Oncogenic KRAS Drives Lipofibrogenesis to Promote Angiogenesis and Colon Cancer Progression

Colorectal cancer is responsible for more than 50,000 deaths in the United States and 600,000 deaths globally each year (1, 2). Although the 5-year survival rate is 90% for localized colorectal cancer, it is a dismal 14% for advanced metastatic disease (3). The signature genetic alterations of colorectal cancer include activating oncogenic mutations in KRAS (42%) and inactivating mutations and/or deletions of APC (75%) and TP53 (60%) tumor suppressors [ref. 4; The Cancer Genome Atlas-Colon Adenocarcinoma (TCGA)-COAD]. In human colorectal cancer, the most common activating KRAS mutation, KRAS^G12D, is a glycine to aspartic acid substitution at codon 12 that correlates positively with advanced disease (5) and shows high cooccurrence in primary and metastatic tumors, supporting a role for KRAS^G12D in colorectal cancer progression (6). Correspondingly, studies of mice engineered with a doxycycline-inducible KRAS^G12D allele and conditional null alleles for APC and TP53 (designated “iKAP” mice) have established that KRAS^G12D (hereafter denoted as KRAS*) represses antitumor immunity and drives increased invasive and metastatic disease (7, 8).

Current targeted growth factor receptor therapies, such as those aimed at EGFR, fibroblast growth factor receptor, and insulin-like growth factor 1 receptor (IGF1R), have shown limited clinical benefit in metastatic colorectal cancers harboring KRAS*, consistent with the actions of KRAS* in regulating cell survival and proliferative signaling downstream of these growth factor receptors. Alternative strategies for KRAS*-driven colorectal cancer have embraced targeting the downstream pathways of KRAS* such as the PI3K/mTOR and KRAS*/MEK pathways. However, the inhibition of mTOR (9) or MEK (10) have failed to demonstrate meaningful clinical activity in patients with KRAS*-driven colorectal cancer, suggesting the presence of alternative or redundant KRAS* signaling pathways. Consistent with the redundancy thesis, combined MEK inhibitor (binimetinib), BRAF inhibitor (encorafenib), and EGFR inhibitor (cetuximab) treatment has shown increased clinical benefit in patients with colorectal cancer with the BRAF^V600E mutant, although recurrence still occurs in most cases, and treatment only extends the median survival duration by 3.6 months (11). The complexity of KRAS* function in colorectal cancer biology has been further substantiated by the discovery that KRAS* suppresses antitumoral immunity via the recruitment of myeloid-derived suppressor cells (8). While new therapies directly targeting KRAS* (e.g., adagrasib and sotorasib) show promising antitumor activity, the tumors of patients treated with these therapies rapidly acquire bypass mechanisms of resistance (12), underscoring the need to better understand the KRAS* signaling network to identify alternative or synergistic therapeutic interventions for patients with KRAS*-driven colorectal cancer.

These continuing challenges prompted further exploration of the function of KRAS* in colorectal cancer tumor biology, particularly its impact on the host cells of the tumor microenvironment, such as immune cells, endothelial cells, and cancer-associated fibroblasts (CAF). CAFs are among the most abundant, albeit the least understood and characterized, cells in the tumor microenvironment. CAF subtypes can be identified by markers, including alpha smooth muscle actin (αSMA), fibroblast activation protein, and platelet-derived growth factor receptor (PDGFR). Myofibroblastic CAFs (αSMA⁺) and non–myofibroblastic CAFs (αSMA⁻) appear to be the most consistently defined subtypes present across diverse cancers (13–16). Myofibroblasts are lipid-sparse cells that transdifferentiate from adipocytes or lipofibroblasts (17, 18). CAF subtypes play opposing roles in the promotion or suppression of cancer (19, 20). Specifically, αSMA⁺ fibroblasts suppress LGR5⁺ cancer stem cells and restrain colorectal cancer progression (21). Conversely, αSMA⁻ CAFs secrete lipid metabolites into the tumor microenvironment via fatty acid translocase, thus promoting colorectal cancer metastasis by enhancing membrane fluidity (22, 23).

Mounting evidence suggests that specific cancer genotypes can influence CAF biology (24). For instance, p53 loss activates JAK2–STAT3 signaling to drive stromal fibrogenesis and promote tumor growth (25). KRAS*-induced sonic hedgehog (SHH) secretion shapes stromal cells to initiate reciprocal signaling of KRAS* via AXL/IGF1R–AKT signaling in pancreatic cancer (26). KRAS* reprograms pancreatic fibroblasts to activate inflammatory genes and polarize protumorigenic macrophages (27). BRCA-mutated cancer cells activate heat shock factor 1 (HSF1) in pancreatic stellate cells (PSC), which transform into CLU⁺ immune regulatory CAFs (28). While KRAS* and other cancer genes can influence aspects of CAF biology, whether and how cancer cell oncogenic events influence CAF subtype biology and heterogeneity remains an area of active investigation.

In this study, our exploration of the biology and mechanisms of KRAS*-driven colorectal cancer progression revealed the capacity of KRAS* to transform the cell state of fibroblasts in the tumor microenvironment. We found that KRAS* activated the transcription factor CP2 (TFCP2) that directly upregulated transcription of proadipogenic cytokines, stimulating the phenotypic transformation of specific CAF subpopulations into lipid-rich CAFs that produced abundant VEGFA to promote angiogenesis and disease progression. In addition, the genetic or pharmacologic neutralization of TFCP2 impeded lipid-rich CAF development, blocked tumor angiogenesis, and inhibited tumor growth in KRAS*-expressing colorectal cancer. These insights, together with consonant human colorectal cancer profiles, provide a testable precision oncology therapeutic option for this intractable disease.

To identify pathways and processes that correlate with both KRAS* signaling and disease progression in colorectal cancer, we performed a gene-set enrichment analysis (GSEA) of 19 bulk tumor RNA sequencing profiles to identify upregulated genes in iKAP murine tumors (in which KRAS* is present and functioning) compared with iAP (control) murine tumors (in which no KRAS* transgene is present) and in invasive primary tumors compared with noninvasive primary tumors. These intersected KRAS* and progression transcriptomes revealed prominent representation of the adipogenesis pathway in invasive KRAS*-expressing colorectal cancer (Fig. 1A; Supplementary Fig. S1A). These in silico findings aligned with tumor histology results, which showed a dramatic increase in lipid droplet abundance in the invasive iKAP tumor stroma compared with noninvasive iAP controls (Fig. 1B; Supplementary Fig. S1B). Lipid-rich iKAP tumor stroma does not colocalize with αSMA and epithelial cellular adhesion molecule (CD326) staining (Fig. 1B); Lipid-rich iKAP tumor stroma colocalizes with delta-like homolog (DLK1), a marker of preadipocytes and lipid-rich fibroblast progenitors, further attesting to the adipocyte-like lineage enrichment in the stroma of KRAS*-expressing colorectal cancer (Supplementary Fig. S1B). Flow cytometry analysis gated on LipidTOX⁺, PDGFRα-positive [PDGFRα⁺ (CD140a, a non–myofibroblast marker)], CD326^– (a colorectal cancer epithelium marker), CD45^– (an immune cell marker), and CD31^– (an endothelial marker) cells revealed an increased percentage of lipid-rich CAFs in KRAS*-expressing invasive iKAP tumors compared with noninvasive iAP tumors (Fig. 1C; Supplementary Fig. S1C). IHC and Western blot analyses corroborated the transcriptomic findings by confirming the enrichment of lipid droplets and adipocyte markers and regulators—DLK1, TCF21, PLIN4, FABP4, and LPL (29–31)—in KRAS*-expressing invasive iKAP tumors compared with noninvasive iAP controls; and the expression of fibroblast markers, COL3A1, THY1 and PDGFRα (Fig. 1D; Supplementary Fig. S1D and S1E). To decipher KRAS dependency and the stage specificity of lipid-rich CAF enrichment during colorectal cancer development, we examined lipid-rich CAFs during tumor progression in high-, moderate- and low-grade invasive and noninvasive iKAP and iAP tumors. The intensity of positive lipid staining is more abundant in high-grade invasive iKAP tumors compared with moderate- and low-grade invasive and noninvasive iKAP and iAP tumors (Fig. 1E; Supplementary Fig. S1F), indicating KRAS* and tumor stage dependency of lipid-rich CAF enrichment in colorectal cancer development.

To further characterize the KRAS*-driven stromal response, we conducted single-cell RNA sequencing (scRNA-seq) comparisons across invasive iKAP, noninvasive iAP, and noninvasive iKAP–DOXoff (KRAS*-off) colorectal cancer tumors. Among the 71,525 cells analyzed, 5,984 stromal cells were identified, including (i) pericytes (RGS4- and RGS5-enriched), (ii) three endothelial cell (PECAM1- and ENG-enriched) subgroups, and (iii) six fibroblast (PDGFRα- and PDPN-enriched) subgroups (Supplementary Fig. S2A). The fibroblast subtypes, including ECM⁺ CAFs (COL12A1, COL1A2, and COL6A2); MMP⁺ CAFs (MMP13, MMP10, and MMP9); PI16⁺ CAFs (PI16, IL33, and CLIP); inflammatory CAFs (CXCL13, TNFSF13B, and C4B); proliferating fibroblasts; and myofibroblasts (ACTG2, TAGLN, and ACTA2; Fig. 1F; Supplementary Fig. S2A), were defined using previously established markers (32–36). Importantly, four CAF subgroups were notable for their expression of adipocyte- and/or adipogenesis-related genes (Supplementary Fig. S2A, right). For example, ECM⁺ CAFs express IGF1, which induces lipid production (37); MMP CAFs express NRG1, which fine-tunes white adipose stem cell differentiation (38); PI16⁺ CAFs express PI16⁺, which is the marker of PDGFRα⁺ adipocyte progenitor cells (39); and inflammatory CAFs express SFRP2, which enhances the adipogenic differentiation of dental mesenchymal stem cells (MSC; ref. 40). To validate these findings, we analyzed scRNA-seq data containing 2,666 stroma cells from six KRAS-mutant and six KRAS wild-type tumors from patients with colorectal cancer. These murine CAF annotations mirrored those in KRAS* and KRAS wild-type human colorectal cancer, as reflected by scRNA-seq analyses showing six fibroblast subgroups including PI16⁺, SFRP⁺, inflammatory, extracellular matrix (ECM), myofibroblasts and KI67⁺ fibroblasts (Supplementary Fig. S2B, left). Moreover, the PI16⁺, SFRP⁺, inflammatory, and ECM fibroblasts express adipocyte and/or adipogenesis-related genes, PI16, SFRP2, RARRES1, and SFRP4, respectively (refs. 41, 42; Supplementary Fig. S2B, right).

Furthermore, we characterized the murine CAF subtypes corresponding to lipid-rich CAFs by using gene-set module score analyses in murine scRNA datasets, which revealed that the adipocyte gene signature (43), and the combination of classic adipocyte and preadipocyte genes (29–31, 44, 45) were enriched in the 4 αSMA^– CAF subgroups—inflammatory, PI16⁺, ECM⁺, and MMP⁺ CAFs (Fig. 1F; Supplementary Fig. S2C)—which express classic adipokine genes (DPT, DCN, and CCL2; Supplementary Fig. S2D, left; refs. 44, 45) and lipid-rich fibroblast regulators and markers (TCF21, FGF10, PLIN3, and LPL; Supplementary Fig. S2E, left; ref. 30). In contrast, αSMA⁺ myofibroblastic CAFs show low or no expression of such signatures and markers, consistent with the lack of lipid droplets in this tumor-suppressive CAF subtype. Correspondingly, the same analyses of human colorectal cancer scRNA-seq revealed an adipocyte gene signature enriched in PI16⁺ and SFRP⁺ fibroblasts as well as adipokine genes and lipid-rich fibroblast markers [Supplementary Fig. S2D and S2E (right) and S2F]. Attesting to the scRNA-seq characterization of lipid-rich CAFs in vivo, transcriptomic analysis of flow-sorted, lipid-rich CAFs (LipidTOX⁺/PDGFRα⁺ and CD326^–/CD45^–/CD31^–) showed a lack of αSMA (ACTA2) expression and enrichment of the differential expression genes in the four CAF subgroups (Fig. 1G), reflecting the association between the lipid-rich fibroblast signature and the four αSMA^– CAF subgroups.

Finally, scRNA-seq analysis showed increased lipid-rich CAFs (adipocyte gene signature and pre-/mature adipocyte genes are enriched in the combination of four αSMA⁻ CAF subgroups, with the PI16⁺ and inflammatory CAFs subtypes showing highest expression) in KRAS*-expressing invasive iKAP tumors compared with noninvasive iAP and KRAS*-off (no doxycycline) iKAP controls (Fig. 1F, bottom right, and 1H; Supplementary Fig. S2G). These results aligned with the enrichment of lipid-rich CAFs (LipidTOX⁺/PDGFRα⁺) in KRAS*-expressing colorectal cancer tumors (Fig. 1C). These murine findings mirror those from KRAS* human colorectal cancer scRNA-seq data showing increased lipid-rich CAFs (PI16⁺ and SFRP⁺ fibroblasts; Supplementary Fig. S2H) and from TCGA analyses comparing APC⁻/TP53⁻/KRAS* versus APC⁻/TP53⁻/KRAS^WT (Supplementary Fig. S2I–S2M) showing a positive correlation between KRAS*-driven colorectal cancer and the lipid-rich CAF (αSMA⁻) gene signatures (Supplementary Table S1; Supplementary Fig. S2N). Together, these observations establish a positive correlation between KRAS*-driven invasiveness and lipid-rich CAF enrichment in human and murine colorectal cancer.

The enrichment of lipid-rich CAFs in KRAS*-expressing tumor stroma prompted us to examine potential for KRAS* to regulate the expression of proadipogenic factors. As noted, lipofibrogenesis appears to coopt the mechanisms driving classical adipogenesis. Specifically, the preadipocyte commitment phase of lipofibrogenesis and adipogenesis is driven by the inhibition of WNT signaling (46) and the activation of BMP signaling (47). PDGFRα⁺ preadipocytes accumulate lipids via an FGF10 autocrine loop (48), and lipid droplet accumulation can be further promoted by insulin, cyclic AMP, or glucocorticoids via PPARγ and C/EBPs upregulation (49).

As shown in Fig. 2A, the intersection of (i) differentially expressed genes in invasive iKAP versus noninvasive iAP, (ii) the human secretome, and (iii) the known pro- and antiadipogenic secreted factors revealed KRAS*-regulated expression of six key BMP and WNT factors. In addition, doxycycline modulation of KRAS* expression produced marked changes in the expression of well-established lipid-rich and proadipogenic regulatory factors and markers (Supplementary Fig. S3A); and GSEA of stromal populations showed KRAS*-dependent enrichment of two major preadipocyte commitment drivers: RESPONSE_TO_BMP and NEGATIVE_REGULATION_OF_WNT (Fig. 2B; Supplementary Fig. S3B). To validate these KRAS*-regulated proadipogenic cytokines, we measured their expression in KRAS*-on (+doxycycline) versus KRAS*-off (−doxycycline) primary two-dimensional iKAP cancer cell cultures, revealing KRAS*-induced upregulation of BMP4 and WNT5B (Fig. 2C; Supplementary Fig. S3C). BMP4 is known to induce the transformation of pluripotent stem cells into adipocytes via the SMAD and p38MAPK pathways (47); and WNT5B participates in noncanonical WNT signaling to inhibit β-catenin nuclear localization and also indirectly promotes the adipogenesis of progenitor cells (49). The KRAS*–WNT5B/BMP4 connection was confirmed in (i) two patient-derived xenograft organoids (PDXO) engineered with inducible KRAS activation (Fig. 2D, (ii) the murine colorectal cancer cell line CMT93 with and without KRAS*, (iii) iAP and iKAP tumor organoids, (iv) the human colorectal cancer cell line DLD1 isogenic for KRAS* and KRAS wild-type cell lines, and (v) the KRAS* and KRAS wild-type human colorectal cancer in TCGA dataset (Supplementary Fig. S3D). In addition, IHC staining confirmed increased BMP4 and WNT5B levels in KRAS*-expressing iKAP tumors compared with iAP controls (Fig. 2E). WNT5B and BMP4 expression levels positively correlate with tumor stages in the iKAP model but not in the iAP model (Fig. 2F; Supplementary Fig. S3E) indicating the KRAS* and stage dependency of Bmp/Wnt signaling in colorectal cancer development. This stage dependency may relate to the subclonal nature of KRAS-mutant cancer cells in early-stage disease (50), as well as changes in KRAS* downstream signaling as a function of tumor progression (51).

We next sought to validate the proadipogenic activity of KRAS*-regulated BMP4 and WNT5B in several model systems. First, the 3T3L1 cell culture model of adipocyte differentiation (52) confirmed that supplementation of insulin with recombinant BMP4/WNT5B increased adipocyte markers/regulators expression relative to insulin-only controls (Supplementary Fig. S3F). Second, undifferentiated 3T3L1 cell treatment with KRAS*-on, but not KRAS*-off, iKAP conditioned medium (CM) stimulated differentiation into adipocyte progenitor cells and lipid-rich fibroblasts (Fig. 2G; Supplementary Fig. S3G), which expressed the adipocyte marker PPARG and lipid-rich fibroblast markers FGF10, TCF21, and PLIN4 (Fig. 2G, right). Third, using two PDXOs engineered with a doxycycline-inducible KRAS* allele and human colorectal cancer DLD1 cell line isogenic for KRAS* or wild-type KRAS showed that treatment of human MSCs (hMSC) with KRAS* CM-stimulated differentiation to adipocyte-like cells expressing lipid-rich fibroblast genes to a greater degree than did treatment with wild-type KRAS CM (Fig. 2H; Supplementary Fig. S3H and S3I). Finally, we observed that lipofibrogenesis was increased specifically in collagen-expressing tumor stroma in tumors generated by the coinjection of 3T3L1 and KRAS*-on iKAP cancer cells and decreased in the tumors after 1-week KRAS* extinction by doxycycline withdrawal (Fig. 2I; Supplementary Fig. S3J).

In addition, insulin pathway signaling is observed in iKAP bulk tumors and KRAS*-driven tumor stroma (Supplementary Fig. S3K), indicating the presence of an active adipogenesis process within the KRAS* stroma. Together, these data support the view that KRAS* drives the expression of the proadipogenic cytokines WNT5B and BMP4 to promote stromal lipofibrogenesis.

These observations prompted us to investigate the potential biological impact of lipid-rich versus lipid-sparse CAFs on tumors. To that end, we isolated primary lipid-rich and lipid-sparse CAFs from iKAP tumors by flow sorting LipidTOX⁺/PDGFRα⁺/CD326^–/CD45^–/CD31^– and LipidTOX^–/PDGFRα^–/CD326^–/CD45^–/CD31^– cell populations for use in coinjection studies (Fig. 3A and B). The small population of lipid-sparse CAFs required brief expansion in vitro (see Methods “Lipid–sparse CAFs/myofibroblasts isolation”) and were subsequently validated for the expression of αSMA and RGS5 to confirm the CAF myofibroblastic cell type (Supplementary Fig. S4A). Orthotopic coinjection of iKAP cancer cells and sorted, lipid-rich CAFs generated larger tumors (Fig. 3C) and reduced overall survival (Fig. 3D); immunofluorescence analysis of these tumors confirmed abundant lipid droplets in the tumor stroma (Fig. 3E). In contrast, orthotopic coinjection of iKAP cancer cells and lipid-sparse CAFs (myofibroblastic CAFs) generated tumors that grew more slowly and were associated with prolonged survival; immunofluorescence analysis revealed a paucity of lipid droplets in the tumor stroma and more prominent representation of αSMA⁺ cells (Fig. 3D and E). These findings are aligned with the above primary tumor single-cell data, showing that αSMA⁺ CAFs are distinct from lipid-rich CAF subpopulations (Fig. 1F; Supplementary Fig. S2A). Consistent with these murine studies, human colorectal cancer (TCGA) analyses established a positive correlation between advanced disease and increased lipid-rich CAF gene signatures (Supplementary Table S1: the combination of top 10 differentially expressed genes in four αSMA⁻ CAF subgroups from mouse scRNA-seq; Supplementary Table S2: the combination of adipocyte and fibroblast genes; PanglaoDB; Supplementary Fig. S4B) and a worse prognosis (Fig. 3F; Supplementary Fig. S4C). Of note, myofibroblast gene signature (GSEA) and myofibroblast differentiation gene signature (GSEA) scores have no impact on colorectal cancer prognosis (Supplementary Fig. S4D). Together, these results suggest that lipid-rich CAFs promote the progression of mouse and human colorectal cancer.

To determine how KRAS* drives proadipogenic cytokine gene expression, we surveyed transcription factor (TF) expression (53), TF signatures, and consensus TF-binding motifs present in the promotors of proadipogenic cytokine genes exhibiting KRAS*-dependent expression in murine and human colorectal cancer (Supplementary Table S3). Two TF candidates emerged from this data integration: sex-determining region Y (SRY, also known as testis-determining factor), which is responsible for the sex determination of males (54); and α-globin TFCP2 (Fig. 4A; Supplementary Fig. S5A). We focused hereafter on TFCP2, given that it was the top candidate and that ERK1, a downstream effector of KRAS, is known to phosphorylate and increase TFCP2's DNA-binding activity (55). SRY was deprioritized because it is a Y chromosome gene expressed solely in males, yet our data indicated that KRAS* drives proadipogenic cytokine gene expression in both sexes.

To validate TFCP2's role in mediating KRAS*-driven, proadipogenic cytokine gene expression, TFCP2 was depleted (via short hairpin RNA) or deleted [via clustered regularly interspaced short palindromic repeats (CRISPR)] in iKAP cell lines. As an experimental control, we generated TFCP2-null clones that were rescued by reexpression of wild-type TFCP2. In the clones with TFCP2 depletion or deletion, WNT5B and BMP4 expression was decreased in the KRAS*-expressing cells yet restored in the TFCP2-rescued null controls (Fig. 4B; Supplementary Fig. S5B). Chromatin immunoprecipitation (ChIP) using anti-TFCP2 antibodies confirmed the direct binding of TFCP2 to its consensus binding motifs in the WNT5B and BMP4 promotors (Fig. 4C; Supplementary Fig. S5C and S5D). In addition, WNT5B and BMP4 reporter expression was decreased in the KRAS*-expressing TFCP2 depletion cells and restored by TFCP2-reexpressed rescue, validating the TFCP2 binding in WNT5B and BMP4 promotors (Supplementary Fig. S5E). The role of TFCP2 in regulating the biology of CAFs and other cells of the tumor microenvironment was reinforced by the strong positive correlation between TFCP2-high gene signature scores and various stromal lineages, including endothelial cells, fibroblasts, and adipocyte signatures (ref. 43; Fig. 4D). Functionally, the syngeneic and xenograft injections of iKAP TFCP2-depleted cancer cells generated smaller tumors that possessed fewer lipid-rich CAFs, whereas the lipid-rich CAFs were abundant in the TFCP2-rescued null controls (Fig. 4E–G; Supplementary Fig. S5F). It is worth noting that TFCP2 depletion had no significant impact on cancer cell proliferation in vitro (Supplementary Fig. S5G and S5H), suggesting that its prime actions may be exerted at the level of the tumor microenvironment.

TFCP2-high gene signature scores correlated positively with advanced invasive and metastatic disease and poor survival in patients with colorectal cancer (Fig. 4H; Supplementary Fig. S5I). This finding motivated us to determine the therapeutic impact of the TFCP2-specific small-molecule inhibitor FQI1, which targets the TFCP2 DNA–binding domain and eliminates downstream transcriptional activity (56). We first determined that FQI1 downregulated the canonical TFCP2-regulated genes SPP1, MMP9, and CFH (57) in cultured FQI1-treated iKAP cells (Supplementary Fig. S5J). Next, we validated that FQI1 treatment of iKAP tumor cell lines decreased the expression of WNT5B and BMP4 (Fig. 4I). In addition, FQI1 treatment of tumor-bearing iKAP mice reduced the accumulation of lipid-rich CAFs, and prolonged survival in autochthonous iKAP mice (Fig. 4J and K). Together, these findings provide strong evidence that KRAS*-dependent colorectal cancer progression is driven in part by the stimulation of lipofibrogenesis via the TFCP2-mediated regulation of proadipogenic cytokines.

To better understand the biological actions of lipid-rich CAFs in KRAS*-expressing colorectal cancer tumors, we surveyed murine and human colorectal cancer transcriptomic profiles for pathways and cell types enriched in tumors with high lipid-rich CAF gene signature scores. The top pathways were the ANGIOGENESIS and KRAS_SIGNALING_UP pathways. Per the cell-type gene signature database, Panglao DB (43), the cell type occurrences associated with the highest lipid-rich CAF signature scores were fibroblasts, adipocytes, and endothelial cells (Fig. 5A; Supplementary Fig. S6A). Notably, a heat map analysis revealed that the endothelial cell signature was enriched in both the lipid-rich CAFs and KRAS*-driven invasive colorectal cancer tumors (Fig. 5B). These in silico data prompted us to hypothesize that a prime action of lipid-rich CAFs is to promote angiogenesis in the tumor microenvironment.

To identify the factors that may promote tumor angio­genesis in lipid-rich CAFs, we intersected several datasets and performed functional assays utilizing iKAP CM–educated 3T3L1 cells and KRAS*-expressing DLD1 CM–educated hMSCs. First, using adipokine array profiling, we determined that several secreted factors with known pivotal roles in tumor angiogenesis—including VEGFA, macrophage colony-stimulating factor, hepatocyte growth factor, lipocalin 2, pentraxin 3 (PTX3), and DLK1, a noncanonical ligand of Notch signaling—were upregulated in iKAP CM–educated 3T3L1 cells (Fig. 5C). Second, we performed a tumor scRNA-seq analysis, which confirmed increased VEGFA, pentraxin 3, and hepatocyte growth factor expression in the KRAS*-expressing iKAP tumor stroma compared with iAP and KRAS*-off (doxycycline-off) iKAP tumor controls (Fig. 5D; Supplementary Fig. S6B). Correspondingly, reverse transcription PCR (RT-PCR) validated the significantly and consistently increased expression of VEGFA in DLD1 CM–educated hMSCs (Fig. 5E). Third, IHC and cytometry analysis confirmed the enrichment of VEGFA expression and the endothelial cell population (CD326^–/CD45^–/CD31⁺) in KRAS*-expressing iKAP and KRAS*-expressing DLD1 tumors (Fig. 5F; Supplementary Fig. S6C). The positive correlation between lipid-rich CAFs and VEGFA expression and endothelial cell counts in iKAP tumors aligned well with the results of our immunofluorescence analysis, which showed an enrichment of VEGFA and an increased number of CD31⁺ endothelial cells in iKAP tumors and in tumors generated by the coinjection of iKAP cancer cells and LipidTOX⁺-sorted cells (Supplementary Fig. S6D). Finally, on the functional level, we assessed the proangiogenic activity of lipid-rich CAFs in human umbilical vein endothelial cell (HUVEC) tube formation assays. The addition of CM from DLD1–KRAS^G12D CM–educated hMSCs stimulated an increase in all conventional endothelial cell metrics, including the covered area, total tube length, total number of branching points, and total number of loops of the endothelial structure (Fig. 5G). Thus, lipid-rich CAFs promote tumor angiogenesis via the secretion of VEGFA.

Finally, the role of TFCP2 in mediating KRAS*-induced lipofibrogenesis and angiogenesis was solidified by demonstrating that TFCP2 depletion decreased VEGFA and endothelial cell density in KRAS*-expressing iKAP tumors compared with controls (Fig. 5H; Supplementary Fig. S6E). In addition, inhibiting TFCP2 with FQI1 reduced VEGFA expression and the abundance of CD326^–/CD45^–/CD31⁺ endothelial cells in KRAS*-expressing iKAP tumors (Fig. 5I). Analysis of the TCGA colorectal cancer dataset reinforced the TFCP2–VEGFA connection, as evidenced by the poor prognosis of patients with colorectal cancer with high VEGFA expression, a TFCP2-high signature, or a combined VEGFA- and TFCP2-high signature (Figs. 4H and 5J). Correspondingly, treatment with FQl1 alone, the murine anti-VEGFA neutralizing antibody B20 alone, or both all impaired colorectal cancer progression in iKAP GEM (Fig. 5K). These survival curves uncover two noteworthy observations. First, treatment with FQl1 alone was equivalent to combination treatment with FQI1 and B20, consistent with VEGFA as a key downstream target for TFCP2. Second, FQI1 or FQI1+B20 treatments were more effective than B20 alone, suggesting that TFCP2 acts on additional targets beyond VEGF such as its additional angiogenesis targets as noted above (Fig. 5C). In contrast, FQI1 treatment of the iAP colorectal cancer model had no impact on improving survival relative to vehicle-treated controls (Supplementary Fig. S6F), reinforcing the specificity of TCP2 inhibition on angiogenesis to extend survival specifically in the context KRAS*. We conclude that KRAS*-activated TFCP2 promotes the development of lipid-rich CAFs which contribute to the progression of colorectal cancer primarily via VEGFA-induced tumor angiogenesis. The preservation of the KRAS*–TFCP–VEGFA axis in human KRAS*-driven colorectal cancer portends its translational potential of clinical research and further raises the possibility that targeting TFCP2 may prove more effective than VEFGA neutralization in colorectal cancer.

In this study, exploration of the mechanisms governing KRAS*-mediated colorectal cancer progression revealed its role in activating TFCP2-induced transcriptional upregulation of proadipogenic cytokines in cancer cells, which in turn stimulates the lipogenic transformation of CAFs in the tumor microenvironment. These lipid-rich CAFs produce abundant VEGFA to promote tumor angiogenesis and disease progression. We found that the genetic or pharmacologic neutralization of TFCP2 impeded CAF adipogenesis, reduced tumor angiogenesis, and inhibited tumor growth in mouse and human KRAS*-expressing colorectal cancer models. In colorectal cancer, KRAS* is known to regulate numerous hallmarks, including those intrinsic to cancer cells and those regulating the tumor immune microenvironment. This study now expands the known actions of KRAS* to the phenotypic transformation of CAFs with enhanced protumorigenic properties. Our findings also indicate that the KRAS*–TFCP2–VEGFA axis provides a testable therapeutic option for patients with KRAS*-driven colorectal cancer.

The lipid-laden CAF cells identified in this study are reminiscent of the lipid interstitial cells present in the developing murine lung (58, 59) and cells in adipogenesis (49). First, analogous to the lipid-rich CAF–endothelial cell heterotypic interactions, lung lipofibroblasts provide a nurturing environment for stem cell growth (60, 61) by enhancing oxidative defenses, providing leptin to stimulate surfactant synthesis in type 2 alveolar epithelial cells, and promoting type 2 cell proliferation (62). Second, additional parallels between lipid–rich CAFs and alveolar lipofibroblasts are seen in lineage-tracing analyses. These analyses show that PDGFRα⁺ progenitor stromal cells differentiate into alveolar lipofibroblasts (63), which contain an abundance of neutral lipids; lipid droplet–associated protein (perilipin); the adipocyte markers TCF21, FABP4, LPL, and ZFP423; membrane-tethered COL13A1; and FGF10 (29–31). Similarly, lipid-rich CAFs and alveolar lipofibroblasts share a similar gene expression profile that includes LPL, perilipin, TCF21, FGF10, PI16, CLO12A, and MMP13 (Fig. 1D; Supplementary Fig. S1D and S1E). Third, in both lipofibrogenesis and adipogenesis, FGF10 has been implicated in the transformation of PDGFRα⁺ preadipocytes into beige adipocytes via a miRNA–327–FGF10–FGFR2 autocrine loop (48). Similarly, KRAS* CM induced the expression of FGF10 which educated and transformed embryonic fibroblast and MSCs into lipid-rich fibroblasts in cell culture (Fig. 2G and H; Supplementary Fig. S3I). KRAS* functioned similarly in the iKAP tumor microenvironment (Supplementary Fig. S2F). Among the similarities between lipid-rich CAFs and adipocytes is their emergence from a common progenitor cell, the PDGFRα⁺ preadipocyte (48). In adipogenesis, the inhibition of canonical WNTs and the activation of BMP signaling facilitates the commitment of preadipocytes toward an adipogenic fate; and insulin stimulates adipocyte maturation with lipid droplet accumulation. Correspondingly, iKAP cell lines exhibit KRAS*-dependent enrichment of the two main proadipogenic cytokines, the key molecules BMP4 and WNT5B (Fig. 2C; Supplementary Fig. S3C); iKAP bulk tumors, and KRAS*-driven tumor stroma exhibit insulin pathway signaling (Supplementary Fig. S3K) consistent with an active adipogenesis process in KRAS* stroma. Together, these findings establish that lipid-rich CAFs coopt the mechanisms governing normal lipofibrogenesis and adipogenesis programs.

Our focus on angiogenesis relates first and foremost to tumor histopathologic analyses showing that KRAS* is associated with increased tumor angiogenesis. Second, an unbiased GSEA analysis of KRAS*-regulated pathways that are shared and enriched in both mouse and human KRAS* colorectal cancer showed a strong and prominent correlation between EMT and angiogenesis pathways and the lipid-rich CAF gene signature. Third, cytokine array screens of the CAF-secreted factors show the secretion of VEGFA, MCSF, DLK1, HGF, and LCN2 in KRAS*–CM cultured–embryonic fibroblasts (lipid–rich CAFs). Together, these data support our prioritization of angiogenesis as a major cancer hallmark regulated by lipid-rich CAFs. More generally, this prioritization is justified on several levels by evidence supporting angiogenesis as a driver of tumor progression including (i) increased VEGFA expression correlating negatively with overall survival in TCGA (Fig. 5J), (ii) higher CD31 staining in invasive iKAP tumors relative to noninvasive iAP tumors (Fig. 5F), and (iii) the clinical benefit of anti–VEGFA mAb in iKAP model (Fig. 5K). Thus, we are confident that angiogenesis is a very important aspect of the tumor biological impact of lipid-rich fibroblasts in KRAS* colorectal cancer, although we acknowledge that we cannot exclude the possibility lipid-rich CAFs modulates addition cancer hallmarks that contribute to tumor progression such as EMT (Fig. 5A).

With respect to the mechanism of VEGF upregulation in lipid-rich CAFs, we speculate that prostaglandin E2 (PGE2) is the upstream metabolite driving VEGFA expression specifically in lipid-rich CAFs. Lipid droplets are known to be the major reservoirs of diverse metabolites including PGE2 (64). Our expression data show that the key PGE2 synthesis enzymes and PGE2 receptors are indeed enriched in the lipid-rich CAFs, but not in the myofibroblasts (Supplementary Fig. S6G). Moreover, PGE2 ELISA analysis shows that PGE2 is enriched in the iKAP tumors relative to iAP tumors and in the CM from KRAS-on iKAP fibroblasts (lipid-rich fibroblasts; Supplementary Fig. S6H and S6I). PGE2 has been shown to bind its receptor PTGER to activate downstream ERK/JNK signaling which can upregulate VEGFA expression (65)—accordingly, PTGER1 and 4 are expressed in lipid-rich CAFs (Supplementary Fig. S6G, bottom) and depletion of the key PGE2 synthesis enzyme PTGS2 in the KRAS-on CM-educated 3T3L1 cells results in decreased VEGFA expression (Supplementary Fig. S6J). Thus, the lipid-rich CAF-derived PGE2 can upregulate VEGF upregulation via a paracrine, autocrine, and/or intracrine manner as proposed by (66) and operates similarly in our model system.

Our work identifies TFCP2 as a key downstream target of KRAS* in driving the proangiogenic adipogenic program. TFCP2 regulates MMP9 expression, which promotes angiogenesis in hepatocellular carcinoma (HCC; ref. 57). TFCP2 binds to YAP as a cofactor to regulate gene transcription, including MMP9, FN1, and TJP1 and contributes to YAP-dependent liver malignancy (67). In colorectal cancer, however, there remained a more limited understanding of TFCP2's actions. Previous work has shown that TFCP2 binds to CPEB1's hypermethylated TF-binding region to suppress the expression of CPEB1 and promote tumor metastasis (68). In our colorectal cancer studies, TFCP2 acted in an entirely different manner by directly upregulating WNT5B and BMP4 gene transcription to spur the development of VEGFA-expressing, lipid-rich CAFs (Fig. 4C and F). In addition, although TFCP2 regulates MMP9 expression in HCC, this was not observed in the colorectal cancer model, likely because MMP9 is mostly expressed in myeloid cells and has low or no expression in cancer cells. Moreover, TFCP2-driven lipid-rich CAFs are the major source of VEGFA (another source being myeloid cells), indicating their important contribution in tumor angiogenesis. Given that a high TFCP2 signature is associated with a worse prognosis in patients with colorectal cancer (Fig. 4H), it is encouraging that, in our preclinical models, FQ1's inhibition of TFCP2 was well tolerated and impaired lipofibrogenesis, tumor angiogenesis, and tumor progression (Figs. 4J, 4K, 5I, and 5K). FQI1 has been shown to impair tumor growth in subcutaneous xenograft and spontaneous Alb–cMyc HCC models (56, 69), pointing to the need for additional investigations of the role of TFCP2 in other tumor types and genotypes. In colorectal cancer, our findings rationalize the testing of adjuvant FQI1 treatment in patients with early-stage KRAS*-driven colorectal cancer at high risk for metastases, that is, stage II patients with large amounts of circulating tumor DNA after surgery. Moreover, the lipid-rich CAF/TFCP2–high signature may serve as a patient selection biomarker to enlist patients with colorectal cancer into such targeted TFCP2 trials designed to reduce disease recurrence. Current therapies targeting KRAS* in cancers show promise, but resistance to KRAS* inhibitor therapies is occurring in human trials, underscoring the need to define the spectrum of targetable biological actions of KRAS* in colorectal cancer.

iKAP primary cell line cultures were sorted from GFP⁺, iKAP-dissociated tumors and cultured in DMEM (Gibco; 11995–065) with or without doxycycline supplementation (2 μg/mL). CMT93 (ATCC; CCL–223), 293T (ATCC; CRL–3216), and 3T3L1 (ATCC; CL–173) cell lines were cultured in DMEM, supplemented as defined by ATCC. DLD1 isogenic KRAS wild-type colorectal cancer cells were cultured in RPMI1640 medium (Gibco; 11875–093). hMSCs (ATCC; SCRC–4000) was obtained from ATCC and cultured in MSC basal medium (ATCC PCS-500–030), supplemented as defined by ATCC. HUVECs (CRL–1730) were obtained from ATCC and cultured in F-12K medium (ATCC; 30–2004), supplemented as defined by ATCC. All cell lines were cultured in the indicated medium containing 10% FBS or bovine calf serum (Sigma) and an 1% antibiotic–antimycotic (Gibco; 15–240–062) following ATCC's instructions. All cells were confirmed to be Mycoplasma–free using Mycoplasma Elimination Kit (Sigma; MP0030) and maintained at 37°C and 5% CO₂.

For supplemental recombinant Bmp4 and Wnt5b proteins used for adipogenic cell differentiation, 100 ng/mL Bmp4 (R&D Systems; 5020–BP–010) and 100 ng/mL Wnt5b (R&D Systems; 3006–WN–025) were added to complete 3T3L1 medium. After 2 weeks, the medium was supplemented with 10 μg/mL insulin for 2 days. The transcription expression of adipogenic genes was validated using RT–qPCR.

CM was collected from iKAP cell lines, DLD1 cell lines, and PDXOs after culturing for 24 hours in complete culture medium as specified for each experiment. Following a published protocol (70), in vitro differentiation was achieved by coculturing 3T3L1 cells or hMSCs with the iKAP CM or DLD1 cell lines’ CM for 16 days. Old CM was exchanged for fresh CM every 2 days. In the last two days, adipo­genic terminal differentiation was initiated by adding insulin to 10 μg/mL (Sigma; 11070–73–8, and Sigma; 11061–68–0), 3–isobutyl–1–methylxanthine 500 μmol/L (Sigma; I7018), and dexamethasone 700 nmol/L (Sigma; D4902). Immunoblotting, RT-qPCR, IHC, and immunofluorescence procedures were performed to validate adipogenic differentiation by confirming the presence of adipocyte and lipid-rich fibroblast-related genes.

Human studies followed Helsinki Declaration ethical guidelines and were conducted under an Institutional Review Board-approved protocol at MD Anderson Cancer Center (Houston, TX). All patients provided written, informed consent for tissue distribution and specimen collection. To generate organoids, primary human tumor xenografts were minced and digested in digestion buffer [DMEM (Gibco; 11995–065) plus 100 U/mL penicillin/streptomycin (Gibco; 15140–122), 500 U/mL Collagenase IV (Thermo Fisher Scientific; 17104019), and 125 mg/mL Dispase type II (Thermo Fisher Scientific; 17105041)] for 30 minutes at 37°C. The samples were washed with DMEM/5% FBS and PBS sequentially, and the digested tumor cells were resuspended in Matrigel and plated onto a 24-well culture plate (approximately 10% confluent in 30 mL droplet per well). After the gel solidified, 500 μL of WENR medium [advanced DMEM/F12 medium (Gibco; 12634-010) plus 1 mmol/L GlutaMAX (Gibco; 35050-061), 1 mmol/L HEPES (Gibco; 15630-080), and 100 U/mL penicillin/streptomycin (Gibco; 15140-122), supplemented with 50% Wnt-3A CM from L-Wnt3A cells (ATCC; CRL-2647), 10% R-spondin 1 CM from R-spondin 1 cells (Sigma; SCC111), 100 ng/mL hNoggin (R&D Systems; 6507-NG), 10 nmol/L hGastrin I (Sigma; G-9145), 500 nmol/L A83-01 (Tocris; 2939), 10 μmol/L SB202190 (Sigma; S7067), 10 mmol/L nicotinamide (Sigma; N0636), 1× B27 supplement (Gibco; 008–0085SA), 1× N2 supplement (Gibco; 17502–048), 1 mmol/L N-acetyl cysteine (Sigma; A9165), and 50 ng/mL hEGF (R&D Systems; 236-EG)] was added to wells and incubated at 37°C.

Established organoids were routinely passaged every 10 to 14 days, at a cell density of approximately 50% confluent. Organoids were released from Matrigel by incubating with Cell Recovery Solution (Corning; 354253) at 4°C for 30 to 40 minutes. Released organoids were broken down into small fragments by physical pipetting, or further dissociated with TrypLE Express Solution (Gibco; 12604013) at 37°C for 3–5 minutes to obtain smaller fragments or homogeneous single cells. After washing with PBS, the resulting fragments and cells were resuspended with Matrigel (Corning; 356231) and plated onto a 24-well culture plates (Thermo Fisher Scientific; 142475) with a typical split ratio = 1:3–1:4. Culture medium was refreshed every 2 to 3 days. For quality control, the organoid cultures were regularly checked short tandem repeat sequencing for Mycoplasma. For long-term storage, organoid fragments or dissociated single cells were resuspended in cold Recovery Cell Culture Freezing Medium (Gibco; 12648010), gradually frozen down to –80°C, and stored in liquid nitrogen.

The iKAP colorectal cancer mouse model (Villin-Cre-ERT;Apclox/lox;Trp53lox/lox;tet-O-LSL-Kras^G12D) and iAP mouse model (Villin-Cre-ERT;Apclox/lox;Trp53lox/lox) were previously established and the histopathologic identification of invasive and noninvasive tumors in the models were previously validated as described (7). Mouse colony management and tumor observations under colonoscopy were conducted as previously described (7). To induce colorectal tumors in iKAP and iAP mice, 20 μL of 4-OHT (1 mg/mL) in 100% ethanol was injected directly into the distal colon lumen followed by doxycycline water feeding (200 mg/kg) at 10 to 16 weeks of age. Treatment as described here was initiated after mice had received one 4-OHT injection and 14 days of doxycycline water. 4-OHT–injected iKAP mice (equal numbers of males and females) were randomly assigned to receive FQI1 or VEGFA mouse antibody (B20) treatment. FQI1 (1 mg/kg) or vehicle (PBS + 0.4% DMSO) were given 5 days a week; B20 (10 mg/kg) or IgG isotype control (10 mg/kg) were given three times a week by intraperitoneal injection. Mice were sacrificed when they became moribund, and their tumors were collected.

All surgical manipulations were performed under isoflurane anesthesia, and all animal experiments were approved by The University of Texas MD Anderson Cancer Center's Institutional Animal Care and Use Committee. For the orthotopic xenograft coinjection model, 10⁵ iKAP cells and 10⁴ fluorescence-activated cell-sorted, lipid-rich or lipid-scarce fibroblasts in growth factor–reduced Matrigel were coinjected into the cecum wall of the mice. Mice were monitored for survival and sacrificed when moribund. A set of mice was sacrificed 4 weeks after injection to measure tumor volumes. For the syngeneic transplantation and xenograft tumor models, 10⁵ Tfcp2-engineered iKAP cells were suspended in growth factor–reduced Matrigel and injected subcutaneously. Syngeneic transplanted tumors were collected at 4 months and xenograft tumors were collected at 1.5 months.

Four mTfcp2 knockout guide RNAs were designed in CHOPCHOP (71) and were screened for their efficacy in reducing mRNA and protein levels by at least 80%. The guide RNA sequences 5′–GACGACGTGATCCAAATCTGCGG–3′ and 5′–GCCTGTCGTGGAACACTACGCGG–3′ were inserted into LentiCRISPR v2 plasmids following the standard protocol (72). LentiCRISPR v2 was a gift from Feng Zhang (Addgene plasmid # 52961, http://n2t.net/addgene:52961; RRID, Addgene_52961). Serial dilution of each mTfcp2 knockout clone in a 96-well plate produced colonies from single cells. After testing the knockout efficacy of Tfcp2 in each colony, two single clones from two guide RNA knockout cells were designated as cell lines Tfcp2–KO1 and Tfcp2–KO2. Four short hairpin RNAs targeting mTfcp2 in the pLKO.1 vector (Sigma) were screened and chosen for their efficacy of reducing mRNA and protein levels by at least 80%. The TRCN0000218063, TRCN0000085495, and TRCN0000225944 clones (Sigma) were selected for further use. mTfcp2 overexpression plasmids were purchased from GenScript (NM_033476.3; cloning mTfcp2 protein-coding sequence into pInducer20 vector, mTfcp2–DYK_pInducer20). Lentiviral particles (8 μg) were generated by transfecting 293T cells with the packaging vectors psPAX2 (4  μg) and pMD2.G (2  μg). Lentiviral particles in medium were collected 48 and 72 hours after the transfection of 293T cells, and the medium was filtered through a 0.45-μm filter (Corning; 430514). Viral medium was used in a 1/3 dilution of complete medium for cell culture. After 24 hours, fresh DMEM was provided. At 48 hours, cells were selected following 72 hours of treatment with puromycin (2 μg/mL). Cell lines were then cultured in complete medium.

Immunoblotting was performed following standard protocol (the R&D Systems Quality Control Western Blot Protocol). Antibodies, including β-actin (Sigma, #A3854), vinculin (Millipore 05–386), Bmp4 (Abcam, ab124715), Bmp2 (Abcam, ab82511), Wnt5a (Abcam, ab229200), Wnt5b (Abcam, ab94914), Wnt10b (Abcam, ab70816), insulin (Abcam, ab181547), KRAS^G12D (Cell Signaling Technology, 14429S), and Tfcp2 (Cell Signaling Technology, 80784S), were purchased. Quantifications were performed using ImageJ software, and the results were normalized using the first control band.

Invasive iKAP, noninvasive iAP, and noninvasive iKAP-DOXoff (KRAS*-off) colorectal cancer tumors were collected from microdissected localized adenomas and invasive adenocarcinomas. These tumors were analyzed by IHC and immunofluorescence as described below.

IHC was performed following R&D Systems’ protocol for the preparation and chromogenic IHC staining of frozen tissue sections and protocol for the preparation and chromogenic IHC staining of paraffin-embedded tissue sections. Lipid droplet staining by Oil Red and/or LipidTOX were performed on frozen tissue sections. Antibody staining was done on paraffin-embedded tissue sections after antigen retrieval using a pressure cooker (95°C for 30 minutes followed by 120°C for 10 seconds) and incubation in antigen-unmasking solution (Vector Laboratories). For frozen tissue sections, antibody and lipid droplet staining was performed after 4% formalin fixation. Oil Red O Stain Kit (Lipid Stain; ab150678) was used for lipid droplet staining. The antibodies used included Pdgfrα (Cell Signaling Technology, #3174), Lpl (Invitrogen, PA585126), KRAS^G12D (Cell Signaling Technology, 14429S), Thy1 (eBioscience, 14–0900–81), Col3a1 (ProteinTech, 22734-1-AP), Fabp4 (Abcam, ab92501), Dlk1 (Abcam, ab119930), Tcf21 (Invitrogen, PA5116012), Vegfa (Invitrogen, MA5-32038), and Cd31 (Invitrogen, PA5-16301). Slides were scanned using a Pannoramic 250 Flash III scanner (3DHISTECH Ltd), and images were captured using Pannoramic Viewer software (3DHISTECH Ltd). IHC quantification was conducted following bioprotocol (73) to measure 3,3′-diaminobenzidine staining and quantify the average pixel intensity. Immunofluorescence was performed using the protocol for the preparation and fluorescent IHC staining of frozen tissue sections (R&D Systems). Purchased lipid-staining dyes included HCS LipidTOX Deep Red Neutral Lipid Stain (Invitrogen, H34477) and HCS LipidTOX Red Neutral Lipid Stain (Invitrogen, H34476). Antibodies included those specific for αSMA (Abcam, ab124964), Cd326 (eBioscience, 11-5791-82), Dlk1 (Abcam, ab119930), Fabp4 (Abcam, ab92501), Col3a1 (ProteinTech, 22734-1-AP), Lpl (Millipore, MABS1270), Pdgfrα (Cell Signaling Technology, #3174), and Cd31 (Invitrogen, PA5–16301). Images were captured using a fluorescence microscope (Leica DMi8). Quantification was performed using Fuji (74) and ImageJ (75).

iKAP and iAP tumors were dissociated following the protocol, dissociation of single cell suspensions from human breast tissues at protocols.io (http://dx.doi.org/10.17504/protocols.io.t3aeqie). Dissociated primary cells were stained with antibodies and/or dyes, including Cd326 (eBioscience, 11-5791-82), Cd45 (BioLegend, 103137), Cd31 (BioLegend, 102427), CD140a (BioLegend, 135906), and LipidTOX Deep Red Neutral Lipid Stain (Invitrogen, H34477). After being washing twice using cold PBS, lipid-rich CAFs and endothelial cells were analyzed using a flow cytometer (BD LSRFortessa X–20 Cell Analyzer) or were flow-sorted using a BD FACSAria Flow Cytometer.

Lipid-sparse CAFs were collected by gating on CD326⁻, CD45⁻, CD31⁻, PDGFRα⁻, and LipidTOX^low. The lipid-sparse CAFs were cultured in a 2D culture dish to expand myofibroblasts and exclude pericytes following the JoVE and MIMB protocols (76, 77). Briefly, the selection was conducting by culturing cells in RPMI1640 medium (Gibco; 11875–093) with 10% FCS, 50 μmol/L 2-mercaptoethanol (Sigma; M6250–10ML), 100 μmol/L asparagine (Sigma; PHR2350), 2 mmol/L glutamine (Sigma; 1294808) and 1% antibiotic–antimycotic (Gibco; 15–240–062) in 2D culture dish (Corning; 430167). Pericyte grows on pericyte growth medium [pericyte growth supplement (ScienCell 1252) in pericyte medium (ScienCell 1201)] and grows on Collagen I–coated plate. After expanding the myofibroblasts for subculturing, cells were validated by measuring ACTA2, RGS5, and PDGFRα expression levels by RT–qPCR and immunofluorescence staining. The validated myofibroblasts were used for in vivo tumor coinjection assay.

Chromatin immunoprecipitation–quantitative PCR (ChIP-qPCR) was performed using the standard protocol described previously (78). Briefly, iKAP cells were crosslinked for 10 minutes using 1% paraformaldehyde, then reactions were quenched for 5 minutes using glycine at room temperature. Cells were lysed for 30 minutes with ChIP lysis buffer (78) on ice. Chromatin fragmentation was performed using a Diagenode BioruptorPico sonicator (45 cycles, each with 30 seconds on and 30 seconds off). Solubilized chromatin was then incubated with a mixture of Tfcp2 antibody (Cell Signaling Technology, 80784S) in 1:100 dilution and Dynabeads (Life Technologies) 4°C overnight. Immune complexes were then washed three times with RIPA buffer, then once with RIPA-500, and once with LiCl wash buffer. Elution and reverse crosslinking were performed overnight at 65°C in a direct elution buffer containing proteinase K (20 mg/mL). Eluted DNA was purified using AMPure beads (Beckman Coulter) and was then used to perform qPCR. Three Bmp4 and Wnt5b primer pairs each were used (listed in Supplementary Table S4).

For luciferase reporter assay, the promoter region of human BMP4 and WNT5B (−1,500 to +500 bp) was amplified by PCR and inserted into the pGL3 vector (Addgene: E1751) to generate the corresponding reporter constructs. The luciferase reporter assay was conducted by transfecting the reporter constructs and Renilla luciferase vector into iKAP TFCP2 knockout and rescued cell lines. Cells were harvested after 24 hours of transfection and luciferase activity was measured.

PGE2 analysis was performed using the Prostaglandin E2 ELISA Kit following the manufacturer's protocol (Abcam, ab133021). Briefly tumors were homogenized and centrifuged to collect the supernatant to measure the amount of PGE2. In addition, CM from iKAP-doxycycline on and -doxycycline off CM-cultured 3T3L1 cells were collected to measure PGE2 levels.

An adipokine array was performed using the Proteome Profiler Mouse Adipokine Array Kit following the manufacturer's protocol (R&D Systems; ARY013). Briefly, cell lysates were diluted, mixed with a cocktail of biotinylated detection antibodies, and incubated with a Proteome Profiler Mouse Adipokine Array. Control antibodies from the kit were used as the reference spots on the top left corner of the arrays.

HUVEC tube formation assay was performed using CM from lipid-rich CAFs. CM from DLD1-KRAS^G12D CM-educated hMSCs was cultured with HUVECs on Matrigel (Corning, 47743–722), and a web-based Wimasis Image Analysis tool (https://www.wimasis.com/en/) was used to analyze tube formation, including the covered area (%), total tube length (px), total branching points, and total loops.

Cell pellets from iKAP cell line, sorted lipid-rich fibroblasts, CM-cultured 3T3L1 and hMSCs, and FQI1 treated iKAP cell lines were collected, and isolated mRNA was recovered using an RNeasy Mini Kit (Qiagen). mRNA was reverse-transcribed into copy DNA using a OneScript Plus cDNA Synthesis Kit (ABM; G236). SYBR Green PCR Master Mix (Thermo Fisher Scientific) was used to perform RT–qPCR in a 7500 Fast Real-Time PCR Machine (Applied Biosystems). The genes were normalized using actin and/or Ywhaz, an adipocyte internal control gene (79). qRT-PCR primers are listed in Supplementary Table S4.

Before performing RNA sequencing, iKAP and iAP tumors were assessed for KRAS* expression by IHC, as well as by IF and IHC of GFP, which is expressed by enema OHT/Cre-induced reporter activation. The tumor stages were verified by a gastrointestinal pathologist, blinded to the sample identities. The tissues for RNA isolation were collected from microdissected localized adenomas and invasive adenocarcinomas (T4).

The RNA sequencing of colorectal cancer GEM tumors was performed as described in Boutin and colleagues (7), and the data were deposited in the National Center for Biotechnology Information under SRA number SRP097890. Briefly, cells were first lysed with Buffer RLT and then purified with TRIzol Reagent (Life Technologies) and chloroform. The remaining steps of the RNeasy Mini Kit were then followed. RNA-sequencing libraries were made, and samples were run on an Illumina HiSeq 2000 at The University of Texas MD Anderson Cancer Center Sequencing and Microarray Core Facility. Transcriptome reads were mapped to the reference mouse genome mm10, normalized, and quantified as fragments per kilobase per million mapped fragments values using HTSeq–count. A differential gene expression analysis was performed on the basis of the results of the negative binomial distribution. Genes were compared for iAP versus iKAP; iKAP versus iKAP doxycycline off; high vs. low lipid-rich CAF scores; and high vs. low Tfcp2 scores). The lipid-rich CAF gene signatures are listed in Supplementary Tables S1 and S2. The TFCP2- and SRY-regulated gene signatures were obtained from GSEA datasets (C3 TFT). The results from the comparisons were subjected to GSEA and heat map analyses.

Three independent samples each of invasive iKAP, noninvasive iAP, and noninvasive iKAP–doxycycline off (KRAS*-off) colorectal cancer tumors were collected from microdissected localized adenomas and invasive adenocarcinomas. These tumors were freshy harvested for tumor dissociation and single-cell sequencing analyses. Tumors were collected 2 months after tamoxifen/doxycycline tumor induction and approximately 1 cm³. For the iKAP-doxycycline off (KRAS*-off) colorectal cancer tumors, doxycycline was discontinued for 7 days following 7 weeks of tamoxifen/doxycycline tumor induction. scRNA-seq was performed by MD Anderson's CPRIT Single Core. Single-cell capture, barcoding, and library preparation were performed by following the Single–Cell Chromium 3′ protocol (10X Genomics; PN–120237) using V3 chemistry reagents (10X Genomics). The final libraries containing barcoded single-cell transcriptomes were sequenced at 100 cycles on an S2 flowcell on the Novoseq 6000 System (Illumina). Data were processed using the CASAVA 1.8.1 pipeline (Illumina), and sequence reads were converted to FASTQ files and UMI read counts using the CellRanger software (10X Genomics). For in silico analyses, raw count matrices were converted to Seurat objects for further analysis. The quality of scRNA-seq data was measured by the percentage of mitochondrial genes (percent.mt) per cell and the number of unique molecular identifiers (UMI count). Low-quality cells with a UMI count less than 800, a percent.mt greater than 25%, and doublets expressing both cell-type markers were removed. The batch effect was counteracted using the reciprocal PCA-based integration method in Seurat. We then performed unsupervised dimension reduction and clustering for cell identification. The annotation of cell identities in each cluster was based on both the canonical marker genes and the cluster differential genes. For module score plots, we perform Seurat's AddModuleScore using adipocyte gene signature (43), and the combination of classic adipocyte and preadipocyte genes (mature adipocyte markers: Fabp4, Lpl, Plin4, Dcn, Dpt, Cfd, Ccl2; mature adipocyte regulatory TFs: Pparg, Cebpa, Cebpb; preadipocyte markers: Dlk1, Dlk2, Lpl, Plin4; preadipocyte regulatory TFs: Zfp423, Tcf21; refs. 29–31, 44, 45). All GEM colorectal cancer scRNA-seq data were deposited in the National Center for Biotechnology Information (GSE229559); R code and R processed data were deposited in Code Ocean https://codeocean.com/capsule/9653675/tree/v1.

Human research was conducted in accordance with the Helsinki Declaration ethical guidelines. Human tumor specimens were collected at The University of Texas MD Anderson Cancer Center under Institutional Review Board approved protocol LAB10–0982 (principal investigator: S. Kopetz) after obtaining written informed consent from each participant. Fresh specimens were shortly transported to MD Anderson's CPRIT Single Core in sterile cold DMEM for single-cell isolation, scRNA-seq, and analyses as described above. Clinical information, including tumor location, gender, age, stage at collection, MS status and genotype are listed in Supplementary Table S5. The raw scRNA-seq and processed data generated from human colorectal cancer specimens are available on Gene Expression Omnibus under accession number GSE231559.

For human colorectal cancer data analyses, gene expression and survival data were obtained from TCGA-COAD datasets and/or datasets from cBioPortal (https://www.cbioportal.org/). We collected data including tumor genotypes, patient survival, tumor stages, tumor subtypes, and patient treatments in cBioPortal. Using R packages (DESeq2, gene-set variation analysis, and pathfindR), we analyzed the enriched pathways in KRAS* patients, the KRAS-related genes, lipid-rich CAF scores, TFCP2 scores, and their roles in patient survival. Briefly, for pathway analyses, we obtain differential expression genes in KRAS* patients by DESeq2 and performed GSEA analyses or used pathfindR to obtain enriched pathways. For lipid-rich CAF scores and TFCP2 scores, we performed gene-set variation analysis using lipid-rich CAF gene signature (Supplementary Tables S1 and S2) and TFCP2-regulated gene signature (GSEA datasets, C3 TFT) from KRAS* and KRAS^WT patients.

The candidates of KRAS* downstream TFs were analyzed by utilizing the consensus TF-binding motifs present in the promoters of BMP4 and WNT5B (PROMO software), along with the TF signatures modulated in a KRAS*-dependent manner in both murine and human colorectal cancers (GSEA analyses). The candidate lists from these three gene sets are listed in Supplementary Table S3.

All two-sample statistical analyses were performed using Student t test, two sided, and were represented as the mean ± SD. The significance analysis of the TCGA colon adenocarcinoma database, GEM RNA sequencing, scRNA-seq, RT–qPCR, IHC quantification, and angiogenesis assay quantification were all performed in EXCEL and GraphPad Prism 9. Analyses of survival data from the TCGA colon adenocarcinoma database and GEM mice were performed using the log-rank (Mantel–Cox) test in GraphPad Prism 9. P values are included in the graphs. P value <0.05 was considered statistically significant.

Additional images for all tissue staining are provided in Supplementary Fig. S7A–S7T; the RNA sequencing data of colorectal cancer GEM tumors were deposited in the National Center for Biotechnology Information under SRA number SRP097890; the GEM colorectal cancer scRNA-seq data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE229559; R code and R processed data were deposited in Code Ocean at https://codeocean.com/capsule/9653675/tree/v1; the human colorectal cancer scRNA-seq data generated in this study are publicly available in Gene Expression Omnibus under accession number GSE231559.

J.P. Shen reports grants from Celsius Therapeutics, personal fees from Engine Biosciences, personal fees from NaDeNo Nanoscience, and grants from BostonGene outside the submitted work. S. Kopetz reports other support from Genentech, other support from EMD Serono, other support from Merck, other support from Holy Stone Healthcare, other support from Novartis, other support from Lilly, other support from Boehringer Ingelheim, other support from AstraZeneca/MedImmune, other support from Bayer Health, other support from Redx Pharma, other support from Ipsen, other support from HalioDx, other support from Lutris, other support from Jacobio, other support from Pfizer, other support from Repare Therapeutics, other support from Inivata, other support from GlaxoSmithKline, other support from Jazz Pharmaceuticals, other support from Iylon, other support from Xilis, other support from AbbVie, other support from Amal Therapeutics, other support from Gilead Sciences, other support from Mirati Therapeutics, other support from Flame Biosciences, other support from Servier, other support from Carina Biotech, other support from Bicara Therapeutics, other support from Endeavor BioMedicines, other support from Numab, other support from Johnson & Johnson/Janssen, other support from Genomic Health, other support from Frontier Medicines, other support from Replimune, other support from Taiho Pharmaceutical, other support from Cardiff Oncology, other support from Ono Pharmaceutical, other support from Bristol Myers Squibb-Medarex, other support from Amgen, other support from Tempus, other support from Foundation Medicine, other support from Harbinger Oncology, other support from Takeda, other support from CureTeq, other support from Zentalis, other support from Black Stone Therapeutics, other support from NeoGenomics Laboratories, other support from Accademia Nazionale Di Medicina, other support from Tachyon Therapeutics, other support from Sanofi, other support from Biocartis, other support from Guardant Health, other support from Array BioPharma, other support from Genentech/Roche, other support from EMD Serono, other support from MedImmune, other support from Novartis, other support from Amgen other support from Daiichi Sankyo, other support from Lutris, other support from Iylon, other support from Frontier Medicines, and other support from Navire outside the submitted work. R.A. DePinho reports grants from National Institutes of Health during the conduct of the study; other support from Tvardi, other support from Nirogy, other support from Asylia, other support from Stellanova, and other support from Sporos Bioventures outside the submitted work. No disclosures were reported by the other authors.

W.-H. Hsu: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft. K.A. LaBella: Conceptualization, data curation, software, formal analysis, validation, investigation, methodology. Y. Lin: Conceptualization, data curation, software, formal analysis, investigation, methodology. P. Xu: Investigation. R. Lee: Investigation. C.-E. Hsieh: Investigation. L. Yang: Investigation. A. Zhou: Investigation. J.M. Blecher: Investigation. C.-J. Wu: Software, investigation, methodology. K. Lin: Software, investigation, methodology. X. Shang: Investigation. S. Jiang: Investigation. D.J. Spring: Funding acquisition, project administration, writing–review and editing. Y. Xia: Conceptualization. P. Chen: Conceptualization. J.P. Shen: Conceptualization, resources. S. Kopetz: Conceptualization, resources. R.A. DePinho: Conceptualization, data curation, supervision, funding acquisition, validation, methodology, project administration, writing–review and editing.

The authors thank Dr. Douglas Hanahan and Dr. Krisztian Homicsko (the Swiss Federal Institute of Technology Lausanne) for invaluable feedback on their manuscript and for providing anti-mouse VEGFA mAb (B20); Dr. Bert Vogelstein (Johns Hopkins) and Dr. Jihye Yun (Baylor College of Medicine) for providing KRAS wild-type DLD1 isogenic colorectal cancer cell lines; Dr. Nicholas Navin (The University of Texas MD Anderson Cancer Center) for performing single-cell sequencing and analysis; Dr. Guillermina Lozano, Dr. Filippo G. Giancotti, and Dr. Traver Hart (MD Anderson Cancer Center) for providing manuscript and data analysis consultations; Dr. Kuo-Shun Hsu and Vincent Li (MD Anderson's TRACTION) for PDXO establishment and maintenance; and Laura L. Russell, scientific editor (Research Medical Library at MD Anderson Cancer Center) for editing this article. This work was supported by MD Anderson Cancer Center's Cancer Prevention & Research Institute of Texas (CPRIT) Single Cell Genomics User Group grant (RP180684), a CPRIT Research Training Program grant (RP210028, to W.-H. Hsu and K.A. LaBella), a NIH F99/K00 grant (CA274661–01, to W.-H. Hsu), MD Anderson Cancer Center UTHealth Graduate School of Biomedical Sciences’ Pauline Altman–Goldstein Foundation Discovery Fellowship (to W.-H. Hsu), an NIH R01 grant (CA231360, to R.A. DePinho), National Center for Advancing Translational Sciences awards (TL1TR003169 and UL1TR003167, to K.A. LaBella), the American Legion Auxiliary Fellowship in Cancer Research (to K.A. LaBella), and the Dr. John J. Kopchick Fellowship (to K.A. LaBella).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.