The intestinal microbiota provide vital contributions to gastrointestinal (GI) and systemic health. The composition of bacterial species and associated metabolites in the human GI tract have been associated with multiple disorders, such as obesity and inflammatory bowel diseases (Ley et al., 2006; Flint et al., 2012; Hooper et al., 2012; Tremaroli and Backhed, 2012; Kabat et al., 2014). Importantly, these bacteria not only serve as indicators for disease but also as targets for therapeutic intervention. Many of the recent developments to our understanding of the intestinal microbiome have been the result of the application of next generation DNA sequencing methods targeting bacterial 16S rRNA genes for phylogenetic and taxonomic assessments of bacterial community membership (Liu et al., 2012). Because of the difficulty in collecting intestinal contents from human subjects, feces are typically used as the source to identify and characterize the intestinal microbiota. Although microbial composition differs along the length of the digestive tract (Eckburg et al., 2005; Stearns et al., 2011), fecal bacteria were shown to have a greater diagnostic value for patients with irritable bowel syndrome than those located in the gut mucosa (Rangel et al., 2015). Therefore, methods used to sample and examine the microorganisms in human fecal material should be optimally designed to ensure precision, accuracy, and a lack of significant bias introduced either by sample collection, homogenization, DNA extraction and sequencing, or data analysis.

Numerous studies have examined the effects of storage conditions, DNA extraction methods, and PCR parameters on the detection and quantification of bacteria in human stools. Several reports concluded that bacterial community composition was not affected by incubating human feces at room temperature for 24 h or 4°C for 48 h or freezing immediately after collection at either -20 or -80°C (Wu G.D. et al., 2010; Cardona et al., 2012; Carroll et al., 2012; Hang et al., 2014; Fouhy et al., 2015; Tedjo et al., 2015). The application of nucleic acid stabilizing reagents (such as RNAlater) have shown promise, but also significant limitations. Although the microbiota of stools stored in RNAlater were stable for 7 days at room temperature (Flores et al., 2015) and were subject-specific over time (Voigt et al., 2015), the incorporation of RNAlater for fecal sample storage decreased DNA quality (Dominianni et al., 2014) and yields (Gorzelak et al., 2015). The microbial diversity (Dominianni et al., 2014) and proportions of certain bacterial taxa were also affected (Nechvatal et al., 2008; Flores et al., 2015; Gorzelak et al., 2015).

Different commercial DNA extraction kits were found to be capable of recovering sufficient quantities of DNA (Scupham et al., 2007; Nechvatal et al., 2008; Salonen et al., 2010; Wu G.D. et al., 2010; Sergeant et al., 2012; Yuan et al., 2012; Hang et al., 2014; Kennedy et al., 2014; Santiago et al., 2014; Vishnivetskaya et al., 2014; Wesolowska-Andersen et al., 2014; Wagner Mackenzie et al., 2015), but yielded proportional changes in identified taxa (Salonen et al., 2010; Wu G.D. et al., 2010; Sergeant et al., 2012; Yuan et al., 2012; Hang et al., 2014; Kennedy et al., 2014; Santiago et al., 2014; Vishnivetskaya et al., 2014; Wesolowska-Andersen et al., 2014). Notably, physical shearing of cells by bead-beating improves DNA recovery (Salonen et al., 2010; Yuan et al., 2012; Santiago et al., 2014; Vishnivetskaya et al., 2014) and generally increases the proportions of Gram-positive bacteria that are difficult to lyse by enzymatic methods (Scupham et al., 2007; Sergeant et al., 2012).

PCR parameters, including the DNA polymerase, oligonucleotides, annealing temperature, and barcoding strategy varied the outcome in bacterial surveys (Frank et al., 2008; Engelbrektson et al., 2010; Wu J.Y. et al., 2010; Sergeant et al., 2012; Brandariz-Fontes et al., 2015; Franzen et al., 2015; Sinclair et al., 2015; Tremblay et al., 2015; Zheng et al., 2015; D’Amore et al., 2016; Derakhshani et al., 2016). The number of PCR amplification cycles altered estimates of low abundance bacterial taxa (Gonzalez et al., 2012) but did not change conclusions of the overall community structure (Wu J.Y. et al., 2010). The impacts of DNA sequencing platforms, including Ion Torrent, 454 pyrosequencing, and Illumina, on the outcomes of assessments of bacterial composition in stool samples were minor (Salipante et al., 2014; Sinclair et al., 2015; Tremblay et al., 2015; D’Amore et al., 2016). In contrast, bioinfomatics approaches can result in significant differences (Bokulich et al., 2013; Majaneva et al., 2015).

Despite these seemingly exhaustive comparisons of microbial analysis techniques, stool preparation methods have not been well-tested. Although the collection of small aliquots of stool is sufficient to perform bacterial analysis (e.g., 200 mg or less), it is not clear how well the bacterial species are distributed throughout the fecal contents. Therefore it is not known whether sampling from different spatial locations in the stool without homogenization is adequate to obtain data that are representative of the stool as a whole. This is particularly relevant for human studies wherein people consumer different foods throughout the day. Moreover, there has not been a comparative analysis of different fecal preparation methods, and less labor-intensive methods have often been favored. In preparation for optimal processing of stool samples from a study of the effects of dietary resistant starch in prediabetic participants, we measured the intra- and inter-individual variations in stool samples with and without freezing and with and without homogenization to evaluate the extent to which the bacterial proportions change as a result of processing steps.

We found that the inter-individual variation in fecal microbiota between subjects was clearly greater than any difference introduced either by freezing or homogenization. Intra-individual comparisons showed that there was no method that reproducibly altered the alpha and beta diversity of the bacteria in the same way among stools from all subjects. Instead, our findings strongly indicate that there are certain families and genera that are more vulnerable to over/under representation in stools based on the processing method and hence might be more susceptible to potential biases caused by the manner in which fecal samples are collected, stored, and processed.

Our findings are in agreement with other studies showing that current protocols for multiplexed, 16S rRNA gene sequence assessments of fecal bacterial diversity are sufficiently robust that inter-individual variation in fecal microbiota is not overcome despite large differences in stool preparation and DNA sequencing and analysis methods (Salonen et al., 2010; Wu G.D. et al., 2010; Carroll et al., 2012; Dominianni et al., 2014; Hang et al., 2014; Santiago et al., 2014; Wesolowska-Andersen et al., 2014; Fouhy et al., 2015; Tedjo et al., 2015; Voigt et al., 2015; Wagner Mackenzie et al., 2015). Importantly, biases introduced by sample preparation and analysis methods are minor compared to the large changes in intestinal bacterial composition caused by diet, antibiotics, and disease (Willing et al., 2011; Perez-Cobas et al., 2013; David et al., 2014; Conlon and Bird, 2015; Arora and Bäckhed, 2016; Matijasic et al., 2016). The greater impact of these biologically, as opposed to technically, induced differences shows that human fecal bacteria can be monitored for species diversity and richness, even with procedural distinctions.

Consistent with this premise, the relative abundances of the dominant bacterial phyla were not consistently affected by freezing or homogenization. Although the proportions of Gram-positive bacteria and the corresponding Firmicutes to Bacteroidetes ratios tended to increase following homogenization (either by blending or pneumatic mixer), no phylum was repeatedly, significantly enriched (or depleted).

At other taxonomic levels, we found that processing methods can have modest effects on study outcomes. By testing for repeated, significant differences that could be attributed to bacterial families and specific genera within those families, we identified six genera susceptible to variation in stools from the majority (≥4) of the six human subjects tested. Most notably, Faecalibacterium was the only microbial taxon for which the relative abundance was significantly affected by one or more processing method in all individuals. Homogenization tended to increase the proportions of this genus, although no single processing method was responsible for these effects. Faecalibacterium is a butyrate-producing genus in the Firmicutes phylum that has been associated with good intestinal health (Miquel et al., 2013). Members of this genus are anti-inflammatory and secrete proteins that reduce activation of NF-κB pathways in rodent and cell culture models (Quevrain et al., 2016). These bacteria are negatively associated with inflammatory bowel diseases and other inflammatory conditions in humans (Sokol et al., 2008; Eppinga et al., 2016; Matijasic et al., 2016). Faecalibacterium was also found in lower proportions in individuals with Type 2 diabetes (T2D; Tilg and Moschen, 2014) and increased in a diet that promoted insulin sensitivity in obese subjects (Haro et al., 2016b).

Similar to Faecalibacterium, the levels of Streptococcus, another genus in the Firmicutes phylum, were increased in stools after homogenization. Streptococcus is commonly used in the production of fermented foods (Veiga et al., 2014) and is a prominent member of the human small intestine microbiota (Zoetendal et al., 2012). It was also recently reported that proportions of this genus are reduced in stools of individuals with T2D on a macrobiotic diet (Candela et al., 2016). Notably, because the homogenization methods applied here lowered the levels of Oscillospira, another genus associated with human health (Konikoff and Gophna, 2016), it should not be expected that homogenization will increase all physiologically relevant members of the Firmicutes phylum.

Other bacteria of medical and ecological significance were also consistently altered by homogenization such that they were either more (Bifidobacterium) or less (Bacteroides and Parabacteroides) abundant. Bifidobacterium is commonly applied as a probiotic and is an indigenous inhabitant of the large intestine of both infants and adults (Bottacini et al., 2014; Jost et al., 2015). Bacteroides and Parabacteroides are also prominent in the intestine and appear to be pivotal, diet-responsive members of the human microbiota (Wu et al., 2011; Del Chierico et al., 2016; Haro et al., 2016a). Taken together, these findings show how the fecal processing method could influence conclusions on the impacts of dietary interventions on bacterial genera with functions important for human health.

Presently, it remains to be shown whether these significantly changed taxa are either more or less vulnerable to homogenization or if there are other factors that influenced the outcomes. One possibility is that certain bacteria are more sensitive to lysis and DNA shearing/degradation and, therefore, were lost with the application of homogenization for sample processing. This effect would lower the detectable numbers of those sensitive taxa and, conversely, increase the proportions taxa able to withstand the sample processing method. Alternatively, homogenization could provide benefits by dislodging bacteria from macromolecules or aggregates and distributing them for increased access during DNA extraction. This change would enable more accurate and reliable detection of those taxa that would otherwise be missed by less rigorous processing approaches. The latter possibility is supported by our finding that homogenization of frozen stools, either by blender or pneumatic mixer, resulted in significant reductions in intra-sample variation. These findings are in agreement with homogenized fecal samples that were examined either by quantitative PCR (Gorzelak et al., 2015) or shot-gun metagenomics (Wesolowska-Andersen et al., 2014). Notably, the outcomes were not solely due to freezing because of the lack of change in bacterial diversity between the fresh and frozen unhomogenized samples.

In summary, fecal sample homogenization is a useful method for characterizing the human intestinal microbiome and reducing intra-sample variation. When at all possible, this method should be used instead of random sampling without homogenization because of the within-stool variation in microbial composition. Moreover, the intra-sample variability was similar for homogenization by blender versus pneumatic mixer. This result is robust evidence both are valid methods for homogenizing stool in research protocols. Because the blender method took only 2 min, in comparison to the more labor-intensive shaker-and-beads method (10–30 min), blending offers a time-saving approach to homogenization. Ultimately, improved understanding of fecal microbiota will be gained from the application of constructed bacterial community “standards” and improved knowledge on the localization of different bacterial taxa at micro- and macro-scales within luminal contents. Because of the sensitivity of certain taxa to be enriched or diminished as a result of fecal processing, comparisons across studies should be viewed with caution and instead more general outcomes provided.

Y-HH performed experiments, analyzed data and wrote the manuscript; CP designed the experiments and revised the manuscript; AR performed experiments; MK, RM, and ER designed the experiments and revised the manuscript; MM designed the experiments and wrote the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.