Article Text

Download PDFPDF

Protocol
Artificial intelligence-derived gut microbiome as a predictive biomarker for therapeutic response to immunotherapy in lung cancer: protocol for a multicentre, prospective, observational study
  1. Fumihiro Shoji1,
  2. Takanori Yamashita2,
  3. Fumihiko Kinoshita1,
  4. Shinkichi Takamori1,
  5. Takatoshi Fujishita1,
  6. Ryo Toyozawa1,
  7. Kensaku Ito1,
  8. Koji Yamazaki3,
  9. Naoki Nakashima2,
  10. Tatsuro Okamoto1
  1. 1Department of Thoracic Oncology, National Kyushu Cancer Center, Fukuoka, Japan
  2. 2Medical Information Center, Kyushu University, Fukuoka, Japan
  3. 3Department of Thoracic Surgery, National Hospital Organisation Kyushu Medical Center, Fukuoka, Japan
  1. Correspondence to Dr Fumihiro Shoji; fumshojifumshoji{at}gmail.com

Abstract

Introduction Immunotherapy is the fourth leading therapy for lung cancer following surgery, chemotherapy and radiotherapy. Recently, several studies have reported about the potential association between the gut microbiome and therapeutic response to immunotherapy. Nevertheless, the specific composition of the gut microbiome or combination of gut microbes that truly predict the efficacy of immunotherapy is not definitive.

Methods and analysis The present multicentre, prospective, observational study aims to discover the specific composition of the gut microbiome or combination of gut microbes predicting the therapeutic response to immunotherapy in lung cancer using artificial intelligence. The main inclusion criteria are as follows: (1) pathologically or cytologically confirmed metastatic or postoperative recurrent lung cancer including non-small cell lung cancer and small cell lung cancer; (2) age≥20 years at the time of informed consent; (3) planned treatment with immunotherapy including combination therapy and monotherapy, as the first-line immunotherapy; and (4) ability to provide faecal samples. In total, 400 patients will be enrolled prospectively. Enrolment will begin in 2021, and the final analyses will be completed by 2024.

Ethics and dissemination The study protocol was approved by the institutional review board of each participating centre in 2021 (Kyushu Cancer Center, IRB approved No. 2021-13, 8 June 2021 and Kyushu Medical Center, IRB approved No. 21-076, 31 August 2021). Study results will be disseminated through peer-reviewed journals and national and international conferences.

Trial registration number UMIN000046428.

  • respiratory tract tumours
  • respiratory medicine (see thoracic medicine)
  • respiratory tract tumours
http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Strengths and limitations of this study

  • The present study is a multicentre, prospective, observational study.

  • This is a study to discover the specific composition of the gut microbiome or combination of gut microbiome comprising a true predictive biomarker of therapeutic response to immunotherapy in patients with lung cancer using artificial intelligence.

  • This study could also perform the subgroup analysis for the different population enrolled such as non-small cell lung cancer versus small-cell lung cancer or mono-immunotherapy versus combined-immunotherapy.

  • As with all observational studies, any association may not be causal, and it will require evaluation in randomised controlled studies.

Introduction

Immunotherapies such as immune checkpoint inhibitors (ICIs) targeting programmed cell death-1 (PD-1), programmed cell death-ligand 1 (PD-L1) and cytotoxic T-lymphocyte-associated protein-4 (CTLA-4) are widely used to treat various malignancies, including lung cancer. Immunotherapy has changed therapeutic approaches to cancer treatment. Anti-PD-1 (eg, nivolumab, pembrolizumab), anti-PD-L1 (eg, atezolizumab, durvalumab) and anti-CTLA-4 antibodies (eg, ipilimumab) enable T-cell activation and induce an immune response to cancer. Many studies reported that cancer immunotherapies including monotherapies or combination regimens featuring platinum-based chemotherapy improved patient survival in both non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC).1–15 Nevertheless, there is no widely accepted optimal biomarker for predicting the efficacy of immunotherapy; therefore, it is necessary to establish a definite predictive biomarker of therapeutic response to cancer immunotherapy. A previous study reported that the gut microbiome can modulate the host immune response (eg, antitumour immunity) and optimise both innate and adaptive immune responses.16 Gori et al17 reviewed the association between the microbiome and host immunoregulation. They introduced that some microbiome could sustain anticancer immunosurveillance through broadening T-cell receptor repertoire or enhancing immune response; on the other hand, some microbiome might suppress host immunity by interaction with T and Natural Killer cells.

Host immunity is clearly associated with the response to ICIs, and the internal microbiome is regarded as a controlling factor of host immunity, which supported our recent finding that the pretreatment host immunonutritional condition was a prognostic biomarker for patients with NSCLC who received immunotherapy.18 In fact, preclinical analyses illustrated that the gut microbiome composition and its modification in murine models could influence the efficacy of ICIs.19 20 Therefore, the microbiome has been emphasised as a predictive biomarker of immunotherapy mainly based on reports from the USA or Europe. In addition, the abundance of specific gut microbiome components or the gut microbiome diversity has also been reported to be related to the efficacy of anti-PD-1 antibodies in patients with melanoma.21 Moreover, faecal microbiome transplantation (FMT) in murine models potentially restored the response to ICIs.22 23 In a recent study, FMT from ICI responders to ICI non-responders resulted in ICI efficacy in patients with melanoma.24

Remarkably, there are definite differences in the microbiome composition among ethnicities.25 Thus, we recently reported that high gut microbiome diversity and the presence of specific microbes such as the genus Blautia and order RF32 unclassified were significantly correlated with the therapeutic response to immunotherapy in Japanese patients with NSCLC. However, the study had several limitations. First, the gut microbiome was examined ‘during’ immunotherapy opposed to ‘before’ treatment. Second, the sample size was small. Therefore, additional large multicentre studies to clarify the ‘genus-level gut microbiome’ as a predictive biomarker of cancer immunotherapy before treatment initiation are needed. In addition, the specific composition and combination of gut microbiome species that influence the response to immunotherapy are numerous, meaning there is limited ability to clarify the specific composition or significant combinations of gut microbiome species using statistical analysis via human intelligence.

Recently, the development of artificial intelligence (AI)-based technologies for medical data has progressed, and evidence has been discovered that could not be clarified by conventional methods.26 27 Such technologies are expected to be used for decision-making in medical fields. It is also expected that AI will be applied in the clinical treatment of lung cancer. Compared with conventional medical statistics methods, machine-learning models can more easily capture the non-linearity of medical data, and they exhibit higher prediction performance. These models are effective for diverse and large amounts of medical data, and extract contribution variables for the purpose.

In this analysis, we will conduct a prospective observational study to clarify the specific gut microbiome composition or combination of gut microbes associated with the response to immunotherapy in patients with lung cancer using AI.

Methods and analysis

Study objectives

Our objective is to discover the specific composition of the gut microbiome or combination of gut species comprising a predictive biomarker of immunotherapy efficacy in patients with lung cancer using AI.

The study was approved by institutional review board of each participating centre.

Study setting

The present study is a multicentre, prospective, observational study.

Observational points

In the present study, we will clarify the following items:

  1. The specific composition of the gut microbiome or combination of gut microbes associated with clinical/graphical therapeutic responses to immunotherapy.

  2. The diversity of the gut microbiome associated with clinical/graphical therapeutic responses to immunotherapy.

  3. The specific composition of the gut microbiome or combination of gut microbes associated with immune-related adverse events (irAEs, ≥grade 3).

  4. The diversity of gut microbes associated with irAEs (≥grade 3).

  5. The association of the composition, combination or diversity of gut microbes with clinicopathological features including PD-L1 expression, haematological data, immunonutritional indices, the use of medicines (corticosteroids, antibiotics, proton pomp inhibitors, and probiotics) within 28 days before the initiation of immunotherapy and patients’ dietary habits such as the consumption of milk and fermented foods.

  6. Progression-free survival (PFS), overall survival, efficacy (response rate, time to treatment and disease control rate) and safety according to the following items: type of pathology, the use of medicines at baseline, sex, age at baseline, Eastern Cooperative Oncology Group performance status at baseline, patients’ dietary habits, regimens, the number of treatments and regimens after the failure of immunotherapy, the number of immunotherapy cycles completed, the best response to immunotherapy, tumour proportional score (tumorous PD-L1 expression), the past history of autoimmune disease, smoking history and the status of driver mutations.

Data collection

Faecal samples will be collected before the initiation of immunotherapy and analysed to assess the correlations of the clinical/graphical therapeutic response to immunotherapy, severity or types of irAEs and the continuation/discontinuation of treatment with patients’ gut microbiomes. The present study design is presented in figure 1 and table 1.

Table 1

Summary of data collection and timeline

Figure 1

Study design. irAE, immune-related adverse event.

Eligibility criteria

Inclusion criteria

  1. Pathologically or cytologically confirmed metastatic or postoperative recurrent NSCLC or SCLC.

  2. ≥20 years old at the time of informed consent.

  3. Planned treatment with immunotherapy including combination regimens and monotherapies as the first immunotherapy.

  4. Ability to provide faecal samples.

  5. Provision of written informed consent.

Exclusion criteria

  1. Diseases eligible for definitive chemoradiotherapy.

  2. Patients deemed inappropriate for the study by the investigator.

Sample size

A sample size and power calculation before study initiation were not performed because this is an exploratory observational study to clarify the predictive biomarker of therapeutic responses to immunotherapy. The present study will be conducted by two high-volume centres. Therefore, we estimate 100 participants per year in each centre. Thus, we plan to include total 400 patients over a 2-year period. This number will be sufficient to be analysed by AI, although there are few data of sample sizes recommended due to few studies similar to the present study. In total, 400 participants will be enrolled prospectively over 2 years.

Registration

Accrual started in September 2021.

Study period

The study started in September 2021 and will end 12 months after recruitment of the last participant, expected in September 2024.

Study population

We are planning to recruit eligible participants from participating hospitals. All enrolled patients will have at least one measurable target lesion based on the Response Evaluation Criteria in Solid Tumours, V.1.1.28 Clinical/pathological stage will be based on the tumour node metastasis (TNM) classification established by the International Union Against Cancer.29 For TNM staging, all patients will undergo CT of the thorax and upper abdomen, as well as bone scintigraphy, brain CT, MRI or fluorodeoxyglucose-positron emission tomography. Postoperative local or distant recurrence will be defined as described previously.30 Immunotherapy will be continued until radiographic progression or discontinuation because of severe irAEs or the patient’s request. PD-L1 protein expression will be evaluated using antibody clone 22C3 (Dako, Agilent Technologies, Santa Clara, California, USA). Adverse events will be graded according to Common Terminology Criteria for Adverse Events, V.5.0.

Treatment plan

The present study is allowing the immunotherapy regimens presented in table 2.

Table 2

Immunotherapy regimens adopted in the present study

Monotherapy: pembrolizumab will be administered at a dose of 200 mg intravenously every 3 weeks.1–3 Atezolizumab will be administered at a dose of 1200 mg intravenously every 3 weeks.4 5 Nivolumab will be administered at a dose of 3 mg/kg intravenously every 2 weeks.6

Combined therapy: KEYNOTE 189 regimen and KEYNOTE 407 regimen: carboplatin (area under the concentration–time curve (AUC)) of 5–6 mg/mL/min plus pemetrexed 500 mg/m2 (KEYNOTE 189) or paclitaxel 200 mg/m2 (KEYNOTE 407). Maintenance therapy with pembrolizumab with/without pemetrexed (KEYNOTE 189) or pembrolizumab alone (KEYNOTE 407) is given every 3 weeks after induction therapy until disease progression, unacceptable toxicity, or death.7 8 IMpower130 regimen: atezolizumab (1200 mg intravenously every 3 weeks) plus carboplatin (AUC of 6 mg/mL/min every 3 weeks) plus nab-paclitaxel (100 mg/m² every week).9 IMpower132 regimen: atezolizumab (1200 mg intravenously every 3 weeks) plus cisplatin (75 mg/ m2) or carboplatin (AUC of 6 mg/mL/min) plus pemetrexed 500 mg/m2 every 3 weeks.10 IMpower150 regimen: atezolizumab (1200 mg intravenously every 3 weeks) plus carboplatin (AUC of 6 mg/mL/min for 4 cycles) plus paclitaxel (200 mg/m2 intravenously every 3 weeks for 4 cycles) plus bevacizumab (15 mg/kg) intravenously every 3 weeks.11 Maintenance therapy with atezolizumab plus bevacizumab (IMpower150), atezolizumab plus pemetrexed (IMpower132) or atezolizumab alone (IMpower130) is given every 3 weeks after induction therapy until disease progression, unacceptable toxicity, or death. CheckMate 227 regimen: nivolumab (3 mg/kg intravenously every 2 weeks) plus ipilimumab (1 mg/kg intravenously every 6 weeks).12 CheckMate 9LA regimen: nivolumab (360 mg intravenously every 3 weeks) plus ipilimumab (1 mg/kg intravenously every 6 weeks) combined with histology-based platinum doublet chemotherapy (intravenously every 3 weeks for two cycles). The intravenous chemotherapy regimens consist of carboplatin (AUC of 6 mg/mL/min) plus paclitaxel (200 mg/m2) for patients with squamous histology or carboplatin (AUC of 5–6 mg/mL/min) or cisplatin (75 mg/m2) plus pemetrexed (500 mg/m2) for patients with non-squamous histology.13 Thereafter, treatment with nivolumab plus ipilimumab continues until disease progression or unacceptable toxicity. IMpower133 regimen: carboplatin (AUC of 5mg/mL/min intravenously on day 1) plus etoposide (100 mg/m2 intravenously on days 1–3) plus atezolizumab (1200 mg intravenously every 3 weeks) for four cycles.14 Maintenance therapy with atezolizumab alone is given every 3 weeks after induction therapy until disease progression, unacceptable toxicity, or death. CASPIAN regimen: carboplatin (AUC of 5–6 mg/mL/min intravenously on day 1) or cisplatin (75–80 mg/m2) plus etoposide (80–100 mg/m2 intravenously on days 1–3) plus durvalumab (1500 mg intravenously every 3 weeks) for four cycles.15 Maintenance therapy with durvalumab alone is given every 4 weeks after induction therapy until disease progression, unacceptable toxicity or death.

Sample collection, DNA extraction, gene amplification, sequencing and data analysis procedures

Faecal samples will be collected in sterile containers, immediately incubated at 4°C, and then frozen at −80°C. The preliminary treatment of faecal samples will follow a previously described method,31 followed by DNA extraction using an automated DNA isolation system (Gene Prep Star PI-480, Kurabo, Japan). DNA will be extracted from faeces using a Mora-Extract kit (Kyokuto Pharmaceutical, Japan). The V3–V4 regions of bacterial 16S rRNA genes will be amplified using the Pro341F/Pro805R primers31 and dual-index method32 under hemi-nested PCR conditions.33 Barcoded amplicons will be 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 will be merged using the fastq-join programme with default settings.34 The joined amplicon sequence reads will be processed through QIIME 2 V.2020.6.35 The chimeric and low quality sequences were filtred using the DADA2 denoise-single plugin V.2017.6.036 based on default configuration. The taxonomy of representative sequences will be assigned using the Ribosomal Database Project Multi-classifier V.2.11.37 38 by training a naïve Bayes classifier using the q2-feature-classifier plugin. To account for compositional artefacts, we transformed relative abundances using the centred log-ratio transformation.39 Alpha diversity indices (Chao1, Shannon and Simpson) will be calculated using the alpha-rarefaction plugin. The statistical significance of the Chao1, Shannon and Simpson indices among the groups will be assessed by the Kruskal-Wallis test using the alpha-group-significance plugin. Beta diversity will be analysed by the weighted UniFrac, unweighted UniFrac and Bray-Curtis distances using the core-metrics-phylogenetic plugin. The Emperor tool will be used to visualise principal coordinates analysis plots. The statistical significance of similarity of bacterial communities among the groups will be assessed by the ANalysis Of SIMilarities (ANOSIM) test using the beta-group-significance plugin. The heatmap and ward clustering from phylum to species will be presented using the feature-table heatmap plugin.

Machine-learning analysis

To develop predictive models, we will use gradient boosting decision trees (GBDTs).40 Additionally, the feature variables obtained using GBDTs will be interpreted using SHapley Additive exPlanation (SHAP). The Shapley value is a solution concept of fairly distributing both gains and costs, and SHAP could also explain the output of any machine-learning model. These methods are the most powerful techniques for building and interpreting predictive models.41 42 All machine-learning analyses will be performed using the R programme.

Statistical analysis

Categorical variables will be analysed using Fisher’s exact test. Continuous variables will be compared using the χ2 test. The Mann-Whitney U test will be used to determine significant differences among the different groups using alpha diversity, which reveals the diversity in each individual sample. Logistic regression analysis will be performed to calculate ORs for the response to ICIs with respect to clinicopathological characteristics. Kaplan-Meier statistics and the log-rank test will be applied to evaluate PFS. Statistical analyses will be performed using JMP software, V.14.0 (SAS Institute). Significance will be indicated by p<0.05.

Patients and public involvement

Patients and/or public were not involved in the design of the present study.

Ethics and dissemination

Study results will be disseminated through peer-reviewed journals and national and international conferences.

Ethics statements

Patient consent for publication

Acknowledgments

We thank Joe Barber Jr, PhD, from Edanz (https://jp.edanz.com/ac) for editing a draft of this manuscript.

References

Footnotes

  • Contributors FS, FK, ST, TF, RT, KI, KY and TO significantly contributed to the conception and design of the study protocol. TY and NN provided statistical expertise for AI. The protocol was written by FS and critically reviewed by TO, and FS drafted the manuscript. All authors read and approved the final paper.

  • Funding This work was supported by Uehara Memorial Foundation, KAKENHI (No. 20K09188), Suzuken Memorial Foundation, and J-milk Foundation.

  • Competing interests None declared.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

  • Provenance and peer review Not commissioned; externally peer reviewed.