Effect of chemotherapy and weight change on the gut microbiome of breast cancer patients during the rst year of treatment

Objective: The effects of chemotherapy and weight changes on the gut microbiome of breast cancer patients are not well understood. Methods: We conducted a 1-year follow-up study of 33 breast cancer patients and investigated gut microbiome before initiation of chemotherapy and after completion of treatment. We compared alpha diversity and mean taxa abundance at baseline and absolute changes (Δ; nal-baseline) in taxa abundance by treatment (16 neoadjuvant- neoADJ, 13 adjuvant- ADJ, 4 no chemotherapy-noC) using Wilcoxon rank sum and negative binomal tests and evalauted whether these changes were affected by weight changes during follow-up. Results: Alpha diversity measures increased in the neoADJ (+16.4% in OTU p =0.03; +51.6% in Chao1 p =0.03; +7.0% in Shannon index P=0.02; +11.0% in PD whole tree p =0.09) but not in the non-neoADJ group (ADJ+noC). The difference in change in Chao1 index between groups was statistically signicant ( p NEOADJ vs ADJ+noC =0.04). Wilcoxon p values of 0.03 to 0.003 were observed for ve taxa: Bacteroidetes (g _Alistipes) , Firmicutes (g_Clostridium, g_Eubacterium, g_Bilophila) and Preteobacteria g_Haemophilus). In the negative binomial analysis, changes in abundance differed at Bonferroni-adjusted p values ≤ 0.0007 for four taxa: two Bacteroidetes taxa ( g_Alistipes, f _S247 ) and two Firmicutes taxa (g_Catenibacterium, g_Eubacterium ). The negative binomial results remained largely the same when we adjusted for weight changes. Relevance: This pilot longitudinal study showed changes in alpha diversity measures and abundance of select taxa that appeared to differ by chemotherapy type. Further investigations are needed to conrm these ndings and to assess the impact of these microbiome changes on patient outcome.


Fecal Specimen Processing And Microbiome Analyses
Microbiome analyses were conducted in the laboratory of Dr. Jacques Ravel [19]. This included DNA extraction, 16S rRNA gene ampli cation of the two barcoded universal primers 319F and 806R for PCR ampli cation of the V3 and V4 hypervariable regions and sequenced the amplicons on the Illumina MiSeq platform. The 16S rRNA genes were ampli ed in 96-well microtiter plates. Negative controls without a template were processed for each primer pair. They performed taxonomic assignments and generated taxa abundance and read counts tables for each of the 144 fecal samples collected from 38 breast cancer patients. Fourteen samples failed (i.e., <100 read counts); 4 were from one patient who was excluded from all subsequent analyses. Of the other failed samples from 37 patients, two were baseline and two were last samples from four different patients (the two subjects with low baseline read counts did not contribute to analysis on alpha diversity but contributed baseline analysis on taxa abundance [19]). This current analysis on baseline and last fecal samples was based on 33 breast cancer patients ( Table 1). The study protocol was approved by the USC Institutional Review Board.

Statistical Analyses
Microbiome alpha diversity was estimated after rarefaction. We used Wilcoxon rank sum test to examine changes in alpha diversity during the 9 months of follow-up by chemotherapy treatment (16 neoADJ vs 17 non-neoADJ (13 ADJ + 4 noC); 16 neoADJ vs 13 ADJ) and by BMI change (16 lost weight vs 17 gained weight).
We conducted permutational multivariate analysis of variance (PERMANOVA) to test statistical signi cance of overall composition between baseline vs nal samples (n=33) by treatment and by BMI change [20]. The relationship of overall gut microbiome composition by treatment and BMI change was assessed by principal coordinate analysis (PCoA) based on the unweighted and weighted phylogenetic UniFrac distance matrix [21]. PCoA plots were generated using the rst two principal components by treatment and BMI change.
Turning to taxonomy, we conducted analyses to examine relationships of speci c taxa to treatment and BMI change during the follow-up. We calculated change in the relative abundance of 74 speci c taxa that had levels above zero, allowing comparison between baseline and last fecal microbiome by subtracting relative abundance of baseline values from nal (end of study) values to represent the absolute percentage change after completion of treatment. Statistical tests by Wilcoxon rank sum were used to compare results for neoADJ vs non-neoADJ (ADJ + noC) and for neoADJ vs ADJ groups. To accommodate the sparse, non-normally distributed count data, we repeated analyses using negative binomial mixed models for longitudinal microbiome data [22], without adjustment and with adjustment for BMI change during the follow-up. To correct for multiple comparisons [23], Bonferroniadjusted signi cance levels were set for 74 genera (0.05/74, p≤0.0007). Because of the modest sample sizes of this pilot study, we also considered the differences in change in genera to be suggestivie if 0.0007<p≤0.007. All data were analyzed using R (R Foundation for Statistical Computing, Vienna, Austria).
Assessment of cancer status and vital status through May 2021 was used as the outcome in proportional hazards regression models, each using one alpha diversity measure at either the baseline or the nal fecal sample collection as the dependent variable. These data were analyzed using SAS 9.4 (SAS, Cary, NC).

Results
The 33 breast cancer patients with baseline and nal (end of study) fecal samples were ages 51.7 ± 12.4 years, obese (mean baseline BMI of 31.4 ± 8.0 kg/m 2 ), mostly Hispanics (73%), and had hormone receptor (HR) positive (ER+PR+) (63.6%) and HER2 negative breast cancer (72.7%) ( Table 1). About half (n=16, 48.5%) received neoADJ, 13 (39.4%) received ADJ, and 4 (12.1%) had no chemotherapy (noC). The three treatment groups were similar in distribution by HR status but all four patients requiring no chemotherapy had early (I/II) stage cancer compared with 61.5% (8 of 13) in the ADJ and 37.5% (6 of 16) in the neoADJ group (p 2df =0.07). Although baseline BMI did not differ by treatment and there were no overall signi cant changes in weight during follow-up, 17 women gained an average of 3.7± 0.5 kg while 16 women lost an average of 4.7± 1.2 kg. Weight change was not uniform by treatment; the mean weight gain was 1.0 ± 3.6 kg in the ADJ and 0.33 ± 3.4 kg in the noC groups while there was a mean weight loss of 1.5 ± 7.0 kg in the neoADJ group.
The neoADJ and non-neoADJ group did not differ in beta diversity at baseline using the unweighted UniFrac distance ( Figure 3). However, the groups appeared to differ in their nal microbiome in unweighted UniFrac distance analysis; separation differed for axis 1 (p=0.11) and axis 2 (p=0.05), each explaining 18.4% and 9.2% of the variance, respectively. The weight loss and weight gain groups did not differ in their baseline microbiome but they appeared to differ in their nal microbiome using unweighted UniFrac distance analysis; separation between the groups differed for axis 1(p=0.28) and axis 2 (p=0.07), each explaining 20.0% and 9.4% of the variance respectively ( Figure 4). None of the differences were statistically signi cant using weighted UniFrac distance (data not shown).
As of May 2021, 8 (6 neoADJ, 2 ADJ) of the 33 breast cancer patients developed a recurrence or metastases of whom ve were deceased or in hospice care at this date. There were no signi cant associations between risk of recurrence/metastases and baseline alpha diversity measures or changes in alpha diversity (p 's were ≥ 0.18).

Discussion
To our knowledge, this is the rst study to follow a group of breast cancer patients with preplanned analyses to examine changes in gut microbiome during their rst year of treatment in relation to type of chemotherapy, taking into account changes in weight during the study period. In all three groups, the nal fecal sample was collected more than 100 days after the end of chemotherapy (for neoADJ and ADJ groups) or radiation (for noC group), representing a recovery period. We found striking differences in gut alpha diversity changes by treatment, increases in all four alpha diversity measures in the neoADJ group that was not observed in the ADJ or noC groups. Increases in alpha diversity were also observed in the weight loss but not in the weight gain groups. However, there were no signi cant changes in alpha diversity in relation to risk of recurrence/metastases. There were notable changes in taxa abundance that differed by treatment; one taxa, p Bacteroidetes (g_Alistipes) showed changes in abundance that reached the Bonferroni threshold of p<0.0007, while changes in abundance of 12 taxa showed p values that ranged from <0.001 to 0.0007 without or with adjustment of weight changes. In contrast, there were far few changes in taxa differences between the weight loss and weight gain groups, but one taxa, p_Firmicutes (g_Lachnobacterium) showed changes that were consistent in all three statistical analyses.
Although average weight changes were modest during this rst year of treatment, our nding of a signi cant increase in the alpha diversity measures in the neoADJ group may be related in part, to weight loss in the neoADJ group (-1.48 kg) but weight gain in the ADJ (+1.01 kg) and noC (+0.33 kg) groups. Alpha diversity measures have been used to assess health habits including body composition, and low gut alpha diversity has been associated with obesity in some studies [24,25]. In a study of 26 cancer patients (7 with breast cancer) who were treated with cytotoxic, targeted chemotherapy, or a combination of chemotherapy with immununotherapy, gut microbiome Shannon index was higher in responders than in nonresponders, who also displayed higher abundance of Alistipes, a genus member of the Rikenellaceae family within the Bacteroidales order [13]. In a study of non-small cell lung cancer patients treated with immune checkpoint inhibitors, responders showed higher diversity of gut microbiome as well as an enrichment of Alistipes [26]. In our analysis, Alistipes emerged to be important in both Wilcoxon rank sum and negative binomal analyses but its abundance decreased in the neoADJ group but not in the ADJ group after completion of treatment. The signi cance of our nding on Alistipes is unclear but this genus has been found to be correlated with both healthy phenotypes as well as having pathogenic roles [27] in colorectal cancer [28] and liver diseases [29]. It has been suggested that decrease in Alistipes contributes to the decrease in short chain fatty acids which have anti-in ammatory properties.
We also found suggestive differences in changes in taxa of select Erysipelotrichaceae genera (Catenibacterium, Eubacterium and Clostridium), abundance increased in the neoADJ but decreased in the non-neoADJ groups. In a small study of patients with breast or gynecological cancers, Erysipelotrichaceae abundance also increased but this was mainly among women who gained weight following treatment [12]. The immunogenic properties of some members of the Erysipelotrichaceae family may lead to gut in ammation and weight gain [30]. Our ndings on changes in taxa abundance of Verrucomicrobia (g_Akkermansai), in particular, a reduction in abundance in the neoADJ group but an increase in the non-neoADJ group adds to the literature of the importance of this butyrate-producing bacteria [31,32]. In a study of breast cancer patients treated by neoadjuvant chemotherapy, breast tumor tissue microbiome pro le was impacted by treatment but a comparable group of patients treated with adjuvant chemotherapy was not included in this study [33]. Nevertheless, results from this [33] and our study suggest that type of chemotherapy may impact breast and gut microbiome changes.
Our nding of a difference in the abundance of Lachnobacterium between the weight gain and weight loss groups needs con rmation. There is scant information on this genus. A recent Swedish cross-sectional study found that high intake of sugar and sweet bevarages was signi cantly inversely associated with abundance of Lachnobacterium [34]. However, another cross-sectional study found abundance levels of Lachnobacterium was higher in obese subjects than normal weight subjects and higher among individuals with low physical activity than those with high physical activity [35].
Strengths of this pilot study include the longitudinal collection of gut microbiome data on 33 breast cancer patients at multiple (baseline, during, and at the completion) time points during the rst year of treatment with either neoADJ, ADJ, or noC treatment. In addition to the detailed information on breast cancer treatment, tumor characteristics, and lifestyle information that was updated at each clinic visit, body composition was assessed using DXA at baseline and at the completion of study. Our results on gut microbiome changes were analyzed using complementary statististical methods, Wilcoxon rank sum test and negative binomial mixed models for longitudinal microbiome data with adjustment for select covariates including changes in weight. We also considered multiple comparisons and used a Bonferroni-adjusted type I error rate to evaluate p-values. Participants included whites and nonwhites, re ecting the catchment area of USC. However, we are limited by a modest sample size and the no chemotherapy group was based on only 4 patients. Although we conducted results separately for the three treatment groups, our main analysis was based on comparing neoADJ to non-neoADJ groups (i.e, ADJ + noC). Because of the inherent differences in timing of treatment between neoADJ, ADJ, and no chemotherapy groups, we were not able to collect fecal samples at a standardized interval and the period of enrollment and length of follow-up were not identical in the three groups ( Figure  1). Nevertheless, the baseline fecal samples were collected before initiation of chemotherapy for the neoADJ and ADJ groups or radiation for the no chemotherapy group and the nal fecal samples were collected when there was a recovery period of at least 100 days after completion of chemotherapy or radiation. Because this was funded as a pilot study, we only monitored patients during the rst year of treatment and did not collect information on additional treatment (e.g., hormone therapy).
In conclusion, this pilot longitudinal study found signi cant increases in gut microbiome alpha diversity measures in the neoADJ group but not in the non-neoADJ group and also intriguing changes in select Bacteroidetes and Firmicutes taxa. The dynamic nature of the gut microbiome in association with chemotherapy and weight changes highlight the need to better understand the signi cance of these ndings and how to harness this information to identify a gut microbiome pro le that would have lasting bene cial effects among women with breast cancer. Given the very modest sample size of this pilot study, we view these taxa changes as potentially informative and worthy of investigation in future studies with larger sample sizes of breast cancer patients and with longer duration of follow-up. Percentage change in alpha diversity measures (OTU, Chao 1 index, Shannon index, and PD Whole tree) by chemotherapy treatment (neoADJ n=16, ADJ n=13, no chemotherapy n=4), and by weight changes (weight loss n=16; weight gain n=17).