An Ambient-temperature Collection and Stabilization Device Performs Comparably to Flash-frozen Collection for Stool Metabolomics in Infants

Background: Stool metabolites provide essential insights into the function of the gut microbiome. The current gold standard for collection and storage of stool samples for metabolomics is ash-freezing at -80°C which can be inconvenient and expensive. Ambient temperature collection of stool is more practical, however no available methodologies adequately preserve the metabolomic prole of stool. A novel sampling kit (OMNImet.GUT; DNA Genotek, Inc.) was introduced for ambient temperature collection and stabilization of feces for metabolomics; we aimed to test the performance of this kit vs. ash-freezing. Methods: Stool collected from an infant’s diaper was divided into two aliquots: 1) ash-frozen and 2) stored in an OMNImet.GUT tube at ambient temperature for 3-4 days. Samples from the same infant were collected at 2 different time points to assess metabolite changes over time. Subsequently, all samples underwent metabolomic analysis by liquid chromatography – tandem mass spectrometry (LC-MS/MS). Results: Paired fecal samples (ash-frozen and ambient temperature) from 16 infants were collected at 2 time points (n= 64 samples). Similar numbers of metabolites were detected in both the frozen and ambient temperature samples (1126 in frozen, 1107 in ambient temperature, 1064 shared between sample types). Metabolite abundances were strongly correlated between collection methods (median Spearman correlation Rs=0.785 across metabolites). Hierarchical clustering analysis and principal component analysis showed that samples from the same individuals at a given time point clustered closely, regardless of the collection method. Repeat samples from the same individual were compared by paired t-test, separately for the frozen and OMNImet.GUT. The number of metabolites in each biochemical class that signicantly changed (p<0.05) at timepoint 2 relative to timepoint 1 was similar in ash-frozen versus ambient temperature collection. Changes in microbiota modied metabolites over time were also consistent across both methodologies. Conclusion: Ambient temperature collection and stabilization of stool in the OMNImet.GUT device yielded comparable metabolomic results to ash freezing in terms of 1) the identity and abundance of detected biochemicals 2) the distinct

time points (n= 64 samples). Similar numbers of metabolites were detected in both the frozen and ambient temperature samples (1126 in frozen, 1107 in ambient temperature, 1064 shared between sample types). Metabolite abundances were strongly correlated between collection methods (median Spearman correlation Rs=0.785 across metabolites). Hierarchical clustering analysis and principal component analysis showed that samples from the same individuals at a given time point clustered closely, regardless of the collection method. Repeat samples from the same individual were compared by paired t-test, separately for the frozen and OMNImet.GUT. The number of metabolites in each biochemical class that signi cantly changed (p<0.05) at timepoint 2 relative to timepoint 1 was similar in ash-frozen versus ambient temperature collection. Changes in microbiota modi ed metabolites over time were also consistent across both methodologies.
Conclusion: Ambient temperature collection and stabilization of stool in the OMNImet.GUT device yielded comparable metabolomic results to ash freezing in terms of 1) the identity and abundance of detected biochemicals 2) the distinct metabolomic pro les of subjects and 3) the biochemical signature of microbiome development over time. This method potentially provides a more convenient, less expensive home collection option for stool metabolomic analysis.

Background
Small-molecule biochemicals, or metabolites, are critical signaling molecules that mediate hostmicrobiome communication and dynamics [1]. Bioactive metabolites produced by gut microbiota, such as short-chain fatty acids, aromatic amino acid metabolites, and bile acids, have been implicated in numerous human health conditions, including obesity [2], insulin resistance [3], cardiovascular disease [4], autism-spectrum disorder [5], and Parkinson's disease [6]. Due to the heterogeneity of human genetics, microbiome composition, and nutrient and other exposures, these studies have made only limited inroads into the richness and diversity of human gut metabolites. In addition to providing fundamental insight, these molecules may offer avenues to therapeutic strategies and biomarker discovery. Consequently, attention in the eld is increasingly turning to metabolomics -the discipline of measuring metabolites in a comprehensive and unbiased manner [7] -as a powerful tool to elucidate how the gut microbiome exerts its effects on human health and disease [8,9,10].
Despite this growing interest, the practical challenge of collecting fecal samples from human volunteers has constrained the application of metabolomics in gut microbiome research. In general, immediately freezing samples at -80°C has long been viewed as the optimal approach for preserving metabolites by quenching enzymatic activity, hydrolysis, oxidation, and other degradative processes ( [11]. Yet, in the case of human feces, obvious considerations of donor privacy and convenience have driven a strong demand for at-home collection. However, storage and shipping of frozen at-home collected samples can be inconvenient for participants and prohibitively expensive for researchers, pushing the need for ambienttemperature storage options. To ll this need, several research groups have recently sought to repurpose currently available sampling devices, such as DNA-stabilizing tubes and fecal immunochemical test (FIT) tubes [12,13], for metabolomic analysis. However, the abundant detergents, buffers, salts, and other additives in these tubes render them incompatible with liquid chromatography and mass spectrometry (LC/MS), the technique of choice for metabolomic analysis. As a result, use of these devices signi cantly distorted the metabolomic pro le of stool samples relative to the gold-standard ash-freezing methodology [12,13].
Recently, OMNImet.GUT tubes (DNA Genotek, Inc.) were launched onto the market as a stool collection and stabilization kit developed speci cally for metabolomics. We aimed to test the delity with which samples collected and stored in OMNImet.GUT tubes recapitulate the metabolomic pro le of those collected by the gold-standard method, i.e., ash-freezing at -80°C. To accomplish this, we conducted a unique hospital-based eld test in which matched samples (n=16 donors, each donor sampled at two timepoints) were both ash-frozen and stored in OMNImet.GUT tubes. This study design enabled us to compare the two sample types directly (within-donor comparisons) as well as to compare groups of donors within each sample type (inter-donor comparisons) to assess whether the OMNImet.GUT samples would yield the same biochemical ndings as the ash-frozen samples.

Methods
As part of a longitudinal microbiome study conducted at the Inova Health System (IRB# 15-1945), infants who were admitted in the neonatal intensive care unit (NICU) were enrolled shortly after birth after written consent from their parent and had serial stool samples collected. Stool was collected from the infant's diaper and divided into two ~ 500 mg aliquots: 1) ash-frozen, i.e., immediately stored in an Eppendorf tube without preservatives at -80 °C, and 2) stored in an OMNImet.GUT tube (DNA Genotek, Inc.) at ambient temperature for 3-4 days prior to freezing at -80 °C. Samples from the same infant were collected at 2 different time points, at a minimum of 5 days apart, to assess changes in the metabolomic pro le over time. OMNImet.GUT devices (product number ME-200) were obtained from DNA Genotek (Ottawa, ON).
To prepare feces for metabolomic analysis, frozen samples were lyophilized while OMNImet.GUT samples were dried in a Genevac evaporator. All dried samples were weighed and then resuspended at a 50:1 (50 µL deionized water for every 1 mg of feces weight) ratio for homogenization as previously described [13]. The homogenates were subjected to automated biochemical extraction and analysis by liquid chromatography and high-resolution tandem mass spectrometry (LC-MS/MS) on Metabolon's Global Platform, as previously described [14,15,16]. Raw data were extracted, peak-identi ed, and processed by Metabolon using proprietary software [17,18,19]. In brief, metabolites -were identi ed by comparison to library entries of puri ed standards or recurrent unknown entities. Metabolon maintains a dynamic and proprietary biochemical reference library of more than 4,500 known metabolites (based on authenticated standards) and more than 2,000 novel metabolites (without an identi ed chemical structure); each library entry contains the retention time/index (RI), mass to charge ratio (m/z), and spectral data (including MS/MS fragmentation). Biochemical identi cations are based on three criteria: retention index within a narrow RI window of the proposed identi cation, accurate mass match to the library ± 10 ppm, and the MS/MS forward and reverse scores between the experimental data and authentic standards. The MS/MS scores are based on a comparison of the ions present in the experimental spectrum to the ions present in the library spectrum. Three types of controls were included: a pool of small portions of each experimental sample serving as a technical replicate throughout the platform run; extracted water samples (process blanks); and a cocktail of standards spiked into every analyzed sample allowing instrument performance monitoring.
After identi cation, metabolites were quanti ed by peak integration. To analyze the Frozen and OMNImet.GUT sample sets separately, the data were scaled to a median of 1 for each biochemical in that sample set. Missing values, if any, were imputed with the observed minimum for that particular metabolite in that sample set. The data were natural log-transformed prior to statistical analyses including paired t-test and principal component analysis (PCA). To directly compare the Frozen and OMNImet.GUT sample sets, each set was normalized to a bridging control sample, which was prepared by pooling the extracts from the Frozen samples and was analyzed in n = 4 technical replicates in parallel with both of the experimental sets. In each set, metabolites were retained if present in at least 3 of the 4 bridging samples; for each metabolite, its raw peak areas were divided by the median of the raw peak areas across the bridging samples. After combining the two sample sets, only metabolites present in both sets were retained; for each metabolite, its missing values were imputed with its minimum value. The resulting merged dataset was natural log-transformed prior to statistical analyses including paired t-test, PCA, hierarchical clustering analysis (HCA) using complete clustering with Euclidean distance and Spearman correlation.
The datasets generated and/or analyzed during the current study are available in the Metabolights repository, (https://www.ebi.ac.uk/metabolights/MTBLS201).

Results
A total of 16 infants provided paired fecal samples ( ash frozen and ambient) collected at 2 different time points (mean time between samples = 23 days, range 5-44 days), for a total of 64 samples. See Table 1 for demographic and clinical data for subjects. were shared between the sample types, i.e. 94.5% of the metabolites detected in ash-frozen samples were also detected in the OMNImet.GUT samples (Fig. 1). These biochemicals were classi ed into 10 super-pathways (all shared between both sample types) and 121 sub-pathways, of which 117 were shared between both sample types. 62 metabolites from 7 super-pathways and 38 sub-pathways were only found in frozen samples; 43 metabolites from 8 super-pathways and 24 sub-pathways were only found in ambient temperature samples (Supplemental table 1). Metabolite abundances were strongly correlated between collection methods, with a median Spearman correlation Rs = 0.785 across metabolites that were detected in at least 50% of samples of each type (Fig. 2, Supplemental Table 2).
PCA and HCA of the 1064 metabolites detected in both frozen and OMNImet.GUT samples showed that the samples from the same individuals at a given time point clustered closely, regardless of the collection method ( Figs. 3 and 4), suggesting that the effect of OMNImet.GUT collection on the individual metabolomic pro les was small compared to the differences among individual patients. Both HCA and PCA also showed a partial separation between samples from the two timepoints, also regardless of collection method, consistent with time-dependent metabolomic changes that occurred across multiple patients and were captured by both collection methods. To further investigate this time-dependent signature, repeat samples from the same individual were compared by paired t-test, separately for the frozen and OMNImet.GUT collections. The number of metabolites in each biochemical class that signi cantly changed (p < 0.05) at timepoint 2 relative to timepoint 1 was similar in ash frozen versus ambient temperature collection (Fig. 5).
To gain better insight into how fecal sample collection and storage for the frozen and OMNImet.GUT samples impacted biological changes related to timepoint 1 and timepoint 2 in the same individual, PCA analysis was performed separately for the frozen and OMNImet.GUT dataset. PCA showed major separation of samples from the two time points on Component 1 for both frozen (Fig. 6A) and OMNImet.GUT samples (Fig. 6B), suggesting substantial changes in metabolomic pro les between the time points mediated by developmental changes in the infants. Together, these observations suggest that time point might be a prominent contributor to variation in metabolomic pro le in both the frozen and OMNImet.GUT tube samples. Importantly, regardless of collection method, the frozen samples and OMNImet.GUT tube-collected samples yielded strikingly similar PCA results, suggesting that the method of fecal collection did not greatly in uence the biochemical signature.
With a clear distinction of metabolomic pro les between timepoint 1 and timepoint 2, we next determined if metabolites known to be modi ed by the gut microbiome were contributing to the time dependent changes. Consistent with an altered gut microbiome, changes in speci c aromatic amino acids metabolites, which are derived from microbial metabolism of these amino acids, were observed between timepoint 1 and timepoint 2 in both frozen and OMNImet.GUT tube samples. Altered levels of phenyllactate (PLA), tyramine, indole and 3-indoxyl sulfate between timepoint 1 and timepoint 2 were seen (Fig. 7A). Furthermore, differences in primary and secondary bile acids (which are also modi ed by the gut microbiota) were seen between timepoint 1 and 2 in both collection methods (Fig. 7B). Overall, these results may represent altered microbial activity between the two developmental time points, captured by both collection methods.

Discussion
This is the rst study to our knowledge to directly compare the delity of the new OMNImet.GUT device, designed for ambient temperature collection and stabilization of stool metabolites, with the current gold standard of ash freezing of stool. We found that metabolite analysis from the OMNImet.GUT device yielded comparable results to ash freezing in terms of 1) the identity and abundance of detected biochemicals, 2) the distinct metabolomic pro les of subjects, and 3) the biochemical signature of gut microbiome development over time.
While the optimal method for metabolic pro ling of stool is likely extraction within 1 hour of collection [20], this method is out of reach in the vast majority of circumstances. It is therefore accepted that the next best method and more practical "gold standard" is ash-freezing of stool below − 20 °C [21].
However, ash-freezing of samples is inconvenient, expensive and not practical for large epidemiological studies. A handful of prior studies have attempted to address the feasibility of ambient temperature collection and storage for metabolite analysis using different collection methods. Loft eld et al. [13] performed metabolomic analysis of samples collected into 95% ethanol, fecal occult blood test cards (FOBT) and fecal immunochemical test tubes (FIT) relative to matched ash-frozen samples from 18 healthy adult volunteers. The 95% ethanol tubes performed the best, providing detection of 89% of the metabolites detected in ash-frozen samples, while collection into either FIT or FOBT led to a signi cant decline in both metabolite coverage and correlation to ash-frozen samples. Another study [12] of three healthy adults investigated whether a DNA-stabilizing device intended for ambient storage of stool for metagenomic analysis (OMNIgene.GUT) would also be suitable for metabolomics. These authors found a median Spearman correlation < 0.5 between ash-frozen samples and samples immediately frozen in the metagenomic analysis tubes (i.e., without room temperature storage), indicating that this collection method signi cantly distorted the metabolomic pro le relative to ash-freezing. In addition to these low correlations, Lim et al. [12] also found that the differences among a single individual's samples across collection methods were greater than those among the three individuals' samples when the collection method was held constant, which precludes the use of these metagenomic tubes for discovery of metabolomic distinctions among individuals. Finally, Wang et al. [22] also compared 95% ethanol and OMNIgene.GUT with ash-freezing. Consistent with Lim et al., in this second study OMNIgene.GUT also performed poorly, providing detection of only 34.3% of the metabolites detected in ash-frozen samples and with a median interclass correlation coe cient (ICC) to ash-frozen of only 0.21. Surprisingly, however, in the study by Wang et al. [22], 95% ethanol also showed a median ICC of only 0.35, possibly due to practical di culties inherent in using a "homemade" sampling methodology rather than a commercially available kit. Together, these studies demonstrated the lack of a commercially available, metabolomics-compatible methodology for ambient fecal collection and storage prior to the development of OMNImet.GUT collection device. In contrast, our study found that ambient temperature collection in the OMNImet.GUT device maintained both the identity and abundance of detected biochemicals (94.5% overlapping coverage, median Spearman R = 0.795) and the distinct metabolomic pro les of the subjects.
Moreover, we examined the metabolic signature re ective of microbiome development over time including microbiome modi ed aromatic amino acid metabolites and bile acids [23,24]. We found that sample collection and storage in the OMNImet.GUT device accurately retained the changes in gut-microbial metabolites such as phenyllactate (PLA) and indole over time compared to the gold standard methodology of ash freezing.

Limitations
Although this eld testing of the OMNImet.GUT device showed highly comparable results to the gold standard of ash freezing, this is not precisely the same as home collection of stool as trained lab staff conducted the aliquoting of samples. Therefore, further validation of the OMNImet.GUT device may be needed in samples collected at home in the by lay study participants. All subjects in this study were infants and the metabolite content of stool likely differs in older children and adults due to microbiome and diet differences. Further validation may be valuable in stool collected from adults.

Conclusions
Ambient temperature collection and stabilization of stool in the OMNImet.GUT device yielded comparable metabolomic results to ash freezing in terms of 1) the identity and abundance of detected biochemicals 2) the distinct metabolomic pro les of subjects and 3) the biochemical signature of microbiome development over time. This potentially provides a more convenient and less expensive home collection option for stool metabolite analysis, with many potential uses, including in longitudinal childhood microbiome research.     Principal Component Analysis (PCA) of samples collected at 2 timepoints. 6A: The PCA was generated using the 1126 metabolites detected in the Frozen samples. 6B:The PCA was generated using the 1107 metabolites detected in the OMNImet.GUT samples.