Applications of machine learning on electronic health record data to combat antibiotic resistance.

There is growing excitement about the clinical use of artificial intelligence and machine learning technologies. Advancements in computing and the accessibility of machine learning frameworks enable researchers to easily train predictive models using electronic health record data. However, there are several practical factors that must be considered when employing machine learning on electronic health record data. We provide a primer on machine learning and approaches commonly taken to address these challenges. To illustrate how these approaches have been applied to address antimicrobial resistance, we review the use of electronic health record data to construct machine learning models for predicting pathogen carriage or infection, optimizing empiric therapy, and aiding antimicrobial stewardship tasks. Machine learning shows promise in promoting the appropriate use of antimicrobials, although clinical deployment is limited. We conclude by describing potential dangers of, and barriers to, implementation of machine learning models in the clinic.

Many purported pseudogenes in bacterial genomes are bona fide genes.

BACKGROUND: Microbial genomes are largely comprised of protein coding sequences, yet some genomes contain many pseudogenes caused by frameshifts or internal stop codons. These pseudogenes are believed to result from gene degradation during evolution but could also be technical artifacts of genome sequencing or assembly. RESULTS: Using a combination of observational and experimental data, we show that many putative pseudogenes are attributable to errors that are incorporated into genomes during assembly. Within 126,564 publicly available genomes, we observed that nearly identical genomes often substantially differed in pseudogene counts. Causal inference implicated assembler, sequencing platform, and coverage as likely causative factors. Reassembly of genomes from raw reads confirmed that each variable affects the number of putative pseudogenes in an assembly. Furthermore, simulated sequencing reads corroborated our observations that the quality and quantity of raw data can significantly impact the number of pseudogenes in an assembler dependent fashion. The number of unexpected pseudogenes due to internal stops was highly correlated (R2 = 0.96) with average nucleotide identity to the ground truth genome, implying relative pseudogene counts can be used as a proxy for overall assembly correctness. Applying our method to assemblies in RefSeq resulted in rejection of 3.6% of assemblies due to significantly elevated pseudogene counts. Reassembly from real reads obtained from high coverage genomes showed considerable variability in spurious pseudogenes beyond that observed with simulated reads, reinforcing the finding that high coverage is necessary to mitigate assembly errors. CONCLUSIONS: Collectively, these results demonstrate that many pseudogenes in microbial genome assemblies are actually genes. Our results suggest that high read coverage is required for correct assembly and indicate an inflated number of pseudogenes due to internal stops is indicative of poor overall assembly quality.

FindNonCoding: rapid and simple detection of non-coding RNAs in genomes.

SUMMARY: Non-coding RNAs are often neglected during genome annotation due to their difficulty of detection relative to protein coding genes. FindNonCoding takes a pattern mining approach to capture the essential sequence motifs and hairpin loops representing a non-coding RNA family and quickly identify matches in genomes. FindNonCoding was designed for ease of use and accurately finds non-coding RNAs with a low false discovery rate. AVAILABILITY AND IMPLEMENTATION: FindNonCoding is implemented within the DECIPHER package (v2.19.3) for R (v4.1) available from Bioconductor. Pre-trained models of common non-coding RNA families are included for bacteria, archaea and eukarya. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Multi-factorial examination of amplicon sequencing workflows from sample preparation to bioinformatic analysis.

BACKGROUND: The development of sequencing technologies to evaluate bacterial microbiota composition has allowed new insights into the importance of microbial ecology. However, the variety of methodologies used among amplicon sequencing workflows leads to uncertainty about best practices as well as reproducibility and replicability among microbiome studies. Using a bacterial mock community composed of 37 soil isolates, we performed a comprehensive methodological evaluation of workflows, each with a different combination of methodological factors spanning sample preparation to bioinformatic analysis to define sources of artifacts that affect coverage, accuracy, and biases in the resulting compositional profiles. RESULTS: Of the workflows examined, those using the V4-V4 primer set enabled the highest level of concordance between the original mock community and resulting microbiome sequence composition. Use of a high-fidelity polymerase, or a lower-fidelity polymerase with an increased PCR elongation time, limited chimera formation. Bioinformatic pipelines presented a trade-off between the fraction of distinct community members identified (coverage) and fraction of correct sequences (accuracy). DADA2 and QIIME2 assembled V4-V4 reads amplified by Taq polymerase resulted in the highest accuracy (100%) but had a coverage of only 52%. Using mothur to assemble and denoise V4-V4 reads resulted in a coverage of 75%, albeit with marginally lower accuracy (99.5%). CONCLUSIONS: Optimization of microbiome workflows is critical for accuracy and to support reproducibility and replicability among microbiome studies. These considerations will help reveal the guiding principles of microbial ecology and impact the translation of microbiome research to human and environmental health.

Evaluating the long-term portrayal of antibiotic resistance in major U.S. newspapers.

BACKGROUND: Popular media play a critical role in informing the public about antibiotic resistance, which has remained a health concern for over seven decades. Media attention increases the notoriety of antibiotic resistance and shapes the public's perception of its severity, causes, and solutions. Therefore, it is critical the media accurately portray scientific knowledge that may shape personal and policy responses to antibiotic resistance. METHODS: We analyzed articles from two major U.S. newspapers, The New York Times and Los Angeles Times, from 1940 to 2019 to assess trends in sentiment and lexicon surrounding antibiotic and antimicrobial resistance. RESULTS: We observed a gradual increase in the number of relevant articles about resistance, although far fewer than other topics with comparable mortality rates. We found a consistently threatening portrayal of antibiotic resistance as a crisis, reflected in the usage of terms such as "superbug" to refer to some pathogens. Governmental agencies responsible for determining antibiotic usage policies were infrequently mentioned in articles. Blame for resistance was almost exclusively attributed to inappropriate antibiotic use, mainly in animals, rather than appropriate uses of antibiotics. CONCLUSIONS: Collectively, our results provide insights into how popular media can more accurately inform the public about antibiotic resistance. Potential changes include increasing news coverage, avoiding fear-mongering, and adequately conveying the multiple uses of antibiotics that can potentiate resistance.

RNAconTest: comparing tools for noncoding RNA multiple sequence alignment based on structural consistency.

The importance of noncoding RNA sequences has become increasingly clear over the past decade. New RNA families are often detected and analyzed using comparative methods based on multiple sequence alignments. Accordingly, a number of programs have been developed for aligning and deriving secondary structures from sets of RNA sequences. Yet, the best tools for these tasks remain unclear because existing benchmarks contain too few sequences belonging to only a small number of RNA families. RNAconTest (RNA consistency test) is a new benchmarking approach relying on the observation that secondary structure is often conserved across highly divergent RNA sequences from the same family. RNAconTest scores multiple sequence alignments based on the level of consistency among known secondary structures belonging to reference sequences in their output alignment. Similarly, consensus secondary structure predictions are scored according to their agreement with one or more known structures in a family. Comparing the performance of 10 popular alignment programs using RNAconTest revealed that DAFS, DECIPHER, LocARNA, and MAFFT created the most structurally consistent alignments. The best consensus secondary structure predictions were generated by DAFS and LocARNA (via RNAalifold). Many of the methods specific to noncoding RNAs exhibited poor scalability as the number or length of input sequences increased, and several programs displayed substantial declines in score as more sequences were aligned. Overall, RNAconTest provides a means of testing and improving tools for comparative RNA analysis, as well as highlighting the best available approaches. RNAconTest is available from the DECIPHER website (http://DECIPHER.codes/Downloads.html).

Balancing Trade-Offs Imposed by Growth Media and Mass Spectrometry for Bacterial Exometabolomics.

The bacterial exometabolome consists of a vast array of specialized metabolites, many of which are only produced in response to specific environmental stimuli. For this reason, it is desirable to control the extracellular environment with a defined growth medium composed of pure ingredients. However, complex (undefined) media are expected to support the robust growth of a greater variety of microorganisms than defined media. Here, we investigate the trade-offs inherent to a range of complex and defined solid media for the growth of soil microorganisms, production of specialized metabolites, and detection of these compounds using direct infusion mass spectrometry. We find that complex media support growth of more soil microorganisms, as well as allowing for the detection of more previously discovered natural products as a fraction of total m/z features detected in each sample. However, the use of complex media often caused mass spectrometer injection failures and poor-quality mass spectra, which in some cases resulted in over a quarter of samples being removed from analysis. Defined media, while more limiting in growth, generated higher quality spectra and yielded more m/z features after background subtraction. These results inform future exometabolomic experiments requiring a medium that supports the robust growth of many soil microorganisms. IMPORTANCE Bacteria are capable of producing and secreting a rich diversity of specialized metabolites. Yet, much of their exometabolome remains hidden due to challenges associated with eliciting specialized metabolite production, labor-intensive sample preparation, and time-consuming analysis techniques. Using our versatile three-dimensional (3D)-printed culturing platform, SubTap, we demonstrate that rapid exometabolomic data collection from a diverse set of environmental bacteria is feasible. We optimized our platform by surveying Streptomyces isolated from soil on a variety of media types to assess viability, degree of specialized metabolite production, and compatibility with downstream LESA-DIMS analysis. Ultimately, this will enable data-rich experimentation, allowing for a better understanding of bacterial exometabolomes.

AssessORF: combining evolutionary conservation and proteomics to assess prokaryotic gene predictions.

MOTIVATION: A core task of genomics is to identify the boundaries of protein coding genes, which may cover over 90% of a prokaryote's genome. Several programs are available for gene finding, yet it is currently unclear how well these programs perform and whether any offers superior accuracy. This is in part because there is no universal benchmark for gene finding and, therefore, most developers select their own benchmarking strategy. RESULTS: Here, we introduce AssessORF, a new approach for benchmarking prokaryotic gene predictions based on evidence from proteomics data and the evolutionary conservation of start and stop codons. We applied AssessORF to compare gene predictions offered by GenBank, GeneMarkS-2, Glimmer and Prodigal on genomes spanning the prokaryotic tree of life. Gene predictions were 88-95% in agreement with the available evidence, with Glimmer performing the worst but no clear winner. All programs were biased towards selecting start codons that were upstream of the actual start. Given these findings, there remains considerable room for improvement, especially in the detection of correct start sites. AVAILABILITY AND IMPLEMENTATION: AssessORF is available as an R package via the Bioconductor package repository. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Stochastic exits from dormancy give rise to heavy-tailed distributions of descendants in bacterial populations.

Variance in reproductive success is a major determinant of the degree of genetic drift in a population. While many plants and animals exhibit high variance in their number of progeny, far less is known about these distributions for microorganisms. Here, we used a strain barcoding approach to quantify variability in offspring number among replicate bacterial populations and developed a Bayesian method to infer the distribution of descendants from this variability. We applied our approach to measure the offspring distributions for five strains of bacteria from the genus Streptomyces after germination and growth in a homogenous laboratory environment. The distributions of descendants were heavy-tailed, with a few cells effectively 'winning the jackpot' to become a disproportionately large fraction of the population. This extreme variability in reproductive success largely traced back to initial populations of spores stochastically exiting dormancy, which provided early-germinating spores with an exponential advantage. In simulations with multiple dormancy cycles, heavy-tailed distributions of descendants decreased the effective population size by many orders of magnitude and led to allele dynamics differing substantially from classical population genetics models with matching effective population size. Collectively, these results demonstrate that extreme variability in reproductive success can occur even in growth conditions that are far more homogeneous than the natural environment. Thus, extreme variability in reproductive success might be an important factor shaping microbial population dynamics with implications for predicting the fate of beneficial mutations, interpreting sequence variability within populations and explaining variability in infection outcomes across patients.

Exclusivity offers a sound yet practical species criterion for bacteria despite abundant gene flow.

BACKGROUND: The question of whether bacterial species objectively exist has long divided microbiologists. A major source of contention stems from the fact that bacteria regularly engage in horizontal gene transfer (HGT), making it difficult to ascertain relatedness and draw boundaries between taxa. A natural way to define taxa is based on exclusivity of relatedness, which applies when members of a taxon are more closely related to each other than they are to any outsider. It is largely unknown whether exclusive bacterial taxa exist when averaging over the genome or are rare due to rampant hybridization. RESULTS: Here, we analyze a collection of 701 genomes representing a wide variety of environmental isolates from the family Streptomycetaceae, whose members are competent at HGT. We find that the presence/absence of auxiliary genes in the pan-genome displays a hierarchical (tree-like) structure that correlates significantly with the genealogy of the core-genome. Moreover, we identified the existence of many exclusive taxa, although individual genes often contradict these taxa. These conclusions were supported by repeating the analysis on 1,586 genomes belonging to the genus Bacillus. However, despite confirming the existence of exclusive groups (taxa), we were unable to identify an objective threshold at which to assign the rank of species. CONCLUSIONS: The existence of bacterial taxa is justified by considering average relatedness across the entire genome, as captured by exclusivity, but is rejected if one requires unanimous agreement of all parts of the genome. We propose using exclusivity to delimit taxa and conventional genome similarity thresholds to assign bacterial taxa to the species rank. This approach recognizes species that are phylogenetically meaningful, while also establishing some degree of comparability across species-ranked taxa in different bacterial clades.

DesignSignatures: a tool for designing primers that yields amplicons with distinct signatures.

UNLABELLED: For numerous experimental applications, PCR primers must be designed to efficiently amplify a set of homologous DNA sequences while giving rise to amplicons with maximally diverse signatures. We developed DesignSignatures to automate the process of designing primers for high-resolution melting (HRM), fragment length polymorphism (FLP) and sequencing experiments. The program also finds the best restriction enzyme to further diversify HRM or FLP signatures. This enables efficient comparison across many experimental designs in order to maximize signature diversity. AVAILABILITY AND IMPLEMENTATION: DesignSignatures is accessible as a web tool at www.DECIPHER.cee.wisc.edu, or as part of the DECIPHER open source software package for R available from BioConductor. CONTACT: kalin@discovery.wisc.edu SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

DECIPHER: harnessing local sequence context to improve protein multiple sequence alignment.

BACKGROUND: Alignment of large and diverse sequence sets is a common task in biological investigations, yet there remains considerable room for improvement in alignment quality. Multiple sequence alignment programs tend to reach maximal accuracy when aligning only a few sequences, and then diminish steadily as more sequences are added. This drop in accuracy can be partly attributed to a build-up of error and ambiguity as more sequences are aligned. Most high-throughput sequence alignment algorithms do not use contextual information under the assumption that sites are independent. This study examines the extent to which local sequence context can be exploited to improve the quality of large multiple sequence alignments. RESULTS: Two predictors based on local sequence context were assessed: (i) single sequence secondary structure predictions, and (ii) modulation of gap costs according to the surrounding residues. The results indicate that context-based predictors have appreciable information content that can be utilized to create more accurate alignments. Furthermore, local context becomes more informative as the number of sequences increases, enabling more accurate protein alignments of large empirical benchmarks. These discoveries became the basis for DECIPHER, a new context-aware program for sequence alignment, which outperformed other programs on large sequence sets. CONCLUSIONS: Predicting secondary structure based on local sequence context is an efficient means of breaking the independence assumption in alignment. Since secondary structure is more conserved than primary sequence, it can be leveraged to improve the alignment of distantly related proteins. Moreover, secondary structure predictions increase in accuracy as more sequences are used in the prediction. This enables the scalable generation of large sequence alignments that maintain high accuracy even on diverse sequence sets. The DECIPHER R package and source code are freely available for download at DECIPHER.cee.wisc.edu and from the Bioconductor repository.

Exploiting extension bias in polymerase chain reaction to improve primer specificity in ensembles of nearly identical DNA templates.

We describe a semi-empirical framework that combines thermodynamic models of primer hybridization with experimentally determined elongation biases introduced by 3'-end mismatches for improving polymerase chain reaction (PCR)-based sequence discrimination. The framework enables rational and automatic design of primers for optimal targeting of one or more sequences in ensembles of nearly identical DNA templates. In situations where optimal targeting is not feasible, the framework accurately predicts non-target sequences that are difficult to distinguish with PCR alone. Based on the synergistic effects of disparate sources of PCR bias, we used our framework to robustly distinguish between two alleles that differ by a single base pair. To demonstrate the applicability to environmental microbiology, we designed primers specific to all recognized archaeal and bacterial genera in the Ribosomal Database Project, and have made these primers available online. We applied these primers experimentally to obtain genus-specific amplification of 16S rRNA genes representing minor constituents of an environmental DNA sample. Our results demonstrate that inherent PCR biases can be reliably employed in an automatic fashion to maximize sequence discrimination and accurately identify potential cross-amplifications. We have made our framework accessible online as a programme for designing primers targeting one group of sequences in a set with many other sequences (http://DECIPHER.cee.wisc.edu).

Automated design of probes for rRNA-targeted fluorescence in situ hybridization reveals the advantages of using dual probes for accurate identification.

Fluorescence in situ hybridization (FISH) is a common technique for identifying cells in their natural environment and is often used to complement next-generation sequencing approaches as an integral part of the full-cycle rRNA approach. A major challenge in FISH is the design of oligonucleotide probes with high sensitivity and specificity to their target group. The rapidly expanding number of rRNA sequences has increased awareness of the number of potential nontargets for every FISH probe, making the design of new FISH probes challenging using traditional methods. In this study, we conducted a systematic analysis of published probes that revealed that many have insufficient coverage or specificity for their intended target group. Therefore, we developed an improved thermodynamic model of FISH that can be applied at any taxonomic level, used the model to systematically design probes for all recognized genera of bacteria and archaea, and identified potential cross-hybridizations for the selected probes. This analysis resulted in high-specificity probes for 35.6% of the genera when a single probe was used in the absence of competitor probes and for 60.9% when up to two competitor probes were used. Requiring the hybridization of two independent probes for positive identification further increased specificity. In this case, we could design highly specific probe sets for up to 68.5% of the genera without the use of competitor probes and 87.7% when up to two competitor probes were used. The probes designed in this study, as well as tools for designing new probes, are available online (http://DECIPHER.cee.wisc.edu).

Mathematical tools to optimize the design of oligonucleotide probes and primers.

The identification and quantification of specific organisms in mixed microbial communities often relies on the ability to design oligonucleotide probes and primers with high specificity and sensitivity. The design of these oligonucleotides (or "oligos" for short) shares many of the same principles in spite of their widely divergent applications. Three common molecular biology technologies that require oligonucleotide design are polymerase chain reaction (PCR), fluorescence in situ hybridization (FISH), and DNA microarrays. This article reviews techniques and software available for the design and optimization of oligos with the goal of targeting a specific group of organisms within mixed microbial communities. Strategies for enhancing specificity without compromising sensitivity are described, as well as design tools well suited for this purpose.

DECIPHER, a search-based approach to chimera identification for 16S rRNA sequences.

DECIPHER is a new method for finding 16S rRNA chimeric sequences by the use of a search-based approach. The method is based upon detecting short fragments that are uncommon in the phylogenetic group where a query sequence is classified but frequently found in another phylogenetic group. The algorithm was calibrated for full sequences (fs_DECIPHER) and short sequences (ss_DECIPHER) and benchmarked against WigeoN (Pintail), ChimeraSlayer, and Uchime using artificially generated chimeras. Overall, ss_DECIPHER and Uchime provided the highest chimera detection for sequences 100 to 600 nucleotides long (79% and 81%, respectively), but Uchime's performance deteriorated for longer sequences, while ss_DECIPHER maintained a high detection rate (89%). Both methods had low false-positive rates (1.3% and 1.6%). The more conservative fs_DECIPHER, benchmarked only for sequences longer than 600 nucleotides, had an overall detection rate lower than that of ss_DECIPHER (75%) but higher than those of the other programs. In addition, fs_DECIPHER had the lowest false-positive rate among all the benchmarked programs (<0.20%). DECIPHER was outperformed only by ChimeraSlayer and Uchime when chimeras were formed from closely related parents (less than 10% divergence). Given the differences in the programs, it was possible to detect over 89% of all chimeras with just the combination of ss_DECIPHER and Uchime. Using fs_DECIPHER, we detected between 1% and 2% additional chimeras in the RDP, SILVA, and Greengenes databases from which chimeras had already been removed with Pintail or Bellerophon. DECIPHER was implemented in the R programming language and is directly accessible through a webpage or by downloading the program as an R package (http://DECIPHER.cee.wisc.edu).

Multi-stable bacterial communities exhibit extreme sensitivity to initial conditions.

Microbial communities can have dramatically different compositions even among similar environments. This might be due to the existence of multiple alternative stable states, yet there exists little experimental evidence supporting this possibility. Here, we gathered a large collection of absolute population abundances capturing population dynamics in one- to four-strain communities of soil bacteria with a complex life cycle in a feast-or-famine environment. This dataset led to several observations: (i) some pairwise competitions resulted in bistability with a separatrix near a 1:1 initial ratio across a range of population densities; (ii) bistability propagated to multi-stability in multispecies communities; and (iii) replicate microbial communities reached different stable states when starting close to initial conditions separating basins of attraction, indicating finite-sized regions where the dynamics are unpredictable. The generalized Lotka-Volterra equations qualitatively captured most competition outcomes but were unable to quantitatively recapitulate the observed dynamics. This was partly due to complex and diverse growth dynamics in monocultures that ranged from Allee effects to nonmonotonic behaviors. Overall, our results highlight that multi-stability might be generic in multispecies communities and, combined with ecological noise, can lead to unpredictable community assembly, even in simple environments.

Accurate annotation of protein coding sequences with IDTAXA.

The observed diversity of protein coding sequences continues to increase far more rapidly than knowledge of their functions, making classification algorithms essential for assigning a function to proteins using only their sequence. Most pipelines for annotating proteins rely on searches for homologous sequences in databases of previously annotated proteins using BLAST or HMMER. Here, we develop a new approach for classifying proteins into a taxonomy of functions and demonstrate its utility for genome annotation. Our algorithm, IDTAXA, was more accurate than BLAST or HMMER at assigning sequences to KEGG ortholog groups. Moreover, IDTAXA correctly avoided classifying sequences with novel functions to existing groups, which is a common error mode for classification approaches that rely on E-values as a proxy for confidence. We demonstrate IDTAXA's utility for annotating eukaryotic and prokaryotic genomes by assigning functions to proteins within a multi-level ontology and applied IDTAXA to detect genome contamination in eukaryotic genomes. Finally, we re-annotated 8604 microbial genomes with known antibiotic resistance phenotypes to discover two novel associations between proteins and antibiotic resistance. IDTAXA is available as a web tool (http://DECIPHER.codes/Classification.html) or as part of the open source DECIPHER R package from Bioconductor.

Identification of antibiotic pairs that evade concurrent resistance via a retrospective analysis of antimicrobial susceptibility test results.

BACKGROUND: Some antibiotic pairs display a property known as collateral sensitivity in which the evolution of resistance to one antibiotic increases sensitivity to the other. Alternating between collaterally sensitive antibiotics has been proposed as a sustainable solution to the problem of antibiotic resistance. We aimed to identify antibiotic pairs that could be considered for treatment strategies based on alternating antibiotics. METHODS: We did a retrospective analysis of 448 563 antimicrobial susceptibility test results acquired over a 4-year period (Jan 1, 2015, to Dec 31, 2018) from 23 hospitals in the University of Pittsburgh Medical Center (Pittsburgh, PA, USA) hospital system. We used a score based on mutual information to identify pairs of antibiotics displaying disjoint resistance, wherein resistance to one antibiotic is commonly associated with susceptibility to the other and vice versa. We applied this approach to the six most frequently isolated bacterial pathogens (Escherichia coli, Staphylococcus aureus, Klebsiella pneumoniae, Enterococcus faecalis, Pseudomonas aeruginosa, and Proteus mirabilis) and subpopulations of each created by conditioning on resistance to individual antibiotics. To identify higher-order antibiotic interactions, we predicted rates of multidrug resistance for triplets of antibiotics using Markov random fields and compared these to the observed rates. FINDINGS: We identified 69 antibiotic pairs displaying varying degrees of disjoint resistance for subpopulations of the six bacterial species. However, disjoint resistance was rarely conserved at the species level, with only 6 (0·7%) of 875 antibiotic pairs showing evidence of disjoint resistance. Instead, more than half of antibiotic pairs (465 [53·1%] of 875) exhibited signatures of concurrent resistance, whereby resistance to one antibiotic is associated with resistance to another. We found concurrent resistance to extend to more than two antibiotics, with observed rates of resistance to three antibiotics being higher than predicted from pairwise information alone. INTERPRETATION: The high frequency of concurrent resistance shows that bacteria have means of counteracting multiple antibiotics at a time. The almost complete absence of disjoint resistance at the species level implies that treatment strategies based on alternating between antibiotics might require subspecies level pathogen identification and be limited to a few antibiotic pairings. FUNDING: US National Institutes of Health.

SubTap, a Versatile 3D Printed Platform for Eavesdropping on Extracellular Interactions.

Communication within the microbiome occurs through an immense diversity of small molecules. Capturing these microbial interactions is a significant challenge due to the complexity of the exometabolome and its sensitivity to environmental stimuli. Traditional methods for acquiring exometabolomic data from interacting microorganisms are limited by their low throughput or lack of sampling depth. To address this challenge, we introduce subtapping (short for substrate tapping), a technique for tapping into extracellular metabolites that are being transferred through the growth substrate during coculture. High-throughput subtapping is made possible by a new coculturing platform, named SubTap, that we engineered to resemble a 96-well plate. The three-dimensional (3D) printed SubTap platform captures the exometabolome in an agar compartment that connects physically separated growth chambers, which permits cell growth without competition for space. We show how SubTap facilitates replicable and quick detection of exometabolites via direct infusion mass spectrometry analysis. Using bacterial isolates from the soil, we apply SubTap to characterize the effects of growth medium, growth duration, and mixed versus unmixed coculturing on the exometabolome. Finally, we demonstrate SubTap's versatility by interrogating microbial interactions in multicultures with up to four strains. IMPORTANCE Improvements in experimental techniques and instrumentation have led to the discovery that the microbiome plays an essential role in human and environmental health. Nevertheless, there remain major impediments to conducting large-scale interrogations of the microbiome in a high-throughput manner, particularly in the field of exometabolomics. Existing methods to coculture microorganisms and interrogate their interactions are labor-intensive and low throughput. This inspired us to develop a solution for coculturing that was (i) open source, (ii) inexpensive, (iii) scalable, (iv) customizable, and (v) compatible with existing mass spectrometry instrumentation. Here, we present SubTap-a 3D printed coculturing platform that permits tapping directly into the growth substrate between physically separated, but interconnected, growth compartments. SubTap allows multiculture (with up to four distinct growth compartments) in spatially mixed or unmixed configurations and enables repeatable results with mass spectrometry, as shown by our validation with known compounds and cultures of one to four organisms.