The Tumor Microbiome Reacts to Hypoxia and Can Influence Response to Radiation Treatment in Colorectal Cancer Open Access

Colorectal cancer is the third most common and second most lethal cancer worldwide, accounting for 1.9 million cases and nearly 900,000 deaths every year (1). More than 60% of all colorectal cancer cases occur in the sigmoid colon and are diagnosed at stage II or above (2). The 5-year survival rate for colorectal cancer typically ranges from 90% for patients diagnosed with localized disease to 14% for those diagnosed with metastatic disease (3). Clinical management of colorectal cancer depends on the stage: surgery is the primary treatment option in the early stages, whereas neoadjuvant chemoradiotherapy (nCRT) followed by surgery is the standard of care for locally advanced stages (4). However, only about 15% of patients with colorectal cancer treated with nCRT achieve complete pathologic responses (5). Because dose escalation dramatically increases the risk of toxicity and exceeds the radiotolerance of adjacent normal tissues, understanding the mechanisms behind colorectal cancer resistance to nCRT therapy is of utmost clinical importance.

The human intestinal microbiome comprises 10¹³ to 10¹⁴ mutualistic microorganisms that play a crucial role in shaping the intestinal epithelium, harvesting nutrients, directing maturation of the host immune system, regulating defense against pathogens, and maintaining gut barrier function (6–10). Increasing lines of evidence suggest that dysbiosis or alteration of the intestinal microbiome composition and function is increasingly involved with the initiation and progression of colorectal cancer (11). Moreover, emerging studies suggest a direct association between intestinal microbiome dysbiosis and sensitivity to anticancer therapy (12–15). In particular, studies in gnotobiotic mice have shown that intestinal microbiota can shape the response to anticancer treatment, suggesting complex host–microbiome cross-talk (16, 17).

The microenvironment of the human gastrointestinal (GI) tract is generally hypoxic (physiologic hypoxia), showing steep oxygen gradients along the radial axis ranging from well-oxygenated subepithelial mucosa to the anaerobic lumen (18, 19). Many anaerobic microbes that thrive under anoxic conditions inhabit the lumen. Microbial colonization of colorectal cancer tumors by microbial taxa is, therefore, highly influenced by the route by which they reach the tumor and by the ability of the microorganism to survive the environmental oxygen within the tumor microenvironment (TME). Because hypoxia directly limits the efficiency of radiotherapy (20), we hypothesized that cross-talk between the colorectal cancer tumor microbiome and the hypoxic TME influences the tumor response to radiotherapy.

In this study, we analyzed RNA sequencing (RNA-seq) data from 141 pretreatment colorectal cancer samples from patients receiving radiotherapy to identify environmental hypoxia– and radiation treatment–dependent variations in the tumor microbiome. We then assessed these patterns for their impact on overall survival. We performed an in vivo validation of these concepts in heterotopic model CT26 tumors to identify microbiome composition and gene expression as a function of tumor oxygenation. A comparison of host animals from BALB/c and athymic nude strains identified hypoxia-dependent microbial populations and adaptive gene-level responses to varying oxygen conditions.

The Oncology Research Information Exchange Network (ORIEN) is an alliance of 18 US cancer centers established in 2014. All ORIEN alliance members utilize a standard Total Cancer Care^® (TCC) protocol. As part of the TCC study, participants agree to have their clinical data followed over time, undergo germline and tumor sequencing, and be contacted by their provider if an appropriate clinical trial or other study becomes available (21). TCC is a prospective cohort study with a subset of patients enrolled in the ORIEN Avatar program, including research-use-only–grade whole-exome tumor sequencing, RNA-seq, germline sequencing, and deep longitudinal clinical data collection with lifetime follow-up. Nationally, more than 325,000 participants have enrolled in TCC. Aster Insights, ORIEN’s commercial and operational partner, harmonizes all abstracted clinical data elements and molecular sequencing files into a standardized, structured format to enable aggregation of de-identified data for sharing across the network. This study included 2,755 ORIEN Avatar patients diagnosed with melanoma, sarcoma, or thyroid, pancreatic, colorectal, or lung cancer who consented to the TCC protocol from the participating members of ORIEN. Of these, 500 patients had colon adenocarcinoma, and 95 had rectal adenocarcinoma (Table 1).

ORIEN Avatar specimens underwent nucleic acid extraction and sequencing at HudsonAlpha or Fulgent Genetics. Frozen and optimal cutting temperature tissue RNA extraction was performed using the Qiagen RNeasy plus mini kit, generating a 216 bp average insert size. For formalin-fixed, paraffin-embedded tissue, the Covaris Ultrasonication FFPE DNA/RNA kit was utilized to extract DNA and RNA, respectively, generating a 165 bp average insert size. RNA-seq was performed using the Illumina TruSeq RNA Exome with single library hybridization, cDNA synthesis, library preparation, and sequencing (100-bp paired reads at HudsonAlpha; 150-bp paired reads at Fulgent) to a coverage of 100 M total reads/50 M paired reads. RNA-seq tumor pipeline analysis was processed according to the workflow outlined in the following section using GRCh38/hg38 human genome reference sequencing and GenCode build version 32. Adapter sequences were trimmed from the raw tumor sequencing FASTQ file. Adapter trimming via k-mer matching was performed along with quality trimming and filtering, contaminant filtering, sequence masking, guanine–cytosine filtering, length filtering, and entropy filtering. The trimmed FASTQ file was used as input to the read alignment process. The tumor adapter–trimmed FASTQ file was aligned to the human genome reference (GRCh38/hg38) and the GENCODE genome annotation v32 using the STAR aligner. The STAR aligner generates multiple output files for gene fusion prediction and expression analysis. RNA expression values were calculated and reported using estimated mapped reads, fragments per kilobase of transcript per million mapped reads, and transcripts per million mapped reads at both the transcript and gene levels based on transcriptome alignment generated using STAR. We used PowerFecal pro kits from Qiagen to extract RNA and DNA for model murine tumors.

Microbes were classified in human and mouse samples using the {exotic} pipeline. This entails removing as many human or mouse reads as possible by aligning to the appropriate genome using STAR. The hg38 genome was used for human samples, the GRCm39 genome was used for mouse samples, and annotated genes were further used in gene expression analyses. Reads left unclassified after STAR alignment were then aligned to a Kraken 2 database customized to include fungi and archaea. Bayesian re-estimation of abundance with Kraken was used to determine the likely species of all classified reads. The R package {exotic} (22) was then used to decontaminate the results and to normalize the human samples to account for processing at different sequencing centers.

Hypoxia scores were generated using the R package {tmesig} (23) for Buffa, Winter, and Leonard signatures. The R package {mt.surv} was used to compare these scores’ relationships with overall survival, and Buffa (24) was chosen to represent hypoxia in further analyses. For all analyses comparing low and high hypoxia scores, samples were defined as having low hypoxia if their hypoxia score was in the lower tertile of the data and high hypoxia if their score was in the upper tertile.

Kaplan–Meier survival curves were generated, and Cox proportional hazards models were applied using the R package {survival} and visualized using {ggsurv}.

The R package {DESeq2} was used to compare the abundance of microbes in low- and high-hypoxia tumors for both The Cancer Genome Atlas (TCGA) and ORIEN data. It was also used to compare microbe abundances in low- and high-hypoxia tumors in the mouse experiments and the behavior of microbe gene expressions in high- and low-hypoxia mouse tumors. Demographic information for this population was similar to the ORIEN dataset (Table 2).

The bacterial 16S rRNA gene was amplified from fresh frozen tumor (n = 31) and adjacent normal (n = 31) tissues from 31 patients. Microbial and tumor cells were lysed on a PowerLyzer 24 at 2,000 rpm for 30 seconds, and then DNA was purified using an AllPrep mini kit (Qiagen). The bacterial rDNA was amplified using V3 to V4 primers and KAPA HiFi enzyme (50°C, 30 seconds; 72°C, 2 × 20 cycles). Magnetic beads cleaned amplicons, and sequencing libraries were generated using a QIAseq kit (Qiagen) following the manufacturer’s instructions. Libraries were sequenced on a MiSeq 2 × 300 (600 cycles) using a V3 reagent kit (Illumina). Demultiplexed FASTQs were filtered for quality and length (340–440 bp), and processed to amplicon sequence variant using {dada2}. Taxonomy was assigned with {exotic} using the same database as for the tumor RNA-seq.

MC38 (RRID: CVCL_B288) and CT26 cells (RRID: CVCL_7254) were obtained from ATCC. Female 7-week-old athymic nude (RRID: IMSR_CRL:088) and BALB/c (RRID: IMSR_APB:4790) mice were purchased from Charles River (housed in groups of five) and were injected subcutaneously into the right flank with either 1 million MC38 or 0.5 million CT26 cells with tumor growth measured using calipers using the formula (L × W²)/2. Radiotherapy was delivered by a single tangential beam from the small animal radiation research platform. Untreated tumors were typically harvested on day 16 after inoculation when most tumors were at removal criteria. Mice stool was collected before tumor harvesting. All animal experiments were done in accordance with the approved Ohio State University Institutional Animal Care and Use Committee protocol (protocol # 2012A00000124-R3). Cells were grown in DMEM containing 25 mmol/L D-glucose, 4 mmol/L glutamine, and 44 mmol/L sodium bicarbonate in 5% CO₂. All cell lines have been authenticated using STR profiling within the past 3 years. All experiments were performed with Mycoplasma-free cells, as determined by routine PCR testing in 3 to 4-month intervals.

Representative bacterial strains Akkermansia muciniphila (ATCC^® BAA-835), Lactobacillus acidophilus (ATCC^® 4356), Streptococcus thermophilus (ATCC^® 19258), Bacteroides ovatus (ATCC^® 8483), and Dorea formicigenerans (ATCC^® 27755) were obtained from ATCC. Fecal bacteria isolates were collected from three healthy adult volunteers as described previously (protocol approval was obtained from the Institutional Review Board, and written informed consent was documented for each individual; ref. 25). All bacterial strains were maintained under standard culture conditions as previously described (26). Briefly, overnight bacterial cultures were established in Gifu anaerobic medium (GAM; HiMedia). The cultures were maintained in an anaerobic environment at 37°C using a Coy Laboratory anaerobic chamber (Coy Lab). To estimate growth, the final optical density of bacterial cultures was measured using an ELx808 absorbance plate reader (BioTek). Drop plate analysis was also used to quantify the colony-forming units in each culture. Before injections, bacterial cultures were centrifuged and then resuspended at a final concentration of 3 × 10⁹ colony-forming units per injection.

RNA-seq data from the mouse tumors were classified using HUMAnN 3.0 with MetaPhlAn and the default ChocoPhlAn database to obtain information on the gene expression of microbes in the mouse experiments.

The Ohio State University Institutional Review Board approved data access in an Honest Broker protocol (2015H0185) and TCC protocol (2013H0199) in coordination with Aster Insights. FASTQs for mouse RNA-seq are available in two batches published to Zenodo at 10.5281/zenodo.11211908 and https://doi.org/10.5281/zenodo.11212525. Analysis scripts and data to regenerate all figures and tables are available at: https://github.com/spakowiczlab/exorien-recrad.

Virtually all solid tumor types are characterized by some level of hypoxia. One way to approximate tumor hypoxia is to use quantitative gene expression signatures of hypoxia-regulated genes. For Fig. 1A , we calculated the values and distribution of the hypoxia expression score (24) across cancer types within the ORIEN dataset (24). This hypoxia signature is similar to others and scores sets of genes induced in response to hypoxia primarily by the hypoxia-inducible factor 1 (HIF1) transcription factor. This gene family can be extracted from tumor mRNA data in publicly available clinical datasets to identify tumors with significant levels of hypoxia. In colorectal cancers, we did not find a significant association between hypoxia and patient characteristics such as tumor stage or sex (Supplementary Fig. S1). We also see no differences in hypoxia correlating with age, gender, body mass index (BMI), stage, race, or when subdividing rectal adenocarcinoma versus colon adenocarcinoma (Table 1).

We processed RNA-seq data from 141 patients with colorectal cancer available through the ORIEN network and calculated hypoxia scores (24) for each sample. Colorectal adenocarcinoma (COADREAD) ranks among the more hypoxic cancer types with a median hypoxia score of 10 (Fig. 1A). Comparing cancers from different sites, the distribution of hypoxia scores ranges from lowest in thyroid cancer (median hypoxia expression score of −16) to highest in lung squamous cell cancer (median hypoxia expression score of 26). We then cross-validated the ORIEN COADREAD hypoxia scores by comparing them with the publicly available TCGA COADREAD dataset. Demographic information for TCGA patients is similar to that of the ORIEN patients and is available in Table 2. We found substantial overlap in the distribution of scores between the independent patient datasets, confirming the robustness of this analysis and indicating that the many COADREAD samples were distributed within the high-end hypoxia expression score region (Fig. 1B). When we compare the average hypoxia scores in the normal colon [from the Genotype-Tissue Expression (GTEx) database of normal tissues] with the scores in the adjacent normal colon or rectum in TCGA database with the colon or rectal cancer, we can see a significant increase in the cancerous samples, and very similar distribution in the colon and rectal cancers (Fig. 1C). Current molecular models of radiation-induced DNA damage specify that oxygen is needed for the “fixation” of damage; thus, hypoxia directly impairs the efficiency of radiotherapy (27). Therefore, we tested whether elevated hypoxia scores would correlate with poor survival of patients with colorectal cancer treated with radiotherapy. We stratified ORIEN patients who underwent radiotherapy at the upper and lower tertiles of the hypoxia expression score, and a Cox proportional hazards model shows that a high hypoxic score significantly correlates with poor overall patient survival (P = 0.028; Fig. 1D; numbers at risk in Supplementary Table S1). Demographic variables were not significantly different between these groups (Supplementary Table S2). To ensure that the association with survival was robust to where we selected the “high” versus “low” stratification and to identify the best expression signature for hypoxia for this dataset, we calculated the hypoxia scores for three published signatures (3–5) and then stratified the data at all possible thresholds as a percentile of the distribution (Fig. 1E). The hypoxia expression score by Buffa and colleagues (24) significantly associated with survival over a wide range of stratification thresholds. By testing all possible thresholds, we created a curve and then integrated the area of P values below a 0.05 threshold. The hypoxia score from Buffa and colleagues showed the greatest area and was used for all further analyses (Fig. 1E, inset).

Significant evidence suggests that the composition of the gut microbiome correlates with the initiation and progression of colorectal cancer and affects tumor sensitivity to anticancer therapy (15). Accordingly, we evaluated tumor microbiota composition in ORIEN and TCGA patient samples using the exogenous sequences in tumors and immune cells {exotic} tool for quantifying microbe abundances in tumor RNA-seq data (22). To identify taxa associated with high levels of tumor hypoxia, we stratified patient samples into tertiles of increasing hypoxia scores (24) and quantified log₂ abundance fold change between the bottom and top ranked tertiles (Fig. 2A). This analysis was performed on both ORIEN and TCGA colorectal cancer datasets, and several microbes were concordant in both datasets (significant and in the same direction; Fig. 2B; all fold changes are available in Supplementary Table S3). Interestingly, Fusobacterium nucleatum has been associated with poor outcomes in patients with colorectal cancer (15, 16) and is elevated in hypoxic tumors (Fig. 2A and B). Hypoxic tumors also contained significantly more M0-type macrophages, resting NK cells, memory B cells, CD4 T cells (regulatory T cells and resting memory), and activated dendritic cells, estimated by a deconvolution approach (Supplementary Fig. S2; Supplementary Table S4). Amplicon-based sequencing of a subset of the ORIEN samples confirmed the presence of Fusobacterium, which had a higher median abundance in hypoxic tumors, though the difference between the distributions was not statistically significant (Supplementary Fig. S3).

We next analyzed the intratumoral microbial populations in these patients for strains that correlated with the outcome of radiotherapy for colorectal cancer. We identified 11 microbes in which presence correlated with significantly worse outcomes after radiotherapy (Fig. 3A; complete results in Supplementary Table S5). Strains showing the strongest statistical correlation included Candida glabrata, a fungus, and Fusobacterium canifelinum and Bulleidia sp. zg 1006, both bacteria. F. canifelinum is an anaerobic, Gram-negative bacilli that is not generally present in a healthy human gut microbiome but typically occurs in the oral cavity (28). The closely related F. nucleatum has been identified in recurrent colorectal cancer, and the presence of the Fusobacterium strain promotes colorectal cancer development (29, 30) and chemoresistance (16). We, therefore, asked whether the combined presence of hypoxia and F. canifelinum affected treatment outcomes in patients with colorectal cancer undergoing radiotherapy. Analysis of patients with colorectal cancer treated with radiotherapy identified that the presence of F. canifelinum or a high hypoxia score (24) is significantly associated with reduced overall survival (P value F. canifelinum = 0.005; Fig. 3B; numbers at risk in Supplementary Table S6).

To experimentally validate our bioinformatic observations in vivo, athymic nude mice were inoculated heterotopically with the mouse colorectal cancer line MC38. Upon reaching 100 mm³, MC38 tumor–bearing mice were randomized into groups receiving either nothing as control, intratumoral bacterial growth medium (GAM), or cultured F. nucleatum grown in GAM to achieve a ratio of 1:1 bacteria to tumor cells. The 1:1 ratio was chosen after control experiments indicated it as a dose that did not elicit an acute inflammatory response. After 48 hours, mice were either sham irradiated or given one dose of 8 Gy external beam radiation using the small animal radiation research platform. We observed no significant change in tumor response to therapy with the introduction of F. nucleatum (Fig. 3C).

We, therefore, decided to test other bacterial strains for effects on radiation response and repeated this experiment using several common gut bacterial strains: (i) microbial coculture representative of a healthy intestinal microbiome (gut mix); (ii) Lactobacillus spp.; (iii) Streptococcus spp.; or (iv) Akkermansia spp., also by intratumoral injection at a 1:1 ratio of microbe to tumor cell, followed 48 to 72 hours later by an 8-Gy dose of local radiation. Figure 3D shows that the injection of Lactobacillus or Streptococcus bacterial strains into heterotopic colorectal cancer tumors resulted in a pronounced increase in tumor growth delay after radiation. Taken together, these findings support the idea that intratumoral microbes can influence the response of model tumors to radiotherapy in a strain-dependent manner.

We next investigated the impact of a hypoxic TME on spontaneous tumor microbial colonization in model tumors in mice. We inoculated a cohort of 10 BALB/c and 10 athymic nude host mice with the CT26 murine colorectal cancer cell line. The mice received 0.5 × 10⁶ CT26 cells subcutaneously in the flank, and their tumors were harvested upon reaching 500 mm³ for nucleic acid extraction. Hypoxia scores (24) were calculated from tumor RNA for each sample, and the tumors were stratified at the mean hypoxia score (24) into normoxic (low hypoxia expression score) and hypoxic (high hypoxia expression score) tumors in each mouse host (Fig. 4A). We did not detect strong differences in estimated immune populations across both datasets (Supplementary Fig. S4; Supplementary Table S7), though deconvolution predicted differences in the number of tumor-infiltrating lymphocytes in the nude mice (Supplementary Table S8). We used metatranscriptomic sequencing of exogenous RNA to identify microbial strains enriched in hypoxic tumors. In the immune-competent model, the bacterial burden was equivalent in both low- and high-hypoxia tumors. However, in the tumors grown in nude mice, the burden was higher in hypoxic tumors (Fig. 4B). Nude mice showed a larger fraction of anaerobic strains in the Firmicutes phylum (Fig. 4C). Microbes that were significantly enriched in the normoxic and hypoxic tumors for each host were identified (Fig. 4D; complete results in Supplementary Table S9). To better understand the differences in the two host strains of mice, we collected stool samples for analysis of the gut microbiome by 16S amplicon sequencing. A comparison of individuals in this experiment by Bray–Curtis distance revealed that the immune-competent host samples were most similar, whereas the immune-deficient hosts were highly variable (Fig. 4E). However, very similar taxa were observed in all mice; approximately 80% of the 16S sequences were common to both hosts, whereas 10% were unique to each of the BALB/c and athymic nude host animals (Fig. 4F). Interestingly, there was an overlap of three of the strains found preferentially in the hypoxic tumors in both humans and mice (Fig. 4G; complete overlap between human and mouse data in Supplementary Table S10). The amount of Pichia significantly correlated with the estimated number of tumor monocytes (Kruskal–Wallis, FDR-corrected P value 0.03; all correlation results in Supplementary Table S11).

However, the most consistent colonization of all tumors was observed with the genus Cutibacterium from the phylum Actinobacteria (Fig. 4D). Cutibacterium is a Gram-positive anaerobic bacillus typically found in the skin and associated with acne vulgaris; however, emerging studies identified a possible role for the microbe in prostate cancer immunosuppression (31). Based on the prevalence of Cutibacterium in all analyzed samples, we performed transcriptomic analysis of this species in both the BALB/c and nude mouse hosts. We analyzed the mechanisms by which the same microbial strain responds to this TME variable (tumor oxygenation as determined by the hypoxia expression score) in different mouse hosts. Differential microbial RNA analysis revealed significant differences between microbial gene expression in normoxic versus hypoxic tumors, indicating microbial adaptation to TME changes (Fig. 5A; complete list in Supplementary Table S12). Of the 1,295 identified RNA species, we found 48 significantly different RNA species in the normoxic versus hypoxic tumors in both datasets (39 induced by hypoxia and 9 repressed).

Even with the paucity of annotation in this strain, we found that manual curation identifies adaptive changes in the genes that are primarily involved in metabolism but also several that are involved in oxygen/redox stress, DNA metabolism, and several transcriptional regulators (Fig. 5B). Hypoxic induction of lactate dehydrogenase or citrate synthase may be analogous to what is seen in higher species adapting their metabolism to reduced oxygen environments. The 34 with UniRef90 annotations are listed in Fig. 5C in order of significance. One of the more interesting induced transcriptional regulators is the WhiB family member, which contains a redox-sensitive [4Fe-4S] cluster. The WhiB‐like protein family is exclusively found in Actinobacteria and has been reported to be nitric oxide–responsive and oxygen-responsive in other strains (32). This transcriptional regulator may be analogous to the hypoxia-responsive HIF1 transcriptional regulator found in higher species that is largely responsible for transcriptional adaptation to hypoxia. The sum of these findings indicates that intratumoral oxygen tension can both favor the growth of specific microbial populations and elicit adaptive transcriptional responses in microbes that can grow in both normoxic and hypoxic tumors.

Several lines of evidence show that host–microbiome interactions can affect the response to anticancer therapy (14, 33–35). The balance between bacterial and fungal microbes and the metabolites produced by these microbes can affect response to radiotherapy (36, 37). Chronic intestinal inflammation associated with microbial dysbiosis and an intrinsically hypoxic microenvironment creates a niche permissive to microbial colonization of all ranges of partial oxygen pressure (pO₂) within the tumor. In line with this expectation, we observed that tumor microbiome composition shows distinct patterns as a function of tumor oxygen levels in datasets of patients with colorectal cancer.

Under normal conditions, the intestinal microbiome consists of 1,000+ species of bacteria, most of which are obligate anaerobes (38). Intestinal dysbiosis triggers the rapid expansion of conditional anaerobic pathogens that can disrupt the intestinal mechanical barrier and result in luminal microbes colonizing the tumor. Mean pO₂ levels within the GI tract range from 42 to 71 mm Hg (7%–10% O₂) in the colonic muscle wall to 3 mm Hg (0.4% O₂) in the sigmoid colon, similar to 2.5 mm Hg levels typically observed in colorectal cancer tumors (39, 40). Distribution characteristics of conditional pathogens in both mouse hosts show an overrepresentation of the Firmicutes phylum, similar to observations made in human colorectal cancer (41).

Much investigation has gone into the composition of the intratumoral immune system to establish its role in determining the microbes that can colonize a tumor (42). However, there are fewer studies of the metabolic parameters influence the microbes that find a tumor hospitable. The partial pressure of oxygen has been long recognized as an important component of culture media for different microorganisms. Bacteria have been found that range from obligate anaerobes that can only grow in the absence of oxygen to aerobes that grow well in an atmosphere with 21% oxygen. Often, this categorization is due to the presence or absence of enzymes within the microbial genome that are capable of metabolizing toxic oxygen species, such as superoxide dismutase or catalase, which can facilitate growth in an oxidizing environment. Many microbes fall between these extremes and can be categorized as facultative anaerobes or microaerophiles capable of growth in low- or no-oxygen environments. These microbes may be most well adapted for growth in intermediate to low oxygen, which is often found within the TME.

Similarly, although much attention has been given to strains of bacteria that exist within the tumor, more study still needs to be completed on the physiologic state of intratumoral microbes. Microbes could have a very different impact on the tumor due to the expression or silencing of different genes. For example, hypoxic induction of lactate dehydrogenase or citrate synthase in Cutibacterium may be analogous to that is seen in higher species adapting their metabolism to reduced oxygen environments. In addition, one of the more interesting induced genes is the transcriptional regulators in the WhiB family, which contains a redox-sensitive (4Fe-4S) cluster. The WhiB‐like protein family is exclusively found in Actinobacteria and has been reported to be nitric oxide–responsive and oxygen-responsive in other strains (32). This transcriptional regulator may be analogous to the hypoxia-responsive HIF1 transcriptional regulator found in higher species that is largely responsible for transcriptional adaptation to hypoxia. Our data suggest that adaptation to Cutibacterium hypoxia could result in higher levels of microbially produced lactate or citrate. Such metabolites may interact with and change the behavior or morphology of the tumor. For example, tumor lactate has been shown to impact macrophage polarization (43). Such microbe signals have also been shown to impact tumor cell response to radiotherapy (37).

Our study also observed several microbial strains correlated with worse outcomes from radiotherapy in rectal cancer. Notably, we identified C. glabrata and F. canifelinum, which share close taxonomic similarity to microbes previously related to cancer (Candida albicans and F. nucleatum, respectively). C. albicans has been linked to oral squamous cell carcinoma (44) and radioresistant rectal cancer (45) and was shown in a large study to be predictive of metastatic disease across multiple GI tumors (46). Additionally, previous studies have reported that species from the Fusobacterium genus are associated with elevated inflammatory response in the colorectal mucosa (47–49). This raises the question of whether the host’s immune response to microbial dysbiosis affects radiotherapy outcomes.

Experimentally, we observed that immunocompromised BALB/c and nude mice hosts show distinct acute immune responses following intratumoral inoculation with several microbial strains found in human colorectal cancer (50). The inoculated mice exhibited mild symptoms of acute immune response associated with changes in tumor growth and response to radiotherapy. We identified that the addition of common commensal bacteria Lactobacillus acidophilus and Streptococcus thermophilus in the tumor enhances the therapeutic efficacy of local radiotherapy. At the same time, inoculation with F. nucleatum or microbial coculture, representative of a healthy intestinal microbiome, does not potentiate radiotherapy. These observations support the model that the presence of Lactobacillus or Streptococcus genera might predict better radiotherapy outcomes in patients with colorectal cancer. However, we note that the 16S-based validation of microbes from additional tumors did not observe either of these microbes. We expect that the well-established differences between the mouse and human gut microbiome could provide a rationale for altered tumor microbiomes (6, 7). In comparison, the presence of Fusobacterium predicts worse outcomes {F. nucleatum in previous publications [e.g., (8)] and F. canifelinum in our study}. Notably, the presence of this organism in tumors has been confirmed by many methods, including 16S-based sequencing in this study.

Perhaps the most intriguing findings come from the metatranscriptomic analysis of conditional pathogen enrichment in hypoxic versus normoxic CT26 colorectal cancer tumors in BALB/c and nude mice. We identified that Cutibacterium, a typically anaerobic opportunistic pathogen, most consistently colonizes both the normoxic and hypoxic tumors, displaying broad adaptive flexibility toward different pO₂ environments. Transcriptomic analysis revealed a systemic reprogramming underlying the adaptive response of the pathogen to varying pO₂ levels. These results support the model that opportunistic pathogens colonizing colorectal cancer tumors can respond to environmental oxygen levels with adaptive gene expression changes, providing a solid rationale for investigating microbial metabolism in future studies. These results also indicate that it may not be just the presence or absence of a specific microbe within a tumor that dictates its effect on clinical outcomes but the physiologic state of the microbe that can be influenced by the microenvironment in which it resides.

All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Y. Zakharia reports that the following disclosures are not related to the current work: advisory board: Bristol Myers Squibb, Amgen, Roche Diagnostics, Novartis, Janssen, Eisai, Exelixis, Castle Biosciences, Genzyme Corporation, AstraZeneca, Array, Bayer, Pfizer, Clovis, EMD Serono, and Myovant; grant/research support from: Institution clinical trial support from NewLink Genetics, Pfizer, Exelixis, and Eisai; and DSMC: Janssen Research and Development Consultant honorarium: Pfizer and Novartis. C.M. Ulrich reports grants from NIH during the conduct of the study, as well as grants from NIH and personal fees from the Cancer Center Adv. Board outside the submitted work. A.A. Tarhini reports grants from Bristol Myers Squibb, Merck, Sanofi, Regeneron, Iovance/Clinigen, Scholar Rock, OncoSec, Genentech-Roche, Pfizer, Acrotech, and Agenus and personal fees from Bristol Myers Squibb, Merck, Genentech/Roche, Novartis, Sanofi, Regeneron, Partner Therapeutics, Bayer, Eisai, Instil Bio, ConcertAI, and BioNTech outside the submitted work. G. Tinoco reports being a part of Deciphera Pharmaceuticals advisory board and provided feedback about vimseltinib, a drug used for the treatment of TGCT, and discussed future clinical trials. The activity lasted 4 hours in total on April 27, 2024, with payment <3,000 USD; being a part of Daiichi Sankyo Company advisory board and provided his opinion about one of their drugs pexidartinib, provided feedback about the new dose of pexidartinib, discussed the real-world implications of the new dose, and provided insights about the new label and strategies for compliance. The activity lasted 5 hours in total on November 16, 2022, with payment <3,000 USD; and being a part of Servier Pharmaceuticals (added June 24, 2022) advisory board and provided feedback on the design of a clinical trial with ivosidenib. The activity lasted 4 hours with payment <3,000 USD. G. Riedlinger reports grants from ORIEN during the conduct of the study, as well as personal fees from AstraZeneca outside the submitted work. B.P. Schneider reports other from Genentech, Pfizer, and Foundation Medicine outside the submitted work. D. Spakowicz reports grants from Lung Cancer Foundation of America, National Institute on Aging, American Lung Association, Pelotonia, National Comprehensive Cancer Center Network, Oncology Nursing Foundation, Sarcoma Foundation of America, NCI, and American Cancer Society outside the submitted work. No disclosures were reported by the other authors.

M. Benej: Data curation, formal analysis, investigation, visualization, writing-original draft, writing-review and editing. R. Hoyd: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-review and editing. M. Kreamer: Data curation, investigation, visualization, writing-review and editing. C.E. Wheeler: Data curation, software, formal analysis, supervision, visualization, methodology, writing-original draft, writing-review and editing. D.J. Grencewicz: Writing-review and editing. F. Choueiry: Data curation, formal analysis, investigation, writing-review and editing. C.H.F. Chan: Resources, writing-review and editing. Y. Zakharia: Resources, writing-review and editing. Q. Ma: Resources, writing-review and editing. R.D. Dodd: Resources, writing-review and editing. C.M. Ulrich: Resources, writing-review and editing. S. Hardikar: Resources, writing-review and editing. M.L. Churchman: Resources, writing-review and editing. A.A. Tarhini: Resources, writing-review and editing. L.A. Robinson: Resources, writing-review and editing. E.A. Singer: Resources, writing-review and editing. A.P. Ikeguchi: Resources, writing-review and editing. M.D. McCarter: Resources, writing-review and editing. G. Tinoco: Writing-review and editing. M. Husain: Writing-review and editing. N. Jin: Writing-review and editing. A.C. Tan: Resources, writing-review and editing. A.E.G. Osman: Resources. I. Eljilany: Writing-review and editing. G. Riedlinger: Resources, writing-review and editing. B.P. Schneider: Resources, writing-review and editing. K. Benejova: Investigation, writing-review and editing. M. Kery: Investigation, writing-review and editing. I. Papandreou: Writing-review and editing. J. Zhu: Resources, data curation, formal analysis, supervision, investigation, writing-review and editing. N. Denko: Conceptualization, writing-original draft, project administration, writing-review and editing. D. Spakowicz: Conceptualization, resources, software, methodology, writing-original draft, project administration, writing-review and editing.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the PIIO.

This project was partly supported by The Ohio State University Comprehensive Cancer Center and the NIH (P30CA016058); The Ohio State University Center for Clinical and Translational Science and the National Center for Advancing Translational Sciences (8UL1TR000090-05); an Alpha Omega Alpha Carolyn L. Kuckein Student Research Fellowship (D.J. Grencewicz); and a Samuel J. Roessler Memorial Scholarship (D.J. Grencewicz). This work was also supported by the Pelotonia Institute of Immuno-Oncology. The authors acknowledge the support and resources of the Ohio Supercomputer Center (PAS1695 and PCON0005). The results published here are in whole or part based upon data generated by TCGA Research Network: https://www.cancer.gov/tcga. The GTEx Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health and by National Cancer Institure, National Health Genome Research Institute, National Heart, Lung, and Blood Institute, National Institute on Drug Abuse, National Institute of Mental Health, and National Institute of Neurological Disorders and Stroke. The data used for the analyses described in this article were obtained from the GTEx portal on June 12, 2023. We would like to thank Angela Dahlberg, Editor, Division of Medical Oncology at The Ohio State University Comprehensive Cancer Center, for editing and proofreading the manuscript.