The gut microbiota as a potential biomarker in patients with EGFR-mutant lung cancer.

Introduction
Lung cancer is the leading cause of cancer-related mortality worldwide1,2, with non-small cell lung cancer (NSCLC) being the major histological subtype. Approximately 50% of Asian NSCLC patients carry epidermal growth factor receptor (EGFR) mutations3, the most common of which are deletion of exon 19 and L858R substitutions in exon 214. Since these mutations are associated with high sensitivity to EGFR-tyrosine kinase inhibitors (EGFR-TKIs), these drugs are the standard front-line treatment and have improved clinical outcomes for patients with advanced disease over and above those achieved by traditional cytotoxic therapies5. Although EGFR-TKIs are highly effective, biomarkers that predict efficacy and adverse events are needed in clinical practice.
Many have described an association between cancer and the gut microbiota. Recently, the relationship between the gut microbiota and the immune system has gained much attention6. Growing evidence supports the concept of the lung–gut axis, a bidirectional communication pathway between the intestinal and respiratory microbiota. Alterations in the gut microbiota can affect pulmonary immunity and inflammatory responses through microbial metabolites, such as short-chain fatty acids (SCFAs) and bile acids, and through systemic cytokine signaling. Conversely, lung inflammation or tumor-associated factors may disrupt intestinal homeostasis7–9. These findings suggest that the gut microbiota may contribute not only to local intestinal physiology but also to the pathogenesis and treatment response of lung diseases, including lung cancer. Indeed, accumulating evidence indicates that modulation of the gut microbiota can influence immune responses and therapeutic efficacy in lung cancer. In the case of NSCLC patients lacking mutations, incorporation of immune checkpoint inhibitors (ICIs) into the treatment strategy has improved survival in some cases10. Thus, the gut microbiota is a useful biomarker for NSCLC without genetic mutations11–13; however, no study has examined its potential as a biomarker in patients with EGFR mutations.
Here, we asked whether the gut microbiota is a potential biomarker for NSCLC with EGFR mutations. To do this, we evaluated the diversity and trends of the gut microbiota before treatment to determine its potential as a biomarker, as well as its association with adverse effects and responses to treatment.

Materials and methods

Study design
This prospective observational study was conducted at the Department of Respiratory Medicine, Hirosaki University Hospital, and approved by the Ethics Committee of Hirosaki University Graduate School of Medicine (approval number: 2020-085-1). All patients provided written informed consent before entry into the study. The clinical grade of diarrhea was evaluated in accordance with the common toxicity criteria for adverse events (CTCAE), version 5.0. All methods were performed in accordance with the relevant guidelines and regulations.

Eligibility criteria
The inclusion criterion was advanced NSCLC scheduled for induction of the first-line treatment with EGFR-TKIs. Patients with concomitant intestinal diseases such as autoimmune or inflammatory bowel disease, infectious diseases requiring systemic treatment, psychosis or psychiatric symptoms that would make participation in the study difficult, poorly controlled diabetes mellitus (HbA1c ≥ 10%), pregnancy or lactation, or active symptomatic interstitial pneumonia or pulmonary fibrosis were excluded.

Sample collection
Fecal samples used for intestinal microflora analysis were collected using a stool collection kit (CYK20201025, Cykinso Inc., Tokyo, Japan). The storage container provided with the kit contained a preservation solution (Guanidine_solution 1), and the collected stool samples were placed in the container so that they were completely soaked in the preservation solution. Subjects were asked to collect stool specimens on the same day of their visit and bring them to the clinic. If it was difficult to collect the specimens on the same day, the subjects were asked to store specimens collected a few days before the visit in a cool and dark place. Specimens were stored at 4 °C until DNA extraction. Stool samples collected within a few days prior to the first day of EGFR-TKI treatment was used for analysis.

DNA extraction and gene amplification
DNA was extracted from fecal samples using a DNeasy PowerSoil Pro Kit (Qiagen, Hilden, Germany) and stored at -20 °C. Each library was prepared according to the Illumina 16S Metagenomic Sequencing Library Preparation Guide using primer set 27Fmod/338R targeting the V1–V2 region of 16S rRNA genes. Next, 251-bp paired end sequencing of the amplicons was performed on a MiSeq system (Illumina, San Diego, CA) using a MiSeq Reagent v2 500 cycle kit.

Bioinformatics analysis
The paired end sequences obtained were merged, filtered, and denoised using DADA2. Taxonomic assignment was performed using the QIIME2 feature-classifier plugin and the Greengenes 13_8 database. The QIIME2 pipeline, version 2020.2, was used as the bioinformatics environment for processing of all relevant raw sequencing data.

Diversity analysis
Analysis of α- and β-diversity was performed for each individual sample. The Shannon index of α-diversity is a common metric that measures the α-diversity of species in a community mathematically. The Shannon index measures both species richness and evenness. β-diversity compares the evenness of the population of each species. Principal coordinate analysis (PCoA) assessment of β-diversity was used to identify differences between two groups (i.e., patients with a partial response (PR) vs. those with stable disease (SD)/progressive disease (PD)). β-diversity analysis was performed for weighted and unweighted UniFrac distances. Analysis of similarities (ANOSIM) is used widely as a multivariate method in microbiome studies. ANOSIM was conducted to observe the degree of separation between sample groups (PR vs. SD/PD). The null hypothesis that the average rank similarity between samples within a group is the same as the average rank similarity between samples belonging to different groups was tested by ANOSIM.

Linear discriminant analysis effect size (LEfSe) analysis
LEfSe determines the bacterial flora classification unit most likely to explain differences among groups. The best overall response, evaluated according to the response evaluation criteria in solid tumors (RECIST ver. 1.1), was determined for the two groups (PR vs. SD/PD), and LEfSe analysis was used to search for strains that differed significantly between the groups.

Statistical analysis
The statistical analysis was performed using JMP Pro 17.0 (SAS Institute Inc., Cary, NC). The difference in the Shannon index and relative abundance of Ruminococcus between the groups was evaluated using the Mann-Whitney U tests. P value < 0.05 were considered statistically significant in univariate analysis. The predictive cut-off value was obtained by constructing a receiver operating characteristic (ROC) curve. Progression free survival and overall survival (OS) were evaluated using log rank tests, in which a P value < 0.05 was considered significant.

Results

Patient characteristics
From July 2020 to July 2023, 21 treatment-naïve patients with EGFR-mutant NSCLC were judged eligible and enrolled in the study. The characteristics of the eligible patients are shown in the Table 1.

All patients included were Japanese and had adenocarcinoma histology. There were seven male and 14 female patients, with a median age of 70 years (range, 43–85 years). In total, 11 patients (52.4%) had never smoked. Most patients had a good Eastern Cooperative Oncology Group performance status (ECOG-PS) 0–1 (95.2%), and only one patient had ECOG-PS 3 (4.8%). Nine patients had the EGFR L858R mutation, and nine had deletions of exon 19 (19del). There were two uncommon mutations (L861Q and G719X), and one compound mutation (L858R + E709G). Seven patients were treated with osimertinib, 13 with afatinib, and one with elrotinib plus ramucirumab (RAM) (Fig. 1). No patient experienced a complete response. A total of 11 (52.4%) patients had a PR, nine (42.8%) had SD, and one (4.8%) had PD. Seven patients had diarrhea of grade 2 or higher, and 14 patients had no diarrhea or diarrhea of grade 1. None of the patients had been taking antibiotics prior to specimen collection.

Analysis of α-diversity of the intestinal gut microbiota
First, we evaluated the patients’ gut microbiota by assessing α-diversity indices. This was done using stool samples collected before introduction of EGFR-TKIs. The Shannon index for α-diversity was calculated to describe the diversity of each species. The data showed that the Shannon index was higher in patients with no diarrhea or grade 1 (G1) diarrhea than those with grade 2 (G2) or higher diarrhea (Shannon, p = 0.0367; Fig. 2), suggesting that a high diversity within the gut microbiota might be associated with no or mild diarrhea.

Association between the microbiota and EGFR-TKI efficacy
Next, we analyzed the β-diversity of the microbiota. β-diversity analysis was performed to compare “high efficacy” patients (the PR group) and “low efficacy” patients (the SD/PD group). ANOSIM revealed no significant differences in unweighted UniFrac distances between the PR and the SD/PD groups (PCoA, unweighted; p = 0.118; Fig. 3A). By contrast, the weighted UniFrac distances differed significantly between the two groups (weighted; p = 0.041; Fig. 3B). These results indicate that the composition of the gut microbiota of patients who respond well to EGFR-TKIs is different from that of those who do not.

LEfSe analysis conducted to investigate the strains responsible for this difference identified g_Ruminococcus.s_ (Ruminococcus) in patients who achieved a PR. A list of strains with a log score of 3 or higher was plotted according to abundance of Ruminococcus, which was the focus of our study (Fig. 4). Univariate analysis revealed significant differences in the relative abundance of Ruminococcus between the PR and the SD/PD groups (p = 0.0018, Fig. 5). These results suggest that EGFR-TKIs may be more effective in those with a higher abundance of Ruminococcus in the gut microbiota.

Furthermore, we determined the percentage of the total microbiota should be made up of Ruminococcus to expect a PR (or better) effect. The predicted cut-off value for Ruminococcus was obtained by drawing a ROC curve with a cut-off value of 1.4% (Area under curve (AUC) 0.96, p < 0.001; the dotted line in Fig. 5). The results suggested that EGFR-TKIs treatment may be more effective when Ruminosoccus makes up more than 1.4% of the total flora in stool samples before EGFR-TKI treatment.

Discussion
Here, we evaluated the gut microbiota profile, and the α- and β-diversity, in NSCLC patients with EGFR mutations. The data show that patients with higher α-diversity may have less severe diarrhea during/after EGFR-TKI treatment. The high abundance of Ruminococcus in those with good response suggests that this genus may be a potential biomarker for predicting favorable efficacy of EGFR-TKIs. Few reports have examined the association between EGFR-mutant NSCLC and the gut microbiota, and our study is the first to examine the gut microbiota of Japanese patients with advanced NSCLC carrying EGFR mutations.
Our findings are also related to the concept of the lung–gut axis, which refers to the bidirectional interaction between the intestinal and pulmonary microbiota7–9. In this context, alterations in the gut microbiota may affect the tumor microenvironment, immune status, and therapeutic response in NSCLC. The observation that higher α-diversity was associated with less severe diarrhea and that Ruminococcus abundance correlated with better EGFR-TKI response suggests that intestinal microbial composition may modulate both gastrointestinal toxicity and systemic treatment efficacy through this axis.
Consistent with this concept, it is important to consider the overall taxonomic composition of the gut microbiota in EGFR-mutant NSCLC. In general, the gut microbiota of EGFR-mutant NSCLC patients is occupied by members of four bacterial phyla: Bacteroidetes, Firmicutes, Actinobacteria, and Proteobacteria7. Another study comparing the gut microbiota of lung cancer patients with that of healthy controls reported that Actinobacteria (Lung cancer (LC): 3%, Healthy control (HC): 1.56%) and Fusobacteria (LC: 1.14%, HC: 0.22%) were predominant in lung cancer patients14. Fusobacteria are linked to malignancy and are associated with unfavorable prognosis and treatment failure15. Herein, we detected Fusobacteria in three patients in the SD/PD group (0.69–5.32% of the total), all of whom had a poor response. Although this number of patients is small, we think that Fusobacteria may be associated with a poor prognosis for NSCLC patients with EGFR mutations.
The gut microbiota is affected not only by drugs but also by other factors such as place of residence and diet, and the composition of the gut microbiota varies among ethnic groups16. Shoji and colleagues evaluated the gut microbiota in three groups of Japanese individuals: patients with lung cancer, patients with lung inflammation, and healthy controls17. Each group was represented by 20 phyla, with the main phyla being Firmicutes, Actinobacteria, and Bacteroidota. There were no significant differences between the three groups in the abundance of these phyla. Similarly, we found that Firmicutes, Actinobacteria and Bacteroidota accounted for more than 90% of the gut microbiota, a trend similar to that in previous reports. While the overall trends in the gut microbiota are consistent, Akkermansia muciniphila and Bifidobacterium breve, which are associated with effective immunotherapy in patients with NSCLC without mutations18,19, were not identified in NSCLC patients with EGFR mutations. We speculate that this may be one reason why immunotherapy is less effective in these patients, although further investigation is necessary.
Our data demonstrated that α-diversity was higher in the group with no or less severe diarrhea. A previous report also showed a trend toward higher α-diversity in Japanese healthy individuals who did not have diarrhea20. Taken together, the data suggest that the α-diversity of the gut microbiota is associated with no/less severe diarrhea, regardless of whether the subject has cancer. Also, our β-diversity and LEfSe analyses identified Ruminococcus in patients who were responsive to EGFR-TKI treatment. Previous studies show that ICI have better anti-tumor effects in patients with abundant Ruminococcus21, and that high levels of Ruminococcus correlate with durable clinical benefit22. It seems therefore that Ruminococcus abundance is associated with favorable outcomes after immunotherapy, making it a potential biomarker. In this study, we found a similar trend in EGFR-mutant NSCLC patients, suggesting that Ruminococcus abundance may also be a biomarker that can be used to estimate the efficacy of EGFR-TKI therapy.
This study has several limitations that should be considered. First, our 16S V1–V2 sequencing approach did not provide accurate species-level resolution (e.g., R. gnavus vs. R. bromii). Previous studies suggest that members of the Ruminococcus genus play important functional roles in the gut, including participating in short-chain fatty acid production, bile acid metabolism, and maintenance of mucosal integrity23–25. These functions may influence drug metabolism, mucosal protection, and ultimately the efficacy of EGFR-TKI therapy. Further studies using metagenomic shotgun sequencing are needed to clarify species-level contributions and their mechanistic impact.
In addition, use of the V1–V2 region may limit the comparability of our data with other studies that employed different target regions. This study did not include a control group. Instead, we referred to the report by Shoji et al., which described characteristic oral and gut microbiota alterations in Japanese patients with lung cancer, as contextual information for the control cohort. Although their analysis targeted the V3–V4 region of the 16S rRNA gene, whereas our study targeted the V1–V2 region, making direct and accurate comparisons difficult, their findings provide useful context for interpreting our results. Inclusion of a control cohort, such as patients with wild-type EGFR, would allow us to better identify trends specific to EGFR-mutant NSCLC, and this is an important future research direction.
Second, with respect to diarrhea severity, the observed difference in the Shannon index between the G0–G1 and G2–G3 groups was borderline significant (p = 0.0367). This was a two-group comparison and thus multiple testing correction was not applied; however, this borderline statistical significance is a limitation of the present study. Further validation in larger cohorts and complementary analyses, including β-diversity comparisons, are necessary to confirm this association between α-diversity and diarrhea severity.
Third, different types of EGFR-TKIs (osimertinib, afatinib, and erlotinib + RAM) were administered in this cohort, and the type of EGFR-TKI could potentially influence the gut microbiota composition. However, the total number of samples was limited (n = 21) and therefore further subgroup analyses according to EGFR-TKI type were not statistically feasible. We acknowledge this is an important limitation and plan to increase the sample size in future studies to enable stratified analyses according to EGFR-TKI type. Finally, we were unable to obtain OS data for all patients because the study period was short (3 years).
In conclusion, we report novel findings regarding the gut microbiota of EGFR-mutant NSCLC patients. Higher α-diversity before the start of treatment may predict no/less severe diarrhea as a side effect. In addition, Ruminococcus is a potential biomarker for predicting a better response. Because stool examination is noninvasive, it is beneficial for patients with EGFR-mutant NSCLC.