Extracellular Vesicle-Packaged circTAX1BP1 from Cancer-Associated Fibroblasts Regulates RNA m6A Modification through Lactylation of VIRMA in Colorectal Cancer Cells

Abstract

The underlying molecular mechanism of patients with colorectal liver metastasis (CRLM) remains unclear. In this study, it is found that cancer-associated fibroblasts (CAFs)-derived extracellular vesicles (EVs) are significantly enriched in circTAX1BP1 in CRLM, which associate with poor prognosis. The disruption of EV-packaged circTAX1BP1 significantly inhibits CRLM in vivo and in vitro. Mechanistically, CAF-derived EV-packaged circTAX1BP1 is delivered to colorectal cancer (CRC) cells, where it binds to VIRMA and promotes its lactylation at lysine residue 1713 by recruiting AARS2. Lactylated VIRMA enhances m⁶A-based modification and stability of SP1 mRNA. SP1 mediates the transcription of TGF-β, enhancing epithelial-mesenchymal transition and paracrine TGF-β of CRC cells. Notably, this study identifies an important subgroup ITGA11⁺ myCAFs through single-cell RNA sequencing data. Paracrine TGF-β of CRC cells specifically targets ITGA11⁺ myCAFs, activating the TGF-β signalling pathway, which contributes to extracellular matrix remodeling and increases delivery of EV-packaged circTAX1BP1, forming a positive feedback loop to promote CRLM. Finally, the combined blockade of EV-packaged circTAX1BP1 and TGF-β can effectively disrupt this feedback loop and significantly inhibit tumor progression in a PDX model. Overall, this study provides an in-depth understanding of tumor cell-CAFs crosstalk and new insights into therapeutic targets for CRLM.

Keywords: cancer‐associated fibroblasts; circRNAs; colorectal cancer; extracellular vesicles; tumor microenvironment.

Figure 1

Elevated circTAX1BP1 expression in CAFs within the colorectal cancer stroma is associated with liver metastasis. A,B) Unsupervised hierarchical clustering of normalized counts from high‐throughput sequencing of normal adjacent tissues (NAT), primary tumors (PT), and liver metastasis (LM). The pseudocolor represents the intensity scale of PT versus NAT or the LM versus PT, generated by a log2 transformation. C) Schematic representation of circTAX1BP1 upregulation in colorectal cancer (CRC) tissues and LM‐positive tissues. D) qRT‐PCR analysis of circTAX1BP1 expression in a 192‐case cohort of freshly collected human CRC samples and NATs. E,F) Correlation of circTAX1BP1 expression in CRC tissues (n = 192) assessed using qRT‐PCR with pathological grade (E) and LM status (F). G) Comparison of circTAX1BP1 expression in primary human CRC samples and paired metastatic LMs. H,I) Kaplan–Meier curves for OS (H) and DFS (I) of patients with CRC with low vs. high expression of circTAX1BP1. The median expression level of circTAX1BP1 was taken as the cutoff value. p‐values were calculated by the log‐rank (Mantel–Cox) test. J) Schematic of circTAX1BP1 formation. The circTAX1BP1 back‐splicing junction was identified through Sanger sequencing. K) circTAX1BP1 was present in HCT116, as determined via qRT‐PCR using convergent and divergent primers. L,M) qRT‐PCR confirmation of circTAX1BP1 stability after RNase R treatment (n = 3). N,O) Following treatment with actinomycin D, the half‐lives of circTAX1BP1 and linear TAX1BP1 were measured (n = 3). P) CRC tissues were stained with pan‐cytokeratins (cancer nest) and α‐SMA (stroma) through immunofluorescence assay. Scale bar, 200 µm. Q) In situ analysis with Cy3‐labelled RNA using a circTAX1BP1 FISH probe in CRC tissues. Scale bar, 200 µm. R) Protein levels of vimentin, α‐SMA, FAP and PDGFR‐α were detected using western blotting in isolated CAFs and NFs. S) Immunofluorescence staining showed the subcellular location and the expression of pan‐cytokeratins and α‐SMA in isolated CAFs and NFs. Scale bar, 50 µm. T) qRT‐PCR was used to analyze circTAX1BP1 level in isolated CAFs and NFs (n = 9). The statistical difference was assessed through the nonparametric Mann–Whitney U‐test in (D–G); and 2‐tailed Student's t‐test in (L–O,T). All data are presented as mean ± SD of experimental triplicates. ns, P > 0.05; **, P < 0.01; ***, P < 0.001.

Figure 2

CAF‐derived exosomal circTAX1BP1 enhances the proliferation, migration, and invasion of colorectal cancer (CRC) cells. A) Representative images of CAF‐ and NF‐derived exosomes analyzed via transmission electron microscopy. Scale bar, 600 nm. B) Size distribution of the isolated exosomes analyzed via nanoparticle tracking. C) Western blotting analysis for exosomal markers CD63, ALIX, TSG101, and GM130 of HCT116 cells and CAF‐ or NF‐derived exosomes. D,E) CAFs were pre‐treated with or without GW4869. CAFs were labelled with CM‐Dil (red) and co‐cultured with HCT116 for 18 h. Immunofluorescence analysis of CM‐Dil distribution in cells. Scale bar, 50 µm. Scale bars insets, 25 µm. F) qRT‐PCR was used to analyze circTAX1BP1 level in the isolated CAF‐ and NF‐derived exosomes of five patients (n = 3). G,H) HCT116 and DLD1 cells were incubated with indicated exosomes or PBS for 48 h, and circTAX1BP1 level was analyzed using qRT‐PCR (n = 3). I) Colony formation assays were performed for HCT116 and DLD1 cells. J,K) Proliferation of HCT116 and DLD1 cells was measured using CCK‐8 assays (n = 3). L) Representative images of wound healing assay using HCT116 and DLD1 cells showing cell motility. M) Representative images of Transwell assays, and migration and invasion were determined for HCT116 and DLD1 cells. Scale bar, 100 µm. N) Schematic of the xenograft nude mice model. O) Images of xenograft tumors harvested from nude mice. P) Tumor volume summary for mice, measured every 7 days (n = 5). Q) Nude mice were examined for tumor weights (n = 5). R) circTAX1BP1 levels were assessed using qRT‐PCR in xenograft tumors harvested from nude mice (n = 5). S) IHC staining for Ki67 was measured (n = 5) in xenograft tumors harvested from nude mice. Scale bar, 100 µm. T) Representative livers were acquired from the nude mice (n = 5). U,V) Representative images of metastatic nodules, and measurement of metastatic nodules of mouse livers were examined using haematoxylin–eosin staining (n = 5). Scale bar, 400 µm. The statistical difference was assessed by a 2‐tailed Student's t‐test in (F–H,Q,R,V); and one‐way ANOVA followed by Dunnett tests in (J,K,P). All data are presented as mean ± SD of experimental triplicates. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Figure 3

circTAX1BP1 in CAF‐derived EVs binds VIRMA and promotes VIRMA lactylation, elevating m⁶A levels in colorectal cancer (CRC). A) LC‐MS/MS analysis of circTAX1BP1 pulldown. B) Interaction of circTAX1BP1 with the m⁶A writer protein VIRMA was determined using RIP assays of HCT116 and DLD1 cells (n = 3). C) Interaction propensity prediction between circTAX1BP1 and VIRMA by CatRAPID. D) circTAX1BP1 and VIRMA interaction was verified using RNA pull‐down assays with HCT116 and DLD1 cells. circTAX1BP1 plasmid (5 µg) was transfected into 5 × 10⁵ cells using Lipofectamine 3000. E) A circTAX1BP1 Mut plasmid or a WT plasmid was transfected into HCT116 and DLD1 cells; subsequently, a circTAX1BP1 probe was used for RNA pull‐down assays. F,G) RNA m⁶A dot blot assays were used to assess the m⁶A levels of total mRNA. Methylene blue staining was used to determine the loading control. H) Relative VIRMA protein level was detected in HCT116 and DLD1 cells using western blotting. I) Western blot analyses were performed, and nuclear and cytoplasmic localizations of VIRMA were quantified for HCT116 and DLD1 cells. J,K) Cell lysates of HCT116 were immunoprecipitated with anti‐VIRMA or control IgG, followed by immunoblotting. L) Identification and quantification of VIRMA K1713 lactylation. LC–MS/MS analysis of modified FFTPPAS (Kla) GNYSR is shown. M) Flag‐VIRMA (WT) and Flag‐VIRMA (K1713R) cells were treated with lactate (25 mm) for 24 h, and whole‐cell extracts were collected for immunoprecipitation with anti‐VIRMA antibody, followed by immunoblotting. N,O) RNA m⁶A dot blot assays were used to assess the m⁶A levels of total mRNA. Methylene blue staining was used to determine the loading control. The statistical difference was assessed by a 2‐tailed Student's t‐test in (B). All data are presented as mean ± SD of experimental triplicates. **, P < 0.01.

Figure 4

CAF‐derived EVs deliver circTAX1BP1 to recruit lactyltransferase AARS2, enhancing VIRMA lactylation and elevating m⁶A levels. A) LC‐MS/MS analysis of anti‐VIRMA IP. B) Effects of AARS2 overexpression or knockdown by two independent shRNAs on VIRMA lactylation expression. C) RNA pull‐down assays showing the association of the in vitro circularized circTAX1BP1 with VIRMA and AARS2. D) Structural model prediction of the circTAX1BP1‐VIRMA‐AARS2 ternary complex by AlphaFold3. E) circTAX1BP1 overexpression in CRC cells increased the association of AARS2 with VIRMA. F) The increased AARS2‐VIRMA association by circTAX1BP1 depends on the direct interaction between VIRMA and circTAX1BP1. G) Immunoblot assays to measure the expression of VIRMA lactylation and AARS2 proteins in HCT116 and DLD1 stable cell lines expressing circTAX1BP1 or AARS2 shRNA. H) RNA m⁶A dot blot assays were used to assess the m⁶A levels of total mRNA of different stable cells expressing circTAX1BP1 or AARS2 shRNA. I) Colony formation assays of different stable cells expressing circTAX1BP1 or AARS2 shRNA. J,K) CCK‐8 assays of different stable cells expressing circTAX1BP1 or AARS2 shRNA (n = 3). L) Wound healing assays of different stable cells expressing circTAX1BP1 or AARS2 shRNA. M) Cell migration and invasion assays of different stable cells expressing circTAX1BP1 or AARS2 shRNA. Scale bar, 80 µm. The statistical difference was assessed by one‐way ANOVA followed by Dunnett tests in (J,K). All data are presented as mean ± SD of experimental triplicates. **, P < 0.01.

Figure 5

CAF‐derived EV‐packaged circTAX1BP1 promotes colorectal cancer (CRC) progression by lactating VIRMA to modulate SP1 m⁶A modification. A) Gene ontology of biological processes at the mRNA level. B,C) KEGG and GSEA enrichment analysis of the number of pathways at the mRNA levels. D) Volcano plots of upregulated and downregulated genes in circTAX1BP1 siRNA compared to those in siRNA con. E) SP1 expression in tumor and normal samples based on the TCGA and GTEx dataset. F) m⁶A modification of SP1 in the SRAMP prediction servers. G,H) MeRIP‐qPCR determinations for m⁶A enrichment on SP1 mRNA in HCT116 cells (n = 3). I,J) Relative activities of the WT and Mut luciferase reporters (n = 3). K) Relative protein levels of SP1 were determined using western blotting. L) Relative protein levels of SP1 and AARS2 were determined using western blotting. M–O) Relative protein levels of AARS2, SP1, TGF‐β1, p‐SMAD2, SMAD2, p‐SMAD3, and SMAD3 were determined using western blotting. P,Q) Relative protein levels of AARS2, SP1, N‐Cadherin, E‐Cadherin, Vimentin, Slug, and Snail were determined using western blotting. R) Proliferation of HCT116 and DLD1 cells was assessed using colony formation assays. S) Proliferation of HCT116 cells was assessed using CCK‐8 assays (n = 3). T) Migration of HCT116 and DLD1 cells was assessed using wound healing assays. U) Migration and invasion of HCT116 and DLD1 cells were assessed using transwell assays. Scale bar, 80 µm. The statistical difference was assessed through nonparametric Mann–Whitney U‐test in (E); and one‐way ANOVA followed by Dunnett tests in (G,I,J,S); and 2‐tailed Student's t‐test in (H). All data are presented as mean ± SD of experimental triplicates. ns, P > 0.05; **, P < 0.01; ***, P < 0.001.

Figure 6

TGF‐β secreted by colorectal cancer (CRC) cells targets ITGA11⁺ myCAFs and activates the TGF‐β signalling pathway. A) Enriched motifs of SP1 binding sites predicted using JASPAR. B) Schematic model of SP1 binding sequences in the TGF‐β promoter region predicted using JASPAR and PROMO. C) Schematic plot of pGL3‐Basic vector containing wild‐type and two mutated motifs of the TGF‐β promoter. D,E) Relative activities of the WT and Mut luciferase reporters (n = 3). F,G) ChIP‐qPCR analysis of SP1‐enriched chromatin in HCT116 and DLD1 cells (n = 3). H–K) Relative TGF‐β1 and SP1 expressions were detected in HCT116 and DLD1 cells using qRT‐PCR (n = 3). L) t‐SNE plot showing the integration of single cells from metastatic and non‐metastatic CRC samples, partitioned into eight distinct cell clusters. M) TGFR1 and TGF‐β expression profiling across eight cell clusters reveals elevated expression in CAFs. N) Re‐clustering of CAFs partitioned this cellular compartment into 14 cell subclusters. O) t‐SNE plot showing clustering of 14 CAF populations into Clusters 1 (TGF‐β‐low myCAF), 2 (iCAF), and 3 (TGF‐β myCAF) based on expression profiles. P) Dimensionality reduction clustering of the three CAF subtypes in CRC with and without LM. Q) Proportions of the three CAF subtypes in CRC with and without LM, showing an elevated Cluster 3 proportion in CRC with LM. R) Heatmap of highly expressed genes in the three CAF subtypes. S) Violin plot shows the expression of the TGFR1, TGFR2, ITGA11, and collagen‐related genes in three CAF subtypes. T) Log2 fold‐change (y‐axis) and percentage difference (x‐axis) of gene expression (dots) between Cluster 3 and Clusters 1 and 2. U) t‐SNE plots of the myCAF cluster (left) showing heterogeneous in TGF‐β score (middle) and expression of ITGA11 (right). V) TGF‐β signalling pathway expression in the three CAF subtypes. W) In single‐cell samples, higher levels of ITGA11⁺ myCAF scores and TGF‐β signalling were observed in CRC with LM than in CRC without LM. X) Correlation between TGF‐β scores and ITGA11⁺ myCAF scores in TCGA‐COAD samples (n = 461). Y) Survival analysis of high (n = 173)‐ and low (n = 288)‐expression ITGA11⁺ myCAF score groups in TCGA‐COAD samples. The median expression level of ITGA11⁺ myCAF scores was taken as the cutoff value. p‐values were calculated by the log‐rank (Mantel–Cox). Spearman correlation analysis was used in (X). The statistical difference was assessed through one‐way ANOVA followed by Dunnett tests in (D,E,H–K,V); and 2‐tailed Student's t‐test in (F,G); and χ ² test in (Q); and nonparametric Mann–Whitney U‐test in (W). Data are presented as mean ± SD of experimental triplicates. ns, P > 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Figure 7

TGF‐β drives ITGA11⁺ myCAFs to upregulate EV‐packaged circTAX1BPP1 and ECM to form a feedback loop. A) CRC tissues were stained with PDPN and ITGA11 in immunofluorescence assays. Scale bar, 100 µm. B,C) Hallmark (B) and KEGG (C) pathway enrichment analysis of highly expressed pathways across three CAF subtypes. D,E) Schematic and representative flow plot of flow cytometric and flow‐sorting strategy of ITGA11⁺ myCAF. F) qRT‐PCR analysis of relative mRNA expression in iCAFs, ITGA11⁺‐d myCAFs, or ITGA11⁺ myCAFs (n = 3). G) Relative protein levels of TGF‐β1, p‐SMAD2, SMAD2, p‐SMAD3, and SMAD3 were determined using western blotting. H) Schematic presentation of the established co‐culture model of fibroblasts and EV‐induced CRC cells. I) qRT‐PCR analysis of relative mRNA expression in ITGA11⁺ myCAFs cultured with PBS or TGF‐β treatment (n = 3). J) qRT‐PCR analysis of relative mRNA expression in ITGA11⁺ myCAFs cultured with indicated CM from CRC cells with or without fresolimumab treatment (n = 3). K) qRT‐PCR analysis of circTAX1BP1 expression in iCAFs, ITGA11⁺‐d myCAFs, or ITGA11⁺ myCAFs (n = 3). L) qRT‐PCR analysis of circTAX1BP1 expression in ITGA11⁺ myCAFs cultured with indicated CM from CRC cells with or without fresolimumab treatment (n = 3). M) Relative protein levels of TGF‐β1, p‐SMAD2, SMAD2, p‐SMAD3, and SMAD3 were determined using western blotting. N) Enriched motifs of smad3 binding sites predicted by JASPAR. O) Schematic model of smad3 binding sequences in the circTAX1BP1 promoter region predicted by JASPAR and PROMO. P) Schematic plot of pGL3‐Basic vector containing wild‐type and mutated two motifs of the circTAX1BP1 promoter. Q) Relative transcriptional activity of circTAX1BP1 in TGF‐β‐treated ITGA11⁺ myCAFs with or without smad3 silencing (n = 3). R) ChIP‐qPCR analysis of smad3‐enriched chromatin in ITGA11⁺ myCAFs (n = 3). S) Relative activities of the WT and Mut luciferase reporters (n = 3). The statistical difference was assessed through one‐way ANOVA followed by Dunnett tests in (F,J–L,Q); and 2‐tailed Student's t‐test in (I,R,S). All data are presented as mean ± SD of experimental triplicates. ns, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Figure 8

Clinical relevance of the circTAX1BP1‐induced positive feedback loop in colorectal cancer (CRC) liver metastasis. A) IHC results of SP1, TGF‐β1, p‐Smad2, and p‐Smad3 in CRC and normal tissues in samples from CRC patients with or without LM. Scale bar, 400 µm. B–E) qRT‐PCR analysis of SP1 and TGF‐β1 expression between CRC tissues and NATs (n = 192) or between LM‐negative (n = 144) and LM‐positive (n = 48) CRC tissues. F,G) Kaplan–Meier curves of the OS and DFS of patients with CRC with low vs. high TGF‐β1 expression levels. The cutoff value is the median. p‐values were calculated by the log‐rank (Mantel–Cox) test. H) Schematic illustration of the establishment of the PDX model. I) Images of PDX tumors harvested from the mice. J,K) Tumor volume summary for mice, measured every 4 days (n = 3). L,M) Mice were examined for tumor weights (n = 3). N) IHC staining for SP1, TGF‐β, p‐Smad2, and p‐Smad3 was measured in PDX tumors harvested from the mice. Scale bar, 200 µm. O,P) qRT‐PCR analysis of circTAX1BP1 expression in plasma EVs from patients with CRC (n = 30) and healthy controls (n = 30) or CRC patients with (n = 8) or without (n = 22) LM. Q) ROC curves for the efficiency of plasma EV‐mediated circTAX1BP1 in diagnosing CRC. The statistical difference was assessed through nonparametric Mann–Whitney U‐test in (B–E,O,P); and one‐way ANOVA followed by Dunnett tests in (J–M). All data are presented as mean ± SD of experimental triplicates. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Figure 9

EV‐packaged circTAX1BP1 from ITGA11⁺ myCAFs regulates RNA m⁶A modification through lactylation of VIRMA.