Authors: Yonaida A. Valentine, Audrey F. Duff, Michael T. Bailey, Leah M. Pyter
Categories: Research Article, Chemotherapy, bacteriome, metabolome, gut-brain Axis, inflammation, neuroinflammation, fatigue, anxiety, paclitaxel, cyclophosphamide, cisplatin, doxorubicin
Source: Gut Microbes
A systematic assessment of the short- and long-term effects of commonly used breast cancer chemotherapeutics on the gut microbiome‒blood‒brain axis of female mice
Authors: Yonaida A. Valentine, Audrey F. Duff, Michael T. Bailey, Leah M. Pyter
Abstract
Chemotherapy affects over 300,000 U.S. breast cancer patients, which disrupts the gut microbiome and induces gut inflammation—an effect hypothesized to drive gastrointestinal side effects (e.g., diarrhea, vomiting) experienced by 50%–80% of patients. Preclinical studies have found causal links amongst chemotherapy-induced gut microbiome disruption, systemic inflammation, and brain-mediated side effects. Therefore, the gut microbiome represents a therapeutic target to attenuate chemotherapy side effects. Because clinical populations are administered multiple chemotherapeutics in combination, a comprehensive understanding of which treatments disrupt the gut microbiome‒blood‒brain axis is lacking. Here, translationally-relevant regimens of four commonly used breast cancer chemotherapies (paclitaxel, cyclophosphamide, cisplatin, and doxorubicin) were given to adult female C57BL/6 mice, and inflammatory, metabolomics and/or bacteriome outcomes were measured in the gut, gut contents, blood, and brain tissues, along with a fatigue and anxiety-like behavioral assessment. Many inter-chemotherapy differences were observed but notable findings include prolonged circulation and central proinflammatory signals by paclitaxel and sustained disruption of the gut microbiome by cisplatin. In contrast, cyclophosphamide and doxorubicin modestly disrupted the gut microbiome‒blood‒brain axis. Taken together, this study systematically identified that paclitaxel and cisplatin most robustly disrupted the gut microbiome‒blood‒brain axis, suggesting that those treated with these drugs may benefit the most from gut-targeted interventions for associated side effects.
Introduction
In the United States, over 300,000 people are diagnosed with breast cancer every year, and chemotherapy is one of the most common treatments.^1^ Chemotherapy treatment (a regimen of anti-cancer chemotherapeutics) is highly effective at killing rapidly dividing tumor cells, contributing to an impressive 91% five-year survival rate following initial breast cancer diagnosis.^1^ However, chemotherapy is not tumor cell-specific and therefore induces many debilitating side effects by damaging healthy dividing cells throughout the body, including intestinal epithelial cells.^2^ Indeed, gastrointestinal side effects (e.g., diarrhea, vomiting) are common (50%–80%) and can result in early termination or dose reductions in these cancer-saving treatments (reviewed in^3^^,^^4^). These side effects are associated with chemotherapy-induced disruption of the gut microbiome in breast cancer patients,^5^^,^^6^ and shifts in microbial metabolites essential for the health of the gastrointestinal tract. Clinical and preclinical studies have demonstrated that chemotherapy differentially alters the relative abundance of bacterial taxa in the gut and thereby shifts the bacteriome community composition.^5^^,^^7^ Some preclinical, but fewer clinical studies, have further identified that chemotherapy also disrupts metabolites produced by gut microbes, such as short-chain fatty acids,^8^ bile acids,^9^ and tryptophan metabolites.^10^^,^^11^ Finally, studies using rodent models demonstrate that chemotherapy increases gut tissue proinflammatory mediators,^7^ an effect that can reduce gut barrier integrity^12^ and tight junction protein expression.^13^
Beyond gastrointestinal side effects, chemotherapy also induces brain-mediated side effects in breast and other cancer populations (e.g., fatigue and symptoms of mood).^14^^,^^15^ Because most chemotherapeutics do not readily cross the blood-brain barrier (reviewed in^16^) these brain consequences are hypothesized to be caused indirectly via signaling from the periphery to the brain (reviewed in^17^). Indeed, mounting evidence suggests that some of the gastrointestinal side effects may be related or contribute to the brain side effects (reviewed in^17^^,^^18^) Leading hypotheses suggest that humoral communication between the gut microbiome and the brain can occur via circulating immune signaling (i.e., proinflammatory mediators) and gut derived metabolites. For example, in our previous research, the transplantation of paclitaxel chemotherapy-altered gut microbiota was sufficient to increase the levels of proinflammatory mediators in circulation and in the hippocampus of germ-free recipient mice, as well as causing anxiety-like behavior.^19^ However, understanding how chemotherapy affects the gut microbiome (including the bacteriome, metabolome, and physiology)-blood‒brain communication is complicated by simultaneous or consecutive administration of drugs as combination therapies in the clinic.^20^ Identifying the chemotherapeutics that have the greatest impact on the gut microbiome‒blood‒brain axis would help to elucidate which regimens may benefit the most from novel, microbiota-directed interventions to attenuate the gastrointestinal and possibly brain-mediated side effects of chemotherapy.
Thus, the present study used controllable mouse models to comprehensively compare the extent to which four common types of breast cancer chemotherapeutics, namely, taxanes (paclitaxel), alkylating agents (cyclophosphamide), platinum-based alkylating agents (cisplatin), and anthracyclines (doxorubicin), disrupt 4 tissues across this axis and, over time, with coincident assessment of fatigue and affective-like sickness behavior. The goal was to identify chemotherapies that may be most suitable for gut microbiome-targeted interventions for future studies. In contrast, current preclinical studies have primarily focused on a single chemotherapeutic agent and outcomes in a single tissue, as well as assessing the consequences cross-sectionally.
Materials and methods
Animals
Female 8–10-week-old C57BL/6 nulliparous mice (Charles River, Wilmington, MA, USA) from mixed litters were group housed (2–5 mice/cage), and acclimated to a 10 dark cycle (lights off at 1300 h) in a temperature-controlled vivarium (22 °C ± 1 °C) for approximately 3 weeks before any experimental procedures were performed. During the 3-week housing acclimation period, soiled bedding was mixed from all the cages and redistributed evenly, twice weekly, to standardize the microbiome amongst all the mice, as previously described.^19^ Female mice were used in all the experiments, as an estimated 99% of all new breast cancer cases in the U.S. are women.^21^ Standard rodent chow (Teklad LM-485, irradiated, Envigo, Indianapolis, IN, USA) and water were available ad libitum throughout the duration of the study. The body mass and average food intake of the mice in each cage were assessed at baseline (i.e., 0% of the chemotherapy regimen completed), the day of each injection, and one day after each injection during treatment, as well as at a 7-d interval after treatment for the longer-term cohort. The average intake was calculated as [(food mass at the time interval before measurement—food mass the day of measurement)/number of mice per cage)]/time interval. All animal experiments were approved by The Ohio State University 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.^22^ All efforts were made to minimize animal suffering and reduce the number of mice used.
Experimental design overview
After the bedding mix, mice were administered one of 4 repeated intraperitoneal regimens of a single chemotherapeutic (paclitaxel, cyclophosphamide [CycloP], cisplatin, or doxorubicin) or one of 2 controls (Cremophor-EL [CremEL] or phosphate buffer saline [PBS]). To assess both the short- and long-term consequences of these chemotherapy regimens, two cohorts were run with final tissues collected either 1 or 28 d after regimen completion, respectively. Body mass and food intake were measured throughout the duration of each experimental cohort. Fatigue and anxiety-like behavior were assessed 6 h and 27 d after the final injections in the short- and longer-term cohorts, respectively. Feces, gut contents, ileums, plasma, and a brain region proximal to the ventricular supply of circulating mediators and a regulator of affective-like and locomotor behavior (hippocampi), were collected from all the mice to assess changes relevant to the gut (e.g., physiology, bacteriome, metabolome), blood (e.g., inflammation, metabolome), and hippocampus (e.g., inflammation, metabolome), as detailed below.
Chemotherapy administration
The four types of chemotherapeutics chosen work differently to inhibit cell division through arresting microtubule dynamics and DNA intercalation (reviewed in^23^). All the groups were treated at the concentrations and schedules listed in Table 1. These regimens were selected to represent the animal equivalent dosage^24^ of each chemotherapeutic when given as a monotherapy to individuals with breast cancer^25^ and were previously reported to disrupt the gut, peripheral immune function, central immune function and/or behavior in mice.^7,26‐28^ Since responses to chemotherapy may differ by the time of day that chemotherapeutics are administered,^29^ all chemotherapies (and their respective control groups) were administered at randomized times across the light cycle.
The sources and catalog numbers of the drugs paclitaxel (#P-9600, LC Labs, Woburn, MA, USA), CycloP (#C0768, Sigma Aldrich, St Louis, MI, USA), cisplatin (#232120, Sigma Aldrich), or doxorubicin (#D1515, Sigma Aldrich). Immediately before administration, paclitaxel was prepared in a 50 CremEL:PBS vehicle solution^19^; all other chemotherapeutics were reconstituted in sterile saline.
Tissue collections
Fecal samples used for 16S rRNA bacterial sequencing analyses were collected immediately before (baseline) and 1 and 28 d after the final dose of vehicle/chemotherapy in the mice from the long-term cohort. Owing to a change in the institutional protocol, fecal samples were not collected from all the mice at the 1 d post treatment timepoint which resulted in a lower sample size pairwise comparison of longitudinal β-diversity shifts. For other tissues collected, the mice from both cohorts were euthanized via rapid decapitation. Blood was collected in EDTA-lined tubes and stored on ice. Whole blood was then centrifuged for 20 min at 2500 × g at 4 °C, and the plasma was collected and stored at −80 °C for subsequent cytokine, lipopolysaccharide binding protein (LBP), and targeted metabolomics assays. To prioritize the effect of chemotherapy on inflammation within the gut, two cm of ileal tissue was dissected immediately proximal to the ileocecal junction as previously described.^30^ The digestate was flushed out with cold PBS and the tissue was immediately frozen on dry ice, then stored at −80 °C prior to RT-qPCR processing to assess the gene expression of markers related to gut barrier integrity, water retention, and inflammation in the ileum, the site within the gut of greatest immune cell density (reviewed in^31^). Each side of the hippocampus was dissected from the sagittally-bisected brain, immediately frozen on dry ice, and then stored separately at −80 °C. One side was chosen at random for RT-qPCR analyses of glial reactivity and inflammatory markers, and the other for metabolomics. Spleens were dissected out and kept at 4 °C until completion of tissue collection and then weighed to assess gross immune activation. All tissues were collected during the light phase (0800–1200 h).
RT-qPCR
Qiagen RNeasy Lipid Tissue Mini Kits (Qiagen, Germantown, MD, USA) were used to extract total RNA from the ileum and hippocampus samples, as previously described.^19^ Briefly, RNA was reverse transcribed using qScript cDNA SuperMix (QuantBio, Beverly, MA, USA) per the manufacturer's instructions. Gene expression of proteins related to gut barrier integrity (Ocln, Tjp1), water retention (Aqp1, Aqp11), and inflammation (Ccl2, Il1b, Il6, Il17a, Il17ra, Tnf) was assessed in the ileum. Gene expression of pro- and anti-inflammatory mediators (Il1b, Il6, Il17ra, Tnf, Ccl2, Cxcl1, Tgfb1), inflammatory cascade mediators (Myd88, Nlrp3, Tlr4), and glial reactivity (Aif1, Gfap) was assessed in the hippocampus based on previous studies.^7^^,^^19^ Gene expression was normalized to the geometric means of B2m, Gapdh, and Hprt and reported as the fold change relative to the control group (2^−^^∆∆CT^). Prior to analysis, it was confirmed that the expressions of B2m, Gapdh, and Hprt were not significantly different among the treatment groups.
16S rRNA bacterial gene sequencing and analyses
Fecal samples were collected at baseline, 1, and 28 d post-chemotherapy from the longer-term cohort were extracted for 16S sequencing as previously described.^19^ Briefly, paired-end (300 bp forward and reverse) sequences of the V4 hypervariable region of the 16S rRNA gene were generated from samples via Illumina NextSeq. The sequences were processed using Quantitative Insights Into Microbial Ecology (QIIME) 2.0,^32^ primers were removed with Cutadapt, and DADA2 was used for downstream amplicon processing, denoising, and quality control. Forward and reverse reads were truncated from the 3′ end to 240 nucleotides to achieve an average quality score of approximately 20, and sequences that failed to meet the quality control criteria were discarded. Taxonomy was assigned using a trained classifier constructed from the SILVAv138.99 ribosomal RNA database, and features denoted as “Eukaryota”, “Unassigned”, “Chloroplast”, “Mitochondria”, not annotated beyond the phylum level, and not present in at least 2 samples were filtered from the dataset. For diversity analyses, a sequencing depth of 79,936 was selected based on rarefaction curves, and data were generated using the core-metrics-phylogenetic QIIME 2.0 pipeline. Relative abundance of taxonomic assigned features were visualized (Supplemental Figures 4 and 5). Faith's index of phylogenetic diversity (PD), Shannon's index, and Pielou's evenness index were used to quantify α-diversity. Weighted UniFrac Distance, Unweighted UniFrac Distance and Bray‒Curtis matrices were used to assess β-diversity. Prior to analyzing differential abundance, all samples were conditionally filtered to retain features with an abundance of at least (1/rarefaction depth) in at least 10% of the samples to minimize noise and underpowered sequences. Microbiome diversity statistics were analyzed in QIIME 2.0 using respective plugins. Alpha-diversity was analyzed by two-way ANOVA with Benjamini and Hochberg post-hoc in GraphPad Prism version 9.3.1 (GraphPad Software, San Diego, CA, USA). Beta-diversity was analyzed by permutational multivariate analysis of variance (PERMANOVA) with 999 randomizations of the data in QIIME 2.0. The differential abundance of bacterial taxa was assessed by ANCOM-BC^33^ using the QIIME 2.0 composition plugin. The data were considered significant when Holm's FDR-adjusted p-values (q-values) ≤ 0.05.
Targeted metabolomics and analyses
The metabolomics data were generated by the Host-Microbiome Metabolomics Facility (HMMF) within the Duchossois Family Institute at the University of Chicago. Gut (cecum/proximal colon) contents, EDTA-collected plasma, and hippocampi were collected terminally at 1 and 28 d post-treatment for targeted metabolomics from both cohorts of mice. Targeted metabolomics measures metabolites that are related to tryptophan metabolism, short-chain fatty acids, and bile acids to systematically assess how chemotherapeutics differentially alter the metabolome and gut-to-brain signaling, as these groups of metabolites have previously been demonstrated to be associated with microbial metabolism, altered by chemotherapy, and may mediate brain consequences ^9^^,^^34^^,^^35^. Metabolites from the tryptophan metabolism and bile acid panels were extracted using LC/MS. Metabolites from the short-chain fatty acid panels were extracted using GC/MS. Metabolite validation (including extraction methods, retention time, and fragmentation), alignment to spectral libraries, metabolite abundance normalization, and metabolite abundance quantification were performed.^36^ A subset of metabolites in each panel were quantified using standards and blanks. Namely, 17 of 36 for the tryptophan panel, 8 of 50 for the short-chain fatty acid panel, and 15 of 62 for the bile acid panel (detailed in Supplemental File 1). For metabolites where greater than 50% of the dataset was not detected (for normalized abundance) or quantified, statistical analyses were not performed. The normalized abundance of all metabolites in each panel assessed was used for principal component analysis (PCA) to determine whether the metabolomes were different from those of the controls. Subsequently, enriched pathways contributing to metabolome composition differences and the quantified metabolites that were differentially abundant were identified. PCAs and quantitatively enriched pathway analyses were completed via MetaboAnalyst 6.0 (McGill University, Montreal, QC, Canada) using standard approaches.^37^ Briefly, for principal component analyses, given that the number of normalized abundance metabolites encompassed a wider array of compounds potentially contributing to these metabolic pathways, the normalized abundance of metabolites was used to determine if the compositional nature of the tissue-specific metabolome differed from that of the controls. The data were autoscaled to weigh all the metabolites equally. PCAs were analyzed by permutational multivariate analysis of variance (PERMANOVA) with 999 randomizations of the data, and the data were significant when Holm's FDR was q < 0.05. Treatment groups that were significantly different from controls by PCA were further investigated to determine which metabolic pathways may drive this difference using standard approaches.^37^ Briefly, for (quantitatively enriched) pathway analyses, normalized abundance data were used as scaled as described for PCAs. Pathways were then annotated via the Small Molecule Pathway Database as it encompassed a 10% greater metabolite set than the KEGG database and also referenced metabolites related to human metabolic pathways.^38^ Pathways were assigned an enrichment ratio based on the number of hits present in the sample relative to all metabolites in the pathway, weighing the statistical significance (via Student's t-test comparing each chemotherapy group to control) of each hit. Pathways were considered significantly enriched when p < 0.05, and the top three pathways from each comparison were visualized. The quantified metabolites were statistically analyzed by one-way ANOVA with Fisher's LSD. Metabolites that were statistically significant between at least one chemotherapeutic and controls were visualized via radarcharts with the R programming language (version 4.3.1) using tidyverse^39^ and fsmb^40^ packages.
Microbe‒metabolite correlations
To gain deeper insight into how chemotherapeutics may change microbe‒metabolite relationships, secondary correlational analysis was performed using 16S rRNA sequencing and targeted metabolomics data from the longer-term cohort at the 28 d timepoint, as these mice had both datasets. Microbe‒metabolite correlations were completed via MicrobiomeAnalyst 2.0 (McGill University, Montreal, QC, Canada) using standard approaches.^41^ Briefly, operational taxonomical unit (OTU) tables from the 16S rRNA sequencing analysis and targeted metabolomics data mentioned earlier were loaded onto the MicrobiomeAnalyst platform. Microbiome data were filtered to remove OTUs where the variance was less than 5% based on the interquartile range and whose abundance in the samples was less than 10% of the samples. The bacteriome data were scaled using total sum scaling to account for the compositional nature of the dataset since MicrobiomeAnalyst 2.0 utilizes the MaAslin2 package^42^ in the R programming language to handle this dataset. The quantified metabolomics data were not filtered or further normalized to reflect the true variance of the absolutely quantified dataset. Spearman's correlations were performed between each chemotherapeutic and combined control group to explore the relationships between the differentially abundant taxa (FDR q < 0.05) and all the metabolites present. Trending and significant relationships across all chemotherapeutic and taxonomic levels were further consolidated and visualized using tidyverse,^39^ readxl,^43^ ggplot2^44^ packages in R programming language.
Plasma immunoassays
Concentrations of LBP in plasma were measured as a reliable marker of gut barrier disruption^45^ using the Hycult Mouse LBP Assay (Hycult Biotech Inc., Wayne, PA, USA), as previously described.^7^ Samples were run in duplicate and diluted 200 in wash buffer, and the concentration of LBP was measured following the manufacturer's instructions. Inter-plate variability was less than 12%, and intra-plate variability was less than 6%.
Concentrations of plasma cytokines and chemokines (CCL2 [MCP-1], CXCL1 [KC/GRO], IL-1β, IL-6, IL-10, and TNFα) that were previously implicated in gut-to-brain signaling^19^^,^^46^ were measured using multiplex fluorescent immunoassays (Meso Scale Discovery, Rockville, MD, USA) and run according to the manufacturer's instructions, as previously described.^19^ Briefly, samples were run in duplicate and diluted 1 with assay diluent. Inter-plate variability was less than 11%, and intra-plate variability was less than 6%.
Open field test
Behavioral testing occurred during the dark phase (1600–000 h) under dim red light and ambient white noise. The mice were tested for fatigue (general locomotion) and anxiety-like behavior (central tendency) using an open field test, as previously described.^19^ Briefly, the mice were placed into one corner of a 16 × 16 in PAS system photobeam arena (San Diego Instruments, San Diego, CA, USA) and allowed to explore the arena for 15 min. Automated quantification of beam breaks was measured using PAS Data Reporter version 1.0.2.5 software (San Diego Instruments). Fatigue was operationally defined as decreased locomotion (total beam breaks) in the whole arena, as compared to controls. Anxiety-like behavior was defined as a reduction in central tendency (i.e., the percentage of locomotion in the centermost 4 × 4 zone of the arena as compared to all locomotion in the entire arena) relative to that of the controls. Each testing apparatus was cleaned with 70% ethanol between mice.
Network analyses
Network analyses were performed to identify relationships between outcomes that differed between the chemotherapy groups and controls across the gut, blood, and brain, using R programming language and Cytoscape 3.10.1 (Consortium, San Diego, CA, USA), as previously described.^47^ Briefly, Spearman's ranked correlation was were utilized to identify relationships between outcomes, adjusting for relationships involving non-parametric outcomes via the ggcorrplot package.^48^ Data were further wrangled using tidyverse,^39^ and readxl^43^ packages to prepare for visualization in Cytoscape. In Cytoscape, topology measures of each network (e.g., betweenness centrality, closeness centrality, and clustering coefficients) were calculated in a non-directed manner as there was no a priori hypothesis of the direction of the relationships between all nodes in the network. Within Cytoscape, networks were made within the same collection (i.e., a single file), and thus, the scales were the same for each network.
Statistical analyses
For all datasets, appropriate statistical tests were first completed to determine whether the two control groups (PBS and CremEL) differed. In most instances, these groups did not differ and were combined and presented as “Combined Controls”. In the rare cases where these control groups differed, the chemotherapies were compared to their respective controls (CremEL for paclitaxel, PBS for all other chemotherapies). Body mass and average food intake were analyzed using a two-way ANOVA with Fisher's LSD post-hoc test. Spleen mass, circulating LBP, circulating cytokines/chemokines, ileal and hippocampal gene expression, and locomotion, and central tendency were analyzed using a one-way ANOVA with Fisher's LSD post-hoc test. For one- and two-way ANOVAs where data among groups were non-parametric or homoscedastic, a Kruskal‒Wallis or a Brown‒Forsythe ANOVA were used respectively. Statistical outliers were removed using ROUT's test for outliers (detailed in Supplemental File 1). To control for false positives, FDR q-values were used to determine significance in non-exploratory analyses from large datasets with multiple hypothesis testing (i.e.,16S rRNA sequencing and targeted metabolomics).^49^ However, where each comparison was independent of each other and we prioritized testing whether groups were different from controls, one- or two-way ANOVAs with Fisher's LSD post-hoc tests were used.^50^ GraphPad Prism was used to calculate these statistics. Statistical significance was reported when p < 0.05. All other statistical analyses and approaches for the dataset not mentioned in this section, including 16S rRNA sequencing, targeted metabolomics, microbe‒metabolite correlation, and systemic network analyses, are described above in their respective sections.
Results
Statistical significance is denoted by post-hoc comparison compared to control where the overall ANOVA results were already different unless otherwise indicated. Statistical outcome details are included in Supplemental File 1.
Gross physiology
Body mass and food intake
To compare general health consequences among varying regimens of chemotherapeutics, body mass and food intake are reported relative to the percentage of the chemotherapy regimen completed. After chemotherapy completion, data are reported 1 d post-chemotherapy for all the mice and at 7-d intervals thereafter until 28 d post-chemotherapy for the longer-term mice (Figure 1A). In general, most chemotherapy groups did not robustly change body mass during treatment or by 28 d post-treatment, except cisplatin. Specifically, at 1 d post-chemotherapy (i.e., 100% chemotherapy was completed), all chemotherapies decreased the body mass relative to that of the controls (p < 0.05, respectively, Figure 1B). This acute mild weight loss was recovered by 28 d post-treatment in paclitaxel-treated mice relative to controls. In contrast, mild weight loss was retained at 28 d post-treatment in the cyclophosphamide and doxorubicin groups, whereas cisplatin markedly decreased body mass during and after treatment (p < 0.05 for all instances, Figure 1B). Food intake slightly decreased over the active treatment period but did not differ amongst treatment groups (Figure 1C).
Spleen mass
Because splenomegaly has been observed in chemotherapy models^7^^,^^51^ and is a gross measure of immune activation, relative spleen masses were compared amongst chemotherapy models. Overall, there were mixed effects of chemotherapy on the relative spleen mass immediately after treatment, which resolved over time in all regimens except paclitaxel. Specifically, at 1 d post-treatment, cisplatin and doxorubicin reduced the spleen mass (post-hoc p < 0.05 for both, Figure 1D), whereas cyclophosphamide did not alter the spleen mass. In contrast, paclitaxel robustly increased the spleen mass 1 d post-chemotherapy and this increase uniquely persisted 28 d post-chemotherapy (post-hoc p < 0.05, respectively, Figure 1D).
Gut physiology
To determine the extent to which various chemotherapeutics disrupt gut physiology, plasma collected 1 and 28 d after treatment was assayed for circulating LBP, and the gene expression of markers related to gut barrier integrity, water retention, and inflammation was quantified in ileal tissues. Overall, all the chemotherapeutics uniformly increased circulating LBP acutely after treatment (post-hoc p < 0.05, in all the cases, Figure 2A top), which was resolved by 28 d post-chemotherapy. In contrast, there were mixed effects of chemotherapy on ileal gene expression of gut barrier integrity, water retention, and inflammatory mediators by chemotherapy, but some differences observed at 1 d post-treatment remained different at 28 d post-treatment. Notably, compared to the other chemotherapeutics, doxorubicin most consistently increased the gene expression of barrier integrity and water retention markers while decreasing inflammatory mediator gene expression.
More specifically, circulating versus ileal markers of gut barrier integrity were not consistent at 1 d post-chemotherapy, as all chemotherapeutics increased circulating LBP, but there were no overall effects of any regimen on ileal Ocln and Tjp mRNA. At 28 d post-chemotherapy, however, doxorubicin increased Ocln gene expression (post-hoc p < 0.05, Figure 2A bottom), and no other significant effects were observed due to other treatments.
Cyclophosphamide, cisplatin, and doxorubicin increased ileal gene expression markers related to water retention, whereas paclitaxel did not affect these markers. At 1 d post-chemotherapy, cisplatin increased Aqp1 and Aqp11 mRNA, whereas cyclophosphamide and doxorubicin increased Aqp1 (post-hoc p < 0.05, respectively, Figure 2B). At 28 d post-chemotherapy, doxorubicin increased Aqp1 and Aqp11 gene expression (post-hoc p < 0.05, respectively, Figure 2B).
With respect to ileal inflammatory responses, cisplatin reduced proinflammatory Il1b and Il17a gene expression acutely after treatment, whereas doxorubicin only reduced Il1b gene expression. By 28 d post-treatment, this reduction in gene expression by these two chemotherapeutics was more widespread (Ccl2, Il1b, Il6, Il17a, and Tnf) (post-hoc p < 0.05 in all cases, Figure 2C). In contrast, paclitaxel increased Il1b mRNA at 28 d post-treatment (post-hoc p < 0.05, Figure 2C). Cyclophosphamide did not alter these ileal inflammatory transcripts.
Gut bacteriome
To compare the disruption of the gut bacteriome among various chemotherapeutics, fecal samples collected at baseline, 1, and 28 d post-chemotherapy from the longer-term cohort were used for 16S rRNA sequencing. In general, chemotherapeutics induced modest mixed effects on bacterial richness and phylogenetic relatedness (α-diversity), as well as mixed effects on bacterial community composition (β-diversity).
α-Diversity
Overall α-diversity was modestly altered by chemotherapy. Using Faith's index of phylogenetic diversity, cisplatin modestly increased α-diversity relative to that of the controls at 1 d post-chemotherapy (post-hoc p < 0.05, Figure 3A). In contrast, doxorubicin moderately decreased α-diversity relative to that of the controls at 28 d post-chemotherapy (post-hoc p < 0.05, Figure 3A). These changes resulted in significant effects of chemotherapy, time, and their interaction for this measure of α-diversity (ANOVA p < 0.05, respectively, Figure 3A). Shannon diversity and evenness, two alternative metrics of α-diversity assessed, showed similar modest perturbations in the absence of a main effect of chemotherapy (Supplemental Figure 1A and B).
β-Diversity
The β-diversity of the controls shifted modestly over time in two of the three measures, but was unchanged over time using the Weighted UniFrac Distance measure, which assesses bacteriome composition based on the abundance and phylogenetic relatedness of the bacterial taxa present (Supplemental Figure 1C–E). Thus, to assess the effects of chemotherapy on the gut bacteriome with minimal noise from the controls, we focused on this measure. Overall, all the groups had a similar gut bacteriome composition at baseline (Figure 3B), and most chemotherapy groups was moderately shifted the gut bacteriome composition at 1 d (Figure 3C) and 28 d post-treatment (Figure 3D) relative to that of the controls. However, only cisplatin induced a robust and persistent shift in bacteriome composition.
At 1 d post-chemotherapy, cyclophosphamide and cisplatin both shifted the gut bacteriome composition away from that of the controls (post-hoc q < 0.05, respectively, Figure 3C). Specifically, only 6 differentially abundant bacterial families contributed to this shift for cyclophosphamide, with the majority of taxa increasing in relative abundance (e.g., Tannerellaceae) and one decreasing (Figure 3E). In contrast, 23 differentially abundant bacterial families contributed to the bacteriome shift caused by cisplatin, which was balanced between relative increases and decreases in specific families (e.g., Ruminococcaceae, Enterococcaceae, Prevotellaceae, and Lactobacillaceae) (Figure 3F). Additionally, both cyclophosphamide and cisplatin increased Akkermansiaceae and Corynebacteriaceae, and decreased an unclassified family of the Coriobacteriales order.
At 28 d post-chemotherapy, cisplatin-mediated shifts in the gut bacteriome remained apparent at 28 d post-chemotherapy, and a separation of doxorubicin-treated mice relative to controls was also observed (post-hoc q < 0.05, respectively, Figure 3D). Indeed, four differentially abundant families contributed to this continued shift after cisplatin, including a persistent increase in Corynebacteriaceae and Erysipelotrichaceae and a persistent decrease in Enterococcaceae as seen at 1 d post-treatment, along with a new increase in Bifidobacteriaceae (Figure 3G). For doxorubicin, 16 differentially abundant families contributed to this late bacteriome shift, which included increases in Akkermansiaceae, and decreases in Ruminococcaceae and Prevotellaceae (Figure 3H), similar to the changes observed from other chemotherapeutics four weeks earlier.
Longitudinal β-diversity changes in response to chemotherapy treatment
To understand how chemotherapy changed the gut microbiome composition within mice, we assessed the Weighted UniFrac Distance over time within each treatment group (Supplemental Figure 2A–D). However, since some measures of β-diversity suggested a slight shift in the bacteriome composition of control mice, we prioritized the comparisons to control mice at each time point mentioned above.
Gut, blood, and brain metabolomes
To determine the extent to which chemotherapy-induced disruptions to the gut bacteriome result in altered metabolite production, targeted panels of short-chain fatty acids, bile acids, and tryptophan metabolites were first analyzed in the gut contents 1 d after chemotherapy. Among all the quantified metabolites in the tryptophan metabolism panel, 90% were altered by at least one chemotherapy regimen, compared to 75% and 40% of the metabolites in the short-chain fatty acid and bile acid panels, respectively (Supplemental File 1). Thus, tryptophan metabolism and short-chain fatty acid panels were selected for assessing the plasma metabolome at 1 d post-treatment. Because only one metabolite (proline) from the short-chain fatty acid panel was differentially abundant in the plasma across groups, only the tryptophan metabolism panel was selected to assess the brain (hippocampal) metabolome 1 d after chemotherapy, as well as the gut contents, plasma, and hippocampal tissues collected 28 d post-treatment.
Overall, the chemotherapy regimens uniformly disrupted the tryptophan metabolome of the gut contents the most, followed by the plasma, with little to no changes observed in the hippocampus. The quantified metabolome disruptions in these tissues were generally characterized by consistent shifts away from serotonin production and towards the kynurenine and indole arms of tryptophan metabolism in all tissues. Another general observation was that the largest disruption in the gut tryptophan metabolome corresponded to when the gut bacteriome differed the most from that of the controls, even though this disruption occurred at different times for the various chemotherapies. Of note, only cisplatin consistently increased indole metabolites in both the gut and blood at 1 and 28 d post-treatment. Detailed observations reported by time and tissue are presented below.
1 d post-chemotherapy
For the gut contents, all chemotherapy treatments shifted the overall metabolome composition from the 3 panels measured away from that of the controls at this acute timepoint (post-hoc q < 0.05, in all cases. Figure 4A left). Among the top three most significantly enriched metabolic pathways identified for each chemotherapy, none were shared amongst the regimens, however, the top pathways were all related to tryptophan or amino acid metabolism (e.g., the phenylalanine metabolism pathway by paclitaxel, the amino sugar metabolic pathway by cisplatin, the selenoacid metabolism pathway by cyclophosphamide, and, broadly, the tryptophan metabolism pathway by doxorubicin) 1 d after treatment (p < 0.05 for all, Figure 4A center). For those metabolites that were quantified (see the Supplementary File for a detailed list), all chemotherapeutics increased tryptophan but differentially altered indoles and kynurenine catabolites. Specifically, paclitaxel increased phenylalanine, tyrosine, tryptophan, kynurenine, kynurenic acid, and indole-3-lactic acid but decreased indole-3-carboxyaldehyde in the gut contents (post-hoc p < 0.05, Figure 4A right). Cisplatin increased tryptophan and kynurenine but decreased kynurenic acid, serotonin, and indole-3-propionate (post-hoc p < 0.05, respectively, Figure 4A right). Cyclophosphamide increased tryptophan and indole-3-lactic acid (post-hoc p < 0.05, respectively, Figure 4A right). Finally, doxorubicin increased tryptophan and decreased indole-3-propionate (post-hoc p < 0.05, respectively, Figure 4A right).
In the plasma, short-chain fatty acids and tryptophan metabolites were measured, but only 1 metabolite was differentially abundant between at least one chemotherapy group and controls (Supplemental File 1) compared to 5 metabolites in the tryptophan panel. Therefore, hereafter, all PCAs were calculated based on the normalized abundance of metabolites in the tryptophan metabolism panel only. Only cisplatin significantly shifted the plasma tryptophan metabolome away from that of the controls (q < 0.05, Figure 4B left). Indeed, the broad tryptophan metabolic pathway was enriched in this group (p < 0.05, Figure 4B center), driven, in part, by decreases in plasma indole-3-proprionate and indole-3-acrylate and increases in indole-3-lactate relative to controls (post-hoc p < 0.05, respectively, Figure 4B right).
In the hippocampal tissue, the PCAs indicated that cyclophosphamide, paclitaxel, and cisplatin each induced differential tryptophan metabolome composition relative to controls (q < 0.05, Figure 4C left). While these three treatment groups each increased enrichment for the phenylalanine and tyrosine metabolic pathways (p < 0.05, Figure 4C center), only cisplatin and paclitaxel specifically increased phenylalanine, tyrosine, and tryptophan and decreased serotonin in the hippocampus (post-hoc p < 0.05, Figure 4C right). On the other hand, the shift in the hippocampal tryptophan metabolome PCA due to cyclophosphamide (q < 0.05) was driven by changes in metabolites that were not quantified (see Supplemental File 1).
28 d post-chemotherapy
At this later timepoint after chemotherapy, the gut metabolome remained significantly shifted away from that of the controls for all the chemotherapy regimens (q < 0.05, respectively, Figure 4D left), similar to that observed more acutely. Indeed, these lasting shifts in the gut tryptophan metabolome enriched metabolic pathways involving amino acid and tryptophan breakdown/catabolism (e.g., the nicotinate/nicotinamide pathway by paclitaxel; the tyrosine metabolism pathway by cyclophosphamide, cisplatin, and doxorubicin) (p < 0.05 for all, Figure 4D center).
Of the quantified, all four chemotherapies increased niacin, whereas three out of four (cyclophosphamide, cisplatin, and doxorubicin) also increased tyrosine, phenylalanine, and tryptophan and decreased melatonin (post-hoc p < 0.05, respectively, Figure 4D right). Additionally, cisplatin and doxorubicin further altered the indole and kynurenine arms of tryptophan metabolism. Specifically, cisplatin decreased kynurenic acid, and increased tryptamine and doxorubicin increased tyramine, indole-3-lactic acid, and anthranilic acid but decreased kynurenic acid (post-hoc p < 0.05 in all cases, Figure 4D right).
In the plasma, only cisplatin shifted the tryptophan metabolome composition away from controls (q < 0.05, Figure 4E left), due to robust increases in indoles and kynurenine catabolites (e.g., indole-3-acetate, indole-3-acrylate, indole-3-propionate, indole-3-lactate, anthranilic acid and 5hiaa; Figure 4E right). The top 3 enriched pathways due to cisplatin were related to tyrosine metabolism (p < 0.05, Figure 4E center) due to an increase of a non-quantified metabolite (see Supplemental File 1). Whereas paclitaxel and doxorubicin did not shift the overall plasma metabolome away from that of the controls, both chemotherapies decreased serotonin and kynurenine levels (post-hoc p < 0.05, Figure 4E). Finally, no groups differed from the controls in the PCA of the hippocampal tryptophan metabolome (Figure 4F); however, cisplatin increased one metabolite, kynurenine, in the brain (Supplemental File 1).
Microbe–metabolite correlation
In the gut, bacteria are related to metabolites through many relationships (e.g., bacterial breakdown of dietary products, cross-feeding, etc.). Exploring these relationships can further contextualize how chemotherapy changes the microbiome. Therefore, the bacteriome and gut metabolome data from 28 d post-treatment were used to assess the relationships between differentially abundant bacterial taxa and metabolites related to tryptophan catabolism using an unbiased approach. Paclitaxel, cisplatin, and doxorubicin induced modest, mixed correlations (positive and negative) between bacterial taxa and metabolites, with the majority observed for paclitaxel and cisplatin. One notable consistency for paclitaxel, cisplatin, and doxorubicin was the trending (p < 0.1) or significant positive (p < 0.05) correlation between the gut Monoglobaceae and indole-3-proprionate and the negative correlation between the gut Erysipelatoclostridiaceae and melatonin. No relationships were identified due to cyclophosphamide treatment (Supplemental Figure 3).
Circulating inflammatory markers
Because circulating inflammatory signals are a leading hypothesized route by which the gut microbiome can communicate with the brain, inflammatory mediators in the blood of all groups were quantified next. Overall, most chemotherapeutics consistently increased circulating proinflammatory mediators (IL-1β, IL-6, TNFα, CCL2, and CXCL1) shortly after chemotherapy, but only paclitaxel and cisplatin remained partly elevated 28 d after chemotherapy.
Specifically, at 1 d post-chemotherapy, paclitaxel and doxorubicin both increased proinflammatory IL-1β, IL-6, TNFα, and CCL2 in the plasma relative to those in the combined controls or the respective controls when the controls were different (post-hoc p < 0.05, respectively, Figures 5A–D). Conversely, cyclophosphamide did not increase these proinflammatory markers but rather decreased the anti-inflammatory cytokine IL-10 (post-hoc p < 0.05. Figure 5F). Cisplatin increased circulating IL-6, TNFα, CCL2, and CXCL1 (post-hoc p < 0.05, Figure 5E), as well as IL-10 (post-hoc p < 0.05, in all cases). Later, at 28 d after treatment, two proinflammatory mediators remained elevated after paclitaxel (IL-6 and CCL2) and cisplatin (IL-6 and TNFα) (post-hoc p < 0.05, in all cases). However, TNFα and IL-10 were increased and decreased by cyclophosphamide and doxorubicin treatment, respectively (post-hoc p < 0.05, respectively), while most circulating inflammatory mediators returned to levels comparable to controls.
Central inflammatory markers
Circulating inflammatory signaling can be propagated across the blood‒brain barrier and induce parallel signaling in the brain parenchyma. Thus, inflammatory gene expression and glial reactivity were assessed in the hippocampus. Overall, there were both mixed and modest changes to inflammatory gene expression caused by paclitaxel and cisplatin only, primarily acutely (1 d) after chemotherapy (Table 2 and Supplemental File 1). Of note, some transcripts were different between the two controls; thus, the groups were compared to their respective controls. Specifically, paclitaxel acutely increased Tnf and Il1b, but decreased Ccl2 and Myd88. Similarly, cisplatin increased Nlrp3 but decreased Aif1 and Ccl2 1 d after chemotherapy. At 28 d post-chemotherapy, only paclitaxel increased proinflammatory mediators (Aif1 and Nlrp3) gene expression (post-hoc p < 0.05 in all cases, Table 2). None of the chemotherapies altered anti-inflammatory Tgfb1 gene expression at either timepoint.
Behavior in an open field
Fatigue (total locomotion) and anxiety-like behavior (central tendency) were assessed using a 15 min open field test. Overall, every chemotherapy regimen reduced both total locomotion (post-hoc p < 0.05, Figure 6A), as well as central tendency (post-hoc p < 0.05, Figure 6B) shortly after chemotherapy, indicative of fatigue and anxiety-like behavior, respectively. By 28 d post-chemotherapy, this evidence of fatigue and anxiety-like behavior resolved.
Network analyses among chemotherapy-altered outcomes in the gut, blood, and brain
Network analyses are useful tools to identify how strongly analytes in a given system might be correlated with each other to generate hypotheses about how they may be mechanistically contributing to a given phenotype. For example, when a subset of nodes (outcomes) are highly interconnected, it may be an indication that they are working together along a shared pathway; or if a node lies in the shortest path that connects two other nodes, it may be an indication of a mediating relationship. These hypotheses can then be tested in subsequent experiments.
In the present study, since paclitaxel and cisplatin induced the most severe and long-lasting disruptions across the various gut, blood, and brain outcomes, with higher severity at 1 d after treatment, we performed network analyses on these nodes. Notably, more outcomes were different between cisplatin and controls than that of paclitaxel. Additionally, after paclitaxel treatment, more nodes in the gut and brain were directly correlated across tissues than through a node in plasma, as well as within respective tissues. Whereas, after cisplatin treatment, nodes in the gut and plasma were directly correlated within their respective tissues, as well as some nodes in the gut and brain were connected through nodes in the plasma (Figure 7A and B).
Some specific relationships observed were the lack of a significant correlation between brain tryptophan and serotonin by either chemotherapeutic. Additionally, one day after paclitaxel treatment, the hippocampal serotonin concentration was positively correlated with central tendency (anxiety-like behavior), whereas the hippocampal phenylalanine concentration was strongly negatively correlated with total locomotion (p < 0.05 in all cases, Figure 7A). Different relationships emerged acutely after cisplatin treatment. Specifically, the plasma 5-hiaa concentration was positively correlated with hippocampal Nlrp3 expression, the plasma LBP concentration was negatively correlated with central tendency, and total locomotion was negatively related to circulating IL-10 due to cisplatin (p < 0.05 in all cases, Figure 7B). Notably, hippocampal Tnf expression due to paclitaxel treatment and the taurocholic acid concentration in the gut contents due to cisplatin stood out as the factors that connected to the most nodes in their respective network (i.e., highest betweenness centrality, Figure 7C). Additionally, paclitaxel induced higher absolute correlation coefficients between its connected nodes (i.e. closeness centrality, Figure 7D) than cisplatin, suggesting stronger relationships amongst outcomes. Furthermore, paclitaxel induced higher clustering coefficients, indicating that more nodes correlated with each other than was seen with cisplatin (Figure 7E).
Discussion
The present study constructed a comprehensive landscape of the gut microbiome‒blood‒brain axis after treatment with four commonly used chemotherapeutics. While all the chemotherapies modulated the gut microbiome and induced gut permeability, there were significant differences in the timing and severity of the disruption, as well as their effects on other tissues along the axis. Pinpointing the chemotherapeutics that most disrupt this axis could reveal which treatment regimens stand to benefit most from innovative, microbiota-targeted strategies aimed at reducing gastrointestinal, and potentially behavioral, side effects of chemotherapy.
All four chemotherapeutics caused weight loss during treatment, as previously reported in mice.^7^^,^^52^^,^^53^ Cisplatin-induced weight loss was most severe, stunting growth despite normal food intake, aligning with clinical evidence of lasting metabolic dysfunction^54^ that may affect the gut microbiome–blood–brain signaling. Cisplatin and doxorubicin also reduced the spleen mass early, likely via lymphoid cell apoptosis,^55^^,^^56^ but recovered by day 28. In contrast, paclitaxel expectedly induced splenomegaly,^7^ which persisted for at least 28 d. These divergent splenic responses highlight how chemotherapeutics vary in their immunomodulatory effects and potential activation of the gut microbiome–blood–brain axis.^56^
Most prior studies investigating the impact of chemotherapy on gut physiology have focused on the colon.^57^^,^^58^ Here, we examined the ileum, the most immune cell-dense region and a training site for circulating immune cells involved in gut microbiome–blood–brain signaling (reviewed in^31^). The gene expression patterns observed for barrier integrity and gut function markers were somewhat unexpected for most chemotherapeutics except for paclitaxel; gene expression patterns for inflammatory markers were, however, more typical. Only paclitaxel increased a proinflammatory mediator (Il1b) at 28 d post-treatment, which aligns with prior findings of minimal colon effects.^7^ In contrast, cisplatin and doxorubicin robustly decreased proinflammatory gene expression, which aligns with previous studies.^58^^,^^59^ Notably, decreased proinflammatory expression by these two chemotherapeutics was coincident with increased gene expression of aquaporins and tight junction proteins within the ileum one whole month after treatment. Given that chemotherapy kills dividing cells, it is possible that gut immune cell populations are slower to recover compared to epithelial cells (reviewed in^3^).
In contrast, all drugs transiently increased circulating LBP, corroborating acute gut barrier disruption in mice and patients.^7^^,^^45^^,^^57^ Indeed, it is possible that this disconnect between the gut transcript and circulating markers of gut barrier injury may be due to the specific gut region (ileum) assessed herein, as different parts of the intestines can have distinct tight junction gene expression profiles.^60^ Alternatively, mislocalization of tight junction proteins within gut epithelial cells may explain the permeability of the gut barrier (i.e., increased LBP) concurrent with unchanged tight junction gene expression.^61^ Future studies that utilize functional assays, such as intragastric delivery of FITC dextran to test permeability of the gut barrier,^62^ are warranted to address these remaining questions. Overall, all chemotherapeutics transiently compromised gut permeability, but cisplatin and doxorubicin reduced markers of ileal inflammation over time, whereas only paclitaxel slightly increased gut inflammation.
These modest changes to gut physiology may inform changes to the bacteria present in the gut since some gut cells provide essential nutrients for some bacteria (reviewed in^3^). Furthermore, a disrupted gut microbiome may have major consequences on the brain since our prior work demonstrated that chemotherapy-induced microbiome alterations causally contribute to disruption of the gut microbiome–blood–brain axis.^19^ Indeed, all chemotherapeutics tested have been reported to shift β-diversity (i.e., a between-group measure of the composition of the bacteria present in the gut) shortly after treatment in mice.^7^^,^^57^^,^^63^ However, to our knowledge, the present study is the first to assess the persistence of these gut bacteriome changes over time. Some remarkable findings include that β-diversity remained significantly disrupted 28 d post-treatment after cisplatin, and the disruption due to doxorubicin was delayed, emerging only at this later time point. A prior report described a transient shift in β-diversity due to doxorubicin, but this study used a lower dosing paradigm and a different strain of mice.^63^ Dynamic shifts in β-diversity were not observed for paclitaxel, in contrast to our previous report.^7^ It is possible that the β-diversity shifts observed over time in the control mice (Supplemental Figure 1F and G) may have masked the paclitaxel-induced bacteriome disruption. Indeed, some factors (e.g., mouse shipment batch, housing density) can lead to small shifts in bacteriome composition of controls over time.^64^^,^^65^ Thus, only the largest effects of paclitaxel chemotherapy on β-diversity were detected, possibly underestimating small microbial shifts. Future studies may require larger sample sizes, as the effect size of paclitaxel-induced gut bacteriome disruption may be relatively small.
In this study, we observed modest changes in α-diversity (a within-group measure of gut bacterial composition) and moderate shifts in differential bacterial taxa, which likely contributed to the β-diversity patterns noted above. One day after cisplatin treatment, greater log-fold enrichment (relative to other chemotherapy groups) of bacterial families may have driven β-diversity changes through an increase in the abundance of phylogenetically unrelated families (i.e., higher α-diversity). In contrast, 28 d after doxorubicin, greater log-fold depletion of bacterial families may have contributed to β-diversity shifts through the reduced abundance of unrelated families. Cyclophosphamide expectedly disrupted the gut bacteriome composition acutely,^57^ but this resolved by 28 d post-treatment. Although cisplatin, doxorubicin, and cyclophosphamide affected gut bacterial taxa enrichment differently, they shared notable patterns linked to adverse health or behavioral side effects of chemotherapy in preclinical and clinical studies, specifically, increases in Akkermansiaceae and decreases in Prevotellaceae,^66^ Coriobacteriales,^67^ and Ruminococcaceae.^19^
Metabolites produced by gut microbes have been shown to impact the gut‒blood‒brain axis, including those involved in tryptophan metabolism, short-chain fatty acids, and bile acids (reviewed in^68^), while rarely assessed, existing preclinical studies of chemotherapy report only the normalized abundance of one of these groups of metabolites in the plasma, gut or fecal contents and focus on the changes that occur shortly after treatment.^9^^,^^69^ Here, the concentrations of these three groups of metabolites were first absolutely quantified in the cecal/proximal colon contents one day after chemotherapy, and tryptophan metabolites were also assessed a month later. The top enriched metabolome pathways across all four chemotherapeutics consistently involved tryptophan or other amino acid (e.g., phenylalanine) metabolism. Of note, phenylalanine metabolism is related to tryptophan metabolism through the shared tetrahydrobiopterin enzyme.^70^ Since tryptophan is primarily derived from dietary sources, shifts in its concentration or its catabolites implicate changes in either bacterial metabolism of tryptophan (i.e., the production of indoles) or host metabolism (i.e., the production of kynurenine or serotonin and their subsequent catabolites) (reviewed in^71^). Indeed, all chemotherapeutics increased tryptophan in the gut contents at one or both timepoints post-treatment, suggesting impaired tryptophan catabolism by the host and/or microbes since all the mice were fed the same diet. Furthermore, the effects of paclitaxel on the gut metabolome were transient but quite robust and coincided with the absence of a significant gut bacteriome β-diversity shift. This suggests that paclitaxel likely shifts microbial function more than, or independent of, major shifts in the bacterial taxa present in the gut, and furthermore implicates that is processed differently after chemotherapy, but metatranscriptomic studies are needed to test this hypothesis. Additionally, cisplatin enriched Akkermansiaceae and Bifidobacteriaceae, known producers of bioactive indoles,^72^^,^^73^ which may have contributed to consistent increases in the gut and blood indoles at both time points. Across all chemotherapeutics, severe disruption of both the bacteriome and metabolome occurred at the same time point (e.g., one day post-treatment for cyclophosphamide and cisplatin and 28 d with doxorubicin). This suggests that future interventions using pro- or post-biotics may be most effective when timed to these periods of greatest disruption.
A few preclinical and clinical studies have suggested that circulating inflammatory mediators and/or metabolites may be key gut-to-brain communication signals after chemotherapy.^6^^,^^11^^,^^19^ Here, paclitaxel, cisplatin, and doxorubicin increased circulating proinflammatory mediators (e.g., IL-1β, IL-6, TNFα, CCL2, and CXCL1) shortly after treatment, which is consistent with these reports.^34^^,^^69^^,^^74^ In contrast, cyclophosphamide did not evoke an acute or delayed proinflammatory response, as previously reported.^28^ To the best of our knowledge, this is the first report showing that paclitaxel and cisplatin sustain elevated circulating proinflammatory mediators for at least 28 d post-treatment. Of note, cisplatin elicited a compensatory increase in the anti-inflammatory mediator, IL-10, acutely after treatment, whereas evidence for this inflammatory “braking” mechanism was not detected for paclitaxel. This, coupled with the differential effects on the spleen mass, further supports the understanding that these two chemotherapeutics may alter Th2 immune responses differently.^75^^,^^76^ In terms of metabolite signaling present in the blood, only cisplatin perturbed the plasma tryptophan metabolome through a prolonged increase in gut-derived indoles. Notably, other chemotherapeutics also increased indoles, but only in the gut. Further studies are needed to determine how various chemotherapies alter indole diffusion from the gut into the bloodstream (reviewed in^77^). Additionally, paclitaxel and doxorubicin decreased kynurenine in plasma at 28 d post-treatment, which is consistent with a study of breast cancer patients who experienced cognitive disruptions, 70% of whom received combination chemotherapy including a taxane.^11^ This link between impaired tryptophan catabolism and negative brain side effects after taxane treatment was also observed in the present study and is described below. In the present study, plasma short-chain fatty acids were difficult to quantify, as previously described.^78^
Behavioral side effects of chemotherapy, including fatigue and mood disruption, are common in patients with breast cancer after chemotherapy.^79^ Based on rodent models of chemotherapy, neuroinflammation is a leading hypothesized mechanism underlying these side effects,^80^ and neuroinflammation has been shown to be caused by gut microbiome disruption after chemotherapy.^19^ Paclitaxel produced mixed effects on hippocampal neuroinflammation markers one day post-treatment, contrasting prior reports of early increases within hours.^7^^,^^25^ Whereas cisplatin had modest effects on transcriptional markers of neuroinflammation, which is consistent with previous reports.^26^^,^^52^ The observed reductions in proinflammatory gene expression may reflect compensatory anti-inflammatory mechanisms (independent of Tgfb1) following an initial proinflammatory response. Unlike other chemotherapies, paclitaxel alone increased markers of microglial and inflammasome activation (Aif1 and Nlrp3) one month post-treatment. Prior work also showed elevated IBA+ staining in the hippocampal dentate gyrus five days after paclitaxel,^81^ suggesting a potential microglial priming effect that warrants further investigation. Additionally, all the chemotherapy paradigms assessed induced both fatigued and anxiety-like behaviors shortly after treatment that resolved by one month, which is consistent with previous reports.^53^^,^^82^^,^^83^ Given the link between microglial reactivity and behavioral disruption after paclitaxel,^81^ we predict that the neuroinflammatory effects on behavior one month post-treatment may be better detected using more sensitive assays (e.g., in-cage movement via emitters for fatigue and the elevated plus maze for anxiety-like behavior).
Understanding how the assessed chemotherapies differentially impact the gut microbiome‒blood‒brain axis may reveal potential intervention targets. To explore this, we performed network analyses focusing on the two regimens that caused the most severe paclitaxel and cisplatin. For paclitaxel, altered outcomes in the gut correlated with altered outcomes in the brain, independent of those in the blood. For example, increased gut content of indole-3-carboxyaldehyde is directly correlated with hippocampal Tnf directly. Of note, increased tryptophan in the gut contents and in the hippocampus one day after paclitaxel treatment were not related to each other. While we have previously identified that changes in the gut microbiome after chemotherapy cause an increase in the level of a circulating proinflammatory cytokine,^19^ the present study suggested that a neural route of gut-to-brain communication may also be at play (reviewed in^18^). In addition, while changes in brain tryptophan and serotonin were not correlated acutely after paclitaxel, network analysis suggested that decreased serotonin may contribute to anxiety-like behavior, consistent with preclinical and clinical studies of stress-related disorders (reviewed in^35^). In addition, the observed negative relationship between increased brain phenylalanine and fatigue after paclitaxel may be due to impaired dopamine biosynthesis from phenylalanine (reviewed in^71^) Furthermore, hippocampal Tnf appears to be most related to perturbations in the gut, blood, and brain and therefore may serve as a robust interventional target within this axis.^19^^,^^34^ Finally, paclitaxel uniquely prolonged proinflammatory signaling along the entire gut microbiome‒blood‒brain axis after 1 month, which may help explain the physiology underlying the reported delayed and unresolving side effects that breast cancer patients experience.^5^
Network analyses based on changes observed one day after cisplatin indicated that there were more correlations between outcomes in the gut (e.g., tryptophan catabolites) and indoles in the blood than to outcomes in the brain. Indeed, circulating tryptophan catabolites, LBP, and the anti-inflammatory IL-10 (rather than proinflammatory cytokines) were each correlated with modest central inflammatory changes^52^ and behavioral side effects. Together, this suggests that circulating gut-derived metabolites and LBP, rather than inflammatory mediators, may be stronger contributors to the behavioral consequences of cisplatin treatment. Furthermore, prolonged disruption of the gut bacteriome and metabolome, independent of brain-mediated side effects demonstrate that brain mediated side effects may not be proportional to changes in the microbiome only.
Cyclophosphamide and doxorubicin modestly disrupted the gut, blood, and brain outcomes and therefore network analyses were not performed. Nevertheless, doxorubicin induced a notably delayed disruption to the gut microbiome and increase in gut barrier gene expression (e.g., Ocln). This slower effect on the gut may explain why fewer changes were observed more broadly in the blood and the brain. Finally, the transient and delayed gut bacteriome disruption seen due to cyclophosphamide and doxorubicin treatment, respectively, may explain why combination treatments of both chemotherapeutics have been shown to cause additive disruption.^53^
Largely, the patterns observed in this study indicate that most chemotherapies more dynamically disrupted the gut microbiome‒blood‒brain axis acutely, with coincident functional disruptions in behavior (Table 3). This finding suggests that gut-driven neuroimmune signals may contribute to these acute behavioral deficits, whereas mechanisms downstream of this acute neuroinflammatory response could contribute to the long-lasting behavioral deficits that are observed in chemotherapy patients.^5^ For example, acute neuroinflammation may shape lasting brain cell energetics, microglial priming, etc.^84^^,^^85^ This hypothesis is partly supported by the increased Aif1 and Nlrp3 hippocampal gene expression observed only after 28 d post-paclitaxel treatment. Indeed, it is possible that other behavioral side effects may have been present at this later timepoint, but assessment was limited to one test. Furthermore, the magnitude of gut microbiome–blood–brain signaling appeared to be buffered along the axis, with a higher threshold in the brain (i.e., greater perturbations in the gut microbiome, less so in the blood, and even less so in the brain). This is likely adaptive such that small daily shifts in the gut microbiome do not proportionately shift behavior. Therefore, it is plausible that in less healthy models where the microbiome is already disrupted (e.g., pre-existing gut disease or a Western diet), the same chemotherapy regimens may drive more robust signaling and/or longer-lasting behavioral side effects. This may also explain the variance in behavioral side effects symptom severity even after similar treatment regimens in clinical populations.^6^ Of note, the simple and transient anxiety-like behavioral test was used here to measure affective-like sickness behaviors; further behavioral testing would be needed to assess more chronic affective-like behaviors. Overall, the present findings of dynamic gut microbiome disruption independent of behavioral side effects is a strong indication of complex gut-to-brain communication that requires further investigation. This study represents a comprehensive approach to systematically assess the contributions of commonly used chemotherapeutics to the gut microbiome‒blood‒brain axis in mice.
One limitation of this research is that the greatest disruption to the gut microbiome‒blood‒brain axis occurred in the chemotherapy paradigms that incorporated more doses (cisplatin 10, paclitaxel 6, doxorubicin 5, and cyclophosphamide 5), although paclitaxel caused marked effects with a similar number of doses to those with fewer effects. In addition, tumor-free female mice have been used, which may influence drug dynamics and the generalizability of these findings to males given the importance of sex in microbiome development.^86^ However, we anticipate that the effects of chemotherapy may generally translate across cancer models and sexes since chemotherapy patients of both genders, as well as those treated neoadjuvantly (before tumor resection) and adjuvantly, experience similar behavioral side effects (e.g. fatigue).^87^ Finally, behavioral assessments were minimal, and network analyses linking the gut, blood, and brain outcomes were correlational and future studies are needed to directly test the causal mechanistic relationships amongst them.
Conclusions
Gut-targeted interventions may be a viable, non-invasive, inexpensive approach to attenuate systemic and brain side effects in the context of chemotherapy.^28^^,^^34^^,^^88^ The present study demonstrated that four chemotherapeutics commonly used to treat breast cancer differentially perturbed the gut microbiome‒brain axis, with paclitaxel and cisplatin causing the most severe and longest-lasting disruption. This work suggests that some chemotherapies that may be most suitable for gut microbiome-targeted interventions in a time-dependent manner. Specifically, the clinical use of taxane-only or -heavy therapies may benefit the most from early gut-targeted interventions (e.g. fecal microbiota transplants, prebiotics, etc.) to alleviate gastrointestinal and possibly brain-mediated side effects, whereas cisplatin or combination therapies of doxorubicin and cyclophosphamide may have a longer window for potential gut-targeted interventions. These inferences warrant future testing in patient cohorts. Finally, this research indicates that studies that investigate the role of tryptophan metabolism in chemotherapy-induced brain and behavioral side effects are warranted.