Ketogenesis Attenuates KLF5-Dependent Production of CXCL12 to Overcome the Immunosuppressive Tumor Microenvironment in Colorectal Cancer

Despite significant improvements in the treatment of colorectal cancer, the prognosis for patients with colorectal cancer with metastasis remains dismal with a median overall survival (OS) of approximately 30 months (1). Immunotherapy such as immune checkpoint blockade (ICB) represents a novel therapeutic approach for a variety of cancers, including colorectal cancer with microsatellite instability-high (2). However, ICB therapy shows little or no clinical activity in the majority of patients with microsatellite-stable colorectal cancer (2). The tumor microenvironment (TME) has a dynamic composition that affects the efficacy of targeted therapies for colorectal cancer (3). Critical interactions between cancer cells occurring in the TME mostly promote tumor initiation, resistance, metastasis, and recurrence (3). Therefore, to achieve an increase in survival, efficient strategies are needed for disruption of these interactions in the TME (4).

Accumulation of cancer-associated fibroblasts (CAF) in TME is associated with poor prognosis and recurrence of colorectal cancer (5). CAFs within the TME have shown to be an important determinant of the tumor immune response (6). The chemokine CXCL12, also known as stromal cell–derived factor-1 (SDF1), is expressed in the TME mainly by CAFs (7). CXCL12 protein, which binds to G protein–coupled receptors CXCR4 and CXCR7, promotes tumor immunosuppression by recruiting specific immune cell populations (8). Activation of the CXCL12/CXCR4 axis is common in many solid tumors and contributes to tumor immunosuppression and metastasis (9). The factors controlling the level of CXCL12 in TME remain largely undefined.

Ketogenic diets (KD), which have high-fat and low-carbohydrate content, have been demonstrated to be a safe and achievable component in the treatment regimen of some cancers (10). Recent studies showed that a KD inhibits growth of certain tumor xenografts in vivo and enhances anticancer effects of ICB in murine colorectal cancer, melanoma, renal, and non–small cell lung cancer (NSCLC) orthotopic tumor models (11, 12). However, the dynamics of ketogenesis and the role of ketogenesis on the immunosuppressive TME are not known.

Our recent studies demonstrate that an increase in ketogenesis induces differentiation and inhibits colorectal cancer cell glycolysis (13, 14). In addition, we showed that inhibition of WNT/β-catenin or mTOR signaling increases the expression of the ketogenic enzyme 3-Hydroxy-3-Methylglutaryl-CoA Synthase 2 (HMGCS2) and the production of the ketone body beta-hydroxybutyrate (βHB) in colorectal cancer cells (13, 14). Here, we show that decreased ketogenesis is a signature of colorectal cancer and that an increase in ketogenesis represses KLF5-dependent CXCL12 expression and thus inhibits immunosuppressive TME, enhances the anticancer effects of PD-1 blockade, and inhibits lung metastasis. Importantly, our study demonstrates that downregulation of de novo ketogenesis in the TME is a critical step in colorectal cancer progression and suggest a potential benefit of using KD to combat colorectal cancer immunoresistance and metastasis.

C57BL/6J and BALB/cJ mice (female, 6-week-old) were obtained from The Jackson Laboratory. All animal procedures were conducted with approval and in compliance with University of Kentucky Institutional Animal Care and Use Committee. Mice were housed in a pathogen-free environment on a 12-hour light/dark cycle and received a normal chow diet or KD (F3666, Bio-serv) with free access to drinking water.

MC38 or CT26 tumors were established by subcutaneously injecting cells (0.5 × 10⁶ cells in 50% Matrigel in PBS, 100 μL/mouse) into the right flank of either C57BL/6J or BALB/cJ female mice. Tumor sizes were measured every 2–3 days by caliper after implantation and tumor volume was calculated in mm³ [mm³ = π/6 × (larger diameter) × (smaller diameter)²]. Four days after tumor cell implantation (mean tumor volume reached approximately 100 mm³), animals were pooled and randomly divided into designated experimental groups with comparable average tumor size and body weight. For ketogenic/anti–PD-1 combination treatment, mice were grouped into IgG control [normal diet (NC) + IgG isotype control (BE0089, Bio-XCell)], ketogenic treatment (KD + IgG isotype control), anti–PD-1 antibody treatment [NC + anti–PD-1 monoclonal antibody (BE0146, Bio-XCell)], and anti–PD-1 antibody plus ketogenic treatment (KD + anti–PD-1). Anti–PD-1 antibody and IgG control treatments were conducted by intraperitoneal injection (200 μg per mouse in 100 μL PBS) every two days from day 5 for a total of 7 injections. For survival studies, C57BL/6J or BALB/cJ mice were monitored for tumor volume every 1–3 days after initial treatment until tumors became necrotic or their volume reached 2,000 mm³.

For the lung metastasis model, GFP-CT26 cells (1 × 10⁶ cells per BALB/cJ mouse) were injected via tail vein as we have described previously (15). Two weeks after cancer cell injection the mice were euthanized, and lungs were removed to examine GFP fluorescence of metastatic tumors using a Lago (Spectral Instruments Imaging) equipped with an excitation source (465 nm) and emission filter (510 nm). GFP signal was quantified in Aura software (Spectral Instruments Imaging).

Single-cell suspensions were prepared following the recommendations from 10X Genomics and CD45⁺tumor-infiltrating leukocytes were enriched for single-cell analysis as described previously (16). Dead cells were removed from the single-cell suspension using the Dead Cell Removal Kit (Miltenyi Biotec). CD45⁺ leukocytes were magnetically labeled and enriched using magnetic beads (Miltenyi Biotec). Cells were stained with Trypan Blue and counted using Cuntess II FL (Invitrogen). Tumors from five different mice were pooled per diet condition. Ten thousand cells were targeted for sequencing and loaded onto 10X Genomics Chromium Controller for Gel Beads-in-emulsion formation. Library was prepared was conducted according to the manufacturer's instructions. Single indexed, paired-end libraries were sequenced on an Illumina HiSeq2500 and data analyzed using Cell Ranger software (10X Genomics).

Collect 2 × 10⁶ or fewer cells per tumor from single-cell suspension and resuspend pellet in 50 μL PBS containing Fc blocking antibody (anti-mouse CD16/32, BioLegend) and fixable live/dead stain (Zombie UV Fixable Viability Kit, BioLegend), incubate sample at room temperature for 15 minutes. Following Fc blocking, add surface antibody (see Supplementary Table S1) mix and incubate at 4°C for 60 minutes. Cells were washed twice with FACS buffer (BioLegend) and resuspended in 200 μL buffer. For cell-cycle analysis, cells were fixed with 70% ice-cold ethanol and resuspended in FxCycle Propidium iodide/RNase staining solution (Thermo Fisher Scientific). Flow cytometry was performed using standard protocol on CytoFLEX LX analyzer (Beckman Coulter) and analyzed with FlowJo software.

Primary human colorectal cancer CAFs (hCAF-05; Neuromics, Cat# CAF05), colorectal cancer CAFs (hCAF-6231; Cell Biologics, Cat# HC-6231) and mouse CAFs (mCAF), isolated from subcutaneous tumors using a Tumor-associated Fibroblast Isolation Kit (Miltenyi Biotec) and identified by means of morphology and immunofluorescence staining for S100A4 using anti-S100A4 antibody (Abcam), were cultured in VitroPlus III, low serum complete medium (Neuromics). Human colorectal cancer cell line HCT116 and mouse colorectal cancer cell line CT26 were obtained from the ATCC. HCT116 cells were maintained in McCoy's 5a supplemented with 10% FCS. The mouse colorectal cancer cell line MC38, purchased from Kerafast, was maintained in DMEM supplemented with 10% FCS, 2 mmol/L glutamine, 0.1 mmol/L nonessential amino acids, 1 mmol/L sodium pyruvate, and 10 mmol/L Hepes. All cell lines were tested for Mycoplasma contamination using a sensitive PCR-based Mycoplasma detection kit (Biovision) and were found to be negative. The mouse colorectal cancer cell line CT26 was tested via short tandem repeat profiling for authentication in 2021 (LabCorp.). Cells were infected with adenovirus vectors encoding either GFP (Ad-GFP; Vector BioLabs) or human KLF5 (Ad-KLF5, SignaGen), or treated with βHB (Sigma) or ML-264 (MedChemExpress). Cells were transfected with either human KLF5, HDAC1 (histone deacetylase), or nontargeting control siRNA SMARTpool (Dharmacon) by electroporation (Gene Pulser, Bio-Rad). The GFP-CT26 cell line was established after transfection with GFP lentivirus as described previously (15). GFP-expressing CT26 cells were selected with 5 μg/mL of puromycin and enriched by two cycles of FACS.

To establish cell lines with inducible overexpression of HMGCS2, human HMGCS2 cDNA (RC208128, OriGene Technologies) was sub-cloned into the pCW57-GFP-2A-MCS lentiviral vector (Addgene). HMGCS2-C166A mutant, an enzymatic dead HMGCS2 (17), was generated from wild-type HMGCS2 by site-directed mutagenesis using the QuikChange II XL Site-Directed Mutagenesis Kit (Agilent) with the primers as described previously (Supplementary Table S2; ref. 17). HMGCS2-C166A mutant was sequenced, and mutation was confirmed. hCAF-05 or hCAF-6231 cells were infected with lentiviral particles containing empty control vector or construct encoding human HMGCS2 or MGCS2-C166A mutant, selected with puromycin (5 μg/mL), and induced by doxycycline (400 ng/mL) for 48 hours as previously described (18).

Cells (5 × 10⁴/well) were seeded in 6-well plates in triplicate. Cell proliferation was analyzed by counting the number of viable cells in response to the treatment of βHB, HMGCS2 overexpression, and knockdown or overexpression of KLF5 as described previously (19). Each experiment was repeated at least three times.

Total protein was resolved by SDS-PAGE and transferred to polyvinylidene difluoride membranes, immunoblotted with specific primary and secondary antibodies (see Supplementary Table S1) as we have described previously (13). αSMA signals from 10 paired human tissues was quantitated densitometrically and expressed as fold change to β-actin.

Total RNA was extracted using the RNeasy Mini Kit (QIAGEN) and DNase-treated. Synthesis of cDNA was performed a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). The TaqMan probe and primers for human CXCL12, KLF5, and GAPDH were purchased from Thermo Fisher Scientific. Quantitative real-time RT-PCR analysis was performed using TaqMan Universal PCR Master Mix (Thermo Fisher Scientific) as described previously (13).

Frozen-fresh human tissue samples were collected from patients who previously had resections at UK Markey Cancer Center with informed consent and Institutional Review Board approval (IRB#13–0753-P2H). Paraffin-embedded tissues were obtained from Markey's tissue repository. The frozen samples were “snap-frozen” in liquid nitrogen and stored at −80°C. The frozen-fresh human tissue samples included 13 paired primary colorectal tumor and adjacent normal colon tissues. The paraffin-embedded tissues included 5 paired primary colorectal tumor, adjacent normal colon, and liver metastases.

IHC staining was performed as described previously (19). Tissue was processed for routine IHC staining using the following antibodies: Anti–CXCR7/RDC-1 and anti-CXCR4 (R&D Systems), anti-SDF1 and anti-HMGCS2 (Abcam), and anti-BTEB2 (KLF5). Negative controls (including no primary antibody or isotype-matched mouse immunoglobulin G) were used in each assessment.

Intracellular and plasma βHB concentration was determined using a Beta-Hydroxybutyrate Assay Kit (Sigma Aldrich) according to the manufacturer's protocol. Each plotted value was normalized to total amount of protein used from cell lines and tissues or volume from plasma.

SDF1 concentrations of cell lysates, cell medium, plasma or tissues were analyzed using SDF1 alpha/CXCL12A Human/Mouse ELISA Kits (Thermo Fisher Scientific). All the assays were performed according to the protocols provided by the manufacturer. Each plotted value was normalized to total amount of protein used from cell lysates and tissues or volumes from media and plasma.

hCAF-05 with empty vector or construct expressing HMGCS2 were seeded and treated with doxycycline (400 ng/mL) for 48 hours to induce HMGCS2 expression before lysis. Chromatin immunoprecipitation (ChIP) assay was performed using the ChIP-IT High-Sensitivity kit (Active motif) in accordance with the manufacturer's protocol as we have described previously (19). The human CXCL12 promoter region containing the KLF5-binding site was amplified by PCR from total (input) or immunoprecipitated chromatin using the primers as described previously in Supplementary Table S2.

Genomic DNA from hCAF-05 cells was extracted. CXCL12 gene promoter fragments 1,070 bp (from −970 bp to +100 bp from the transcription start site), 944 bp (−844 bp to +100 bp), and 767 bp (−667 to +100 bp) were produced by PCR using primers containing NheI and HindIII restriction sites as described previously in Supplementary Table S2. The primers were synthesized on the basis of the genomic sequence lying upstream of the human CXCL12 cDNA (NCBI, NC_000010.11). PCR products were cloned into the reporter vector (pGL3-basic), sequenced, and confirmed to contain contiguous genomic sequences. The pRL-Tk-luc plasmid was cotransfected together with CXCL12 promoter constructs to normalize for variation in transfection efficiency. Luciferase assays were performed as previously described (18).

Migration assays were performed in chambers coated with collagen I (Thermo Fisher Scientific). Invasion assays were performed in chambers coated with Matrigel, as instructed by the manufacturer (BD Biosciences). Briefly, HCT116 cells were incubated in serum-free medium for 24 hours. Cells in serum-free medium were seeded 5 × 10⁴/well into the upper chamber of the 24-well Transwell plate (Corning, Inc.). 600 μL conditioned medium (CM) was introduced to the lower wells and cultured for 6 hours. The number of migrated cells were counted in five areas at ×200 magnifications using an inverted microscope as described previously (20). A similar procedure was carried out to detect the invasion ability of HCT116 cells except for using the chambers being coated with Matrigel and cultured for 24 hours.

For in vitro experiments and in vivo studies, pairwise comparisons between 2 groups were performed using two-sample t test or analysis of variance for multiple groups with contrast statements. Adjustments for multiple testing between groups within an experiment were performed using the Holm P_adjustment method. Kaplan–Meier curves were estimated for survival of in vivo studies and compared using the log-rank test. All data from animal samples with measurement of study endpoints were included in the analysis. Raw sequence reads from single-cell RNA sequencing (scRNA-seq) data were processed and aligned to the GRCm38 (2020-A) mouse reference transcriptome using the Cell Ranger v6.0.0 pipeline (10X Genomics) with default parameters. The gene expression matrices were processed using scDblFinder (v1.1.8) R package for doublets prediction and removal.

All materials generated in this study are available from the corresponding author, Q. Wang (qingding.wang@uky.edu). The scRNA-seq data reported in this study are available from Gene Expression Omnibus (GEO) with accession code GSE180110. All other data are available from the corresponding author (Q. Wang) upon reasonable request.

We have previously shown that an increase in ketogenesis contributes to colorectal cancer cell differentiation (13). To determine whether an increase in ketogenesis affects the proliferation of colorectal cancer cells, we treated human colorectal cancer cell lines HCT116, DLD1, LS174T, and RKO and mouse colon cancer cell lines MC38 and CT26 with βHB. Treatment with βHB (5 or 10 mmol/L) for 3 days had minor effects on the proliferation of these cells (Supplementary Fig. S1A and S1B). To further determine the role of ketogenesis in colorectal cancers, we next analyzed the antitumor effects of a KD using MC38 xenograft models in immune competent C57BL/6J mice (Fig. 1A and B) and athymic nude mice (Supplementary Fig. S1C). As shown in Fig. 1B, MC38 tumor grew slower in C57BL/6J mice fed KD compared with mice fed NC. In contrast, KD showed minor effects on MC38 tumor growth in athymic nude mice (Supplementary Fig. S1C), suggesting that the reduction in tumor growth by KD may be due to the altered immunosuppressive TME. To test this hypothesis, we used flow cytometry to profile tumor-infiltrating T-cell and macrophage populations in MC38 tumors in immune competent C57BL/6J mice. We found that feeding tumor-bearing mice with a KD changes the immune landscape of MC38 tumors. Tumor tissues from mice fed a KD contained more T cells as a fraction of the CD45⁺ leukocyte infiltrate (Fig. 1C). Specifically, increased infiltration of CD8α⁺ T cells were noted in KD tumors (Fig. 1D and E). In contrast, the percentage of total macrophages, especially immunosuppressive M2 macrophages, in the CD45⁺ leukocyte infiltrate decreased with KD (Fig. 1F and G). Our results suggest a potential role for ketogenesis in the inhibition of the immunosuppressive TME in colorectal cancers.

To further determine KD-induced alterations in tumor-infiltrating immune populations, we used scRNA-seq of tumor-infiltrating CD45⁺ leukocytes to map the tumor immune transcriptional landscape in an unbiased and comprehensive manner (Fig. 1H–K). Mice with MC38 tumors were fed with NC or KD as described previously (Fig. 1A) and CD45⁺ leukocytes in tumors were magnetically labeled and enriched by positive selection using CD45 MicroBeads. To define major cell populations, we performed unsupervised clustering analysis on integrated single-cell datasets from NC and KD tumors, which identified 11 distinct clusters (Fig. 1H), where each of the 11 clusters contained cells from both diet conditions. We annotated clusters based on the expression of known genetic markers and categorized those cell populations into groups (see more details in Supplementary Fig. S2A–S2C and Supplementary Methods; refs. 16, 21). In agreement with the findings from flow cytometry analysis, increased CD8⁺ T-cell accumulation was found in tumors from mice fed a KD (Fig. 1I). Conversely, a marked decrease in total macrophage populations were found in KD tumors (16.8%) compared with NC tumors (30%). Moreover, significantly increased accumulation of natural killer (NK) cells was found in KD tumors (32.2%) compared with NC tumors (19.7%; Fig. 1I). In addition, increased CD4⁺ T-cell populations were found in tumors from mice fed a KD. Further analysis demonstrated the increased percentage of antitumor conventional CD4⁺ T-cell population and decreased percentage of CD4⁺ Treg (regulatory T cells) population, which mediate tumor immunosuppressive functions (22), in total CD4⁺ T-cell populations of KD tumors (Fig. 1J). Tumor-infiltrating CD8⁺ T cells and NK cells perform as effector cells against tumor cells (23). These data suggest an enhanced antitumor immune response in KD tumors. Moreover, reduced immunosuppressive populations were found in KD tumors as noted by the markedly decreased myeloid-derived suppressor cells (MDSC; Fig. 1I) and M2 macrophage populations and M2/M1 ratio in KD tumors (Fig. 1K). A higher ratio of M2/M1 macrophages in the TME is clinically associated with a poorer prognosis in colorectal cancer (24). Together, these results demonstrate that KD induces the reversal of an immunosuppressive TME.

CAFs, a key constituent of the TME, promote immunosuppression in TME via secretion of chemokines such as CXCL12 (25). To determine how an increase in ketogenesis inhibits an immunosuppressive TME, we have established human primary CAFs, hCAF-05, and hCAF-6231 with inducible overexpression of HMGCS2. HMGCS2 expression in hCAF-05 cells was induced by treatment with doxycycline and RNA-seq analysis was performed. Overexpression of HMGCS2 decreased CXCL12 mRNA expression and increased βHB production in hCAF-05 without affecting the expression of CAF biomarkers ACTA2, S100A4, and MMP2 (Supplementary Fig. S3A–S3C; ref. 26). Consistent with the results from RNA-seq analysis, overexpression of HMGCS2 in hCAF-05 and hCAF-6231 cells (Supplementary Fig. S3D) resulted in a marked decrease in the expression of CXCL12 mRNA (Fig. 2A; left), CXCL12 protein (SDF1) in cell lysates (Fig. 2B) and secretion in culture medium (Fig. 2C), as determined by RT-PCR and ELISA, respectively. In contrast, overexpression of HMGCS2-C166A (HMGCS2-Mut), an enzymatic dead HMGCS2 (17), did not affect either the expression of CXCL12 mRNA (Fig. 2A; right) and protein or the levels of βHB in cell lysates (Supplementary Fig. S3E). Moreover, βHB treatment significantly inhibited CXCL12 mRNA (Fig. 2D) and protein (Fig. 2E) expression in human primary CAFs (hCAF-05 and hCAF-6231) and mouse primary CAFs (mCAFs; isolated from CT26 tumors). Consistently, reduced CXCL12 mRNA expression was found in mouse primary CAFs isolated from KD MC38 tumors (Fig. 2F). In contrast, treatment with βHB did not affect CXCL12 mRNA levels in human normal colon fibroblasts CCD-18Co (Supplementary Fig. S3F), suggesting a specific inhibition of CXCL12 expression in CAFs. As CXCL12, expressed in the TME mainly by CAFs (Supplementary Fig. S4; ref. 7), plays a critical role in the immunosupressive TME (4), our results demonstrate that increased ketogenesis overcomes the immunosupressive TME through the inhibition of CXCL12 expression.

Our RNA-seq analysis showed decreased KLF5 mRNA levels in CAFs with overexpression of HMGCS2 (Supplementary Fig. S3A). To further determine the inhibition of KLF5 expression by HMGCS2, HMGCS2 expression in hCAF-05-HMGCS2 cells was induced by treatment with doxycycline. As shown in Fig. 3A and B, overexpression of HMGCS2 significantly repressed KLF5 mRNA expression. Moreover, HMGCS2, but not HMGCS2-Mut, markedly repressed KLF5 protein expression without affecting KLF4 protein expression in hCAF-05 cells. In agreement with HMGCS2 overexpression, treatment of hCAF-05, hCAF-6231, and mCAFs with βHB significantly repressed the protein expression of KLF5, but not KLF4, in these CAFs (Fig. 3C). Recently, βHB has been shown to act as an endogenous inhibitor of class I HDACs (27). βHB inhibited HDAC activity in CAFs as noted by the increased acetylation of histone H3 lysine 9 (H3K9ac). Moreover, treatment with sodium butyrate (NaBT), which inhibits class I HDACs (27), or MS-275, a selective HDAC1 inhibitor (28), or knockdown of HDAC1, repressed the protein expression of KLF5, but not KLF4, in hCAF-05 cells (Fig. 3C). These data demonstrate that an increase in ketogenesis inhibits KLF5 expression through the inhibition of HDAC1 in CAFs.

Next, we determined whether KLF5 regulates CXCL12 expression. Overexpression of KLF5 by infection with Ad-KLF5 significantly increased the expression of CXCL12 mRNA and protein compared with Ad-GFP control in hCAF-05 cells (Fig. 3D and E) and hCAF-6231 cells (Supplementary Fig. S5A–S5C). Moreover, βHB reduced CXCL12 mRNA and protein expression in control GFP hCAF-05 cells but not in KLF5 hCAF-05 cells (Fig. 3D; Supplementary Fig. S5D). In contrast, knockdown of KLF5 by transfection with siRNA targeting KLF5 (Fig. 3F) or inhibition of KLF5 by treatment with ML-264 (Fig. 3G), a selective KLF5 inhibitor (29), decreased CXCL12 mRNA expression in hCAF-05 cells. Moreover, knockdown of KLF5 repressed CXCL12 protein (SDF1) expression and secretion (Fig. 3H). These results demonstrate that an increase in ketogenesis represses CXCL12 expression through inhibition of KLF5 expression in CAFs.

KLF5 binds to GC-box sequence near the core promoter elements and amplifies transcription of its target genes (30). An essential binding site for KLF5 is 5′-GGGGTGGG-3′ (30). We identified two KLF5-binding sites (motif 1 and motif 2) in the CXCL12 promoter (Fig. 3I). To determine whether KLF5 binds to CXCL12 promoter, we performed a ChIP assay. Crosslinked chromatin was prepared from hCAF-05 cells and immunoprecipitation was performed using either the anti-KLF5 antibody or IgG; the sequences containing no KLF5-binding site (negative control; top row), the putative KLF5-binding site 1 (motif 1; middle row), and binding site 2 (motif 2; bottom row) were amplified (Fig. 3J). We found that KLF5 mainly binds to the CXCL12 promoter region containing KLF5-binding motif 1; this binding was markedly attenuated by HMGCS2 overexpression. To determine whether KLF5 regulates CXCL12 expression through its binding to this KLF5 motif, we next cloned the CXCL12 promoter (Fig. 3K). hCAF-05 cells were transfected with CXCL12 promoter deletion constructs together with control vector or a plasmid encoding KLF5 (Fig. 3L). Overexpression of KLF5 induced CXCL12 promoter activity, with deletion of KLF5-binding site 1 resulting in a marked decrease in KLF5-induced CXCL12 promoter activation. Deletion of the two KLF5-binding sites eliminated KLF5 induction. Together, these results demonstrate that KLF5 regulates CXCL12 expression through promoter binding and activation.

To determine whether an increase in ketogenesis affects CAF proliferation, hCAF-05, hCAF-6231 or mCAFs cells were treated with βHB for 6 days. As shown in Fig. 4A–C, βHB markedly inhibited the growth of these CAFs. Moreover, overexpression of HMGCS2 significantly repressed the proliferation of hCAF-05 cells (Fig. 4D). In agreement with the inhibition of proliferation, treatment with βHB induced cells to accumulate at the G₀–G₁ cell-cycle checkpoint (Fig. 4E) and increased expression of p53 and p21^Waf1 in hCAF-05 cells (Fig. 4F). To determine the effect of KD on CAFs in MC38 and CT26 tumors, the expression of αSMA, a marker of CAFs (26), was determined by Western blot. Feeding a KD repressed the expression of αSMA in MC38 and CT26 tumors (Fig. 4G and H). Together, these results demonstrate that increased ketogenesis inhibits CAF proliferation. Because βHB, at the same concentration that was used to treat CAFs, did not inhibit colorectal cancer cell proliferation (Supplementary Fig. S1), our results emphasize the functional effects of ketogenesis on the TME. To determine whether KLF5 contributes to CAF growth, hCAF-05 cells were infected with Ad-GFP or Ad-KLF5 followed by treatment with βHB. Overexpression of KLF5 increased hCAF-05 proliferation and reversed the inhibition of hCAF-05 cell proliferation by βHB treatment (Fig. 4I). In contrast, knockdown of KLF5 resulted in the inhibition of hCAF-05 cell proliferation (Fig. 4J). Together, our results demonstrate the regulation of CAF growth by the HMGCS2/βHB/KLF5 signaling axis.

We have shown that increased ketogenesis inhibits KLF5-dependent CXCL12 expression and improves the immunosuppressive TME. We next determined the levels of βHB in human colorectal cancers. As shown in Fig. 5A, the level of βHB was decreased in 12 of total 13 tumor samples analyzed compared with the adjacent normal controls. Conversely, increased CXCL12 protein (SDF1) levels were noted in the colorectal cancers (Fig. 5B). In agreement with the decreased βHB, decreased HMGCS2 protein levels were noted in eight of 10 colorectal cancers compared with matched normal colonic mucosa (Fig. 5C). High KLF5 expression is associated with poor prognosis in patients with colorectal cancer and liver metastasis (31). Moreover, elevated expression of CXCL12 protein (SDF1), CXCR4, and CXCR7 in colorectal cancers has been reported (32, 33). To further examine whether decreased expression of HMGCS2 was correlated with increased KLF5 and SDF1 expression in human colorectal cancers, samples of 5 pair-matched colorectal cancers with liver metastasis were analyzed using IHC staining. As shown in Fig. 5D, the intensity of HMGCS2 is negatively correlated with KLF5, SDF1, CXCR4, and CXCR7 in normal colon mucosa, primary and metastatic tumors, thus demonstrating that the decreased levels of HMGCS2/βHB are present in association with increased CXCL12/CXCR4/CXCR7 signaling in colorectal cancers.

We showed that HMGCS2/βHB inhibits CXCL12 expression in CAFs. To next determine whether KD inhibits the expression of CXCL12 in tumors, BALB/cJ or C57BL/6J mice with subcutaneously implanted CT26 cells or MC38 tumors were fed an NC or KD for 14 days (Fig. 6A), and CXCL12 protein (SDF1) and βHB levels in plasma and tumors were determined. βHB production was markedly elevated in the plasma (Fig. 6B and D) and tumors (Fig. 6C and E) from mice fed with KD. Conversely, KD significantly repressed the levels of SDF1 in the plasma (Fig. 6B and D) and tumor tissues (Fig. 6C and E). As CXCL12 contributes to colorectal cancer resistance to ICB therapy (34), we next determined whether KD enhances the efficacy of anti-PD1 therapy (Fig. 6F). Administration of KD enhanced the antitumor effects of anti-PD1 therapy in CT26 (Fig. 6G) and MC38 (Fig. 6H) tumor models. These effects on tumor growth correlated well with the improved survival observed in the tumor-bearing mice, establishing that anti–PD-1 antitumor activity is impaired with a decrease in ketogenesis but is restored with feeding a KD (Fig. 6I and J). CT26 tumor eradication was noted in 2/10 surviving mice in the group treated with anti-PD1. Importantly, tumor eradication was observed in all 7 surviving mice that received the combined KD and anti-PD1 antibody treatment.

Activation of CXCL12/CXCR4/CXCR7 signaling promotes tumor growth, migration, invasion, angiogenesis, and metastasis in various cancers, including colorectal cancers (34). To test whether alteration of ketogenesis affects colorectal cancer cell migration and invasion, we used CM collected from cultures of CAFs that overexpress HMGCS2 or HMGCS2-Mut. CM from CAFs overexpressing HMGCS2 (Fig. 7A and B), but not HMGCS2-Mut (Supplementary Fig. S6A and S6B), inhibited migration and invasion of HCT116 cells compared with CM from CAFs with control vector as determined by Boyden chamber assays and by a scratch migration assay (Supplementary Fig. S6C). To determine whether KD inhibits cancer metastasis, we used an in vivo lung metastasis model (Fig. 7C), and lung metastasis was assessed by GFP imaging (Fig. 7D and E). Compared with NC, KD remarkably repressed lung metastases in mice. The most common metastatic sites of colorectal cancer are the liver and the lung. We next determined whether KD affects the levels of CXCL12 in lung and liver tissues. KD repressed the expression of SDF1, which was associated with an increase of βHB in lung (Fig. 7F and G) and liver (Supplementary Fig. S6D and S6E) tissues of mice bearing tumors. Taken together, these findings demonstrate that an increase in ketogenesis results in the inhibition of colorectal cancer cell migration, invasion, and metastasis.

Metabolic reprogramming plays a critical role in the inhibition of antitumor immunity and thereby in tumor progression and metastasis (35). The TME is a critical factor that affects targeted therapies for colorectal cancers (3). A KD has been demonstrated to be a safe and achievable component in the treatment strategy for some cancers (10, 36). Recently, KDs have been shown to enhance the anticancer effects of ICB in colorectal cancer, melanoma, renal, and NSCLC orthotopic tumor models (11, 12). However, the activity of ketogenesis and the role of ketogenesis on the immunosuppressive TME are not known. Interestingly, our current study uncovers a novel metabolic fitness and flexibility of the immunosuppressive TME in colorectal cancers. We show that ketogenesis is downregulated in the TME; increased ketogenesis represses KLF5-dependent CXCL12 expression and improves immunosuppressive TME as noted by increased intratumoral CD8⁺ T-cell and NK cell infiltration in conjunction with decreased MDSC and M2-like macrophage populations in the TME, which leads to the enhanced efficacy of immunotherapy and inhibited lung metastasis (Fig. 7H). Importantly, our findings demonstrate that impaired ketogenesis in the TME is critical for colorectal cancer immune escape.

Our current study provides insight into the status of ketogenesis in the immunosuppressive TME. We show that KD inhibits MC38 tumor growth in immune competent mice whereas the ketone body βHB had minor effects on the proliferation of colorectal cancer cell lines, suggesting that KD may inhibit tumor growth through the TME. Indeed, we found that ketogenesis is downregulated in colorectal cancers and that feeding a KD inhibited the immunosuppressive TME. Our results are the first to show that downregulated ketogenesis in colorectal cancers contributes to the immunosuppressive TME. Activation of CXCL12/CXCR4 induces the recruitment of immunosuppressive cells in the colorectal cancer TME (37). Moreover, CXCL12 promotes recruitment of other CXCR4⁺ immunosuppressive cells such as MDSCs and Tregs toward the TME (4). Blocking CXCR4 and/or CXCR7 improves immunosuppressive TME, inhibits colorectal cancer lung metastasis, promotes infiltration of cytotoxic T cells, and synergizes with anti–PD-1 therapy in murine colorectal cancer models (38–41). CXCL12 has been shown to suppress the proliferation of blood-derived NK cells (42). We showed that KD repressed CXCL12 expression, improved the immunosuppressive TME, and enhanced the efficacy of anti-PD1 therapy for colorectal cancers. Our results suggest that KD may reduce the recruitment of CXCR4⁺ immunosuppressive cells by repression of CXCL12 expression, thus allowing for increased cancer-specific NK and CD8⁺ T-cell infiltration and enhanced efficacy of ICB therapy for colorectal cancers.

Although CXCL12 is increased in TME and is critical for tumor immunosuppression (9), regulation of CXCL12 in TME remains largely undefined. We demonstrated that an increase in ketogenesis (i.e., overexpression of HMGCS2, treatment with either βHB or KD) repressed the expression of CXCL12 through the inhibition of HDAC1-dependent KLF5 expression. CXCL12, which is expressed in the TME mainly by CAFs (7), plays a crucial role in the immunosuppressive TME. KLF5, a zinc-finger transcription factor that contributes to colorectal cancer cell growth (43), has recently been shown to inhibit intratumoral accumulation of cytotoxic T cells in pancreatic ductal adenocarcinoma (44). However, it is not known whether KLF5 contributes to the immunosuppression in colorectal cancers. CXCL12 is upregulated in numerous cancers; however, very little is known regarding the mechanisms for this increase. Our data identify CXCL12 as a direct target of KLF5, which binds to the CXCL12 promoter and transcriptionally controls CXCL12 expression in CAFs. Moreover, we show that an increase in ketogenesis or inhibition of KLF5 resulted in the inhibition of CAF proliferation. An increased percentage of tumor stromal cells, such as CAFs, is associated with a poor prognosis for colorectal cancer (5). Our results demonstrate that an increase in ketogenesis represses CXCL12 levels in the TME through both of inhibition of KLF5-dependent CXCL12 induction and inhibition of CAF proliferation. Our findings that KLF5 plays a critical role in promoting immunosuppressive TME in colorectal cancers provide an important step in unraveling the precise molecular regulators of CXCL12 expression in cancer.

CXCL12 plays a critical role in cancer metastasis and is a potential target for developing effective therapeutics against metastatic cancers (7). Tumor cells expressing CXCR4 or CXCR7 migrate following a CXCL12 gradient; organs with high CXCL12 expression are common terminals of CXCL12-guided cell migration (45). In addition to primary tumors, there is also sufficient expression of CXCL12 in the bone marrow, lungs, lymph nodes, and liver attracting CXCR4-positive cancer cells to these CXCL12-rich mesenchymal stroma niches to initiate metastasis (4). CXCL12 is mainly expressed by mesenchymal stromal cells in these organs (7). Moreover, there are number of reports demonstrating that mesenchymal stem cells can function as precursors for CAFs (46). We showed that increased ketogenesis inhibits CXCL12 expression in CAFs. These results suggest that increased ketogenesis may also repress CXCL12 in these mesenchymal stem cells. The most common metastatic sites of colorectal cancers are the liver and the lung. Reduction of CXCL12 in liver stroma has been shown to reduce the potential of colorectal cancer liver metastasis (47). Moreover, cancerous and noncancerous lung tissues showed increased expression of CXCL12 in patients with colorectal cancer, suggesting a contribution of CXCL12 to this site of metastasis for colorectal cancers (48). We showed that feeding a KD repressed CXCL12 expression in primary tumors, plasma, liver, and lung tissues, demonstrating the potential of using a supplemental KD to directly inhibit CXCL12 stimulation of cancer cells, thus interfering with the tumor immune response and the cancer biology driving the establishment of metastasis.

Hyperactivation of the CXCL12/CXCR4 axis is common in many solid tumors and plays a critical role in tumor immunosuppression and metastasis (9); therefore, the reduction of CXCL12 concentrations has the potential to inhibit the immunosuppressive TME and to decrease tumor and immunosuppressive cell activation/migration within the primary tumor and metastatic niche.

Inhibition of the CXCR4/CXCL12 or CXCR7/CXCL12 axis is an attractive strategy for treatment of colorectal cancers overexpressing these receptors (4). AMD3100 is an FDA-approved CXCR4 antagonist used for peripheral blood stem cell transplantation; however, its application to solid tumors is limited by its poor pharmacokinetics and toxic adverse effects after long-term administration (49). Thus, an alternative strategy to repress CXCL12 expression is necessary. Our findings provide a novel metabolic strategy to repress CXCL12 expression in the primary and metastatic niche, demonstrating a potential of using KD to enhance the efficacy of ICB for metastatic colorectal cancers and thus increase OS. Because increased CXCL12 in the primary tumor and metastatic niche is a general phenomenon in a number of cancers, our findings suggest a potential benefit of using KD not only for colorectal cancers but also for other cancers with increased CXCL12.

P. Rychahou reports grants from NIH during the conduct of the study. B.M. Evers reports grants from National Institutes of Health and National Cancer Institute during the conduct of the study. No disclosures were reported by the other authors.

R. Wei: Investigation. Y. Zhou: Investigation. C. Li: Investigation. P. Rychahou: Investigation. S. Zhang: Investigation. W.B. Titlow: Investigation. G. Bauman: Investigation. Y. Wu: Investigation. J. Liu: Investigation. C. Wang: Formal analysis. H.L. Weiss: Formal analysis, methodology. B.M. Evers: Conceptualization, formal analysis, supervision, funding acquisition, writing–original draft. Q. Wang: Conceptualization, supervision, writing–original draft.

This work was supported by National Institutes of Health grants R01 DK48498 and P30 CA177558 (to B.M. Evers), and a pilot grant from P20 GM121327 (to Q. Wang). The authors thank Donna Gilbreath for the article preparation; Dana Napier for tissue preparation, sectioning, and staining; Siva Gandhapudi and Jerry Woodward for Flow Cytometric analysis; the Flow Cytometry and Immune Monitoring core, OncoGenomics Shared Resource, Biospecimen Procurement and Translational Pathology, and Biostatistics and Bioinformatics shared resource facilities of the University of Kentucky Markey Cancer Center.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.