Abstract

   T cell dysfunction enables tumor immune evasion, understanding its
   mechanism is crucial for improving immunotherapy. Here we show, by
   RNA-sequencing analysis of human colon adenocarcinoma and
   triple-negative breast cancer tissues, that expression of Adipocyte
   Enhancer-Binding Protein 1 (AEBP1) positively correlates with T cell
   dysfunction and indicative of unfavorable patient outcomes. Subsequent
   single-cell RNA sequencing identifies cancer-associated fibroblasts
   (CAF) as the primary AEBP1 source. Fibroblast-specific AEBP1 deletion
   in mice enhances T cell cytotoxicity and suppresses tumor growth.
   Mechanistically, autocrine AEBP1 binds CKAP4 on CAFs, activating
   AKT/PD-L1 signaling to drive T cell dysfunction. By
   molecular-docking-based virtual screening we identify Chem-0199, a drug
   that disrupts the interaction between AEBP1 and CKAP4, thereby
   enhancing antitumor immunity. Both genetic and pharmacological AEBP1
   inhibition synergize with immune checkpoint blockade in syngeneic
   models. Our study establishes AEBP1 as a key regulator of CAF-mediated
   T cell dysfunction and a therapeutic target.

   Subject terms: Cancer microenvironment, Gene regulation in immune
   cells, Cancer, Lymphocyte activation
     __________________________________________________________________

   The tumour microenvironment often suppresses immune cell function and
   cancer-associated fibroblasts (CAF) are involved in the process. Here
   authors show in human tumours and in mouse models that AEBP1 is highly
   expressed in CAFs, and via autocrine stimulation of its receptor CKAP4,
   it induces the upregulation of PD-L1, which inhibits the anti-tumor T
   cell response.

Introduction

   T cell dysfunction in tumors is a complex process driven by various
   mechanisms operating within the tumor microenvironment (TME). One
   critical factor contributing to T cell dysfunction is the expression of
   immune checkpoint molecules on T cells, including programmed cell death
   protein 1 (PD-1) and cytotoxic T lymphocyte-associated protein 4
   (CTLA-4)^[60]1. Interaction of these molecules with their ligands on
   tumor cells or other cells in the TME results in T cell exhaustion and
   decreased antitumor activity^[61]2. Moreover, cancer-associated
   fibroblasts (CAF) and other cells in the TME secrete immunosuppressive
   factors that further inhibit T cell function and hamper their ability
   to effectively target and eliminate tumor cells^[62]3,[63]4.
   Additionally, the physical barriers created by CAFs and the
   extracellular matrix (ECM) in the TME also hinder T cell infiltration
   and migration towards the tumor site, exacerbating T cell
   dysfunction^[64]5. The interplay of these mechanisms creates a hostile
   TME, perpetuating tumor immune evasion and impeding the efficacy of
   immunotherapy interventions designed to boost T cell-mediated antitumor
   responses. Understanding and targeting these mechanisms are crucial for
   overcoming T cell dysfunction in the context of cancer.

   Adipocyte enhancer-binding protein 1 (AEBP1), also known as aortic
   carboxypeptidase-like protein, is a secretory carboxypeptidase involved
   in various cellular processes, including adipogenesis, inflammation,
   and cancer progression^[65]6,[66]7. In the context of tumors, AEBP1
   functions as an oncogene, driving tumorigenesis and
   metastasis^[67]8,[68]9. However, its impact on tumor immunity remains
   largely underexplored.

   Our study reveals that AEBP1 is positively associated with T cell
   dysfunction and predominantly expressed in CAFs. By employing
   multiomics analyses and utilizing mice with fibroblast-specific gene
   ablation, we elucidates the pivotal role of AEBP1 in facilitating
   immune evasion mediated by CAFs. Inhibiting the AEBP1 pathway emerges
   as a promising approach to enhance the efficacy of immune checkpoint
   blockade therapy (ICT).

Results

Fibroblast-derived AEBP1 closely correlates with T cell dysfunction and poor
patient survival

   To investigate the mechanisms underlying T cell dysfunction, we
   conducted RNA-seq on 16 colon adenocarcinoma (COAD) and 14
   triple-negative breast cancer (TNBC) samples (Fig. [69]1A). The Gene
   Set Variation Analysis (GSVA) scoring system was employed to assess the
   overall expression of genes related to T cell function within the
   transcriptome data. Based on the computed GSVA scores, the samples were
   stratified into two distinct categories: T cell activation and T cell
   dysfunction, reflecting the relative levels of T cell functional
   activity as inferred from the mRNA expression profiles (Supplementary
   Fig. [70]1A, B). By comparing the gene differences between the two
   categories, we identified 5 genes (AEBP1, MMP2, MMP11, ISLR and SFRP4)
   associated with T cell dysfunction. Then, we employed The Tumor Immune
   Dysfunction and Exclusion (TIDE) to further elucidate the association
   between these genes and T cell dysfunction. AEBP1 exhibited the most
   robust association with T cell dysfunction in pan-cancer compared to
   other genes (Fig. [71]1B). Subsequent single-cell RNA sequencing
   (scRNA) analysis of COAD and TNBC samples based on Tumor Immune
   Single-cell Hub 2 (TISCH2) indicated the predominant expression of
   AEBP1 in human cancer-associated fibroblasts (hCAF) within the tumor
   microenvironment (TME) (Fig. [72]1C, D). Consistently, the dominant
   expression of AEBP1 in hCAFs was validated by double IF staining of
   AEBP1 and fibroblasts marker α-SMA in the tumor samples from two
   independent cohorts of cancer patients with COAD or TNBC (Fig. [73]1E,
   F). In both in-house cohorts, we observed a negative correlation
   between AEBP1 expression in hCAFs and the expression of Granzyme B
   (GZMB) in the TME, indicating a potential association between AEBP1 and
   T cell dysfunction (Fig. [74]1G, H). Finally, we performed survival
   analysis using the Kaplan-Meier Plotter to explore the correlation
   between AEBP1 expression and the clinical outcomes of cancer patients.
   Our findings revealed that individuals with lower AEBP1 expression
   exhibited increased overall survival (OS) rates in cases of COAD or
   TNBC (Supplementary Fig. [75]1C, D). Consistent with this, our in-house
   COAD and TNBC cohort also demonstrated that patients with higher AEBP1
   expression in tumors had shorter overall survival times (Fig. [76]1I,
   J). Together, our data showed that AEBP1 in hCAFs was positively
   associated with T cell dysfunction and closely related to poor
   prognosis of cancer patients.

Fig. 1. CAF-derived AEBP1 is associated with T cell dysfunction and closely
related to poor survival.

   [77]Fig. 1
   [78]Open in a new tab

   A Illustration of the workflow, including the cohorts of tumor samples
   and an overview of analytical approaches used. Created in BioRender.
   Xiaoyu, W. (2025) [79]https://BioRender.com/5uvgprp. B T cell
   dysfunction scores of indicated genes in pan-cancer based on TIDE
   system. C Expression levels of AEBP1 in the cell clusters in TME based
   on scRNA-seq data from COAD (CRC_GSE146771_Smartseq2). D Expression
   levels of AEBP1 in the cell clusters in TME based on scRNA-seq data
   from TNBC (BRCA_GSE114727_inDrop). IF staining showed representative
   images of AEBP1 (green) and α-SMA (red) in human COAD (E) or TNBC (F)
   tissues. Scale bar, 50 μm. n = 3 biological replicates. Correlation
   between GZMB and AEBP1 expression in human COAD (G) or TNBC (H) samples
   was detected by IHC (two-sided Spearman correlation analysis). I, J
   Overall survival of COAD or TNBC patients with AEBP1^high or AEBP1^low
   expression levels in an in-house cohort. Source data are provided as a
   Source Data file.

Ablation of Aebp1 in CAFs enhances T cell antitumor function

   The clinical findings above suggested that AEBP1 in hCAFs may
   facilitate tumor progression by modulating T cell dysfunction. To
   elucidate this, we first investigated the direct impact of AEBP1 in
   CAFs on the proliferation and migration capacities of cancer cells.
   Co-culturing AEBP1-knockdown hCAFs, Aebp1-knockdown mouse CAFs (mCAF)
   or Aebp1^-/- mCAFs with cancer cells, or treating cancer cells with
   recombinant mouse AEBP1 protein (rmAEBP1), did not exhibit more
   influence on growth and invasive potential of cancer cells in vitro,
   compared with matched controls (Supplementary Fig. [80]2A–F).
   Similarly, co-implantation of Aebp1-knockdown mCAFs with 4T1 mammary
   tumor cells in immunodeficient nude mice did not result in substantial
   changes in tumor growth and metastasis, compared to the counterpart
   group receiving control mCAFs and tumor cells (Supplementary
   Fig. [81]2G, H). In contrast, in immunocompetent mice, co-implantation
   of Aebp1-knockdown mCAFs with 4T1, CT26 colon tumor cells, or EMT6
   mammary tumor cells led to slower tumor growth compared to control
   groups, suggesting Aebp1 in mCAFs exerts pro-tumor activities by
   regulating T cell-based tumor immunity (Supplementary Fig. [82]2I–K).

   To better clarify the relationship between CAF-derived AEBP1 and T cell
   function, we next crossed Aebp1^flox/flox mice with the
   S100a4/fibroblast-specific protein 1 (Fsp1)^Cre strain to specifically
   deplete Aebp1 on fibroblasts (hereafter Aebp1^flox/flox denoted as wild
   type [WT] and Aebp1^flox/flox S100a4^Cre as Aebp1 cKO). The specific
   knockout of Aebp1 in mCAFs was verified through genetic identification
   and western blot analysis (Supplementary Fig. [83]2L, M). The MC38 COAD
   cells, EO771 mammary tumor cells or B16-F10 melanoma cells were
   inoculated orthotopically into Aebp1 cKO mice and WT mice. It was
   observed that all three types of tumors grew significantly slower in
   Aebp1 cKO mice than in WT mice (Fig. [84]2A, B and Supplementary
   Fig. [85]2N). Subsequent flow cytometry analysis revealed an elevation
   in the levels of IFN-γ^+ and TNF-α^+ CD8^+ T cells in the tumors of
   Aebp1 cKO mice as compared to WT mice (Fig. [86]2C, D and Supplementary
   Figs. [87]2O, [88]3A). Immunohistochemical (IHC) analysis of MC38
   tumor-bearing mice demonstrated marked intensification of CD8a^+ and
   GZMB^+ signals in Aebp1 cKO specimens relative to WT controls
   (Supplementary Fig. [89]3B), evidencing amplified cytotoxic lymphocyte
   (CTL) infiltration within TME. Furthermore, the TIDE database analysis
   revealed a critical dichotomy: while elevated CTL infiltration strongly
   correlated with improved OS in AEBP1-low COAD and TNBC, this survival
   advantage was completely abrogated in AEBP1-high cases (Supplementary
   Fig. [90]3C, D). This suggested that AEBP1 functionally compromise the
   prognostic benefit of tumor-infiltrating lymphocytes, consistent with
   our proposed mechanism of AEBP1-driven T cell dysfunction.

Fig. 2. Blocking Aebp1 in fibroblasts enhances T cell antitumor immunity.

   [91]Fig. 2
   [92]Open in a new tab

   MC38 (A) and EO771 (B) tumor growth from WT versus Aebp1 cKO mice
   (p < 0.0001). Percentage of IFN-γ^+ and TNF-α^+ CD8^+ T cells in MC38
   (C) and EO771 (D) tumors from WT versus Aebp1 cKO mice. Percentage of
   CD8^+ T cells in MC38 (E) and EO771 (F) tumors from WT versus Aebp1 cKO
   mice, p < 0.0001 (F). Representative images (G) and correlation between
   AEBP1^+α-SMA^+ cells and CD8^+ or GZMB^+CD8^+ T cells expression in
   human TNBC tissues (two-sided Spearman correlation analysis) (H). I
   MC38 tumor growth with CD8^+ T cells depleted by anti-CD8 antibodies. J
   tSNE plot of tumor infiltrating lymphocytes (TILs) overlaid with the
   expression of indicated markers from WT or Aebp1 cKO group. K Frequency
   of clusters of indicated immune cell subsets in MC38 tumors from WT and
   Aebp1 cKO group, p < 0.0001 for WT vs. cKO in cluster 1. n = 5
   mice/group (A–F, I, K). Data are presented as the mean ± SEM (A–F, I,
   K). Data were analyzed by two-sided unpaired Student’s t-test (C–F, K),
   and two-way ANOVA with Sidak’s (A, B) or Tukey’s (I) multiple
   comparisons test. Source data are provided as a Source Data file.

   Moreover, enhanced infiltration of CD8^+ T cells was showed in tumors
   from Aebp1 cKO mice compared to those from WT mice (Fig. [93]2E, F and
   Supplementary Fig. [94]3A). Similar results were also observed in 4T1,
   EMT6 or CT26 tumors co-implanted with Aebp1-knockdown mCAFs compared to
   those co-implanted with control mCAFs (Supplementary Fig. [95]3E, F).
   Furthermore, in our in-house TNBC cohort, an inverse correlation was
   identified between AEBP1 expression levels in hCAFs and the expression
   of CD8a in the TME (Supplementary Fig. [96]3G). Using a TNBC tissue
   microarray (n = 70), our multiplex immunofluorescence analysis revealed
   close spatial proximity between CD8^+ T cells and AEBP1-expressing CAFs
   within the TME (Fig. [97]2G). This spatial interaction was further
   contextualized by our quantitative findings showing a moderate inverse
   correlation between high AEBP1 expression and total CD8^+ T cell
   infiltration levels and cytotoxic GZMB^+ CD8^+ T cell subpopulations
   compared to low AEBP1-expressing samples (Fig. [98]2H). These
   observations supported the hypothesis that CAF-derived AEBP1 fosters T
   cell dysfunction through localized immunosuppression. Notably, the
   inhibition of tumor growth in Aebp1 cKO mice was reversed by depleting
   CD8^+ T cells with anti-CD8 antibodies (Fig. [99]2I), indicating that
   the enhanced antitumor activity by Aebp1 blockade is dependent on CD8^+
   T cells.

   To evaluate the overall influence of Aebp1 blockade on the tumor immune
   landscape, we characterized CD45^+ immune cells isolated from MC38
   tumor tissues in both WT and Aebp1 cKO mice through mass cytometry
   (CyTOF). We identified a total of 10 distinct cell clusters
   (Supplementary Fig. [100]3H, I). MC38 tumors from Aebp1 cKO mice had
   significantly increased CD8^+ T cells (cluster 1), whereas
   immunosuppressive immune cells like M2 macrophages (cluster 6) was
   decreased in tumors from Aebp1 cKO mice (Fig. [101]2J, K), suggesting
   enhanced antitumor immunity.

Single-cell RNA sequencing revealed that AEBP1-educated CAFs inhibited T cell
function

   To further confirm the impact of AEBP1 in the TME, we profiled
   scRNA-seq analysis on EO771 tumors from WT and Aebp1 cKO mice. In
   total, we identified 10 cell subsets based on classic cluster markers,
   including malignant cells, epithelial, monocytes, macrophages, T cells,
   natural killer (NK) cells, fibroblasts, neutrophil, endothelial and B
   cells (Supplementary Fig. [102]4A–C). Consistent with our above
   findings (Fig. [103]1C, D), Aebp1 expression was notably enriched in
   the fibroblast population (Fig. [104]3A). Remarkably, lower numbers of
   pro-tumor cells, such as fibroblasts, macrophages, and neutrophils,
   were observed in the tumors of Aebp1 cKO mice in comparison to those in
   WT mice (Fig. [105]3B). Conversely, increased levels of anti-tumor
   cells, specifically T cells and NK cells, were detected in the tumors
   of Aebp1 cKO mice (Fig. [106]3B). To further explore the effect of
   Aebp1 on mCAFs, we analyzed the fibroblast subtypes and identified four
   clusters: inflammatory CAFs (iCAF), extracellular CAFs (eCAF),
   antigen-presenting CAFs (apCAF), and myofibroblast CAFs (myCAF)
   (Supplementary Fig. [107]4D, E). The expression levels of Aebp1 are
   elevated in iCAFs and apCAFs, while showing relatively lower expression
   in myCAFs (Supplementary Fig. [108]4F). Furthermore, our scRNA-seq
   revealed reduced fibroblast population in EO771 tumors of Aebp1 cKO
   mice compared to WT control (Fig. [109]3B). This observation was
   corroborated by functional assays demonstrating significantly
   attenuated proliferation in Aebp1-deficient mCAFs relative to WT cells,
   coupled with diminished α-SMA immunoreactivity in Aebp1 cKO tumors
   (Supplementary Fig. [110]4G, H). Notably, scRNA-seq analysis further
   showed a marked decrease in myCAF proportion and concomitant
   accumulation of iCAFs in Aebp1-deficient tumors (Fig. [111]3C).
   Intriguingly, Aebp1 cKO-derived iCAFs exhibited enhanced secretion of
   chemokines Cxcl9 and Cxcl10 (Supplementary Fig. [112]4I). Our findings
   suggest that Aebp1 might support myCAF-mediated stromal barrier
   formation to restrict CD8^+ T cell infiltration, while its ablation
   promotes chemokine-dependent T cell recruitment through iCAF
   activation.

Fig. 3. scRNA-seq uncovers that CAF-derived Aebp1 reshapes the tumor immune
microenvironment.

   [113]Fig. 3
   [114]Open in a new tab

   A Expression level of Aebp1 in the indicated cell clusters of EO771
   tumors based on scRNA-seq. B Percentage of immune cell clusters in the
   EO771 tumors from WT or Aebp1 cKO group. C Percentage of fibroblast
   subtypes in the EO771 tumors from WT or Aebp1 cKO group. D Trajectory
   analysis of fibroblast subtypes. Cell types were color-coded and
   arranged by pseudo-time (left). Blue colors were based on pseudo-time
   (middle). Change of Aebp1 expression in the cell types based on
   pseudotime (right). E The expression of Aebp1 in indicated fibroblast
   subtypes based on pseudotime. F Percentage of indicated T cell clusters
   in the EO771 tumors from WT or Aebp1 cKO group. G Expression levels of
   Ifng and Gzmb in the Teff cell clusters of EO771 tumors based on
   scRNA-seq analysis (n = 79 for WT, n = 191 for cKO, two-sided unpaired
   Student’s t-test).

   To explore the potential impact of Aebp1 on the CAFs origin in the
   tumorigenesis, we performed scRNA-seq trajectory analysis on the
   fibroblast subsets. The trajectory analysis showed the eCAFs cluster at
   the trajectory’s start, myCAFs at the terminus of trajectory branch 1,
   iCAFs at the endpoint of trajectory branch 2, and apCAFs at the
   endpoint of trajectory branch 3 (Fig. [115]3D). Notably, eCAFs
   exhibited low Aebp1 expression at the early trajectory stages, whereas
   elevated Aebp1 expression was evident in apCAFs and iCAFs at later
   stages (Fig. [116]3D, E and Supplementary Fig. [117]4J). These findings
   support the notion that Aebp1 derived from iCAFs or apCAFs could
   significantly contribute to tumor progression.

   To further confirm the interaction between AEBP1-educated mCAFs and
   other cell clusters, we performed cell-cell communication analysis
   using scRNA-seq data. The interactions between CAFs and T cells were
   diminished in TME of EO771 tumors from Aebp1 cKO mice compared to WT
   mice (Supplementary Fig. [118]4K). To investigate how AEBP1-educated
   mCAFs affect T cells, we analyzed T cell subsets, identifying five main
   subsets: including tissue resident regulatory T cells (tr-Treg),
   proliferation T cells (Tproli), Th17, effector T cells (Teff) and NKT
   cells (Supplementary Fig. [119]4L, M). A lower number of
   tumor-infiltrating tr-Tregs and a higher number of Teff cells were
   observed in EO771 tumors derived from Aebp1 cKO mice compared to those
   from WT mice (Fig. [120]3F). As expected, Aebp1 loss significantly
   increased the production of cytokines, including Ifng and Gzmb
   (Fig. [121]3G). These data suggest that Aebp1 deficiency significantly
   enhances the cytotoxic effect of T cells by inhibiting the interaction
   between CAFs and T cells.

Blocking AEBP1 down-regulates PD-L1 expression on CAFs

   In order to further elucidate how AEBP1 in CAFs impairs antitumor
   immunity, we conducted a comparative analysis of global transcriptomic
   profiles between the control mCAFs and Aebp1 knockdown mCAFs using RNA
   sequencing. Compared with control group, Aebp1 knockdown mCAFs showed
   higher expression of genes associated with immune activation and T cell
   activation but lower expression of many genes related to immune
   suppression, indicating a regulatory role of Aebp1 in mCAFs’ immune
   function (Fig. [122]4A). Among the affected genes, PD-L1 (CD274), a
   crucial immune checkpoint protein implicated in CAF-mediated immune
   suppression^[123]5, was notably downregulated in Aebp1 knockdown mCAFs
   compared to controls (Fig. [124]4A). Consistent results were observed
   in the scRNA analysis of WT and Aebp1^-/- mCAFs, highlighting a
   significant decrease in PD-L1 expression in Aebp1-deficient mCAFs
   (Fig. [125]4B). Additionally, spatial transcriptomic sequencing
   (stRNA-seq) of colon cancer samples demonstrated reduced PD-L1/PD-1
   signaling in the AEBP1 low-expression group (Supplementary
   Fig. [126]5A). Moreover, the expression levels of AEBP1 and PD-L1
   showed a positive correlation trend within the tumor stroma of human
   COAD and TNBC samples, as well as in mouse MC38 tumors (Fig. [127]4C
   and Supplementary Fig. [128]5B, C). This correlation was further
   supported by reduced PD-L1 expression in CAFs from MC38, EO771, and
   B16-F10 tumors in Aebp1 cKO mice compared to WT mice (Fig. [129]4D, E
   and Supplementary Fig. [130]5D). Subsequent confirmation of
   downregulated PD-L1 expression in AEBP1-knockdown hCAFs, Aebp1^-/- and
   Aebp1-knockdown mCAFs was conducted via real-time PCR and flow
   cytometry in vitro (Fig. [131]4F, G and Supplementary Fig. [132]5E, F).
   Additionally, the recombinant human or mouse AEBP1 (rhAEBP1/rmAEBP1)
   protein stimulated PD-L1 expression in both hCAFs and mCAFs
   (Fig. [133]4H, I and Supplementary Fig. [134]5G). By pre-treating WT
   hCAFs and mCAFs with a PD-L1-blocking antibody and then co-culturing
   them with T cells, levels of IFN-γ^+ and TNF-α^+ CD8^+ T cells were
   elevated to levels comparable to those co-cultured with AEBP1-knockdown
   hCAFs and Aebp1-deficient mCAFs, underscoring the role of PD-L1 as a
   key downstream mediator of pro-tumor immunity induced by AEBP1
   (Fig. [135]4J and Supplementary Fig. [136]5H).

Fig. 4. Blocking AEBP1 down-regulates PD-L1 expression on CAFs.

   [137]Fig. 4
   [138]Open in a new tab

   A Gene ontology analysis by RNA sequencing of shNC or shAebp1 mouse
   CAFs (mCAFs) (n = 3 mice/group). Heatmap shows the differentially
   expressed genes (DEGs) and associated signatures. B scRNA-seq analysis
   showing the expression of DEGs including Cd274 in fibroblasts in EO771
   tumors from WT (n = 184) or Aebp1 cKO (n = 445) group. C The expression
   of PD-L1 in AEBP1^high or AEBP1^low human COAD samples was detected by
   IHC (two-sided Spearman correlation analysis). Flow cytometry analysis
   of PD-L1 expression on CAFs in WT or Aebp1 cKO tumors of MC38 (D) and
   EO771 (E) models (n = 5 mice/group). Flow cytometry analysis of PD-L1
   expression on shNC and shAEBP1 hCAFs (F), or WT and Aebp1^-/- mCAFs
   (G). Flow cytometry analysis of PD-L1 expression on hCAFs or mCAFs
   treated with IgG or rhAEBP1 (2 μg/ml) (H)/rmAEBP1 (1 μg/ml) (I). J
   Percentages of IFN-γ^+ or TNF-α^+ CD8^+ T cells co-cultured with shNC
   or shAEBP1 hCAFs pre-treated with IgG, or anti-PD-L1 antibodies
   (10 μg/mL). n = 3 biological replicates (F–J). Data are presented as
   the mean ± SEM (D–J). Data were analyzed by two-sided unpaired
   Student’s t-test (B, D, E, G–I), and one-way ANOVA with Dunnett’s (F)
   or Tukey’s (J) multiple comparisons test. p < 0.0001 (E, F). Source
   data are provided as a Source Data file.

   Given the secretory property of AEBP1 and the established contribution
   of tumor cell-intrinsic PD-L1 to immune evasion, we investigated
   AEBP1-mediated crosstalk between CAFs and malignant cells. Spatial
   transcriptomics profiling revealed lower PD-L1 in tumor cells from
   AEBP1^low versus AEBP1^high niches (Supplementary Fig. [139]5I).
   Consistently, tumor cells co-cultured with Aebp1^-/- mCAFs showed
   reduced PD-L1 levels compared to WT controls, while supplementation
   with rmAEBP1 effectively induced PD-L1 expression in tumor cells
   (Supplementary Fig. [140]5J, K). Crucially, the regulatory effect of
   Aebp1 on tumor PD-L1 (approximately 20% reduction upon knockout) proved
   substantially weaker than its autocrine control over CAF-intrinsic
   PD-L1 (nearly 50% reduction in Aebp1^-/- mCAFs) (Fig. [141]4G),
   directing our focus to CAF-specific PD-L1 modulation.

CKAP4 is a receptor for AEBP1 activities in CAFs

   To understand how AEBP1 regulates PD-L1 expression in CAFs, we
   conducted Kyoto Encyclopedia of Genes and Genomes (KEGG) signaling
   pathway enrichment analysis based on scRNA-seq data from WT and Aebp1
   cKO tumors. We found that the activity of the PI3K/Akt signaling
   pathway was most pronounced in fibroblasts (Fig. [142]5A and
   Supplementary Fig. [143]6A), which was also corroborated in KEGG
   analysis of mRNA-seq data from control mCAFs and Aebp1-knockdown mCAFs
   (Fig. [144]5B). We then confirmed that the activation of the PI3K/Akt
   signaling pathway was significantly suppressed in both AEBP1-knockdown
   hCAFs and Aebp1^-/- and Aebp1-knockdown mCAFs, while recombinant AEBP1
   notably activated this pathway in both hCAFs and mCAFs (Fig. [145]5C, D
   and Supplementary Fig. [146]6B). Consistent with previous findings
   (Supplementary Fig. [147]5J, K), AEBP1 also regulated PI3K/Akt
   signaling in tumor cells, though less prominently than in CAFs
   (Supplementary Fig. [148]6C). Since PI3K/Akt signaling axis is known to
   positively mediate PD-L1 expression^[149]10, we interrogated the role
   of PI3K/Akt signaling pathway in AEBP1-regulated PD-L1 expression. We
   observed that PD-L1 in mCAFs was significantly down-regulated by a PI3K
   inhibitor LY294002 (Supplementary Fig. [150]6D). Furthermore, the Akt
   activator SC79 efficiently reversed the suppression of PD-L1 expression
   caused by Aebp1 blockade (Fig. [151]5E). These findings suggest that
   PI3K/Akt signaling pathway is responsible for the function of AEBP1 in
   both hCAFs and mCAFs.

Fig. 5. Ckap4 is a receptor for AEBP1 activities in CAFs.

   [152]Fig. 5
   [153]Open in a new tab

   A KEGG analysis of scRNA-seq data showing the enriched signaling
   pathways in CAFs from WT tumors, in comparison to those from Aebp1 cKO
   tumors. B KEGG analysis of RNA-seq data showing the enriched signaling
   pathways in shNC CAFs compared with shAebp1 CAFs. Hypergeometric test
   (one-sided) with BH-FDR (A, B). C Western blot analysis of total AKT
   and p-AKT expression in shNC and shAEBP1 hCAFs, or WT and Aebp1^-/-
   mCAFs. GAPDH was used as a loading control. D Western blot analysis of
   total AKT and p-AKT expression in hCAFs treated with IgG or rhAEBP1
   (2 μg/ml). GAPDH was used as a loading control. E Flow cytometry
   analysis of PD-L1 expression on WT, Aebp1^-/-, or SC79 (20 μM) treated
   Aebp1^-/- CAFs, p < 0.0001. F The binding status of AEBP1 and CKAP4 in
   hCAFs or mCAFs based on IP. G IF showed representative images of AEBP1
   (red) and CKAP4 (green) in shNC and shCkap4 mCAFs, or CAFs treated with
   IgG or rmAEBP1 (1 μg/ml). Scale bar, 25 μm. H Construction of AEBP1
   serial deletion mutants. I IP and western blot were performed to
   analyze the co-expression of CKAP4 with either AEBP1 or its deletion
   mutants in 293 T cells. J Western blot analysis of total AKT and p-AKT
   expression in shNC or shCKAP4 hCAFs treated with vehicle or rhAEBP1
   (2 μg/ml). GAPDH was used as a loading control. K Flow cytometry
   analysis of PD-L1 expression in shNC or shCKAP4 hCAFs treated with
   vehicle or rhAEBP1 (2 μg/ml), p < 0.0001. Percentage of IFN-γ^+ (L) or
   TNF-α^+ (M) CD8^+ T cells cocultured with shNC or shCkap4 mCAFs treated
   with vehicle or rmAEBP1 (1 μg/ml), p < 0.0001. n = 3 biological
   replicates (E, G, K–M). Blot is representative of n = 2 biological
   replicates (C, D, F, I, J). Data are presented as the mean ± SEM, and
   were analyzed by one-way ANOVA with Tukey’s multiple comparisons test
   (E, K–M). Source data are provided as a Source Data file.

   As AEBP1 can induce CAFs to generate immunosuppressive effects, we next
   evaluated whether a potential receptor for AEBP1 exists on the CAFs
   membrane. The proteins of mCAFs were extracted, immunoprecipitated with
   anti-AEBP1 antibody, and further analyzed by mass spectrometry (MS)
   (Supplementary Fig. [154]6E). Cytoskeleton-Associated Protein 4
   (CKAP4), known to activate Akt signaling^[155]11, emerged as one of the
   top proteins binding to AEBP1 (Supplementary Fig. [156]6F). This
   binding was validated through immunoblotting of the
   co-immunoprecipitate (co-IP) with an anti-AEBP1 antibody
   (Fig. [157]5F).

   To further confirm whether AEBP1 colocalizes with CKAP4 on CAFs
   membrane, we incubated mCAFs with rmAEBP1 at 4°C for 3 hours. Confocal
   microscopy using fluorescence-labeled antibodies for AEBP1 and CKAP4
   demonstrated that AEBP1 colocalizes with CKAP4 on the plasma membrane
   of mCAFs (Fig. [158]5G). shRNA-mediated knockdown of Ckap4 expression
   in mCAFs efficiently abrogated AEBP1 localization on the cell surface,
   suggesting that AEBP1 interacts with CKAP4 at the plasma membrane of
   mCAFs (Fig. [159]5G). To map the CKAP4-interacting region, we generated
   four AEBP1 mutants based on structural domains annotated in UNIPROT
   (Fig. [160]5H). Co-IP assays demonstrated complete loss of CKAP4
   binding in Mutant 1 compared to overexpressed WT AEBP1 (Fig. [161]5I).
   These results identified the Mutant 1 region as the primary
   CKAP4-binding domain essential for AEBP1. Moreover, the stRNA-seq
   analysis unveiled the concurrent expression of AEBP1 and CKAP4 in the
   TME (Supplementary Fig. [162]6G). ScRNA-seq analysis demonstrated that
   the expression of Ckap4 was notably enriched in the fibroblast
   population, consistent with Aebp1 (Supplementary Fig. [163]6H).
   Trajectory analysis of fibroblast subtypes revealed a parallel
   expression pattern between Ckap4 and Aebp1. Specifically, low Ckap4
   expression was detected in the eCAFs at the initial stage, while high
   Ckap4 expression was observed in the apCAFs and iCAFs at the later
   stages Supplementary Fig. [164]6I). To verify whether CKAP4 is a
   functional receptor for AEBP1, we examined whether silencing of CKAP4
   influenced the biological effects of AEBP1 in CAFs. Indeed,
   shRNA-mediated knockdown of CKAP4 significantly suppressed Akt
   activation and PD-L1 expression induced by rhAEBP1/rmAEBP1 in both
   hCAFs and mCAFs (Fig. [165]5J, K and Supplementary Fig. [166]6J, K).
   Inhibiting Ckap4 in mCAFs restored cytotoxic activities of CD8^+ T
   cells that were blocked by rmAEBP1-treated mCAFs (Fig. [167]5L, M).
   Collectively, these data suggest that AEBP1 could modulate PD-L1
   expression by activating CKAP4/Akt pathway in CAFs.

Loss of Aebp1 in CAFs enhances therapeutic responses to ICT

   To investigate whether AEBP1 targeting could enhance immune checkpoint
   blockade (ICB) efficacy, we performed a retrospective biomarker
   analysis using the TIDE platform to evaluate correlations between AEBP1
   expression and clinical outcomes in ICB-treated patients. Analyses
   revealed that elevated AEBP1 expression significantly correlated with
   reduced survival in melanoma patients undergoing anti-CTLA-4 therapy
   (Supplementary Fig. [168]7A). Extending this observation to
   anti-PD-1-treated cohorts, we observed a similar trend toward poorer
   survival in both melanoma and glioblastoma patients with high AEBP1
   levels, though statistical significance was not achieved, potentially
   influenced by limited cohort sizes (Supplementary Fig. [169]7A). Based
   on these clinical insights, we further explored combinatorial
   strategies by integrating Aebp1 targeting with anti-CTLA-4 therapy in
   the MC38 syngeneic model. Notably, Aebp1 cKO mice receiving anti-CTLA-4
   treatment showed the most effective tumor inhibition and dramatically
   prolonged survival (Fig. [170]6A), accompanied by increased CD8^+ T
   cell infiltration (Supplementary Fig. [171]7B), and more IFN-γ^+ and
   TNF-α^+ CD8^+ T cells compared to anti-CTLA-4-treated WT mice
   (Fig. [172]6B, C). Next, we examined whether inhibiting AEBP1 could
   also enhance therapeutic efficacy of anti-PD-1 treatment. Similarly, in
   EO771 model, AEBP1 inhibition plus anti-PD-1 combinatorial treatment
   achieved the highest therapeutic response compared with AEBP1
   inhibition or ICT alone (Fig. [173]6D), along with enhanced T-cell
   function and infiltration (Fig. [174]6E, F and Supplementary
   Fig. [175]7C). Moreover, a significant decrease was observed in the
   population of TIM-3^+ PD-1^+ T cells in the combination therapy groups
   (Supplementary Fig. [176]7D–F). Together, AEBP1 blockade combined with
   ICT greatly improved the therapeutic response of multiple tumor models.

Fig. 6. Depletion of AEBP1 in CAFs improves the efficacy of ICT.

   [177]Fig. 6
   [178]Open in a new tab

   A MC38 tumor growth and survival analysis of IgG or anti-CTLA-4-treated
   WT and AEBP1 cKO mice. n = 5 mice/group for tumor volume analysis;
   n = 7 mice/group for survival analysis; log-rank test for survival
   comparison. Percentage of IFN-γ^+ (B),TNF-α^+ (C) CD8^+ T cells in MC38
   tumors from IgG, Aebp1 cKO, anti-CTLA-4, and the combination groups
   (n = 5 mice/group). D EO771 tumor growth and survival analysis of IgG
   or anti-PD-1-treated WT and Aebp1 cKO mice. n = 5 mice/group for tumor
   volume analysis; n = 7 mice/group for survival analysis; log-rank test
   for survival comparison. Percentage of IFN-γ^+ (E),TNF-α^+ (F) CD8^+ T
   cells in EO771 tumors from IgG, Aebp1 cKO, anti-PD-1, and the
   combination groups (n = 5 mice/group). Data are presented as the
   mean ± SEM (A–F), and were analyzed by two-way (A, D) or one-way ANOVA
   (B, C, E, F) with Tukey’s multiple comparisons test. Source data are
   provided as a Source Data file.

A small molecule inhibitor Chem-0199 targeting AEBP1 enhances antitumor
immunity

   To investigate AEBP1 more directly as a potential therapeutic target in
   cancer treatment, we explored the systemic inhibition of AEBP1 with a
   small molecule to potentially enhance antitumor immunity. Through a
   multi-step virtual screening process of compounds sourced from ChemDiv
   and ChemBridge, we identified series of small molecules as AEBP1
   inhibitors (Fig. [179]7A). The top 5 candidate compounds with the
   highest docking scores to AEBP1’s catalytic pocket were selected from
   each source and evaluated for their ability to inhibit AEBP1-CKAP4
   binding activity using co-IP (Fig. [180]7B and Supplementary
   Fig. [181]8A, B). Following validation, ChemBridge-5340199 (hereafter
   referred to as Chem-0199) was chosen for further investigation based on
   its significant impact on inhibiting the interaction between AEBP1 and
   CKAP4 in CAFs (Fig. [182]7C and Supplementary Fig. [183]8B, C).

Fig. 7. Chem-0199 suppresses the interaction between AEBP1 and CKAP4.

   [184]Fig. 7
   [185]Open in a new tab

   A Workflow of inhibitor screening. Created in BioRender. Xiaoyu, W.
   (2025) [186]https://BioRender.com/s97i4nk. B Overall structures of
   AEBP1 (shown in silver) and CKAP4 (shown in pink). The figure
   demonstrates the formation of a hydrogen bond (represented by yellow
   dashed lines) between the amino acid Tyr874 in the AEBP1 protein and
   the amino acid Gln527 in the CKAP4 protein. C Close-up views of
   AEBP1-Chem-0199 complexes. The figure illustrates the competitive
   binding of Chem-0199 to AEBP1, forming hydrogen bonds and π-π
   interactions specifically with the amino acid Tyr874 in the AEBP1
   protein. The green arrows indicate Chem-0199. Two perpendicular views
   are shown. D Western blot analysis of total AKT and p-AKT expression in
   vehicle or Chem-0199 treated WT and Aebp1^-/- CAFs. GAPDH was used as a
   loading control. Blot is representative of n = 2 biological replicates.
   E Flow cytometry analysis of PD-L1 expression on vehicle or Chem-0199
   treated WT and Aebp1^-/- mCAFs. F Percentage of IFN-γ^+ CD8^+ T cells
   cocultured with WT and Aebp1^-/- mCAFs treated with vehicle or
   Chem-0199. G MC38 tumor growth of vehicle or Chem-0199-treated WT and
   Aebp1 cKO mice. H Tumor growth of MC38 tumor-bearing WT mice treated
   with Chem-0199 alone, anti-CTLA-4 alone, or Chem-0199 + anti-CTLA-4.
   Percentage of IFN-γ^+ (I) or TNF-α^+ (J) CD8^+ T cells in MC38 tumors
   from mice treated with indicated regimens, p < 0.0001. n = 3 biological
   replicates (E, F), n = 5 mice/group (G–J). Data are presented as the
   mean ± SEM (E–J), and were analyzed by one-way (E, F, I, J) or two-way
   ANOVA (G, H) with Tukey’s multiple comparisons test. Source data are
   provided as a Source Data file.

   We next checked whether Chem-0199 could exert biological effect via
   AEBP1/CKAP4 pathway in CAFs. Chem-0199 dose-dependently suppressed the
   activation of Akt pathway and PD-L1 expression in CAFs (Supplementary
   Fig. [187]8D, E). Remarkably, Chem-0199 treatment of Aebp1^-/- CAFs
   didn’t further inhibit Akt activation, PD-L1 expression and
   CAF-mediated T-cell dysfunction compared to vehicle-treated Aebp1^-/-
   CAFs (Fig. [188]7D–F and Supplementary Fig. [189]8F), suggesting that
   the in vitro effects of Chem-0199 on CAFs was mediated by AEBP1. We
   then assessed the antitumor activity of Chem-0199 in vivo.
   Intraperitoneal administration of Chem-0199 dose-dependently reduced
   the size of MC38 tumors compared to the vehicle group (Supplementary
   Fig. [190]8G). To test whether the direct effect of Chem-0199 on tumor
   cells contributes to its inhibition on tumor growth in vitro, we
   treated multiple tumor cells with Chem-0199 and observed similar
   proliferation rates (Supplementary Fig. [191]9A–C), suggesting that the
   attenuation of tumor growth by Chem-0199 in mice was unlikely to result
   from direct suppression of Chem-0199 on tumor cells. To be noted, the
   inhibition of tumor growth by Chem-0199 could not be further improved
   in Aebp1 cKO mice, indicating that the suppression of tumor growth by
   Chem-0199 might be attributable to AEBP1 inhibition in CAFs
   (Fig. [192]7G). Then, we determined whether the response to Chem-0199
   might be enhanced further by ICT. Similar to experiments using genetic
   KO of Aebp1 in CAFs, Chem-0199 plus anti-CTLA-4 exhibited significantly
   superior efficacy in inhibiting MC38 tumor growth compared to Chem-0199
   or ICT treatment alone (Fig. [193]7H). Furthermore, we examined the
   effect of Chem-0199 on tumor immunity. The population of IFN-γ^+ and
   TNF-α^+ CD8^+ T cells were markedly higher in tumor sections from the
   Chem-0199 treatment group than in the control group (Fig. [194]7I, J).
   Importantly, the combination of Chem-0199 and ICT substantially boosted
   the cytotoxic activity of T cells compared to individual treatment
   (Fig. [195]7I, J).

   Considering that adverse effects may lead to intolerance in treatment
   procedures and negatively impact its clinical efficacy, we next
   evaluated the in vivo toxicity of Chem-0199 in mouse models. Complete
   blood count (CBC) results showed no marked discrepancies in white blood
   cells (WBC), red blood cells (RBC), hemoglobin (Hb), and platelets
   (PLT) between the control group and the Chem-0199 treatment group
   (Supplementary Fig. [196]9D). Hematoxylin-eosin (H&E) staining
   indicated that Chem-0199 exhibited no significant toxicity to the
   heart, liver, lung, kidney, and intestine tissues in mice
   (Supplementary Fig. [197]9E).

Discussion

   CAFs are a type of crucial stromal cells present in the TME, playing
   indispensable roles in promoting tumor growth, progression, and
   therapeutic resistance^[198]12. Increasing studies have indicated that
   CAFs can promote tumor immune escape by releasing cytokines, expressing
   immune checkpoint molecules, and remodeling the ECM, although the
   detailed molecular mechanisms require further
   clarification^[199]13–[200]15. While previous investigations have
   established AEBP1’s role in driving fibroblast activation and its
   inverse correlation with CD8^+ T cell infiltration, its precise
   immunomodulatory mechanisms within the TME remain
   undefined^[201]16,[202]17. This study elucidates AEBP1 as a central
   upstream regulator controlling CAF-mediated immunosuppression. We
   demonstrate that AEBP1 induces CD8^+ T cell dysfunction by upregulates
   PD-L1 expression, a critical immune checkpoint molecule on CAFs.
   Concurrently, we found that AEBP1 expression in CAFs restricts T cell
   infiltration into the TME. By collectively suppressing T cell function
   and abundance, AEBP1 impairs anti-tumor immunity and significantly
   contributes to tumor immune evasion.

   In the TME, CAFs demonstrate significant heterogeneity and can be
   classified into different subtypes based on their phenotypic
   characteristics and functions^[203]18. Among them, myCAFs express high
   levels of α-SMA, while eCAFs exhibit high expression of ECM proteins
   such as collagen and fibronectin, both participating in ECM remodeling
   to enhance tumor cell migration and invasion abilities^[204]19,[205]20.
   In contrast, iCAFs display an inflammatory phenotype and can influence
   tumor progression by secreting cytokines and chemokines^[206]21.
   Additionally, apCAFs are a unique subtype expressing major
   histocompatibility complex (MHC) class II molecules, capable of
   presenting antigens to T cells to regulate immune responses within the
   TME, potentially either promoting or suppressing antitumor
   immunity^[207]22. A positive correlation was observed between elevated
   levels of iCAFs and improved prognosis, while increased levels of
   myCAFs were associated with a poorer prognosis^[208]23. Our study here
   revealed that inhibiting AEBP1 significantly increased the proportion
   of iCAFs and decreased the proportion of myCAFs within CAF populations.
   Furthermore, Aebp1 exhibited higher expression levels in iCAFs and
   apCAFs, which are closely associated with tumor immunity, suggesting
   the potential important role of Aebp1 in modulating CAF immune
   functions. Critically, Aebp1 inhibition fosters an iCAF-enriched
   microenvironment that amplifies Cxcl9/10 secretion, chemokines critical
   for cytotoxic T cell recruitment and macrophage polarization
   regulation^[209]24,[210]25. Consistent with this mechanism, CyTOF
   analysis revealed reduced M2-like macrophage infiltration upon Aebp1
   suppression, a phenomenon potentially driven by elevated Cxcl9/10.
   Collectively, these findings position AEBP1 as a modulator of the tumor
   immunosuppressive microenvironment through CAF-dependent pathways,
   though mechanistic details await further elucidation.

   PD-L1 plays a crucial role in the tumor immune microenvironment. Tumor
   cells and various immune cells in the TME, including macrophages,
   dendritic cells, and B cells, inhibit antitumor immunity by expressing
   PD-L1^[211]26. Accumulating studies have shown significant expression
   of PD-L1 in CAFs, implicating their participation in tumor immune
   evasion, although the regulatory mechanisms remain
   unclear^[212]13,[213]27. Our findings unveiled AEBP1 as a vital
   upstream regulator of PD-L1 expression in CAFs. PD-L1 expression is
   regulated by signaling pathways such as JAK/STAT, NF-κB, and
   PI3K/Akt^[214]28. Our study revealed that AEBP1 upregulates PD-L1
   expression by activating the PI3K/Akt signaling pathway, highlighting
   the crucial role of PI3K/Akt signal pathway in promoting tumor
   progression mediated by CAFs. Furthermore, our study confirmed that
   CKAP4 is a key receptor for AEBP1 in CAFs, mediating the activation of
   the PI3K/Akt signaling pathway. CKAP4 was shown to interact with
   various signaling pathways, including Wnt, PI3K/Akt, and MAPK,
   influencing the biological behavior of tumor cells^[215]11,[216]29.
   Increased CKAP4 expression was associated with aberrant tumor cell
   proliferation, hastening tumor progression and
   metastasis^[217]30,[218]31. Notably, while CKAP4 demonstrates broad
   expression across multiple TME components, our analyses identify CAFs
   as showing the highest Ckap4 expression levels. Furthermore, despite
   CAF-derived AEBP1 exhibiting weak paracrine control over tumor cell
   Akt/PD-L1, its CAF-intrinsic autocrine signaling dominates functional
   output. This CAF-restricted hierarchy positions CAFs as central
   coordinators of the AEBP1/CKAP4 axis.

   Contrasting previous tumor cell-centric models of AEBP1 in
   malignancy^[219]9^,^[220]32^,^[221]33^,^[222]34, we unveil its
   CAF-selective operational mode. Systematic comparisons revealed
   predominant AEBP1/CKAP4 co-enrichment in CAFs rather than tumor cells.
   CAF-specific AEBP1 suppression failed to alter tumor cell proliferation
   or metastasis in vitro or immunocompromised hosts, yet immunocompetent
   models coupled with CD8^+ T cell depletion revealed its essential
   immunosuppressive role. These results recalibrate understanding of the
   AEBP1/CKAP4 axis, positioning CAF-specific expression as the linchpin
   of its TME-modulatory function.

   ICT has demonstrated promising potential in cancer treatment, yet its
   efficacy remains limited to a subset of patients, underscoring the need
   for enhanced strategies^[223]35. Our study demonstrates that AEBP1
   inhibition synergizes with both anti-CTLA-4 and anti-PD-1 therapies
   through complementary mechanisms, despite their distinct cellular
   targets. For the observed synergy with anti-PD-1, while PD-L1
   regulation in CAFs constitutes a primary pathway, AEBP1 also exerts
   broader immunosuppressive effects by driving CAF proliferation and
   remodeling the tumor microenvironment. These coordinated modifications
   amplify anti-PD-1 efficacy through spatial synergy: neutralizing PD-L1
   on tumor/myeloid cells with ICB while simultaneously targeting
   CAF-mediated PD-L1 regulation and stromal immunosuppression via AEBP1
   inhibition. This multi-layered intervention across distinct cellular
   compartments enhances immune activation, underscoring AEBP1’s potential
   as a combinatorial immunotherapy target. Additionally, our study
   identified Chem-0199, a small molecule inhibitor targeting the
   AEBP1/CKAP4 pathway, which effectively inhibited tumor growth and
   enhanced ICT efficacy without substantial adverse effects. Further
   investigation is warranted to determine the clinical translational
   value of this compound.

   In summary, our study revealed that CAF-derived AEBP1 binds to the
   downstream ligand CKAP4 in an autocrine manner, triggering the
   activation of the PI3K/Akt pathway and the upregulation of PD-L1 on
   CAFs, thereby resulting in T cell dysfunction. Genetic or pharmaceutic
   inhibition of AEBP1 in CAFs restrains tumor progression and enhances
   the efficacy of immunotherapy (Fig. [224]8). These findings highlight
   the crucial role of AEBP1 in CAF-mediated tumor immune evasion and
   demonstrates its potential as a diagnostic biomarker and therapeutic
   target in cancer treatment.

Fig. 8. Schematic illustration of the mechanism by which CAF-derived AEBP1
induces T cell dysfunction in tumors.

   [225]Fig. 8
   [226]Open in a new tab

   CAF-derived AEBP1 activates the CKAP4/AKT/PD-L1 pathway, then leads to
   T cell dysfunction. Chem-0199, an AEBP1 inhibitor, disrupts the
   AEBP1-CKAP4 complex, suppresses the activation of Akt pathway and PD-L1
   expression in CAFs, and finally enhances the cytotoxic activity of
   CD8^+ T cells. Furthermore, genetic or pharmaceutical inhibition of
   AEBP1 synergizes with ICT.

Methods

Cell culture

   Murine colorectal carcinoma cell lines MC38 (3101MOUTCM46) was sourced
   from National Infrastructure of Cell Line Resource (NICR, Beijing,
   China). Murine colorectal carcinoma cell lines CT26 (CRL-2638), murine
   mammary carcinoma cell lines EO771 (CRL-3461), EMT6 (CRL-2755), and 4T1
   (CRL-2539), human mammary cell line MDA-MB-231 (HTB-26), the melanoma
   cell line B16-F10 (CRL-6475), and human embryonic kidney cell line
   293 T (CRL-3216) were all sourced from the American Type Culture
   Collection (ATCC). The mCAFs were isolated from the EO771/MC38 tumors
   and hCAFs were isolated from human breast cancer tissues. The cells
   were consistently cultured at 37 °C with 5% CO[2], with the medium
   being refreshed daily. Treatments comprised recombinant mouse AEBP1
   (1 µg/mL; Abcam, ab225975), human AEBP1 (2 µg/mL; R&D, 6425-AC),
   LY294002 (T2008), and SC79 (T2274) (20 µM each; TargetMol).

RNA sequencing analysis

   COAD and TNBC tumor tissues were subjected to RNA extraction employing
   Trizol reagent (15596026, Invitrogen). The RNA samples were then
   prepared for sequencing at ANNOROAD, where they underwent library
   construction and were sequenced using the NovaSeq platform. The
   resulting FASTQ files from RNA-Seq were analyzed with htseq to
   determine gene read counts, which were then converted to TPM values.
   Subsequently, principal component analysis (PCA) was conducted using R
   software. The GSVA package was downloaded from
   [227]http://www.bioconductor.org. GSEA analysis was performed utilizing
   the Java Web Start platform. To identify differentially expressed
   genes, an analysis was conducted using the R statistical package DESeq2
   (version 1.42.0). The thresholds for significance were set to a fold
   change > 2, an adjusted p value < 0.05, and the mean log[2] expression
   level for the high expression group exceeding 0.

Identification of genes associated with T cell dysfunction

   To identify genes correlated with T cell dysfunction, we first
   established a T cell activation signature comprising key
   immunoregulatory markers (RAB27A, IL12A, XCL1, IFNG, PRF1, GZMB, CCL5,
   TBX21, CD8A, GZMA, IL2, PTPRC, ICOS, GZMM, GNLY)^[228]36,[229]37,
   followed by GSVA scoring of samples. Cohort stratification into T cell
   activation (high-score) and dysfunction (low-score) groups was
   performed using median GSVA values as the threshold.

Human samples

   Tumor tissues from patients diagnosed with COAD and TNBC were procured
   from the First Affiliated Hospital of Chongqing Medical University. The
   acquisition of pathological samples and the examination of pertinent
   patient medical records received approval from the Research Ethics
   Committee of the First Affiliated Hospital of Chongqing Medical
   University (Project Approval No. 2022-K121). All participants or their
   legal representatives provided written informed consent as necessary.

IF and IHC staining

   For IF, sections were blocked with goat serum for 20 minutes, incubated
   with primary antibodies at 4 °C overnight, followed by incubation with
   appropriate fluorescent secondary antibodies. Nuclei were
   counterstained with DAPI during mounting. Images were captured using a
   LEICA (DMI4000B) microscope. For IHC, endogenous peroxidase was blocked
   with 3% H[2]O[2], followed by blocking with goat serum for 20 minutes.
   Sections were incubated with primary antibodies at 4 °C overnight,
   followed sequentially by a biotinylated secondary antibody and
   horseradish peroxidase-conjugated streptavidin. Color development was
   achieved using 3,3’-Diaminobenzidine (DAB), followed by hematoxylin
   counterstaining, dehydration, and mounting. Slides were scanned for
   analysis.

   For IF staining, the primary antibodies employed encompassed anti-α-SMA
   (ab7817, Abcam, 1:100) and anti-AEBP1 (ab254973, Abcam, 1:100). In IHC
   staining, anti-AEBP1 (ab254973, Abcam, 1:200; sc-271374, Santa, 1:100),
   anti-CD8a (ab199016, Abcam, 1:200; 98941, CST, 1:200), anti-GZMB
   (ab4059, Abcam, 1:100; 44153, CST, 1:50), and anti-PD-L1 (66248-1-Ig,
   Proteintech, 1:5000) were applied. Secondary antibodies conjugated with
   DyLight 488 or DyLight 594, targeting mouse or rabbit IgG were sourced
   from Thermo Fisher Scientific.

   The TNBC tissue microarray (AF-BrcSur2202) was constructed and
   multiplex immunofluorescence staining was performed by Hunan Aifang
   Biotechnology Co., Ltd (China). For each sample, five randomly selected
   fields of view were used to count the number of target cells, with the
   help of HALP software (India Labs).

   In our IHC analysis, AEBP1 expression was quantified using a
   standardized scoring system evaluating both staining intensity
   (0 = negative, 1 = pale yellow, 2 = brown-yellow, 3 = dark brown) and
   percentage of positive cells (0 = 0-5%, 1 = 6-25%, 2 = 26-50%,
   3 = 51-75%, 4 = > 75%), with the final score calculated as the average
   of (intensity × percentage) across five representative high-power
   fields (400×). PD-L1 expression was assessed by averaging the
   percentage of positive cells across five fields. Similarly, GZMB^+ and
   CD8^+ cell counts were determined by averaging counts from five
   specified fields (400×/200×). All scoring was performed independently
   by the pathologist.

Gene modulation by lentiviral transduction and plasmid transfection

   Lentiviral particles for human or mouse AEBP1-targeting shRNAs, mouse
   CKAP4-targeting shRNAs were sourced from Genecopoeia. The lentiviral
   expression vector was co-transfected with lentivirus packaging vectors
   into 293 T cells using the Lipofectamine™ 2000 (11668019, Invitrogen).
   Subsequently, CAFs were stably transfected with viral particles for
   48-72 hours.

   Human AEBP1 (with 3× HA tag), four AEBP1 mutants (with 3× HA tag), and
   human CKAP4 (with 3× FLAG tag) overexpression (OE) plasmids were
   obtained from Genecopoeia. These plasmids were expressed in 293 T cells
   using Lipofectamine™3000 (ThermoFisher) for transfection. 293 T cells
   were sequentially transfected with HA-tagged AEBP1 (OE/mutants) and
   FLAG-tagged CKAP4 (24 h interval; total 60-72 h). Mutants were designed
   using Uniprot domain mapping.

Animals

   We procured 6 to 8-week-old C57BL/6-background Aebp1^flox/flox mice
   (S-CKO-17395) from The Cyagen Biosciences, and S100A4^CreERT mice
   (NM-KI-200024) from Shanghai Model Organisms Center, Inc. Wildtype (WT)
   C57BL/6 (stock number: N-0005), WT BALB/c mice (stock number: N-0004)
   and BALB/c nude mice (stock number: N-0007) were sourced from Ensiweier
   (Chongqing, China). The Aebp1^flox/flox mice were mated with
   S100a4^CreERT mice to create Aebp1^flox/floxS100a4^CreERT mice. Prior
   to tumor cell inoculation, these mice received intragastric
   administration of tamoxifen (T5648, 60 mg/kg, Sigma-Aldrich) over five
   consecutive days. Compliance with the laboratory animal care and use
   guidelines was strictly maintained throughout all experimental
   processes, and the protocols were vetted and endorsed by the
   Institutional Animal Care and Use Committee of Chongqing Medical
   University (IACUC-CQMU-20210222). The mice were housed in a specific
   pathogen-free (SPF) animal facility, with experimental and control
   animals housed separately. The housing conditions were maintained at a
   stable temperature of 21-23 °C, humidity levels within the range of
   40-60%, and a 12-hour light/dark cycle. Carbon dioxide euthanasia was
   performed immediately when the longest diameter of the tumor in their
   bodies reaches 20 mm or the tumor volume reaches 2000 mm^3. The maximum
   tumor burden permitted by our institutional animal care protocol was
   not exceeded.

Tumor models

   MC38 cells and CT26 cells (1.0×10^5 cells), along with B16-F10 cells
   (1.5×10^5 cells), were implanted subcutaneously into the proximal
   region of the mice’s right hind legs. In the mammary tumor model, 4T1
   cells, EMT6, and EO771 (1.0×10^5 cells) cells were implanted into the
   mammary fat pad of WT BALB/c, Aebp1^flox/flox, or
   Aebp1^flox/floxS100a4^CreERT C57BL/6 mice. Both male and female C57BL/6
   and BALB/c mice were used for colon cancer models (MC38, CT26), and
   melanoma model (B16-F10), and female C57BL/6 and BALB/c mice were used
   for breast cancer models (4T1, EMT6, EO771). Before tumor formation,
   mCAFs were mixed with CT26, 4T1, or EMT6 cells at a ratio of 3:1. Tumor
   size estimation utilized the formula: volume (mm^3) = (long axis) ×
   (short axis)^2/2. Anti-PD-1 antibody (RMP1-14, 10 mg/kg, Bioxcell) and
   anti-CTLA-4 antibody (9D9, 5 mg/kg, Bioxcell) were intraperitoneally
   administered on days 5, 7, 9, 11, and 13 post-tumor inoculation. In
   cases of in vivo CD8^+ T cell depletion, the mice received treatment
   with anti-CD8 antibody (YTS 169.4, 10 mg/kg, Bioxcell) every 3 days,
   beginning 3 days before MC38 tumor inoculation. The depletion efficacy
   of CD8^+ T cell was confirmed through flow cytometry analysis.

Cell isolation

   Tumor samples were initially mechanically dissociated using a scissor
   device, followed by enzymatic digestion with Liberase (5401119001,
   2 mg/mL, Roche) in DMEM at 37 °C for 30 minutes to create a homogenate.
   Subsequently, the homogenate was sieved through a 70 µm filter
   (bs-70-xbs, BioSharp) to isolate a single cell suspension, which was
   then treated with red blood cell lysis buffer for 2 minutes, washed,
   and resuspended in flow cytometry buffer for subsequent analysis. For
   CAF isolation, after red blood cell lysis, cells were resuspended in
   complete medium and plated in culture dishes. Following 15-min
   incubation at 37 °C, non-adherent cells were discarded. The adherent
   fraction, enriched for stromal cells including CAFs, was expanded in
   culture. Secondary purification of CAFs was achieved via morphological
   selection and immunophenotypic validation. Following the generation of
   a single cell suspension from the tumor specimens utilizing the
   aforementioned procedure, cells were cultured in Dulbecco’s Modified
   Eagle Medium (DMEM, 11965092, Gibco), enriched with a 10% concentration
   of fetal bovine serum (04-001-1Acs, BI) and an additional 1%
   penicillin-streptomycin (15140122, Gbico) mixture. CAFs were identified
   as cells adhering to the culture vessel within 15 minutes, whereas
   non-adherent cells were discarded.

Flow cytometry analysis

   Using the Fixable Viability Dye eFluor 450 (65-0863-14, Invitrogen,
   1:1000), live cells were stained at 4 °C for 30 minutes to identify
   viable cells. Cells were pre-treated with anti-CD16/32 (101302,
   BioLegend) to block Fc receptors, followed by a 10-minute incubation on
   ice. Following washing, cells were incubated with a combination of
   primary antibodies against various cell surface markers. For mouse
   markers: CD45 (30-F11), CD4 (GK1.5), CD8a (53-6.7), CD11b (M1/70),
   CD140a (APA5), PD-L1 (10 F.9G2), Tim-3 (B8.2C12), PD-1 (MH5A); for
   human markers: CD45 (2D1), CD8a (SK1), CD11b (M1/70), PD-L1 (29E.2A3);
   all sourced from BioLegend. After stimulating cells with the Cell
   Stimulation Cocktail (00-4975-93, Invitrogen) at 37 °C for 4-6 hours,
   cytokines were stained with anti-IFN-γ (XMG1.2-mouse, 4S.B3-human,
   BioLegend) and anti-TNF-α (MP6-XT22-mouse, MAb11-human, BioLegend)
   antibodies. All antibodies were diluted at a ratio of 1:100 unless
   otherwise specified. Stained cells were analyzed using the BD FACSCanto
   II Flow Cytometer and BD FACSDiva software (BD Biosciences), with data
   processed using FlowJo software (version 10.5.3).

CyTOF analysis

   As previously mentioned, individual live cells were extracted from
   tumor tissues for CyTOF analysis. The cells underwent a viability
   assessment using cisplatin (HY-17394, 25 µM, MedChemExpress) for
   1 minute, followed by labeling with a metal-tagged cocktail of
   monoclonal antibodies targeting cell surface molecules. After fixation
   and permeabilization with Buffer (88-8824-00, eBioscience), cells were
   labeled with a panel of monoclonal antibodies targeting intracellular
   antigens. Analysis was performed on the CyTOF 2 platform at the Liver
   Disease Institute, Beijing You’an Hospital. The data files generated
   were standardized and processed with Cytobank (version 9.1). Data were
   transformed using the cytofAsinh method before Phenograph clustering
   analysis, performed with the R cytofkit package version 0.99.0, for
   categorizing immune subsets from consolidated samples. The heatmaps
   depicted the mean intensity for markers across distinct clusters, where
   the frequency of cells within a cluster was determined by the ratio of
   the count of identified cells to the overall count of CD45^+ cells
   within the identical specimen. For the mass cytometry procedure,
   antibodies from Fluidigm were employed, encompassing a range of markers
   including anti-NK1.1 (142Ce), anti-CD40 (160Dy), anti-KI67 (161Dy),
   anti-TIM3 (162Dy), anti-LY6C (163Dy), anti-CD68 (164Dy), anti-GATA3
   (167Er), anti-CD8 (168Er), anti-CD103 (151Eu), anti-CD69 (153Eu),
   anti-LY6G (152Gd), anti-CTLA4 (154Gd), anti-TIGIT (155Gd), anti-CD14
   (156Gd), anti-LAG3 (158Gd), anti-FOXP3 (165Ho), anti-MHCII (175Lu),
   anti-B220 (176Lu), anti-CD11B (143Nd), anti-SIGLECF (144Nd), anti-CD4
   (145Nd), anti-CD127 (141Pr), anti-CD25 (147Sm), anti-TBET (148Sm),
   anti-CD19 (149Sm), anti-CD11C (150Sm), anti-F4/80 (159Tb), anti-CD206
   (169 T), anti-CD45 (89Y), anti-CD80 (171Yb), anti-CD86 (172Yb),
   anti-PD1 (173Yb), and anti-CD3 (174Yb).

Single cell RNA sequencing

   Single-cell suspensions of EO771 tumors were obtained from WT or Aebp1
   cKO mice. Samples of three tumors from the same cohort were amalgamated
   at random to form a composite sample, which was then shipped to Beijing
   Annoroad (ANNOROAD) for sequencing. Subsequent to cell quantification,
   the samples were processed and sequenced on the Illumina NovaSeq 6000
   platform in San Diego, California, following the manufacturer’s
   protocol. Demultiplexing, barcode decoding, sequence alignment, data
   sifting, tallying of unique molecular indices, and consolidation of
   sequencing data were executed using version 1.2 of the Cell Ranger
   suite.

   Subsequent analyses were conducted within the R environment (version
   4.3.2), employing the Seurat toolkit (version 5.0.1) for data
   processing. Cells exhibiting elevated expression of multiple
   cluster-specific marker genes were omitted, suggesting potential
   polyploidy based on gene expression patterns. Additionally, cells with
   gene counts below 200 or exceeding 7000, along with those showing
   mitochondrial-encoded transcripts comprising over 10% of the total
   library, were excluded. Find Variable Features was utilized to pinpoint
   genes with significant variability, subsequently selecting the 2000
   most fluctuating genes for execution in principal component analysis.
   Uniform Manifold Approximation and Projection (UMAP) was employed to
   condense the data dimensions, thereby rendering the deduced cellular
   groupings discernible, which were derived from an analysis of the
   fifteen most salient principal components.

   Immunocyte types were distinguished by comparing gene expression across
   clusters with the Wilcoxon test, then aligning the most variant genes
   with established immunocyte markers to create a standardized marker set
   for 10 clusters.

   We systematically investigated intercellular communication networks
   using CellChat, a computational framework for analyzing ligand-receptor
   interactions, applied to scRNA-seq data from Aebp1 cKO and WT tumors.
   The workflow involved three sequential phases: first, preprocessing raw
   sequencing data and constructing genotype-specific CellChat objects;
   second, calculating interaction probabilities using the tool’s curated
   ligand-receptor interaction database to quantify communication
   likelihoods; third, performing comparative analyses of interaction
   network strengths and patterns between genotypes.

Spatial transcriptomics data analysis

   The stRNA-seq slides of two human COAD samples were printed with two
   identical capture regions. The Visium Spatial platform was employed
   under the default protocol to capture gene expression information in
   the ST slides by combining spatial barcodes with mRNA oligonucleotides.
   The raw sequencing reads of stRNA-seq underwent quality checking and
   alignment using Space Ranger (version 1.1). Subsequently, the gene-spot
   matrix derived from processing the ST and Visium samples was analyzed
   with Seurat (version 5.0.1) in R. A detection threshold of 200 was
   established for genes in individual spots, leading to the exclusion of
   genes with fewer than 10 reads or those detected in less than three
   spots. Spot normalization was conducted using the LogVMR method.
   Dimensionality reduction and clustering using the top 30 principal
   components were performed with Independent Component Analysis (PCA) at
   a resolution of 0.8. Finally, the SpatialFeaturePlot function in Seurat
   was utilized to create a spatial feature expression plot. The library
   preparation was completed, followed by sequencing on the NovaSeq 6000
   system.

Real-time quantitative PCR

   WT, Aebp1^-/- mCAFs, alongside shNC/shAEBP1 mCAFs and hCAFs, were grown
   in DMEM for 48 hours prior to RNA extraction with Trizol reagent
   (15596026, Invitrogen). Subsequently, the RNA was converted to cDNA
   using Superscript III and generic primers, following the manufacturer’s
   guidelines. This cDNA was then used for amplifying specific gene
   transcripts via qPCR with SYBR Green I Master Mix on an ABI PRISM
   7300HT analyzer. For standardization purposes,
   glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was selected as a
   reference gene. The primer sequences used were as follows:

   Mouse Gapdh,

   F: 5′-AGGTCGGTGTGAACGGATTTG-3′. R: 5′-TGTAGACCATGTAGTTGAGGTCA-3′.

   Mouse Cd274,

   F: 5′-GCTCCAAAGGACTTGTACGTG-3′. R: 5′-TGATCTGAAGGGCAGCATTTC-3′.

   Human GAPDH,

   F: 5′-ACAACTTTGGTATCGTGGAAGG-3′. R: 5′-GCCATCACGCCACAGTTTC-3′.

   Human PD-L1,

   F: 5′-TGGCATTTGCTGAACGCATTT-3′. R: 5′-TGCAGCCAGGTCTAATTGTTTT-3′.

Co-culture with T cells in vitro

   Spleens from C57BL/6 mice were filtered through a 40-μm mesh to yield a
   homogenous cell suspension. Post-erythrocyte lysis, cells were
   enumerated and cultured in Roswell Park Memorial Institute (RPMI) 1640.
   Human blood was layered onto Ficoll (abs930, Absin) and centrifuged
   (acceleration 1/deceleration 0) to isolate peripheral blood mononuclear
   cell (PBMCs) from the interface layer. Prior to T cell stimulation,
   12-well plates were coated with anti-CD3 (17A2 for mouse, OKT3 for
   human, Invitrogen) and anti-CD28 (37.51 for mouse, CD28.2 for human,
   Invitrogen) at 2.5 and 3 µg/ml, respectively. After 48 hours, T cells
   were co-cultured with CAFs, and then harvested post another 48 hours
   for flow cytometry.

Bioinformatics

   In the TME, we performed pathway enrichment using the KEGG database via
   R clusterProfiler (v4.10.0) to pinpoint pathways linked to AEBP1. GSEA
   (v4.0.3) was applied for further analysis, complemented by Seurat’s
   capabilities in processing single-cell RNA sequencing (scRNA-seq) data.
   Significance was determined for terms with a p value < 0.05 and at
   least 3 enriched genes.

Western blot

   Cells were lysed in ice-cold buffer containing 20 mM Tris-HCl pH 7.0,
   250 mM NaCl, 1% Triton X-100, 0.5% NP-40, 3 mM EDTA, and protease
   inhibitors. Lysates were resolved by SDS-PAGE, transferred to
   nitrocellulose membranes, and blocked with 5% milk for 30 minutes at
   room temperature. Membranes were incubated overnight at 4 °C with
   primary antibodies, followed by 1-hour room temperature incubation with
   HRP-conjugated secondary antibodies. Protein signals were visualized
   using enhanced chemiluminescence reagent (34580, Thermo). The primary
   antibodies used in this study included anti-AEBP1 (ab254973, Abcam,
   1:1000; sc-271374, Santa, 1:1000), anti-CKAP4 (sc-393544, Santa,
   1:1000), anti-AKT (4691, Cell Signaling Technology, 1:1000), anti-p-AKT
   (4060, Cell Signaling Technology, 1:1000), anti-GAPDH (60004-1-Ig,
   Proteintech, 1:10000).

Co-immunoprecipitation and mass spectrometry

   Cells (human/mouse CAFs and 293 T) were lysed in IP buffer, and
   supernatants were pre-cleared with Protein A/G beads. For
   immunoprecipitation, CAF lysates were incubated with 2.5 μg anti-AEBP1
   (sc-271374; Santa Cruz) while 293 T lysates received anti-HA (M20003S;
   Abmart), with IgG (B30010S; Abmart) as control. After overnight
   rotation (4 °C), Protein A/G beads were added (2 h), washed with IP
   buffer, and eluted in SDS loading buffer for western blot. For mass
   spectrometry, mCAF samples were electrophoresed, Coomassie-stained
   (P0017; Beyotime), and analyzed by Shanghai Huaying Biotechnology Co.,
   Ltd. (China).

Docking-based virtual screening and molecular docking

   A software package from Schrödinger (Schrödinger, LLC, New York, 2015)
   was utilized for virtual screening of small-molecule inhibitors
   targeting the AEBP1-CKAP4 interaction. The protein structure
   preparation involved the use of the Prep Wiz module, while the
   identification of active sites, defined as docking pockets, was done
   using the SiteMap module. Screening of compounds from Chembridge and
   Chemdiv libraries excluded PAINS compounds through the Canvas 1.1
   program. ADME properties were predicted with the QikProp 3.2 program.
   Molecular docking employed the Glide algorithm and analysis was based
   on docking scores. Cluster analysis was conducted using FCFP_6
   fingerprints. The most effective small-molecule inhibitor was
   identified, and its mechanism of inhibition was investigated.
   Protein-protein docking calculations were carried out with the ZDOCK
   module in Discovery Studio 4.0 software. Suggestions for edits to avoid
   plagiarism include paraphrasing and restructuring sentences, which
   maintain the original content without textual duplication. The
   compounds were sourced from the commercial chemical libraries of
   ChemDiv (San Diego, CA, USA) and ChemBridge Corporation (San Diego, CA,
   USA).

BioRender statement

   Some figures were generated using BioRender.com (Fig. [230]1A,
   [231]7A). Publication licenses were secured in compliance with
   BioRender’s academic licensing terms, and appropriate attribution is
   provided within the corresponding figure legends.

Statistics

   Data were statistically analyzed using GraphPad Prism (version 8.4.0)
   and R (version 4.3.2). Significance was calculated using an unpaired
   student t-test unless otherwise specified. For comparisons involving
   more than two groups, one-way ANOVA with Dunnett’s or Tukey’s multiple
   comparisons test, or two-way ANOVA with Sidak’s or Tukey’s multiple
   comparisons test were applied. Survival analysis utilized the log-rank
   test. A significance level of p <  0.05 was deemed statistically
   significant, with exact p values displayed on all graphs. Data were
   presented as mean ± SEM.

Reporting summary

   Further information on research design is available in the [232]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [233]Supplementary Information^ (2.6MB, pdf)
   [234]Reporting Summary^ (114KB, pdf)
   [235]Transparent Peer Review file^ (1.3MB, pdf)

Source data

   [236]Source Data^ (1MB, xlsx)

Acknowledgements