64]^5 ^] Recent studies have indicated that the
   interaction between tumor cells and other cells in the microenvironment
   mediated by CX3CL1, such as macrophages and mesenchymal stem cells, is
   related to developing an immunosuppressive environment.^[ [65]^6 ,
   [66]^7 ^] Therefore, a comprehensive analysis of the mechanism of
   CX3CL1 production and function contributes to unraveling its effector
   network in TME.

   Regulatory T cells (Tregs) are one of the main cell types that exert
   immunosuppressive effects in the tumor microenvironment. Their
   infiltration is closely associated with adverse clinical phenotypes
   such as treatment resistance, immune evasion, and metastasis. Recent
   studies have shown that tumor cells can recruit Treg cells to exert
   immunosuppressive effects and promote tumor progression by secreting
   certain cytokines, such as CCL17 and CCL20.^[ [67]^8 , [68]^9 ^]
   Notably, with the emergence of single‐cell RNA sequencing (scRNA‐seq)
   technology, it has become increasingly recognized that Treg cells
   exhibit heterogeneity in the tumor environment. A pan‐cancer analysis
   indicated that TNFRSF9^+ Treg (TNF Receptor Superfamily Member 9, also
   known as 4‐1BB and CD137) cells are widely present in various types of
   tumor tissues and exert strong immunosuppressive effects.^[ [69]^10 ^]
   The transcription factor TCF‐1 (encoded by TCF7) plays a crucial role
   in the development of Treg cells, but in CRC, TCF‐1‐deficient Treg
   cells exhibit stronger immunosuppressive effects than WT Treg cells.^[
   [70]^11 ^] These studies suggest that further elucidating the impact of
   cytokines on Treg phenotypes and functions is highly important.

   In this study, we discovered that SIRT1 promotes the upregulation of
   CX3CL1 in CRC cells. CX3CL1, in turn, enhances the function of Tregs
   and suppresses antitumor immunity. Our findings suggested that tumor
   metabolic reprogramming and immune evasion cooperate to promote tumor
   progression.

2. Results

2.1. SIRT1 is Closely Related to the Immunosuppressive Tumor Microenvironment

   The rapid energy provision and the production of key precursors for the
   biosynthesis of macromolecules make glycolysis the preferred energy
   supply method for cancer cells. However, metabolomic analysis of a
   large cohort study has found that the progression of CRC was highly
   dependent on lipid metabolism,^[ [71]^2 ^] suggesting that glucolipid
   metabolic reprogramming is involved in the development of CRC. To
   explore the specific roles of glycolysis and lipid metabolism in CRC
   progression, we conducted a univariate Cox regression analysis of all
   glycolysis‐ and lipid metabolism‐related genes in CRC samples from the
   TCGA database (n = 392). We identified 126 genes that were
   significantly associated with poor or prolonged survival in CRC
   (univariate Cox p < 0.05). The expression matrix of these genes served
   as the input for consensus clustering. Four groups were identified by
   consensus clustering: C1, C2, C3, and C4 (Figure [72] 1A). Kaplan‒Meier
   (K‒M) survival analysis indicated that the prognosis of the C4 group
   was significantly worse than that of the other groups (Figure [73]1B).
   Subsequently, we performed gene set enrichment analysis (GSEA) of the
   differentially expressed genes (DEGs) between patients in the C4 group
   and those in the C1 group. The results revealed a significant
   enrichment of lipid metabolism‐related gene sets and a markedly
   decreased enrichment of effector immune response‐related gene sets. At
   the same time, changes in glycolysis‐related pathways were not
   prominent (Figures [74]1C,D and [75]S1A, Supporting Information).
   Correspondingly, we observed increased enrichment of pathways
   associated with macrophage differentiation and regulatory T‐cell
   differentiation related to immune suppression in patients in the C4
   group (Figure [76]S1B,C, Supporting Information). These results
   indicated that the enhanced lipid metabolism resulting from glucolipid
   metabolic reprogramming might promote the progression of CRC by
   reshaping the tumor immune microenvironment (TIME).

Figure 1.

   Figure 1
   [77]Open in a new tab

   SIRT1 is closely related to the immunosuppressive tumor
   microenvironment. A) Consensus clustering of glucolipid
   metabolism‐related genes, which are significantly associated with
   prognosis, identifying four groups in the TCGA colorectal cancer cohort
   (TCGA‐COAD, n = 392). B) Kaplan‐Meier estimates of survival for
   patients with different glucolipid subtypes in the TCGA‐COAD cohort. C)
   The enrichment score of the “Lipid metabolic process” in C4 patients
   versus C1 patients was analyzed by GSEA using RNA‐seq data from the
   TCGA‐COAD cohort. D) Enrichment score of the “Immune effector process”
   in C4 patients versus C1 patients, analyzed by GSEA using RNA‐seq data
   of TCGA‐COAD cohort. E) The correlation of 10 main tumor‐infiltrating
   cell types with SIRT1 in CRC tissues. Red: positive correlation; blue:
   negative correlation. F) Representative gross appearance (left) of the
   subcutaneous allografts in the indicated groups. Tumor growth curves
   (right) of C57BL/6J mice inoculated with Sirt1‐KO or control MC38
   allografts. Statistical significance was assessed by student t‐test of
   variance. ***p < 0.001. G) Flow cytometry analysis of the ratio of
   Cd206^+ cells in Cd11b^+ cells. n = 7, student t‐test of variance,
   ***p < 0.001. H) Flow cytometry analysis of the ratio of Cd25^+Cd127^−
   cells in Cd4^+ T cells. n = 7, student t‐test of variance,
   ***p < 0.001. I–K) Flow cytometry analysis of the ratio of Pd‐1^+ cells
   in Cd8^+ T cells (I), Tim‐3^+ cells in Cd8^+ T cells (J), and Ctla4^+
   cells in Cd8^+ T cells (K). n = 7, student t‐test of variance,
   ***p < 0.001. L) Elisa of effector cytokines IFNγ (left), TNFα (right)
   from Sirt1‐KO or control MC38 allografts. n = 7, student t‐test of
   variance, ***p < 0.001.

   Our previous study revealed that SIRT1 was the hub that facilitated the
   metabolic shift from glycolysis to lipid metabolism in CRC cells.^[
   [78]^3 ^] Here, we assessed whether SIRT1 shaped the tumor immune
   environment. First, we examined the correlation between SIRT1
   expression and the levels of tumor‐infiltrating immune cells using the
   CIBERSORT algorithm.^[ [79]^12 ^] We found that SIRT1 was positively
   correlated with the infiltration of regulatory T cells (Tregs), but
   negatively correlated with CD8^+ T cells (Figure [80]1E). Moreover, we
   discovered a positive correlation between SIRT1 mRNA expression and the
   expression of T‐cell exhaustion signature genes in CRC samples by
   performing Gene Expression Profiling Interactive Analysis (GEPIA)
   (Figure [81]S1D, Supporting Information). Moreover, SIRT1 expression
   was positively correlated with markers of M2 macrophages and Tregs
   (Figure [82]S1E,F, Supporting Information). These results suggested
   that SIRT1 participated in shaping the immunosuppressive tumor
   microenvironment.

   Next, we established Sirt1 knockout (Sirt1‐KO) MC38 mouse CRC cell
   lines using lentiviral vectors and confirmed gene knockout efficiency
   by western blot analysis (Figure [83]S1G, Supporting Information).
   Next, we constructed a subcutaneous allograft tumor model in C57BL/6
   mice and generated tumor growth curves for the wild‐type (WT) MC38 and
   MC38‐Sirt1KO groups. When the living tumor volume in any tumor‐bearing
   mice exceeded 2000 mm^3, all the mice were euthanized, and the tumor
   tissues were harvested and weighed. The experimental results indicated
   a significant inhibition of tumor progression in the Sirt1‐KO group
   (Figure [84]1F). Furthermore, we processed all tumor tissues by
   digesting them into single cells and then analyzed the infiltration of
   immune cells using flow cytometry. The results showed that the Sirt1‐KO
   group exhibited a significant decrease in the infiltration of
   immunosuppressive cells, such as M2‐type macrophages (Cd11b^+Cd206^+)
   and Treg cells (Cd4^+Cd25^+Cd127^−), within the tumor tissue
   (Figure [85]1G,H). Meanwhile, we found that the expression level of
   SIRT1 did not affect the infiltration of myeloid‐derived suppressor
   cells (MDSCs, Cd11b^+Gr1^+) and dendritic cells (DCs, Cd11c^+Cd86^+)
   (Figure [86]S1H,I, Supporting Information). Moreover, the expression of
   exhaustion‐related immune checkpoint markers (Pd‐1, Tim‐3, and Ctla4)
   was notably reduced in CD8^+ T cells within the Sirt1‐KO cohort
   compared to the control group (Figure [87]1I–K). Correspondingly, we
   detected significantly increased levels of IFN‐γ and TNF‐α in the
   supernatants of the Sirt1‐KO single‐cell cultures (Figure [88]1L).
   These findings were consistent with previous bioinformatics analysis
   results, indicating that SIRT1 plays a crucial role in shaping the
   immunosuppressive tumor microenvironment.

2.2. SIRT1 Promotes the Expression of CX3CL1 in CRC Cells

   We subsequently evaluated the impact of alterations in SIRT1 expression
   levels within tumor cells on tumor immunity. Previous studies have
   highlighted the pivotal role of chemokines in facilitating the
   interplay between tumor cells and immune cells, orchestrating the
   generation and recruitment of immune cells that foster a
   tumor‐promoting microenvironment.^[ [89]^13 ^] Therefore, we
   hypothesized that SIRT1 might be associated with the secretion of
   certain chemokines. To test this hypothesis, we first constructed
   overexpression and knockdown plasmids for SIRT1, and transduced them
   into two human CRC cell lines, HCT116 and SW480, to obtain cell models
   with either SIRT1 overexpression (OE) or SIRT1 knockdown (KD).
   Subsequently, we isolated T cells from human peripheral blood
   mononuclear cells (PBMCs) and activated them with anti‐CD3, anti‐CD28
   antibodies, and IL‐2 for 72 h. Then, we seeded tumor cells in the
   basolateral chamber of a transwell system and added the activated T
   cells to the apical chamber (Figure [90]S2A, Supporting Information).
   The coculture mixture was maintained for 72 h, after which cell culture
   supernatants were collected to assess the effector cytokines IFN‐γ and
   TNF‐α expression in the different groups (Figures [91] 2A,B and
   [92]S2B,C, Supporting Information). We also evaluated the expression
   levels of PD1 and Ki67 in CD8^+T cells to assess their exhaustion state
   (Figures [93]2C,D and [94]S2C,D, Supporting Information). Compared to
   those cultured with tumor cells from the control group, T cells
   cocultured with tumor cells from the SIRT1‐OE group exhibited reduced
   IFN‐γ and TNF‐α production, decreased Ki‐67 expression, and increased
   PD‐1 expression. Coculture of T cells with SIRT‐KD tumor cells had the
   opposite effect. These experiments indicated that SIRT1 in tumor cells
   could suppress anti‐tumor immunity through paracrine signaling.
   Additionally, we found that the expression level of SIRT1 was
   negatively correlated with the apoptosis rate of tumor cells (Figure
   [95]S2E,F, Supporting Information). These data suggested that the
   co‐culture system served as an excellent in vitro model for
   investigating the one‐way influence of tumor cells on T cells.

Figure 2.

   Figure 2
   [96]Open in a new tab

   SIRT1 promotes the expression of CX3CL1 in colorectal tumor cells. A,B)
   Elisa of effector cytokines IFNγ (A), TNFα (B) in the supernatant of
   activated T cells co‐culture with indicated HCT116 cells. n = 6,
   one‐way ANOVA test of variance, *p <0.05, **p < 0.01, ***p < 0.001. C)
   Activated T cells were co‐cultured with indicated HCT116 cells. Flow
   cytometry analysis of the ratio of PD‐1^+ cells in CD8^+ T cells.
   n = 7, one‐way ANOVA test of variance, **p < 0.01, ***p < 0.001. D)
   Activated T cells were co‐cultured with indicated HCT116 cells. Flow
   cytometry analysis of the ratio of Ki67^+ cells in CD8^+ T cells.
   n = 6, one‐way ANOVA test of variance, **p < 0.01, ***p < 0.001. E) The
   correlation of SIRT1 with all chemokines, analyzed by GEPIA. F) The
   SIRT1 and CX3CL1 levels of SIRT1‐KD and control HCT116 (left) or
   SIRT1‐OE and control HCT116 (right) were assessed using immunoblotting.
   G,H) Confocal assay of SIRT1 (green) and CX3CL1 (red) in SIRT1‐OE,
   SIRT1‐KD, or control HCT116, and the fluorescence intensity was
   calculated by Image J software. Representative images of SIRT1‐KD,
   SIRT1‐OE, or control HCT116 (F) and correlation matching between SIRT1
   and CX3CL1 (G) are shown. Scale bars, 20µm. I) Based on the median mRNA
   expression value, TCGA‐COAD samples were divided into two groups:
   SIRT1^LowCX3CL1^Low and SIRT1^HighCX3CL1^High. Kaplan‐Meier analysis of
   the overall survival rate of patients in different groups. Red line:
   SIRT1^LowCX3CL1^Low; Blue line: SIRT1^HighCX3CL1^High. J) Multivariate
   logistic regression analysis of odds ratio (ORs) of different
   clinic‐pathological characteristics showed that compared with
   SIRT1^LowCX3CL1^Low profile, SIRT1^HighCX3CL1^High profile had a higher
   risk of large tumor grade, invasive depth, and metastasis.

   Next, we analyzed the correlation between SIRT1 and the expression of
   all chemokines using GEPIA in the TCGA database. The results
   demonstrated that among chemokines, CX3CL1 was the most strongly
   correlated with SIRT1 in CRC (Figure [97]2E). Subsequently, we
   performed immunoblotting and immunofluorescence analyses to confirm the
   positive correlation between the expression levels of SIRT1 and CX3CL1
   further (Figure [98]2F–H). We then assessed whether SIRT1 and CX3CL1
   could be combined to indicate CRC prognosis. CRC samples from the TCGA
   database were divided into high and low‐expression groups based on the
   median expression levels of SIRT1 and CX3CL1, respectively. The
   intersection of expression groups for both genes revealed that 188
   samples had the same expression patterns for SIRT1 and CX3CL1 (96
   samples had high expression of both SIRT1 and CX3CL1, and 92 samples
   had low expression of both SIRT1 and CX3CL1). Survival analysis of
   these samples revealed that patients with a SIRT1^HighCX3CL1^High
   profile had a significantly lower overall survival rate than those with
   a SIRT1^LowCX3CL1^Low profile (Figure [99]2I). Moreover, the
   upregulation of SIRT1 and CX3CL1 was significantly correlated with
   several aggressive clinicopathological features of CRC, including
   advanced tumor stage, increased lymphatic and distant metastasis rates,
   and increased tumor invasion depth (Figure [100]2J). These findings
   suggested that the SIRT1‐CX3CL1 axis was critical in promoting the
   progression and spread of CRC and could serve as a prognostic biomarker
   for identifying patients with a high likelihood of poor outcomes.

2.3. SIRT1 Activates FOXO1 to Promote CX3CL1 Expression

   Next, we explored the mechanisms by which SIRT1 enhances the expression
   of CX3CL1. First, we performed RNA sequencing (RNA‐seq) on WT and
   SIRT1‐OE HCT116 cells. To identify signaling pathways that SIRT1 might
   directly regulate, we conducted a Kyoto Encyclopedia of Genes and
   Genomes (KEGG) analysis of the DEGs in the SIRT1‐OE group. We found
   significant enrichment of the FoxO signaling pathway in the SIRT1‐OE
   cells (Figures [101] 3A and [102]S3A, Supporting Information). The FoxO
   transcription factor family consists of four members: FOXO1, FOXO3,
   FOXO4, and FOXO6. Screening based on the cBioPortal website revealed
   that FOXO1 and FOXO3 exhibited the strongest correlation with SIRT1
   (Figure [103]3B). GEPIA analysis indicated that the expression of
   CX3CL1 was significantly positively correlated with that of FOXO1 but
   not with that of FOXO3 in CRC patients (Figure [104]3C). Moreover, the
   expression of SIRT1 was positively correlated with that of FOXO1 in
   various gastrointestinal cancers (Figure [105]S3B, Supporting
   Information). To further validate the FOXO1 function, we separately
   performed FOXO1 or FOXO3 overexpression or knockdown experiments in
   HCT116 and SW480 cells and detected the expression of CX3CL1 in the
   culture supernatant via ELISA (Figure [106]S3C,D, Supporting
   Information) and western blot (Figures [107]3D,E and [108]S3F,G,
   Supporting Information). We observed that FOXO1 significantly affects
   the level of CX3CL1, while FOXO3 has a relatively minor impact on the
   level of CX3CL1 (Figure [109]S3E,H Supporting Information). Moreover,
   treatment of CRC cells with a FOXO1‐specific inhibitor (AS1842856)
   significantly suppressed the secretion of CX3CL1 (Figure [110]S3I,
   Supporting Information). These data indicated that SIRT1 promoted
   CX3CL1 secretion by regulating FOXO1.

Figure 3.

   Figure 3
   [111]Open in a new tab

   SIRT1 activates FOXO1 to promote CX3CL1 expression. A) The upregulated
   DEGs of SIRT1‐OE relative to SIRT1‐WT HCT116 cells were subjected to
   KEGG pathway enrichment analysis. The top 20 KEGG pathways with the
   most significant p‐values were displayed. B) Screening graph showing
   the relationship between SIRT1 and FOXO family members using cBioPortal
   in CRC samples. C) GEPIA analysis of the correlation between CX3CL1 and
   FOXO1 or FOXO3. D,E) FOXO1 and CX3CL1 levels of FOXO1‐KD, FOXO1‐OE, and
   control HCT116 (D) or SW480 (E) were assessed using immunoblotting.
   F,G) SIRT1 immune complexes were immunoprecipitated from HCT116 (F) or
   SW480 (G) cells and subjected to immunoblotting of FOXO1. H,I) Acetyl
   level of FOXO1 in SIRT1‐KD (H) or SIRT1 inhibitor EX527 (20 µM, 48 h)
   treated HCT116 cells (I) was assessed using immunoblotting. J,K)
   Confocal assay of SIRT1 (green) and FOXO1 (red) in SIRT1‐OE, SIRT1‐KD,
   or control HCT116 and the fluorescence intensity of FOXO1 in nuclear
   was calculated by Image J software. Representative images of SIRT1‐KD,
   SIRT1‐OE, or control HCT116 (J) and the statistical analysis of nuclear
   fluorescence intensity of CX3CL1 (K) are shown. One‐way ANOVA test of
   variance, ***p < 0.001. Scale bars, 20µm. L) The indicated plasmids
   were transfected into HCT116 cells, and the acetylation of FOXO1^WT and
   mutants was measured by immunoblotting. (M) CX3CL1 levels of cells in
   (L) were assessed using immunoblotting.

   Then, we used coimmunoprecipitation experiments to validate the
   endogenous SIRT1‐FOXO1 interaction in CRC cell lines
   (Figure [112]3F,G). Given that SIRT1 primarily functions as a
   deacetylase within cells, we treated CRC cells with shSIRT1 or the
   SIRT1‐specific inhibitor EX527. Both treatments resulted in elevated
   acetylation levels of the transcription factor FOXO1 (Figures [113]3H,I
   and [114]S3J,K Supporting Information). This finding indicates that
   SIRT1 acts as a deacetylase for FOXO1. Next, we assessed whether the
   acetylation level of FOXO1 affected its nuclear localization. We
   performed immunofluorescence staining to evaluate the subcellular
   localization of FOXO1 in SIRT1‐WT, SIRT1‐KD, and SIRT1‐OE HCT116 cells.
   The results showed that deacetylation promoted the nuclear accumulation
   of FOXO1, while acetylated FOXO1 exhibited significantly reduced
   stability within the nucleus (Figure [115]3J,K). To directly examine
   the impact of FOXO1 acetylation status on its function, we constructed
   five Flag‐tagged plasmids: FOXO1^K245R‐Flag, FOXO1^K248R‐Flag,
   FOXO1^K262R‐Flag, and FOXO1^K265R‐Flag and FOXO1^WT‐Flag, to simulate
   the deacetylated state of FOXO1. We knocked down the endogenous FOXO1
   in tumor cells to minimize its impact on the experimental outcomes
   (Figure [116]S3L, Supporting Information). Then we transfected five
   plasmids and found that each of the mutated plasmids reduced the
   acetylation levels of FOXO1, with the arginine mutation at position 262
   showing the most pronounced effect (Figure [117]3L). Subsequently,
   through western Blot detection, we found that the expression of CX3CL1
   is negatively correlated with the acetylation levels of FOXO1
   (Figure [118]3M). These results suggested that SIRT1 stabilizes FOXO1
   in the nucleus via deacetylation, allowing it to exert its
   transcriptional effects. In summary, the above experiments demonstrated
   that SIRT1 deacetylates FOXO1 to promote the expression of CX3CL1 in
   CRC cells.

2.4. CX3CL1 Promotes Immune Suppression by Mediating the Function of Treg
Cells

   Next, we investigated the potential role of CX3CL1 in developing an
   immunosuppressive microenvironment. First, GEPIA analysis revealed a
   positive correlation between the expression of CX3CL1 and T‐cell
   exhaustion markers in human CRC samples (Figure [119]S4A, Supporting
   Information). Then, we cocultured HCT116 cells with activated T cells
   in transwell chambers as previously described. We found that the
   exogenous addition of CX3CL1 significantly increased the expression of
   exhaustion‐related markers on CD8^+ T cells (Figure [120] 4A,B). Recent
   studies have indicated that early exhausted T cells exhibit strong
   proliferative capacity and retained tumor reactivity. In contrast,
   terminally exhausted T cells exhibit reduced proliferative capacity and
   decreased effector cytokines production.^[ [121]^14 ^] Correspondingly,
   we found that the exogenous addition of CX3CL1 reduced the
   proliferative capacity and suppressed the production of TNF‐α and IFN‐γ
   by CD8^+ T cells (Figures [122]4C and [123]S4B,C, Supporting
   Information). These results suggest that CX3CL1 regulates T‐cell
   exhaustion.

Figure 4.

   Figure 4
   [124]Open in a new tab

   CX3CL1 promotes immune suppression by mediating the function of Treg
   cells. A–C) Activated T cells co‐cultured with HCT116 cells received
   indicated treatment (CX3CL1, 200 ng ml^−1, 48 h), and flow cytometry
   analysis of the ratio of PD‐1^+ cells in CD8^+ T cells (A); the ratio
   of TIM‐3^+ cells in CD8^+ T cells (B); the ratio of Ki67^+ cells in
   CD8^+ T cells (C). n = 6, student t‐test of variance, ***p < 0.001. D)
   The seven‐step cancer immunity cycle scores of CX3CL1^High and
   CX3CL1^Low CRC samples from the TCGA database. E) Using the CICS as a
   phenotype, upregulated DEGs of CX3CL1High versus CX3CL1Low CRC samples
   were used to construct a WGCNA coexpression network; each color
   represents a type of module. The heatmap shows the correlation of each
   module with CICS. Red represents a positive correlation, while blue
   represents a negative correlation. *p <0.05, **p < 0.01, ***p < 0.001.
   F) The ClueGO plug‐in was used to perform GO enrichment for the core
   genes in brown modules. Each node represents a biological process;
   nodes with the same color have similar functions. Node size shows the
   significance of enrichment. G) The ClueGO plug‐in was used to perform
   KEGG enrichment for the core genes in brown modules. Each node
   represents a signaling pathway; nodes with the same color have similar
   functions. Node size shows the significance of enrichment. H) Activated
   T cells co‐cultured with HCT116 cells received indicated treatment, and
   flow cytometry analysis of the ratio of CD25^+CD127^− (Treg) cells and
   CD25^−CD127^+ (memory) cells in CD4^+ T cells. n = 6, student t‐test of
   variance, ***p < 0.001. I) Schematic diagram of constructing a
   humanized mouse subcutaneous transplant tumor model. J) NCG mice with
   subcutaneous xenografts (10^6 HCT116 cells) have tail vein injected
   CD4^+/CD8^+ mixed T cells. All mice were divided into two groups and
   received indicated treatment, and tumor tissues were extracted at the
   experimental endpoint (left). The tumor size was measured every three
   days using calipers to plot the tumor growth curve (right). Statistical
   significance was assessed by student t‐test of variance. n = 6,
   **p < 0.01. K–N) Tumor tissues in (J) were digested into single cells,
   and flow cytometry analysis of the ratio of PD‐1^+ cells in CD8^+ T
   cells (K); the ratio of TIM‐3^+ cells in CD8^+ T cells (L); the ratio
   of LAG‐3^+ cells in CD8^+ T cells (M) and the ratio of CD39^+ cells in
   CD8^+ T cells (N). n = 6, student t‐test of variance, ***p < 0.001. O)
   Tumor tissues in (J) were digested into single cells, and flow
   cytometry analysis of the ratio of CD25^+CD127^− (Treg) cells and
   CD25^−CD127^+ (memory) cells in CD4^+ T cells. n = 6, student t‐test of
   variance, ***p < 0.001. P) TGF‐β (left) and IL‐10 (right) levels of
   JMS‐17‐2 treated or control HCT116 xenografts (single cell culture
   supernatant after tissue lysis) were assessed by ELISA assay. n = 6,
   student t‐test of variance, ***p < 0.001.

   To investigate the mechanism by which CX3CL1 mediates T‐cell
   exhaustion, we calculated the cancer‐immunity cycle score (CICS)
   ([125]http://biocc.hrbmu.edu.cn/TIP/)^[ [126]^15 ^] using mRNA
   expression data for the 28 samples with the highest CX3CL1 expression
   and the 28 samples with the lowest CX3CL1 expression among TCGA CRC
   samples. The results showed that the high‐CX3CL1‐expression group had a
   significantly greater infiltration level of Tregs and lower T‐cell
   effectiveness (steps 5–7) in tumor tissue than did the low‐expression
   group (Figure [127]4D). These data suggested that the enhanced
   immunosuppressive phenotype of the CX3CL1 high‐expression group might
   be related to Tregs. Next, we identified the DEGs upregulated in the
   CX3CL1 high‐expression group compared to the low‐expression group and
   used the CICS as a clinical marker for weighted gene co‐expression
   network analysis (WGCNA). We identified 18 co‐expressed gene modules
   and performed a correlation analysis between these modules and the CICS
   phenotype. We found that most modules were related to Treg recruitment,
   with ten modules showing positive correlations and 5 being
   significantly positively correlated (Figure [128]4E). Furthermore, we
   conducted GO and KEGG analyses of the genes in the brown module, which
   was most closely associated with Treg infiltration. GO analysis
   revealed that these genes were enriched in processes related to
   immunosuppression, such as “negative regulation of leukocytes,”
   “negative regulation of immune effector process,” and “myeloid
   leukocyte‐mediated immunity” (Figure [129]4F). On the other hand, KEGG
   analysis revealed that these genes were enriched in classical pathways
   that activate Treg function, including the “JAK‐STAT signaling pathway”
   and “MAPK signaling pathway” (Figure [130]4G). We also found a
   significant positive correlation between the expression of CX3CL1 and
   that of the key transcription factor FOXP3, which regulates Treg
   function, as well as between the expression of the Treg effector
   molecules IL‐10 and TGFβ in human CRC tissues via GEPIA (Figure
   [131]S4D, Supporting Information). These results suggest that CX3CL1
   helps tumor cells establish an immunosuppressive microenvironment by
   promoting the infiltration and enhancing the function of Tregs.

   We then proceeded to validate the effect of CX3CL1 on Treg cells
   directly via in vitro experiments. T cells isolated from PBMCs were
   activated with anti‐CD3, anti‐CD28 antibodies, and IL‐2. Then activated
   T cells were cocultured with exogenously added CX3CL1 for 48 h. The
   results showed that, compared to those in the control group, the
   proportion of Treg cells (CD4^+CD25^+CD127^−) was significantly higher,
   and the proportion of memory T cells (CD4^+CD25^−CD127^+) was
   significantly lower in the CX3CL1‐treated group (Figure [132]4H).
   Although CX3CL1 enhances the function of Treg cells, we could not
   determine whether the increase in T‐cell exhaustion induced by CX3CL1
   is due to its effect or mediated by Treg cells. To address this issue,
   we sorted ex vivo‐activated T cells into CD4^+ T cells and CD8^+ T
   cells. Compared to adding CX3CL1 only in CD8^+ T cells, adding CX3CL1
   in mixed CD4^+ T and CD8^+ T cells significantly increased the
   expression levels of exhaustion markers (Figure [133]S4E,F, Supporting
   Information) and attenuated proliferative capacity (Figure [134]S4G,
   Supporting Information) on CD8^+ T cells.

   We further validated that CX3CL1 promotes T‐cell exhaustion by
   regulating Treg function in humanized mice subcutaneously engrafted
   with CRC cells. Specifically, one week after the injection of HCT116
   cells into the flanks of NOD/ShiLtJGpt (NCG) mice to induce
   subcutaneous tumor formation, in vitro activated CD8^+ T cells (1  ×
    10^7 per mouse) or mixed CD4^+ T cells (5 × 10^6 per mouse) and CD8^+
   T cells (5  ×  10^6 per mouse) cells from the same healthy donor were
   injected via the tail vein every week (Figure [135]4I). We also treated
   mice injected with CD8^+ T cells or mixed T cells with the
   CX3CL1‐specific inhibitor JMS‐17‐2 (10mg kg^−1 per day). We discovered
   that in mice intravenously injected with a mixture of CD4^+ T and CD8^+
   T cells via the tail vein, JMS‐17‐2 significantly inhibited tumor
   progression (Figure [136]4J). However, the antitumor effect of JMS‐17‐2
   was not pronounced in mice that were solely administered CD8^+ T cells
   (Figure [137]S4H, Supporting Information). At the endpoint of the in
   vivo experiment in which mice were injected with mixed T cells, we
   euthanized all the mice, extracted the tumors, and digested them into
   single cells for flow cytometric analysis. Firstly, we evaluated the
   infiltration of T cells in tumor tissues and found that JMS‐17‐2 could
   significantly increase the level of T cell infiltration (Figure
   [138]S4I, Supporting Information). The proportion of T cells was
   between 18%‐25%, which is essentially consistent with the abundance of
   T cell infiltration in surgically resected tissues from patients.^[
   [139]^16 ^] Next, we found that applying JMS‐17‐2 treatment could
   significantly increase the proliferative capacity of CD8^+ T cells,
   suggesting a stronger tumor reactivity (Figure [140]S4J, Supporting
   Information). Notably, the administration of JMS‐17‐2 significantly
   alleviated CD8^+ T cell exhaustion in the mice, as evidenced by the
   decreased expression of exhaustion markers such as PD‐1, TIM‐3, LAG‐3,
   and CD39 (Figure [141]4K–N). Correspondingly, we found that JMS‐17‐2
   treatment reduced Treg infiltration and a significant increase in the
   proportion of memory T cells (Figure [142]4O). We also conducted ELISA
   tests on the primary cell supernatant and found that IL‐10 and TGFβ
   levels were significantly decreased in the JMS17‐2‐treated group,
   indicating diminished Treg function (Figure [143]4P). To further
   validate the immunomodulatory effects of CX3CL1‐CX3CR1 signaling
   blockade, we administered neutralizing antibodies targeting CX3CL1
   (Quetmolimab, 200 µg  per mouse/3 days) to assess the immunosuppressive
   effects of CX3CL1‐CX3CR1 signal blocking in humanized mouse models
   injected with a mixture of CD4^+ T and CD8^+ T cells. We observed that,
   compared to the control group, Quetmolimab treatment exhibited a
   significant anti‐tumor effect (Figure [144]S5A, Supporting
   Information). The immune microenvironment analysis revealed that
   Quetmolimab treatment significantly reduced the expression of
   exhaustion markers (Figure [145]S5B,C, Supporting Information) and
   increased the tumor reactivity of CD8^+ T cells (with elevated
   expression of Ki67, TNF, and IFN) (Figure [146]S5D–F, Supporting
   Information), indicating a reduction in CD8^+ T cell exhaustion.
   Correspondingly, the blockade of CX3CL1 led to a decrease in the
   infiltration of Treg cells (Figure [147]S5G, Supporting Information).
   Taken together, these findings indicated that CX3CL1 enhanced the
   function of Treg cells to suppress anti‐tumor immunity.

2.5. CX3CL1 Enhances the Function of Treg Cells by Promoting the TNFRSF9
Phenotype

   To further investigate the role of CX3CL1 signaling in Treg cells, we
   first examined the expression of CX3CL1 and its unique receptor CX3CR1
   in different Treg populations. We examined scRNA‐seq data
   ([148]GSE108989) from tissues of 12 CRC patients and found that Treg
   cells from PBMCs and adjacent normal tissue hardly expressed CX3CR1. In
   contrast, a subset of Treg cells infiltrating the tumor tissue
   exhibited high expression of CX3CR1 (Figure [149]S6A, Supporting
   Information). Next, we integrated all pairs of Treg cell samples. The
   UMAP analysis revealed distinct origins of Treg cell clusters: Cells in
   Clusters 0 and 3 were primarily from PBMCs and adjacent normal tissues,
   making up 31.67% and 38.47% of these subgroups, respectively, with only
   10.8% from CRC tissues. In contrast, Clusters 1 and 4 were mainly tumor
   tissue‐infiltrating Treg cells, with 29.72% from CRC, while
   contributions from adjacent tissues and PBMCs were minimal at 3.48% and
   5.47%, respectively (Figures [150] 5A and [151]S6B, Supporting
   Information). The above results indicated that the phenotype of
   tumor‐infiltrating Treg cells significantly differed from that of Tregs
   derived from PBMCs or adjacent normal tissues.

Figure 5.

   Figure 5
   [152]Open in a new tab

   CX3CL1 enhances the function of Treg cells by promoting the TNFRSF9
   phenotype. A) UMAP plot based on the scRNA‐seq data ([153]GSE108989) of
   Treg cells sourced from peripheral blood, adjacent normal tissue, and
   cancer tissue. Create visual representations that reflect the origin of
   the samples. Determine and quantify the percentages of cells within
   Clusters 0, 3 (encircled by the purple dotted line) and Cluster 1, 4
   (encircled by the blue dotted line) for each sample category in
   relation to the overall number of cells. NTR, Tregs from adjacent
   normal colorectal tissues; PTR, Tregs from peripheral blood; TTR, Tregs
   from CRC. B–D) Feature plots of indicated genes across all Treg
   subclusters in (A). E) scRNA‐seq of CD45^+ immune cells extracted from
   resected tumor tissues of three primary‐care CRC patients. UMAP plot of
   Tregs (CD4^+FOXP3^+, n = 4628), identifying nine distinct subgroups.
   Each subgroup was named by marker genes calculated using the Seurat
   package. F) Feature plots of indicated genes across all Treg
   subclusters in (E). G) UMAP and extrapolated future state of cells
   (overlaid arrows) based on RNA velocity. H) Putative driver genes of
   TNFRSF9^+ Treg are identified by high likelihoods. Phase portraits
   (left) and expression dynamics (right, green represents upregulation,
   while red represents downregulation) for these driver genes
   characterize their activity. I) Heatmap shows the GEPIA analysis of the
   correlation between putative driver genes in (H) and all chemokine
   receptors. J) The expression level of putative transcription factors in
   CX3CL1 treated or control CD4^+ T cells was assessed using
   immunoblotting. K) The expression level of BTG2 and SATB1 in CX3CL1
   inhibitor JMS‐17‐2 treated or control CD4^+ T cells was assessed using
   immunoblotting. L,M) Co‐culturing CD4^+ T cells from different groups,
   as illustrated, with HCT116 cells. The ratio of TNFRSF9^+ cells in
   CD25^+ T cells was analyzed by flow cytometry. n = 6, student t‐test of
   variance, **p < 0.01, ***p < 0.001. (N) Exogenous addition of CX3CL1
   significantly increased the IL‐10 (left) and TGF‐β (right) levels of
   CD4^+ T cells co‐cultured with HCT116. n = 6, student t‐test of
   variance, ***p < 0.001.

   A previous study reported that TCF‐1‐deficient Treg cells strongly
   suppressed T‐cell proliferation and cytotoxicity.^[ [154]^11 ^]
   Correspondingly, Cluster 0 and Cluster 3 Treg cells exhibited high TCF7
   expression (encoding the TCF1), while Cluster 1 and Cluster 4 Treg
   cells barely expressed TCF7 (Figure [155]5B). Moreover, we found that
   CX3CR1 was primarily expressed in Treg cells from Cluster 4
   (Figure [156]5C). Notably, TNFRSF9^+ Tregs are a subset that has been
   demonstrated to play a crucial immunosuppressive role in various
   tumors.^[ [157]^10 ^] Our analysis showed that Treg cells in both
   Cluster 1 and Cluster 4 were in the TNFRSF9^+ Treg subset
   (Figure [158]5D). Because of the overlap between CX3CR1^+ cells and
   TNFRSF9^+ cells, these results suggested that the CX3CL1‐CX3CR1
   signaling pathway might be involved in shaping the TNFRSF9^+ Treg
   phenotype. Lipid synthesis and metabolism are closely associated with
   Treg differentiation and functional enhancement.^[ [159]^17 ^] We
   identified DEGs between Treg cells in Cluster 4 and those in Clusters 0
   and 3, and subsequently conducted GSEA. The results indicated that
   lipid biosynthetic and lipid metabolism processes are more active in
   Tregs from Cluster 4 (Figure [160]S6C, Supporting Information).
   Moreover, genes associated with the “receptor signaling pathway via
   JAK‐STAT” pathway were significantly enriched in Cluster 4 (Figure
   [161]S6D, Supporting Information). The above data suggested that
   CX3CL1‐CX3CR1 signaling might promote Treg function.

   To explore the impact of CX3CL1‐CX3CR1 signaling on the phenotype of
   Tregs, we performed scRNA‐seq on CD45^+ immune cells sourced from three
   resected CRC samples. We then integrated and clustered the CD4^+FOXP3^+
   cells (n = 4628) and identified nine subgroups after batch effect
   removal (Figure [162]5E). Consistent with the above results, we
   observed similar expression patterns of CX3CR1 and TNFRSF9, and the
   CX3CR1^+ cells did not express TCF7 (Figure [163]5F). Subsequent RNA
   velocity analysis of the scRNA‐seq data indicated that TCF7^+ Treg
   cells served as the starting point for the differentiation of various
   phenotypes of Treg cells in the tumor microenvironment. TNFRSF9^+ Treg
   cells were identified as a terminal differentiation phenotype
   (Figure [164]5G). Next, by individual gene dynamics screening of
   transcription factors, we identified five potential transcription
   factors that correlated most strongly with the differentiation of
   TCF7^+ Treg cells to TNFRSF9^+ Treg cells: NFKBIA, BTG2, JUND, SATB1,
   and FOSB (Figure [165]5H). To further clarify the transcription factors
   associated with the CX3CL1‐CX3CR1 signaling pathway, we calculated the
   correlation of these transcription factors with the expression of all
   chemokine receptors in CRC samples from the TCGA database. The results
   indicated that CX3CR1 strongly correlates with BTG2 and SATB1
   (Figure [166]5I). Next, we activated T cells isolated from PBMCs and
   cocultured them with HCT116 cells after sorting the CD4^+ T cells. We
   observed that the exogenous addition of CX3CL1 increased the expression
   of SATB1 and BTG2 in CD4^+ T cells, while its impact on the other three
   transcription factors was minimal (Figures [167]5J and [168]S6E,
   Supporting Information). Simultaneous JMS17‐2 treatment reduced the
   expression of SATB1 and BTG2 (Figures [169]5K and [170]S6F, Supporting
   Information). To further validate the promoting effect of BTG2 and
   SATB1 on the TNFRSF9^+ Treg phenotype, we constructed lentiviruses for
   overexpressing these two transcription factors. T cells activated by
   anti‐CD3, anti‐CD28, and IL‐2 were transfected with empty vector,
   BTG2‐OE, SATB1‐OE, and a combination (BTG2‐OE+SATB1‐OE) lentivirus. The
   overexpression of transcription factors was verified by western
   blotting (Figure [171]S6G, Supporting Information). After sorting
   CD4^+T cells by flow cytometry, we co‐cultured them with tumor cells
   according to the illustrated treatment. We then detected the proportion
   of TNFRSF9^+ Treg cells after 72 h. The results indicated that
   overexpressing BTG2 and SATB1 significantly increased the
   differentiation of the TNFRSF9^+ Treg phenotype, and the combined
   overexpression achieved an effect comparable to that of CX3CL1
   stimulation (Figures [172]5L,M and [173]S6H,I, Supporting Information).
   These results suggested that BTG2 and SATB1 are downstream effector
   molecules of the CX3CL1‐CX3CR1 signaling pathway. Finally, we
   co‐cultured CD4^+ T cells with tumor cells for 48 h and found that the
   exogenous addition of CX3CL1 significantly promoted the secretion of
   IL10 and TGFβ (Figures [174]5N and [175]S6J, Supporting Information).
   Therefore, these results indicated that the CX3CL1‐CX3CR1 signaling
   pathway facilitates the induction of the TNFRSF9^+ Treg phenotype by
   activating BTG2 and SATB1.

2.6. The Clinical Relevance of CX3CL1 Signaling

   To explore the potential of CX3CL1 as a prognostic indicator in
   clinical samples and validate its correlation with Treg infiltration,
   we conducted multiplex immunofluorescence (mIHC) staining of a tissue
   microarray containing 100 CRC tissue samples with antibodies against
   CX3CL1, CD4, CD25, and PANCK (a tumor cell marker). Given that CD25^+
   Tregs are an independent risk factor for a poor prognosis in patients
   with CRC, we used CD4 and CD25 as Treg markers instead of CD4 and
   FOXP3.^[ [176]^18 ^] By the 7th edition of the American Joint Committee
   on Cancer (AJCC) cancer staging system, we categorized tumor tissues
   with a grade <T3 as low‐grade tumors and those with a grade ≥T3 as
   high‐grade tumors. We observed a lower CX3CL1^+/PANCK^+ ratio in
   adjacent noncancerous tissues and low‐grade tumor tissues, while the
   CX3CL1^+ tumor cell ratio was significantly greater in high‐grade tumor
   tissues (Figure [177] 6A,B). We divided the samples into a
   high‐expression group (CX3CL1^High) and a low‐expression group
   (CX3CL1^Low). K‒M survival analysis revealed a significantly poorer
   prognosis in the high‐expression group than in the low‐expression group
   (Figure [178]6C).

Figure 6.

   Figure 6
   [179]Open in a new tab

   The clinical relevance of CX3CL1 signaling. A) Representative mIHC
   images of different‐stage tumor tissue showing the expression level of
   CX3CL1 in the PANCK^+ tumor area. Green, CX3CL1; Purple, PANCK; Blue,
   DAPI. Scale bars, 200 µm. B) Statistics analysis of the ratio of
   CX3CL1^+/PANCK^+ in the adjacent normal, low‐grade tumor, and
   high‐grade tumor tissues in mIHC‐stained tissue microarrays. One‐way
   ANOVA test of variance, ns, non‐significant, *p < 0.05, **p < 0.01. C)
   All samples were divided into two groups, CX3CL1^High and CX3CL1^Low,
   based on the fluorescence intensity of CX3CL1. Kaplan‐Meier estimates
   of survival for patients in different groups. *p < 0.05. D)
   Representative mIHC images of different stage tumor tissue showing
   co‐localization of CX3CL1 and Treg cells. Green, CX3CL1; Red, CD4;
   Orange, CD25; Purple, PANCK; Blue, DAPI. Scale bars, 200 µm. E)
   Statistics analysis of the ratio of CD4^+CD25^+/CD4^+ in the adjacent
   normal, low‐grade tumor, and high‐grade tumor tissues in mIHC‐stained
   tissue microarrays. One‐way ANOVA test of variance, ns,
   non‐significant, *p < 0.05. F) Linear matching of the correlation
   between CX3CL1^+/PANCK^+ ratio and CD4^+CD25^+ /CD4^+ ratio. G) Tumor
   tissues in (Figure [180]S7A, Supporting Information) were digested into
   single cells, and flow cytometry analysis of the ratio of CD25^+CD127^−
   (Treg) cells in CD4^+ T cells. n = 5, one‐way ANOVA test of variance,
   ns, non‐significant, *p < 0.05, ***p < 0.001. H) Tumor tissues in
   (Figure [181]S7B, Supporting Information) were digested into single
   cells, and flow cytometry analysis of the ratio of CD25^+CD127^− (Treg)
   cells in CD4^+ T cells. n = 6, one‐way ANOVA test of variance, ns,
   non‐significant, **p < 0.01, ***p < 0.001.

   Additionally, we observed significant colocalization of CX3CL1 and
   CD4^+CD25^+ Treg cells in the PANCK^− stromal region (Figure [182]6D).
   Upon statistical analysis, we found an increase in Treg infiltration in
   high‐grade tumors and a positive correlation between the fluorescence
   intensity of CX3CL1 and the extent of Treg infiltration
   (Figure [183]6E,F). Therefore, the above results demonstrated that
   CX3CL1 served as a prognostic indicator in CRC and promoted the
   infiltration of Treg cells in the tumor microenvironment.

   Next, we validated the clinical relevance of CX3CL1 as a therapeutic
   target through in vivo experiments in mice. We divided humanized CRC
   subcutaneous mice (as described in Figure [184]4I) into four groups.
   One week after tumor transplantation, we initiated drug treatments and
   measured tumor size using calipers every three days. One group served
   as the control, while the remaining three groups received different
   treatments: one group received the CX3CR1 inhibitor (JMS‐17‐2,
   10 mg kg^−1) once daily, another group received the PD1 antibody
   (250 µg per mouse) every three days, and the third group received a
   combination of both drugs. When the living tumor burden reached
   2000 mm^3 in any mouse, we euthanized all the mice and extracted the
   tumors. The experimental results showed that the CX3CR1 inhibitor
   significantly suppressed tumor growth and achieved treatment effects
   similar to the PD‐1 antibody. Furthermore, the CX3CR1 inhibitor
   combined with the PD‐1 antibody exhibited enhanced effects, resulting
   in the best antitumor efficacy (Figure [185]S7A,B, Supporting
   Information). We digested the tumor tissues into single cells for flow
   cytometry analysis. We found that only the group receiving the CX3CR1
   inhibitor exhibited a significant reduction in Treg cells. In contrast,
   the PD‐1 antibody group did not show a significant change in the
   proportion of Treg cells from that in the control group
   (Figure [186]6G,H). These data further confirmed the critical role of
   CX3CL1‐CX3CR1 signaling in promoting the infiltration and function of
   Treg cells.

3. Conclusion

   Tumor cells continuously undergo metabolic reprogramming to adapt to
   the microenvironment's extreme and unstable energy supply conditions,
   which reshapes the physical and chemical characteristics of the
   microenvironment.^[ [187]^19 ^] Previous studies have indicated that
   metabolic reprogramming byproducts, such as lactate and reactive oxygen
   species (ROS), promote tumor progression by suppressing the function of
   immune cells.^[ [188]^20 , [189]^21 ^] However, this represents a
   long‐term and macroscopic impact. From the perspective of dynamic
   regulation within micro‐communities, tumor cells in extreme conditions
   need to swiftly adapt to perturbations in metabolism and immune
   challenges. Our previous research indicates that SIRT1 helps tumor
   cells cope with glucose‐deficient environments by mediating the
   glucolipid metabolic reprogramming, however, the impact of this process
   on anti‐tumor immunity remains unclear.^[ [190]^3 ^] Thus, further
   exploration of whether SIRT1 participates in immune suppression is
   warranted. This study revealed that SIRT1 promotes the secretion of
   CX3CL1 in CRC cells by deacetylating FOXO1. CX3CL1‐CX3CR1 signaling
   plays a significant role in driving modestly functional TCF7^+ Treg
   cells to a more potent immunosuppressive TNFRSF9^+ Treg phenotype.
   Therefore, we identified SIRT1 as one of the key molecules that
   contributes to the formation of a pro‐tumorigenic CRC microenvironment
   through both metabolic and immunological ways.

   Transcription factors often drive metabolic reprogramming, and studying
   the immediate effects of these factors on the paracrine secretion of
   tumor cells and subsequent immune effects is important for restoring an
   authentic metabolic‐immune regulatory network. During certain
   “vulnerable” moments in tumor progression, such as nutrient
   deprivation, hypoxia, and metastasis, robust metabolic adaptation and a
   suitable immune environment are crucial for the survival of malignant
   cells. Emerging research suggests that these stress conditions are
   constantly changing, but our current understanding of the
   metabolic‐immune coordination process in the tumor microenvironment is
   superficial. While tumor cells employ variable mechanisms to carry out
   metabolic reprogramming, several classical transcription factors, such
   as members of the STAT family, the Sirtuin family, and the HIF family,
   have been demonstrated to play crucial roles in metabolic conversion.^[
   [191]^22 , [192]^23 , [193]^24 ^] We believe these transcription
   factors' effects are not limited to metabolism alone. Since releasing
   cytokines or chemokines is one of the most direct and efficient ways
   for tumor cells to influence the immune system, we explored the
   specific mechanisms by which metabolic reprogramming of CRC cells
   affects the immune environment through a transcription factor‐chemokine
   model.

   With the increased usage of high‐throughput sequencing technologies,
   researchers have begun to re‐examine immune cells traditionally
   considered to exhibit definite phenotypes. A recent study indicated
   that exhausted CD8^+ T cells exhibit intrinsic heterogeneity. Through
   transcription factor analysis via a single‐cell CRISPR screening
   system, the RBPJ‐IRF1 axis was shown to drive terminal differentiation
   toward exhausted cell phenotypes.^[ [194]^25 ^] However, major
   knowledge gaps remain regarding the gene regulatory networks and
   corresponding differentiation directions that govern other
   immunosuppressive cells, such as Tregs, tumor‐associated macrophages,
   and MDSCs. Pancancer analysis revealed that TNFRSF9^+ Tregs exert
   potent immunosuppressive effects in the tumor microenvironment across
   multiple cancer types, but there are still questions regarding their
   origin and regulatory factors. Chemokines are considered to play a
   crucial role in differentiating CD4^+ T cells into Tregs and the
   functional differentiation of Tregs.^[ [195]^26 ^] Through the analysis
   of single‐cell RNA sequencing data for T cells from CRC patients in
   public databases, we determined that the CX3CL1‐CX3CR1 signaling
   pathway may be involved in inducing the TNFRSF9^+ phenotype in Treg
   cells. These findings were further validated via analysis of our
   sequencing data, as well as through in vivo and in vitro experiments.
   Moreover, we found that CX3CL1‐CX3CR1 exert its functional effects by
   activating the transcription of BTG2 and SATB1. In a healthy state,
   BTG2 is associated with maintaining a quiescent state of T cells,
   aligning with the functional characteristics of Tregs.^[ [196]^27 ^]
   Moreover, SATB1 has been demonstrated to be closely associated with the
   development of Tregs.^[ [197]^28 ^] In conclusion, our study reveals a
   potential mechanism wherein metabolic reprogramming of tumor cells
   drives the functional differentiation of Tregs.

4. Experimental Section

Cell Culture

   HCT116 and SW480 cell lines were purchased from ATCC, and the MC38 cell
   line was obtained from NCI/NIH. All cell lines were propagated and
   passaged as adherent cell cultures. MC38 cells were maintained in a
   complete DMEM (Gibco) medium containing 10% fetal bovine serum (FBS)
   (Gibco) and 100 U ml^−1 penicillin/streptomycin. HCT116 cells were
   maintained in complete McCOY's 5A (Gibco) medium containing 10% fetal
   bovine serum (FBS) and 100 U ml^−1 penicillin/streptomycin (Gibco).
   SW480 cells were maintained in a complete L15 (Gibco) medium containing
   10% fetal bovine serum (FBS) and 100 U ml^−1 penicillin/streptomycin.
   MC38 and HCT116 cells were maintained in adherent conditions at 37 °C
   in a humidified atmosphere containing 5% CO[2]. SW480 cells were
   maintained in adherent conditions in airtight culture bottles at 37 °C.
   The medium was replaced three times weekly, and the cells were passaged
   using 0.05% trypsin/EDTA (Gibco) and preserved at early passages.

Transcriptomic Analysis with the Cancer Genome Atlas (TCGA) Datasets—Data
Preparing

   The transcriptomic and clinical data for colorectal cancer (COAD) were
   retrieved using the GDC download Tool from The Cancer Genome Atlas
   (TCGA) database ([198]https://portal.gdc.cancer.gov). All data were
   normalized by log2 transformation.

Transcriptomic Analysis with the Cancer Genome Atlas (TCGA)
Datasets—Glucolipid Metabolism Consensus Clustering Analysis

   The 126 glucolipid metabolism genes significantly associated with the
   survival prognosis of the CRC sample in the TCGA database were obtained
   by univariate Cox regression analysis. The mRNA expression matrix was
   used as the input of consensus clustering analysis to identify the
   glucolipid metabolic phenotype. Consensus clustering was performed
   using the ConsensusClusterPlus R package (v.1.58.0). Consequently, the
   consensus matrix at k = 4 demonstrated distinct delineation between
   clusters. The empirical cumulative distribution function (CDF) plot
   initially revealed pronounced optimal separation. Selecting k = 4
   resulted in a minimal proportion of ambiguous clustering (PAC) and
   notably correlated with patient survival outcomes.

Transcriptomic Analysis with The Cancer Genome Atlas (TCGA)
Datasets—Evaluation of the Correlation between SIRT1 and Immune Cell Subtype
Infiltration

   The CIBERSORT algorithm
   ([199]https://github.com/jason‐weirather/CIBERSORT) assessed immune
   cell infiltration in CRC samples from the TCGA database. LM22 was
   employed as a reference expression profile. The LM22 signature matrix
   defined 22 infiltrating immune cell components. Cell types were
   selected that were more abundant in the tumor microenvironment and were
   potentially involved in immune suppression, including macrophage
   subtypes (M0 macrophages, M1 macrophages, and M2 macrophages) and T
   cells (CD8^+ T cells, naïve CD4^+ T cells, memory CD4^+ T cells, Tfh
   cells, regulatory T cells, and γδ T cells). P‐values and root mean
   square errors were determined for each expression file in CIBERSORT.
   Only data with CIBERSORT p‐values < 0.05 were retained for subsequent
   analysis. The output was directly integrated to generate a
   comprehensive matrix of immune cell scores. After integrating this
   matrix with the gene expression matrix, Spearman correlation
   coefficients between SIRT1 and immune cells were calculated and
   visualized.

Transcriptomic Analysis with The Cancer Genome Atlas (TCGA) Datasets—WGCNA
Based on Cancer–Immunity Cycle Scores

   A weighted gene co‐expression network was constructed utilizing the
   “WGCNA” package in R software. Standardized transcriptome expression
   data from colorectal adenocarcinoma (COAD) was employed as the input
   matrix for the WGCNA process. Immune cycle scores for COAD specimens,
   derived from the TIP database ([200]http://biocc.hrbmu.edu.cn/TIP/),
   were integrated as clinical phenotypes for the WGCNA. The expression
   matrix was converted into an adjacency matrix and subsequently
   transformed into a topological overlap matrix. Gene clustering was
   conducted based on topological overlap using the average‐linkage
   hierarchical clustering approach. Adhering to the hybrid dynamic
   tree‐cutting criteria, a minimum threshold of 30 genes was established
   for each network module. Key genes within each module were pinpointed,
   and the modules were subjected to clustering with a height cutoff
   threshold of 0.25; the grey module was designated for genes that did
   not neatly fit into any other category. Employing the key genes from
   each module, the correlation between module membership and clinical
   phenotypes was calculated, culminating in creating a module‐phenotype
   correlation heatmap.

Multiplex Immunohistochemical (mIHC) Staining and Analysis

   The tissue microarray was stained with a PANO 7‐plex IHC kit, cat
   0004100100 (Panovue, Beijing, China), followed the standard protocol.
   The slides were deparaffinized in xylene, rehydrated, and washed in tap
   water before boiling in antigen retrieval. Use tissue markers to draw
   tissue areas and Antibody Diluent / Block for protein blocking (CST).
   All antigens were labeled separately, including primary antibody
   incubation (CX3CL1: Abcam, CD25: Abcam, CD4: ZSGB‐BIO, PNCK: MERCK),
   secondary antibody incubation, and TSA visualization, followed by
   labeling the next antibody. Primary antibodies were incubated for 1 h
   at room temperature. Next, incubation with Opal Polymer HRP was
   performed at 37 °C for 10 min. TSA visualization was performed with the
   PANO 7‐plex IHC kit. Microwave treatment was performed to remove the Ab
   TSA complex with antigen retrieval. The slide was finished with DAPI
   for 5 min and was enclosed in an antifade mounting medium. Slides were
   scanned using the Olympus VS200 MTL (Olympus Germany). Image analysis
   software Qupath was used to build the algorithm by checking, training,
   and confirming the steps. TUMOR, STROMA, and other tissue
   classifications were constructed in the algorithm construction, and
   cases were recorded. Accurately recognize and count cells using Qupath
   software and set reasonable thresholds to identify positive cells.
   Counting the number of all cells and the number of positive cells.

Single‐Cell RNA Sequencing and Analysis—RT & Amplification and Library
Construction

   Surgically resected tumor samples were dissociated into single cells,
   and CD45^+ cells were isolated using flow cytometry for single‐cell RNA
   sequencing. Single‐cell suspensions, prepared at a concentration of
   2 × 10^5 cells ml^−1 in PBS (HyClone), were loaded onto a microwell
   chip using the Singleton Matrix Single Cell Processing System.
   Barcoding Beads were then harvested from the chip, allowing for the
   reverse transcription of mRNA captured by these beads to synthesize
   cDNA, followed by PCR amplification. The resulting amplified cDNA was
   fragmented and subsequently ligated with sequencing adapters. Libraries
   for scRNA‐seq were prepared per the GEXSCOPE Single Cell RNA Library
   Kits (Singleton) protocol. Individual libraries, normalized to a 4 nM
   concentration, were pooled and subjected to sequencing on an Illumina
   Nova Seq 6000 platform, employing 150 bp paired‐end reads.

Single‐Cell RNA Sequencing and Analysis—Quality Control, Dimension Reduction,
and Clustering

   Cells were filtered by gene counts of more than 200 or less than 6000
   and mitochondrial content of no more than 20%. Post‐filtering, 33130
   cells met the criteria and were carried forward for subsequent
   analyses. Tools were employed from Seurat (v.4.2.0) to facilitate
   dimensionality reduction and clustering. The NormalizeData and
   ScaleData functions were utilized to standardize gene expression levels
   across all cells, followed by identifying the top 2000 highly variable
   genes using the FindVariableFeatures function, which was then used for
   principal component analysis (PCA). The batch effect between samples
   was removed by Harmony (v.0.1.0). Finally, the UMAP algorithm was
   applied to visualize cells in a 2D space.

Single‐Cell RNA Sequencing and Analysis—RNA velocity

   For RNA velocity, the BAM file and reference genome GRCh38 (hg38) were
   used in the analysis with velocity (v.0.6) and scVelo (v.0.3.0) in
   Python. The result was projected to the UMAP plot from the Seurat
   clustering analysis for visualization consistency.

GSEA Analysis

   Gene Set Enrichment Analysis was executed via R‐based clusterProfiler
   (v.4.2.2) on normalized expression data. GSEA ranked genes by phenotype
   correlation, calculating Enrichment Scores (ES) and Normalized
   Enrichment Scores (NES) through 1000 permutations. Significance was
   determined by a False Discovery Rate (FDR) < 0.25, with significant
   gene sets visualized using enrichplot.

T Cell Isolation and Cultivation

   The Transfusion Department at Southwest Hospital provided human
   peripheral blood mononuclear cells (PBMCs). PBMCs were diluted with the
   same volume of PBS (Solarbio), and then lymphocytes were separated by
   gradient centrifugation with Lymphocyte Separation Medium (Human)
   (Solarbio). After resuspending the lymphocytes with a complete
   RPMI‐1640 medium (Gibco), cells were placed in a culture dish and
   maintained overnight. Collecting supernatant containing T cells and
   discarding adherent cells. The sorted cells were stimulated with
   anti‐CD3 (2 µg ml^−1) (Biolegend), anti‐CD28 (2.5 µg ml^−1) (Biolegend)
   antibodies and IL‐2 (1000 U ml^−1) (Biolegend) for 48 h and cultured in
   RPMI‐1640 (10% FBS).

In Vivo Experiments—Humanized Xenograft Mice

   Six‐week‐old male NOD/ShiLtJGptPrkdc^em26Cd52Il2rg^em26Cd22/Gpt (NCG)
   mice were purchased from the GemPharmatech Co., Ltd (Jiangsu, China)
   and acclimated for one week. Each mouse was injected subcutaneously in
   the right groin with 1 × 10^6 HCT116 cells, and then the mice were
   divided into two groups randomly. After one week, 1 × 10^7/200 ul
   CD3/CD28 antibody‐activated human T cells were injected into the tail
   vein of each mouse every week. The PD‐1 treated group was receiving
   intraperitoneal injections of anti‐PD‐1 (BioXcell) 200 ug per mouse
   every three days. The JMS‐17‐2 treated group received intraperitoneal
   injections of the CX3CR1 inhibitor JMS‐17‐2 (MCE) 10 mg k^−1g daily.
   The Quetmolimab treated group received intraperitoneal injections of
   the CX3CL1 neutralizing antibody Quetmolimab (Selleck) 200 µg per mouse
   every 3 days. The control group received intraperitoneal injections of
   the same volume solvent. Tumor sizes were measured every three days in
   two dimensions with the caliper and calculated using the formula
   (L × W^2)/2, where L is the length, and W is the width. When the living
   tumor volume in any subject mouse reached 2000 mm^3, this was
   designated the experimental endpoint. Subsequently, all mice were
   humanely euthanized, and tumor specimens were harvested for subsequent
   flow cytometry analysis.

In Vivo Experiments—Allograft Mice

   Six‐week‐old male C57BL/6 mice were purchased from the Hubei Benente
   Biotechnology Co., Ltd (Hubei, China) and acclimated for one week. Each
   mouse was injected subcutaneously in the right groin with 1 × 10^6 MC38
   or 1 × 10^6 Sirt1‐KO MC38 cells. The experimental and control groups
   were injected subcutaneously in the right groin with 1 × 10^6 Sirt1‐KO
   MC38 and WT‐MC38 cells, respectively; no additional treatment was
   applied. Tumor sizes were measured every three days in two dimensions
   with the caliper and calculated using the formula (L×W^2)/2, where L is
   the length, and W is the width. When the living tumor volume in any
   subject mouse reached 2000 mm^3, this was designated the experimental
   endpoint. Subsequently, all mice were humanely euthanized, and tumor
   specimens were harvested for subsequent flow cytometry analysis.

Primary Cell Isolation and Flow Cytometry Analysis (FACS)

   After the mice were sacrificed, the separation subcutaneous transplant
   tumors were cut into ≈1 × 1 mm^3 pieces with scissors. Digesting tumors
   in a shaker for 1 h with digest diluent, RPMI‐1640 containing 1 g L^−1
   collagenase I (BioFroxx), 1 g L^−1 collagenase IV (BioFroxx),
   10 mg L^−1 DNase I (BioFroxx) at 37 °C. Collecting single‐cell
   suspension after filtering with 70 um filter. Cells were stained with
   fluorescent antibodies and analyzed by flow cytometry. Cells were first
   stained with Zombie NIR Fixable Viability Kit (BioLegend) following the
   manufacturer's instructions for live/dead criteria. Subsequent surface
   markers staining was performed in FACS buffer containing 1×PBS with 2%
   FBS. Intracellular staining for nuclear proteins was performed using
   the eBioscience FoxP3/Transcription Factor Staining Buffer Set
   (ThermoFisher Scientific), and cytoplasmic proteins, such as cytokines,
   were stained by the Fixation/Permeabilization Solution Kit (BD
   Biosciences). Intracellular cytokine staining was performed after a 4 h
   stimulation with Cell Activation Cocktail (with Brefeldin A) (Biolegnd)
   at 37 °C.

Western Blotting

   Cells were lysed with an ice‐cold protein lysis solution containing
   phenylmethanesulfonyl fluoride (PMSF) (Beyotime). Total protein
   concentration was measured by the BCA Protein Assay Kit (Beyotime)
   according to the manufacturer's instructions. Protein was loaded
   10–30 ug equally for separating by SDS‐PAGE and transferred to
   nitrocellulose membranes (0.45 µm). After blocking with TBST containing
   5% BSA, membranes were incubated with primary and secondary antibodies.
   Blots were visualized with Affinity ECL Reagent (femtogram) (Affinity).
   The grayscale values of western blotting bands were analyzed using
   ImageJ.

Co‐Immunoprecipitation

   Proteins interacting with A/G Dynabeads were pulled down for Western
   blotting analysis using Pierce Classic Magnetic IP/Co‐IP Kit (Thermo
   Fisher). Cells were lysed with an ice‐cold protein lysis solution
   containing phenylmethanesulfonyl fluoride (PMSF). The protein
   supernatant was incubated with a specific antibody overnight at 4 °C
   for target protein immunoprecipitation. Then, the supernatant was
   incubated with dynasts for 2 h at 4 °C. Sorting bead‐coupled proteins
   by magnetic frame and eluting protein for Western blotting analysis.

Confocal Microscopy

   Tumor cells were seeded on the confocal dish. Cells were fixed using 4%
   PFA (Biosharp) and permeabilized using 0.3% Triton (Biosharp). After
   permeabilization, cells were blocked with 1% BSA (BioFroxx) in PBS for
   30 min and incubated with primary antibody overnight at 4 °C. Then,
   probed with Alexa 488‐conjugated rabbit anti‐mouse (CST) or Alexa
   555‐conjugated mouse anti‐rabbit (CST) secondary. Finally, staining
   nuclear using DAPI (CST). An LSM900 fluorescence confocal microscope
   (Zeiss) under a 40x lens was used to obtain confocal fluorescence
   images.

Statistical and Quantitative Analysis of Staining and Microscopy Images

   For the statistical analysis of the confocal fluorescence images, the
   relevant information was first quantified using ImageJ software and
   then performed graphical and statistical analysis using PRISM.
   Differences were tested by using a one‐way ANOVA followed by a Tukey's
   multiple comparisons test (*p < 0.05; **p < 0.01, ***p < 0.001).
   Average values were indicated in the plots.

Other Statistical Analysis

   The data statistic was performed using PRISM. Paired or unpaired
   Student's t‐test (two tails) was used to compare experiments with only
   two groups. One‐way ANOVA with multiple comparisons was performed for
   experiments with more than two groups. Kaplan‐Meier assay and Log‐rank
   test were used for survival analysis (ns: no significance; *p < 0.05;
   **p < 0.01, ***p < 0.001).

Ethics Approval and Consent to Participate

   All the animal experiments were performed in accordance with
   institutional guidelines and received ethical approval from the Animal
   Ethics Committee of Third Military Medical University. Human tumor
   tissue samples were approved by the Institutional Review Board of
   Southwest Hospital (#KY2023115). Patients were informed of the trial
   content and given consent.

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   R.Z., X.Z., and L.L. contributed equally to this work. Z.W. and Y.D.
   worked on conceptualization. R.Z., Z.W., and Y.D. worked on data
   curation. Z.W., R.Z., X.Z., and L.L. worked on formal analysis. Z.W.,
   Y.D., J.L., and H.L. worked on funding acquisition. R.Z., X.Z, Y.W.,
   Z.B., H.J., T.L., Y.S., X.W., F.L., C.Z., F.Z., and Q.Q. worked on
   investigation. Z.W., R.Z., X.Z., Y.W., and H.J. worked on methodology.
   Z.W. and Y.D. worked on project administration. X.Z., Z.W., and Y.D.
   worked on resources. Z.W., R.Z., and H.P. worked on software. Y.D.,
   Z.W., J.L., and H.L. worked on supervision. R.Z., X.Z., Y.W., Z.W., and
   Y.D. worked on validation. Z.W., R.Z., X.Z., L.L., and H.P. worked on
   visualization. Z.W., R.Z., X.Z., L.L, H.L., J.L., and Y.D. worked on
   wrote the original draft. Z.W., R.Z., X.Z., L.L, H.L., J.L., and Y.D.
   worked on wrote, reviewed, and edited the draft.

Supporting information

   Supporting Information
   [201]ADVS-12-2404734-s001.docx^ (4.3MB, docx)

Acknowledgements