Prediction of high fecal propionate-to-butyrate ratios using 16S rRNA-based detection of bacterial groups with liquid array diagnostics

The human gut microbiome affects the health of the host through a variety of mechanisms, including the fermentation of nondigestible carbohydrates that escape small intestinal digestion and absorption [1]. The end products of this fermentation, short-chain fatty acids (SCFAs), serve a variety of functions, including acting as the main energy source for colonocytes [2], enhancing the intestinal epithelial barrier [3], regulating mucus production [4], modulating inflammatory responses [5], inducing apoptosis in colon cancer cells [6], regulating blood pressure [7], mediating gut–brain cross-talk [8], regulating glucose homeostasis [9] and lipid metabolism and adjusting satiety levels [10].

In healthy adult populations, it is estimated that the three major SCFAs (acetic, propionic and butyric acid) accumulate in a 3:1:1 molar ratio [11–13]. A deviation in such proportions, with a significant decrease in butyrate levels, has been observed in people consuming a diet high in protein and low in carbohydrates [14]. Butyrate production is solely dependent on the intake of nondigestible fiber, whereas the major propionate producers, such as Bacteroidetes, metabolize peptides as well, thus leaving propionate levels unaltered [15]. Lower butyrate levels have also been linked to a slower fecal transition time, and both are associated with a higher colonic pH, which in turn promotes the production of propionate [16]. A low pH environment protects against the overgrowth of pathogens [17]; thus, in this context, an increase in the propionate-to-butyrate (P:B) ratio may indicate a vulnerable gastrointestinal state.

A deviant ratio in favor of propionate was proposed to act as a diagnostic marker for irritable bowel syndrome (IBS) [18]. Increased levels of this acid (but not butyrate) were also reported in overweight and obese people [11], individuals with an increased risk of type 2 diabetes [19], patients with Alzheimer's disease [20] and those with nonalcoholic fatty liver disease [21]. Additionally, a reduced butyrate (but not propionate) concentration was observed in people with a high risk of stroke [22]. Although the evidence linking disproportionately low levels of butyrate and/or high levels of propionate with various diseases is expanding, routine diagnostic measurement for SCFA content remains challenging, mainly due to the high volatility of SCFAs and the complex sample clean-up procedure [23,24].

Here we aimed to infer SCFA levels by targeting a limited number of key bacteria using a novel quantitative PCR (qPCR)-compatible method, liquid array diagnostics (LAD) [25], circumventing the need to utilize GC-based methods. A LAD test targets variable regions within the 16S rRNA gene and allows the detection of up to 25 bacterial markers in a single tube. We focused on the P:B ratio, a single variable with the potential of providing an indication of functional dysbiosis in clinical samples. The analytical strategy followed in this study is outlined in Figure 1.

Materials & methods

Fecal samples & gDNA extraction

In total, 115 anonymized adult fecal samples, biobanked at Genetic Analysis AS, Oslo, Norway (research biobank no. 4071), were used for this study. Samples were collected and anonymized in accordance with the ruling by the Norwegian Regional Committee for Medical and Health Research Ethics (reference no. 2010/3209).

All fecal samples were stored at -40°C prior to gDNA extraction or GC sample prep. The gDNA of all fecal samples was extracted using a mag midi kit (LGC Biosearch Technologies, Hoddesdon, UK) following the steps suggested by the manufacturer. Genomic extracts were further analyzed with LAD, PacBio single-molecule real-time technology (PacBio, CA, USA) [26], or whole-genome shotgun sequencing (Illumina, CA, USA).

Measurement of SCFA content with GC

The SCFA content of 115 fecal samples was measured with GC (TRACE™ 1310 with autosampler; Thermo Fisher Scientific, MA, USA). Fecal samples were diluted in water (1:10) to a total volume of 1500 μl and then homogenized for 2 × 40 s at 1800 rpm using a Fastprep^®-96 (MP Biomedicals, CA, USA). After a gentle spin, 300 μl of supernatant was transferred to a new tube, to which 300 μl of internal standard was added. The internal standard consisted of 0.4% formic acid and 2 mM 2-methylvaleric acid. The samples were centrifuged at 13,000 rpm for 10 min. Subsequently, 300 μl of supernatant was transferred to spin columns (0.2-μm filters) and centrifuged at 10,000 rpm for 5 min. The solution that passed the membrane was transferred to GC vials for SCFA measurement. An internal standard (1 mM 2-methylvaleric acid) was used as a reference for sample-to-sample normalization of results. A total of nine samples did not pass quality control by failing to produce a measurement on acetic acid. Given that this acid is the most volatile, its depletion was taken as an indication that the samples were compromised; therefore, they were excluded from further processing. In addition, a sample erroneously handled during laboratory work was removed. The P:B ratio of the 105 remaining samples was computed and used for further data analyses.

PacBio sequencing of 16S rRNA gene

Ninety-six samples randomly selected from 115 with SCFA content determined by GC were sent for PacBio Sequencing (Full-Length 16S Amplification SMRTbell® Library Preparation and Sequencing) at the Norwegian Sequencing Center (Oslo, Norway) (www.sequencing.uio.no). The first round of amplification was performed using the in-house 16S primers (GA-map^® forward primer 5′-TCCTACGGGAGGCAGCAG-3′, GA-map^® reverse primer 5′-CGGTTACCTTGTTACGACTT-3′, both protected by patent US20110104692 A1) tailed with universal sequences, as recommended in the PacBio protocol.

On average, there were approximately 28,480 sequencing reads per sample with an average length of 1175 nucleotides. The reads sharing at least 0.97 sequence identity were clustered into operational taxonomic units (OTUs) using the open-source metagenomics tool VSEARCH [27]. Excluding singletons, clustering resulted in 598 total OTUs, and the average number of OTUs per sample was 184. The OTU read counts were centered log ratio-transformed [28] (after the addition of one pseudo-read count) prior to further processing. One of the samples was unfortunately mislabeled when sent for sequencing, and two of the 96 samples did not meet the GC criteria (no measured acetate) and were thus excluded from the downstream analysis.

Identification of bacterial targets through combined partial least squares & linear discriminant analysis modeling

Centered log ratio-transformed OTU read counts from 93 samples were used as input for a combined partial least squares and linear discriminant analysis algorithm, with the aim of selecting variables (OTUs) to act as markers for classifying samples with normal versus high P:B ratios [29]. The aim was to correctly identify and classify the samples with the highest ratios, as they represent the deviation from the norm. The border between the two types of samples (normal vs high ratio) was allowed to go as low as possible without losing model prediction accuracy. The highest accuracy was reached at a P:B border of 2.5, with 37 OTUs acting as predictors, spanning 15 dimensions (leave-one-out cross-validated sensitivity = 90%, specificity = 99%, positive prediction rate = 90% and negative prediction rate = 99%). These OTUs were subsequently considered targets for LAD assay development.

Probe design for LAD

Eight-mer sequences containing a C at their 3′ ends – shared between only 16S in silico amplicons of target organisms – were computed using the in-house TNTProbeTool [30]. These sequences were considered the 3′ end segments of potential LAD labeling probes (LPs). Probes had to have a minimum melting temperature (Tm) of 70°C (computed by the nearest neighbor method) hybridizing to the target group and a maximum Tm of 30°C hybridizing to a nontarget group. The final LP sequences did not contain a C at their 3′ ends. In this way, the presence of the corresponding bacterial target would ensure that they became extended with a quencher-labeled ddCTP.

A reverse complementary reporter probe (RP) was designed for each of the LPs. The RPs were designed with a fluorophore tag on their 5′ ends, ensuring proximity to the quencher in duplexes harboring a 3′-labeled, RP-complementary LP. Duplexes containing the same fluorophore were designed with varying lengths to produce distinct temperature-dependent signals on the same qPCR channel of detection. The quenching effect of a longer and more stable duplex is observed as a dissociation curve with a higher Tm. The DNA duplex Tms were calculated using the web-based OligoAnalyzer Tool^™ 3.1 (Integrated DNA Technologies, Inc., IA, USA) based on the nearest neighbor method.

Basic Local Alignment Search Tool searches (nucleotide collection [nt/nr] database with Blastn) with each OTU sequence as query were performed to infer OTU taxonomy. Initially, 21 probes were designed, covering all 37 OTUs. However, six of the probes, targeting 11 OTUs (Coprococcus eutactus, Alistipes indistinctus, Bacteroides eggerthii, [Clostridium] spiroforme, Ruthenibacterium lactatiformans, [Clostridium] glycyrrhizinilyticum), failed to produce a signal; therefore, they were excluded from the assay.

Because of sequence similarity between F_3_1 and R_12_1 LPs, which were designed to detect Dorea longicatena and Fusicatenibacter saccharivorans, respectively, it was impossible to keep them in a single test tube, as this would risk producing double signals when only one target was present. Therefore the test was split into two tubes and the number of probes was divided proportionally between them (eight probes in group 1, seven probes in group 2). We used ROX_12_1 as an RP for both LPs. A list of final probes and their Tms and target species are presented in Table 1.

Generation of templates for LAD labeling reaction

Genomic DNA from 71 available samples was PCR-amplified. The SCFA content of these samples had been measured in previous steps; however, nine of the samples were not PacBio-sequenced. Each PCR reaction, with a total volume of 25 μl, consisted of 5 μl bacterial lysate (catalog no. 01-02-00500; Solis BioDyne, Tartu, Estonia), 3.75 U HOT FIREPol^® DNA Polymerase (catalog no. 01-02-00500; Solis BioDyne), 1× B1 buffer (catalog no. 01-02-00500; Solis BioDyne), 2.5 mM MgCl₂ (catalog no. 01-02-00500; Solis BioDyne), 0.2 mM dNTPs (catalog no. 18427088; Thermo Fisher Scientific) and 0.2 μM in-house primers (GA-map^® forward primer 5′-TCCTACGGGAGGCAGCAG-3′, GA-map^® reverse primer 5′-CGGTTACCTTGTTACGACTT-3′, both protected by patent US20110104692 A1). The amplification was carried out using an Applied Biosystems Veriti^™ Thermal Cycler (Thermo Fisher Scientific) with an initiation period of 15 min at 95°C followed by 30 cycles of 30-s denaturation at 95°C, 30-s annealing at 55°C and 80-s elongation at 72°C, ending with a final step of elongation at 72°C for 7 min. PCR products were then treated with 2.7 U Exonuclease I (catalog no. M0293L; New England Biolabs, MA, USA) and 7.36 U recombinant shrimp alkaline phosphatase (catalog no. M0371L; New England Biolabs) and set for incubation at 37°C for 10 min, followed by 15 min at 80°C to inactivate the enzymes.

Single nucleotide extension of LPs & melting curve analysis with LAD

A total of 10 μl of PCR products treated with Exonuclease I–shrimp alkaline phosphatase (14.5–25.6 ng/μl) were used as templates for LP labeling. The labeling reaction also comprised LPs at a final concentration of 0.1 μM (biomers.net GmbH, Ulm, Germany), 1× buffer C (catalog no. 01-06-00500; Solis BioDyne), 1 mM MgCl₂ (catalog no. 01-06-00500; Solis BioDyne), 7.5 U HOT TERMIPol^® DNA Polymerase (catalog no. 01-06-00500; Solis BioDyne) and 0.96 μM ddCTP-DYQ660 (catalog no. NU-850–660Q; Jena Bioscience, Jena, Germany). The reaction was performed in a PCR instrument with an initiation step at 95°C for 12 min followed by 40 cycles of denaturation (96°C for 20 s) and annealing/elongation (68°C for 40 s).

Following labeling, a mixture of RPs and MgCl₂ was added to the reactions to achieve final concentrations of 0.01 μM and 5 mM, respectively. Reagent S, available from Inland Norway University of Applied Sciences (Hamar, Norway), was also added to a final concentration of 1%. The melting curve analysis (31–85°C) was performed using a CFX96 qPCR instrument (Bio-Rad Laboratories, Inc., CA, USA).

The extraction of peaks and determination of positive signals were performed as described by Hiseni et al. [25] with a slight modification. Prior to extracting the signals, the fluorescence measurements within each channel were centered with the purpose of minimizing the range of measurements across wells at any given temperature. Next, the baseline within each channel was corrected (flattened) by subtracting the centered values of each sample from the average no template control centered values. As an ultimate step, for group 1 samples only, a further correction of FAM and CY5 baselines was performed by subtracting the values from one another (FAM = FAM - CY5 and CY5 = CY5 - FAM).

Bioinformatics evaluation of probe specificity

OTU sequences (PacBio) were used as subjects to check for sequences complementary to 3′ C-labeled probes. A search for the occurrence of probes, allowing one mismatch anywhere along the sequence was performed (excluding the probe 3′-C). The intention of this step was to prove that probes precisely targeted the intended bacteria. OTU sequences containing sites complementary to probe sequences were considered to act as ‘labeling templates’. The read counts of all such sequences were considered in silico signals, which were then used to compute the correlation with real LAD signals.

Whole-genome shotgun sequencing

A total of 24 samples were sent for whole-genome shotgun sequencing at the Norwegian Sequencing Center. Libraries were prepared using a Nextera^™ DNA Flex Library Preparation Kit (Illumina) following the protocol recommended by the manufacturer. Samples had different SCFA levels that spanned well the P:B values. One of the samples failed the GC quality check (no measured acetate) and was excluded from further analysis.

Processing of whole-genome shotgun sequencing results

DIAMOND software [31] was used to search for genes related to propionate and butyrate. Raw whole-genome shotgun sequencing reads were used as an input. For propionate, we searched for the genes mmdA, lcdA and pduP (markers for the succinate, acrylate and propanediol pathways, respectively [32]). For butyrate, the process involved searching for but and buk genes. For each read, only the hit with the highest bit score per pathway was kept (e-value ≤1e-05). For each sample, the reads that got a hit with one of the genes were counted and then grouped and summed according to the SCFAs to which they were related. After normalizing for the query sequence size and sequencing depth, the total number of hits related to propionate and butyrate was compared with the relative abundance of these acids. Taxonomic assignment of the sequencing reads was performed with a combination of Kraken2 [33], KrakenUniq [34] and Bracken [35] using HumGut_975 as a custom database, as described by Hiseni et al. [36].

Results & discussion

Identification of taxonomic biomarkers for P:B ratio

We examined the microbiome composition (PacBio sequencing of 16S rRNA gene) and SCFA content of 93 adult fecal samples. The aim was to identify potential associations between different members of the microbiome and levels of propionate and butyrate, and use this information to build a simple, predictive test based on LAD technology.

We computed the correlation between centered log ratio-transformed OTU read counts and relative abundance of propionate and butyrate. Only OTUs with >0.2 or <-0.2 correlations (p < 0.05) were considered. A total of 65 OTUs correlated with propionate levels (39 positively, 26 negatively), and 62 correlated with butyrate levels (28 positively, 34 negatively). Of these, 11 correlated with both butyrate and propionate, albeit in opposite directions. A simplified network of SCFA/OTU relationships is presented in Supplementary Figure 1.

We performed a Basic Local Alignment Search Tool search for highly similar sequences (nucleotide collection [nt/nr] database) using OTU sequences as queries. Among the OTUs positively correlated with butyrate, we found some that shared high sequence identity (>98.5%) with typical butyrate producers, including Faecalibacterium prausnitzii [37,38] (correlation = 0.21; p < 0.05), Agathobaculum butyriciproducens [39] (correlation = 0.23; p < 0.05) and Coprococcus catus [40] (correlation = 0.21; p < 0.05). However, we also found a positive relationship between butyrate and the read counts of sequences sharing high identity (>99%) with Lactobacillus acidophilus (correlation = 0.22; p < 0.05), Fusicatenibacter saccharivorans (correlation = 0.33; p < 0.005) and Blautia wexlerae (correlation = 0.26; p < 0.05) – species not known to produce this acid [41–43]. Furthermore, Dysosmobacter welbionis (correlation = -0.25; p < 0.05) and Flavonifractor plautii (correlation = -0.3; p < 0.005) – both butyrate producers [44,45] – exhibited a negative correlation with the relative abundance of butyrate. Similarly, propionate levels did not exclusively correlate with well-described propionate producers.

In light of this complex outcome, we decided to build a model based on a binary classification system (i.e., classifying samples as having a high or normal acid level). Aiming for a simple method, we chose to detect and classify samples based on a single variable that inferred information about both acid concentrations: the P:B ratio. Classification of samples based on this ratio makes biological sense, as the molar ratio between propionate and butyrate in healthy adults is nearly 1.0 [11–13]. Given the role of butyrate in maintaining human health [2,6,46–48], our goal was to detect samples with depleted butyrate levels, inferred by a deviant P:B ratio in favor of propionate (i.e., P:B ratio >>1.0).

We computed the P:B ratio from GC data for all samples. We then built a model combining partial least squares and linear discriminant analysis (PLS + LDA) using OTU read counts as predictors and aimed to find the ratio that best separated the two groups (normal vs high ratio) while selecting a reasonably small number of OTUs to act as markers. These marker OTUs did not exclusively represent propionate and butyrate producers. GC measurements for each sample are presented in Supplementary Table 1, and a list of all OTUs correlated with propionate and/or butyrate is presented in Supplementary Table 2.

Building a predictive LAD-based test

We designed 21 LAD probes to cover the 37 OTUs selected by the combined partial least squares and linear discriminant analysis model, with the intention of converting the dry lab results to a routine molecular diagnostic tool for classification. Six of the probes failed to produce a signal, so they were removed from the assay. The remaining LAD probes were used to analyze 71 random samples, nine of which were not PacBio-sequenced. The performance of the LAD probes is presented in Supplementary Figure 2.

When signals from the 15 LAD probes were used as an input, the best separation, yielding the highest model prediction accuracy (leave-one-out cross-validated), was observed at 2.6 (Figure 2A). This value corresponded well to the value derived by applying a formula designed to find outliers in positively skewed data like ours (i.e., median value + 3 × median absolute deviation → 0.92 + 3 × 0.54 = 2.54) [49]. A detailed distribution of P:B ratios among the samples tested is presented in Figure 2B.

To ensure that a high P:B ratio (≥2.6) implied increased levels of propionate relative to butyrate (and not acetate), we computed the average levels of these acids within the different groups. Indeed, the average butyrate concentration for the normal ratio group was 20%, whereas the average butyrate concentration for the high ratio group was 7.2%. Samples with a normal P:B ratio had, on average, a propionate level of 16.6%, whereas samples with a high P:B ratio had a level of 29.8% (Figure 2C). These results support our theory that a disturbed ratio between propionate and butyrate elucidates information about the levels of both acids.

Validation of the prediction model using LAD

Given the limited number of samples, we validated the LAD test by performing leave-one-out cross-validation; that is, classification of each sample was performed by taking the rest of the samples into account, excluding from the training set the one to be classified. Of nine samples with a P:B ratio ≥2.6, the algorithm correctly classified seven and missed two; however, of 62 samples with a P:B ratio <2.6, 61 were classified correctly (Figure 3). The positive predictive value showed that for any sample classified as having a high ratio, the chance of that sample indeed having a ratio >2.6 was 87.5%. The negative predictive value was 97%. All nine samples that were not PacBio-sequenced and therefore not included in the initial model for selecting OTU markers were correctly classified (all normal ratio).

We acknowledge that the number of tested samples not forming the basis for marker selection by the PLS + LDA algorithm is low (i.e., nine of 71). Therefore, testing of more independent samples will be crucial in the next phase of LAD characterization as well as further development and implementation. However, we here present a solid proof of concept to serve as a foundation for future work.

We do not possess clinical details regarding the individuals whose samples were tested, and that may present another limitation of this study. It would be of particular interest to learn whether these people suffer from health conditions for which high propionate or low butyrate has been reported. Nevertheless, we screened the metadata of 130 samples used by Zeng et al., who reported that significantly increased propionate levels were associated with a high risk of stroke [22]. We found that, on average, people with a low risk of stroke had a P:B ratio <2.6, whereas significantly higher P:B ratios were observed in people with a medium and high risk of stroke (average P:B ratios of 2.04, 3.22 and 2.84 for low, medium and high risk, respectively) (p < 0.05).

A P:B ratio threshold of approximately 2.6 was determined using two different approaches (PLS + LDA algorithm and outlier formula). It represents a limit separating normal samples from biological outliers in terms of both SCFAs and microbiome composition. It is tempting to speculate that this threshold may very well reflect an important biological threshold with a direct implication for the etiology of complex diseases.

Functional & strain resolution associations with P:B ratio

We chose to further analyze 23 randomly selected samples of various P:B ratios (17 normal, six high) by performing whole-genome shotgun sequencing in an attempt to further explore the biological differences between the two classes (i.e., normal and high ratio). On average, samples with a normal P:B ratio displayed 205 species, whereas samples with a high ratio harbored ten fewer species, suggesting a lower diversity in the latter. However, this difference did not exhibit an acceptable significance level (p > 0.1).

Looking deeper into the composition, we found that high ratio samples were significantly richer in Escherichia coli, Phocaeicola dorei (a known propionate producer, formerly named Bacteroides dorei [50]), Enterocloster sp001517625 (named Clostridium bouchedurhonense at National Center for Biotechnology Information), Blautia_A sp900066165 and Anaerotruncus colihominis (butyrate producer [51]). There was also a tendency for lower richness of F. prausnitzii_C (butyrate producer [37,38]) and Eisenbergiella sp900066775 and higher richness of Akkermansia muciniphila (propionate producer [52]) (p < 0.1) (Figure 4).

Among these strains, our test is designed to detect both E. coli and F. prausnitzii. These two species have commonly been found to act as markers in a wide range of diseases [53–55]. Next, we used the sequencing reads to search for genes related to propionate and butyrate production using DIAMOND software [31], and no linear relationship was found (Figure 5).

This finding corroborates those retrieved from PacBio sequencing, where the majority of OTUs correlated to either propionate or butyrate were not known to be producers of such acids. Furthermore, the abundance of bacteria known to produce specific SCFAs was not always in a positive correlation with the fecal levels of such short-chain fatty acids. This was the case with D. welbionis and F. plautii – both butyrate producers – where relative abundance was found to be negatively correlated with butyrate levels. The latter species was instead found in a positive correlation with propionate levels, and is a target in our assay.

The seemingly complicated relationship between bacterial species and butyrate and propionate levels suggests that levels of SCFAs in fecal samples cannot be inferred by quantifying known acid producers alone, presumably because of the complex cross-feeding mechanisms involved [56]. For example, we believe that the inclusion of Bifidobacterium adolescentis (lactate and acetate producer) as a target of our test, is tightly related to cross-feeding between this bacterium and well-described butyrate producers (i.e., the production of butyrate is enhanced by B. adolescentis activity) [57,58].

Clinical utility

Currently, it appears that the most relevant clinical application of the P:B ratio would be related to neurodegenerative diseases such as Alzheimer's [20] and Parkinson's [59]. A contributing cause of neurodegenerative disease in elderly individuals is their reduced ability to metabolize propionate as a result of decreased methylmalonyl-CoA mutase activity [60]. This leads to potential accumulation of toxic methylmalonic acid, which has been associated with decreased cognitive function in older adults [61]. By contrast, it has been shown that butyrate, a histone deacetylase inhibitor, can act as a therapeutic agent by reducing levels of abnormally deposited brain amyloid-β [62,63]. Flagging samples with a high P:B ratio in a timely manner would assist clinicians to offer the necessary dietary advice to the elderly.

Other diseases and disorders can also potentially be linked to high P:B levels. An association with a significant propionate increase or butyrate decrease has been reported for the ailments listed in Table 2. The most pronounced association was reported between a high P:B ratio and IBS [18]. In addition to being a biomarker, there could also be a causality between the P:B ratio and IBS severity. Thus, this ratio could potentially have utility in treatment of these patients through, for example, dietary advice.

To the best of our knowledge, detecting samples with high P:B ratios can only be achieved by directly quantifying SCFAs and computing the ratios afterward. Measuring the level of SCFAs in fecal samples is usually accomplished by employing GC, LC, capillary electrophoresis or NMR [68]. However, given the complex sample clean-up and preparation procedure combined with high volatility of these acids, SCFA measurement using today's technology remains a challenging task [23,24,69]. This is why knowledge in the field continues to be derived from fragmented, small-scale studies that are insufficiently standardized across laboratories.

The lack of robust methods for use in clinical settings creates a gap between the state-of-the-art knowledge in the field and its practical utility and application. A simple molecular diagnostic method like the LAD test presented here allows inexpensive, high-throughput screening of fecal samples, bridging this gap. The major benefits of LAD in a clinical setting are related to simplicity and cost as well as the ability to detect the microorganisms underlying the P:B ratio, which in turn can be used in therapeutics.

Our approach offers a solution for at least two problems. First, it focuses on the ratio between propionate and butyrate, ignoring their absolute values, which are known to fluctuate based on the time of day a sample is collected and processed [70]. Second, it circumvents the need to measure SCFA levels, utilizing a robust molecular diagnostic system instead.

We offer an indirect way of detecting both propionate and butyrate levels by identifying biological outliers, that is samples with highest propionate and/or lowest butyrate ratios. The tool we present here is not aimed at replacing other conventional 16S rRNA gene or SCFA analyses; nor does it have the capacity to do so, as it is strictly focused on inferring a narrow segment of microbial functionality. Rather, it represents an applicable solution that integrates both types of methodologies into a single measurement of high clinical utility.

Conclusion

Here we present a novel qPCR-compatible, single-tube multiplex test that predicts samples with increased ratios of propionate relative to butyrate. Circumventing the need to directly measure the SCFA content in fecal samples, a robust and simple test like LAD will enable high-throughput analysis and regular monitoring of functional dysbiosis in the gut.