Background & Summary

Fiber is naturally present in plants, fungi, animals, bacteria, and can also be synthetically made1,2. Dietary fibers are carbohydrates that resist digestion by the small intestine and have physiological health benefits to humans3,4. High fiber diets show a risk reduction for or amelioration of various illnesses such as constipation, obesity, diabetes, high cholesterol, heart disease, allergies, among others5,6,7,8,9. Furthermore, they are associated with improving mineral absorption, insulin responses, gut barrier permeability, immune system defense, production of beneficial metabolites, and inducing changes in the gut microbiome1,10. Fiber can modify the gut microbiome by affecting host secretions and transit stool time. It also serves as fermentative substrate for specific microbes and in turn, alters microbial activity more broadly (e.g., through cross-feeding and competition)11.

To understand the influence of dietary fiber on the gut microbiota, researchers have performed dietary fiber interventions among both healthy and unhealthy individuals12. These studies usually take a fecal sample from a person before and after their dietary change to assess shifts in the composition of the gut microbiome. Currently, the most common approach to assess microbial taxonomic composition is amplicon sequencing of a portion of the universal bacterial 16S ribosomal RNA (rRNA) marker gene13 because of the relatively low cost of next generation sequencing and the variety of tools available for bioinformatic processing. However, it still is challenging to access and harmonize such data to compare across studies, especially when its corresponding metadata is missing or hard to decipher14.

Motivated by the investigation of fiber-induced shifts in microbiota and the potential for re-analyzing sequencing data, we screened more than 1,500 abstracts and obtained data from 11 fiber intervention studies performed in healthy human subjects, for a total of 2,368 samples from 488 subjects. The purpose of publishing this data descriptor is to provide a detailed description of these valuable datasets, allow others to re-use the data that was carefully curated, and to promote data accessibility. Here, we present 1) the next generation 16S rRNA amplicon sequencing data which have been pre-processed and checked for quality scores, 2) its corresponding metadata which has been harmonized across studies, and 3) the operational taxonomic unit (OTU) tables that contain the number of reads per sample for each taxonomic unit. The sequencing data was primarily produced by Illumina platforms, but also includes 454 and Ion Torrent technologies. All metadata was curated to include similar columns across studies that are clearly defined in the metadata dictionary. The availability of scientific data and its corresponding metadata in comparable and reusable forms will allow researchers to re-analyze and synthesize these data in new ways to better understand the role of fiber in gut health.

Methods

Data collection and harmonization

We conducted a keyword search of published literature through the PubMed search engine (keywords: dietary, fiber, and microbiome) under the Best Match algorithm recommended by PubMed on May 9th, 2020. The search yielded 977 abstract hits from 2010 to 2020 (https://pubmed.ncbi.nlm.nih.gov/). We also searched through all the records available in the database of open-source microbial management site Qiita15 (https://qiita.ucsd.edu) on April 7th, 2020 and found 528 microbiome studies including human and animal studies. From both sources, each abstract was carefully read to select studies with fiber interventions in healthy humans that included 16S rRNA amplicon sequencing data from fecal microbial communities (n = 34). We excluded studies in animals and unhealthy humans (Fig. 1). Corresponding authors and first authors were contacted up to 4 times requesting their sequencing data and metadata when not publicly available. We were able to obtain 16S rRNA amplicon sequencing data and their corresponding metadata from 11 studies (Table 1). Data was shared to us via accession number16,17,18,19,20,21,22,23 or, if not publicly available, via virtual box. For the studies that did not make their datasets available at the time of publication (Dahl_2016_V1V2, Hooda_2012_V4V6, and Morales_2016_V3V4), we received consent to deposit their data under the BioProject ID: PRJNA891951 to the NCBI Sequence Read Archive24. For these studies, we recommend downloading the raw data through the SRA Run Selector Tool that allows users to see the Library Name. Each Library Name includes the study name followed by an underscore and the Sample ID. These Sample IDs are described in the metadata files created for this manuscript (see Data Records and Harmonization of datasets for more information). All studies included in this data repository complied with their relevant ethical regulations and have consent from their human participants to collect and share the data. For more information regarding guidelines for study procedure and trial registration numbers we refer our readers to the individual studies referenced in Table 1 and Table 2. The naming scheme for each of the studies included in this data collection is the following: Last name of the first author in the publication, followed by the year the study was published, and ending with the amplified region of the 16S rRNA bacterial gene (e.g., Liu_2017_V4).

Fig. 1
figure 1

Data collection workflow.

Full size image
Table 1 Data collected and available for eleven fiber intervention studies.
Full size table
Table 2 Summary of data collected by study.
Full size table

We provide Table 2 with a summary of each of the studies which includes: number of interventions per study, fibers used and their amounts, length of interventions, number of colletion timepoints, subjects and total samples. Because the metadata available was heterogeneous across studies, we performed harmonization across the datasets, so that common variables across studies could be easily identified. The metadata dictionary (Table 3) contains the definition for the data collected across studies.

Table 3 Metadata dictionary. Explains each column in the metadata files.
Full size table

To provide as much information on the dietary fiber interventions as possible, we investigated the specific fibers that were used in each study. Table 4 shows all the dietary fibers that were used in the interventions and their manufacturer or recipe (when available) including controls.

Table 4 Fibers and placebos given in the interventions.
Full size table

Sequencing processing

Individual studies used different methods for sequencing processing and bioinformatic pipelines, and such differences can influence the diversity and composition of microorganisms detected in a sample as well as the variation observed across samples25. Thus, to compare the sequences directly across studies, we obtained the raw sequencing reads for each study and then processed them in a similar manner.

First, we assessed the quality of the 16S rRNA sequencing data using FastQC software26 (version 0.11.8). The sequencing reads were cleaned from poor quality sequences using the Fastp program27 (version 0.20.0). The cleaned sequences were imported into the QIIME2 platform28 (version 2020.11.1), and primers were removed using Cutadapt29 plugin when necessary. We then denoised the reads using DADA230 plugin, obtaining an OTU table depicting the number of reads per sample for each taxonomic unit (Fig. 2).

Fig. 2
figure 2

Bioinformatics pipeline for data processing.

Full size image

Next, the taxonomic classification of the reads was also performed in the QIIME2 platform by training the SILVA31 (version 132_99_16S) and the Genome Taxonomy Database32 (GTDB; version bac120_ssu_reps_r95) databases to each respective study based on the primers that were originally used (Fig. 2). The SILVA database was used to remove chloroplast and mitochondrial DNA. Then, the cleaned reads were assigned to a final taxonomic group using the GTDB trained database. Reads that were not classified at least to the phylum level were removed from the analysis; sequences were classified to the finest level when possible (e.g., species and/or strain). The sequencing processing and taxonomic classification was performed with both the forward and reverse reads when paired-end data was available. We also repeated the analyses with only the forward reads, and found that both gave very similar results. We provide the OTU tables obtained with both procedures (e.g., baxter_OTU_table_paired_reads.tsv and baxter_OTU_table_forward_reads.tsv) to allow the reader to choose either option for further analysis.

Data Records

The following data have been deposited in the Figshare33 repository: 1) The compressed 16S rRNA sequencing reads (.fastq.gz) containing the amplicon data that were quality filtered as described above; 2) the metadata files per study in tab-delimited format (.txt) describing their corresponding samples serving as a reference to help identify and sort the DNA sequences by different metrics (e.g., timepoint, treatment, individual, etc.); 3) the OTU tables with taxonomic assignment per study (.tsv) presenting the number of reads per sample for each taxonomic unit. As mentioned in the Data collection section and in Table 1, the raw reads for the studies mentioned here can be found in publicly available databases16,17,18,19,20,21,22,23. For the studies that did not make their datasets available prior to this publication (Dahl_2016_V1V2, Hooda_2012_V4V6, and Morales_2016_V3V4), we received consent to deposit their data under the BioProject ID: PRJNA891951 to the NCBI Sequence Read Archive24.

Technical Validation

Data integrity

For quality assurance of the sequencing reads, we utilized the FastQC tool26 as it provides quality control statistics such as sequence length, per base quality scores, and adapter contamination34. We used the Fastp software27 to ensure data integrity: we removed low quality reads from all datasets, only keeping reads with an average quality score of 30, the average score of 25 was chosen in only two occasions (Rasmussen_2017_V1V3 and Liu_2017_V4) because read counts dropped dramatically with a higher threshold (−average_qual 30 or 25); we discarded sequences shorter than 100 bp (−length_required 100) to remove small sequences that could not complete 16S rRNA amplicon fragments. We only had to remove adapter contamination from one study (Deehan_2020_V5V6) using the detection of adapter correction tool in Fastp (−detect_adapter_for_pe). When paired-end data was available, we enabled base correction in overlapped regions of paired reads (−correction). When corrupted data, having characters that did not belong to the sequencing reads, was found (Hooda_2012_V4V6) we discarded those samples (n = 10).

Harmonization of datasets

To ensure the datasets were comparable, we converted sequencing reads from all studies into .fastq extension files (when necessary). Furthermore, we followed the same pipeline using consistent software and versions (Fig. 2) and cross-validated our results by visually inspecting the sequences after each clean-up step using Geneious prime (version 2020.2.4; https://www.geneious.com). For instance, after removing primers from reads using the Cutadapt plugin in QIIME2, we extracted the reads and imported them into Geneious to verify that sequences had been properly trimmed. Moreover, to ensure clarity and consistency of metadata across datasets, we created a metadata dictionary (Table 3) to explain the data type (categorical, numerical, text, etc.). In most cases, the metadata files available for the studies did not follow a consistent report of variables. For example, there was a big difference in how the timepoints were described (e.g., “before”/“after” vs “post”/“pre” vs numeric) and in most instances the fiber type and grams of fiber were not included. To remedy this, we carefully curated the data collected per sample across studies to have similar naming schemes.