Colorectal cancer screening programs with fecal sample collection may provide a platform for population-based gut microbiome disease research. We investigated sample collection and storage method impact on the accuracy and stability of the V3-V4 region of the 16S rRNA genes and bacterial quantity across seven different collection methods [i.e., no solution, two specimen collection cards, and four types of fecal immunochemical test (FIT) used in four countries] among 19 healthy volunteers.
Intraclass correlation coefficients (ICC) were calculated for the relative abundance of the top three phyla, the most abundant genera, alpha diversity metrics, and the first principal coordinates of the beta diversity matrices to estimate the stability of microbial profiles after storage for 7 days at room temperature, 4°C or 30°C, and after screening for the presence of occult blood in the stool. In addition, accuracy was estimated for samples frozen immediately compared to samples with no solution (i.e., the putative gold standard).
When compared with the putative gold standard, we observed significant variation for all collection methods. However, interindividual variability was much higher than the variability introduced by the collection method. Stability ICCs were high (≥0.75) for FIT tubes that underwent colorectal cancer screening procedures. The relative abundance of Actinobacteria (0.65) was an exception and was lower for different FIT tubes stored at 30°C (range, 0.41–0.90) and room temperature (range, 0.06–0.94).
Paper-based collection cards and different types of FIT are acceptable tools for microbiome measurements.
Our findings inform on the utility of commonly used fecal sample collection methods for developing microbiome-focused cohorts nested within screening programs.
This article is featured in Highlights of This Issue, p. 303
Evidence on the role of the human microbiome in the development of chronic diseases such as obesity, diabetes, and, potentially cancer, is growing (1–3). However, most of the current epidemiologic literature is based on cross-sectional studies that used diverse methods (4). Most established prospective cohorts did not collect fecal samples and repeated, prospectively collected samples are likely necessary for advancing understanding of the relationship between the microbiome and chronic disease development. Colorectal cancer screening programs, which provide screening through fecal tests such as the fecal immunochemical test (FIT), may offer great potential for establishing population-based cohort studies with multiple fecal specimens and epidemiologic data.
The impact of fecal sample collection methods on gut microbiome parameters has been investigated recently (5–10). Microbial populations in fecal samples collected using fecal occult blood tests and FIT have been found to be stable at room temperatures for up to 4 to 7 days, with similar microbial communities compared with samples collected without an additive and frozen immediately. For example, in one study conducted among 52 healthy volunteers in the United States, all fecal sample collection methods yielded microbial data that appeared relatively reproducible, stable, and accurate, when compared with the putative gold standard, and provided evidence that these collection methods can be employed for microbiome analyses in population-based studies (8).
However, to potentially inform on the establishment of international studies based in colorectal cancer screening programs, additional methodologic work is needed to test FIT and specimen collection cards used in other countries where different FIT methods and screening procedures are employed. Therefore, we evaluated microbial stability in fecal specimens stored at different temperatures and accuracy of microbiome metrics from two different specimen collection cards (used in Afghanistan) and four different FIT tubes used in ongoing international colorectal cancer screening programs (in France and most European countries, Morocco, Turkey, and Iran). In addition, we investigated the impact of colorectal cancer screening procedures and alternative shipping temperatures (e.g., summer and winter temperatures) on microbiome accuracy and stability ascertained from FIT samples.
Materials and Methods
Nineteen healthy participants from the International Agency for Research on Cancer (IARC) personnel were recruited in Lyon, France. Eligible participants were at least 18 years of age or older and had not taken antibiotics in the past 3 months. This study was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent, and the study was approved by the IARC Ethics Committee. On average, participants were 40.2 years old and had a body mass index of 22.7 kg/m2. Most participants were female (78.9%), had no weight variation in the past 6 months (73.7%) and reported having a regular bowel movement at least once per day (94.8%; Supplementary Table S2 in Supplementary Materials and Methods).
Fecal sample collection
At recruitment, participants were provided with a fecal sample collection kit including all materials needed for sample collection at the workplace or at home. Participants were asked to provide a fecal sample in the fecal collection containers (Sarstedt) by filling the scoop contained in each tube. The participant collected the fecal sample at the workplace (n = 9) or at home (n = 10) and returned it to the study coordinator within a few hours after collection. The study coordinator then delivered it to the laboratory for immediate processing. The participants completed a questionnaire that was used to obtain information on the time and date of sample collection, typical bowel movements, and general information.
Following the collection step, the fecal samples were manually homogenized and aliquoted for the different collection methods (Supplementary Fig. S1 in Supplementary Materials and Methods). For each participant, fecal samples were aliquoted into two cryotubes without solution (considered to be the putative gold standard, average weight per cryotube 208.1 mg), two GenSaver specimen collection cards (Ahlstrom-Munksjö) used in Afghanistan, two GenCollect specimen collection cards (Ahlstrom-Munksjö) used in Afghanistan, four OC-Auto Sampling tubes (Eiken Chemical) used in France, two Hemotrust tubes (Biosynex) used in Morocco, two One-Step FOB tubes (Padyabteb) used in Iran, and two Specimen Collection Container A tubes (Alfresa Pharma) used in Turkey. For specimen collection cards, a disposable spatula was used to smear a small portion of the homogenized feces on each window of the cards; the flaps on the cards were closed and then each card was placed in a separate biohazard bag with desiccant. The FIT tubes were filled following the instructions provided by the different colorectal cancer screening programs. Specifically, the FIT probes were dipped into the homogenized fecal specimen and returned to the FIT tubes and were then shaken to mix as instructed. All steps were performed by the same laboratory technician.
Fecal sample storage
The study samples are outlined in Supplementary Table S1. After sample collection and processing, the two cryotubes and one of each sample type were immediately stored at −80°C. One of the OC-Auto Sampling tubes was placed in a blue mailer for colorectal cancer screening and mailed to the Reference Centre for Epidemiology and Cancer Prevention in Piemonte (CPO Piemonte) laboratory within 24 hours after processing. To mimic mailing during seasonal temperatures in France, one of the OC-Auto Sampling tubes was stored at a winter temperature (4°C) for 7 days in a refrigerator, returned to room temperature for at least 4 hours, and then frozen at −80°C. The remaining OC-Auto Sampling tubes was stored at a summer temperature (30°C) for 7 days in a water bath, returned to room temperature for at least 4 hours, and then frozen at −80°C. The remaining FIT tubes (i.e., Hemotrust, two One-Step FOB and Specimen Collection Container A tubes) were stored for 7 days at room temperature and then frozen at −80°C. The remaining half of the specimen collection cards (i.e., GenCollect and GenSaver cards) remained at room temperature in a closed cupboard from the date of collection until the date of DNA extraction (average 70.2 days).
On arrival at the CPO Piemonte laboratory, the FIT tubes were processed using standard colorectal cancer screening procedures (11). After testing for occult blood, the FIT tubes were immediately closed with parafilm and shipped at room temperature to the Micalis Institute (INRAE/AgroParisTech) in Jouy-en-Josas, France for 16S rRNA gene profiling. Upon receipt of all the specimens at the same time, the samples were removed from the tube using pliers and a sterile pipette, transferred to a 1.5 mL Eppendorf tube, shaken, and then two 250 μL aliquots were transferred to a sterile tube and frozen at −80°C.
DNA extraction, real-time quantitative PCR, and 16S rRNA sequencing
DNA extraction, real-time PCR, and 16S rRNA-sequencing methods are described in the Supplementary Materials and Methods. Briefly, the samples were shipped to the Micalis Institute (INRAE/AgroParisTech) in Jouy-en-Josas, France and DNA was extracted with the PowerFecal DNA Isolation kit after the reception of the samples. The V3-V4 region of the 16S rRNA gene was PCR amplified and sequencing was performed with Illumina technology with MiSeq kit V2 2 × 250 bp. Quality control procedures and DNA concentrations available from the different sampling methods are described in the Supplementary Materials and Methods (Supplementary Tables S3 and S4).
Data were stored on secured servers at INRAE MAIAGE (Jouy-en-Josas, France) and IARC (Lyon, France). Sequencing data were analyzed with Find, Rapidly, OTUs with Galaxy Solution (FROGS) v3.1.0 (12). Briefly, this pipeline included a preprocessing step where reads were merged with Paired-End Read Merger (PEAR; ref. 13), dereplicated, and filtered according to their length, mismatches in primers with cutadapt (14), and N content. This step was followed by Swarm clustering (15) with an agglomeration distance of d = 3. Chimera detection was then performed using VSEARCH (16) before applying an operational taxonomic unit (OTU) abundance filter (OTUs <0.005% of the total abundance are discarded). The most abundant sequence of each OTU was then affiliated with 100% similarity with blastn against the SILVA v138 database (17). After processing, a total number of 8,239,370 sequences was found with an average number of 26,751 sequences per sample without any major deviation by sampling type. The average number of OTUs and genera per samples were 521 and 115, respectively. Two samples were discarded because of low sequencing depth and all remaining samples were rarefied to 7,144 reads per sample using R package phyloseq. Diversity metrics were then computed to represent the diversity of OTUs in each sample (alpha diversity: Shannon index, number of observed OTUs, and inverse Simpson index) and the differences between samples (beta diversity: Jaccard, Bray–Curtis, weighted, and unweighted UniFrac distances).
Statistical analyses were conducted using R, version 3.6.2 and the packages DESeq2, icc, phyloseq, and vegan (18–21). Descriptive characteristics of study participants were based on the questionnaire provided by the participants. We performed visualization using multidimensional scaling (MDS) plots and all samples from a given participant tended to group together. To assess the impact of the protocol on the alpha diversity, comparison of measures of alpha diversity between each fecal collection method was performed and a linear mixed-effects models with the collection method as fixed effect and the participant as random effect was fitted to the data and used to calculate least-squares means of the alpha diversity metrics. Mean diversities for the collection methods were then compared using Tukey HSD (honestly significant difference) tests. To estimate a distance-based coefficient of determination (R2) explained by participant and collection method from unweighted UniFrac, weighted UniFrac, Jaccard and Bray–Curtis distance matrices, permutational multivariate analysis of variance was performed [adonis function, vegan package, R] (21). We calculated intraclass correlation coefficients (ICC) using the variance components from a one-way ANOVA to evaluate the stability and accuracy of the different fecal collection methods. The ICCs were calculated on the basis of (i) the square root of the relative abundances of the three most dominant phyla (Actinobacteria, Bacteroidetes, and Firmicutes) and the most abundant genera (Faecalibacterium, Bacteroides, UCG-002, Subdoligranulum, Roseburia, Eubacterium eligens group, Blautia, Christensenellaceae R-7 group, Ruminococcus), which were present in at least 50% of fecal samples with relative abundance of ≥0.1%, (ii) our three alpha diversity metrics, and (iii) the first MDS axis, also called first principal coordinate (PC1), of our four beta diversity metrics. The first axis explained 8.2%, 15.8%, 17.6%, and 49.7% of the variability for Jaccard, Bray–Curtis, unweighted UniFrac, and weighted UniFrac distances, respectively. To calculate accuracy ICCs, we compared one replicate of samples without solution frozen immediately (considered as the gold standard), selected randomly, to one sample from each of the other collection methods for each participant. To calculate stability ICCs at different temperatures and procedures for each fecal collection method, we compared one sample frozen immediately with one stored at different conditions for each participant. The 95% confidence interval (CI) was estimated using the ICCest() function from the R ICC package with default option confidence interval = “Smith.”
Microbial alpha diversity by collection method
The Shannon index and inverse Simpson index values were, on average, highest for the immediately frozen samples without solution, (i.e., the putative gold standard) and the number of observed OTUs appeared highest in the specimen collection cards (Fig. 1). However, the different storage conditions did not seem to have a statistically significant impact on these alpha diversity metrics. From the linear mixed-effects model, we found that compared with the immediately frozen samples without solution [296, 95% CI (257–335)], the observed number of OTUs was significantly higher in GenSaver cards at room temperatures [347, 95% CI (308–387), P = 0.01] and in OC-Auto Sampling FIT tubes at −80°C [344, 95% CI (305–383), P = 0.03]. Compared with the defined gold standard [4.03, 95% CI (3.81–4.24)], the Shannon index was significantly lower in OC-Auto Sampling FIT tubes at −80°C [3.73, 95% CI (3.51–3.94), P = 0.005], One-Step FOB tubes at −80°C and at room temperatures [3.76, 95% CI (3.54–3.98), P = 0.03 and 3.49, 95% CI (3.27–3.72), P < 0.001, respectively], and in Specimen Collection Container A tubes at −80°C and at room temperatures [3.60, 95% CI (3.38–3.82), P < 0.001 and 3.70, 95% CI (3.48–3.91), P < 0.001, respectively].
Percent variability explained by participant and collection method
All ordinations suggested that the between participant variability was higher than the technical variability (Supplementary Fig. S2 in Supplementary Materials and Methods). In MDS plots, each point represents one microbiome sample for 19 participants and seven collection methods; all samples from a given participant tended to group together (as shown by ellipses, one per participant). In addition, based on four distance matrices (i.e., Jaccard, Bray–Curtis, unweighted UniFrac, and weighted UniFrac), the overall variability in diversity was largely explained by between participants (55%–79%) and only marginally by collection methods (4.8%–14.8%; Supplementary Fig. S3). For example, based on the Bray–Curtis distance matrix, the protocol variability accounted for roughly 6.9% of the overall variability whereas the biological variability accounted for 79.0%.
Relative abundance comparisons
At the phylum level, the distributions of relative abundances of each phylum were consistent for all the collection methods; samples were mainly represented by Firmicutes followed by Bacteroidetes (Fig. 2). At the genus level, when compared to the putative gold standard, the relative abundance of Faecalibacterium was greater in the other collection methods, especially in the OC-Auto Sampling FIT tubes. Consequently, when compared with gold standard, the relative abundance of other genera, such as Bacteroides or Blautia, was lower in other collections methods. However, there was also substantial interindividual variability at the genus level (Supplementary Fig. S4 in Supplementary Materials and Methods).
Accuracy compared with putative gold standard
Samples collected without solution and immediately frozen at −80°C were considered as the putative gold standard and compared with samples collected using other methods in different storage conditions using ICCs (Fig. 3; Supplementary Table S5 in Supplementary Materials and Methods). Accuracy ICCs were variable for the relative abundance of Actinobacteria (range, 0.28–0.83), Bacteroidetes (range, 0.24–0.84), Firmicutes (range, 0.26–0.82) as well as for the inverse Simpson index (range, 0.23–0.73) and the weighted UniFrac (range, 0.46–0.93). Specifically, for the relative abundance of Actinobacteria, ICCs were ≥75% for OC-Auto Samples tubes stored at 30°C and those that went through screening, for One Step FOB tubes stored at room temperatures and for all the specimen collection cards. For the relative abundance of Bacteroidetes, ICCs were ≥75% for OC-Auto Samples tubes that went through screening, for One-Step FOB tubes immediately frozen and for Specimen Collection Container A tubes immediately frozen. For the relative abundance of Firmicutes, ICCs were ≥75% for One-Step FOB tubes immediately frozen and for Specimen Collection Container A tubes immediately frozen. ICCs were ≥75% for PC1 scores of weighted UniFrac for all the methods, except for OC-Auto Samples tubes stored at 30°C and for Hemotrust tubes stored at room temperatures. At the genus level, we did not detect any links between the accuracy differences and the Gram positive or Gram negative (Fig. 4). As expected, genera that represent a phylogenetically narrow group of species, UGC-002 and Christenellaceae RT7 group, showed accuracies above 0.75. Conversely, ICCs were higher for observed OTUs (≥0.74) except for Hemotrust at room temperatures (0.31), OC-Auto Sampling tubes immediately frozen (0.66) and at 30°C (0.40), and One-Step FOB at room temperatures (0.60). Specimen collection cards showed the highest ICCs for the Shannon index (range, 0.79–0.88). ICCs were ≥75% for PC1 scores of Jaccard, Bray–Curtis and unweighted UniFrac distances for all collection methods. In addition, technical variability was also quantified on the basis of beta diversity distances between the two replicates of the putative gold standard and used as baseline to assess the accuracy of each method. For each participant, the distance between the putative gold standard and each collection method was computed (Supplementary Fig. S5 in Supplementary Materials and Methods). All of those distances were slightly higher than the baseline, suggesting that the effect of the collection methods exceeds the technical variability.
In each collection method, samples frozen immediately were compared with the samples stored in different conditions using ICCs (Fig. 5; Supplementary Table S6 in Supplementary Materials and Methods). When compared with specimen collection cards frozen directly, collection cards (i.e., GenCollect and GenSaver) kept at room temperatures showed high stability, with ICCs ≥89% for all seven metrics. Stability ICCs were generally lower and more variable in samples collected using Hemotrust tubes (range, 0.06–0.94). For One-Step FOB tubes, the confidence intervals were wide, indicating high variability in stability at room temperatures. Specimen Collection Container A tubes at room temperatures showed stability ICCs ≥75% except for the relative abundance of Actinobacteria (0.58) and Firmicutes (0.66). Stability ICCs were lower and inconsistent for OC-Auto Sampling FIT tubes stored at 30°C (range, 0.41–0.90). However, stability ICCs were higher for OC-Auto Sampling FIT tubes stored at 4°C (≥0.75) except for the number of observed OTUs (0.65). Stability ICCs were also high for OC-Auto Sampling FIT tubes that went through colorectal cancer screening procedures, with ICCs ≥75%, except for the relative abundance of Actinobacteria (0.65). Interestingly, at the genus level, the stability remained very high, especially for Gencollect and Genesaver. Of note, the Blautia genus, including the homoacetogen Blautia hydrogenotrophica, was significantly lower at room temperature (Fig. 2B; Fig. 6). In addition, UGC-002, but also Faecalibacterium, Roseburia, and Ruminococcus were also less stable when stored at 30°C.
We compared microbiome stability and accuracy across different fecal sample collection methods used in ongoing colorectal cancer screening programs. We found that the overall variability in diversity was largely explained by differences between participants and less by the collection method. In addition, accuracy and stability ICCs were generally very high for PC1 of beta diversity matrices, except for OC-Auto Sampling tubes stored at 30°C and for Hemotrust tubes stored at room temperatures. Accuracy measures were very inconsistent for the relative abundance of the three phyla and alpha diversity, and in particular, very low for the inverse Simpson. This highlights the importance of using one consistent method for study comparisons. Overall, microbial profile stability was very high for specimen collection cards and seemed generally acceptable for FIT tubes, except for Hemotrust tubes. Colorectal cancer screening tests did not impact microbiome stability in FIT tubes; however, exposure to summer temperatures (i.e., >30°C) did influence stability. These results are informative for the development of future population-based cohorts with fecal sample collection within colorectal cancer screening programs.
As shown in prior studies, microbial composition and diversity were largely explained by between-participants differences and only marginally by the collection methods (8, 9). Fecal specimen collection cards have been previously tested for microbial analysis and have shown moderate to excellent accuracy compared with the putative gold standard and stability at room temperatures (5, 9, 22–27). In this study, the specimen collection cards (i.e., GenCollect and GenSaver) stored at room temperature for 10 weeks showed excellent stability when compared with the immediately frozen cards, although other studies detected lower amounts of DNA among fecal samples from humans and animals collected on FTA cards (Whatman) after several weeks (25, 28). Also consistent with previous findings, we found that specimen collection cards tended to differ in bacterial taxa composition and OTUs than when using the putative gold standard (8, 22, 28), but we did not observed higher Shannon and inverse Simpson. Firmicutes, often spore formers, were the most represented phylum in samples collected with specimen collection cards, supporting the hypothesis that chemical cell lysis induced by the card matrix might be one explanation for these differences (22). Furthermore, Blautia genus, including the homoacetogen Blautia hydrogenotrophica, is also adequately represented in the collection cards, despite being highly sensitive to oxygen.
Fecal samples collected with FIT tubes have previously shown moderate to excellent accuracy compared with the gold standard and stability at room temperatures (10, 29, 30). However, the different types of FIT tubes in our study did not seem to perform equally. For the specimen cards and some of the FIT tubes, especially OC-Auto Sampling and Hemotrust tubes stored at −80°C, we detected significant differences in the relative abundance of phyla and genera with higher levels of Firmicutes and Faecalibacterium, including the anti-inflammatory butyrate producing Faecalibacterium prausnitzii, when compared with the gold standard, which support observations from a previous study (30). Similar to previous results (24, 29), when compared with the gold-standard, FIT tubes showed good to excellent accuracy for beta diversity metrics; however, some FIT tubes, including OC-Auto Sampling, Hemotrust and Specimen Collection Container A tubes, revealed lower accuracy for alpha diversity metrics, especially those stored for 7 days at room temperatures or at 30°C.
When compared with those directly frozen, Hemotrust tubes stored at room temperature for 7 days showed poor stability for alpha diversity metrics and for One-Step FOB tubes the confidence intervals were wide, indicating high variability in stability at room temperature. In this study, we detected different stabilities at room temperature for Eubacterium eligens and Roseburia, other butyrate producers from the human gut microbiota. As OC-Auto Sampling tubes seemed to be less stable at 30°C, the collection and shipping of samples during high temperatures might have an impact on gut microbiome composition. These variations in accuracy and mostly in stability between the types of FIT tubes might be due to differences in DNA-stabilizing and anti-microbial properties of the solution inside the tubes, impacting the stabilization of DNA, prevention of bacterial growth, and preservation of microbial profiles.
This study has several limitations. First, we included principally female, healthy participants, which might limit the inference of our results to the general population. However, previous studies have found that stability and accuracy for comparison of samples that were frozen immediately without solution to other fecal sample collection methods were similar between different populations (7, 8, 10). Second, we used 16S rRNA gene sequencing to characterize the microbial composition, while other profiling methods such as whole-genome shotgun metagenomics are becoming more commonly employed in high-income settings (31). However, 16S rRNA gene sequencing remains the most affordable method to study the gut microbiome diversity, especially in the context of large epidemiologic cohorts. In addition, in low-to-middle income countries, where sequencing technologies are not always available, these collection methods, and more specifically specimen collection cards, could be used to detect specific biomarker species that have been associated with diseases, such as Fusobacterium nucleatum or Parviromonas micra which have been associated with colorectal cancer development (4, 32). Finally, freezing procedures might have an impact on relative abundance of Gram-positive and Gram-negative bacteria (33, 34), and therefore, considering samples without solution frozen immediately as the gold standard method might be suboptimal. However, in the context of large population-based cohorts, immediate DNA extraction after defecation is likely unfeasible, and therefore, standardization of storage protocols is necessary.
Our study included many different collection methods, including two different specimen collection cards and four different FIT tubes, which are currently being used in ongoing colorectal cancer screening programs around the world. The different collection methods tested in this study have allowed us to highlight a number of important considerations for sample collections in large population settings such as acceptability from participants, processing safety, the volume of kit necessary, the storage logistics and cost. For instance, due to the rigidity of the tubes, the solution in OC-Auto Sampling tubes was not easy to sample safely; however, the amount of bacterial DNA obtained by qPCR indicated that sufficient material was available in these tubes. On the basis of qPCR data, the collected volume could be lowered from 500 to 200 μL to obtain sufficient DNA for several analyses and rendering it unnecessary to extract all of the liquid, which might be logistically challenging. In addition, storage of specimen collection cards is easier and cheaper than tubes, which might help low-to-middle income countries to develop infrastructure for microbiome research. Finally, because of the small size of the kits, the acceptance is expected to be higher in the general population, compared with the gold standard, where provision of a whole fecal sample could make subjects uncomfortable and reluctant to participate. Furthermore, in this study, stability was assessed over the course of several days to several weeks at room temperature, but also for different conditions directly reflecting settings in colorectal cancer screening (i.e., shipping at different temperatures reflecting seasonal variation, colorectal cancer screening procedures including mailing and occult blood detection test). This is the first study in which the impact of colorectal cancer screening procedures on fecal samples using FIT tubes has been demonstrated. Importantly, OC-Auto Sampling FIT tubes that went through colorectal cancer screening procedures and tests had good stability, opening opportunities for establishing prospective cohorts within screening populations.
In conclusion, our study supports previous findings indicating that microbial data obtained from FIT tubes and specimen collection cards are relatively stable and accurate and may be appropriate methods to collect fecal samples for gut microbiome analysis in population-based cohort studies. Furthermore, our findings suggest that opportunistic collection of fecal samples in FIT tubes after colorectal cancer screening is feasible, thereby permitting the potential establishment of cohorts within such screening programs. Because different collection methods and high temperatures impact the stability and accuracy, it is important for future investigators, in the context of the implementation of large-scale epidemiologic studies, to coordinate their efforts and follow standardized protocols to accurately compare the microbiome between sites, groups or countries and be able to pool microbial data.
M.-C. Boutron-Ruault reports personal fees from Mayoli-Spindler and Gilead outside the submitted work. No disclosures were reported by the other authors.
S. Zouiouich: Conceptualization, formal analysis, methodology, writing–original draft. M. Mariadassou: Formal analysis, methodology, writing–review and editing. O. Rué: Data curation, software, formal analysis, methodology, writing–review and editing. E. Vogtmann: Conceptualization, writing–review and editing. I. Huybrechts: Conceptualization, writing–review and editing. G. Severi: Conceptualization, writing–review and editing. M.-C. Boutron-Ruault: Conceptualization, writing–review and editing. C. Senore: Resources, writing–review and editing. A. Naccarati: Resources, writing–review and editing. G. Mengozzi: Resources, formal analysis, writing–review and editing. Z. Kozlakidis: Resources, formal analysis, writing–review and editing. M. Jenab: Writing–review and editing. R. Sinha: Conceptualization, writing–review and editing. M.J. Gunter: Conceptualization, formal analysis, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing. M. Leclerc: Conceptualization, resources, formal analysis, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing.
Where authors are identified as personnel of the IARC/World Health Organization, the authors alone are responsible for the views expressed in this article and they do not necessarily represent the decisions, policy, or views of the IARC/World Health Organization.
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors; this research was jointly funded by IARC and INRAE. The authors of this article would like to thank Christophe Lallemand from IARC and Laurence Brazeau from INRAE for their help with laboratory work.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.