A metagenome association study of gut microbiome revealed biomarkers for chemotherapy ecacy in locally advanced and advanced lung cancer

Accumulating evidence has conrmed the important role of the gut microbiome in the development and immunotherapy ecacy of lung cancer. However, little is known about the relationship between intestinal ora and chemotherapy. This study investigates the correlation between intestinal ora and chemotherapy ecacy in lung cancer. We analyzed baseline stool samples from patients with locally advanced and advanced lung cancer before chemotherapy treatment, through metagenomics of the gut microbiota. The composition, diversity, function, and metabolic pathway analysis of the microbial communities were compared, using the R statistical programming language, among patients with different clinical outcomes.

antineoplastic chemotherapy drug, could be converted to toxic metabolites by bacterial betaglucuronidase [21]. A lung cancer mouse model constructed by Gui et al. showed that normal intestinal ora, such as Lactobacillus, may enhance the antitumor effect of cisplatin by increasing serum levels of interferon gamma (IFN-γ) [22]. Previous studies have focused on exploring the role of gut microbes in few individual chemotherapeutic drugs. However, the relationship between the metabolism of most chemotherapeutic drugs and intestinal microorganisms has not been clari ed, and neither has the exact role of the gut microbiome in systemic chemotherapy schemes for patients with locally advanced and advanced lung cancer. Moreover, the correlations between gut bacteria and certain chemotherapy prognostic indices have not yet been observed. Therefore, to help to elucidate this new and extremely interesting eld, we conducted a metagenome sequencing analysis of microbiota in stool samples to analyze the correlation between clinical outcomes and the gut microbiome and explore whether speci c ora can predict the e cacy of chemotherapy in patients with lung cancer. This work may lay a theoretical foundation for improving the e cacy of chemotherapy by regulating intestinal ora or even ora transplantation in the future.

Research patients and subjects
The study design was a single-site, correlative study on the effects of the gut microbiome on the e cacy of rst-line systematic chemotherapy in locally advanced and advanced lung cancer patients. This study was approved by the Research Ethics Board of the Cancer Hospital of the Chinese Academy of Medical Sciences (Beijing, China), was conducted in accordance with the principles of the Helsinki Declaration, and informed consent was obtained from all subjects.
Locally advanced and advanced lung cancer patients treated with rst-line chemotherapy were enrolled in the study between September 2018 and September 2019. The following exclusion criteria were strictly followed: (i) the clinical diagnosis of mental disorders; (ii) a history of gastrointestinal surgery; (iii) a previous diagnosis of a gastrointestinal disease, any autoimmune or metabolic disease; (iv) combined with treatments for other cancers; (v) the occurrence of acute or chronic infections in the past six months; (vi) use of antibiotics, probiotics, or steroids within the past six months. All patients enrolled in our study had a de nite histologic pathological diagnosis of lung cancer, for the rst time, according to the diagnostic criteria proposed by the 8th edition of the American Joint Committee on Cancer in 2018 [23], and had been prescribed systemic rst-line chemotherapy. Additionally, no patients received radiotherapy, targeted therapy, surgery, or immunotherapy for lung cancer before sample collection. Furthermore, patients enrolled had to have measurable lesions according to the response evaluation criteria in solid tumors version 1.1 (RECIST v1.1) [24]. Tumor size was assessed by computerized tomography scan and/or magnetic resonance imaging within 4 weeks before the start of treatment. The treatment strategies for patients enrolled include: (i) pemetrexed combined with cisplatin or carboplatin ± bevacizumab for patients with lung adenocarcinoma; (ii) paclitaxel or gemcitabine in combination with cisplatin or carboplatin for lung squamous cell carcinoma; (iii) etoposide in combination with cisplatin or carboplatin for small cell lung cancer; (iv) paclitaxel combined with cisplatin or carboplatin for lung adenosquamous carcinoma. Repeat exams and scans were taken every 2 cycles of chemotherapy.
Curative effects were evaluated using the RECIST v1.1 criteria [24], and patients were divided into responders (R) and non-responders (NR) groups according to treatment e cacy. Progression-free survival (PFS) was de ned as the interval (in months) from the date of chemotherapy to the date of progression. Moreover, chemotherapy-related bone marrow suppression, nausea, vomiting, and other adverse events throughout the treatment period were recorded according to the National Cancer Institute's common terminology criteria adverse events version 3.0 [25] for each patient.

Samples collection and storage
Stool samples were collected from all patients after diagnosis of locally advanced and advanced lung cancer and before any treatment. All participants had a bland diet and did not smoke or consume alcohol the day prior to sample collection. Fecal samples from all participants were taken at a fresh feces center using a sterile cotton swab, placed in a sterile plastic vial mixed with phosphate-buffered saline, and immediately transferred to a freezer at -80 °C. All samples were stored at -80 °C until further processing.

DNA extraction and sequencing
Fecal bacterial DNA was extracted using the QIAamp PowerFecal Pro DNA Kit (QiaGen, Venlo, Netherlands). Sodium dodecyl sulfate-Tris solution, glass beads (diameter 0.1 mm) (BioSpec), and EDTA-Tris-saturated phenol were added to the suspension, and the mixture was vortexed vigorously by a FastPrep-24 (MP Biomedicals) at 5.0 power level for 30 seconds, with a collection of the supernatant after centrifugation at 20 000 g for 5 minutes. Subsequently, a phenol-chloroform extraction was performed and the supernatant was subjected to isopropanol precipitation. Finally, the DNA was stored at -20 °C. The concentration and purity of the DNA was tested on 2% agarose gels. The ampli ed DNA was further subjected to library preparation (KAPA HyperPlus PCR-free (96 rxn)) and sequenced on the Illumina MiSeq platform as per the manufacturer's instructions (Illumina technologies, USA).

Data Quality Control
All raw data passed quality control by MOCAT2 and low-quality reads were discarded [26]. Cutadapt software (version v1.14, -m30) was used to remove the sequencing adapter. Then, clean reads were obtained by ltering out low quality reads < 20 or short reads < 30 base pairs (bp) with the SolexaQA package [27]. Finally, SOAPaligner (version v2.21, -M 4 -l 30 -v 10) was applied to get high-quality clean reads for analysis, which were aligned to the human genome (H. sapiens, UCSC hg19) without contaminated host reads [28].

De Novo Assembly
The clean data were assembled by the SOAPdenovo software (version v2.04, an iterative De-Bruijn Graph De Novo Assembler), with parameters as follows: -D 1, -M 3, -L500, for constituting scaftigs of at least 500 bp.

Non-Redundant Metagenomic Gene Catalogue Construction
Genetic structure predictions were carried out with MetaGeneMark [29]. A non-redundant gene catalogue of prediction genes was constructed with CD-HIT [30]. High quality reads were mapped onto the gene catalogue using Burrows-Wheeler Alignment tool for calculation of gene abundance.

Statistical Analyses
MetaPhlAn2 was used to determine the microbial components, including the relative abundance of each level, that is kingdom, phylum, class, order, family, genus, and species [31]. Statistical analyses, such as the composition, diversity, difference, function, and metabolic pathway analyses, were performed using R (version 3.4.3) statistical programming language. Spearman correlation analyses between microbiome and clinical phenotypes were performed with R. Moreover, variation analysis in microbiome between different chemotherapy e cacy groups was identi ed using Wilcoxon rank-sum permutation test and P value with adjustments according to Benjamini-Hochberg. Non-linear unsupervised clustering analysis was used for further veri cation. Heatmap showing the unsupervised clustering of the microbiota relative abundance data was performed by ComplexHeatmap in R, in which the cluster_rows and cluster_columns were clustered by Euclidean [32]. After classifying into clusters, we determined the microbiome biomarkers at the species level that showed chemotherapy e cacy combined with clinical e cacy. Finally, we performed gene set enrichment analysis (GSEA) using R, and the genes annotated by the HUMAnN2 gene database were ranked. The gene sets were de ned according to the HUMAnN2 pathway for functional analysis of the metabolic pathways.

Patients characteristics
From September 1, 2018 to September 30, 2019, a total of sixty-four patients with locally advanced and advanced lung cancer undergoing de nitive chemotherapy were enrolled in the study, provided pretreatment fecal samples, and had follow-up exams and scans. Clinical characteristics, including age, gender, smoking and drinking history, pathological tumor type, clinical stage, and chemotherapy regimen, e cacy, and adverse reactions, were all fully recorded (Table 1). Patients were predominantly male (n = 48, 75%), the median age was 60 years with a range of 33 to 78. Forty-ve patients were in advanced stage with different sites of metastases, including bone, brain, pulmonary, pleura, and liver. The pathological types were adenocarcinoma, squamous cell carcinoma, adenosquamous carcinoma, and small cell lung cancer with a ratio of 34 (53.125%), 10 (15.625%), 2 (3.125%), and 18 (28.125%), respectively. The chemotherapy regimens were pemetrexed + platinum ± beacizumab for adenocarcinoma, paclitaxel/gemcitabine + platinum for squamous cell carcinoma and adenosquamous carcinoma, and etoposide + platinum for small cell lung cancer. The clinical outcomes were that 33 patients showed RECIST response to chemotherapy, whom we classify as R in this study, and 31 patients did not (NR). In addition, the median progression-free survival was 7 months (range, 1.5-14.5). No apparent discrepancy in fecal bacterium diversity We constructed a non-redundant gene set from 64 stool samples. The number of genes in R and NR groups were 1 930 858 and 1 984 255, respectively, of which 1 683 584 were part of the universal gene set ( Fig. 1a-b). Species richness, shown by species accumulation curves, indicated that the reads obtained from both groups represented most of the microbiome present in the samples (Fig. 1c). Alpha diversity was determined, by Shannon index, to analyze the complexity of species diversity in each sample. No differences were found in the alpha diversity indices between the R and NR groups (Fig. 2a).
In addition, there was no signi cant difference in beta diversity as constructed by the principal coordinates analysis (PCoA) based on the bray-curtis distance of the top several ora species (Fig. 2b), indicating that the primary differences may lie in the less abundant microbiota. The TOP20 microbiome species correlated with the difference between R and NR groups are shown in Table 2, which were obtained using the similarities percentage method (SIMPER). An additional tabular data le shows this in more detail [see Additional le 1]. Correlation between gut microbiome and clinical phenotypes Spearman correlation analyses between microbiome and clinical phenotypes showed that different kinds of clinical manifestations, such as age, body mass index (BMI), pathology, and metastatic sites, were associated with speci c ora. Spearman rank correlation coe cients were represented by a heatmap (Fig. 3), with red and blue representing positive correlation and negative correlation, respectively. The results showed that age was inversely related to Prevotella disiens (P < 0.01) and Enterococcus gallinarum (P < 0.05); BMI was inversely related to Clostridium hylemonae (P < 0.01) and had positive correlation with Streptococcus thermophilus and Coprococcus comes (P < 0.05). Patients who reported long-term smoking, were associated with higher abundance of Campylobacter concisus (P < 0.05) and lower abundance of Streptococcus thermophilus (P < 0.01) and Dorea longicatena (P < 0.05). Five species, such as Dorea longicatena and Streptococcus parasanguinis (P < 0.001), were reduced in patients with a long history of drinking. Collinsella intestinalis was inversely related to lung adenocarcinoma (P < 0.05), while it presented at a higher abundance in patients with small cell lung cancer (P < 0.05). Mitsuokella multacida (P < 0.05) and Alloscardovia omnicolens (P < 0.01) were enriched in patients with squamous cell lung carcinoma. In addition, baseline metastatic sites also had obvious correlation with different ora. Eleven species, including Rothia dentocariosa (P < 0.001) and Solobacterium moorei (P < 0.01), were more abundant in lung cancer patients with pleural metastasis at baseline. Porphyromonas uenonis (P < 0.01) and three other ora were enriched in patients with pulmonary metastasis. While patients with hepatic metastases had higher abundance of Pseudomonas mandelii (P < 0.001), Campylobacter hominis (P < 0.001), and six other species. Moreover, both clinical e cacy and adverse events after chemotherapy were associated with certain bacteria. The enrichment of Bacteroides nordii and Ruminococcus sp_5_1_39BFAA were associated with severe adverse events after chemotherapy (P < 0.01). However, Gardnerella vaginalis was inversely related to adverse events (P < 0.01). For treatment, Eubacterium siraeum (P < 0.01), Leuconostoc lactis (P < 0.01), Rothia dentocariosa (P < 0.05), and two other ora had negative correlations with e cacy. Besides, Rothia dentocariosa showed signi cant correlations with poorer e cacy and shorter PFS (P < 0.05).

Differences in bacterial communities among between different e cacy groups
To characterize the differences in the gut microbial communities between the R and NR groups, we conducted a variance analysis using the Wilcoxon rank-sum permutation test to determine the differences in the relative abundance of ora (Table 3). An additional tabular data le shows this in more detail [see Additional le 2]. The relative abundances of 13 species were signi cantly different between the two groups. Streptococcus mutans (P = 0.026) and Enterococcus casseli avus (P = 0.049) were enriched in R group, while 11 bacteria, including Leuconostoc lactis (P = 0.002) and Eubacterium siraeum (P = 0.006), were enriched in NR group. Signi cantly different predominant taxa (TOP10) among gut microbiota species were shown in a box comparison plot (Fig. 4a). In addition, the metagenomic biomarker discovery approach was used to identify the phylotypes responsible for the greatest differences in gut bacteria at the operational taxonomic unit level for distinguishing between the R and NR groups by Lefse analysis, showing that the responders to chemotherapy were associated with signi cantly higher levels of Acidobacteria and Granulicella. While Streptococcus oligofermentans, Megasphaera micronuciformis, and Eubacterium siraeum were more abundant in non-responders ( Fig. 4b-c). Identi cation of signi cant taxa clusters at the species level as bacterial markers for the e cacy of chemotherapy using unsupervised clustering The discrepancies at the species level were comprehensively assessed by the deconvolution of the metagenome data. Unsupervised clustering was performed using cluster_rows and cluster_columns with euclidean by ComplexHeatmap in R, which classi ed the species into 5 clusters. Two additional tabular data les show these in more detail [see Additional le 3 and 4]. Finally, a heatmap based on the unsupervised hierarchical clustering displayed species differences between individuals (Fig. 5).
Comparing the clustering data with the treatment e cacy for each patient allowed for the identi cation of signi cant taxa clusters at the species level as bacterial markers for chemotherapy e cacy in locally advanced and advanced lung cancer patients. Streptococcus mutans and Enterococcus casseli avus were signi cantly enriched in the R group, consistent with the result of variance analysis, and can be used as biological biomarkers for e cacy of chemotherapy. On the other hand, Leuconostoc lactis and Eubacterium siraeum were the bacterial markers for the NR group, whose P values also consistently showed statistical signi cance.

Differences in metabolic pathway analysis for functional analysis
To fully understand the differences in the metabolic networks between the R and NR groups, all the metabolites with differential expression were submitted to HUMAnN2 for metabolic pathway enrichment analysis. An additional tabular data le shows this in more detail [see Additional le 5]. Three metabolic pathways were enriched in NR group, including "PWY-241: C4 photosynthetic carbon assimilation cycle, NADP-ME type" (P < 0.001), "P23-PWY: reductive TCA cycle I" (P = 0.007) and "P461-PWY: hexitol fermentation to lactate, formate, ethanol, and acetate" (P = 0.025). While the metabolic pathway called "PWY-5088: L-glutamate degradation VIII (to propanoate)" was abundant in the R group (P = 0.014).
Additionally, we calculated the metabolite potential based on the relative abundance of the metabolic micro ora producing or digesting metabolites and conducted a variation analysis. The heatmap of signi cantly different metabolites is shown in Fig. 6, and the results suggested that the aliphatic acid or carbohydrate pathways may be used to distinguish between the two groups.

Discussion
The gut microbiome may be a modi able factor that affects cancer treatment e cacy and toxicity [33]. In the present study, we characterized the composition and differences in the gut bacteria associated with different chemotherapeutic outcomes from 64 lung cancer patients. This is the rst detailed report of human gut microbiome metagenomic pro ling in lung cancer patients treated with rst-line chemotherapy. In this study, we have demonstrated the correlation of gut microbiota with clinical phenotypes in lung cancer patients and identi ed speci c microbial candidates that might contribute to predicting the chemotherapy e cacy. These ndings provided a broader understanding of the effect of the gut microbiome on chemotherapy e cacy in lung cancer patients, paving the way for further investigation in this research area.
Our results showed that the abundance of Prevotella disiens and Enterococcus gallinarum declined with age. Previous studies have shown that the translocation of gut micro ora, such as Enterococcus gallinarum, to systemic tissues triggers intense autoimmune responses [34]. However, the correlation between the abundance of these species and age was rst proposed in this study, which deserves to be further investigated. Smoking is a recognized risk factor for lung cancer, whose pathogenic mechanism has been extensively studied [35]. The abundance of Streptococcus thermophilus was obviously reduced in patients with a history of long-term smoking. Streptococcus thermophilus possess in vitro probiotic properties along with anticancer activity [36]. Therefore, taking into consideration previous microbiome studies and our correlation analysis, we recommend quitting smoking to maintain the abundance of bene cial bacteria. Different pathological types of cancer appear to have their own unique gut microbial characteristics [37]. In our study, lung cancer patients with different types of pathology also showed distinct microbial signatures. For example, Collinsella intestinalis was reduced in lung adenocarcinoma, while enriched in small cell lung cancer, and Mitsuokella multacida and Alloscardovia omnicolens were enriched in squamous cell lung carcinoma. These differences may help distinguish the pathological cancer type of patients who do not have adequate conditions for puncture biopsy diagnosis. Several microbes such as Blautia obeum and Akkermansia muciniphila have been proved to be increased only in metastatic lung cancer patients [37]. Our study further reveals the differences in the ora associated with different metastatic sites (Fig. 3), which may lay the foundation for the individual management of patients.
Data from a previous report has suggested that the treatment effects, including chemotherapy or a combination of chemotherapy with immunotherapy, in cancer patients are positively correlated with some speci c types of gut microbiota, such as Bacteroides ovatus and Bacteroides xylanisolvens [38]. However, the correlation between the e cacy of chemotherapy in lung cancer patients and certain speci c ora is still unclear. In our study, we proposed microbial biomarkers to predict the e cacy of chemotherapy in patients with locally advanced and advanced lung cancer for the rst time. Through correlation analysis, variance analysis, and unsupervised clustering, we gradually determined the gut microbes that might predict the e cacy of chemotherapy in lung cancer patients. Our results showed that Streptococcus mutans and Enterococcus casseli avus were biomarkers for better chemotherapy outcomes. Streptococcus mutans is an organism from carious lesions, whose natural habitat is the human oral cavity, with possible translocation to other tissues [39]. The production of glycan, a hydrolytic proteoglycan, by this microbe gives it the ability to adhere to epithelial cells, which can affect the cell-cell adhesions of the host, by in uencing the level of functional E-cadherin at the cell-cell border, and enhance the process of tumor cell dissociation and invasion [40]. However, there are no studies on the impact of treatment e cacy. Interestingly, in our study, Streptococcus mutans was found to be enriched at baseline in patients with better outcomes, indicating that it may potentially contribute to chemotherapy e cacy in lung cancer. Gut microbiota, including Enterococcus, has been found to be signi cantly higher in cancer patients, especially those with colorectal cancer, than in healthy people, demonstrating the relevance of this ora to cancer [41]. Moreover, Enterococcus is more abundant in metastatic melanoma patients that responded to immunotherapy [42], which may lead to improved tumor control and greater e cacy of immunotherapy by augmented T cell responses. In our study, Enterococcus casseli avus was also shown to be related to better chemotherapy e cacy in lung cancer. Therefore, the combination of Streptococcus mutans and Enterococcus casseli avus seemed to be biomarkers for greater chemotherapy e cacy. On the other hand, Leuconostoc lactis and Eubacterium siraeum were the bacterial markers associated with poor chemotherapy effect. Leuconostoc bacteria were initially considered for a wide range of uses in the food industry due to their fermentation properties and odorproducing compounds [43]. In recent years, the role of probiotics in anti-tumor therapy has been explored [44]. Leuconostoc mesenteroides isolated from traditional dairy products promoted apoptosis in colon cancer by regulation of MAPK1, AKT, NF-kB, and some key oncomicroRNAs [44]. Interestingly, we found another species, Leuconostoc lactis, that had a negative effect on lung cancer chemotherapy, which contradicts the probiotic characteristics of this genus shown in previous studies, and is worth being explored in depth. Eubacterium rectalie seemed to be more abundant in healthy persons than in patients with prostate cancer [45]. However, the relationship with anti-tumor therapy has not been investigated. Eubacterium siraeum was negatively corelated with chemotherapy e cacy for lung cancer patients in our study, which poses a new challenge for the role of this ora in anti-tumor therapy. Overall, we proposed that enrichment of Leuconostoc lactis and Eubacterium siraeum represented a biomarker for poorer chemotherapy e cacy in lung cancer. Finally, the detailed molecular mechanism for enhancement of chemotherapy e cacy by any of the bacteria discussed in this study remains unknown, which needs exploring with in-depth experiments in a mouse model.
Taking the challenges of the lability of mRNA and di culty in standardized sample collection for metatranscriptomics into consideration, we have attempted a metagenomic functional pathway analysis to gain functional insight into changes in the gut microbiome. However, the enriched or reduced abundance of genes in a certain microbiome is still di cult to explain. As studied previously, anacardic acid had an antitumor effect, mediated by enhanced T-cell recruitment, in several preclinical models [46].
Similarly, based on the analysis of metabolic pathways and metabolites in our study, hexitol fermentation and the aliphatic acid or carbohydrate pathways may be reasonable potential targets for metabolic intervention combined with chemotherapy.
Our study had some limitations. First, the sample size of lung cancer patients treated with chemotherapy was relatively small. Therefore, we could not comprehensively and systematically pro le the microbial biomarkers for chemotherapy e cacy in lung cancer patients. Thus, a larger number of individuals is needed to verify our ndings. Second, we did not monitor the dynamic bacterial community structure of patients during chemotherapy, which might lead to a better understanding of the alterations in gut bacteria associated with chemotherapy. Therefore, further research on this subject is required, and studies with a longitudinal design using the lung cancer animal model treated with chemotherapy drugs to investigate the underlying mechanisms of the relationship between gut bacteria and chemotherapy e cacy in lung cancer are preferable.

Conclusions
In conclusion, we present the description of the gut microbiota in patients with different chemotherapy outcomes, providing a signi cant rst step in understanding the relationship between the gut microbiome and chemotherapy e cacy in lung cancer. Our work not only extends this observation to monitoring the therapeutic response in lung cancer patients treated with chemotherapy, but also may facilitate clinical therapeutic strategies from a microbial perspective.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests.   Figure 1 The number of sequenced genes and species accumulation curve in responders and non-responders. The Wayne diagram shows the non-redundant metagenomic gene catalogues constructed by high quality reads in the responder and non-responder groups, presented in a pie chart (a) and a box comparison chart (b). Species accumulation curve, with a tendency to gradually atten out, was projected using vegan in R programming language (c).

Figure 1
The number of sequenced genes and species accumulation curve in responders and non-responders. The Wayne diagram shows the non-redundant metagenomic gene catalogues constructed by high quality reads in the responder and non-responder groups, presented in a pie chart (a) and a box comparison chart (b). Species accumulation curve, with a tendency to gradually atten out, was projected using vegan in R programming language (c).

Figure 1
The number of sequenced genes and species accumulation curve in responders and non-responders. The Wayne diagram shows the non-redundant metagenomic gene catalogues constructed by high quality reads in the responder and non-responder groups, presented in a pie chart (a) and a box comparison chart (b). Species accumulation curve, with a tendency to gradually atten out, was projected using vegan in R programming language (c).

Figure 2
Alpha and beta diversity between responders and non-responders. Boxplots showing the alpha diversity were evaluated by Shannon index using vegan in R programming language (a). Principal coordinates analysis (PCoA) revealing the beta diversity for responders and non-responders were exhibited with braycurtis distance (b). The rst two principal coordinates (PCs) were labeled with the percentage of variance explained (12.7 % and 9 %).

Figure 2
Page 23/36 Alpha and beta diversity between responders and non-responders. Boxplots showing the alpha diversity were evaluated by Shannon index using vegan in R programming language (a). Principal coordinates analysis (PCoA) revealing the beta diversity for responders and non-responders were exhibited with braycurtis distance (b). The rst two principal coordinates (PCs) were labeled with the percentage of variance explained (12.7 % and 9 %).

Figure 2
Alpha and beta diversity between responders and non-responders. Boxplots showing the alpha diversity were evaluated by Shannon index using vegan in R programming language (a). Principal coordinates analysis (PCoA) revealing the beta diversity for responders and non-responders were exhibited with braycurtis distance (b). The rst two principal coordinates (PCs) were labeled with the percentage of variance explained (12.7 % and 9 %).    Differences in the microbiomes of patients with different chemotherapy e cacy outcomes. Signi cant differences in relative abundance of the predominant taxa among gut microbiota species between responders and non-responders were projected as box comparison charts (a). Differential taxonomic abundance between responders and non-responders (b, c) were analyzed by linear discriminate analysis coupled with effect size measurements (Lefse) as a histogram (c) and a cladogram (b). The listed bacterial oras were signi cantly (P < 0.05, Kruskal-Wallis test) gathered for their respective groups.

Figure 4
Differences in the microbiomes of patients with different chemotherapy e cacy outcomes. Signi cant differences in relative abundance of the predominant taxa among gut microbiota species between responders and non-responders were projected as box comparison charts (a). Differential taxonomic abundance between responders and non-responders (b, c) were analyzed by linear discriminate analysis coupled with effect size measurements (Lefse) as a histogram (c) and a cladogram (b). The listed bacterial oras were signi cantly (P < 0.05, Kruskal-Wallis test) gathered for their respective groups.

Figure 4
Differences in the microbiomes of patients with different chemotherapy e cacy outcomes. Signi cant differences in relative abundance of the predominant taxa among gut microbiota species between responders and non-responders were projected as box comparison charts (a). Differential taxonomic abundance between responders and non-responders (b, c) were analyzed by linear discriminate analysis coupled with effect size measurements (Lefse) as a histogram (c) and a cladogram (b). The listed bacterial oras were signi cantly (P < 0.05, Kruskal-Wallis test) gathered for their respective groups.

Figure 5
Heatmap of species differences between individuals based on unsupervised hierarchical clustering.
Unsupervised hierarchical clustering was applied to draw the heatmap, with red, blue, and white indicating enrichment, reduction, and no correlation, respectively, which showed the microbial species abundance was different between the responders (Effect = 1, red dot) and non-responders (Effect = 0, green dot). Color of boxes indicates relative abundance, and darker colors indicate greater abundance.

Figure 5
Heatmap of species differences between individuals based on unsupervised hierarchical clustering.
Unsupervised hierarchical clustering was applied to draw the heatmap, with red, blue, and white indicating enrichment, reduction, and no correlation, respectively, which showed the microbial species abundance was different between the responders (Effect = 1, red dot) and non-responders (Effect = 0, green dot). Color of boxes indicates relative abundance, and darker colors indicate greater abundance.

Figure 5
Heatmap of species differences between individuals based on unsupervised hierarchical clustering.
Unsupervised hierarchical clustering was applied to draw the heatmap, with red, blue, and white indicating enrichment, reduction, and no correlation, respectively, which showed the microbial species abundance was different between the responders (Effect = 1, red dot) and non-responders (Effect = 0, green dot). Color of boxes indicates relative abundance, and darker colors indicate greater abundance.

Figure 6
Heatmap of signi cantly different metabolites responders and non-responders. Heatmap shows differentially abundant metabolites between responders and non-responders, in which columns represent samples (green = responders, purple = non-responders) and rows metabolites. The heatmap visualization is used to encode individual abundance of the metabolites for each sample as colors (red, relative enrichment; blue, relative reduction; white, no correlation). Designation of metabolites were indicated on the right hand-side of the gure. Heatmap of signi cantly different metabolites responders and non-responders. Heatmap shows differentially abundant metabolites between responders and non-responders, in which columns represent samples (green = responders, purple = non-responders) and rows metabolites. The heatmap visualization is used to encode individual abundance of the metabolites for each sample as colors (red, relative enrichment; blue, relative reduction; white, no correlation). Designation of metabolites were indicated on the right hand-side of the gure. Heatmap of signi cantly different metabolites responders and non-responders. Heatmap shows differentially abundant metabolites between responders and non-responders, in which columns represent samples (green = responders, purple = non-responders) and rows metabolites. The heatmap visualization is used to encode individual abundance of the metabolites for each sample as colors (red, relative enrichment; blue, relative reduction; white, no correlation). Designation of metabolites were indicated on the right hand-side of the gure.