Dietary pattern and the corresponding gut microbiome in response to immunotherapy in Thai patients with advanced non-small cell lung cancer (NSCLC)

Abstract

Gut microbiota is considered a key player modulating the response to immune checkpoint inhibitors (ICI) in cancer. The effects of dietary pattern on this interaction is not well-studied. A prospective multicenter cohort of 95 patients with advanced non-small cell lung cancer (NSCLC) undergoing ICI therapy were enrolled. Stool shotgun metagenomic sequencing was performed. Three-day dietary patterns before ICI were assessed. Patients were categorized as hyperprogressive disease (HPD) if they exhibited a time to treatment failure of less than 2 months. All others were categorized as non-hyperprogressive disease (non-HPD). The correlation between dietary patterns, gut microbiome, and response to ICI therapy was analyzed. In the multivariate analysis, a high abundance of Firmicutes unclassified and the Ruminococcaceae family correlated with a significantly diminished progression-free survival (PFS) with an HR of 2.40 [P = 0.006] and 4.30 [P = 0.005], respectively. More specifically, within the subset of NSCLC patients treated solely with ICI therapy, a high abundance of Intestinimonas and the Enterobacteriaceae family were associated with substantially reduced PFS with an HR of 2.61 [P = 0.02] and HR 3.34 [P = 0.005], respectively. In our comprehensive dietary pattern analysis, the HPD group showed increased consumption of cholesterol, sodium, and fats beyond recommended levels compared to the non-HPD group. This group also displayed a tendency towards higher food pattern scores characterized by a high intake of fat and dairy products. Our study revealed a distinct association between the gut microbiome composition and treatment outcomes. The overall composition of diet might be related to ICI therapeutic outcomes.

Introduction

The treatment paradigm of lung cancer has dramatically changed in recent years with the introduction of immune checkpoint inhibitors (ICI). ICI that target programmed cell death 1 (PD-1) or its ligand 1 (PD-L1) have shown promising efficacy for the treatment of advanced non-small cell lung cancer (NSCLC), with better outcomes and less toxicity when compared to traditional chemotherapy^1–7. Although durable responses have been noted, only a limited proportion of patients benefit from ICI⁸. Most patients with NSCLC develop resistance or occasional rapid acceleration of the disease, called hyper-progression⁹. Therefore, predictive biomarkers are required to optimize patient selection for treatment. Available predictive biomarkers have been recently explored, such as tumor mutation burden, PD-L1 expression, and microsatellite instability that are generally focused on the cancer^10–12. However, current evidence has shown that these tumor-based biomarkers are not sufficient to identify the patients who will best benefit from this treatment.

Recently, the role of the gut microbiome in the modulation of immune functions has been increasingly studied. Unfavorable changes in gut microbial composition and function have recently been linked to alterations in the balance of host immunity and the effectiveness of ICI. Routy B. et al. analyzed stool samples from patients receiving PD-1 inhibitors for treatment across various cancer types and found that the intestinal abundance of Akkermansia muciniphila can stimulate the production of interleukin-12 (IL-12) and enhance the recruitment of tumor-specific cytotoxic T lymphocytes (CTLs) that affect the clinical efficacy of anti-PD-1 treatment¹³. In contrast, a study from the University of Texas MD Anderson Cancer Center examined the gut microbiome of melanoma patients undergoing anti-PD-1 immunotherapy and observed that Bacteroides can promote the maturation of regulatory T (Treg) cells and the secretion of anti-inflammatory cytokine IL-10, resulting in suppression of the antitumor response¹⁴.

It is now understood that diet plays an important role in shaping the microbiome. Animal experiments demonstrated that foods containing high levels of saturated fats, sugar, and low levels of fiber can induce shifts in the microbiome with secondary effects on host immunologic response. Given this association, there may be significant therapeutic utility in altering microbial composition through diet. However, there is still a limited understanding of how dietary patterns impact the gut microbiome and the clinical efficacy of ICI.

We aimed to investigate the association between dietary patterns, gut microbiome, and clinical outcomes in patients with advanced NSCLC who were treated with ICI. Part of this study was presented at the 2023 American Society of Clinical Oncology (ASCO) annual meeting [10.1200/JCO.2024.42.23_suppl.204].

Materials and methods

Study population

A prospective multicenter cohort of 95 patients with histologically confirmed recurrent or metastatic NSCLC who received ICI therapy were enrolled over a period of three years from August 1, 2019, to May 31, 2022. Enrollment locations included two academic cancer centers in Thailand, King Chulalongkorn Memorial Hospital and Siriraj Hospital. Demographic data including age, gender, Eastern Cooperative Oncology Group (ECOG) performance status (PS), histological type, smoking status, and nutritional status were recorded. The tumor response and follow-up were assessed every two to three months as the standard protocol of our institutions. Biomarker testing was conducted as standard practice, including EGFR, ALK, and PD-L1. Blood and stool samples were collected before starting ICI therapy. Objective response rate (ORR) and progression-free survival (PFS) were determined according to the Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST v1.1) and classified per 2-tier classification including hyperprogressive disease (HPD) and non-hyperprogressive disease (non-HPD). Despite several definitions used worldwide¹⁵, the authors defined HPD as patients who had radiographic evidence of progressed disease (PD) by RECIST 1.1 at first evaluation or time-to-treatment failure of < 2 months. Non-HPD was defined as those with complete or partial response (CR or PR) or stable disease (SD) by RECIST 1.1 at first evaluation and time-to-treatment failure of more than 2 months. Non-HPD was further sub-classified into durable response (DR), which was defined as a patient who had disease control for more than 12 months, and intermediate response (IR), defined as those not meeting DR criteria. All patients provided written informed consent. I confirm that all experiments were performed in accordance with the Declaration of Helsinki. This study was approved by the Institutional Review Board of the Faculty of Medicine at Chulalongkorn University (IRB No. 385/63) and Mahidol University (IRB No. 374/2564).

Fecal sample collection and process

All participants received oral and written instructions regarding the stool collection procedure. Fresh fecal samples of approximately 1 gram were collected in DNA/RNA Shield™ Fecal Collection Tubes to ensure sample stability. Samples were transported at ambient temperatures and stored at − 20 °C immediately prior to further process. Bacterial genomic DNA was extracted using QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany). After checking the integrity and concentration of DNA, the sequencing libraries were constructed using Nextera^® DNA Flex Library Prep Kit (Illumina, San Diego, CA) according to the manufacturer’s protocol and were sequenced on the platform NovaSeq^® (Illumina, San Diego, CA). The samples were processed and analyzed with the Shotgun Metagenomic Sequencing Service (Zymo Research, Irvine, CA). A total of 57 individual data sequences were generated. Overall, 93.6% of the raw reads were regarded as high-quality reads and retained for further analysis.

Bioinformatics analysis

Raw sequence reads were trimmed to remove low-quality fractions and adapters with Trimmomatic-0.33¹⁶. Quality trimming by sliding window with 6 bps window size and quality cutoff of 20 and reads with size lower than 70 bp were removed. Antimicrobial resistance and virulence factor gene identification was performed with the DIAMOND sequence aligner¹⁷. Microbial composition was profiled using MetaPhlAn4¹⁸. MetaphlAn4 is a current tool for profiling the composition of microbial communities (Bacteria, Archaea, and Eukaryotes) from metagenomic shotgun sequencing. MetaPhlAn4 relies on 5.1 M unique clade-specific marker genes identified from ~ 1 M microbial genomes (including reference genomes and metagenome-assembled genomes) spanning 26,970 species-level genome bins (SGBs). However, 4992 of them are taxonomically unidentified at the species level. This workflow filters out low-abundant (< 0.01%). The threshold of standard depth detection was 10Gbases and 20% read mapping of SGB fraction. Relative abundance was quantified based on the within-sample-normalized average coverage estimation¹⁸. A non-parametric Wilcoxon rank-sum test was employed to analyze the statistical significance of diversity indices and taxa between HPD and non-HPD subjects. Bray-Curtis metrics were applied to calculate pairwise dissimilarities between samples and were used in subsequent beta diversity evaluation by principal coordinate analysis (PCoA). The LEfSe algorithm was applied to identify the phylotypes significantly different in relative abundance between all HPD and non-HPD samples¹⁹. Phylotypes with a linear discriminant analysis (LDA) score cut-off of > 2.0 and p-value < 0.05 in the built-in rank sum test were regarded as statistically significant.

Dietary assessment

Dietary intake was assessed using a 3-day food record (two weekdays and one weekend day) that was collected concordantly with the fecal sample. An information session was delivered by a trained dietitian to train each study participant on how to record their daily food intake and how to use a household scale. Written instructions were incorporated in the food diaries. Information on the type (including brand names) and amounts of food consumed were collected. For the verification and estimation of the size of individual food portions, the participants were instructed with the help of three-dimensional food models and household utensils to enhance the accuracy of portion size estimation. The importance of maintaining regular diets and recording all food and drink consumed was emphasized. The quality control of all food diaries was handled and reviewed by an experienced observer to avoid inconsistency and to maintain accurate data entries. After completion, the diaries were processed to record food quantities by an experienced dietitian using a standard manual on food portions and household measures. The food codes were those used in the Thai Dietary Database (fourth edition), Institute of Nutrition, Mahidol University. The calculation of nutrients was done using INMUCAL-Nutrient Software Version 4. Incomplete food records were excluded from the analysis.

Dietary pattern analysis

Food items were systematically categorized into 22 distinct food groups, with their daily consumption measured in grams. To evaluate the quality of participant diets, we employed the Diet Quality Index-International (DQI-I), which encompasses four key dimensions of a high-quality and healthy diet: variety, adequacy, moderation, and overall balance²⁰. The total DQI-I scores, ranging from 0 to 100, served as an indicator of diet quality, with higher scores indicating better adherence to a healthy dietary pattern. For subsequent analyses, the continuous total DQI-I scores were divided into tertiles, allowing for comparisons of diet quality between higher and lower tertiles. Factor analysis through minimal residual analysis with the R package “psych”²¹ was used to distinguish the dietary patterns. The selection of the optimal number of factors was determined by examining the scree plot by parallel analysis, comparing the true dataset with randomly generated “parallel” matrices, considering factor interpretability. Ultimately, a five-factor solution was chosen, collectively explaining 53% of the total variance in food intake frequencies. Subsequently, alternative rotations and transformations using varimax were applied to achieve a mutually independent structure that would facilitate interpretation. Factor score coefficients were estimated by regression approach and saved for individual values of food pattern. This clustering approach allowed for the identification of distinct dietary patterns within the population under study.

Statistical analysis

Categorical variables were summarized by frequencies and percentages, while continuous variables were reported by the median and interquartile range (IQR). Variables include age, gender, ECOG performance status, histology, smoking status, histology, presence of liver or brain metastasis, number of metastatic sites, treatment type (single or combination), and treatment outcome were analyzed between different subgroups using Chi-square or Fisher’s exact test for categorical data, while the unpaired t-test and Mann-Whitney test were used for continuous data. Progression-free survival (PFS) was calculated from the first day of ICI therapy to disease progression or death. Overall survival (OS) was calculated from the date of diagnosis of recurrent or metastatic disease until the date of death or last follow-up. Patients who did not develop the event at the end of the study were censored at the date of the last observation which was defined on April 30, 2023. Survival analysis was performed by Kaplan–Meier survival analysis and log-rank test. Multivariate analysis was performed by binary logistic regression or Cox proportional hazards regression model, as appropriate. The level of statistical significance was selected as a p-value less than 0.05. All statistical analyses were conducted using SPSS 29.0 (SPSS Inc, Chicago, Illinois, USA), R-package version 3.6.3. or GraphPad Prism 9.4 (GraphPad Software, San Diego, CA, USA).

Results

Patient baseline characteristics

Between August 1, 2019, and May 31, 2022, 95 patients with histologically confirmed recurrent or metastatic NSCLC who were treated with ICI alone or in combination with chemotherapy were enrolled. Of these, 57 patients who obtained stool collection and 34 patients who completed the validated 3-day food record were included in the final analyses. Patient baseline characteristics are summarized in Table 1. The median age of the patients was 69 years (interquartile range [IQR] 60 to 73.5 years). Approximately 80% of the patients were men. Most of the patients were ever-smokers (76%) and had an ECOG performance status of 0–1 (93%). The majority had adenocarcinoma (88%), presented with metastatic disease (82%), and had 0–2 metastatic sites (68%). Baseline brain and liver metastases were present in 18% and 9% of the overall population, respectively. Oncogenic alterations were found in 9 patients (9.4%), including one with EGFR mutations (Del19), two with ROS1 rearrangement, four with KRAS mutation, one with MET exon 14 mutation, and one with BRAF mutation. Among patients who had EGFR and ROS1 alteration, molecular testing was not known before ICI therapy either due to inadequate tissue or negative ctDNA results. PD-L1 testing was conducted in 37 patients (39%), composed of PD-L1 tumor proportion score (TPS) of ≥ 1% for 46%.

Table 1.

Baseline characteristics in the overall population and according to response to ICI therapy status.

Characteristics	All (N = 57)	HPD (N = 22)	Non-HPD (N = 35)	P-values
Age at diagnosis, n (%)				0.05
Median (IQR)	69 (60-73.5)	68.5 (56-73.8)	69 (61–74)
< 60 years	13 (22.8)	8 (36.4)	5 (14.3)
≥ 60 years	44 (77.2)	14 (63.6)	30 (85.7)
Gender, n (%)				1.00
Male	46 (80.7)	18 (81.8)	28 (80)
Female	11 (19.3)	4 (18.2)	7 (20)
ECOG PS, n (%)				0.64
0–1	53 (93)	20 (90.9)	33 (94.3)
≥ 2	4 (7)	2 (9.1)	2 (5.7)
Smoking status, n (%)				0.76
Never	14 (24.6)	6 (27.3)	8 (22.9)
Current/former	43 (75.4)	16 (72.7)	27 (77.1)
Histology, n (%)				1.00
Non-squamous cell carcinoma	50 (87.7)	19 (86.4)	31 (88.6)
Squamous cell carcinoma	7 (12.3)	3 (13.6)	4 (11.4)
PD-L1 status, n (%)				0.08
< 1%	20 (35.1)	12 (54.5)	8 (22.9)
1–49%	9 (15.8)	2 (9.1)	7 (20)
≥ 50%	8 (14.0)	2 (9.1)	6 (17.1)
Unknown	20 (35.1)	6 (27.3)	14 (40)
Number of metastatic sites, n (%)				0.72
< 3 sites	48 (84.2)	18 (81.8)	30 (85.7)
≥ 3 sites	9 (15.8)	4 (18.2)	5 (14.3)
Brain metastasis, n (%)	10 (17.5)	3 (13.6)	7 (20)	0.73
Liver metastasis, n (%)	5 (8.8)	3 (13.6)	2 (5.7)	0.36
Line of ICI, n (%)				< 0.001*
First line	15 (26.3)	0	15 (42.9)
Later line	42 (73.7)	22 (100)	20 (57.1)
Type of ICI, n (%)				0.04*
Monotherapy	39 (68.4)	19 (86.4)	20 (57.1)
ICI and chemotherapy	18 (31.6)	3 (13.6)	15 (42.9)
Received antibiotics 30 days before ICI initiation, n (%)				0.39
Yes	6 (10.5)	1 (4.5)	5 (14.3)
No	51 (89.5)	21 (95.5)	30 (85.7)

ECOG PS Eastern Cooperative Oncology Group Performance Status, HPD Hyper Progressive Disease, ICI Immune Checkpoint Inhibitor, IQR Interquartile Range, PD-L1 Programmed-cell Death Ligand 1.

A total of 15 patients (26%) received ICI therapy as first-line treatment, while the remaining 42 patients (74%) were treated in subsequent lines (2nd or 3rd line). ICI therapy was administered as monotherapy in 39 patients (68.4%) and in combination with chemotherapy in 18 patients (31.6%). The objective response rate (ORR) of ICI therapy in our cohort was 31.6%, with 18 patients showing partial response (PR), 18 patients with stable disease (SD), and 21 patients with progressive disease (PD). The median duration of treatment was 4.6 months (IQR 1.5 to 10.4 months), and the median PFS for all lines of ICI therapy was 5.2 months (95% confidence interval [CI], 2.6 to 7.7 months). As of the data cut-off on April 30, 2023, 5 out of 57 patients (8.8%) were still undergoing treatment. Overall, 2 patients completed the 24-month treatment as a current recommendation. Median ICI treatment in this cohort was 6 cycles (IQR 3 to 14 cycles).

Responses to treatment were assessed using radiographic imaging in those with evaluable treatment responses (n = 57) according to RECIST 1.1 criteria. We further classified patients as either having HPD (n = 22) or non-HPD (n = 35). The baseline demographic and disease characteristics were generally well-balanced between the groups, except the HPD group was significantly more likely to have received ICI therapy as monotherapy (86.4% vs. 57.1%, p-value = 0.04) and in later lines of treatment compared to the non-HPD group (100% vs. 57.1%, p-value < 0.001) (Table 1).

Taxonomic profiles of the gut microbiota

The composition of the gut microbiota in NSCLC patients was analyzed using shotgun metagenomic sequencing before the first administration of ICI therapy. Classification and analysis were conducted at the phylum, class, order, family, genus, and species levels. The most abundant microbiota compositions at the phylum level were Bacteriodetes (49%), Firmicutes (36.6%) and Proteobacteria (8.9%). Microbiota classification at the phylum level according to treatment response evaluation are presented in Fig. 1B.

Fig. 1.

Diversity of gut microbiome in response to ICI therapy. Comparison of alpha diversity of the gut microbiome was conducted using richness (observed indices) and diversity (Shannon indices) by the Mann-Whitney U rank sum test. (A). Taxonomic profiles at phylum level of the gut microbiome of the study cohort according to response of treatment; HPD (hyper-progressive disease) and non-HPD (non-hyper-progressive disease) (B). Principal coordinate analysis of gut microbiome by response using Bray-Curtis metrics pairwise dissimilarities. Red and blue dots represented HPD and non-HPD group, respectively. The gray sphere represented the coefficient of significant taxa of our cohort (C). Diverse eigenvector plot of top 5 most abundance at species level according to response status; HPD vs. non-HPD (D and E).

Differences in gut microbial composition and responses to the ICIs therapy

The alpha diversity or within-sample diversity of the gut microbiomes between the two groups was significantly higher in the HPD group compared to the non-HPD group, as indicated by the Shannon indices (p-value < 0.01; Fig. 1A). The authors analyzed beta diversity, between sample diversity, using principal coordinate analysis on paired-wise distances calculated by the Bray-Curtis dissimilarity at the strain level (Fig. 1C). We observed a diverse eigenvector plot of the top 5 most abundant microbiomes at species level per response status (HPD vs. non-HPD) (Fig. 1D and E).

Because compositional differences in the gut microbiome may influence the response to ICI therapy, the enrichment of abundance for each taxon in HPD versus non-HPD was performed. Bacteroidetes and Firmicutes were the most abundant in both the HPD (45% and 40.4% respectively) and non-HPD groups (52.8% and 34.2% respectively). At the species level, Phocaeicola vulgatus was the most common species in the HPD group, whereas Prevotella copri clade A was dominant in the non-HPD group. To further explore these findings, we performed high-dimensional class comparisons using linear discriminant analysis of effect size (LEfSe)¹⁹. Using LDA score cut-off 2.0 and significant p-value < 0.05, this analysis revealed differentially abundant bacteria in the gut microbiome of the HPD and non-HPD groups. Specifically, the Clostridia unclassified genus, Ruminococcaceae family (GGB9699), Lachnospiraceae unclassified, Ruthenibacterium genus, Alistipes genus, Firmicutes unclassified genus, Candidatus Cibiobacter, and Alphaproteobacteria family (GGB6612) were enriched in the HPD group, while the Pasteurellales order, Pasteurellaceae family, Anaerostipes, Veillonella, Mediterraneibacter, and Haemophilus genera were enriched in the non-HPD group (Fig. 2A and B). We then performed pairwise comparisons of species abundances based on the response status. Shotgun metagenomic sequencing identified the enrichment of the Faecalibacillus and Enterobacter genera as well as the Coprococcus catus, Ruminococcaceae family (GGB9730_SGB15291), and Enterobacter hormaechei in the HPD group. In comparison, the Desulfovibrio genus, among others, were enriched in the non-HPD group (Fig. 2C, Table S1).

Fig. 2.

Taxonomic cladogram from LEfSe showing differences in gut microbiome taxa. The circle radiating inside-out demonstrated the classification from the phylum to the genus. Dot size is proportional to the abundance of the taxon. Red and blue dots denoted the core bacterial populations in each respective group (A). Linear discriminant analysis (LDA) scores computed for differentially abundant taxa in the gut microbiomes of HPD (red) and non-HPD (blue). Length indicates the effect size associated with a taxon (B). Pairwise comparisons by Mann-Whitney U rank sum test of abundances of metagenomic species identified by metagenomic WGS sequencing for HPD (red) and non-HPD (blue) (C). *p-value < 0.05; ** p-value < 0.01; *** p-value < 0.005.

Next, our investigation focused on elucidating the precise composition of the gut microbiome in the non-HPD group of patients. To achieve this, we further classified the non-HPD group based on the durability of their treatment response. Specifically, we defined the durable response (DR) group as comprising patients who demonstrated a PFS greater than 12 months¹⁵. The intermediate response (IR) group encompassed patients who had duration of disease control of more than 2 months but less than 12 months. Patient baseline characteristics are summarized in Table S2. The taxonomic profiles of the gut microbiota composition at the phylum and species levels are presented in Figure S1, highlighting the distinctions among the three groups. Regarding the alpha diversity of the gut microbiome, as measured by the Shannon indices, a statistically significant difference was observed between the HPD group and the IR group (P = 0.002) (Figure S2 A-B). Furthermore, the beta diversity of the gut microbiome was assessed by principal coordinate analysis based on the response status of these patients shown in Figure S2C. Compositional differences in the gut microbiome were assessed through the utilization of LEfSe analysis and pairwise comparisons of metagenomic species abundances based on the response status of the patients between the three groups. This comprehensive analysis unveiled additional differentially abundant bacteria within the gut microbiome of the IR and DR groups. Specifically, the Methanomassiliicoccales order, Odoribacteraceae family, Culturomica, Odoribacter genera, and Tannerellaceae family (GGB1407) were found to be enriched in the DR group, while the Pasteurellales order, Pasteurellaceae family, and Klebsiella genera were enriched in the IR group (Figure S3 A-D and Table S3). After excluding the HPD group, patients with a higher abundance of the gut microbiome Klebsiella genus experienced a significantly shortened PFS compared to those with a lower abundance (P = 0.02).

Univariate and multivariate analysis of specific gut microbiome abundance and response to ICI therapy

To explore how specific bacterial taxa affect patient treatment response, progression-free survival (PFS) of ICI therapy was applied to the “top hits” gut microbiome that were consistently observed across our analyses. We focused on the Firmicutes unclassified, Alphaproteobacteria family (GGB6612), and Ruminococcaceae family (GGB9730_SGB15291) and stratified patients into high versus low categories based on the median relative abundance of these taxa in the gut microbiome. Patients with a high abundance of Firmicutes unclassified genus, Alphaproteobacteria family. (GGB6612), and Ruminococcaceae family (GGB9730_SGB15291) had a significantly shortened PFS compared to those with a low abundance (Fig. 3A-C) with univariate hazard (HR) ratios of 2.04 [95% CI 1.13–3.69, P = 0.02], 3.92 [95% CI 1.16–13.24, P = 0.03], and 4.72 [95% CI 2.06–10.81, P < 0.0001], respectively (Table 2). Consistent with previous published findings²², there was no significant difference in PFS among patients who had received antibiotics 30 days before the initiation of ICI therapy. The multivariate Cox proportional hazards regression model showed that high abundance of Firmicutes unclassified genus and Ruminococcaceae family (GGB9730_SGB15291) remained independent factors for shorten PFS with an HR of 2.40 [95% CI 1.28–4.50, P = 0.006] and 4.30 [95% CI 1.56–11.83, P = 0.005], respectively (Table 2).

Fig. 3.

Progression-free survival of overall cohort according to abundance of top hits gut microbiomes is shown. Using median relative abundance as the cut-off level, Kaplan-Meier analysis of high abundance (black) or low abundance (blue) of Firmicutes unclassified genus (P = 0.02) (A), Alphaproteobacteria family (GGB6612) (P = 0.02) (B) and Ruminococcaceae family (GGB9730) (P < 0.001) (C) is shown. For subset of ICIs monotherapy (68.4%), we found differences in gut microbiome composition related to outcome between ICI monotherapy vs. overall cohort. Abundance of Intestinimonas genus (P = 0.04) (D), Ruthenibacterium genus (P = 0.01) (E), and Enterobacteriaceae family (P = 0.0005) (F) were significantly correlated with PFS.

Table 2.

Univariate and multivariate Cox regression analyses of PFS in the overall population.

Variablesa	Univariate		Multivariate
	HR (95% CI)	P-value	HR (95% CI)	P-value
Age (< 60/≥60)	1.43 (0.75–2.76)	0.28
Gender (male/female)	0.85 (0.42–1.70)	0.64
ECOG PS (≥ 2/0–1)	1.10 (0.34–3.56)	0.88
Smoking (ever/never)	0.74 (0.39–1.41)	0.36
Histology (SQ/non-SQ)	1.49 (0.63–3.56)	0.37
Number of metastatic sites (≥ 3/<3)	1.35 (0.66–2.80)	0.41
Brain metastasis (yes/no)	0.48 (0.21–1.07)	0.07	0.42 (0.18–1.01)	0.05
Liver metastasis (yes/no)	1.94 (0.75–5.01)	0.17
PD-L1 expression (negative/positive)	1.20 (0.60–2.38)	0.61
Line of ICI (later/first)	1.93 (0.99–3.72)	0.05	1.29 (0.63–2.61)	0.49
Type of ICI (single/combo)	1.59 (0.86–2.94)	0.14
Received antibiotics (yes/no)	0.87 (0.35–2.20)	0.77
Alpha diversity by Shannon (above/below median)	1.33 (0.76–2.34)	0.32
Abundance of Firmicutes unclassified (above/below median)	2.04 (1.13–3.69)	0.02*	2.40 (1.28–4.50)	0.006*
Abundance of Alphaproteobacteria family (GGB6612) (above/below median)	3.92 (1.16–13.24)	0.03*	1.27 (0.29–5.51)	0.72
Abundance of Ruminococcaceae family GGB9730_SGB15291 (above/below median)	4.72 (2.01–10.81)	< 0.0001*	4.30 (1.56–11.83)	0.005*

^aCategory after the slash (/) was set as a reference category.

ECOG PS Eastern Cooperative Oncology Group performance status, HPD hyper progressive disease, SQ Squamous cell carcinoma.

Differences in gut microbial composition particular subset of ICIs monotherapy

Given the observed differences in response among patients receiving ICI therapy as monotherapy (68.4%) or in combination with chemotherapy (31.6%), our investigation aimed to elucidate the precise composition of the gut microbiome solely in response to ICI therapy. The HPD group comprised of a higher proportion of patients who received ICI monotherapy (86.4%) compared to the non-HPD group (57.1%). We conducted an analysis of the gut microbiota composition in NSCLC patients treated with ICI monotherapy with a summary of their baseline characteristics provided in Table S5.

Through the assessment of compositional differences in the gut microbiome, we unveiled distinct bacterial abundance patterns among the three groups. Specifically, the Clostridia unclassified (GGB9347), Candidatus Cibiobacter, Ruthenibacterium, and Intestinimonas genera exhibited enrichment in the HPD group. In contrast, the Enterobacteriaceae family, Flavonifractor, Enterocloster, Tyzzerella and Lachnoclostridium genera were enriched in the IR group, and the Fusobacteriaceae unclassified, Alphaproteobacteria family (GGB6608), Bacteroidaceae family (GGB1154), and Paraprevotella genus were enriched in the DR group (Figure S4-S6 and Table S4). Notably, patients with a high abundance of Ruthenibacterium, Intestinimonas genera, and the Enterobacteriaceae family exhibited significantly shortened PFS compared to those with a low abundance (Fig. 3D-F). Furthermore, our multivariate Cox proportional hazards regression model revealed that the high abundance of Intestinimonas genus and Enterobacteriaceae family remained the independent factors of PFS with an HR = 2.61 [95% CI 1.16–5.87, P = 0.02] and HR = 3.34 [95% CI 1.43–7.80, P = 0.005], respectively (Table S6).

Dietary pattern and response to ICI Therapy

Given that dietary patterns play an important role in shaping the microbiome with secondary effects on host immunologic response, we next sought to assess the effect of dietary patterns on response to ICI therapy. Dietary intake was assessed using a 3-day food record (two weekdays and one weekend day) and analyzed for food composition using the Thai Dietary Database. Of 95 study participants, 33 patients completed the validated 3-day food record. Baseline characteristics are summarized in Table S7. Caloric intake was 1256 (± 377.5) kcal/day with a daily intake of 170 (± 54.4) grams/day carbohydrates, 54 (± 22.0) grams/day protein, and 40 (± 18.3) grams/day fat, representing 54%, 17%, and 28% of the total daily energy intake, respectively. Free sugar intake was 13% of total energy, much higher than the Thai Recommended Daily Intake (free sugar ≤ 5% of total energy). In contrast, the dietary fiber intake was 6 g per 1000 kcal, lower than the Thai Recommended Daily Intake (≥ 14 g per 1000 kcal). Only 11.8% and 2.9% of patients met the recommendations for free sugar, and dietary fiber intake, respectively. No differences in total daily energy, macro, micronutrient intake, and type of diet consumption were found between the HPD and non-HPD groups (Table S8-S10).

Instead of analyzing individual nutrients or foods, we conducted a comprehensive dietary pattern analysis to investigate the broader impact of the overall diet. To accomplish this, we employed the Diet Quality Index-International (DQI-I), a robust cross-national diet quality assessment tool. We utilized factor analysis to discern and characterize distinct dietary patterns within our study cohort. The total score of the DQI-I approached approximately 60% of the maximum achievable score in both the HPD and non-HPD groups. Notably, no significant differences were observed in the scores of each component, except for the overall balance of macronutrient intake components in energy and fatty acid composition (Table S11). Upon analysis, the lowest quartile (0-25th percentile) of DQI-I balance scores, which represent poor quality dietary, were distinct among the two groups (p-value = 0.001) (Figure S7). The HPD group displayed higher energy and fatty acid intake than the non-HPD group. The HPD group also failed to meet the recommended proportionality in macronutrient intake, primarily due to excessive consumption of fats above the recommended maximum (P = 0.03) (Figure S7).

A correlation matrix showing paired correlations of all food items was produced (Fig. 4A). Using factor analysis, we identified five distinct dietary patterns, collectively accounting for 53% of the total variance. Pattern 1 was characterized by high consumption of vegetables, poultry, meat, and legumes. Pattern 2 was distinguished by its prominent intake of sugar and bread. In Pattern 3, there was a marked consumption of fats and dairy products (Fig. 4B). Pattern 4 was primarily characterized by a heightened intake of rice. In contrast, Pattern 5 was notable for its considerable consumption of fish and seafood (Figure S8). Notably, the HPD group demonstrated higher individual factor scores for food pattern 3 than the non-HPD group. However, differences did not meet the significance threshold in our sample (P = 0.07) (Fig. 4C).

Fig. 4.

Correlation matrix shows paired correlations of all food items. Red color indicates positive correlations, and blue color indicates negative correlations (A). Dietary patterns derived from factor analysis using 15 food groups. Food pattern 3 factor loading which predominate fat (0.67) and daily products (0.88) is shown (C). Boxplots show the individual scores of dietary pattern 3 for HPD (red) and non-HPD (blue). The HPD group demonstrated a trend non-significantly higher scores in the food pattern 3 compared to the non-HPD groups by Wilcoxon test (p-value = 0.07).

Discussion

The effects of dietary habits on the composition of gut microbiome have been previously published^23,24. We conducted whole metagenome shotgun sequencing to profile gut microbial composition in a prospective cohort of Thai patients with advanced NSCLC who received ICI therapy. The most abundant microbiota composition in our cohort at the phylum levels was Bacteriodetes (49%), Firmicutes (36.6%), and Proteobacteria (8.9%), which contrasted with healthy middle-aged Thai participants, non-smokers with regular bowel habits, which had been previous published, whose phylum levels were predominant Firmicutes (82.1%), Bacteriodetes (6.7 and Proteobacteria 4.2%²⁵. Consistent with other reports²⁶, environmental influences on the gut microbiome such as age, smoking, diet, and disease factors should be considered as part of the interpretation.

Our findings revealed that individuals with a high abundance of Firmicutes unclassified, and Ruminococcaceae family (GGB9730_SGB15291) experienced a significantly reduced PFS after ICI therapy compared to their counterparts who showed a lower abundance of these microbial genera. More specifically, within the subset of NSCLC patients treated solely with ICI therapy, those exhibiting an elevated abundance of Intestinimonas genus and the Enterobacteriaceae family also faced a notably shorter PFS after ICI therapy. Our taxonomic analysis at the species level further confirmed these findings by multivariate analysis. Prior investigations evaluating the gut microbial composition in relation to ICI response have revealed considerable variability in their findings. Previous studies on NSCLC patients have employed 16 S rRNA sequencing to elucidate the relationship between the gut microbiome and ICI response. Such approaches often rely on distinct reference datasets and offer a limited taxonomic profile in comparison to shotgun metagenomics sequencing^27–29. A study by Bertrand et al., employing shotgun metagenomics sequencing to profile the gut microbiota of advanced NSCLC patients receiving ICI therapy, identified a correlation between gut microbiomes Firmicutes unclassified/classified, Akkermansia muciniphila, Alistipes genus, and ICI response²⁶. These findings presented a distinct gut microbiome comparable to our findings. Several preclinical studies found a diverse composition of commensal gut microbiome influencing the therapeutic response of anti-PD-1 therapy by affecting the tumor microenvironment^13,30–32. Differing definitions of ICI response or geographical differences in the gut microbiome³³ might be the factor of diverse results of taxa between each study. Our purpose was to document the association between the Thai population and response to ICI, especially patients who had progression of disease within 2 months. We believe that this cut-off point reflects the biological reality of patients who do not receive any benefit from ICI therapy.

Our research underscored the association between dietary patterns and differential outcomes in response to ICI therapy. Rather than focusing on individual nutrients or specific food items, we conducted a comprehensive dietary pattern analysis to investigate the broader impact of overall diet. Upon evaluation of the DQI-I scores, we observed distinct patterns among the two groups. Notably, the HPD group manifested a pronounced deficiency in the intake of essential nutrients such as iron, calcium, and vitamin C. Their dietary patterns were characterized by notably elevated intakes of cholesterol and sodium. A significant deviation was observed from the recommended macronutrient proportionality, largely attributable to the ingestion of fats beyond the recommended limits. It was further observed that the HPD group demonstrated a potential to show elevated scores in food pattern 3, which is predominantly defined by its considerable intake of fats and dairy products. Several prior studies have underscored the modulatory effect of dietary fiber intake on the gut microbiome, suggesting its potential association with enhanced clinical response to ICI therapy³⁴. High fat diets also modulate gut microbiome composition and affect systemic inflammatory response^35,36. Combining direct and indirect effects of gut microbiome and diet intake may notably influence the impact of ICI therapy in advanced NSCLC patients.

Despite being the first study examining the association between a Thai diet and ICI therapy, there are some limitations. First, the number of patients who returned their food diary is limited which made it difficult to reach statistical significance. Subsequent mediation analysis on gut microbiome was not able to be performed. Second, a full understanding of the potential food pattern that might affect outcomes of ICI therapy and host inflammatory response is ongoing and needs additional study and research to fill this knowledge gap. Lastly, the combination of chemotherapy and ICI therapy (31.6%) in our cohort might not reflect the true biology of gut microbiome after ICI therapy. However, we believe that our findings shed novel light on the complex relationship between the gut microbiome, dietary pattern, and ICI treatment efficacy.

In conclusion, our study of NSCLC patients, which mainly focused on those with rapid disease progression during ICI therapy, revealed a distinct association between the gut microbiome composition and treatment outcomes. Additionally, distinct dietary patterns in the HPD group revealed deficiencies in key nutrients and a propensity towards elevated fat and dairy consumption. Our findings suggested that the overall quality of one’s diet plays a pivotal role in ICI therapeutic outcomes.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

PD-1

Programmed death-1

PD-L1

Programmed death-ligand 1

Author contributions

Piyada Sitthideatphaiboon contributed to the conception, investigation, analysis, and drafted the manuscript. Nicha Somlaw, Nicha Zungsontiporn and Narittee Sukswai contributed to data acquisition. Pongsakorn Ouwongprayoon, Krittiya Korphaisarn and Naravat Poungvarin contributed specimen collection. Chatchawit Aporntewan contributed to the analysis of data. Chanida Vinayanuwattikun contributed to the design of the research, conception, analysis of the data, funding acquisition, and drafted the manuscript. Nattiya Hirankarn contributed to funding acquisition and consultation. All authors critically revised the manuscript, agreed to be fully accountable for ensuring the integrity and accuracy of the work, and approved the final manuscript.

Funding

This project is funded by the National Research Council of Thailand (NRCT) to PS, CV and NH [Grant number N35A660426] and Rachadapisek Sompote Matching fund, Faculty of Medicine, Chulalongkorn University [Grant number RA-MF-04/67] to NH. HN is a member of the Thailand Hub of Talents in Cancer Immunotherapy (TTCI). The academic endeavors of TTCI receive support from the National Research Council of Thailand [Grant number N35E660102]. The biospecimen collection was supported by Biobank, Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand.

Data availability

The data that support the findings of this study are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

All patients provided written informed consent. This study was approved by the Institutional Review Board of the Faculty of Medicine at Chulalongkorn University (IRB No. 385/63) and Mahidol University (IRB No. 374/2564).

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-79339-6.