Evaluating gut microbiota profiles from archived fecal samples

The BCSN trial is a comparative effectiveness research trial randomizing 140.000 persons aged 50–74 years to be screened for CRC either with flexible sigmoidoscopy (FS) or iFOBT [17]. About 70.000 individuals have been invited for iFOBT screening and will deliver iFOBT samples every other year for 10 years. The sampling kits for iFOBT are mailed to the participants and collected by return mail with a shipping time of 2–6 days. The stool samples are collected on plastic sticks designed to catch about 10 μg stool and then stored in 2 ml buffer containing HEPES buffer (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid), BSA (Bovine serum albumin) and sodium azide. The participants were asked to mail the sample to the lab as soon as possible. On the day of arrival to the laboratory, the samples were analyzed with an immunochemical test for human blood (globin) (Eiken Chemicals Ltd., Tokyo, Japan) and the perforated iFOBT tube was frozen at − 80 °C immediately thereafter. We randomly selected 50 anonymized iFOBT samples among the participants in the BCSN trial for this study. The samples have not previously been thawed.

The NORCCAP study was a randomized flexible sigmoidoscopy screening clinical trial of 100,210 individuals aged 50 to 64 years from the populations in two Norwegian counties (Oslo and Telemark). The trial was performed during 1999–2000 (55–64–year age group) and in 2001 (50–54–year age group) [15, 16]. The intervention group received flexible sigmoidoscopy and delivered stool samples for iFOBT. To-date the trial has about 4800 stool samples stored at − 30 °C. 50 feces samples randomly selected from NORCCAP participants with normal sigmoidoscopy were included in this study. Stools were sampled in 20 ml vials by the screening participants at home and kept for 1–7 days in their home freeze (− 20 °C) before delivery to the screening centre when attending FS screening. Samples were transported in the 20 ml vials with an outer transport tube. Most participants had less than 30 min travel to the screening centre, but partial thawing during this transport is likely. At the screening centers, stools were stored at − 20 °C for up to 1 week before transport for up to 2 h in insulated transport boxes to a central − 30 °C storage room with alarm systems. Some samples may have been partially thawed to provide material for non-microbial research purposes.

The collection of iFOBT and archived samples was approved by the Regional Committees for Medical and Health Research Ethics in South-Eastern Norway (2011/1272 and 2010/3087 A, respectively). An additional ethical approval for the feasibility study was not required since all samples were anonymized and the purpose of the study was method related, as stated by the Norwegian regional ethics committee (Regional Committees for Medical and Health Research Ethics in South-Eastern Norway, ref. 2015/9).

In the automatized analysis of iFOBT, the same instrument needle extracts liquid from all samples in a consecutive order with intermediary washing. To investigate if this may result in cross-contamination of the samples, a test sample series alternating between 200 ng/μl human DNA (Roche #11691112001) and water was prepared. This sample series was run on the desktop instrument OC-Sensor Diana (Eiken Chemicals,Tokyo, Japan). A qPCR measurement of human DNA beta-globin in the 40 water samples and two positive controls were analyzed in two replicates in 25 μl total TaqMan universal PCR master mix (Thermofisher Scientific, Waltham, USA) reactions using 5 μl template [23].

The iFOBT tubes are perforated during iFOBT analysis. Evaporation during the freezing process, sublimation during storage or condensation during thawing of these “open” tubes might increase or decrease sample volume indicating that contamination of samples due to these processes might be possible. To asses if samples quality could be diminished by such processes we tested for changes in sample volume by freezing 50 perforated iFOBT tubes with and without parafilm and parafilm with rubber bands. We compared the weight of the 50 tubes before and after 6 months in the freezer (− 80 °C), 3 h after thawing.

All samples were processed in Lysing Matrix E tubes with silica beads. Sample input was about 10 μl, approximately 10 mg, for the solid archived fecal samples and 500 μl for the iFOBT samples. Dry weight of feces varies in the iFOBT samples, however, we estimate that about 2.5–5 μg feces were used in the protocol. PBS (phosphate-buffered saline) buffer was added to a total liquid volume of 1.1 ml prior to mechanical lysis on a FastPrep 24 and centrifugation at 400×g for 1 min. Nucleic acids were extracted on an automated platform, QIAsymphony, using a standard nucleic acid extraction protocol that included pretreatment with ProteinaseK and a final elution in 85 μl. To account for the level of microbiological contamination from the environment and reagents an extraction-negative control (ENC) containing all the reagents except for sample material was processed in each batch of samples. Two additional negative control samples were included in the PCR, one with water and one with 10 ng/μl human DNA. Due to experimentally encountered inhibition of PCR in the archived samples, these were diluted 1:5 before used as template for PCR. All other samples were used directly in PCR without dilution.

A two-step PCR protocol was used. The primers S-D-Bact-0341-b-S-17 and S-D-Bact-0785-a-A-21 [24] were first used to amplify a 16S rDNA in the V3-V4 region, to produce a 464 nucleotide (nt) long amplicon. PCR was performed in 20 μl reaction volumes with 1× Phusion Master Mix, 0.25 μM each 16S primer and 2 μl template under the following conditions; initial denaturation 98 °C for 30 s and then 27 cycles of 98 °C for 10 s, 62 °C for 30 s and 72 °C for 15 s, before a final extension at 72 °C for 10 min and storage at 10 °C. PCR cleanup was done using partly modified PERFORMA DTR V3 96-well Short Plates and QuickStep™2, 96 Well PCR Purification kit protocols. All samples PCR products were tested on 1% agarose gels using 5 μl sample. Then in the second round of PCR, indexes were added. The index PCR was also performed in 20 μl volumes using 1× Phusion Master Mix, 0.5 μM each index primer, and 1 μl of template DNA from the previous PCR, diluted 1:100.

Sequencing of the amplicons was performed on an Illumina MiSeq (Illumina) desktop sequencer using V3 chemistry 2 × 300 cycle kits. This protocol has shown high reproducibility and repeatability and very little contamination [25].

Sequencing quality filtering and processing were carried out using the MiSeq SOP and the Mothur pipeline (v.1.36.1) [26, 27]. To ensure high-quality data for analysis, sequence reads containing ambiguous bases, homopolymers > 8 bp, more than one mismatch in the primer sequence, or sequences under the default per base quality score were removed. Assembled reads > 460 bp in length, singletons and chimeric sequences identified with the UCHIME algorithm [27] were excluded from the analyses. The high-quality assembled sequences (contigs) were aligned to the Silva 16S rRNA database (v.119) [28] and clustered into operational taxonomic units (OTUs) at a cut-off value > 98% in a closed reference OTU approach. OTUs were calculated at distance 0.02, and alpha diversity (Shannon index and inverse Simpson index) was calculated per sample. The number of assembled sequences for each sample was rarefied to 2000 to minimize the impacts of uneven sampling. Inverse Simpson’s index is weighted on dominant species, whereas the Shannon index assumes all species are represented [29]. To identify differences in species diversity between the groups, a two-way ANOVA and post hoc comparisons (Tukey Honest test of Significant Differences - TukeyHSD) of the observed number of OTUs, Inverse Simpson diversity and the Shannon index were used. Beta diversity (i.e., the variation in community composition between microbiota samples) was calculated between the sample types using the Bray–Curtis dissimilarity index [30]. Dissimilarities between sample groups were tested using permutational analysis of variance (PERMANOVA) implemented in the vegan (v. 2.5–2) R package and the “adonis” function. Differentially abundant OTUs were identified using the zero-inflated Gaussian model (fitZig), implemented in the metagenomeSeq package [31, 32]. The community composition data was transformed using the “decostand” function and divided by the margin total. Principal components analyses were done using the prcomp function in R. Heatmaps of the log transformed OTUs (for OTUs with more than 10 in counts) were produced using hierarchical clustering and Euclidean distance. To improve visualization only OTUs with a log sum > 20 for all samples, and > 2 for the fresh samples were shown. The clustering of these heatmaps were indistinguishable from heatmap with all OTUs included.

All of the ENC and negative controls, but also four iFOBTs and 14 archived samples did not produce sufficient aligned reads to calculate diversity.

Potential transfer of DNA from sample to sample during iFOBT analysis was tested with qPCR. 79 of the 80 qPCR measurements were negative. i.e. no contamination. One sample had a cycle threshold (Ct) value of 40.7, also considered a negative result. The positive controls were positive on the qPCR (Ct values < 30).

The weight difference of 50 perforated iFOBT tubes before freezing (383.6 g) and after 6 months in the freezer (383.7 g) was 0.1 g or 0.03%. The weight difference of 50 perforated iFOBT tubes packed in parafilm to prevent change in volume before freezing (384,5 g) and after freezing (384,9 g) was 0.4 g or 0.1%. Some condensation on the tube surface was observed.

In total, 4.2 million 16S rDNA contigs were assembled from the sequencing data. The fresh samples were sequenced separately, with a larger fraction of the sequencing pool per sample than the other sample groups, thus producing more reads, contigs and OTUs (Fig. 1a, Additional files 1, 2 and 3). Archived samples produced less OTUs than fresh and iFOBT samples (Table 1). Gel electrophoresis of the 16S rDNA PCR product from the archived fecal samples showed increased PCR efficiency when the DNA solutions were diluted 5-fold compared to undiluted DNA, signifying PCR inhibition. There were no signs of PCR inhibition in iFOBT samples.

The mean, maximum and minimum numbers of contigs (assembled reads) assigned to OTUs per sample group are shown in Table 1. The ENC and negative control samples produced very few OTUs. 96% of the iFOBT, 81% of the archived and 100% of the fresh samples passed a minimum relative abundance of 200 OTUs. The minimum relative OTU abundance chosen as a threshold was based on the sequencing depth and distribution of OTU counts (Additional file 1).

We compared the diversity in iFOBT, fresh and archived samples using observed OTUs, and the Inverse Simpson and Shannon indexes from rarefied OTU tables (Fig. 1a, b, c and Additional file 4). Fresh and iFOBT samples were similar in mean diversity while the archived samples had statistically significant less diversity compared to iFOBT (P-value = 0.027 for Inverse Simpson and P-value = 0.06 for Shannon) and fresh (P-value < 0.001 for both Inverse Simpson and Shannon) in the ANOVA and TukeyHSD test. The Bray-Curtis dissimilarity index showed that the microbiota in iFOBT, archived and fresh samples differ (Fig. 1d and Additional file 5) and the PERMANOVA analysis showed that this difference is significant (P-value < 0.001, F = 1.4587). The first and second principal components showed overlapping community composition (Fig. 2).

The microbiota profiles differed between fresh and iFOBT vs. archived fecal samples (Fig. 3). More than 62% of the archived fecal samples clustered together, while the fresh and iFOBT samples were intermixed (Fig. 3). In particular, the OTUs 1 to 6 (rightmost vertical clade), including Lachnospiraceae and Ruminococcaceae family OTUs, were of high abundance in fresh and iFOBT samples, whereas OTU 7 and 11, representing the Peptostreptococcaceae and Ruminococcaceae family, were abundant in archived samples.

The zero-inflated log-normal mixture model identified 11 OTUs that significantly (P-value < 0.05) differed between archived and iFOBT samples (Table 2 and Additional file 6). The OTUs of the Ruminococcaceae family (OTU4, 14 and 9) and OTU5 Lachnospiraceae were significantly less abundant in archived samples. Peptostreptococcaceae (OTU11) and Clostridiaceae (OTU18) were more abundant in archived fecal samples.

The mean alpha-diversity did not differ significantly between samples that had been directly frozen and samples that were frozen after 48 h in room temperature (Observed OTU count P-value = 0.18, Inverse Simpson P-value = 0.65 and Shannon P-value = 0.94) (Fig. 4a and Additional file 4). Dissimilarities (Bray-Curtis) between sample pairs from the same individual with different room temperature storage time, was lower than between samples from different individuals (Fig. 4b). Unsupervised clustering of the log-transformed OTU relative abundance supports similarities between sample pairs (Fig. 4c). The log-normal mixture model identified no significant differences in relative abundance (Additional file 6) between samples stored in room temperature for 48 h and samples directly frozen (Additional file 7).