The purpose of this study is to investigate the impact of a high-salt diet (HSD), which is commonly found in Western countries, on the progression of glioma. Our research shows that the alterations in gut microbiota caused by an HSD facilitated the development of glioma. Mice fed an HSD have elevated levels of intestinal propionate, which accelerated the growth of glioma cells. We also find that propionate supplementation enhanced the response of glioma cells to low oxygen levels. Moreover, we identify a link between TGF-β signaling, response to low oxygen levels, and invasion-related pathways. Propionate treatment increases the expression of HIF-1α, leading to an increase in TGF-β1 production. Additionally, propionate treatment promotes glioma cell invasion through TGF-β signaling. Our findings suggest that an HSD-induced increase in propionate plays a crucial role in glioma progression by facilitating invasion through the hypoxic response and TGF-β signaling pathways, thereby establishing a significant connection between gut microbiota and the progression of glioma.

The importance of the gut microbiome in human health is increasingly recognized. In the context of cancer research, gut microbiota have gained attention because of their potential to promote cancer development and modulate the efficacy of cancer therapy (Blake et al., 2023; El Tekle and Garrett, 2023). Dietary patterns substantially influence the composition and function of the gut microbiome (Perler et al., 2023). The Western diet, which is prevalent in developed countries and is characterized by high consumption of processed foods rich in fat, sugar, and sodium, is a contributing factor to the increased prevalence of noncommunicable diseases, including cancer (Varlamov, 2017). According to a global analysis of noncommunicable diseases, high sodium intake is the leading cause of mortality (Afshin et al., 2019). A high-salt diet (HSD) alters the intestinal microbiome of both mice and healthy humans (Ferguson et al., 2019; Wilck et al., 2017). In particular, HSD modifies the composition of the gut microbiome and aggravates disease severity in mouse models of neuroinflammation (Wilck et al., 2017). Also, a human study shows an increase in the abundance of Bacteroides in the gut in response to HSD (Ferguson et al., 2019). These studies demonstrate the potential of high sodium intake to alter gut microbiota, which can influence disease progression.

Glioblastoma (GBM) is a highly malignant brain tumor that poses a significant challenge to medical practitioners owing to its aggressive nature and poor prognosis. Although surgery and chemotherapy are the standard treatments, the overall survival rate of GBM patients is only ∼15 mo, which is much lower than that of patients with most other tumor types (Khasraw et al., 2022). The treatment of GBM is further complicated by tumor cell heterogeneity, an immunosuppressive microenvironment, and invasive behavior, leading to recurrence and treatment failure (van Solinge et al., 2022). Therefore, understanding the glioma microenvironment is crucial for developing effective therapies. Recent research suggests that alterations in gut microbiota composition and metabolites occur in both mice and humans, which could have significant implications for patient care (Dono et al., 2020; Patrizz et al., 2020). However, further investigation is required to understand the underlying mechanisms.

This study explored the influence of HSD on glioma progression by altering gut microbiota. Our results indicate that HSD reduced survival in glioma mice by modifying the composition of the gut microbiota, leading to increased levels of intestinal propionate. Propionate supplementation exacerbated glioma progression by promoting hypoxic responses and stimulating the production of TGF-β1, which facilitated the glioma cell invasion and expression of type I collagen. Our findings suggest that HSD and changes in gut microbiota may be linked to glioma progression, highlighting the role of gut microbiota–derived metabolites in glioma development.

HSD worsens glioma via the intestinal microbiome

To evaluate the impact of HSD on glioma progression, we fed mice either a normal salt diet (NSD) or HSD and monitored the sodium consumption levels of the mice while maintaining a similar body weight to those fed NSD (Fig. S1, a–d). Mice were injected with GL261 cells, a murine glioma cell line, 0, 2, or 4 wk after switching to HSD (Fig. 1 a). Only mice that switched to HSD 4 wk before tumor injection showed a significant decrease in survival rate compared with those that switched at 0 or 2 wk (Fig. 1 b and Fig. S1 e). A similar result was observed in mice injected with CT2A cells, another murine glioma cell line, 4 wk after switching to HSD (Fig. S1 f). We also injected heterogeneous glioma cells of GBM organoids (GBMOs) derived from GBM tissues of a spontaneous glioma mouse model and observed a similar decrease in the survival of HSD-fed mice compared with NSD-fed mice (Fig. S1 g). We confirmed that HSD-fed mice had larger tumor burdens 15 days post injection (dpi) of green fluorescent protein (GFP)–expressing GL261 (GL261-GFP) cells (Fig. 1, c and d). Overall, these results demonstrate that an HSD of sufficient duration before tumor injection promotes glioma progression.

Since HSD can cause alterations in the gut microbiota, we sought to investigate the role of gut microbiota in the progression of gliomas promoted by HSD. Initially, we administered an antibiotic cocktail (ABX) to deplete the gut microbiome (Fig. 1 e). Administration of ABX eliminated the disparities in survival rates and tumor sizes between NSD- and HSD-fed mice (Fig. 1, f–h). Furthermore, we performed fecal microbiota transplantation (FMT) in germ-free (GF) mice to confirm the effects of HSD-induced changes in the gut microbiota. We transferred fecal microbiota from NSD- or HSD-fed mice into GF mice (NSD-FMT or HSD-FMT mice) 2 wk before the injection of GL261 cells (Fig. 1 i). HSD-FMT mice showed a significant decline in median survival and an increase in tumor burden compared with NSD-FMT mice (Fig. 1, j–l). These findings suggest that HSD promotes glioma progression by modulating the gut microbiota.

HSD alters the gut microbiome, leading to increased intestinal propionate levels

We next investigated changes in gut microbiota induced by HSD under our experimental conditions because the composition of gut microbiota can be influenced by environmental factors. To accomplish this, we conducted 16S ribosomal DNA (rDNA) sequencing of fecal DNA from mice injected with GL261 cells. We compared the composition of gut microbiota between NSD and HSD groups for both time points to determine changes induced by HSD independent of glioma formation (Fig. 2, a–h). At the family level, Muribaculaceae and Bacteroidaceae were dominant in both groups, with higher abundance of Bacteroidaceae in the HSD group than in the NSD group (Fig. 2 a). Alpha diversity analysis revealed no significant difference in species richness (i.e., abundance-based coverage estimator [ACE] index) between groups, whereas species diversity (i.e., non parametric [NP] Shannon index) was lower in the HSD group than in the NSD group (Fig. 2, b and c). In addition, principal coordinate analysis (PCoA) showed that NSD and HSD groups clustered separately, with a significant between-group difference (Fig. 2 d). To identify bacteria with substantial differences between groups, we performed linear discriminant analysis (LDA), and most selected bacteria belonged to the higher taxonomic levels of Bacteroides vulgatus (Fig. 2 e). Moreover, using quantitative PCR (qPCR), we confirmed the increases in Bacteroides and B. vulgatus in the HSD group compared with the NSD group (Fig. 2, f and g). These results suggest that HSD induces compositional changes in gut microbiota, particularly an increase in the Bacteroides and B. vulgatus.

Several species of the Bacteroides, such as B. vulgatus, which are present in human microbiota, contribute to the production of propionate, a short-chain fatty acid (SCFA), through the succinate pathway (Louis and Flint, 2017). Therefore, we hypothesized that the elevated abundance of Bacteroides and B. vulgatus caused an increase in the levels of propionate in HSD-fed mice compared with NSD-fed mice. To validate this hypothesis, we employed Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) analysis of fecal DNA 16S rDNA sequencing data from NSD- or HSD-fed mice. This approach allowed us to estimate the abundance of genes involved in propionate production, and we found a significant increase in the expression of these genes in the HSD group compared with the NSD group (Fig. 2 h). We then validated the abundance of propionyl-CoA carboxylase beta chain (Pccb), which is involved in the final step of the propionate-producing pathway (Reichardt et al., 2014), using qPCR, and confirmed a significant increase in abundance in the HSD group compared with the NSD group (Fig. 2 i). Additionally, the abundance of Pccb was significantly correlated with the abundance of Bacteroides (Fig. 2 j) and B. vulgatus (Fig. 2 k).

We also verified the presence of Bacteroides and B. vulgatus in fecal DNA of GF mice after FMT, and there was a substantial increase in B. vulgatus, but not Bacteroides, in HSD-FMT mice compared with NSD-FMT mice (Fig. 2, l and m). Additionally, HSD-FMT mice showed increased abundance of Pccb compared with NSD-FMT mice (Fig. 2 n). The abundance of Pccb was significantly correlated with the abundance of B. vulgatus but not Bacteroides (Fig. 2, o and p).

We then confirmed levels of propionate in cecal tissues from NSD-fed or HSD-fed mice and observed a significant increase in propionate. However, the levels of acetate and butyrate, the other types of SCFAs, remained unchanged in HSD-fed mice (Fig. 2 q). Additionally, FMT from HSD-fed mice also increased the level of cecal propionate, but not acetate or butyrate, in GF mice compared with GF mice receiving FMT from NSD-fed mice (Fig. 2 r). These results demonstrate that HSD induces alterations in gut microbiota, specifically promoting the production of propionate.

Interestingly, both NSD and HSD increased the abundance of fecal B. vulgatus within 1 wk, and maintained the increased abundance until 2 wk. However, the abundance of B. vulgatus was decreased in NSD over time, eventually reaching levels that were no longer statistically significantly different from those observed before the dietary change (P = 0.0772, Fig. S1 h). This result indicates that microbiota require more time to adapt to host’s dietary change, even though a week of increased salt consumption was sufficient to alter the microbiota composition, and explains the survival difference between mice injected with GL261 at 2 and 4 wk after the dietary change (Fig. S1 e). In addition, we compared compositional changes between NSD and HSD groups in non–tumor- and tumor-bearing mice (Fig. S1 i). At both non–tumor- and tumor-bearing mice, NSD and HSD groups clustered separately with significant differences (Fig. S1, j and k), and the Bacteroides and B. vulgatus were selected as bacteria with significant differences between groups (Fig. S1, l and m). The abundance of the Bacteroides in the HSD group was significantly higher at both non–tumor- and tumor-bearing mice, but the fold increase decreased after glioma formation (Fig. S1 n). This trend was also observed for the abundance of B. vulgatus and Pccb (Fig. S1, o and p). When comparing levels of cecal SCFAs depending on the presence of glioma, there was no significant change in levels of acetate or butyrate at either time point. However, similar to changes in the abundance of B. vulgatus and Pccb, propionate levels were significantly higher in the HSD group than the NSD group at the non-tumor mice, but this significance declined after glioma formation (Fig. S1, q–s). These results suggest that glioma formation decreases changes in gut microbiota induced by HSD.

Propionate supplementation promotes glioma progression and enhances the transcriptional hypoxic response in cancer cells

To explore the effects of propionate on glioma, we administered propionate, acetate, butyrate, or sodium chloride as a control in drinking water 4 wk before intracranial GL261 injection (Fig. 3 a). Unlike acetate and butyrate, propionate supplementation decreased the survival rate compared with the control group (Fig. 3 b). Additionally, we observed a significant reduction in the survival rate in the PROP compared with the CTRL after GBMO-derived glioma cell injection (Fig. S2 a). Moreover, a larger tumor burden was observed in the propionate supplemented group (PROP) than in the control group (CTRL) 20 days after GL261-GFP injection (Fig. 3, c and d). These results suggest that propionate supplementation accelerates glioma progression.

To assess the effect of propionate supplementation on various cell types within the glioma microenvironment, we conducted single-cell RNA-sequencing (scRNA-seq) analysis. Intriguingly, the glioma cell exhibited the highest number of differentially expressed genes (DEGs) (Fig. 3 e), indicating that propionate supplementation primarily affects glioma cells within the glioma microenvironment. Among the immune cell types, monocytes and macrophages had the highest number of DEGs (Fig. 3 e and Fig. S2 b).

To investigate the potential role of immune cells in promoting glioma in response to propionate supplementation, we employed two distinct mouse models: immunodeficient mice lacking functional lymphoid cells (NOD-scid IL2rgnull, NOG) and CCR2 KO mice, which have impaired monocyte recruitment. Even in the NOG, propionate supplementation resulted in decreased mouse survival compared with that in the CTRL (Fig. 3 f). Similarly, CCR2 deficiency did not affect the survival rate of the mice supplemented with propionate (Fig. 3 g). Moreover, we validated the decreased survival of mice in the PROP compared with that in the CTRL by injecting clodronate liposomes to deplete macrophages in glioma tissues (Fig. S2 c). These findings suggest that propionate supplementation significantly affects glioma cells with a minimal impact on immune cells, thereby contributing to glioma progression in an immune cell–independent manner.

To examine detailed transcriptomic changes in glioma cells induced by propionate supplementation, we reclustered cells into eight distinct clusters (Fig. 3 h). The structures of these clusters differed between groups, with clusters 5, 6, and 7 being more prevalent in the PROP than in the CTRL. We then used gene set enrichment analysis (GSEA) to compare characteristics of these clusters with all other clusters and found that clusters 5 and 6 commonly exhibited an increase in hypoxia within the hallmark gene sets (Fig. S2 d). Further analysis of all clusters revealed that hypoxia was the most significantly increased gene set in the PROP (Fig. 3 i and Fig. S2 e) and hypoxia-related genes (Liberzon et al., 2015), including Slc2a1 and Vegfa, were mainly expressed by cells in clusters 5, 6, and 7 (Fig. S2 f). We confirmed the high hallmark hypoxia gene set score in clusters 5 and 6, which was also significantly higher in the PROP group than in the CTRL (Fig. S2 g). In addition, we examined the score for hypoxic genes in GBM (GBM hypoxia), which were selected through scRNA-seq analysis of cells from GBM patients (Zhang et al., 2023). This score was also high in clusters 5 and 6 and was significantly higher in the PROP than in the CTRL (Fig. 3 j). These findings suggest that propionate supplementation enhances the hypoxic response in glioma cells at the transcriptional level.

In addition to hypoxia, cluster 7 showed distinct expression patterns for myogenesis and transforming growth factor-β (TGF-β) signaling (Fig. S2 d). However, there was no significant difference in the myogenesis gene set score between the PROP and CTRL (Fig. S2 h). Instead, we then evaluated similar gene sets to hallmark myogenesis gene set among canonical pathways and found that the Kyoto Encyclopedia of Genes and Genome (KEGG) focal adhesion set was the most significant, excluding those gene sets related to muscle (Table S1). We confirmed that the score for focal adhesion was high in clusters 5, 6, and 7 and significantly increased in the PROP compared with the CTRL (Fig. 3 k). Additionally, the focal adhesion score was positively correlated with the hypoxia score in glioma cells (Fig. 3 l). Similarly, the TGF-β signaling score was highly expressed in clusters 5 and 7, significantly increased in the PROP compared with the CTRL (Fig. 3 m), and positively correlated with the hypoxia score in glioma cells (Fig. 3 n). There was also a positive correlation between the focal adhesion score and the TGF-β signaling score (Fig. S2 i). These results demonstrate an interactive correlation among hypoxia, focal adhesion, and TGF-β signaling sets and suggest that propionate supplementation enhances their expression.

Propionate supplementation enhances hypoxia markers in glioma cells regardless of whether hypoxia is caused by oxygen deprivation

Glucose transporter protein type 1 (GLUT1), encoded by Slc2a1, and B-cell lymphoma 2–interacting protein 3 (BNIP3), encoded by Bnip3, are commonly used hypoxia markers induced by hypoxia-inducible factor-1 (HIF-1) (Pietrobon and Marincola, 2021). We confirmed that Slc2a1 and Bnip3 were densely expressed in hypoxic clusters of glioma cells and that their expression was significantly higher in all clusters in the PROP than in the CTRL (Fig. S2, j and k). Before investigating the change in the hypoxic response of glioma cells following propionate supplementation, we examined the hypoxic status of glioma with pimonidazole, which forms covalent bonds with macromolecules in hypoxic cells and accumulates in areas of low oxygenation (Masaki et al., 2016). Interestingly, propionate supplementation did not cause a notable difference in the tumor vessels and hypoxic area, despite the difference in tumor size at an earlier time point (11 dpi) (Fig. S3 a). However, tumor progression made difference in hypoxic regions with necrosis in glioma from mice supplemented with propionate (15 dpi) (Fig. S3 b). We additionally analyzed in the later tumor progression stage (28 dpi for CTRL and 22 dpi for PROP), when the hypoxic regions became more distinct, to exclude the effects caused by the difference in tumor size (Fig. S3 c). Although hypoxic and necrotic areas in tumor were larger in the propionate-supplemented group, they were observed in both conditions. These data suggest that hypoxic areas in glioma result from the faster growth of tumor mass, rather than being directly caused by propionate supplementation. To avoid a severe hypoxic status due to glioma progression, we measured the intensity of these markers at an earlier time point (i.e., 11 dpi, Fig. 4) than the day of sample collection for scRNA-seq analysis (i.e., 20 dpi, Fig. 3).

We compared the intensities of GLUT1 and pimonidazole in GFP-positive tumor regions (Fig. 4 a). Although there was no difference in pimonidazole intensity between two groups, we observed a significant increase in GLUT1 intensity in the PROP compared with the CTRL (Fig. 4 b). To determine whether the enhancement of GLUT1 intensity within the PROP occurred in the hypoxic region, we compared GLUT1 intensity between regions with low and high pimonidazole intensity (Fig. 4 c). We found a significant increase in GLUT1 intensity only in the region with low pimonidazole intensity but not in the region with high pimonidazole intensity by propionate supplementation. Furthermore, we found that BNIP3 intensity displayed a similar trend (Fig. 4, d–f), with a notable increase in BNIP3 intensity in the region with low pimonidazole intensity but not in the region with high pimonidazole intensity. Additionally, we analyze the GLUT1- or BNIP3-positive area in the GFP-positive tumor area to compensate for the heterogeneity of their expression in tumor. GLUT1 was expressed in more tumor cells in the propionate-treated group, but there were no differences in the pimonidazole-high region (Fig. S3, d and e). BNIP3-positive area also increased by propionate supplementation only in the pimonidazole-low region (Fig. S3, f and g). At a later time point (i.e., 15 dpi), GLUT1 intensity was significantly increased in both pimonidazole-low and pimonidazole-high areas, and BNIP3 intensity was increased only in pimonidazole-high areas (Fig. S3, h–n). We observed enhanced BNIP3 and GLUT1 expression in pimonidazole-low regions of HSD-fed mice compared with NSD-fed mice (Fig. S3, o–u). These findings indicate that both propionate supplementation and HSD enhanced GLUT1 and BNIP3 expression in glioma cells regardless of oxygen deprivation–induced hypoxia in the tumor microenvironment. These results further support the hypothesis that propionate supplementation contributes to the enhanced hypoxic response in glioma cells.

Propionate directly enhances the hypoxic response of glioma cells under normal oxygen conditions

To examine cellular dynamics in glioma cells, we utilized RNA velocity and partition-based graph abstraction (PAGA) analyses (Fig. 5, a and b). RNA velocity analysis showed that the streams started with cluster 2 and ended in clusters at sites opposite to cluster 2 (Fig. 5 a). Analyzing the PAGA graph based on RNA velocity demonstrated that cluster 2 was the starting point of dynamics in glioma cells (Fig. 5 b). Cluster 2 highly expressed proliferation-related genes, including Cdk1 and Top2a (Fig. S2 f). Thus, we further analyzed the proliferation gene scores (Yuan et al., 2019). Glioma cells at the start of the dynamics had a high proliferation gene score that decreased toward the end of the dynamics (Fig. 5 c). Compared with cluster 5, which harbored a hypoxic signature, cluster 2 exhibited a significantly higher proliferation score but lower GBM hypoxia score (Fig. 5, d and e). Under hypoxic conditions, cellular metabolism generally changes from oxidative phosphorylation (OXPHOS) to glycolysis due to low oxygen concentrations (Al Tameemi et al., 2019). Cluster 2 also exhibited a significantly higher score for OXPHOS gene set of Gene Ontology Biological Process (GO BP) than cluster 5 (Fig. 5 f). Thus, cluster 2 was considered to reflect a type of the non-hypoxic and proliferating cell.

After assessing cluster 2, which exhibited the characteristics of non-hypoxia and proliferation and represented the start of dynamics, we analyzed changes in the hypoxia score following propionate supplementation. We discovered that cluster 2 had higher scores for both hallmark hypoxia and GBM hypoxia gene sets in the PROP than in the CTRL (Fig. 5, g and h). Additionally, cluster 2 showed a significant increase in focal adhesion and TGF-β signaling scores in the PROP compared with the CTRL (Fig. S4, a and b). We then compared scores related to two major HIFs, HIF-1 and HIF-2, which are the master transcription factors in response to hypoxia. Cluster 2 of the PROP showed higher scores for the PID HIF-1 transcription factor pathway than the CTRL (Fig. 5 i). Furthermore, Semenza HIF-1 targets, the score for another gene set related to HIF-1 signaling, was also increased in cluster 2 of the PROP compared with the CTRL (Fig. 5 j). However, the enhancement of the HIF-1 pathway was limited to clusters 2 and 6 (Fig. S4 c). Moreover, in cluster 2 of the PROP, HIF-2 pathways were upregulated compared with those in the CTRL, whereas upregulation of the HIF-2 pathway was observed across all clusters (Fig. S4 d). These results suggest that propionate supplementation augments the HIF-1 pathway in non-hypoxic proliferating cells.

To evaluate whether propionate could enhance the response to hypoxia in glioma cells, we performed bulk RNA-seq in GL261 cells treated with sodium chloride (CTRL) or propionate (PROP) for 4 h. We confirmed that propionate treatment did not promote cell growth (Fig. S4 e). Analysis of hallmark gene sets using GSEA showed that hypoxia and OXPHOS were upregulated in the propionate-treated group compared with the control group (Fig. 5 k and Fig. S4 f). To verify the increased expression of OXPHOS-related genes, we measured the oxygen consumption rate (OCR) after propionate treatment (Fig. S4 g) and found a slight, but not statistically significant, increase in the OCR after propionate treatment (Fig. S4 h).

We next investigated HIF-1α expression in the nuclei of various glioma cell lines and GBMOs after propionate treatment. HIF-1α expression was enhanced in GL261 cells in both protein and transcript levels (Fig. 5, l and m; and Fig. S4, i and j). However, the expression of GLUT1 and BNIP3 remained unchanged in the protein level, although there was an enhanced expression at transcript levels. Furthermore, we measured HIF-1α expression 4 h after propionate treatment in GBMOs derived from murine GBM tissues (Fig. 5, n and o). HIF-1α expression was increased on the central regions of GBMOs owing to limited oxygen delivery. However, following treatment with propionate, we observed an increase in HIF-1α expression in both the surface and center regions of GBMOs (Fig. 5 o). We also observed an increase in HIF-1α intensity in U87MG cells, a human glioma cell line, following propionate treatment (Fig. 5, p and q). When comparing HIF-1α intensity between two patient-derived glioma (PDG) cell lines, we observed that PDG-S cells exhibited higher HIF-1α intensity than PDG-W cells, and propionate treatment induced an increase in HIF-1α intensity in PDG-W cells but not in PDG-S cells (Fig. 5, r and s). These results suggest that despite normal oxygen conditions, propionate directly induces an increase in HIF-1α in glioma cells originally harboring the low expression of HIF-1α, leading to a hypoxic response.

Propionate promotes the invasion of glioma cells via HIF-1TGF-β signaling

To investigate whether the propionate-induced hypoxic response leads to activation of specific pathways, we performed bulk RNA-seq in GL261 cells treated with sodium chloride (CTRL) or propionate (PROP) for 24 h, which was a later time point. GSEA demonstrated that the hypoxic response was still upregulated among the hallmark gene sets in the PROP compared with the CTRL (Fig. 6 a). In addition to hypoxia, we observed an increase in gene sets associated with TGF-β signaling in the PROP compared with the CTRL among hallmark and canonical and genetic perturbation gene sets (Fig. 6 b and Fig. S4 k). GSEA results for the hallmark TGF-β signaling gene set exhibited core enrichment genes in the PROP (Fig. S4 l), and we found that these genes were associated with several Smad proteins, including Smad2 and Smad3 (Fig. S4 m). Based on these findings, we postulated that propionate enhances TGF-β signaling in glioma cells by increasing HIF-1α expression.

To verify our hypothesis, we measured levels of TGF-β1 in the culture media of GL261 cells 4 and 24 h after treatment with propionate. We found no change in TGF-β1 production after 4-h treatment but a significant increase in TGF-β1 levels 24 h after treatment (Fig. 6 c). Transcriptomic analysis confirmed the upregulated expression of Tgfb1, Tgfb3, and Serpine1, but not Tgfb2, in GL261 cells 24 h after propionate treatment (Fig. S4 n). Furthermore, we found that the propionate-induced increase in TGF-β1 production by GL261 cells was abolished when Hif1a was knocked down using lentivirus-mediated shRNA (GL261-shHIF-1α) or after treatment with the HIF-1α inhibitor Cay10585 (Fig. 6 d; and Fig. S4, o and p). We also observed that propionate induced an increase in TGF-β1 production in PDG-W cells but not in PDG-S cells (Fig. 6, e and f), similar to the changes in HIF-1α intensity induced by propionate treatment (Fig. 5, r and s). Additionally, we confirmed that the level of TGF-β1 increased at 20 days compared with 15 days after tumor injection, and propionate supplementation augmented these levels at both time points (Fig. 6 g).

To explore the effect of TGF-β signaling on the survival of mice with glioma, we administered a TGFβRI inhibitor (TGFβRi) after mice developed glioma. The inhibitor did not appear to have a considerable impact on the survival rate compared with that in the CTRL group (Fig. 6 h). However, when mice were administered propionate in addition to the inhibitor, we observed a significant reversal of the decreased survival rate typically associated with propionate supplementation (Fig. 6 h). HSD-fed mice displayed a similar recovery in the survival rate after TGFβRi administration (Fig. S4 q). These findings imply that inhibiting TGF-β signaling using a blocking agent can counteract the detrimental effects of propionate on the survival of mice with glioma. Additionally, in contrast to the scrambled RNA control (GL261-shControl), GL261-shHIF-1α tumors did not show an increase in TGF-β1 upon propionate supplementation (Fig. 6 i). Likewise, the size of tumors injected with GL261-shHIF-1α was not affected by propionate supplementation (Fig. 6, j and k). These suggest that enhanced HIF-1α signaling by propionate is crucial for the increase in TGF-β1 in tumor, and indicate that tumor cells are the main cellular source of the propionate-induced increase in TGF-β1.

As integrins activate TGF-β secreted in a latent form, the expression of integrins in tumor cells is one of the strategies for evading host immunity (Takasaka et al., 2018). In scRNA-seq, Itgav was expressed in GL261 tumor cells and several immune cells including T cells, macrophages, and dendritic cells (Fig. S4 r). Under propionate supplementation, the expression level of Itgav decreased in tumor cells (0.6290–0.4738 ± 0.007792), but increased in T cells (0.6755–0.7409 ± 0.02526) and macrophages (0.3574–0.4736 ± 0.01769) (Fig. S4 r). However, in vitro, propionate treatment did not alter the Itgav level in GL261 (Fig. S4 s), suggesting that the altered expression of Itgav in vivo under propionate supplementation is not a direct effect of propionate.

Our scRNA-seq analysis revealed correlations among hypoxia, TGF-β signaling, and focal adhesion in glioma cells (Fig. 3, l and m; and Fig. S2 i). Focal adhesions are intracellular structures that mediate adhesion to the extracellular matrix, promoting cell migration (Fierro Morales et al., 2022). We observed that the PROP group exhibited increased adhesion between cells and the matrix but decreased adhesion between cells and tight junctions as compared to the CTRL group (Fig. S5, a–c). Based on this analysis, we hypothesized that propionate enhances glioma cell invasion through hypoxia and TGF-β signaling. To assess the potential of propionate to enhance glioma cell invasion, we conducted invasion assays using GL261 cells and GBMOs. In the GL261 cell invasion assay, we pretreated GL261 cells in the bottom chambers with propionate for 24 h before initiating the invasion assay. Subsequently, we seeded additional GL261 cells onto Matrigel-coated Transwell inserts to initiate invasion. Propionate-treated GL261 cells exhibited increased invasiveness in Matrigel compared with the control group, and this enhancement was significantly suppressed when a TGFβRi was introduced (Fig. 6, l and m). For GBMOs, the invasion assay employed a matrix composed of collagen type I and Matrigel. After 24 h of treatment, there was no significant difference in invasion distance between groups; however, after 48 h of propionate treatment, GBMOs exhibited an increased invasion distance, which was mitigated by the addition of a TGFβRi (Fig. 6, n and o). These findings indicate that propionate promotes the invasion of both GL261 cells and GBMOs through TGF-β signaling.

Our scRNA-seq analysis revealed an increase in collagen-related Reactome gene sets in the PROP compared with the CTRL (Fig. S5 d). Additionally, we observed a positive correlation between scores for Reactome collagen degradation and KEGG focal adhesion (Fig. S5 e). We found that Col1a1, a member of the KEGG focal adhesion and collagen-related gene set, was significantly higher in the PROP than in the CTRL (Fig. S5 f). Type I collagen often increases during the development of tumors, and its expression is linked to the poor prognosis of various cancer types (Shi et al., 2022). Also, COL1A1 expression in glioma cells is a key factor for collective motility (Comba et al., 2022). Thus, we measured the expression of the COL1A1 protein in glioma tissues and observed that COL1A1-expressing glioma cells were mainly located at the peripheral site of the tumor and appeared to spread beyond the tumor border (Fig. 6 p and Fig. S5 g). To determine whether TGF-β signaling mediated the increase in COL1A1-expressing glioma cells following propionate supplementation, we injected a TGFβRi and measured the intensity of COL1A1 in glioma tissues. We observed that the increase in COL1A1-positive glioma cells due to propionate supplementation was reduced by TGFβRi injection (Fig. 6 q and Fig. S5 g). When comparing the proportion of COL1A1-positive area in the GFP-positive tumor area between groups, although not statistically significant, we noticed a trend toward a larger COL1A1-positive area following propionate supplementation, which was decreased by TGFβRi injection (Fig. S5 h). These findings suggest that propionate supplementation promotes COL1A1 expression in glioma cells via TGF-β signaling.

Transcriptional differences in human gliomas associated with COL1A1 expression are reminiscent of those observed in mouse gliomas induced by propionate supplementation

To evaluate whether human glioma cells showed a tendency for transcriptional changes similar to those caused by propionate in our experimental models, we analyzed publicly available scRNA-seq data from tumor samples of GBM patients (GSE182109) deposited in Gene Expression Omnibus (GEO) of the National Center for Biotechnology Information (NCBI). We reclustered glioma cells from four samples of newly diagnosed GBM patients into 11 clusters (Fig. S5 i). COL1A1 was one of the most significantly increased genes and was associated with invasion, which appeared to be induced by HIF and TGF-β signaling via propionate supplementation in our experimental model. We found that there were COL1A1-expressing cells in patients GBM02 and GBM04 but nearly no COL1A1-expressing cells in patients GBM01 and GBM03 (Fig. S5 j). Therefore, we classified GBM01 and GBM03 patients as the COL1A1low and GBM02 and GBM04 patients as the COL1A1high. We confirmed COL1A1 expression and observed a significant difference between groups (Fig. 7, a and b).

As the key signatures of GBM hypoxia, hallmark TGF-β signaling, and KEGG focal adhesion were present in murine glioma cells induced by propionate supplementation (Fig. 3, j–n), we examined the scores for these signatures between COL1A1low and COL1A1high and observed that all scores were significantly higher in the COL1A1high (Fig. 7 c). Additionally, there were significant positive correlations between these three signatures in human glioma cells (Fig. S5, k–m). Similar to mouse glioma cells, we confirmed scores for proliferation and GBM hypoxia in human glioma cells. Cluster 4 exhibited the highest score for proliferation, whereas cluster 1 exhibited the highest score for GBM hypoxia (Fig. 7, d and e). Comparing these two clusters, we found a significant difference in proliferation and GBM hypoxia scores (Fig. 7 f), with cluster 4 considered to be a non-hypoxic and proliferating type of cells. We also observed a significant increase in scores for GBM hypoxia, hallmark TGF-β signaling, and KEGG focal adhesion in cluster 4 of the COL1A1high compared with the COL1A1low (Fig. 7 g). These findings demonstrate that transcriptomic changes in human glioma cells observed when classified according to COL1A1 expression are similar to those in mouse glioma cells after propionate supplementation.

To determine which genes were commonly upregulated in non-hypoxic and proliferating mouse and human glioma cells, we compared the DEGs in cluster 2 of mouse glioma cells and cluster 4 of human glioma cells and identified 16 genes, including COL1A1, that were commonly upregulated (Fig. 7 h). We used Enrichr to determine which gene sets were related to these common DEGs among GO BP 2023. We observed that six gene sets were significantly associated with the common DEGs and were primarily related to cell cycle and migration (Fig. 7 i). We also identified specific genes related to each gene set, such as COL1A1, SPRY2, and SEMA3E, that were associated with cell migration (Fig. 7 j). We then investigated the association between the expression of common DEGs and overall survival in GBM patients by analyzing overall survival in The Cancer Genome Atlas GBM dataset provided by GEPIA using the Kaplan–Meier analysis based on the expression of common DEGs. No significant correlation was observed between COL1A1 expression and overall survival (Fig. S5 n). We observed a nonsignificant negative correlation between common DEG expression and the overall survival of GBM patients (Fig. 7 k). Additionally, there was a nonsignificant negative correlation between the expression of cell cycle–related genes CDKN1A, TCIM, and TOP2A and overall survival (Fig. 7 l). However, the expression of migration-related genes, including COL1A1, SPRY2, and SEMA3E, was significantly and negatively correlated with overall survival (Fig. 7 m). These findings imply that our migration signature including COL1A1, which was typically upregulated in both mouse and human glioma cells, could be a crucial element in the progression of GBM and may have a detrimental impact on the prognosis of GBM patients.

Our research uncovered that altering diet can impact glioma development by influencing gut microbiota composition. Specifically, we observed that HSD resulted in an increase in B. vulgatus and propionate-producing enzymes, leading to a rise in propionate levels in the gut environment. Furthermore, propionate supplementation promoted glioma progression by enhancing the hypoxic response of glioma cells, even under normal oxygen conditions. This increased expression of HIF-1α resulted in TGF-β1 production, which aided glioma cell invasion by expressing type I collagen. Additionally, we found that modifications in human glioma cells categorized based on COL1A1 expression mirrored those triggered by propionate supplementation in mouse glioma cells. Both human and mouse glioma cells exhibited a common upregulated migration signature that included COL1A1, and this signature was inversely correlated with the prognosis of GBM patients.

We found that administering an HSD of appropriate duration (i.e., 4 wk) before tumor injection accelerated glioma progression. The period between the ages of 4 and 8 wk in mice corresponds to the human juvenile and puberty stages (Wang et al., 2020). Juvenile diets have long-lasting effects on the adult microbiome, and restoring them to a healthy state requires considerable time (De Filippo et al., 2010). In our mouse model, sodium intake was equivalent to consuming 44 mg/kg of sodium per day, or 3,520 mg of sodium per day for an 80-kg adult. As of 2019, the global sodium intake level was estimated to be ∼4,310 mg/day (WHO, 2023), which is similar to or even higher than that in our model. Therefore, our mouse model has a sodium intake comparable to that of individuals on an HSD and may also be applicable to people during their teenage years.

We also found that HSD led to an increase in intestinal propionate level and that different bacteria can produce SCFAs via diverse pathways (Deleu et al., 2021). One study shows that the succinate pathway, which utilizes the Pccb, is the most prevalent pathway for generating propionate in a human microbiota dataset (Reichardt et al., 2014). Under our experimental conditions, HSD increased the abundance of Bacteroides, B. vulgatus, and Pccb, and there were significant positive correlations between the abundance. Interestingly, FMT from HSD did not alter the abundance of Bacteroides but increased the abundance of B. vulgatus and Pccb, as well as levels of propionate. Therefore, B. vulgatus utilizing Pccb may be the major propionate producer under our experimental conditions.

Recent studies provide conflicting evidence regarding the role of SCFAs. Although many studies suggest that SCFAs predominantly induce antitumor responses, other reports indicate the opposite. For example, one study demonstrates that systemic treatment with SCFAs inhibits the efficacy of anti-cytotoxic T lymphocyte–associated protein 4 antibody therapy against tumors (Coutzac et al., 2020), and another study reports that SCFAs enhance insulin-like growth factor 1 production in prostate cancer, thereby promoting tumor growth (Matsushita et al., 2021). Our investigation revealed another mechanism that promotes the growth of brain tumors, involving propionate, a SCFA.

Previous studies have shown that propionate inhibits cancer cell growth in vitro (Bindels et al., 2012; Ryu et al., 2022). In these studies, treatment of cells with 2.5–10 mM propionate resulted in reduced cell proliferation. Moreover, a previous study demonstrates that 4 mM propionate increases oxygen consumption by activating the β-oxidation–like pathway in a human goblet cell line (Kawaguchi et al., 2022). However, the concentrations used in these studies may not reflect conditions within the glioma microenvironment. A previous study reported that mice receiving propionate supplementation had circulating propionate levels of ∼8–10 μM, whereas the control mice had levels of ∼3–5 μM (Bartolomaeus et al., 2019). Therefore, we investigated the effects of additional propionate on glioma cells at a physiological concentration under 5 μM. Interestingly, this concentration did not significantly affect glioma cell growth or oxygen consumption but induced OXPHOS expression. Therefore, physiological concentrations of propionate may increase HIF-1α expression in glioma cells without suppressing their proliferation by altering other pathways.

TGF-β plays a key role in promoting the invasion and metastasis of cancer cells, including gliomas (Derynck et al., 2021), and hypoxia and TGF-β signaling have a synergistic effect on various types of cancer (Tam et al., 2020). Our findings support the interactive effect of HIF-1α and TGF-β signaling in glioma cells, which promotes invasion. Studies demonstrate that HIF-1α expression activates Smad2/3 protein, which in turn strengthens TGF-β signaling (Mallikarjuna et al., 2019; Zhang et al., 2003). Our transcriptomic analysis also revealed the upregulation of genes related to the Smad2/3 pathway. This analysis also suggested the possibility that propionate supplementation modifies not only the amount of TGF-β1, but also its usability in glioma by altering the expression of Itgav, integrin-activating TGF-β1. This suggests that propionate-induced HIF-1α increases the production of TGF-β1 via the Smad2/3 pathway, but further investigation into the precise mechanisms is still needed.

There are practical difficulties in conducting research on the effects of diet, metabolites, and gut microbiota in humans. Instead, to determine the relevance of our findings to human disease, we performed a comparative analysis using scRNA-seq data from GBM patients and our data from mouse glioma cells. We found notable variation in COL1A1 expression among GBM patients, leading to its classification into high- and low-expression groups. Surprisingly, this classification revealed a pattern similar to that observed in mouse glioma cells supplemented with propionate. We also analyzed migration signature genes, including COL1A1, by comparing changes in human and mouse glioma cells, which showed a negative correlation with the prognosis of patients with GBM. Our analysis indicated that the modifications seen in the mouse model are also present in human patients, which may play a role in the progression of glioma.

We found that propionate increased HIF-1α in the nucleus and conducted transcriptomic analysis to identify potential mechanisms; however, we did not find any concrete evidence. Propionate may exert its effects by altering protein-level mechanisms, such as by directly inhibiting the degradation of HIF-1α. Therefore, further research is necessary to investigate this phenomenon in detail. Additionally, we noted an increase in the number of cancer cells expressing Col1a1 in glioma tissues in vivo after propionate administration. Although COL1A1 is suggested to play a role in invasion based on previous research, its usefulness as an invasion marker requires further investigation. Although our experimental models provide significant insights into the link of HSD, microbiota-derived propionate and the activation of the HIF-1α–TGF-β1 axis in GBM progression, there is currently a lack of sufficient research addressing these mechanisms in human patients. Further clinical studies are needed to validate these preclinical findings and to better understand the impact of dietary and microbiota-related factors on GBM progression in humans. Additionally, exploring the role of the gut–brain axis in glioma progression and its potential modulation through microbiota-targeted strategies holds promise. Overall, addressing these future research directions will help expand our understanding of gliomas and may lead to the discovery of novel therapeutic strategies for glioma patients.

Study design

We referred to prior experiments conducted in our laboratory to determine the appropriate sample size in advance for identifying differences between groups, performing statistical analyses, and ensuring reproducibility. The sample size was determined based on previously published studies (Kim et al., 2022; Park et al., 2021), and the minimum number of animals per group was three. Each sample was obtained from a biologically independent mouse. All experiments were repeated at least twice, with the number of replications specified in each figure legend.

All mice were grouped according to their genotype or treatment. Mice within each group were randomly assigned. For in vitro experiments, samples were organized into groups based on treatment. Before treatment, mice were randomly allocated to each group. Samples from animals in different groups were collected in an alternating manner. Data collection and analysis were performed in a blinded fashion using randomized samples. No data were excluded from the results.

Animals

Male mice were used for all animal experiments. C57BL/6J mice (4 wk old, RRID:IMSR_JAX:000664) were purchased from the Korea Advanced Institute of Science & Technology (KAIST) Laboratory Animal Resource Center and DBL Co. Ltd. NOG mice (NOD.Cg-PrkdcscidIL2rgtm1Sug/JicKoat, RRID:IMSR_TAC:NOG) were purchased from Koatech. GF C57BL/6J mice were purchased from the POSTECH Biotech Center. CCR2 KO mice (C57BL/6J background, B6.129S4-Ccr2tm1Ifc/J, RRID:IMSR_JAX:004999) were purchased from the Jackson Laboratory. For experiment with CCR2 KO mice, C57BL/6J mice purchased from DBL Co. Ltd were used as controls. LsL-TdTomato mice (C57BL/6J background, B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J, RRID:IMSR_JAX:007914) and LsL-EGFRvIII mice (NCI Mouse Repository:01X68, FVB background) were kindly provided by Dr. Jeong Ho Lee (KAIST, Daejeon, Republic of Korea), and were backcrossed with C57BL/6J mice five times before tumor induction. Mice were housed in a specific pathogen-free facility at the KAIST Laboratory Animal Resource Center and maintained under a 12-h light/dark cycle at 18–24°C and 30–70% humidity. All animal experiments and determination of humane endpoints were performed in accordance with guidelines and protocols approved by the KAIST Institutional Animal Care and Use Committee (KA2022-006).

For the HSD model, 4-wk-old mice were assigned to two dietary groups: the NSD group, which received standard chow (0.5% NaCl, D18012201; Research Diets, Inc.) and 0.5% NaCl (7548-4105; Daejung) in tap water, and the HSD group, which received 4% NaCl-supplemented HSD chow (D09022509; Research Diets, Inc.) with 1% NaCl in tap water. The mice in both groups had ad libitum access to their respective diets and water throughout the experiment.

For SCFA administration, 4-wk-old mice were provided with drinking water containing 200 mM sodium propionate (P0512), 200 mM sodium acetate (S0559), and 200 mM sodium butyrate (S0519; Tokyo Chemical Industry Co. Ltd.) or 200 mM NaCl ad libitum until the end of the experiment.

Cells

Mouse glioma cell lines GL261 (RRID:CVCL_Y003) and GL261-GFP were kindly provided by Dr. Injune Kim (KAIST, Daejeon, Republic of Korea) (Kim et al., 2018). The human GBM cell line, U87MG (HTB-14, RRID:CVCL_0022; ATCC), was kindly provided by Dr. Ji Ho Park (KAIST, Daejeon, Republic of Korea). The cells were passaged using trypsin/ethylenediaminetetraacetic acid (EDTA, 25-052-CV) and cultured in Dulbecco’s modified Eagle’s medium (DMEM, 10-013-CVRC) containing 10% fetal bovine serum (FBS, 35-015-CV) and 1% penicillin–streptomycin (30-002-CI; Corning) at 37°C. No Mycoplasma contamination was confirmed with the e-Myco plus Mycoplasma PCR kit (25237; Intron Biotechnology).

Heterogeneous tumor cells were obtained from GBMOs. GBMOs were ground and filtered through a 70-μm strainer (93070; SPL). Cells were cultured in DMEM containing 10% FBS and 1% penicillin–streptomycin at 37°C.

Generation of PDG cell

All tissue samples were obtained from Seoul St. Mary’s Hospital, with approval from our institutional review boards (KC22SISI00079) (Choi et al., 2022). Freshly resected GBM samples were processed immediately: the tumor tissue was minced and placed in DMEM with 20% FBS and 1% penicillin–streptomycin. The cells were then cultured at 37°C in a humidified atmosphere containing 5% CO2.

GBM organoids

Spontaneous murine high-grade glioma (EGFRvIII+TP53PTEN) expressing TdTomato was induced in LsL-TdTomato × LsL-EGFRvIII mice as previously described (Park et al., 2022). Briefly, 0- to 2-day-old pups were anesthetized via hypothermia and injected with a plasmid containing sgRNA-Trp53/Pten-Cas9-Cre (backbone: pU6-[BbsI]_CBh-Cas9-T2A-BFP, RRID:Addgene_64323; Addgene) into the lateral ventricle of the brain. Electric pulses (100 V for 50 ms, five times at 950-ms intervals) were applied.

GBMOs were generated with modified protocol described previously (Jacob et al., 2020). Briefly, tumor tissues were isolated from mice 12–14 wk after tumor induction. The whole brain was isolated and kept in H+GPSA medium containing Hibernate A (A1247501), 1× GlutaMAX (35050-061), 25 μg/ml amphotericin B (15290-026; Thermo Fisher Scientific), and 1× penicillin–streptomycin. Under a stereoscopic microscope, the normal brain tissue was removed using fine-tip forceps. The tumor tissues were chopped into pieces, 0.5–1-mm diameter using a blade and fine scissors. Small pieces and debris were removed by washing tissues with H-GPSA. Tumor pieces were suspended in GBM organoid medium (GBOM), 50% F12/DMEM (10-092-CV; Corning), and 50% Neurobasal medium (21103-049) supplemented with N-2 (17502-048), B-27 (12587-010), MEM-NEAA (11140-050), and GlutaMAX (Thermo Fisher Scientific). The suspensions were incubated in ultra-low attachment plates with shaking at 120 rpm. For the first 3 days, the culture medium was replaced daily with fresh GBOM to remove debris, and thereafter, it was replaced every 48 h. Organoids were incubated for at least 2 wk before freezing.

To prepare organoids for freezing, they were incubated with 10 μM Y-27632 (13624S; Cell Signaling Technology) in GBOM for 1 h at 37°C. Subsequently, the cells were incubated in 10% dimethyl sulfoxide (DMSO, 036480; Thermo Fisher Scientific) for 15 min at room temperature. The organoids were frozen in GBOM supplemented with 10 μM Y-27632 and 10% DMSO.

Mouse glioma model

Tumor cell injection was conducted 2 or 4 wk after the initiation of the experiments. Cultured glioma cells were washed with Dulbecco’s phosphate-buffered saline (DPBS, 21-031-CV; Corning) to remove the residual medium and detached using trypsin/EDTA solution. After centrifugation at 440 rcf at 4°C for 5 min to remove the supernatant, the cells were resuspended in DPBS at 1 × 105 cells/2 μl.

To prepare the mice for injection, anesthesia was induced using vaporized isoflurane, and the mice were fixed on a stereotaxic instrument (505314; World Precision Instrument). Mice were intracranially injected with 1 × 105 glioma cells as previously described (Park et al., 2021). Briefly, tumor cells were injected at coordinates of 2 mm anterior, 2 mm right from the bregma, and 3 mm inferior to the surface. Injection was conducted at a rate of 0.4 μl/min for 5 min using a 10-μl Hamilton syringe (803; Hamilton Company) and a syringe pump (LEGATO 130; KD Scientific). After the injection, the skin around the wound was sutured to facilitate wound closure.

To measure tumor volume, a 60-μm-thick coronal section of the brain was obtained from the tumor injection site, and tumor volume was calculated according to the following formula: 0.52=(longestdiameter)×(perpendiculardiameter)2. Fiji software (RRID:SCR_002285) was used to measure the length of the tumor area (Schindelin et al., 2012).

Mouse treatment

For antibiotic treatment, mice were orally administered 200 μl of an antibiotic cocktail for five consecutive days per week for 2 wk before and 2 wk after glioma challenge, using a Zonde needle (Jeong do Bio & Plant Co. Ltd; 0.9 × 50 mm [20G]). The antibiotic cocktail consisted of 10 mg/ml ampicillin (A-1414), 5 mg/ml vancomycin (V-1065), 10 mg/ml neomycin sulfate (N-1053), 10 mg/ml gentamycin (G-1067; AG Scientific), and 10 mg/ml metronidazole (M1547; Sigma-Aldrich) in distilled water.

For macrophage depletion, mice were intravenously injected with 200 μl of clodronate liposomes (LIP-CP-010-010; Liposoma) per mouse one day before the glioma challenge and on days 3, 7, 11, and 15 after glioma challenge.

To block TGF-β signaling, mice were intraperitoneally injected with 0.25 mg LY-364947 (3965362; PeproTech, Inc.) on days 8, 11, 14, and 17 after glioma challenge. DMSO was used as the carrier control.

FMT

Feces were harvested from mice fed an NSD or HSD for five consecutive days each week during the 2 wk following the diet change. Four fecal samples per group were diluted with 2 ml of DPBS and passed through a 70-μm strainer. Homogenates were centrifuged at 1,900 rcf and 4°C for 5 min, and the supernatants were harvested. GF mice were orally administered 100 μl of the supernatant per day for five consecutive days each week for 2 wk. The administration schedule was aligned with the day of fecal sample collection from NSD- or HSD-fed mice. After bacterial colonization, GF mice were injected intracranially with 1 × 105 GL261 or GL261-GFP cells.

qPCR

GL261 cells (2 × 105 cells/well) were treated with 1 μM sodium chloride or propionate in 6-well plates (30006; SPL), and total RNA was extracted using RNeasy Plus Mini Kits (74136; Qiagen) 4 h or 24 h after treatment. cDNA was synthesized using ReverTra Ace qPCR RT Master Mix (FSQ-201; Toyobo). For analysis of bacterial abundance from fecal samples, fecal DNA was extracted from mouse fecal samples using QIAamp Fast DNA Stool Mini Kit (51604; Qiagen).

For quantitative analysis, 10 μl of SYBR Green Real-time PCR Master Mix (TOQPK-201; Toyobo) and each 25 μM of forward and reverse primer were mixed with 1 μl of cDNA or 30–50 ng of fecal DNA. RT-qPCR was performed using CFX96 Touch Real-Time PCR System 3.1 (RRID:SCR_018064; Bio-Rad) as follows: initial denaturation at 95°C for 15 min, followed by 40 cycles of 94°C for 15 s, 60°C for 30 s, and 72°C for 30 s.

The relative expression levels of mouse genes were normalized to Hprt expression, and the abundance of bacteria and bacterial genes was normalized to the value for universal bacteria using the 2(−∆∆C(T)) method. Primer sequences used in this study are obtained from the previous study (Gradisteanu Pircalabioru et al., 2022; Kim et al., 2022; Reichardt et al., 2014; Seystahl et al., 2017; Song et al., 2014; Wu et al., 2023) or primer bank (Spandidos et al., 2010), and provided in Tables S2 and S3.

Bulk RNA-seq

Total RNA was extracted from GL261 cells (2 × 105 cells/well in 6-well plates) 4 h after treatment with 1 μM NaCl or propionate using RNeasy Plus Mini Kit. Similarly, total RNA was extracted from GL261 cells (1 × 105 cells/well in 24-well plates [30024; SPL]) 24 h after treatment with 1 mM NaCl or propionate using RNeasy Plus Mini Kit. The sample quality was assessed using a TapeStation D1000 Screen Tape (NC1780350; Agilent). The extracted RNA samples were processed using TruSeq Stranded Total RNA LT Sample Prep Kit (20020599; Illumina) and sequenced on an Illumina HiSeq 4000 platform (RRID:SCR_016386; Illumina).

The sequencing reads were trimmed for quality using Fastp (RRID:SCR_016962) (Chen et al., 2018), and the trimmed reads were mapped to the reference sequence of Mus musculus (GRCm38) using HISAT2 (RRID:SCR_015530) (Kim et al., 2019). After mapping, StringTie (RRID:SCR_016323) was used to assemble and merge transcripts, allowing their abundance to be estimated (Pertea et al., 2015). The DEG analysis was performed using DESeq2 (RRID:SCR_015687) (Love et al., 2014). GSEA (RRID:SCR_003199) was conducted based on the normalized counts.

HIF-1α knockdown

The pLKO.1-puro empty vector (RRID:Addgene_8453; Addgene) served as the backbone for constructing the lentiviral plasmid. To enable magnetic or flow cytometric sorting, the selection marker puromycin resistance gene (PuroR) was replaced with Thy1.1 (CD90.1), resulting in a modified vector called pLKO.1-Thy1.1. The shRNA sequences targeting HIF-1α or a scrambled sequence were inserted downstream of the hU6 promoter in the pLKO.1-Thy1.1 vector. The shRNA sequence targeting mouse HIF-1α was 5′-GCC​GCT​CAA​TTT​ATG​AAT​ATT-3′. The scrambled shRNA sequence was 5′-ATA​CCG​GTA​TAC​GTC​ACG​GAC-3′.

A Lenti-X 293T cell line (632180; Takara Bio) was transfected with lentivirus plasmids and packaging vectors using Lipofectamine 3000 (L3000001; Invitrogen). After 48 h, the viral supernatant was collected, filtered using 0.45-μm syringe filters (Sartorius), and used to transduce GL261 or GL261-GFP cells.

16S rRNA sequencing analysis

Fecal DNA was extracted from mouse fecal samples using QIAamp Fast DNA Stool Mini Kits, as described above. Sequencing was performed using Illumina MiSeq (RRID:SCR_016379; Illumina). Bacterial DNA amplification was conducted using PCR targeting the V3-V4 regions of the 16S rRNA gene under the following conditions: initial denaturation at 95°C for 3 min, followed by 30 cycles at 95°C for 30 s, 55°C for 30 s, and 72°C for 30 s, and final elongation at 72°C for 5 min.

Low-quality reads with average quality scores <25 were removed using Trimmomatic v.0.32 (RRID:SCR_011848). Quality-controlled paired-end sequence data were merged using VSEARCH version 2.13.4 (RRID:SCR_024494), and primer sequences were trimmed using an alignment algorithm. Nonspecific amplicons that did not encode 16S rRNA were identified using the HMMER software package ver.3.2.1 (RRID:SCR_005305). Unique reads were extracted, and redundant reads were clustered with unique reads using the derep_fulllength command of VSEARCH. Taxonomic assignment was performed using the EzBioCloud 16S rRNA database (Yoon et al., 2017) with precise pairwise alignment. Chimeric reads were filtered using the UCHIME algorithm (RRID:SCR_008057) and non-chimeric 16S rRNA database from EzBioCloud, removing reads with <97% similarity. After chimeric filtering, reads that could not be identified at the species level with <97% similarity in the EzBioCloud rRNA database were compiled and de novo clustering was performed using the cluster_fast command to generate additional operational taxonomic units (OTUs). OTUs consisting of single reads (singletons) were excluded from further analysis.

Secondary analysis, including diversity calculations and biomarker discovery, was conducted using in-house programs from CJ Bioscience, Inc. The alpha diversity indices ACE and NP Shannon were estimated, and beta diversity distances were calculated using the Jensen–Shannon method to visualize sample differences. Taxonomic and functional biomarkers were identified using statistical comparison algorithms (RRID:SCR_014609, LDA Effect Size, LefSe) with functional profiles predicted by PICRUSt (RRID:SCR_016855). All mentioned analyses were performed using the EzBioCloud 16S-based MTP, CJ Bioscience’s bioinformatics cloud platform.

SCFA measurement

SCFA levels in the mouse cecum tissues were quantified using a previously described method (Suzuki et al., 2006). Briefly, 0.1 g of the cecum tissue was treated with 0.25 ml 1N of HCL and 0.1 M of isobutanol. Fatty acids were extracted by shaking the mixture in 1 ml of diethyl ether for 10 min. After centrifugation at 13,500 rpm for 5 min, the supernatant was collected. A 1-μl aliquot of the supernatant sample was loaded into a gas chromatography/mass spectrometry system (7890A; Hewlett Packard/Agilent) equipped with a DB-FATWAX Ultra Inert column (Agilent) for analysis.

Cytokine measurements

The culture medium from GL261 cells treated with 1 μM NaCl or propionate for 4 or 24 h was harvested. The culture medium from PDG cells treated with 1 μM NaCl or propionate for 24 h was harvested. GL261 cells were treated with 100 μM Cay10585 (CAY-10012682; Cayman Chemical) or DMSO as a control. Brain tissue homogenates were obtained by grinding the hemisphere of mice bearing glioma with 0.5 ml of DPBS 20 dpi with GL261, GL261-shControl, or GL261-shHIF-1α cells. TGF-β1 levels in the culture medium from GL261 cells or brain tissue homogenates were measured using Mouse TGF-β1 DuoSet ELISA Kit (DY1679-05; R&D Systems) following the manufacturer’s instructions. TGF-β1 levels in the culture medium from PDG cells were measured using Human TGF-beta 1 DuoSet ELISA Kit (DY240-05; R&D Systems). Absorbance at 450 nm was measured using a SpectraMax microplate reader, and data were processed using SoftMax Pro version 7.1 (RRID:SCR_014240; Molecular Devices).

Oxygen consumption measurement

OCR was measured in GL261 cells as previously described (Oh et al., 2021). Cells (10,000 per well) were plated on Seahorse culture plates (Agilent) and supplemented with minimal DMEM containing 25 mM glucose (LS001-02), 1 mM pyruvate (LS013-01), and 2 mM glutamine (LS002-01; Wellgene). A Seahorse XFe96 Bioanalyzer (RRID:SCR_019545; Agilent) was used to measure the OCR. Cells were treated with 1 μM NaCl or propionate. Data were analyzed using Wave software (RRID:SCR_024491; Agilent).

Immunofluorescence

To assess hypoxia in glioma, the mice were intraperitoneally injected with 60 mg/kg pimonidazole (HP1-100; Hypoxyprobe) in DPBS 90 min before sacrifice. For brain tissue collection, the mice were sacrificed and perfused with DPBS and 4% paraformaldehyde (PFA) solution (BPP-9004; Tech & Innovation) through the heart. Brain tissues were fixed in 4% PFA solution at 4°C for 1 day, followed by dehydration in 30% sucrose (SR1030-250-00; Biosesang) solution at 4°C for 2 days. The tissues were embedded in OCT solution (4583; Sakura Finetek) and sectioned at a thickness of 20 μm with HM525 Cryostat (HM525; Thermo Fisher Scientific). Tissue sections were blocked and stained in a staining buffer containing 5% goat serum (005-000-121; Sigma-Aldrich, RRID:AB_2336990) and 0.3% Triton X-100 (TR1020-500-00; Bio Basic, Inc.) in DPBS. Primary antibodies, including allophycocyanin-conjugated anti-pimonidazole (4.3.11.3, HP1-100; Hypoxyprobe, RRID:AB_2811309), anti-GLUT1 (ARC0304, A11727, RRID:AB_2897763), anti-BNIP3 (polyclonal, A5683; ABclonal, RRID:AB_2766443), and anti-COL1A1 (polyclonal, PA5-29569; Invitrogen, RRID:AB_2547045), were used at 1:200 dilution. After primary staining, the sections were washed with 0.3% Triton X-100 in DPBS and incubated with Alexa Fluor 594 (AF594)–conjugated goat anti-rabbit IgG secondary antibody (polyclonal, 111-585-144; Jackson ImmunoResearch, RRID:AB_2307325) at a 1:1,000 dilution. Sections were mounted using DAPI-containing mounting solution (ab104139; Abcam). Tumor size was measured by sectioning embedded tissues at a thickness of 40 μm, followed by washing with 0.3% Triton X-100 in DPBS and mounting with DAPI-containing mounting solution.

For glioma cell cultures, GL261 cells (1 × 104 cells/0.5 ml/well), U87MG cells (1 × 105 cells/0.5 ml/well), and PDG-W and PDG-S cells (1 × 104 cells/0.5 ml/well) were seeded in 8-well culture slides (30408; SPL). The glioma cells were treated with 1 μM NaCl or propionate for 4 h. The cells were then washed with DPBS and fixed with a 4% PFA solution for 10 min at room temperature. Following fixation, the cells were blocked and stained with an anti-HIF-1α antibody (polyclonal, PA1-16601; Invitrogen, RRID:AB_2117128) at a 1:200 dilution in staining buffer. Samples were washed with 0.3% Triton X-100 in DPBS and incubated with AF594-conjugated goat anti-rabbit IgG secondary antibody or Cy5-conjugated goat anti-rabbit IgG secondary antibody (polyclonal, 111-175-144; Jackson ImmunoResearch, RRID:AB_2338013) at a 1:1,000 dilution. The samples were mounted using DAPI-containing mounting solution.

Images were acquired and processed using an LSM800 confocal microscope (RRID:SCR_025048) and ZEN software (RRID:SCR_013672; Zeiss). Images acquired using tile scanning were stitched using the stitching function in ZEN software. Fiji software was used to measure the intensities and lengths of the tumor areas (Schindelin et al., 2012).

Western blotting

GL261 cells (2 × 105 cells/2 ml/well) were incubated in a 6-well culture plate for 24 h with 1 μM NaCl or propionate for 24 h. For whole-cell fractions, cells were lysed with 100 μl of RIPA buffer (CBR002; LPS Solution) containing 1x protease inhibitor cocktail (PIC) (P3100; GenDEPOT). Lysates were incubated on ice for 30 min and centrifuged (13,000 rpm, 10 min, 4°C). The protein concentration was measured in supernatants using the DC protein assay (BR5000115; Bio-Rad).

For nuclear fractions, cells were incubated in 400 μl of Buffer A (10 mM HEPES [BB001; Wellgene], 10 mM KCl, 0.1 mM EDTA [ML-005; Wellgene], 1 mM DTT [D1037; Biosesang], and 1x PIC) on ice for 15 min. After the addition of 25 μl of 10% NP-40, cells were vortexed vigorously for 10 s and centrifuged (13,000 rpm, 10 min, 4°C). After washing the pellets with Buffer A by repeated incubation and centrifugation, pellets were resuspended in 50 μl of Buffer B (20 mM HEPES, 400 mM NaCl 1 mM EDTA, 1 mM DTT, and 1x PIC) by vigorous vortexing and incubated for 10 min on ice. Nuclear protein-containing supernatants were obtained by centrifugation (13,000 rpm, 10 min, 4°C). The protein concentration was measured in supernatants.

Proteins (50 μg) and a prestained protein marker (P8502-050; GenDEPOT) were loaded into the wells of 7.5% precast gels (LP00070G; LumiNano). After electrophoresis at 60 V for 120 min, proteins were transferred to a Trans-Blot Turbo Mini 0.2-μm polyvinylidene difluoride (PVDF) membrane (1704156; Bio-Rad) using Trans-Blot Turbo Transfer (1704150; Bio-Rad). The membrane was incubated with 1:1,000 diluted anti-HIF-1α antibody (polyclonal; Invitrogen), anti-BNIP3 antibody (polyclonal; ABclonal), anti-GLUT1 antibody (polyclonal; ABclonal), or anti-GAPDH antibody (14C10; Cell Signaling, RRID:AB_1903993) overnight at 4°C. The membrane was then incubated with 1:5,000 diluted StarBright Blue 700 goat anti-rabbit IgG (12004162; Bio-Rad, RRID:AB_2721073) for 1 h at room temperature. The images were acquired using a ChemiDoc MP imaging system (RRID:SCR_019037; Bio-Rad).

Single-cell preparation

To prepare brain tissue samples from tumor-challenged mice, cardiac perfusion was performed with cold DPBS following the sacrifice of mice to remove blood. The right hemisphere of the brain containing the glioma was dissected into small pieces using a blade. For tumor cell isolation, brain tissues were mechanically and enzymatically dissociated using a tumor dissociation kit for mice in a C-tube (130-093-237) with gentleMACS Octo Dissociator (RRID:SCR_020280; Miltenyi Biotec). The dissociation program used was 37C_m_TCK_1, and the process was carried out for 40 min. The resulting cell suspensions were passed through a 70-μm strainer.

For immune cells, single cells were isolated from brain tissues, as previously described (Park et al., 2021). Briefly, brain tissue samples were digested with a mixture of 2 mg/ml collagenase IV (LS004189; Worthington Biochemical Corporation) and 30 μg/ml DNase I (10104159001; Roche) at 37°C for 30 min. The samples were filtered through a 70-μm strainer.

Both tumor and immune cells were isolated using density gradient centrifugation in media containing 30 and 70% Percoll (17-0891-01; GE Healthcare). After isolation, cells were incubated with ammonium chloride–potassium lysis buffer for 5 min at room temperature to remove red blood cells.

Single-cell transcriptomic analysis

The mice were injected with either GL261-GFP cells for glioma cell analysis or GL261 cells for immune cell analysis. As described above, single cells were isolated from the glioma tissues. To block Fc receptors, isolated cells were treated with a 1:100 dilution of anti-CD16/32 (2.4G2; produced from a HB-197 cell line [RRID:CVCL_9148; ATCC]). Cells were then stained with a 1:200 dilution of anti-CD45.2-phycoerythrin (104, 12-0454-82; Thermo Fisher Scientific, RRID:AB_465678) and 5 μl of 7-AAD solution (559925; BD Biosciences, RRID:AB_2869266). Target cells were sorted using FACSAria II (RRID:SCR_018934; BD Biosciences). Glioma cells were identified as 7-AAD GFP+ CD45.2, while immune cells were identified as 7-AADCD45.2+.

A single-cell library was generated using Chromium Single Cell 3′ Library Kit (10X Genomics) and sequenced on a HiSeq X Ten platform (Illumina) with 10,000 cells per sample. Sequencing data were processed and aligned into a count matrix using Cell Ranger software (RRID:SCR_017344; 10X Genomics). Subsequent analysis was conducted using Seurat v4.3.0 (RRID:SCR_007322) (Hao et al., 2021). Gene scores were determined using the AddmoduleScore_UCell function (Andreatta and Carmona, 2021).

For data analysis, cells with RNA counts and features ranging within 0.025–0.975 quantiles and mitochondrial counts below 5% (for glioma cells) or 10% (for immune cells) in each sample were selected. Selected cells were normalized using the NormalizeData function with a scale factor of 10,000, and variable gene expression was determined using the FindVariableFeatures function.

For glioma cells, the samples from the NaCl- and propionate-supplemented groups were named CTRL and PROP, respectively. Integration of CTRL and PROP data was performed using the FindIntegrationAnchors function (dims = 1:50), followed by integration with the IntegrateData function (dims = 1:50). Cell clustering was performed using FindNeighbors (dims = 1:50) and FindClusters (resolution = 1.5). Clusters expressing Pecam1, Pdgfrb, and AY036118 were considered to contaminate endothelial cells or technically abnormal cells and were subsequently excluded from the analysis. Cells with Ptprc > 0.01 were also excluded to remove immune cells. The remaining cells were reclustered using FindNeighbors (dims = 1:50) and FindClusters (resolution = 0.7).

RNA velocity and PAGA analyses were performed on glioma cells using Velocyto (RRID:SCR_018167) and scVelo (RRID:SCR_018168) packages (Bergen et al., 2020; La Manno et al., 2018). Spliced and unspliced cell counts were calculated using the Velocyto package, and the data were filtered based on the reclustered cells using the Seurat package. The filtered cells were used for RNA velocity analysis using the scVelo package. The resulting RNA velocity streams were visualized using t-distributed stochastic neighbor embedding (t-SNE) plots generated using the Seurat package. The PAGA was generated using the scVelo package.

For immune cell analysis, samples from the NaCl- and propionate-treated groups were named CTRL and PROP, respectively. Data from the CTRL and PROP groups were integrated using the FindIntegrationAnchors function (dims = 1:50) followed by integration with the IntegrateData function (dims = 1:50). Cell clustering was performed using FindNeighbors (dims = 1:50) and FindClusters (resolution = 1.2). These processes resulted in an estimation of 23,609 cells.

For analysis of human data, cells with RNA counts and features ranging from 0.025 to 0.975 quantiles and mitochondrial counts <10% in each sample were selected. Selected cells were normalized using the NormalizeData function with a scale factor of 10,000, and variable gene expression was determined using the FindVariableFeatures function. Cells with Ptprc > 0.01 were also excluded to remove immune cells. Cells expressing glioma markers (OLIG1, SOX2, NES, PDGFRA, GFAP, S100B, and EGFR > 0.2) were considered glioma cells. Data were integrated using the FindIntegrationAnchors function (dims = 1:50) followed by integration with the IntegrateData function (dims = 1:50). Cell clustering was performed using FindNeighbors (dims = 1:30) and FindClusters (resolution = 0.7). These processes resulted in 29,861 cells.

Invasion assay

For the Transwell invasion assays, GL261 cells were seeded in the bottom chamber at a density of 5 × 104 cells/well in 24-well plates (37024; SPL). Cells were treated with 1 μM NaCl or propionate, along with 10 μg/ml LY-364947 (3965362; PeproTech). After 24 h of treatment, fresh GL261 cells were seeded on Transwell inserts coated with Matrigel (CLS354234; Corning) and incubated for 28 h. Before the invasion assay, 100 μl of Matrigel solution (0.3 mg/ml), diluted in serum-free DMEM, was added to the Transwell insert and allowed to gel for 5 h. Following the invasion assay, cells in the Transwell insert were fixed with 4% PFA solution, permeabilized using methanol, and stained with 0.05% crystal violet (C1066; Samchun Pure Chemical Co., Ltd.) solution. Noninvasive cells on the top side of the Matrigel matrix were gently removed using a cotton swab. After allowing the Transwell membrane to dry, images of the remaining cells were captured using a microscope, and the number of cells was counted using Fiji software.

The GBMO invasion assay protocol was adapted from a previously published protocol for tumor spheroids (Berens et al., 2015). To prepare the matrix, a mixture of collagen type I (final concentration 2 mg/ml; 637-00653; Wako) and Matrigel (final concentration 2 mg/ml) was diluted in serum-free DMEM. GBMOs were suspended in this matrix mixture, and 40 μl of the organoid suspension was seeded at the bottom of each well in 24-well plates. The plates were incubated for 1 h to allow gel hardening. Next, GBMOs treated with 1 μM NaCl or propionate, with or without 10 μg/ml LY-364947, were added to the wells to cover the gel and organoids. Images of organoids were captured using a microscope at 24 and 48 h after treatment. The maximal length of organoids was measured using Fiji software.

Statistical analysis

Statistical analyses, excluding scRNA-seq and PCoA data, were performed using GraphPad Prism version 10 (RRID:SCR_002798; GraphPad Software, Inc.). Results are presented as the mean ± standard error of the mean (SEM). For scRNA-seq data analysis, one-way ANOVA was used to compare each gene or gene score, and DEGs were identified using the FindMarker function. PCoA data were analyzed using PERMANOVA provided by EzBioCloud (CJ Bioscience). Survival data were analyzed using the log-rank test. All the data are representative of more than two independent experiments. Statistically significant differences are indicated as follows: ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001; and ****P < 0.0001. NES, normalized enrichment score; FDR, false discovery rate; R, Pearson’s correlation coefficient.

Online supplemental material

Fig. S1 illustrates changes in the gut microbiota induced by HSD and the survival of glioma-bearing mice. Fig. S2 demonstrates enhanced hypoxic responses in glioma cells following propionate supplementation. Fig. S3 displays hypoxia during tumor progression and the propionate-induced hypoxic response in a non-hypoxic tumor microenvironment. Fig. S4 presents enhanced HIF-1α–dependent TGF-β signaling under normoxic conditions in glioma cells following propionate treatment. Fig. S5 shows the correlation between a COL1A1-high signature and poor prognosis in patients. Table S1 lists the canonical pathways overlapping with hallmark myogenesis genes. Table S2 provides the primer sequences for mouse genes, and Table S3 provides the primer sequences for bacterial genes.

scRNA-seq and bulk RNA-seq data are deposited in the GEO series GSE233411, GSE233412, and GSE260955 of the National Center for Biotechnology Information. Human scRNA-seq data were obtained from publicly available data in GSE182109. All data are available from the lead contact upon request.

Resource availability

Requests for resources, reagents, and further information should be directed to the lead contact Heung Kyu Lee (heungkyu.lee@ kaist.ac.kr).

Materials availability

All the mouse lines used in this study were purchased from or provided by the indicated companies or researchers.

The authors thank the members of the Laboratory of Host Defenses for their helpful discussions. Inclusion and diversity: We support the inclusive, diverse, and equitable conduct of research.

This work was supported by the National Research Foundation of Korea grant (RS-2022-NR071193 to H.-J. Kim, RS-2021-NR056438, RS-2023-NR077244, and RS-2024-00439735).

Author contributions: C.W. Kim: conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, validation, visualization, and writing—original draft, review, and editing. H.-J. Kim: conceptualization, investigation, and writing—original draft. I. Kang: investigation. K.B. Ku: investigation. Y. Kim: investigation. J.H. Park: investigation. J. Lim: investigation. B.H. Kang: investigation. W.H. Park: investigation. J. La: investigation. S. Chang: investigation. I. Hwang: investigation. M. Kim: investigation. S. Ahn: methodology and resources. H.K. Lee: conceptualization, data curation, formal analysis, funding acquisition, methodology, project administration, supervision, validation, visualization, and writing—original draft, review, and editing.

Afshin
,
A.
,
P.J.
Sur
,
K.A.
Fay
,
L.
Cornaby
,
G.
Ferrara
,
J.S.
Salama
,
E.C.
Mullany
,
K.H.
Abate
,
C.
Abbafati
,
Z.
Abebe
, et al
.
2019
.
Health effects of dietary risks in 195 countries, 1990-2017: A systematic analysis for the global burden of disease study 2017
.
Lancet
.
393
:
1958
1972
.
Al Tameemi
,
W.
,
T.P.
Dale
,
R.M.K.
Al-Jumaily
, and
N.R.
Forsyth
.
2019
.
Hypoxia-modified cancer cell metabolism
.
Front. Cell Dev. Biol.
7
:
4
.
Andreatta
,
M.
, and
S.J.
Carmona
.
2021
.
UCell: Robust and scalable single-cell gene signature scoring
.
Comput. Struct. Biotechnol. J.
19
:
3796
3798
.
Bartolomaeus
,
H.
,
A.
Balogh
,
M.
Yakoub
,
S.
Homann
,
L.
Markó
,
S.
Höges
,
D.
Tsvetkov
,
A.
Krannich
,
S.
Wundersitz
,
E.G.
Avery
, et al
.
2019
.
Short-chain fatty acid propionate protects from hypertensive cardiovascular damage
.
Circulation
.
139
:
1407
1421
.
Berens
,
E.B.
,
J.M.
Holy
,
A.T.
Riegel
, and
A.
Wellstein
.
2015
.
A cancer cell spheroid assay to assess invasion in a 3D setting
.
J. Vis. Exp.
:
53409
.
Bergen
,
V.
,
M.
Lange
,
S.
Peidli
,
F.A.
Wolf
, and
F.J.
Theis
.
2020
.
Generalizing RNA velocity to transient cell states through dynamical modeling
.
Nat. Biotechnol.
38
:
1408
1414
.
Bindels
,
L.B.
,
P.
Porporato
,
E.M.
Dewulf
,
J.
Verrax
,
A.M.
Neyrinck
,
J.C.
Martin
,
K.P.
Scott
,
P.
Buc Calderon
,
O.
Feron
,
G.G.
Muccioli
, et al
.
2012
.
Gut microbiota-derived propionate reduces cancer cell proliferation in the liver
.
Br. J. Cancer
.
107
:
1337
1344
.
Blake
,
S.J.
,
Y.
Wolf
,
B.
Boursi
, and
D.J.
Lynn
.
2023
.
Role of the microbiota in response to and recovery from cancer therapy
.
Nat. Rev. Immunol.
24
:
308
325
.
Chen
,
S.
,
Y.
Zhou
,
Y.
Chen
, and
J.
Gu
.
2018
.
fastp: An ultra-fast all-in-one FASTQ preprocessor
.
Bioinformatics
.
34
:
i884
i890
.
Choi
,
H.
,
Y.
Lee
,
S.A.
Park
,
J.H.
Lee
,
J.
Park
,
J.H.
Park
,
H.K.
Lee
,
T.G.
Kim
,
S.S.
Jeun
, and
S.
Ahn
.
2022
.
Human allogenic γδ T cells kill patient-derived glioblastoma cells expressing high levels of DNAM-1 ligands
.
OncoImmunology
.
11
:
2138152
.
Comba
,
A.
,
S.M.
Faisal
,
P.J.
Dunn
,
A.E.
Argento
,
T.C.
Hollon
,
W.N.
Al-Holou
,
M.L.
Varela
,
D.B.
Zamler
,
G.L.
Quass
,
P.F.
Apostolides
, et al
.
2022
.
Spatiotemporal analysis of glioma heterogeneity reveals COL1A1 as an actionable target to disrupt tumor progression
.
Nat. Commun.
13
:
3606
.
Coutzac
,
C.
,
J.-M.
Jouniaux
,
A.
Paci
,
J.
Schmidt
,
D.
Mallardo
,
A.
Seck
,
V.
Asvatourian
,
L.
Cassard
,
P.
Saulnier
,
L.
Lacroix
, et al
.
2020
.
Systemic short chain fatty acids limit antitumor effect of CTLA-4 blockade in hosts with cancer
.
Nat. Commun.
11
:
2168
.
De Filippo
,
C.
,
D.
Cavalieri
,
M.
Di Paola
,
M.
Ramazzotti
,
J.B.
Poullet
,
S.
Massart
,
S.
Collini
,
G.
Pieraccini
, and
P.
Lionetti
.
2010
.
Impact of diet in shaping gut microbiota revealed by a comparative study in children from Europe and rural Africa
.
Proc. Natl. Acad. Sci. USA
.
107
:
14691
14696
.
Deleu
,
S.
,
K.
Machiels
,
J.
Raes
,
K.
Verbeke
, and
S.
Vermeire
.
2021
.
Short chain fatty acids and its producing organisms: An overlooked therapy for IBD?
eBioMedicine
66
:
103293
.
Derynck
,
R.
,
S.J.
Turley
, and
R.J.
Akhurst
.
2021
.
TGFβ biology in cancer progression and immunotherapy
.
Nat. Rev. Clin. Oncol.
18
:
9
34
.
Dono
,
A.
,
A.
Patrizz
,
R.M.
McCormack
,
N.
Putluri
,
B.P.
Ganesh
,
B.
Kaur
,
L.D.
McCullough
,
L.Y.
Ballester
, and
Y.
Esquenazi
.
2020
.
Glioma induced alterations in fecal short-chain fatty acids and neurotransmitters
.
CNS Oncol.
9
:
CNS57
.
El Tekle
,
G.
, and
W.S.
Garrett
.
2023
.
Bacteria in cancer initiation, promotion and progression
.
Nat. Rev. Cancer
.
23
:
600
618
.
Ferguson
,
J.F.
,
L.A.
Aden
,
N.R.
Barbaro
,
J.P.
Van Beusecum
,
L.
Xiao
,
A.J.
Simmons
,
C.
Warden
,
L.
Pasic
,
L.E.
Himmel
,
M.K.
Washington
, et al
.
2019
.
High dietary salt–induced DC activation underlies microbial dysbiosis-associated hypertension
.
JCI Insight
5
:e126241.
Fierro Morales
,
J.C.
,
Q.
Xue
, and
M.
Roh-Johnson
.
2022
.
An evolutionary and physiological perspective on cell-substrate adhesion machinery for cell migration
.
Front. Cell Dev. Biol.
10
:
943606
.
Gradisteanu Pircalabioru
,
G.
,
I.
Ilie
,
L.
Oprea
,
A.
Picu
,
L.M.
Petcu
,
L.
Burlibasa
,
M.C.
Chifiriuc
, and
M.
Musat
.
2022
.
Microbiome, mycobiome and related metabolites alterations in patients with metabolic syndrome-a pilot study
.
Metabolites
12
:
218
.
Hao
,
Y.
,
S.
Hao
,
E.
Andersen-Nissen
,
W.M.
Mauck
,
S.
Zheng
,
A.
Butler
,
M.J.
Lee
,
A.J.
Wilk
,
C.
Darby
,
M.
Zager
, et al
.
2021
.
Integrated analysis of multimodal single-cell data
.
Cell
.
184
:
3573
3587.e29
.
Jacob
,
F.
,
R.D.
Salinas
,
D.Y.
Zhang
,
P.T.T.
Nguyen
,
J.G.
Schnoll
,
S.Z.H.
Wong
,
R.
Thokala
,
S.
Sheikh
,
D.
Saxena
,
S.
Prokop
, et al
.
2020
.
A patient-derived glioblastoma organoid model and biobank recapitulates Inter- and intra-tumoral heterogeneity
.
Cell
.
180
:
188
204.e22
.
Kawaguchi
,
S.
,
K.
Kawahara
,
Y.
Fujiwara
,
K.
Ohnishi
,
C.
Pan
,
H.
Yano
,
A.
Hirosue
,
M.
Nagata
,
M.
Hirayama
,
J.
Sakata
, et al
.
2022
.
Naringenin potentiates anti-tumor immunity against oral cancer by inducing lymph node CD169-positive macrophage activation and cytotoxic T cell infiltration
.
Cancer Immunol. Immunother.
71
:
2127
2139
.
Khasraw
,
M.
,
Y.
Fujita
,
C.
Lee-Chang
,
I.V.
Balyasnikova
,
H.
Najem
, and
A.B.
Heimberger
.
2022
.
New approaches to glioblastoma
.
Annu. Rev. Med.
73
:
279
292
.
Kim
,
D.
,
J.M.
Paggi
,
C.
Park
,
C.
Bennett
, and
S.L.
Salzberg
.
2019
.
Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype
.
Nat. Biotechnol.
37
:
907
915
.
Kim
,
H.J.
,
J.H.
Park
,
H.C.
Kim
,
C.W.
Kim
,
I.
Kang
, and
H.K.
Lee
.
2022
.
Blood monocyte-derived CD169+ macrophages contribute to antitumor immunity against glioblastoma
.
Nat. Commun.
13
:
6211
.
Kim
,
I.K.
,
K.
Kim
,
E.
Lee
,
D.S.
Oh
,
C.S.
Park
,
S.
Park
,
J.M.
Yang
,
J.H.
Kim
,
H.S.
Kim
,
D.T.
Shima
, et al
.
2018
.
Sox7 promotes high-grade glioma by increasing VEGFR2-mediated vascular abnormality
.
J. Exp. Med.
215
:
963
983
.
La Manno
,
G.
,
R.
Soldatov
,
A.
Zeisel
,
E.
Braun
,
H.
Hochgerner
,
V.
Petukhov
,
K.
Lidschreiber
,
M.E.
Kastriti
,
P.
Lönnerberg
,
A.
Furlan
, et al
.
2018
.
RNA velocity of single cells
.
Nature
.
560
:
494
498
.
Liberzon
,
A.
,
C.
Birger
,
H.
Thorvaldsdóttir
,
M.
Ghandi
,
J.P.
Mesirov
, and
P.
Tamayo
.
2015
.
The molecular signatures database (MSigDB) hallmark gene set collection
.
Cell Syst.
1
:
417
425
.
Louis
,
P.
, and
H.J.
Flint
.
2017
.
Formation of propionate and butyrate by the human colonic microbiota
.
Environ. Microbiol.
19
:
29
41
.
Love
,
M.I.
,
W.
Huber
, and
S.
Anders
.
2014
.
Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2
.
Genome Biol.
15
:
550
.
Mallikarjuna
,
P.
,
T.S.
Raviprakash
,
K.
Aripaka
,
B.
Ljungberg
, and
M.
Landström
.
2019
.
Interactions between TGF-β type I receptor and hypoxia-inducible factor-α mediates a synergistic crosstalk leading to poor prognosis for patients with clear cell renal cell carcinoma
.
Cell Cycle
.
18
:
2141
2156
.
Masaki
,
Y.
,
Y.
Shimizu
,
T.
Yoshioka
,
F.
Feng
,
S.
Zhao
,
K.
Higashino
,
Y.
Numata
, and
Y.
Kuge
.
2016
.
Imaging mass spectrometry revealed the accumulation characteristics of the 2-nitroimidazole-based agent “pimonidazole” in hypoxia
.
PLoS One
.
11
:e0161639.
Matsushita
,
M.
,
K.
Fujita
,
T.
Hayashi
,
H.
Kayama
,
D.
Motooka
,
H.
Hase
,
K.
Jingushi
,
G.
Yamamichi
,
S.
Yumiba
,
E.
Tomiyama
, et al
.
2021
.
Gut microbiota-derived short-chain fatty acids promote prostate cancer growth via IGF1 signaling
.
Cancer Res.
81
:
4014
4026
.
Oh
,
D.S.
,
J.H.
Park
,
H.E.
Jung
,
H.-J.
Kim
, and
H.K.
Lee
.
2021
.
Autophagic protein ATG5 controls antiviral immunity via glycolytic reprogramming of dendritic cells against respiratory syncytial virus infection
.
Autophagy
.
17
:
2111
2127
.
Park
,
J.H.
,
C.W.
Kim
,
H.J.
Kim
,
H.J.
Kim
,
J.H.
Lee
, and
H.K.
Lee
.
2022
.
Protocol to analyze antitumor immunity of orthotopic injection and spontaneous murine high-grade glioma models using flow cytometry and single-cell RNA sequencing
.
STAR Protoc.
3
:
101607
.
Park
,
J.H.
,
H.-J.
Kim
,
C.W.
Kim
,
H.C.
Kim
,
Y.
Jung
,
H.-S.
Lee
,
Y.
Lee
,
Y.S.
Ju
,
J.E.
Oh
,
S.-H.
Park
, et al
.
2021
.
Tumor hypoxia represses γδ T cell-mediated antitumor immunity against brain tumors
.
Nat. Immunol.
22
:
336
346
.
Patrizz
,
A.
,
A.
Dono
,
S.
Zorofchian
,
G.
Hines
,
T.
Takayasu
,
N.
Husein
,
Y.
Otani
,
O.
Arevalo
,
H.A.
Choi
,
J.
Savarraj
, et al
.
2020
.
Glioma and temozolomide induced alterations in gut microbiome
.
Sci. Rep.
10
:
21002
.
Perler
,
B.K.
,
E.S.
Friedman
, and
G.D.
Wu
.
2023
.
The role of the gut microbiota in the relationship between diet and human health
.
Annu. Rev. Physiol.
85
:
449
468
.
Pertea
,
M.
,
G.M.
Pertea
,
C.M.
Antonescu
,
T.-C.
Chang
,
J.T.
Mendell
, and
S.L.
Salzberg
.
2015
.
StringTie enables improved reconstruction of a transcriptome from RNA-seq reads
.
Nat. Biotechnol.
33
:
290
295
.
Pietrobon
,
V.
, and
F.M.
Marincola
.
2021
.
Hypoxia and the phenomenon of immune exclusion
.
J. Transl. Med.
19
:
9
.
Reichardt
,
N.
,
S.H.
Duncan
,
P.
Young
,
A.
Belenguer
,
C.
McWilliam Leitch
,
K.P.
Scott
,
H.J.
Flint
, and
P.
Louis
.
2014
.
Phylogenetic distribution of three pathways for propionate production within the human gut microbiota
.
ISME J.
8
:
1323
1335
.
Ryu
,
T.Y.
,
K.
Kim
,
T.-S.
Han
,
M.-O.
Lee
,
J.
Lee
,
J.
Choi
,
K.B.
Jung
,
E.-J.
Jeong
,
D.M.
An
,
C.-R.
Jung
, et al
.
2022
.
Human gut-microbiome-derived propionate coordinates proteasomal degradation via HECTD2 upregulation to target EHMT2 in colorectal cancer
.
ISME J.
16
:
1205
1221
.
Schindelin
,
J.
,
I.
Arganda-Carreras
,
E.
Frise
,
V.
Kaynig
,
M.
Longair
,
T.
Pietzsch
,
S.
Preibisch
,
C.
Rueden
,
S.
Saalfeld
,
B.
Schmid
, et al
.
2012
.
Fiji: An open-source platform for biological-image analysis
.
Nat. Methods
.
9
:
676
682
.
Seystahl
,
K.
,
A.
Papachristodoulou
,
I.
Burghardt
,
H.
Schneider
,
K.
Hasenbach
,
M.
Janicot
,
P.
Roth
, and
M.
Weller
.
2017
.
Biological role and therapeutic targeting of TGF-β3 in glioblastoma
.
Mol. Cancer Ther.
16
:
1177
1186
.
Shi
,
R.
,
Z.
Zhang
,
A.
Zhu
,
X.
Xiong
,
J.
Zhang
,
J.
Xu
,
M.-S.
Sy
, and
C.
Li
.
2022
.
Targeting type I collagen for cancer treatment
.
Int. J. Cancer
.
151
:
665
683
.
Song
,
X.
,
H.
Gao
,
Y.
Lin
,
Y.
Yao
,
S.
Zhu
,
J.
Wang
,
Y.
Liu
,
X.
Yao
,
G.
Meng
,
N.
Shen
, et al
.
2014
.
Alterations in the microbiota drive interleukin-17C production from intestinal epithelial cells to promote tumorigenesis
.
Immunity
.
40
:
140
152
.
Spandidos
,
A.
,
X.
Wang
,
H.
Wang
, and
B.
Seed
.
2010
.
PrimerBank: A resource of human and mouse PCR primer pairs for gene expression detection and quantification
.
Nucleic Acids Res.
38
:
D792
D799
.
Suzuki
,
N.
,
Y.
Aiba
,
H.
Takeda
,
Y.
Fukumori
, and
Y.
Koga
.
2006
.
Superiority of 1-kestose, the smallest fructo-oligosaccharide, to a synthetic mixture of fructo-oligosaccharides in the selective stimulating activity on bifidobacteria
.
Biosci. Microflora
.
25
:
109
116
.
Takasaka
,
N.
,
R.I.
Seed
,
A.
Cormier
,
A.J.
Bondesson
,
J.
Lou
,
A.
Elattma
,
S.
Ito
,
H.
Yanagisawa
,
M.
Hashimoto
,
R.
Ma
, et al
.
2018
.
Integrin alphavbeta8-expressing tumor cells evade host immunity by regulating TGF-beta activation in immune cells
.
JCI Insight
3
:e122591.
Tam
,
S.Y.
,
V.W.C.
Wu
, and
H.K.W.
Law
.
2020
.
Hypoxia-induced epithelial-mesenchymal transition in cancers: HIF-1alpha and beyond
.
Front. Oncol.
10
:
486
.
van Solinge
,
T.S.
,
L.
Nieland
,
E.A.
Chiocca
, and
M.L.D.
Broekman
.
2022
.
Advances in local therapy for glioblastoma - taking the fight to the tumour
.
Nat. Rev. Neurol.
18
:
221
236
.
Varlamov
,
O.
2017
.
Western-style diet, sex steroids and metabolism
.
Biochim. Biophys. Acta Mol. Basis Dis.
1863
:
1147
1155
.
Wang
,
S.
,
X.
Lai
,
Y.
Deng
, and
Y.
Song
.
2020
.
Correlation between mouse age and human age in anti-tumor research: Significance and method establishment
.
Life Sci.
242
:
117242
.
WHO
.
2023
.
WHO Global Report on Sodium Intake Reduction
.
World Health Organization
,
Geneva, Switzerland
. https://iris.who.int/handle/10665/366393.
Wilck
,
N.
,
M.G.
Matus
,
S.M.
Kearney
,
S.W.
Olesen
,
K.
Forslund
,
H.
Bartolomaeus
,
S.
Haase
,
A.
Mähler
,
A.
Balogh
,
L.
Markó
, et al
.
2017
.
Salt-responsive gut commensal modulates TH17 axis and disease
.
Nature
.
551
:
585
589
.
Wu
,
Y.
,
S.S.
Hu
,
R.
Zhang
,
N.P.
Goplen
,
X.
Gao
,
H.
Narasimhan
,
A.
Shi
,
Y.
Chen
,
Y.
Li
,
C.
Zang
, et al
.
2023
.
Single cell RNA sequencing unravels mechanisms underlying senescence-like phenotypes of alveolar macrophages
.
iScience
.
26
:
107197
.
Yoon
,
S.H.
,
S.M.
Ha
,
S.
Kwon
,
J.
Lim
,
Y.
Kim
,
H.
Seo
, and
J.
Chun
.
2017
.
Introducing EzBioCloud: A taxonomically united database of 16S rRNA gene sequences and whole-genome assemblies
.
Int. J. Syst. Evol. Microbiol.
67
:
1613
1617
.
Yuan
,
H.
,
M.
Yan
,
G.
Zhang
,
W.
Liu
,
C.
Deng
,
G.
Liao
,
L.
Xu
,
T.
Luo
,
H.
Yan
,
Z.
Long
, et al
.
2019
.
CancerSEA: A cancer single-cell state atlas
.
Nucleic Acids Res.
47
:
D900
D908
.
Zhang
,
H.
,
H.O.
Akman
,
E.L.
Smith
,
J.
Zhao
,
J.E.
Murphy-Ullrich
, and
O.A.
Batuman
.
2003
.
Cellular response to hypoxia involves signaling via Smad proteins
.
Blood
.
101
:
2253
2260
.
Zhang
,
Y.
,
B.
Zhang
,
C.
Lv
,
N.
Zhang
,
K.
Xing
,
Z.
Wang
,
R.
Lv
,
M.
Yu
,
C.
Xu
, and
Y.
Wang
.
2023
.
Single-cell RNA sequencing identifies critical transcription factors of tumor cell invasion induced by hypoxia microenvironment in glioblastoma
.
Theranostics
.
13
:
3744
3760
.

Author notes

*

C.W. Kim and H.-J. Kim contributed equally to this paper.

Disclosures: The authors declare no competing interests exist.

This article is distributed under the terms as described at https://rupress.org/pages/terms102024/.