Rectal Swabs from Critically Ill Patients Provide Discordant Representations of the Gut Microbiome Compared to Stool Samples

Rectal swabs have been proposed as potential alternatives to stool samples for gut microbiome profiling in outpatients or healthy adults, but their reliability in assessment of critically ill patients has not been defined. Because stool sampling is not practical and often not feasible in the intensive care unit, we performed a detailed comparison of gut microbial sequencing profiles between rectal swabs and stool samples in a longitudinal cohort of critically ill patients. We identified systematic differences in gut microbial profiles between rectal swabs and stool samples and demonstrated that the timing of the rectal swab sampling had a significant impact on sequencing results. Our methodological findings should provide valuable information for the design and interpretation of future investigations of the role of the gut microbiome in critical illness.


Expanded Methods
Clinical cohort: For comparisons with healthy controls, we also included samples from donated stool for fecal microbiota transplant from 15 healthy adult donors.

Sample Collection:
From consented critically-ill patients, we collected rectal swabs and/or stool samples at three time intervals starting at the time of intubation and continuing up to 10 days if the patient remained in the ICU (defined as baseline [days 0-2], middle [days [3][4][5][6] and late intervals of follow-up [days 7-10]). Rectal swabs were collected as per routine clinical practices (i.e. placing the patient in a lateral position, entering the tip of the 6" cotton tip swab in the rectal canal and rotating gently for 5 seconds), unless clinical reasons precluded movement of the patient (e.g. severe hemodynamic or respiratory instability). Stool samples were collected when available, either by taking a small sample from an expelled bowel movement (before cleaning of the patient and disposal of stool), or through a fecal management system (rectal tube) placed for management of diarrhea and liquid stool collection.
Stool samples or rectal swabs were placed in sterile collection tubes and then capped, labeled and stored at -80°C until further processing.

Clinical Data Recording:
Detailed data was obtained from the electronic medical record on the day of enrollment including: 1) Demographic information: age, sex, height, weight, body mass index 2) Laboratory results: white blood cell count, hemoglobin, platelets, serum bicarbonate, arterial pH, partial pressure of oxygen, partial pressure of carbon dioxide. Cycle conditions were 98˚C for 30s, then 33 cycles of 98˚C for 10s, 57˚C for 30s, 72˚C for 30s, with a final extension step of 72˚C for 2 min. We combined triplicates and purified with the AMPure XP beads (Beckman) at a 0.8:1 ratio (beads:DNA) to remove primer-dimers. We quantitated eluted DNA on a Qubit fluorimeter (Life Technologies). We performed sample pooling on ice by combining 20 ng of each purified band. For negative controls and poorly performing samples, we used 20 microliters of each sample. We purified the sample pool with the MinElute PCR purification kit. The final sample pool underwent two more purifications -AMPure XP beads to 0.8:1 to remove all traces of primer dimers and a final cleanup in Purelink PCR Purification Kit (Life Technologies). We quantitated the purified pool in triplicate on the Qubit fluorimeter prior to preparing for sequencing. We prepared the sequencing pool according to instructions by Illumina, with an added incubation at 95˚C for 2 minutes immediately following the initial dilution to 20 picomolar. We then diluted the sequencing pool to a final concentration of 7 pM + 15% PhiX control. Amplicons were sequenced on the Miseq platform.

Analytics:
16S Sequence quality control: Sequences from the pooled sequencing run were demultiplexed into individual sample/replicate fastq files. Each fastq file was then processed through the Center for Medicine and the Microbiome (CMM) custom modular read QC pipeline that was configured to perform the following steps: low complexity filtering, QV trimming, Illumina sequencing adapter trimming, and 16S primer trimming. Low complexity filtering utilized NCBI BLAST's dustmasker (6). Reads with greater than 80% low complexity regions were filtered. QV trimming and filtering utilized the FASTX Toolkit. The trimming threshold was QV>25 from the 3' end, and length>125 bp after trimming. The subsequent filtering threshold was >25 across >95% of the read. The Illumina sequencing adapter and 16S primer trimming was performed with cutadapt (7).

16S Clustering and annotation:
Paired sequences with forward and reverse reads passing the QC filtering and trimming steps were then mated (end aligned and a consensus sequence computed) using the make.contigs function of Mothur. Consensus sequences were screened to limit the overlap mismatch to no more than 20%. The maximum number of N's allowed in the overlap was 4 and the minimum overlap was required to be greater than 25bp. Consensus sequences passing screening were then passed through CMM's 16S clustering and annotation pipeline, a Mothur-dependent wrapper designed to streamline and automate the execution of the following Mothur steps in version v.1.39.1: unique.seqs, align.seqs, screen.seqs, filter.seqs, second uniq.seqs, pre.cluster, chimera.uchime, remove.seqs, classify.seqs, dist.seqs, cluster, make.shared, and classify.otu.
Mothur output files were then reformatted to sample x category (taxonomic levels or operational taxonomic units [OTU] at 97% sequence similarity). For downstream statistical modeling and analyses, we utilized taxonomic tables at the genus level, with taxonomic assignments performed using a naïve Bayes k-mer classifier in conjunction with the Ribosomal Database Project (RDP) 16S rRNA gene sequences (8). Taxa table edits: The taxa table was filtered for low abundance taxa (relative abundance, <0.005%) and singletons. We did not filter clinical samples for any taxa detected in the negative control samples. Five samples were subjected to repeat PCR amplification due to poor performance in initial reactions, and were then removed from analyses as duplicates. Samples with very few reads (<200) were excluded from analyses (n=5).

Statistical Analysis
We calculated descriptive statistics of baseline characteristics and performed nonparametric comparisons (Wilcoxon test for continuous and Fisher's exact test for categorical variables) using the R software (R Foundation for Statistical Computing, 2016). We performed ecological analyses of alpha-diversity (richness-Shannon), beta diversity (Bray-Curtis dissimilarity index), and taxonomic relative abundance at the phylum and genus level with R vegan package (9). Beta-diversity comparisons with permutation analysis of variance (Permanova at 1000 permutations) were visualized with Principal Coordinates Analyses.
Taxonomic abundance differences between groups were calculated following transformation of abundance data with the additive log ratio, performed for the top 4 phyla (Firmicutes, Bacteroidetes, Actinobacteria and Proteobacteria) and the top 20 genera (10).
In order to assess for longitudinal changes of alpha-diversity over time as well as to account for the effects of potential confounders on the associations between sample type and gut microbiota profiles, we constructed a set of multivariate models as described below.
As confounders, we considered potential variables that may impact gut motility and bowel content in critically-ill patients, as well as having potential associations with the gut microbiome: 1. Early Enteral Nutrition (defined as initiation of gastric or enteral feeds within 48hrs of ICU admission).
2. SOFA score on continuous scale, as a marker of severity of illness.
3. Presence of rectal tube, as an indicator of severe diarrhea requiring a fecal management system.
4. Obesity, defined as a BMI>30, which has been associated with gut microbiome profiles, but for practical purposes may also affect our ability to perform rectal swabbing especially in the subset of patients with morbid obesity. 5. Clinical diagnosis of sepsis, associated with administration of broad-spectrum antibiotics, which can cause diarrhea, but may also affect the gut microbial communities.
We then constructed the following multivariate models: