Manipulations of the gut microbiome alter chemotherapy-induced inflammation and behavioral side effects in female mice

Abstract

Chemotherapy treatment is associated with acute behavioral side effects (fatigue, anorexia) that significantly reduce patient quality of life and are dose-limiting, thereby increasing mortality (Kidwell et al., 2014). Disruptions to gut homeostasis (diarrhea, constipation, microbial dysbiosis) are also observed in patients receiving chemotherapy. In non-oncological patients, facets of mental health (fatigue, anxiety, depression) correlate with alterations in the gut microbiome, suggestive of a contribution of the gut in CNS disease etiology. The potential gut-to-brain pathway is poorly understood in patients receiving chemotherapy. Our prior studies have demonstrated a correlation between chemotherapy treatment, gut changes, peripheral and central inflammation, and behavioral symptoms in mice. Here we aimed to determine the extent to which chemotherapy-associated gut manipulations modulate the behavioral and biological consequences of chemotherapy. We measured sickness behaviors, peripheral and central inflammatory mediators, and anxiety in conventional or germ-free female mice: 1) cohabitating with mice of the opposite treatment group, 2) pre-treated with broad-spectrum antibiotics, or 3) given an intra-gastric gavage of gut content from chemotherapy-treated mice. In cohabitation studies, presumed coprophagia promoted body mass recovery, however strong associations with inflammation and behavior were not observed. Reduction of gut microbial alpha diversity via antibiotics did not prevent chemotherapy-associated side effects, however the relative abundances of the genera Tyzzerella, Romboutsia, and Turicibacter correlated with circulating inflammatory (IL-1β) and behavioral outcomes (lethargy, anxiety-like behavior). A gut microbiota transplant from chemotherapy-treated mice decreased central locomotion in open field testing, increased circulating CXCL1, and increased hippocampal Il6 and Tnfa in germ-free mice compared to germ-free mice that received a transplant from vehicle-treated mice. Taken together, these data provide further evidence that the gut microbiota likely contributes to the development of chemotherapy-associated side effects. This work has significant implications in the future treatment of anxiety in patients, and warrants future studies using microbe-based treatment options.

1. INTRODUCTION

It is estimated that there were over 16.9 million cancer survivors in the US as of January 2019 (Miller et al., 2019). The most prevalent cancers, including colon, prostate, and breast, are routinely treated with chemotherapy. Chemotherapy is a broad term for a class of drugs that non-specifically prevents cell replication and causes cell death. While often extending the lifespan of patients, chemotherapy can cause a reduction in patient quality of life through its devastating short- and long-term side effects including weight loss, skeletal muscle loss, fatigue, and psychiatric comorbidities (reviewed in Santos and Pyter, 2018). Such symptoms decrease adherence to treatments and increase patient mortality (Satin et al., 2009; Stilley et al., 2010; Griffin et al., 2019). Chemotherapy-associated metabolic and behavioral changes can also be observed in rodent models (Fardell et al., 2012; Loman et al., 2019; Winocur et al., 2018). In our previous studies, weight loss, anxiety-like behavior, and fatigue are observed with the use of paclitaxel chemotherapy, therefore enabling pre-clinical investigations of this complex, multi-organ system phenomenon. Mechanisms underlying the development of these side effects remain incompletely understood, and are poorly managed by patients and physicians. Therefore, it is pertinent to elucidate the causes of these symptoms to aid in the development of effective interventions.

Taxanes, including paclitaxel, are one of the most common classes of chemotherapeutics used to treat breast, ovarian, and other cancers (Ojima et al., 2016). Behavioral symptoms induced by paclitaxel treatment appear to be caused indirectly, as entry of the drug into the brain is limited due to the high expression of the p-glycoprotein efflux pump at the blood brain barrier (Fellner et al., 2002). Instead, correlative studies suggest a role for inflammation in the development of these behavioral symptoms in breast cancer patients (Bagnall-Moreau et al., 2019; Van Der Willik et al., 2018). Pro-inflammatory mediators also play a role in body composition changes associated with chemotherapy treatment in mice (Elsea et al., 2015).

Chemotherapy treatment is associated with gastrointestinal discomfort: diarrhea (Stein et al., 2010), disruption of the gut microbiome (Loman et al., 2019; Montassier et al., 2015; Rashidi et al., 2019), and damage to the gut epithelium (Keefe et al., 2000; Loman et al., 2019). Outside of the context of chemotherapy, gut microbial dysbiosis is consistently associated with fatigue (Giloteaux et al., 2016) and anxiety (Jang et al., 2018; Li et al., 2019; Yi-huan et al., 2019). Furthermore, microbiota alpha and beta diversity differences are observed among individuals that differ by body condition (anorexia nervosa patients, athletes, obese, etc.) (Mörkl et al., 2017). There has been a growing initiative to identify and isolate methods of communication from the gut microbiota to the brain (reviewed in Cryan, et al. 2019). This may be particularly relevant in the context of chemotherapy (Jordan et al., 2018; Loman et al., 2019). Indeed, gut microbes are necessary for the development of mechanical allodynia in oxaliplatin chemotherapy-treated mice (Shen et al., 2017). Multiple gut-to-brain pathways have been proposed, including direct neural communication and/or humoral communication through inflammatory mediators or bacterial metabolites (e.g., short-chain fatty acids) released into circulation (Bajic et al., 2018; Collins et al., 2012; Cryan et al., 2019). Specifically, butyrate, one of the major end metabolites of gut microbial energy metabolism, can influence brain and behavior (Stilling et al., 2016, Arnoldussen et al., 2017, Liu et al., 2015). In the present study, a significant role of the gut microbiome in the development of chemotherapy-associated inflammation and behavioral side effects is tested. Building upon our prior work demonstrating a correlation between paclitaxel-induced gut microbial changes, inflammation, and behavioral symptoms in female mice (Loman et al., 2019), the present study utilizes three approaches to assess the necessity or sufficiency of chemotherapy-induced gut microbial changes in circulating inflammation, neuroinflammation, short-chain fatty acids, anxiety-like behavior, and fatigue. First, female mice were cohoused with cage mates that received the same or different treatments (paclitaxel or vehicle) to determine the extent to which microbial transfer via presumed coprophagia among cage mates (Laukens et al., 2016) alters the development of inflammation or behavioral side effects after chemotherapy. Next, mice were pre-treated with broad-spectrum antibiotics to determine how reducing gut microbial content prior to paclitaxel treatment would impact symptom development and body composition. Lastly, germ-free mice were given an intra-gastric gavage of gut content from mice treated with vehicle or paclitaxel to determine the extent to which direct transfer of gut microbes transfers physiological and behavioral consequences of chemotherapy. Together, the data presented here demonstrate that chemotherapy-associated gut microbial changes contribute to the development of inflammation and psychological side effects.

2. MATERIALS AND METHODS

2.1. Experimental design overview.

Experiment 1: Cohabitation with same and mixed cage mates:

Adult, female C57BL/6 mice were housed 4/cage for at least 2 weeks prior to the start of injections. Between-cage differences in the gut microbiome as a result of coprophagia are often regarded as troublesome when designing experiments to study the gut microbiome (Laukens et al., 2016). Therefore, during the 2-week acclimation period, while it may be a mild stressor to mice, the bedding from all of the cages were combined, mixed, and redistributed among all of the cages to minimize the baseline cage effects. Mice were assigned to 1 of 4 groups; Vehicle Same, Vehicle Mixed, Chemo Same, Chemo Mixed (n=11-12). The groups that are identified as “Same” correspond to mice housed with cage mates that received the same treatment. The groups labeled “Mixed” indicate that the test mouse was housed with 3 other mice that received the opposing treatment (e.g. a mouse in the Chemo Mixed group received chemotherapy, but its cage mates all received vehicle injections). Mice were injected according to a clinically-relevant, multicycle chemotherapy treatment paradigm or vehicle (see “Chemotherapy Treatment” below). Six hours after the final injection, plasma was collected via the retro-orbital sinus (see “Tissue Collections” below). On the third day after the final injection, open field behavioral testing was completed, and on the sixth day after the final injection, body composition was measured in a subset of mice in the morning, then animals were euthanized and tissues were collected in the afternoon. Body mass and food intake was measured throughout the course of all experiments to monitor weight loss.

Experiment 2: Antibiotic treatment in combination with chemotherapy:

Mice were given antibiotic or control chow (see “Antibiotic Treatment” below) for 8 days prior to the start of and throughout the chemotherapy or vehicle regimen (see “Chemotherapy Treatment” below). The four groups of mice are identified as control chow plus vehicle treatment (Veh, n=15), control chow plus chemotherapy treatment (Chemo, n=12), antibiotic chow plus vehicle treatment (Veh+Abx, n=15), and antibiotic chow plus chemo treatment (Chemo+Abx, n=9). Six hours after the final injection, behavioral testing for lethargy and anxiety-like behavior in an open field was completed followed immediately by tissue collection to measure peripheral and central inflammatory mediators and colon content bacteriome analysis.

Experiment 3: Gut-microbial transfer from chemotherapy-treated mice:

Conventional mice were first injected with chemotherapy or vehicle and 6 hours after the final injection, lethargy and anxiety-like behaviors were assessed. In one cohort, 24 h after the final injection, mice were euthanized and cecal and proximal colon contents were collected in an anaerobic broth (see “Gut microbiota transplant” below, n=5-6). Plasma was collected at this time. In the second cohort of mice (n=12), 24 h after the final dose of chemotherapy mice were euthanized, and blood and brain tissues were collected for assessment of “donor” peripheral and central inflammatory mediators. Technical error in the isolation of hypothalamic RNA from the second cohort reduced hypothalamic inflammatory gene sample size. Germ-free recipient mice received microbial transplants from chemotherapy- or vehicle-treated donors via gastric gavage (see “Gut microbiota transplant” below, n=9 per group) within 3 h following content collection. Seven days after the gut microbial transplant, germ-free mice underwent fatigue and anxiety-like behavioral testing followed immediately by euthanization and collection of brain tissue and plasma to measure inflammatory markers.

2.2. Animals.

Female, randomly-cycling, 7-8 week old C57BL/6 mice (Charles River, Wilmington, MA, USA) from mixed litters were housed in groups of 3-4 and acclimated to a 14:10 light:dark cycle (lights off at 1400 h) in a temperature-controlled vivarium (22 ± 1 °C) for 1 week. This strain was chosen as a gold standard in mouse behavioral testing, including open-field testing used in our studies (Crawley, 2007). Furthermore, it has been shown that C57BL/6 mice travel a greater percentage of distance in the center than 7 other mouse strains (in a panel of 14 mouse strains) (Bothe, et al., 2005). This allows us to identify increases or decreases in central tendency (See Behavioral Testing below). Additionally, female mice were used in this study because paclitaxel is a mainstay in the treatment of breast cancer, which overwhelmingly occurs in women. The use of randomly cycling mice were used because result variability of randomly cycling females is not higher than that of male mice in 30 categories of behavioral, morphological, physiological, and molecular traits (Prendergast et al., 2014). All mice were acclimated to handling 3 times prior to experimental procedures to reduce stress during behavioral testing and chemotherapy injections. Standard rodent chow (Teklad LM-485, irradiated, Envigo, Indianapolis, IN, USA) and water were available ad libitum throughout the duration of the study unless otherwise indicated. Female, randomly-cycling, germ-free 6-7-week-old C57BL/6 mice (Charles River, Wilmington, MA, USA) were housed in groups of 3 and acclimated for 1 week to a 14:10 light:dark cycle (lights off at 1400 h) in a temperature-controlled gnotobiotic vivarium at Nationwide Children’s Hospital. Animals housed at Nationwide Children’s Hospital were provided rodent chow (Teklad Global Diets, 2020SX that had been autoclaved, Envigo) and water ad libitum. All animal experiments were approved by the Ohio State University and Nationwide Children’s Hospital Institutional Animal Care and Use Committees and carried out in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (Council, 2011). All efforts were made to minimize animal suffering and to reduce the number of mice used.

2.3. Chemotherapy treatment.

Solid paclitaxel (Experiment 2: >97%, Sigma-Aldrich, St. Louis, MO, USA or Experiments 1 and 3: >99.5%, LC Labs, Woburn, MA, USA), was dissolved in a 50:50 mixture of Cremophor-EL and 200-proof EtOH, then diluted 1:1 with sterile PBS. Mice were injected with a clinically-relevant injection protocol of 30 mg/kg i.p., every other day, unless otherwise indicated, for a total of 6 injections (Loman et al., 2019; Sullivan et al., 2021). Body mass and 48-h food intake was recorded starting at least two days before injections (baseline) and every injection day thereafter to assess cachexia and anorexia. Food intake was averaged for the number of animals per cage to obtain a proxy single mouse food intake over the designated time period. Per IACUC guidelines, animals were removed from the study and immediately euthanized if they lost >20% of their baseline body mass.

2.4. Antibiotic treatment.

For Experiment 2, mice were given non-irradiated control chow, Teklad LM-485, or the same chow with the addition of 0.5 g/kg vancomycin, 1g/kg metronidazole, and 1 g/kg neomycin trisulfate (Envigo, Madison, WI, USA) for 8 days prior to the start of chemotherapy, and throughout the remainder of the experiment to broadly target Gram-positive and –negative aerobic and anaerobic bacteria.

2.5. Gut microbiota transplant.

For Experiment 3, gut content was harvested from the cecum and proximal colon of conventional donor mice after chemotherapy or vehicle treatment (see experimental design overview; n=6/group), combined within treatment group, and suspended in Schaedler broth (30 mg/mL) that was prepared under anaerobic conditions and then stored on ice. Germ-free mice received 100 μL content mixture by intra-gastric gavage within 3 h of isolating donor content. Cecal and proximal colon content were chosen as the donor material because these portions of the lower gut are the sites of highest bacterial abundance (Donaldson et al., 2016). To minimize the chance of infecting germ-free mice with environmental microbes, one intra-gastric gavage was administered, which has been demonstrated to be robustly effective at populating the gut (Bercik et al., 2011). The concentration of paclitaxel in the gut content was determined to be 16.2 ng/ml by mass spectrometry (Supplemental Methods), which equates to 0.083 μg/kg via the gavage. In addition to being >300,000 times less than the dose used for direct injections in these experiments (30 mg/kg, i.p.), this amount of paclitaxel is considered negligible as paclitaxel has very low oral bioavailability (Sparreboom et al., 1997).

2.6. Behavioral testing.

Open Field.

Testing occurred during the early dark phase (1800-2100 h) under dim red light. Mice were placed into one corner of a 16 x 16 inch arena with a sparse coating of corncob bedding, and allowed to explore for 30 min to measure anxiety-like behavior (central tendency) and general locomotion as previously described (Pyter et al., 2017). The apparatus was cleaned with 70% ethanol between mice. Fatigue was operationally defined as decreased locomotion in the open field for this study. Anxiety-like behavior was defined as a reduction in central tendency, or the % of locomotion in the central 4 x 4 inch zone of the arena. The estrous stage of animals was not assessed prior to behavior testing as C57BL/6 mice have a stable performance in the open field test throughout the estrous cycle (Meziane et al., 2006). For Experiment 1 and 2, the arena was placed inside a photobeam frame (San Diego Instruments, San Diego, CA, USA) and measurements of locomotion, XY distance traveled or beam breaks, were analyzed using PAS Data Reporter (San Diego Instruments) automated software. In Experiment 3, corncob was excluded from the arena for donor and recipient mice as a precaution against infection in the gnotobiotic housing. Activity in the arena was video-recorded, and measurements of locomotion were acquired using automated AnyMaze Behavioral Tracking Software (Stoelting Co., Wood Dale, IL, USA).

2.7. EchoMRI Body Composition Analysis.

Body composition measurements were completed in Experiment 1 on a subset of animals (n=7-8/group) on the 6^th day after the final injection to identify potential changes in lean and fat mass caused by chemotherapy treatment or housing conditions. An EchoMRI analyzer in the OSU Small Animal Imaging Core Facility (EchoMRI L.L.C., Houston, TX, USA) was used per manufacture’s protocol. Mice were placed in a cylinder with a loose-fitting plunger to limit movements that can interfere with readings. The cylinder was inserted into the instrument and 2 measurements were taken per animal lasting 1-3 min per measurement. During each reading, fat, lean, free, and total water mass was recorded. The average of the two measurements was determined, and results are reported as percent of body mass.

2.8. Tissue collections.

All tissues were collected during the early dark phase (1400-1700 h). Mice were euthanized by rapid decapitation, trunk blood was collected in heparin-lined Natelson tubes and stored on ice. Whole blood was then centrifuged for 20 min at 600 RCF at 4 °C and platelet-free plasma was collected and stored at −80 °C for cytokine analyses. Hemolysis was observed in some plasma samples, though occurred equally among samples from all groups and experiments. Hypothalamus, hippocampus, and frontal cortex, were dissected and immediately frozen on dry ice and stored at −80 °C prior to RNA or protein isolation. Distal and proximal colon content were collected from the proximal and distal halves respectively, frozen, and stored at −80 °C for storage prior to 16S rRNA bacterial gene sequencing and analysis.

2.9. Plasma cytokine/chemokine concentrations.

Based on our previous studies and studies of breast cancer patients, protein concentrations of circulating (Experiment 1 and Experiment 3) and hippocampal, hypothalamic, and frontal cortex (Experiment 2) cytokines and chemokines (IL-1β, IL-2, IL-6, TNFα, CXCL1 [KC/GRO]) were measured using multi-plex fluorescent immunoassays (Meso Scale Discovery, Rockville, MD, USA) and run according to the manufacturer’s instructions (Sullivan et al., 2021; Loman et al., 2019; Bagnall-Moreau et al., 2019). Proinflammatory cytokines, including these, are associated with fatigue and anxiety (DeSanctis, V. et al., 2014; Perez-Tejada, J. et al., 2021). Samples were run in duplicate and analyzed using a QuickPlex SQ instrument (Meso Scale Discovery). Values below detection limits were reported as the lowest value on the standard curve. This occurred only in Experiment 3 in GMT recipients; IL-1β = 0.49, 6/7 vehicle, 9/9 chemo, IL-6 = 0.61, 2/7 vehicle, 3/9 chemo. Due to a technical error in creating the standard curve on one plate, the sample size for brain regions associated with Experiment 2 are limited. Intra-assay variation was <20%.

2.10. Quantitative RT-PCR.

Qiagen RNeasy mini kits (Qiagen, Germantown, MD, USA) were used to extract total RNA from the hypothalamus, frontal cortex, and hippocampus. RNA was reversed transcribed using qScript cDNA SuperMix (QuantBio, Beverly, MA, USA) per manufacturer’s instructions with 1μg of RNA per 20 μL reaction. Expression of pro-inflammatory cytokines/chemokines (II-1β, Il-6, Tnfα, and Cxcl1) in the brain was assessed based on previous studies.(Loman et al., 2019) Gene expression was normalized to the geometric mean of Gapdh and Hprt and reported as 2^−ΔΔCT. Prior to analysis, it was confirmed that the geometric mean of Gapdh and Hprt was not different between groups.

2.11. 16S rRNA bacterial gene sequencing and analysis.

Colon content samples were sent to The Environmental Sample Preparation and Sequencing Facility at Argonne National Laboratory for DNA extraction, library preparation, and high-throughput sequencing. Paired-end (250 nt forward and reverse) sequences of the V4 hypervariable region of the 16S rRNA gene (515F-806R) were generated on the Illumina MiSeq. Quantitative Insights into Microbial Ecology (QIIME) 2.0 (Bolyen et al., 2019) was utilized for amplicon processing, quality control with DADA2, downstream taxonomic assignment using the SILVAv132 database, and diversity analyses. Sequencing of samples in the antibiotic study initially resulted in 2,488,123 paired-end sequences (median=86,223; maximum=128,714, minimum=39,963). After quality control, 1,850,936 high-quality sequences remained (median=64,045; maximum=93,533; minimum=24,111). Samples were rarefied to 24,000 sequences per sample (no samples excluded). Sequencing of samples in the antibiotic study initially resulted in 2,381,881 paired-end sequences (median=114,274.5; maximum=159,018, minimum=80,250). After quality control, 329,274 high-quality sequences remained (median=16020.5; maximum=20,754; minimum=11,819). Samples were rarefied to 11,000 sequences per sample (no samples excluded). Genus-level taxonomic differences were determined utilizing Wilcoxon tests, and metadata relationships via Spearman’s correlations in JMP v13.0.0 statistical software (SAS Institute Inc., Cary, NC, USA). Raw 16S rRNA bacterial gene sequencing data is available upon request.

2.12. Short-chain fatty acid concentrations.

Short chain fatty acids were assessed in intestinal contents using the method described by Garcia-Villalba. et al. (2012), with the following modifications. Intestinal samples were suspended in 0.5% phosphoric acid (at 200 mg/ml) prior to column filtration. Filtrates were then assessed using a Shimadzu QP 2010 SE with AOC-20s autosampler that was fitted with a Stabilwax-DA column (30 m, 0.25 mm; Restek, College Station, PA). Injection was made in split mode (using a ratio of 1:75), and premade free fatty acid standards (Restek) were used to create standard curves for quantification of acetate, propionate, and butyrate. All other methods are as described in Garcia-Villalba. et al. (2012).

2.13. Statistical analyses.

Statistical analyses of body mass, behavior measures, cytokine concentrations and protein levels, microbe relative abundance, and SCFA concentration were analyzed using a two-way ANOVA with Tukey’s multiple comparisons test. Student’s t-tests when variance was normal or nonparametric Mann-Whitney tests were used to analyze behavioral measures, circulating cytokines, and inflammatory gene expression in Experiment 3. Statistical analyses for gene expression was completed using 2^−ΔΔCT values. Spearman correlations were calculated for predicted pairs of data. Statistical significance was reported when p≤0.05 or indicated as trending/tending when p≤0.09. GraphPad Prism version 8.2.1 (GraphPad Software, San Diego, CA, USA) was used for all statistics except for the sequencing analyses mentioned above.

3. RESULTS

3.1. Healthy mice housed with mice treated with chemotherapy demonstrated select chemotherapy-like symptoms.

Mice were housed with cage mates that were either treated the same or injected with the opposite treatment: 1 vehicle mouse (the subject) housed with 3 chemotherapy mice or 1 chemotherapy mouse (the subject) housed with 3 vehicle mice. This cohabitation arrangement resulted in a reduction in body mass of vehicle mice that were housed with paclitaxel-treated mice as compared to vehicle-treated mice housed only with other vehicle-treated mice (Figure 1A; Veh Same vs Veh Mixed: day 5 t(20)=2.4, p<0.05; day 11 t(20)=2.3, p<0.05, three-way ANOVA main effect of day F_7,77=5.6, p<0.0001). Mice in mixed housing conditions tended to have intermediate weight loss on days 11 and 13 relative to mice housed only with same-treated cage-mates. Furthermore, mixed housing resulted in a significant reduction body fat mass (as a percentage of total body mass) in the vehicle-treated mice housed with chemotherapy-treated cage mates as compared to vehicle-treated mice housed together (Figure 1B; Veh Same vs Veh Mixed, t(13)=2.4, p<0.05, two-way ANOVA no main effect).

Figure 1. Effect of mixed housing condition on development of chemotherapy-associated symptoms.

A) The percent change in body mass as compared to body mass on the first day of injections. *p<0.05, t-test Veh Same vs. Veh Mixed. B) The percent of body mass that was fat was measured on the 6^th day after the final injection (i.e. the morning before afternoon tissue collection) via EchoMRI. *p<0.05, t-test. C) The percent of locomotion, measured by beam breaks, in the center of the open field chamber. Significance determined by two-way ANOVA and Tukey’s multiple comparisons test, *p<0.05, **p<0.01 D) IL-1β, E) TNFα, and F) IL-6 measured in the plasma of mice collected 6 h after the final injection by MSD multi-plex cytokine array. Significance determined by two-way ANOVA and Tukey’s multiple comparisons test, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Mean ± SEM

Chemotherapy increased anxiety-like behavior as reflected by a decreased central tendency in the open field test (Figure 1C; F_1,40=8.9, p<0.005, Veh Mixed vs Chemo Mixed, p<0.05). Peripheral inflammation was assessed by measuring the concentration of pro-inflammatory cytokines, IL-β, TNFα, and IL-6, in the plasma of mice 6 h after the last dose of chemotherapy (Figure 1D, 1E, and 1F, respectively). Chemotherapy treatment significantly increased circulating IL-β, TNFα, and IL-6 (Figure 1D, IL-1β, F_1,33=9.9, p<0.005, Veh Same vs Chemo Same p<0.005; Figure 1E, TNFα, F_1,34=7.5, p<0.01, Veh Mixed vs Chemo Mixed p<0.05; Figure 1F, F_1,33=36.5, p<0.0001, Veh Same vs Chemo Same p<0.005, Veh Mixed vs Chemo Mixed p<0.001). Mixed housing significantly attenuated chemotherapy-induced increases in IL-1β, driven by the reduction in the Chemo Mixed group versus the Chemo Same group (Figure 1D, housing effect, F_1,33=5.9, p<0.05; drug x housing interaction, F_1,33=5.9, p<0.05; Chemo Same vs Chemo Mixed p<0.01).

3.2. Antibiotics significantly altered the gut microbiome without changing chemotherapy-associated neuroinflammation and behavioral consequences.

As anticipated, prolonged antibiotic chow consumption resulted in a robust reduction in the overall alpha diversity (Supplemental Figure 1, Supplemental Figure 2A) and specific reductions in the relative abundances of approximately 100 genera in the colon (Supplemental Table 1). Furthermore, chemotherapy- and antibiotic-treated animals clustered by Bray-Curtis (beta-diversity) distances (Figure 2A; antibiotic effect p=0.001, chemo effect p=0.02). Antibiotics, regardless of drug treatment, tended to decrease the relative abundance of Romboutsia and Turicibacter (Figure 2B and 2C; antibiotic effect A. F_1,23=3.9, p=0.06, Chemo vs Chemo+Abx, U=8 p<0.005 B. F_1,21=3.9, p=0.06, Chemo vs Chemo+Abx, U=0 p<0.001, Supplemental Figure 2B and 2C) and greatly decreased the relative abundance of Tyzzerella (Figure 2D; F_1,22=8.2, p<0.01, Supplemental Figure 2D). Chemotherapy tended to increase Romboutsia, Turicibacter, and Tyzzerella (Figure 2B-D; chemotherapy effect B. F_1,23=3.9, p=0.06, Veh vs Chemo, U=5 p<0.05 C. F_1,21=3.6, p=0.07, Veh vs Chemo, U= 1 p<0.01 and D. F_1,22=4.1, p=0.06, Veh vs Chemo, U= 6.5 p=0.05, Supplemental Figure 2B-D). Conversely, the relative abundances of Escherichia coli and Lactobacillus were dramatically increased in antibiotic-treated mice, regardless of chemotherapy treatment (Supplemental Figure 1E and 1F, and Supplemental Figure 3).

Figure 2. Chemotherapy and antibiotic effects on the gut microbiome, behavior, and inflammation.

A) Bray-Curtis distances (a measure of beta-diversity) principle coordinate plot. The relative abundance of B) Romboutsia, C) Turicibacer, and D) Tyzzerella in the distal colon of mice was assessed on the same day as the last injection of vehicle or chemotherapy. n=8-13. E) Central tendency, calculated by beam breaks, as measured 6 h after the final injection over a 30 min test. n=6-9. F) Locomotion measured by beam breaks in the open field arena 6 h after the final injection over a 30 min test. n=6-9 G) IL-1β in the hippocampus of mice the same day after the final injection. No significant comparisons, n=5-9. H) Correlation plot of relative abundance of Romboutsia vs. Central Tendency in Chemo group, Spearman r=−0.88, p=0.0034. n=9. I) Correlation plot of relative abundance of Turicibacter vs. Locomotion in Chemo group, Spearman r=−0.73, p=0.031. n=9. J) Correlation plot of relative abundance of Tyzzerella vs. hippocampal IL-1β in Chemo group, Spearman r=0.85, p=0.0061, n=9. Mean ± SEM. *p<0.05, **p<0.01, ***p<0.005 by post-hoc test

Again, chemotherapy increased anxiety-like behavior, as measured by central tendency in the open field (Figure 2E. F_1,26=11.5, p<0.01, Veh vs. Chemo, p<0.005). Antibiotics tended to decrease central tendency, and there was a significant interaction between antibiotics and chemotherapy driven by the reduction in central tendency in the vehicle plus antibiotics group compared to vehicle control group (Figure 2E. antibiotic effect, F_1,26=3.1, p=0.090; chemotherapy by antibiotic interaction, F_1,26=7.1, p<0.05, Veh vs Veh+Abx, p<0.001). Chemotherapy significantly reduced locomotion, a measure of fatigue, in the open field test (Figure 2F. F_1,25=19.9, p<0.001, Veh vs Chemo, p=0.06, Veh+Abx vs Chemo+Abx, p<0.01), while antibiotics had no effect on locomotion.

The concentrations of inflammatory cytokines, IL-1β, IL-2, IL-6, and TNFα, and chemokine CXCL1, were measured in the hippocampus, hypothalamus and frontal cortex, brain regions implicated in memory and mood. As anticipated, Chemo increased CXCL1 in all brain regions, IL-6 in the hippocampus, TNFα in the hypothalamus, and IL-1β in the frontal cortex (Supplemental Table 2) (Sullivan et al., 2021). Antibiotic chow increased IL-2 in the hypothalamus and IL-1β and CXCL1 in the frontal cortex compared to control chow (Supplemental Table 2).

Correlation analyses were completed between the bacterial genera, inflammatory outcomes, and behavioral outcomes that were significantly altered by treatment to test the a priori hypothesis that microbial abundance would be predictive of chemotherapy-associated inflammation and behavioral side effects. The relative abundance of Romboutsia, which was enhanced in the distal colon of chemotherapy-treated mice without antibiotics, was associated with greater anxiety-like behavior (i.e., reduced central tendency) measured in the open field test (Figure 2H; Chemo group, Spearman r=−0.88, p<0.01). Additionally, the relative increase in Turicibacter in the distal colon of chemotherapy-treated mice without antibiotics was associated with lethargy (i.e., decreased locomotion) in the open field test (Figure 2I; Chemo group, Spearman r=−0.733, p<0.05). The increased relative abundance of Tyzerella positively associated with neuroinflammation (i.e., hippocampal IL-1β concentration) (Figure 2J; Chemo group, Spearman r=0.85, p<0.01).

Compositional differences in gut microbiota can result in differences in their metabolic activity. Given the alterations in microbial composition and behavior of chemotherapy-treated mice (Figure 2B-D, Supplemental Figure 1B-C, Supplemental Figure 2C-D, and Loman et al. 2019), modifications in metabolites known to signal along the gut-brain axis were investigated. Chemotherapy did not impact the colonic concentrations of total short-chain fatty acids, acetate, propionate, or butyrate (Table 1). As expected, antibiotic treatment drastically reduced concentrations of total short-chain fatty acids, acetate, propionate, and butyrate in both the proximal and distal colon (Table 1). Interestingly, there was a chemotherapy by antibiotic interaction in the propionate concentration in the distal colon; chemotherapy lowered concentrations of propionate, however, Chemo+Abx elevated propionate concentrations relative to mice treated with Veh+Abx (Table 1. F_1,18=6.3, p<0.05).

Table 1.

Short chain fatty acids concentration in colon contents

	Proximal Colon				Distal Colon
	Total (μmol/g)	Acetate (μmol/g)	Propionate (μmol/g)	Butyrate (μmol/g)	Total (μmol/g)	Acetate (μmol/g)	Propionate (μmol/g)	Butyrate (μmol/g)
Main effect	Antibiotic**** F1.36=100.9	Antibiotic**** F1.36=68.9	Antibiotic**** F1.36=26.1	Antibiotic**** F1.36=114.2	Antibiotic*** F1.24=20.4	Antibiotic* F1.24=6.2	Antibiotic** F1.24=10.0	Antibiotic**** F1.24=104.4
Vehicle	30.7 ± 1.9	20.2 ± 1.3	4.8 ± 0.4	5.7 ± 0.5	18.8 ± 3.2	11.1 ± 2.2	3.7 ± 0.5	4.0 ± 0.7
Vehicle + Antibiotic	12.8 ± 0.8****	10.1 ± 0.7****	2.5 ± 0.4***	0.18 ± 0.18****	10.3 ± 1.3*	8.9 ± 0.9	1.4 ± 0.6*	0.00 ± 0.0****
Chemo	34.0 ± 3.2	22.5 ± 2.4	4.42 ± 0.33	7.06 ± 0.85	21.3 ± 2.1	13.4 ± 1.7	3.66 ± 0.33	4.28 ± 0.44
Chemo + Antibiotic	116 ± 0.97****	8.11 ± 0.57****	2.64 ± 0.030*	0.803 ± 0.51****	10.2 ± 1.0*	7.64 ± 1.0	2.56 ± 0.0066	0.00 ± 0.0****

3.3. Gut-microbiota transplant (GMT) from chemotherapy-treated mice lead to anxiety-like behavior and inflammation in recipient mice.

To directly test the effects of microbial composition on inflammatory and behavioral responses, gut microbes (cecal and proximal colonic contents) from chemotherapy- or vehicle-treated conventional donor mice were transplanted as a single intra-gastric gavage to germ-free recipient mice (Figure 3A). As expected, donor mice receiving chemotherapy lost body mass from beginning to the end of treatment, whereas vehicle-treated donor mice maintained their body mass (Figure 3B, F_1,19=24.55, p<0.0001, Veh Donor vs Chemo Donor, p<0.0001). In previously germ-free recipients, neither Veh Recipient nor Chemo Recipient mice lost weight from the time of GMT until the day of behavior testing (7 d post-gavage). Donor mice receiving chemotherapy decreased their food consumption, but there was no difference in food intake between Veh GMT and Chemo GMT mice (Figure 3C,Veh Donor vs Chemo Donor, U=13, p<0.05). To ensure successful GMT transmission of microbiome composition and assess treatment-induce differences, 16S rRNA gene sequencing was performed on samples from Veh GMT and Chemo GMT mice as well as the GMT material. Indicative of successful GMT, each group of recipients clustered by Bray-Curtis (beta diversity) distances (Figure 3D). Differences in beta diversity were underpinned by altered microbial relative abundances at the genus level. Chemo GMT mice had significantly different relative abundances of Eysipelato-clostridium, Rhodospirllales uncultured, Ruminococcus 1, Lachnospiraceae UCG-006, Eysipelotrichaceae uncultured, Intestinomonas, Ruminococcus 2, and Eisenbergiella in the proximal colon relative to Veh GMT mice (Figure 3E).

Figure 3. Intra-gastric gavage of cecum and proximal colon content.

A) Graphic of timeline for donor mice injection of vehicle or chemo and intra-gastric gavage (gut microbiota transplant, GMT) of cecum and proximal colon content of mice to germ-free recipient mice, behavior testing, and tissue collection. B) The percent change in body mass in donor (from first injection to final injection) and recipient (from GMT to post-GMT time point) mice. Drug effect, p<0.0001 by two-way ANOVA and Bonferroni’s multiple comparisons test ****p<0.0001, n=9-12. C) Average food intake/mouse/cage/24 h calculated over the previous 48 h in donor mice and from the day of GMT to the day of tissue collection in recipient mice. Mann-Whitney non-parametric test to determine difference between Veh Donor and Chemo Donor, *p<0.05. n=3-8. Mean ± SEM D) Brey-Curtis distance principle coordinate plot A) The relative abundance (%) of multiple bacterial genera in Veh GMT and Chemo GMT mice proximal colon content (grey and blue circles, respectively), and in the Veh Donor and Chemo Donor material (black circles).

The GMT of chemotherapy-treated mice to germ-free mice resulted in the transfer of a behavioral phenotype in the open field test. Consistent with the previous experiments, chemotherapy reduced total, peripheral, and center distance traveled in conventional donor mice as compared to vehicle-treated donor mice (Figure 4A, Veh Donor vs Chemo Donor; total, t(17)=5.0, p=0.0001; periphery, t(17)=4.4, p<0.001; center, U=11, p<0.005). GMT Chemo did not decrease total or peripheral distance traveled, however there was a reduction specifically in the distance traveled in the center of the open field chamber when compared to GMT Veh mice, similar to that of the chemotherapy-treated donor mice (Figure 4A,Veh GMT vs Chemo GMT, t(13)=2.3, p<0.05).

Figure 4. Transmission of chemotherapy-treated phenotype by GMT in germ-free mice.

A) Total distance, peripheral distance, and center distance traveled in a 16 x 16 open field arena over 30 min. Unpaired t-test or Mann-Whitney non-parametric test, *p<0.05, **p<0.01, ***p<0.005, n=7-15. B) IL-1β, IL-6, TNFα, and CXCL1 measured in plasma of donor and recipient mice on the day of tissue collection by multi-plex cytokine array. IL-1β not detected in Veh GMT or Chemo GMT groups. Unpaired t-test or Mann-Whitney non-parametric test, *p<0.05, n=7-12. C) Hippocampal mRNA expression of Il1b, Il6, Tnfa, and Cxcl1 measured by RT-qPCR and reported as fold change as compared to Veh Donor or Veh GMT. Unpaired t-test, *p<0.05. n=7-11. Mean ± SEM

Elevated levels of peripheral and central inflammatory mediators were detected in both donor and recipient mice. For example, chemotherapy treatment increased circulating concentrations of IL-6 when compared to vehicle treatment in donor mice (Figure 4B, Veh Donor vs Chemo Donor, U=15, p<0.05) and chemo GMT elevated CXCL1 plasma concentration as compared to Veh GMT (Figure 4B, Veh GMT vs Chemo GMT, t(14)=2.7, p<0.05). Similarly, chemotherapy increased hippocampal mRNA levels of Il1b and Tnfa compared to vehicle in donor mice (Figure 4C; Veh Donor vs Chemo Donor; Il1b, U=22 p=0.01, Tnfa, U=26 p<0.05) and chemotherapy GMT increased hippocampal mRNA of Il6 and Tnfa in recipient mice compared to vehicle GMT (Figure 4C; Veh GMT vs Chemo GMT; Il6, t(13)=2.5 p<0.05, Tnfa, t(13)=2.0 p=0.07). Chemotherapy nor Chemotherapy GMT increased inflammatory gene expression in the hypothalamus or frontal cortex of donor or recipient mice (Supplemental Figure 4A and B).

4. DISCUSSION

Cancer treatment significantly reduces patient quality of life and treatment adherence due to side effects including fatigue, anxiety, and gastrointestinal distress. It is evident that alterations in gut microbial populations can impact these mental and physical health outcomes (Lee et al., 2014), although this relationship had not yet been studied in the context of chemotherapy. Our previous study established a correlation between changes in gut microbial community, inflammation, and chemotherapy-induced behavioral side effects in mice (Loman et al., 2019), whereas in the present study we used a multi-pronged approach to identify the role of the gut microbiome in the development of acute chemotherapy-associated inflammation and fatigue.

One limitation of using fecal microbiota transplant approaches for studies of brain and behavior is the confounding stress effects of intra-gastric gavage on these endpoints (Brown et al., 2000; Raio and Phelps, 2015). Therefore, in our first experiment, we took advantage of the propensity of mice to naturally engage in coprophagia, and thereby non-invasively exchange fecal microbes. Coprophagia significantly impacts the microbial makeup of the gastrointestinal tract, increasing the microbial load and altering biodiversity compared to mice prevented from engaging in coprophagia (Bogatyrev et al., 2020). To increase the likelihood of a test mouse ingesting feces from mice of the opposite treatment, the experiment was designed with 1 test mouse to every 3 oppositely-treated cage mates. It should be noted, however, that coprophagia may occur between all mice in a single cage and the study reported here did not quantify the amount of microbial transfer, though, the transfer of microbes has been previously documented (Ridaura et al., 2013). Here, vehicle-treated mice that were housed with chemotherapy-treated mice had less body fat than vehicle-treated mice that were housed together and approximately the same body fat composition as chemotherapy-treated mice. These data suggest that the consumption of feces can transfer chemotherapy-associated body composition, and based on previous work, potentially transfer chemotherapy-associated metabolic hormone signaling (Sullivan et al., 2021). This finding is supported by the report that mixed housing of recipients of FMT from obese and lean twin mice prevents the capacity to relay the adiposity phenotype, which is observed when housing is not mixed between treatments (Ridaura et al., 2013). Furthermore, neoadjuvant chemotherapy reduces fat mass in oesophagogastric cancer patients (Awad et al., 2012).

Consistent with our previous data in Balb/c and C57BL/6 mice (Loman et al., 2019; Sullivan et al., 2021), proinflammatory cytokines, IL-1β, TNF-α, and IL-6, were elevated following paclitaxel treatment in the present study. In chemotherapy-treated mice that were cohoused with vehicle-treated mice, there was a significant reduction in circulating concentrations of IL-1β, suggesting that microbes acquired from healthy control mice via coprophagia may attenuate the IL-1β response to chemotherapy. Aside from affecting body composition and circulating markers of inflammation, there were few other robust effects of cohousing; however, the present results warranted the following subsequent investigations using more invasive approaches.

As a more controlled approach to investigating the extent to which the gut microbiome drives chemotherapy-induced changes in inflammation and behavior, mice were fed chow containing an antibiotic cocktail prior to and during vehicle or paclitaxel injections. Continuous antibiotic consumption robustly altered the gut microbial composition and significantly dampened the relative abundances of microbes that were increased in antibiotic-free chemotherapy-treated mice as compared to vehicle-treated mice.

Alterations in colon microbial profiles in antibiotic- and chemotherapy-treated mice provided the opportunity to correlate behavioral and inflammatory profiles with microbes in chemotherapy-treated mice. Similar to our previous study, chemotherapy treatment caused an increase in Romboutsia (Loman et al., 2019). The correlation of anxiety-like behaviors and Romboutsia and Turicibacter suggest a role for these microbes in the development of an anxiety-like phenotype. Further experiments that specifically increase the relative abundance of Romboutsia and Turicibacter in mice would be needed to support a causal role of these microbes in the development of behavioral side effects. In a mouse model of anxiety induced by chronic unpredictable mild stress (CMS), Romboutsia was also increased (Sun et al., 2019), further implying a relationship between this genus and anxiety. The presently observed positive correlation between hippocampal IL-1β and Tyzzerella relative abundance suggests that a relative increase in this genus after chemotherapy may contribute, in part, to neuroinflammation. Taken together, these correlations further bolster evidence for a chemotherapy-gut microbe-brain axis and exposes new microbial taxa as potential targets of intervention. It should be noted that very few bacterial species survived chronic treatment with the broad-spectrum antibiotic chow. There was, however, a marked antibiotic-induced increase in the relative abundance of Lactobacillus, consistent with its documented relatively high rate of resistance to antibiotics (Gueimonde et al., 2013). Interestingly, Lactobacillus is frequently associated with gut health (Sanders et al., 2019). Conversely, the relative abundance of Escherichia coli, a common pathogen, increased in antibiotic-treated animals, and was further elevated by chemotherapy. In parallel, the pro-inflammatory mediator, IL-1β, increased in the frontal cortex of antibiotic-treated mice, which was exacerbated by chemotherapy. Together this suggests that chemotherapy may selectively potentiate the growth of gut bacteria that modulate brain inflammation. Unlike our previous study, we did not demonstrate a reduction in α-diversity induced by chemotherapy treatment (Loman et al., 2019). However, this is not unexpected considering the well-documented differences in gut microbiome among inbred mouse strains (Korach-Rechtman et al., 2019; Xiao et al., 2015), and specifically between BALB/c (Loman et al., 2019) and C57BL/6 (presented here) mice. Of note, antibiotics alone altered behavior in control mice, potentially caused by coincident alterations in SCFA concentrations, as demonstrated by an increase in anxiety-like behavior (i.e., decreased central tendency). This is consistent with a previous study in which ampicillin decreased central distance in an open field chamber and increased the amount of time spent in the closed arm of the elevated plus maze, both anxiety-like behaviors, in male, juvenile BALB/c mice (Ceylani et al., 2018). Similarly, antibiotics increased select proinflammatory cytokines in the brain. This too is consistent with findings by Leclercq et al. (2017), in which they demonstrated early life exposure to penicillin caused increased proinflammatory cytokines in the brains of mice. While the hypothesis of this experiment was that antibiotic-induced reductions in gut microbes would lessen the effects of chemotherapy on inflammation and brain-mediated side effects, it is conceivable that the prolonged antibiotic dosing masked any mitigation of chemotherapy side effects driven by the microbiome.

Expanding to other potential gut-brain pathways that may be involved with chemotherapy beyond inflammation, prominent short chain fatty acids (SCFAs), acetate, propionate, and butyrate were analyzed in the colon contents of mice. Bacterial fermentation in the lower gut produces SCFAs that are implicated in brain functions including microglial maturation and the maintenance of blood brain barrier integrity (Den Besten et al., 2013; Silva et al., 2020). Importantly, ingesting fructo- and galacto-oligosaccharides increase cecal contents of SCFAs in mice, and decrease anxiety-like behaviors in the open field test and elevated plus maze (Burokas et al., 2017). Here, we hypothesized that chemotherapy decreases SCFAs in the colon of mice and associate with increased anxiety-like behaviors. However, the only chemotherapy-driven difference observed was an antibiotic-induced decrease in distal colon propionate concentration of vehicle-treated mice that was absent in chemotherapy-treated mice. To fully rule out the possibility of a SCFA gut-brain pathway with chemotherapy, future studies should assess SCFAs in other gastrointestinal regions and in circulation. As mentioned above, it should be noted that mice in this experiment were subjected to moderately high doses of antibiotics for 8 days prior to receiving chemotherapy and through the 11-day treatment regimen. Therefore, potential chemotherapy by antibiotic interactions may be uncovered if a lower dose of antibiotics were employed or if a shortened regimen was followed.

The final approach to identify potential causal relationships between the gut microbiome and the development of chemotherapy-associated neuroinflammation and behavioral side effects was an intra-gastric gavage (i.e., gut microbiota transplant, GMT) of cecal and proximal colon content from chemotherapy- or vehicle-treated conventional donor mice into germ-free recipient mice. Notably, this gut content could contain more than just bacteria (e.g., drug metabolites, bacterial metabolites, viruses, etc.), although paclitaxel concentrations were confirmed to be negligible. Sickness behaviors were not transferred by the GMT as determined by the absence of weight loss and anorexia in germ-free mice that received chemotherapy GMT. However, the gut microbial composition of recipient mice that received the GMT clustered with the microbial composition of the donor material as determined by Bray-Curtis distances for both Veh GMT and Chemo GMT groups. Remarkably, GMT was sufficient to transfer select behavioral side effects of chemotherapy treatment. Consistent among all 3 experiments, the donor chemotherapy-treated mice displayed anxiety-like behavior, which was passed on to their GMT recipients. In contrast, the lethargic phenotype of donors was not transferred via GMT. Taken together, these data indicate that anxiety-like behavior associated with chemotherapy treatment may be more directly mediated by gut microbes while lethargy may be a direct result of paclitaxel treatment. The ability of gut microbes to encode signals for anxiety-like behavior is also demonstrated by the colonization of germ-free NIH Swiss mice (less anxious) with microbes from BALB/C mice (more anxious) causing an increase in the step-down latency of NIH Swiss mice (Bercik et al., 2011). Furthermore, CMS induces anxiety-like behavior as determined by open field and elevated plus maze tests, and that phenotype is transferred to mice that received a fecal microbiota transplant from CMS donors (Li et al., 2019). Elevated inflammatory cytokines and chemokines in circulation and elevated hippocampal inflammatory gene expression were observed in donor mice and notably circulating inflammatory chemokine elevation was transferred into recipient mice via GMT. This transference of an inflammatory phenotype by microbiota transplant has been observed in a study of high-fat diet; transplant of control mouse feces attenuated inflammation (Zhou et al., 2017). This suggests that gut microbes contribute to the development of peripheral and central inflammation due to chemotherapy treatment. It is not surprising that the exact same inflammatory markers are not elevated in the donor and recipient mice, as inflammatory pathways are highly redundant and compensatory (Haddad, 2002), and dependent on enteric microbial composition (Schirmer et al., 2016). In fact, the duration required for germ-free mouse gut physiology and microbiome to resemble that of donor mice may require up to a 9 week colonization period (Le Roy et al., 2019).

5. CONCLUSION

The data presented here significantly contribute to the field of psychoneuroimmunology because, for the first time, a multifaceted approach was taken to demonstrate a clear association between paclitaxel chemotherapy-induced gut microbiome disruption and subsequent inflammation and behavioral changes. Future studies adding tumor exposure and surgery into the model would identify potential interactions among these key cancer treatment events on the relationship between the microbiome and side effects of chemotherapy. The present results warrant further studies on the potential prevention and treatment of chemotherapy-induced behavioral symptoms through the use of prebiotics, probiotics, or microbial transplant therapies. The use of such treatment strategies would revolutionize cancer treatment leading to a significant increase in patient quality of life and decrease in mortality.

HIGHLIGHTS.

• Passive FMT by coprophagia attenuates paclitaxel-induced inflammation.

• Gut microbes associate with chemotherapy-induced inflammation and behavior.

• Gut microbial transplants from paclitaxel-treated mice cause anxiety-like behavior.

• Chemotherapy GMT causes neuroinflammation in germ-free recipient mice.