Background

Cancer is a major contributor to global mortality, with colorectal cancer ranking as the second leading cause of cancer-related deaths in 2020, according to the Global Cancer Epidemiological Statistics (GCES) released by the World Health Organization (WHO) [1, 2].

Growing evidence indicates that the composition of humans’ gut microbiota significantly determines our overall health and response to disease [3, 4]. The composition of the gut microbiota, consisting of various microorganisms such as bacteria, archaea, viruses, fungi, and protozoa, has been linked to the emergence and progression of gastrointestinal cancers like colorectal cancer [5, 6]. Recent studies have indicated a significant relationship between the gut microbiota and the efficacy of cancer therapy [1]. The field of cancer immunotherapy is advancing rapidly, with increasing attention to the reciprocal influence between the gut microbiome and cancer treatment outcomes; thus, investigating the role of the gut microbiome in anti-cancer immune surveillance and immunotherapy holds great promise in optimizing treatment responses in cancer patients [7, 8]. The human gut microbiota composition consists primarily of two major phyla: Bacteroidetes and Firmicutes [3]. Various factors such as diet, culture, residence, and medications influence the makeup of the gut microbiota [4, 9]. However, longitudinal studies have shown that individuals who respond favorably to cancer treatment regimens are loaded in specific bacterial taxa than those who do not respond well [10, 11].

The immune system has an essential role in controlling cancer growth and progression. Tumor cells can evade immune system recognition by multiple mechanisms. Immunotherapy augments the human immune defense against cancer. The immune checkpoints pathway encompasses various proteins that exert activating and inhibitory effects to maintain immunological self-tolerance. These checkpoints are instrumental in regulating cytotoxic T-lymphocyte activation, preserving self-tolerance, preventing autoimmunity, and modulating the intensity and duration of immune reactions to limit inflammation-induced tissue damage [12].

The most well-described checkpoints are the cytotoxic T lymphocyte-associated molecule-4 (CTLA-4), programmed cell death receptor-1 (PD-1), and programmed cell death ligand-1 (PD-L1). Monoclonal antibodies block these pathways to enhance the host immunologic activity against tumors and become the standard of care for many malignancies [12]. Currently, the immune checkpoint inhibitors approved by the FDA for CRC are anti-(CTLA-4) (ipilimumab), PD-1 programmed cell death protein 1 (pembrolizumab and nivolumab), and PDL-1(atezolizumab, avelumab, and durvalumab). ^11–13

Several studies have investigated gut microbiota’s impact on cancer patients’ response to immune checkpoint inhibitors [9–11, 13]. However, limited data is available to ascertain whether gut microbiota can serve as a reliable biomarker for assessing the effectiveness of ICIs in colorectal cancer therapy [14].

The present systematic review seeks to build upon the findings from recent systematic reviews in 2021, one of which evaluated the efficacy of ICIs in treating advanced solid cancers, including CRC, and found that combined therapy significantly improved disease control rates and survival time at the cost of increased severe adverse reactions compared to monotherapy [15]. The other systematic review explored the potential of tumor mutational burden as a predictive biomarker for CRC patients receiving ICIs therapy [16]. With these foundations, our systematic review aims to extend these understandings focusing on the association between gut microbiota and the clinical efficacy of ICIs in CRC patients.

Study selection and inclusion criteria

We conducted a thorough literature search in five biomedical databases, including PubMed, Scopus, Web of Science, Embase, and Cochrane Library, until July 18, 2023, S2 Table. The search terms used were: (Gastrointestinal Microbiome OR Gastrointestinal Microbiota OR Gastrointestinal Microflora OR Gastrointestinal Flora OR Gut Microbiome OR Gut Microbiota OR Gut Microflora OR Gut Flora OR Intestinal Microbiome OR Intestinal Microbiota OR Intestinal Microflora OR Intestinal Flora) AND (nivolumab OR pembrolizumab OR atezolizumab OR avelumab OR durvalumab OR ipilimumab OR tremelimumab OR immunotherapy OR PD-1 OR PD-L1 OR CTLA-4 OR immune checkpoint OR checkpoint blockade OR immune checkpoint inhibitors OR ICI OR ICIs OR immune checkpoint blockers OR ICB OR ICBs) AND (colon cancer OR colorectal cancer OR colorectal carcinoma OR colon carcinoma OR colonic neoplasm OR colorectal neoplasm) (S1 Table). The study protocol provides a detailed description of the search strategy and has been made publicly accessible [18].

Reviewers (SA, SZ, MO, LL, and ZA) conducted the study selection and evaluation process via Covidence (Melbourne, Australia) Disagreements were resolved through discussion and consensus between (SA, SZ, and ZA). Duplicate publications were removed before applying five criteria in the PICOS format to select the articles. These criteria included: (1) Participants included patients of any age and sex who were diagnosed with colorectal cancer; (2) The intervention involved the use of ICIs for cancer immunotherapy, such as anti-PD-1, anti-PD-L1, anti-CTLA-4, or a combination of ICIs at various doses and timings, (3) The comparison focused on the gut microbiota composition reported from molecular analysis in CRC patients, (4) The outcomes were evaluated based on the classification of patients into “responders” and “non-responders” to ICIs, following the RECIST 1.1 criteria. and (5)

In our systematic review, we concentrated on published observational human studies that had been subjected to the peer-review process, ensuring the inclusion of only the most robust and directly applicable data. We excluded experimental animal studies, recognizing the challenges in translating such findings to human physiology. Additionally, reviews, comments, case reports, editorials, and conference abstracts were omitted due to their lack of original empirical research, which is essential for the integrity of a rigorous systematic review. This approach was designed to furnish the clinical oncology community with an analysis grounded in reliable and broadly applicable evidence.

Despite our initial assertion to exclude conference abstracts, we made an exception for one abstract from a large-scale clinical trial (Checkmate 142) [19]. Considering the nascent nature of the systematic review topic and the limited number of clinical studies available, including this abstract was deemed necessary to augment data synthesis. Additionally, a chosen study incorporated mouse experiments [20]. Its clinical importance and the limited number of full-fledged studies on CRC patients in this niche area warranted its inclusion. Consequently, after evaluating the full texts, we selected studies that fulfilled the pre-determined eligibility criteria.

Data extraction and quality assessment

The selected studies were reviewed (SA, SZ, HO, ZR, OA, SHKU, SHNU), and data was extracted independently by (SA, SZ, HO, and ZA). The collected data consisted of the author’s information, publication year, number of patients, ICI therapy type, patient recruitment location, gut microbiota sequencing and analysis methods used, outcomes related to diversity and taxonomy of the gut microbiota, as well as survival outcomes as in responders versus non-responders to cancer ICIs therapy. The included studies consisted of two prospective cohort studies, one case-control, and a snapshot (cross-sectional) analysis from a randomized controlled trial (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Clinical studies evaluating the association between gut microbiome and clinical efficacy of the ICIs.

https://doi.org/10.1371/journal.pone.0307639.t001

To assess the rigor of the studies, we utilized standardized assessment tools from the National Institutes of Health (NIH) that align with each specific study type (Table 2). These tools provided scores ranging from 0 to 14. Studies were categorized as having intermediate quality if they scored between 5 and 11, while those scoring between 12 and 14 were deemed to have good quality [21].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Risk of bias for included studies using NIH quality assessment tool.

https://doi.org/10.1371/journal.pone.0307639.t002

Literature search results

After utilizing the systematic review software Covidence [22] to eliminate 1,887 duplicated articles, we identified 5,132 papers from PubMed, Scopus, Web of Science, Embase, and Cochrane Library databases. After thoroughly examining the full texts, we excluded 19 studies deemed irrelevant to the topic or did not meet the criteria for proper comparison, indication, intervention, study design, or patient population. Ultimately, four studies were included in our analysis (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA flowchart and decision-making process for the inclusion criteria of studies.

https://doi.org/10.1371/journal.pone.0307639.g001

Characteristics of studies

Four studies have met the criteria for examining the gut microbiota in patients with colorectal cancer who underwent ICIs cancer therapy. These studies were published between 2017 and 2022 and involved sample sizes ranging from 60 to 89 participants. Cheng et al. 2022 [23], Peng et al. 2020 [24], and Pi et al. 2020 [20] conducted their studies in China. On the other hand, the study by Koptez et al. 2017 [19] is a sub-study of CheckMate 142, an international clinical trial conducted in eight countries: Australia, Belgium, Canada, France, Ireland, Italy, Spain, United States, registered under ClinicalTrials.gov identifier NCT02060188 (Table 1).

Cheng et al. and Peng et al. were prospective cohort studies that included 72 and 89 patients, 5 and 19 of whom had CRC, respectively [19, 20]. The third study, Pi et al., involved a case-control design that examined 60 patients with colorectal cancer who underwent various treatment regimens, including immune checkpoint inhibitors [20]. Kopetz et al. enrolled a total of 72 patients with colorectal cancer [19].

Studies conducted by Cheng et al., Peng et al., and Kopetz et al. examined the relationship between gut microbiome composition and the response to anti-PD-1/PD-L1 immunotherapy in colorectal cancer patients following the RECIST 1.1. In Pi et al.’s analysis, bacterial fluctuations were noted; however, the association with anti-PD-1/PD-L1 immunotherapy response in CRC patients was not distinctly articulated. Although the study suggested a correlation, the inadequate labeling and explanation in the figures constrained the interpretation of the data. Consequently, the results were inferred by contrasting shifts in bacterial taxa within both groups, drawing on previous studies that utilized similar bioinformatics and visualization tools for comparison.

Predictive taxonomic profiles of CRC patients in response to ICIs

The taxonomic distribution or differential abundance of microbial communities differed between CRC patients responding to immunotherapy and non-responders.

Kopetz and colleagues [19] investigated colorectal cancer patients who had undergone ICIs, nivolumab, and ipilimumab and possessed deficiency in DNA mismatch repair (dMMR), causing high microsatellite instability (MSI-H). The study observed shifts in the gut microbiome using 16S rRNA sequencing on baseline stool samples. This investigation was a component of a broader clinical trial series, CheckMate 142. Remarkably, Rothia and Rothia mucilaginosa, from the Micrococcaceae family within the Actinobacteria phylum, exhibited a 2.7-fold and 3-fold lower abundance, respectively, in patients partially responding to CRC ICIs compared to patients with progressive disease (Figs 2 and 3) [19].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Phylogenetic representation of gut microbiome association with ICIs treatment response in CRC patients.

The tree was constructed via phyloT software and visualized using iTOL enabling a comprehensive classification of bacterial taxa identified in the systematic review, presented in an inward manner from phylum to species. Differentially enriched bacterial taxa associated to better response were labelled in green while those associated with poorer response are in red. Taxa with mixed findings are denoted in yellow (i.e. in responder and non-responder groups). Lowercase alphabets (a, b, and c) adjacent to each bacterial taxon indicate the source study, asterisk (*) indicates identification at the family taxonomic level (S6 Table).

https://doi.org/10.1371/journal.pone.0307639.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Comparison of bacterial taxonomy in ICIs-responsive vs non-responsive colorectal cancer patients.

This comparison is conducted regardless of the sequencing approach reported in Peng et al., Kopetz et al., and Pi et al. The arrows indicate whether each taxon is differentially enriched (up arrow) or less abundant (down arrow) in the patient groups. The image was created with mindthegraphBioRender.com. ICIs-responsive vs non-responsive colorectal cancer patients © 2024 by Suad Ajab is licensed under CC BY 4.0.

https://doi.org/10.1371/journal.pone.0307639.g003

Pi and colleagues [20] performed 16S rRNA sequencing on CRC patients’ fecal specimens at the time of diagnosis to identify microbes in the patients’ gut. The researchers investigated the correlation between the number of genes (gene count) or the quantity of these identified metagenomic species (MGS) abundance and the patients’ 3-month progression-free survival. The term’ 3-month PFS’ refers to the time after treatment initiation, during and after which a patient lives without the disease worsening. In this instance, the researchers found that they could not predict the 3-month PFS based on the gene count or the abundance of MGS.

According to Pi et al.’s findings, CRC ICIs responders with six months of progression-free survival demonstrated a high abundance of the Akkermansia muciniphila, from the Verrucomicrobiota phylum, in addition to a unique mix of co-abundant bacterial gene (CAG) groups. These include, but are not limited to, Alistipes, Intestinimonas, Bacteroides nordi, Bacteroides xylanisolvens, Blautia, Lachnospiraceae, Firmicutes bacterium CAG:129, and Ruminococcus sp. CAG:353 (Fig 2 and S6 Table).

Conversely, CRC ICIs non-responders were found to harbor a different set of bacterial CAGs that included Clostridiales, Parabacteroides distasonis, Firmicutes bacterium CAG:227, and Anaerotruncus colhominis (Figs 2 and 3) [20].

Peng et al. [24] implemented 16S rRNA gene sequencing to investigate the link between gut microbiota and the clinical effectiveness of ICIs in advanced-stage gastrointestinal cancer patients. Their research focused on identifying differential abundant Operational Taxonomic Units (OTU) among responders and non-responders to ICIs. In CRC patients, the Prevotellaceae in the responder group were not significantly abundant; however, there was more Prevotellaceae relative to the quantity of Bacteroidaceae. Interestingly, this pattern did not extend to gastric cancer patients, underlining unique gut microbiota interplays specific to cancer type. Also, an increased abundance of Prevotellaceae demonstrated greater chances of disease stabilization (PFS) within 12 weeks after ICIs therapy. Moreover, an increased abundance of bacterial taxa such as Akkermansia muciniphila, Eubacterium rectale, Ruminococcaceae, Bacteroideceae, and Lachnospiraceae were identified in ICIs responders. Conversely, CRC ICIs non-responders exhibited a higher abundance of Bifidobacteria, Coprococcus, and other OTUs within the Ruminococcaceae and Bacteroides family (Figs 2 and 3) [24].

A consistent correlation between Prevotella and Bacteroides was observed in both 16S rRNA sequencing data and metagenomics shotgun sequencing data, as reported by Peng et al. [24]

Lastly, using machine learning tools, Peng et al. aimed to predict patient responses to therapy based on the presence of bacteria. These predictive models, especially the ExtraTrees and elastic net methods, identified 15 of the top 20 predictors belonging to the Firmicutes phylum, indicating that they may serve as predictive markers in ICIs therapy effectiveness.

In a comprehensive study conducted by Cheng et al. [23] in 2022, they discovered notable differences in the gut microbial biodiversity of patients with advanced stages of various cancer types. However, they have provided a pooled summary of gut microbiota composition across all cancer types studied, and a data repository was not provided to access the specific microbiota variations in response to CRC ICIs therapy. The bacterial family Prevotellaceae, focusing on the genus Prevotella, was at a higher concentration in ICIs responders. Furthermore, Cheng and colleagues used the Linear discriminant analysis Effect Size (LefSe) statistical method to illustrate variances in the presence of microbial taxa within distinct groups. Prior ICIs therapy, pan-cancer responders tended to have an abundance of specific microbes such as Archaea, Lentisphaerae, Victivallaceae, Victivallales, Lentisphaeria, Methanobacteriaceae, Methanobacteria, Euryarchaeota, Methanobrevibacter, Methanobacteriales, and Leuconostoc (S5 Table). On the other hand, the pan-cancer non-responders predominantly exhibited a high presence of the Clostridiaceae family within the phylum Firmicutes [23]. Six weeks post treatment, the microbial profile among patients with various types of cancer changed; the responders had an increase in Lachnospiraceae and Thermus, while in non-responders, the levels of Faecalibacterium and Bifidobacterium significantly decreased (S5 Table) [23].

Cheng et al. carried out a network analysis on the gut microbiota, including 16 different bacterial phyla from 72 advanced cancer patients. This analysis aimed to determine how changes in the abundance of one bacterial group could affect the quantity of another. A positive correlation signals a simultaneous rise in the presence of two bacterial groups, while a negative correlation suggests an inverse relationship; an increase in one bacterium coincides with a reduction in another bacterium. For instance, Bacteroidetes were negatively correlated with Firmicutes, Proteobacteria, and Actinobacteria. Conversely, Proteobacteria and Actinobacteria displayed a positive correlation [23].

Prediction of microbial functional pathways in response to ICIs

Peng et al. employed the HUMAnN2 tool to mine microbiota data for insights into associated metabolic and biological processes [24]. Advanced cancer patients responding to ICIs treatment exhibited enhanced bacterial activity associated with producing short-chain fatty acids, unsaturated fatty acids, vitamins, and starch. Indeed, the Eubacterium rectale, known for producing short-chain fatty acid and associated with fiber-rich diets, was vastly abundant in the CRC ICIs responsive group. Conversely, patients who responded poorly to ICIs showed elevated bacterial activity related to lipopolysaccharide production, sugar catabolism, and protein synthesis [24].

Using PICRUSt (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States) function prediction, Cheng et al. found that metabolic pathways L3_Stilbenoid diaryl-heptanoid, gingerol biosynthesis, and the L3_Glycosaminoglycan degradation pathway was more prominent in responders before all cancer type ICIs treatment, S4 Table [23].

Comparative analysis of methodologies utilized in microbiota profiling

The four studies [19, 20, 23, 24] used taxa-targeted methods focused on the 16S rRNA gene to identify bacterial presence in gut microbiota. The 2022 study by Cheng et al. utilized the 16S rRNA-V4 amplicon pyrosequencing method, which allows for genus-level classification but is unable to distinguish between species or strains of a genus in the 16S rRNA gene. Through this method, an analysis was conducted on 9708 OTUs from 144 samples, identifying 10 genera without extending to the species level, S3 Table and S1 Fig [25].

The 2020 studies by Peng et al. and Pi et al. utilized 16S rRNA-V3-V4 amplicon sequencing. Though this method can identify and compare bacteria, it is limited in resolving bacterial taxa (see S3 Table and S1 Fig for more details). Nevertheless, applying metagenomics with 16S rRNA gene sequencing leads to high taxonomic resolution results, as observed in Peng et al.’s study, where the Bacteroides genus is present in both CRC responder and non-responder groups. However, a more refined perspective emerges when examining at the species level. For instance, Pi et al.’s metagenomic study identifies that specific Bacteroides species such as Bacteroides nordii, Bacteroides caccae, and Bacteroides xylanisolvens are associated with improved responses in CRC patients (Fig 2 and S6 Table).

Peng et al.’s study utilized metagenomic shotgun sequencing on fecal samples from 55 out of the 74 advanced cancer patients. This method not only strengthened the initial 16S rRNA analysis findings but also introduced 10 additional differentially enriched species, see S3 Table and S1 Fig.

The applied methodologies in microbiota profiling and the choice of reference databases for taxonomic assignments can both introduce variation in the resolution of results obtained [26]. Cheng et al. and Pi et al. employed GreenGenes, while Peng used the SILVA database. GreenGenes, with almost 202,421 reference sequences, is widely used for its tree-based taxonomy placement. On the other hand, SILVA, with almost 9 million reference sequences, possesses a biannual updated comprehensive coverage of both small and large subunit ribosomal RNA sequences (S3 Table) [27].

Furthermore, Peng chose to use DADA2 for filtering and denoising raw reads, whereas Pi and Cheng opted for the QIIME tool. DADA2 is recognized for its ability to model and correct Illumina-sequenced amplicon errors and detect single-nucleotide polymorphisms (SNPs) across similar species. On the other hand, QIIME is an integrated pipeline that permits analysis of raw sequencing data, diversity, and taxonomic composition, and it is widely used due to its ability to seamlessly integrate various microbiome-specialized algorithms and databases, S4 Table [25, 27].

While striving for accuracy in gut microbiota identification, every methodology contends with unique challenges, including selecting sequencing regions, analytical pipelines, and choice of reference libraries or sequence databases that can introduce variations in study outcomes (S1 Fig and S3 Table). Variations in microbial identification were observed based on the sequencing methodology; Cheng et al. applied 16S rRNA sequencing to identify 9,708 OTUs across 10 genera [23]. On a different approach, Pi et al. used shotgun metagenomic sequencing, uncovering 72 CAGs, 19 genera, and 11 species [20]. Peng et al., employing 16S rRNA sequencing and shotgun metagenomics, discerned 3,852 OTUs across 58 genera and subsequently predicted differential metabolic pathways and 18 differential species, respectively (S1 Fig) [24]. It is imperative for future research to consider the significant influence of sequencing techniques on microbiota characterization, enhancing the accuracy and comprehensiveness of microbiota profiling analyses.

Assessment of risk of bias

Koptez et al. and Pi et al. were judged ‘Fair,’ signifying scores of 5 to 10 on a scale of 14 in response to NIH Quality Assessment Tools, Fig 4 and Table 2. Peng et al. and Cheng et al. garnered a ‘Good’ rating, with scores between 11 and 14 out of 14 questions, Fig 4 and Table 2. Upon evaluating the studies included in this systematic review, several potential sources of bias were identified, affecting the study’s internal validity. These biases arise from methodological and reporting limitations within the individual studies, including the lack of a thorough description and clearly defined study population sample size, incomplete or ambiguous reporting of recruitment details, and inconsistency in applying the inclusion and exclusion criteria. Furthermore, the non-availability of raw data repositories was a notable limitation, hindering a comprehensive and transparent meta-analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Traffic light plot illustrating the risk of bias, assessed using the National Institutes of Health (NIH).

The colors green, yellow, and red represent low, unclear, and high risk of bias respectively in each domain/question of the included studies.

https://doi.org/10.1371/journal.pone.0307639.g004

Only the investigation spearheaded by Peng et al. and colleagues considered all significant potential confounding variables. Cheng et al. conducted an exploratory study without statistical adjustments commonly used to balance the likelihood of chance findings in multiple comparisons. The aim was to find new insights and trends rather than validating hypotheses, emphasizing generating possible research directions over drawing statistically significant definitive conclusions.

Strengths and limitations

The reviewed studies [19, 20, 23, 24] exhibit robustness through rigorous methodologies, providing substantial evidence of a connection between gut microbiota and immune checkpoint inhibitors (ICIs) across various cancers. These studies contribute significantly to cancer treatment research by offering valuable insights gained through microbial identifications. However, it is essential to acknowledge the limitations of these studies, including small sample sizes, single-center origins, uncontrolled variability, and brief observational periods. Additionally, the unequal distribution of responders and non-responders across the studies poses a challenge to drawing comprehensive conclusions.

While Cheng and Kopetz’s study employed population selection based on histopathological examinations, Peng and Pi did not clearly outline the criteria for selecting cancer patients. Furthermore, except for Cheng et al., all studies in this systematic review predominantly relied on baseline gut microbiota profiling without considering the dynamic nature of the microbiome. The lack of continuous assessment of microbiome changes throughout ICI treatment limits the generalizability of findings. Consequently, further comprehensive research is essential to establish more definitive conclusions.

The four studies included in this systematic review focused on colorectal cancer (CRC) patients, with 54% from China and the remaining 46% from various countries (Australia, Belgium, Canada, France, Ireland, Italy, Spain, and the United States), as indicated in Table 1. However, the predominance of data from one geographical area raises concerns about potential geographic bias and the influence of ethnocultural factors, given the likelihood of a single ethnic, dietary, and cultural background. To enhance the validity and reliability of findings, future research in this oncology and microbiota interplay domain must consider a more geographically diverse patient population, addressing these potential biases. This recommendation considers the inherent challenges and complexities associated with designing and executing clinical studies involving cancer patients and collecting stool samples while accounting for a wide range of variables.

The systematic review’s limitations revolve around excluding non-English language publications, increasing the risk of language bias. Additionally, this review may exhibit publication bias, as it only includes published studies, potentially omitting relevant data. The limited number of studies in this niche area and the variability among patient populations and cancer types in the included studies may have further confounded the results.