Gut Microbiome Diversity and Specific Composition During Immunotherapy in Japanese Responders with Non-Small Cell Lung Cancer


 Cancer immunotherapy including immune checkpoint inhibitors (ICI) has revolutionized non-small cell lung cancer (NSCLC) therapy. Recently, the microbiome status before initiation of ICI therapy has been emphasized as a predictive biomarker in patients undergoing ICI therapy. However, the microbiome diversity and composition during ICI therapy is unknown. This multicenter, prospective observational study analyzed both saliva and feces from 28 patients with NSCLC. We performed 16S ribosomal RNA gene sequencing, then analyzed associations of oral and gut microbiome diversity or composition with ICI response. Seventeen patients (responders) had a partial ICI response, and the remaining 11 patients (non-responders) had stable or progressive disease. At the genus level, the alpha diversity of the gut microbiome was significantly greater in ICI responders than in non-responders (Chao 1, p = 0.0174; PD whole tree, p = 0.0219; observed species, p = 0.0238; Shannon, p = 0.0362), while the beta diversity of the gut microbiome was significantly lower in ICI responders (principal coordinates analysis, p = 0.035). Compositional differences in the gut microbiome were observed between the two groups; in particular, Blautia was enriched in ICI responders, whereas RF32 unclassified was enriched in ICI non-responders. There were no significant differences between groups in the oral microbiome. This study revealed a strong association between gut microbiome diversity and ICI response in Japanese NSCLC patients. Moreover, specific gut microbiome compositions may influence the ICI response. These findings might be useful in identifying biomarkers to predict ICI response, as well as in developing biotic therapies to enhance the ICI response.


Background
Immunotherapy with immune checkpoint inhibitors (ICI) is widely used to treat various malignancies, including non-small cell lung cancer (NSCLC); it has revolutionized therapeutic approaches to cancer.
Although tumorous PD-L1 expression is a potential biomarker of the ICI therapeutic response, there is no widely accepted optimal biomarker to predict the e cacy of ICI, because ICI response and survival outcomes show heterogeneity in NSCLC patients receiving ICI therapy, regardless of PD-L1 expression level.
We recently reported that the pretreatment host immune-nutritional condition was a prognostic marker for NSCLC patients receiving ICI therapy [2]. Host immunity is clearly associated with the ICI response. The internal microbiome is regarded as a controlling factor in host immunity. In particular, the gut microbiome can modulate the host immune response (e.g., anti-tumor immunity) and optimize both innate and adaptive immune responses [3]. Recently, preclinical studies have shown that the gut microbiome composition and its modi cation in murine models could in uence the e cacy of ICI [4,5]. Therefore, the microbiome has been emphasized as a predictive biomarker of ICI therapy, mainly in studies from the USA or Europe. Additionally, the gut microbiome diversity or abundance of speci c gut microbiome components has been correlated with the e cacy of anti-PD-1 antibody in melanoma patients [6].
Moreover, fecal microbiome transplantation (FMT) in murine models might restore the ICI response [7,8]. In a recent study, FMT from ICI responders to ICI non-responders produced ICI e cacy in melanoma patients [9]. Furthermore, the oral microbiome has been associated with several diseases (e.g., in ammatory bowel disease and allergic diseases) through its in uence on the gut microbiome [10][11][12]. A recent study revealed that variation in the oral microbiome was associated with a risk of lung cancer [13]. However, samples were collected prior to ICI therapy in most previous studies, and thus minimal information has been available regarding the microbiome status during ICI therapy. Accordingly, FMT or biotics therapy approaches are needed to investigate changes in the microbiome during ICI therapy. Notably, there are de nite differences in microbiome composition among ethnicities [14]; to the best of our knowledge, few reports have been published regarding Japanese NSCLC patients. Here, we performed a prospective study to clarify the microbiome diversity and composition in Japanese NSCLC patients by analyzing samples collected during ICI therapy.

Patient characteristics
The results were determined in follow-up examinations over a mean duration of 598 days (range, 81-1225 days) after initial ICI therapy. Patient characteristics are shown in Table 1. The study group included seven women and 21 men, with a mean age at surgery of 71 years (range, 56-88 years). Fifteen patients (53.6%) had ECOG performance status (PS) 0 and 13 (46.4%) had ECOG-PS 1. Seven patients (25.0%) had never smoked, and the remaining 21 patients were current or FORMER smokers. The histological types were adenocarcinoma in 16 patients (57.1%) and squamous cell carcinoma in 12 patients (42.9%).
Therefore, 17 patients (PR) were regarded as ICI responders; the remaining 11 patients (SD or PD) were regarded as ICI non-responders ( Supplementary Fig. 1).

Relative abundances in oral and gut microbiomes
We analyzed the relative abundances of oral bacteria at the phylum and genus levels. At the phylum level, Firmicutes, Bacteroides, Proteobacteria, Actinobacteria, Fusobacteria, and TM 7 were the main taxa; these taxa comprised more than 99% in all groups. At the genus level, 169 species were detected in the ICI responder saliva microbiome, while 152 species were detected in the ICI non-responder saliva microbiome ( Fig. 2A, B). Among the taxa with < 1% relative abundance, 157 taxa (92.9%) were identi ed in the ICI responder saliva microbiome, while 138 taxa (90.8%) were identi ed in the ICI non-responder saliva microbiome. Taxa in both groups mainly included Streptococcus, Veillonella, Prevotella, Haemophilus, and Neisseria (Fig. 2C, D). Regarding the gut microbiome, at the phylum level, Firmicutes, Bacteroides, Actinobacteria, Proteobacteria, Fusobacteria, and Verrucomicrobia were the main taxa (comprising more than 99%) in ICI responders. Firmicutes, Bacteroides, Actinobacteria, Proteobacteria and Fusobacteria were the main taxa (comprising more than 99%) in ICI non-responders. At the genus level, 180 species were detected in the ICI responder gut microbiome, while 136 species were detected in the ICI nonresponder gut microbiome (Fig. 2E, F). Among the taxa with < 1% relative abundance, 167 taxa (92.8%) were identi ed in the ICI responder gut microbiome, while 122 taxa (89.7%) were identi ed in the ICI nonresponder gut microbiome. Taxa in ICI responders mainly included Bacteroides, Ruminococcaceae, Lachnospiraceae, Streptococcus, and Blautia; taxa in ICI non-responders mainly included Bacteroides, Ruminococcaceae, Lachnospiraceae, Prevotella, and Bi dobacterium (Fig. 2G, H).

Diversity metrics in oral and gut microbiomes
We used alpha and beta diversity indices to evaluate the intersample and intrasample relationships in the oral and gut microbiomes. Figure 3 and Fig. 4 show the alpha diversity metrics of oral and gut microbiomes at the genus level. In the oral microbiome, there were no signi cant differences between the two groups (Chao 1, p = 0.632; Observed species, p = 0.523; PD_whole_tree, p = 0.471; and Shannon, p = 0.474) (Fig. 3A-D). In the gut microbiome, all alpha diversity metrics were signi cantly higher in ICI responders than in ICI non-responders (Chao 1, p = 0.017; Observed species, p = 0.024; PD_whole_tree, p = 0.022; and Shannon, p = 0.036) ( Fig. 3E-H). PCoA assessment of beta diversity was conducted based on weighted UniFrac distance (Fig. 4). In both oral and gut microbiomes, similar patterns were evident in the two groups. The greatest variations in the oral and gut microbiomes of the two groups were 4.39% (PC1) and 4.32% (PC2), and 4.51% (PC1) and 4.22% (PC2), respectively. PERMANOVA based on unweighted UniFrac distance con rmed signi cant differences between the two groups in the gut microbiome alone

LEfSe results
We used LEfSe to perform high-dimensional genus comparisons regarding oral and gut microbiomes between ICI responders and non-responders. Figure 5 shows that the ICI responder gut microbiome was signi cantly enriched for Blautia, compared with the ICI non-responder gut microbiome. In contrast, the ICI non-responder gut microbiome was signi cantly enriched for RF32 unclassi ed, compared with the ICI responder gut microbiome. There were no signi cant differences in the oral microbiome between the two groups.

Discussion
Similar studies regarding ICI therapy have focused on the microbiome in patients who have not yet received ICI therapy. Those studies revealed that the microbiome diversity and composition before ICI therapy was a predictive biomarker for ICI response. Although variations in gut microbiome composition were observed in the previous studies, there has been minimal information regarding gut microbiome status during ICI therapy. This information is important for efforts to enhance ICI therapy through biotics therapy (e.g., pre-, pro-, and synbiotics) and/or FMT. This analysis of oral and gut microbiome pro les in Japanese NSCLC patients during ICI therapy produced several novel ndings.
The gut microbiome might have an important role in the ICI response in NSCLC patients, although the oral microbiome conveyed information distinct from the gut microbiome. Greater numbers of both oral and gut microbiome species were observed in ICI responders than in ICI non-responders. Additionally, those microbiomes mainly consisted of minor species (< 1%) at the genus level. Moreover, the 4th and 5th majority of gut microbiome species in ICI responder were differed from that in ICI non-responder, while 1st to 5th majority of oral microbiome were same between two groups.
An important tool for objective evaluation of the above data involves analysis of microbiome diversity: the numbers or abundances of microorganisms colonizing the gut. Greater alpha diversity indicates larger numbers of species in the gut, which implies a distinct gut microbiome composition. Several studies have reported that the ICI response was in uenced by the alpha diversity of the gut microbiome before ICI therapy [8,27]. In the present study, high alpha diversity was observed in the ICI responder gut microbiome. Thus, our results indicated that ICI responders had more abundant gut microbes, compared with ICI non-responders, during ICI therapy.
Our study also revealed a signi cant association between beta diversity and ICI response. Higher beta diversity indicates a signi cant difference in gut microbiome composition between two samples. In this study, the intersample distance was signi cantly shorter in ICI responders than in ICI non-responders.
Therefore, the ICI responder gut microbiome had signi cant similarity, compared with the ICI nonresponder gut microbiome, which implies a simple approach to control the gut microbiome by adding biotics therapy to ICI therapy.
Additionally, we identi ed speci c gut microbiome species in Japanese NSCLC patients receiving ICI therapy by using LEfSe.  [29] and Firmicutes [29]) and clinical response to ICI therapy. These results differed among individual studies, including the present study, presumably due to factors such as patient disease (e.g., melanoma, renal cell carcinoma, and NSCLC) and ICI regimen. Indeed, Frankel et al [28] indicated that the gut microbiome composition depended on the ICI regimen. Moreover, these differences might be caused by ethnicities. Nishijima et al [14] compared the gut microbiome between the Japanese population and individuals from 11 other nations. Notably, the Japanese gut microbiome was considerably different from the microbiomes of other populations; it was characterized by the highest abundances of Blautia, Bi dobacterium, Collinsella, Streptococcus, and an unclassi ed Clostridiales genus, compared with the microbiomes from the remaining 11 countries.
In addition to its status as a species characteristic of the gut microbiome in Japanese individuals, Blautia coccoides is regarded as an effective probiotic species. B. coccoides is an anaerobe and Gram-positive species found in human fecal samples. Reduced numbers of B. coccoides are associated with several benign diseases (e.g., hepatic cirrhosis and encephalopathy, irritable bowel syndrome, acute diarrhea, idiopathic in ammatory bowel disease, intestinal in ammation, and diabetes mellitus) and some malignancies (e.g., colorectal and breast cancers) [30,31]. Furthermore, some reports have revealed that increased numbers of B. coccoides might be bene cial for human health. B. coccoides is increased among individuals with diets high in resistant starch and arabinoxylan [32]; moreover, it can reduce NF-κB activity in human colon cancer cells [33]. Additionally, Blautia obeum is a gut microbiome component involved in the transformation of carcinogenic heterocyclic amines, and reduced abundance of this species may increase heterocyclic amine-induced colorectal cancer risk [34]. Myles et al [35] reported that high omega-3 intake altered colonic in ammation and increased Blautia abundance in a murine model. Therefore, Blautia may be a key gut microbiome component involved in the ICI response in Japanese NSCLC patients. Conversely, the relative abundance of RF32 unclassi ed has been positively correlated with colonic damage and in ammation [36], which are presumably negative in uences on immunity in NSCLC patients.
The ndings in this study indicate that controlling both gut microbiome diversity and the abundances of speci c gut microbiome species during ICI therapy might lead to ICI response enhancement in Japanese NSCLC patients. Future therapies targeting the gut microbiome by means of pre-, pro-, or synbiotics to enhance the ICI response might be considered from our ndings.
This study had the following limitations. First, only a single sample collection was performed. Thus, it was unclear how the microbiome diversity and composition might have changed before and after ICI therapy. Second, the sample size was small, which may have interfered with meaningful conclusions. Third, this study allowed various ICI regimens with or without combination chemotherapy. In the future, we plan to perform a large, multicenter, prospective observational study to evaluate the association between ICI response and changes in gut microbiome by collecting samples at multiple points (before and during ICI therapy) for NSCLC patients receiving a speci c ICI regimen.
In conclusion, this study revealed a strong association between gut microbiome diversity and ICI response in Japanese NSCLC patients. Moreover, speci c gut microbiome compositions may in uence the ICI response. These ndings might be useful in identifying biomarkers to predict ICI response, as well as in developing biotic therapies to enhance the ICI response. underwent computed tomography (CT) of the thorax and upper abdomen, as well as bone scintigrams, brain CT scans, magnetic resonance imaging (MRI), or uorodeoxyglucose-positron emission tomography (FDG-PET). Postoperative local or distant recurrence was de ned as described previously [17]. ICI therapy was continued until radiographic progression. PD-L1 protein expression was evaluated using antibody clone 22C3 (Dako, Agilent Technologies, Santa Clara, CA, USA).

Sample Collection
Salivary and fecal samples were collected in sterile containers and immediately placed at 4°C, then frozen at -80°C. Individual periods of sample collection are shown in Fig. 1 [20]. Barcoded amplicons were paired-end sequenced on a 2×284-bp cycle using the MiSeq system with MiSeq Reagent Kit chemistry, version 3 (600 Cycle). Paired-end sequencing reads were merged using the fastq-join program with default settings [21].
Only reads with quality value (QV) scores of ≥ 33 were extracted with split_libraries_fastq.py command in QIIME, version 1.8.0 [22]. Chimeric sequences were removed using USEARCH61 [23] with the identify_chimeric_seqs.py command in QIIME [22]. Operational taxonomic units (OTUs) were aligned using the pick_open_reference_otus.py command in QIIME [22]. OTUs with 97% similarity were identi ed with the Greengenes database, version 13.8 [24]. Alpha diversity indices (e.g., observed species, Chao-1, Shannon, and PD_whole_tree) and beta diversity indices (e.g., principal coordinates analysis [PCoA]) were analyzed using the alpha_rarefaction.py and beta_diversity.py commands in QIIME, respectively [23]. The Chao-1 index was used to determine community richness and the Shannon index was used to determine community diversity. The PD_whole_tree index was used to compute Faith's phylogenetic diversity. PCoA was used to show differences between the two groups. Unweighted UniFrac metrics were used for beta diversity [25]. Linear discriminant analysis effect size (LEfSe) measurements were used to quantify differential taxonomic and functional pathway abundances between responders and non-responders [26].

Statistical analysis
Categorical variables were analyzed using Fisher's exact test. Continuous variables were compared using the chi-squared test. Statistical analyses were performed using JMP software, version 14.0 (SAS Institute, Inc., Cary, NC, USA). The adonis function in the vegan package of R software, version 3.6.1, was used to conduct permutational multivariate analysis of variance (PERMANOVA) with respect to microbiome composition. P-values < 0.05 were considered statistically signi cant.  Principal coordinates analysis (A and C) and boxplots (B and D) of microbiome data based on unweighted UniFrac distances between ICI non-responders and responders at the genus level.