Dietary Impacts on Changes in Diversity and Abundance of the Murine Microbiome during Progression and Treatment of Cancer

The intestinal microbial population is recognized for its impact on cancer treatment outcomes. Little research has reported microbiome changes during cancer progression or the interplay of disease progression, dietary sugar/fat intake, and the microbiome through surgery and chemotherapy. In this study, the murine gut microbiome was used as a model system, and changes in microbiome diversity, richness, and evenness over the progression of the cancer and treatment were analyzed. Mice were categorized into four diet cohorts, combinations of either high or low sucrose and high or low omega-3 fatty acids, and two treatment cohorts, saline vehicle or chemotherapy, for a total of eight groups. Fecal samples were collected at specific timepoints to assess changes due to diet implementation, onset of cancer, lumpectomy, and chemotherapy. Akkermansia muciniphila abundance was very high in some samples and negatively correlated with overall Amplicon Sequence Variant (ASV) richness (r(64) = −0.55, p = 3 × 10−8). Throughout the disease progression, ASV richness significantly decreased and was impacted by diet and treatment. Alpha-diversity and differential microbial abundance were significantly affected by disease progression, diet, treatment, and their interactions. These findings help establish a baseline for understanding how cancer progression, dietary macronutrients, and specific treatments impact the murine microbiome, which may influence outcomes.


Introduction
The prevalence of cancer remains high [1]. Currently, it is expected that there will be 1,918,030 incident cases of cancer and 609,360 cancer deaths, this year [1]. Breast cancer is considered "fairly common" by the National Institute of Health [2], and is anticipated to account for 15% (287,850) of those new cases and 7.1% (43,250) of cancer deaths [1]. Women have a 13% lifetime chance of developing breast cancer, requiring them to receive therapeutic treatment [2].
An important aspect of achieving the best cancer treatment outcomes is maintaining patient health against infectious disease and opportunistic pathogens throughout treatment. When patients receive any systemic treatment against breast cancer (e.g., chemotherapy or chemo-radiation), a surgery, or a combination of treatments, their immune system becomes depressed [3]. These therapies target rapidly dividing cells, such as aggressive cancer cells, and many cells which are vital to host immunity (red blood cells, lymphocytes, macrophages, and neutrophils) [4], leading to patients' increased susceptibility to infections [5] including, among others, foodborne pathogens such as Salmonella spp., STEC (Shiga-toxin producing E. coli), and Listeria monocytogenes [5,6]. As both the treatment and disease progress, maintaining gastrointestinal health becomes increasingly difficult [7]. In

Experimental Design
Mice were acclimated to facilities for approximately two weeks, then underwent an ovariectomy to mimic the postmenopausal state common to the majority of breast cancer patients [30]. Following ovariectomy, mice were weighed and housed individually. After recovering from ovariectomy surgery for one week, the semi-purified diets were introduced ( Figure S1).
One week following initiation of semi-purified diets, all mice were inoculated with 100 µL of 1 × 10 5 E0771 murine mammary cancer cells into right abdominal (4th) mammary fat pad. Twelve days after inoculation with cancer cells, all mice underwent lumpectomy in which the intact tumor was excised along with any visible fat pad. Eight days following the lumpectomy procedure, body weight assessment was performed again. Ten days following lumpectomy, mice from each dietary cohort were treated with either a saline vehicle injection or a chemotherapy injection (n = 16 per chemo/diet group and n = 13 per vehicle/diet group). Chemotherapy composition was 9 mg/kg body weight of doxorubicin and 90 mg/kg body weight of cyclophosphamide (50% of human equivalent dose). The experiment ended early, one week after either saline vehicle or chemotherapy injection, because tumor regrowth was detected. All mice were euthanized, according to IACUC guidelines. Briefly, mice were injected with euthasol (270 mg/kg) and when deeply anesthetized, cardiac perfusion was performed, and tissues were collected. Tumors and organs were weighed and stored in −80 • C until further analysis.

Sample Collection
Throughout the course of the study, tumors were palpated, to assess growth, and were measured using calipers. During the experiment, mouse feces were collected, and samples were stored in pre-labeled collection tubes. Each mouse had its own cage, and a sterile pair of forceps was used for each collection. Up to 2 mL of feces was collected for each mouse on days D-1 (baseline), D6 (diet effect), D16 (cancer effect), D29 (surgery effect), and D35 (treatment effect). Bedding was changed after each fecal collection and three days following administration of chemotherapy/vehicle injections. Individual day/mouse tubes were stored in labeled and sealed Whirl-Pak bags and preserved at −80 • C.

DNA Isolation and Processing
Using a random number generator, four representative samples were selected for each collection timepoint, diet, and treatment combination, with the exception of samples collected one day before diet introduction, for which six samples were collected. Selected fecal samples were processed according to instructions given for the QIAmp ® PowerFecal ® DNA isolation kit (Qiagen; Hilden, Germany, 2017). Upon completion of DNA isolation, we measured the concentration and purity of each sample using NanoDrop (Thermo Scientific Nanodrop 1000 Spectrophotometer). Each sample was then diluted to a concentration of 5 ng/µL and suspended in Tris buffer solution (ThermoFisher Scientific; Waltham, MA, USA). Following isolation, samples were sent to the Molecular and Cellular Imaging Center (MCIC) core facility at the Ohio State University, Wooster campus, for 16S ribosomal genetic profiling. A genomic library was generated after amplification with the following primers, as in previous studies [31,32]: Forward-S-D-Bact-0564-a-S-15 (5 -AYT GGG YDT AAA GNG) and Reverse-S-D-Bact-0785-b-A-18 (5 -TAC NVG GGT ATC TAA TCC), amplifying a fragment of 206 bp.

Morphological Analysis
Diet and treatment effects on morphological measurements of mice were analyzed over the course of the experiment, using a regular Analysis of Variance (ANOVA) for body weight and body weight change, and an ANOVA using M-estimators (pbad2way() function from the R package WRS2, version 1.1.3 [33]) for the non-normally distributed measurements of tumor weight and tumor weight as a proportion of body weight.

Microbiome Data Analysis
All code used for the analyses can be found in our GitHub repository for this study (https://github.com/jelmerp/mouse-cancer-metabarcoding, accessed on 26 January 2023). Computation was performed at the Ohio Supercomputer Center [34] using project PAS0471. After confirming that FASTQ files for all samples were of an appropriate quality for analysis using FastQC (version 0.11.8 [35]) and MultiQC (version 1.11 [36]), primers were removed using cutadapt (version 3.4 [37]).
A phylogenetic tree for all ASVs was inferred using the R package phangorn (version 2.8.1 [43]). The R/Bioconductor package phyloseq (version 1.38.0 [44]) was used to store the resulting count table, taxonomy table, phylogenetic tree, and a metadata table as a single R object and perform a number of downstream analyses, such as calculating weighted UniFrac distances among samples, performing a Principal Coordinate Analysis (based on UniFrac distances), and agglomerating counts at higher taxonomic levels. To examine overall differences in microbiome composition between days, diet, and treatment, we used the adonis2() function from the R package vegan (version 2.5.7 [45]) to run a PERMANOVA with the weighted UniFrac distance between samples as the responding variable. Pairwise PERMANOVA comparisons were done with the pairwise.adonis2 function from the R package pairwiseAdonis (https://github.com/pmartinezarbizu/pairwiseAdonis, accessed on 28 December 2022 version 0.4.1).
In the Supplementary, sequence data removal and filtering in different steps of the pipeline are visualized in Figure S2a and taxonomic assignment success for ASVs is visualized in Figure S2b. An average of 158,455 reads per sample (minimum: 54,644; maximum: 301,865) were sequenced, and an average of 142,898 (minimum: 47,581; maximum: 284,075) were retained after quality control measures at the end of the ASV inference pipeline. Taxonomy was assigned to ASVs at high rates at the order level (97.48%), declining at the family (81.68%), genus (55.66%), and especially species (5.59%) levels.
Taxonomic richness was estimated for each sample and compared between days, diet, and treatment using the R package breakaway (version 4.7.6 [46,47]). Other measures of alpha-diversity (Simpson and Shannon indices) were estimated and compared among days, diet, and treatment using the R package DivNet version 0.4.0 [48]). The approaches implemented in these packages robustly estimate diversity measures while accounting for unobserved species and not requiring rarefaction [49].
Differential abundance analysis at the ASV, genus, and family level was performed using the R/Bioconductor package ALDEx2 (version 1.24.0 [50]). Reported p-values were adjusted using the Benjamini-Hochberg multiple-testing correction method. For consistency with this analysis, ASV counts of differentially abundant taxa were visualized after normalization with the ALDEx2 function aldex.clr().
A list of taxa of interest (n = 43) was compiled, based on previously reported commensal human bacteria which are associated with influencing cancer and treatment outcomes or are opportunistic commensal pathogens associated with foodborne disease. Representatives of these taxa or the closest higher taxonomic level that was present in our data were singled out to assess differential abundance, plot abundances, and, where appropriate, perform separate ordination analysis.

Microbial Diversity, Richness, and Abundance
Overall, microbial diversity significantly changed over the course of the experiment (i.e., by day) and differed by diet. The model for day and diet was found to be the best fit; effects of day, diet, and their interactions on richness are shown in Table 1. With LSHF acting as the intercept, significant differences were found between the intercept and LSLF

Microbial Diversity, Richness, and Abundance
Overall, microbial diversity significantly changed over the course of the experiment (i.e., by day) and differed by diet. The model for day and diet was found to be the best fit; effects of day, diet, and their interactions on richness are shown in Table 1. With LSHF acting as the intercept, significant differences were found between the intercept and LSLF  . Other alpha-diversity measures, such as the Shannon index, also take into account evenness in abundance among ASVs. The Shannon index was also significantly affected by day and diet (Figure 3), as well as by treatment. Day-only and treatment-diet interaction reporting can be found in the Supplementary ( Figure S4). Other alpha-diversity measures, such as the Shannon index, also take into account evenness in abundance among ASVs. The Shannon index was also significantly affected by day and diet (Figure 3), as well as by treatment. Day-only and treatment-diet interaction reporting can be found in the Supplementary ( Figure S4).

Overall Microbiome Composition
A PERMANOVA analysis by day and diet, using UniFrac distance as the respondent variable, showed significant differences in overall microbiome composition by day (F(3, 76) = 7.0097, R 2 = 0.2075, p = 0.0001), diet (F(3, 76) = 2.1859, R 2 = 0.0647, p = 0.0001), but not by the interaction between the two (F(9, 70) = 1.087), R 2 = 0.0965, p = 0.32457). Pairwise comparisons among diets in the above model revealed significant differences between the two low sucrose diets (R 2 = 0.0707, p = 0.006), and between LSLF and HSHF (R 2 = 0.04921, p = 0.024). When considering the chemotherapy treatment in a model that also included diet, significant differences were observed between mice that had received chemotherapy versus placebo (F(1, 30) = [2.7814], R 2 = 0.0974, p = 0.04170). Ordination plots using principal coordinate analysis indicated that one ASV, which was classified as Akkermansia muciniphila, had a large impact on this analysis ( Figure S5). This was overall by far the most abundant ASV (ASV1 mean proportional abundance = 0.233; 26 of 86 samples showed ASV1 proportional abundance > 0.5), but its pattern of abundance among samples was clearly bimodal after baseline (Figure 4). Of individual mice that had been randomly sampled at more than one timepoint, 11 experienced an increase of A. muciniphila abundance, while four experienced a decrease, and two experienced both an increase and a decrease at separate time points ( Figure S6). The abundance of A. muciniphila and overall sample taxonomic richness was negatively correlated (r(64) = −0.55, p = 3.00 × 10 −8 ).

Overall Microbiome Composition
A PERMANOVA analysis by day and diet, using UniFrac distance as the respondent variable, showed significant differences in overall microbiome composition by day (F(3, 76) = 7.0097, R 2 = 0.2075, p = 0.0001), diet (F(3, 76) = 2.1859, R 2 = 0.0647, p = 0.0001), but not by the interaction between the two (F(9, 70) = 1.087), R 2 = 0.0965, p = 0.32457). Pairwise comparisons among diets in the above model revealed significant differences between the two low sucrose diets (R 2 = 0.0707, p = 0.006), and between LSLF and HSHF (R 2 = 0.04921, p = 0.024). When considering the chemotherapy treatment in a model that also included diet, significant differences were observed between mice that had received chemotherapy versus placebo (F(1, 30) = [2.7814], R 2 = 0.0974, p = 0.04170). Ordination plots using principal coordinate analysis indicated that one ASV, which was classified as Akkermansia muciniphila, had a large impact on this analysis ( Figure S5). This was overall by far the most abundant ASV (ASV1 mean proportional abundance = 0.233; 26 of 86 samples showed ASV1 proportional abundance > 0.5), but its pattern of abundance among samples was clearly bimodal after baseline (Figure 4). Of individual mice that had been randomly sampled at more than one timepoint, 11 experienced an increase of A. muciniphila abundance, while four experienced a decrease, and two experienced both an increase and a decrease at separate time points ( Figure S6). The abundance of A. muciniphila and overall sample taxonomic richness was negatively correlated (r(64) = −0.55, p = 3.00 × 10 −8 ).

Differential Abundance
An analysis of differential abundance showed that a total of 65 ASVs significantly differed in abundance by day, as did 13 genera and 11 families ( Figure 5).

Differential Abundance
An analysis of differential abundance showed that a total of 65 ASVs significantly differed in abundance by day, as did 13 genera and 11 families ( Figure 5).

Taxa of Interest
Genera identified in the samples included Bacteroides, Bifidobacterium, Enterococcus, Lactobacillus, Legionella, Mycobacterium, Proteus, Pseudomonas, and Streptococcus. Bilophila wadsworthia, Staphylococcus aureus, Klebsiella spp., and Peptococcus spp. were identified to the genus and species level. These four taxa did not show significant differences in abundance by disease progression, diet, or treatment.
At the order level, Campylobacterales and Bacteroidales were detected and Bacteroidales abundance was significantly impacted by day (9 ASVs). At the class level, Gammaproteobacteria (189 ASVs) was impacted by day and diet.  When treatment and diet were considered, the family Streptococcaceae was less abundant in mice that received chemotherapy (p = 4.00 × 10 −2 ) and Oxalobacteraceae were significantly lower in the low sucrose, low omega-3 diet group, regardless of treatment (p = 3.66 × 10 −2 ).

Taxa of Interest
Genera identified in the samples included Bacteroides, Bifidobacterium, Enterococcus, Lactobacillus, Legionella, Mycobacterium, Proteus, Pseudomonas, and Streptococcus. Bilophila wadsworthia, Staphylococcus aureus, Klebsiella spp., and Peptococcus spp. were identified to the genus and species level. These four taxa did not show significant differences in abundance by disease progression, diet, or treatment.
The identified families included Peptostreptococcaceae, Clostridiaceae, Enterobacteriaceae, Oscillospiraceae, and Listeriaceae. Three ASVs from the genus Romboutsia¸ in the family Peptostreptococcaceae, increased by day, as did the overall family (p = 8.58 × 10 −13 ). Additionally, the abundance of the Peptostreptococcaceae family was impacted by the interaction between

Discussion
In this study, we have demonstrated changes in diversity and composition of the mouse microbiome during cancer progression and treatment. As anticipated, alphadiversity decreased as the disease progressed and differed between diet types. However, though there were significant differences between the diets, conclusive results could not be drawn on if certain nutrients protected or exacerbated this decrease in diversity. It is well documented that diet is a major determinant of the richness and diversity of the microbiome [51], as supported by our results. Oligosaccharides are known to be used by opportunistic pathogens within the gut, and can lead to, or worsen, the state of disease and limit treatment opportunities [52]. Though no pathogenic commensals were able to be identified to the species level, there were multiple ASVs, families, and classes of focal taxa which were impacted by dietary patterns, which warrants further investigation. Diets high in sugar or fat have been previously shown to promote inflammation associated with the gut microbiome [16]. Sucrose, in particular, has been associated with changes in the microbiome and to permeability of the intestinal mucosal barrier, an important factor in maintenance of host immunity [53]. This study used a high sucrose, low omega-3 diet to simulate common Western dietary patterns, as the Western diet typically involves obtaining fatty acids from plants, rather than marine sources [27]. While some dietary effects were seen, more investigation is required, as dietary intake and its effects on the microbiome may be much more intricately linked to additional factors, such as genetics and environment. The changes in abundance, from this study, can largely be attributed to a decrease in overall richness throughout the disease progression, rather than explicitly to diet.
The abundance of several ASVs, genera, and families were significantly impacted by diet, treatment, disease progression, and their interactions. However, a distinct link was not able to be determined between disease progression or diet and an increase in pathogenic commensals or inflammatory bacterial species, as hypothesized. Unexpectedly, Akkermansia muciniphila demonstrated a bimodal polarization after baseline, without regard for diet. Akkermansia is a mucin-degrading bacterium [54] and its presence is highly associated with favorable outcomes for cancer patients [55]. It is usually a minor resident of the gut, and is the only known bacterium in the Verrucomicrobia phylum [14]. It is generally associated with positive health outcomes and increased longevity [8], and its presence in the murine gastrointestinal microbiome has previously been shown to increase in mice fed fish-oil [17]. Akkermansia populations have been shown to be negatively impacted, in mouse model systems, by the administration of cyclophosphamide [10]. In this study, no association was found between Akkermansia and diet or treatment. The abundance distribution was bistable and switched randomly, in direction and magnitude, within individual mice, for reasons unable to be explained by the scope of this experimental design. Further analysis is needed to determine why this occurred.
The microbiome influences innate immunity [56] and affects the response of the host to cancer therapies, such as chemotherapy or immunotherapy [57]. A cyclical relationship exists between the composition and diversity of the microbiome and cancer therapy [58]. In this study, we specifically investigated the combination of doxorubicin and cyclophosphamide and found significant changes in richness and alpha-diversity associated with chemotherapy treatment, dependent on diet. Some microbial communities contribute to resilience against doxorubicin toxicity [59], but the administration of doxorubicin still causes changes in the composition of the gut microbiome [60]. Cyclophosphamide has been shown to drive the murine microbiome towards dysbiosis, significantly decreasing Lactobacilli and Enterococci populations, while increasing Clostridium group IV [11]. Depletion of the diversity of the intestinal microbial population is associated with negative cancer outcomes, while in a healthy individual, the composition of the microbiome remains stable for 60% of bacterial species and diversity is commonly high [61]. Many commensals, such as Bifidobacteria, have been shown to be susceptible to the toxicity of chemotherapeutic treatments and can become depleted [62], allowing for an overabundance of opportunistic pathogens. Despite the clear risk that this combination may pose to the host microbial population, combinations of anthracyclines with cyclophosphamide have been the primary chemotherapeutic treatment for metastatic breast cancer since the 1990s [63]. Yet, there are still relatively few studies investigating the interaction between chemotherapy and the gut microbiome [64].
Beyond this analysis of the interplay between cancer treatment and the gastrointestinal microbiome, there are also very few studies that assess the relationship between the microbiome and cancer itself. In one study, cancer progression was associated with a decreased bacterial load in the microbiome: patients with stage 1 cancer showed a greater total bacterial DNA load than patients with stage 2 or stage 3 cancer [25]. However, these results were analyzed from the perspective of the impact of the microbiome on tumorigenesis, rather than assessing the interaction between cancer progression and the microbial population [65]. In this study, we found that microbiome richness significantly decreased after onset of cancer, as did alpha-diversity. Future studies should consider that onset of cancer correlates with a decrease in microbiome diversity, which may further exacerbate tumor progression. Additionally, this study was performed only with breast cancer cells, and so results should not be extrapolated to different types and sites of cancer.
Despite general profiles being documented, interactions between the gut microbiota and the host are specific to the host, and are dependent on genetic, historical, and physical factors, making the composition and cross-talk of the microbiome unique [66]. Therefore, there is no one specific "optimal" microbiome population, as there are many factors which shape the microbial population specifically for the host [67], and it is commonly accepted that a healthy microbiome is one in which the pathogenic bacterial populations are in balance with the symbiotic populations [68].
In this study, the richness of the gastrointestinal microbiome began decreasing prior to inoculation with breast cancer cells, and the presence of Akkermansia muciniphila occurred independently of diet and prior to cancer inoculation. These results may be linked to the ovariectomy performed on the mice prior to sample collection and diet administration. Previous studies have indicated that surgery allows for bacterial translocation, and this is associated with changes in the composition of the gastrointestinal microbiome [69]. Therefore, the changes in richness observed in this study may also be attributable to the performance of surgery prior to sample collection.
Mice are effective preclinical models and help to establish an understanding of the mechanism of changes in the microbiome due to treatment [51]. However, this model is limited by differences in anatomy and genetics, and so is not a perfect comparator for humans [66]. This limitation can be addressed in future studies by using other model systems and, eventually, human patients in order to more accurately understand relevant microbiome shifts throughout the onset and progression of cancer and treatment. Additionally, fecal samples have been accepted as an adequate proxy for assessing the microbial community of the gut microbiome and are less invasive than taking samples directly from the GI tract. However, community structure is frequently lost, leading to an incomplete picture of the microbiome [70]. Future studies would benefit from taking samples directly from the GI tract, when possible.

Conclusions
In this study we determined how richness, evenness, and composition of the murine gastrointestinal microbiome are impacted by the progression of cancer, macronutrient dietary composition, chemotherapeutic treatment, and the interplay between the three. The microbiome has been well analyzed for its contribution to cancer treatment outcomes, but little focus has yet been given to how it is cyclically impacted by cancer, even prior to treatment. Projections for new cancer cases remain high, which means that, at any given time, there are a large number of patients receiving treatment. As the composition and diversity of the microbiome contributes to host immunity and resilience, understanding the complex relationship between disease progression and microbial populations is vital to improving treatment outcomes. These results should be considered, when designing clinical studies, with regards to how the richness and diversity are negatively impacted by changes in diet and the progression of cancer. The overgrowth of commensal pathogens and the depletion of beneficial commensals needs to be quantified and addressed. This study contributes to a better understanding of how the microbial population is impacted by cancer treatment and the course of the disease, leading to potentially better patient support and improved treatment outcomes.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/nu15030724/s1, Figure S1: Experimental design for (a) assigning mice to cohorts, (b) experiment timeline and fecal collection points, and (c) dietary macronutrient balance; Figure S2: Quality control and sample processing (a) filtering and read removal along the data processing pipeline (b) proportion of ASVs assignments by taxonomic level; Figure S3: Estimated richness by diet and treatment (a) microbiome population richness significantly decreased over time (b) microbiome population richness differed by diet and treatment interactions. Figure S4: Richness and evenness of ASV/per-sample, weighted by the Shannon and Simpson diversity indices (a) demonstrate differences in alpha-diversity modeled by study timepoint (b) demonstrate differences in alpha-diversity modeled by diet and treatment. Figure S5: Ordination from principal