Machine learning optimization of peptides for presentation by class II MHCs.

SUMMARY: T cells play a critical role in cellular immune responses to pathogens and cancer and can be activated and expanded by Major Histocompatibility Complex (MHC)-presented antigens contained in peptide vaccines. We present a machine learning method to optimize the presentation of peptides by class II MHCs by modifying their anchor residues. Our method first learns a model of peptide affinity for a class II MHC using an ensemble of deep residual networks, and then uses the model to propose anchor residue changes to improve peptide affinity. We use a high throughput yeast display assay to show that anchor residue optimization improves peptide binding. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Physical and sensory properties of lemon-flavored acidic beverages formulated with nonfat dry milk during storage.

Sensory and physical properties of 2 lemon-flavored beverages with 5% and 7.5% wt/wt nonfat dry milk (NFDM) at pH 2.5 were studied during storage. The 2 beverages had similar volatile compounds, but the 5% NFDM had higher aroma and lemon flavor, with a preferred appearance by consumers due to the lower turbidity and viscosity. After 28 d of storage at 4°C, lemon flavor decreased in the 5% NFDM beverage but was still more intense than the 7.5% one. During 70 d of storage, no microorganisms were detected, and the beverages were more stable when stored at 4°C than at room temperature according to changes of physical properties measured for appearance, turbidity, color, particle size, zeta potential, rheological properties, and transmission electron microscopy morphology. Findings of the present study suggest that NFDM may be used at 5% wt/wt to produce stable acidic dairy beverages with low turbidity when stored at 4°C.

Antibody complementarity determining region design using high-capacity machine learning.

MOTIVATION: The precise targeting of antibodies and other protein therapeutics is required for their proper function and the elimination of deleterious off-target effects. Often the molecular structure of a therapeutic target is unknown and randomized methods are used to design antibodies without a model that relates antibody sequence to desired properties. RESULTS: Here, we present Ens-Grad, a machine learning method that can design complementarity determining regions of human Immunoglobulin G antibodies with target affinities that are superior to candidates derived from phage display panning experiments. We also demonstrate that machine learning can improve target specificity by the modular composition of models from different experimental campaigns, enabling a new integrative approach to improving target specificity. Our results suggest a new path for the discovery of therapeutic molecules by demonstrating that predictive and differentiable models of antibody binding can be learned from high-throughput experimental data without the need for target structural data. AVAILABILITY AND IMPLEMENTATION: Sequencing data of the phage panning experiment are deposited at NIH's Sequence Read Archive (SRA) under the accession number SRP158510. We make our code available at https://github.com/gifford-lab/antibody-2019. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Efficiency of removal of whey protein from sweet whey using polymeric microfiltration membranes.

Our objective was to measure whey protein removal percentage from separated sweet whey using spiral-wound (SW) polymeric microfiltration (MF) membranes using a 3-stage, 3× process at 50°C and to compare the performance of polymeric membranes with ceramic membranes. Pasteurized, separated Cheddar cheese whey (1,080 kg) was microfiltered using a polymeric 0.3-µm polyvinylidene (PVDF) fluoride SW membrane and a 3×, 3-stage MF process. Cheese making and whey processing were replicated 3 times. There was no detectable level of lactoferrin and no intact α- or ß-casein detected in the MF permeate from the 0.3-µm SW PVDF membranes used in this study. We found BSA and IgG in both the retentate and permeate. The ß-lactoglobulin (ß-LG) and α-lactalbumin (α-LA) partitioned between retentate and permeate, but ß-LG passage through the membrane was retarded more than α-LA because the ratio of ß-LG to α-LA was higher in the MF retentate than either in the sweet whey feed or the MF permeate. About 69% of the crude protein present in the pasteurized separated sweet whey was removed using a 3×, 3-stage, 0.3-µm SW PVDF MF process at 50°C compared with 0.1-µm ceramic graded permeability MF that removed about 85% of crude protein from sweet whey. The polymeric SW membranes used in this study achieve approximately 20% lower yield of whey protein isolate (WPI) and a 50% higher yield of whey protein phospholipid concentrate (WPPC) under the same MF processing conditions as ceramic MF membranes used in the comparison study. Total gross revenue from the sale of WPI plus WPPC produced with polymeric versus ceramic membranes is influenced by both the absolute market price for each product and the ratio of market price of these 2 products. The combination of the market price of WPPC versus WPI and the influence of difference in yield of WPPC and WPI produced with polymeric versus ceramic membranes yielded a price ratio of WPPC versus WPI of 0.556 as the cross over point that determined which membrane type achieves higher total gross revenue return from production of these 2 products from separated sweet whey. A complete economic engineering study comparison of the WPI and WPPC manufacturing costs for polymeric versus ceramic MF membranes is needed to determine the effect of membrane material selection on long-term processing costs, which will affect net revenue and profit when the same quantity of sweet whey is processed under various market price conditions.

Determination of the efficiency of removal of whey protein from sweet whey with ceramic microfiltration membranes.

Our research objective was to measure percent removal of whey protein from separated sweet whey using 0.1-µm uniform transmembrane pressure ceramic microfiltration (MF) membranes in a sequential batch 3-stage, 3× process at 50°C. Cheddar cheese whey was centrifugally separated to remove fat at 72°C and pasteurized (72°C for 15 s), cooled to 4°C, and held overnight. Separated whey (375 kg) was heated to 50°C with a plate heat exchanger and microfiltered using a pilot-scale ceramic 0.1-µm uniform transmembrane pressure MF system in bleed-and-feed mode at 50°C in a sequential batch 3-stage (2 diafiltration stages) process to produce a 3× MF retentate and MF permeate. Feed, retentate, and permeate samples were analyzed for total nitrogen, noncasein nitrogen, and nonprotein nitrogen using the Kjeldahl method. Sodium dodecyl sulfate-PAGE analysis was also performed on the whey feeds, retentates, and permeates from each stage. A flux of 54 kg/m2 per hour was achieved with 0.1-µm ceramic uniform transmembrane pressure microfiltration membranes at 50°C. About 85% of the total nitrogen in the whey feed passed though the membrane into the permeate. No passage of lactoferrin from the sweet whey feed of the MF into the MF permeate was detected. There was some passage of IgG, bovine serum albumen, glycomacropeptide, and casein proteolysis products into the permeate. ß-Lactoglobulin was in higher concentration in the retentate than the permeate, indicating that it was partially blocked from passage through the ceramic MF membrane.

The effect of spray drying on the difference in flavor and functional properties of liquid and dried whey proteins, milk proteins, and micellar casein concentrates.

Traditionally most protein ingredients are sold as a powder due to transport ease and longer shelf life. Many high-protein powder ingredients such as milk protein concentrate with 85% protein and micellar casein concentrate have poor rehydration properties (e.g., solubility) after storage, which might limit their use. An alternative to the production of dried protein ingredients is the option to use liquid protein ingredients, which saves the cost of spray drying, but may also improve flavor and offer different functional properties. The objective of this study was to determine the effect of spray drying on the flavor and functionality of high-protein ingredients. Liquid and dried protein ingredients (whey protein concentrate with 80% protein, whey protein isolate, milk protein concentrate with 85% protein, and micellar casein concentrate) were manufactured from the same lot of milk at the North Carolina State University pilot plant. Functional differences were evaluated by measurement of foam stability and heat stability. Heat stability was evaluated by heating at 90°C for 0, 10, 20, and 30 min followed by micro-bicinchoninic acid and turbidity loss measurements. Sensory properties were evaluated by descriptive analysis, and volatile compounds were evaluated by gas chromatography-mass spectrometry. No differences were detected in protein heat stability between liquids and powders when spray dried under these conditions. Whey protein concentrate with 80% protein (liquid or spray dried) did not produce a foam. All powders had higher aroma intensity and cooked flavors compared with liquids. Powder proteins also had low but distinct cardboard flavor concurrent with higher relative abundance of volatile aldehydes compared with liquids. An understanding of how spray drying affects both flavor and functionality may help food processors better use the ingredients they have available to them.

Genetic diversity of the unicellular nitrogen-fixing cyanobacteria UCYN-A and its prymnesiophyte host.

Symbiotic interactions between nitrogen-fixing prokaryotes and photosynthetic eukaryotes are an integral part of biological nitrogen fixation at a global scale. One of these partnerships involves the cyanobacterium UCYN-A, which has been found in partnership with an uncultivated unicellular prymnesiophyte alga in open-ocean and coastal environments. Phylogenetic analysis of the UCYN-A nitrogenase gene (nifH) showed that the UCYN-A lineage is represented by three distinct clades, referred to herein as UCYN-A1, UCYN-A2 and UCYN-A3, which appear to have overlapping and distinct geographic distributions. The relevance of UCYN-A's genetic diversity to its symbiosis and ecology was explored through combining flow cytometric cell sorting and molecular techniques to determine the host identity, nifH expression patterns and host cell size of one newly discovered clade, UCYN-A2, at a coastal site. UCYN-A2 nifH expression peaked during daylight hours, which is consistent with expression patterns of the UCYN-A1 clade in the open ocean. However, the cell size of the UCYN-A2 host was significantly larger than UCYN-A1 and host, suggesting adaptation to different environmental conditions. Like the UCYN-A1 host, the UCYN-A2 host was closely related to the genus Braarudosphaera; however, the UCYN-A1 and UCYN-A2 host rRNA sequences clustered into two distinct clades suggesting co-evolution of symbiont and host.

A pan-variant mRNA-LNP T cell vaccine protects HLA transgenic mice from mortality after infection with SARS-CoV-2 Beta.

Licensed COVID-19 vaccines ameliorate viral infection by inducing production of neutralizing antibodies that bind the SARS-CoV-2 Spike protein and inhibit viral cellular entry. However, the clinical effectiveness of these vaccines is transitory as viral variants escape antibody neutralization. Effective vaccines that solely rely upon a T cell response to combat SARS-CoV-2 infection could be transformational because they can utilize highly conserved short pan-variant peptide epitopes, but a mRNA-LNP T cell vaccine has not been shown to provide effective anti-SARS-CoV-2 prophylaxis. Here we show a mRNA-LNP vaccine (MIT-T-COVID) based on highly conserved short peptide epitopes activates CD8+ and CD4+ T cell responses that attenuate morbidity and prevent mortality in HLA-A*02:01 transgenic mice infected with SARS-CoV-2 Beta (B.1.351). We found CD8+ T cells in mice immunized with MIT-T-COVID vaccine significantly increased from 1.1% to 24.0% of total pulmonary nucleated cells prior to and at 7 days post infection (dpi), respectively, indicating dynamic recruitment of circulating specific T cells into the infected lungs. Mice immunized with MIT-T-COVID had 2.8 (2 dpi) and 3.3 (7 dpi) times more lung infiltrating CD8+ T cells than unimmunized mice. Mice immunized with MIT-T-COVID had 17.4 times more lung infiltrating CD4+ T cells than unimmunized mice (7 dpi). The undetectable specific antibody response in MIT-T-COVID-immunized mice demonstrates specific T cell responses alone can effectively attenuate the pathogenesis of SARS-CoV-2 infection. Our results suggest further study is merited for pan-variant T cell vaccines, including for individuals that cannot produce neutralizing antibodies or to help mitigate Long COVID.

Using deep learning to annotate the protein universe.

Understanding the relationship between amino acid sequence and protein function is a long-standing challenge with far-reaching scientific and translational implications. State-of-the-art alignment-based techniques cannot predict function for one-third of microbial protein sequences, hampering our ability to exploit data from diverse organisms. Here, we train deep learning models to accurately predict functional annotations for unaligned amino acid sequences across rigorous benchmark assessments built from the 17,929 families of the protein families database Pfam. The models infer known patterns of evolutionary substitutions and learn representations that accurately cluster sequences from unseen families. Combining deep models with existing methods significantly improves remote homology detection, suggesting that the deep models learn complementary information. This approach extends the coverage of Pfam by >9.5%, exceeding additions made over the last decade, and predicts function for 360 human reference proteome proteins with no previous Pfam annotation. These results suggest that deep learning models will be a core component of future protein annotation tools.

Predicted Cellular Immunity Population Coverage Gaps for SARS-CoV-2 Subunit Vaccines and Their Augmentation by Compact Peptide Sets.

Subunit vaccines induce immunity to a pathogen by presenting a component of the pathogen and thus inherently limit the representation of pathogen peptides for cellular immunity-based memory. We find that severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) subunit peptides may not be robustly displayed by the major histocompatibility complex (MHC) molecules in certain individuals. We introduce an augmentation strategy for subunit vaccines that adds a small number of SARS-CoV-2 peptides to a vaccine to improve the population coverage of pathogen peptide display. Our population coverage estimates integrate clinical data on peptide immunogenicity in convalescent COVID-19 patients and machine learning predictions. We evaluate the population coverage of 9 different subunits of SARS-CoV-2, including 5 functional domains and 4 full proteins, and augment each of them to fill a predicted coverage gap.

Targeting ischemia-induced KCC2 hypofunction rescues refractory neonatal seizures and mitigates epileptogenesis in a mouse model.

Neonatal seizures pose a clinical challenge in their early detection, acute management, and long-term comorbidities. They are often caused by hypoxic-ischemic encephalopathy and are frequently refractory to the first-line antiseizure medication phenobarbital. One proposed mechanism for phenobarbital inefficacy during neonatal seizures is the reduced abundance and function of the neuron-specific K+/Cl− cotransporter 2 (KCC2), which maintains chloride homeostasis and promotes GABAergic inhibition upon its phosphorylation during postnatal development. Here, we investigated whether this mechanism is causal and whether it can be rescued by KCC2 functional enhancement. In a CD-1 mouse model of refractory ischemic neonatal seizures, treatment with the KCC2 functional enhancer CLP290 rescued phenobarbital efficacy, increased KCC2 abundance, and prevented the development of epileptogenesis, as quantified by video electroencephalogram monitoring. These effects were prevented by knock-in expression of nonphosphorylatable mutants of KCC2 (S940A or T906A and T1007A), indicating that KCC2 phosphorylation regulates both neonatal seizure susceptibility and CLP290-mediated KCC2 functional enhancement. Our findings therefore validate KCC2 as a clinically relevant target for refractory neonatal seizures and provide insights for future drug development.

TrkB agonists prevent postischemic emergence of refractory neonatal seizures in mice.

Refractory neonatal seizures do not respond to first-line antiseizure medications like phenobarbital (PB), a positive allosteric modulator for GABAA receptors. GABAA receptor-mediated inhibition is dependent upon electroneutral cation-chloride transporter KCC2, which mediates neuronal chloride extrusion and its age-dependent increase and postnatally shifts GABAergic signaling from depolarizing to hyperpolarizing. Brain-derived neurotropic factor-tyrosine receptor kinase B activation (BDNF-TrkB activation) after excitotoxic injury recruits downstream targets like PLCγ1, leading to KCC2 hypofunction. Here, the antiseizure efficacy of TrkB agonists LM22A-4, HIOC, and deoxygedunin (DG) on PB-refractory seizures and postischemic TrkB pathway activation was investigated in a mouse model (CD-1, P7) of refractory neonatal seizures. LM, a BDNF loop II mimetic, rescued PB-refractory seizures in a sexually dimorphic manner. Efficacy was associated with a substantial reduction in the postischemic phosphorylation of TrkB at Y816, a site known to mediate postischemic KCC2 hypofunction via PLCγ1 activation. LM rescued ischemia-induced phospho-KCC2-S940 dephosphorylation, preserving its membrane stability. Full TrkB agonists HIOC and DG similarly rescued PB refractoriness. Chemogenetic inactivation of TrkB substantially reduced postischemic neonatal seizure burdens at P7. Sex differences identified in developmental expression profiles of TrkB and KCC2 may underlie the sexually dimorphic efficacy of LM. These results support a potentially novel role for the TrkB receptor in the emergence of age-dependent refractory neonatal seizures.

Computationally Optimized SARS-CoV-2 MHC Class I and II Vaccine Formulations Predicted to Target Human Haplotype Distributions.

We present a combinatorial machine learning method to evaluate and optimize peptide vaccine formulations for SARS-CoV-2. Our approach optimizes the presentation likelihood of a diverse set of vaccine peptides conditioned on a target human-population HLA haplotype distribution and expected epitope drift. Our proposed SARS-CoV-2 MHC class I vaccine formulations provide 93.21% predicted population coverage with at least five vaccine peptide-HLA average hits per person (≥ 1 peptide: 99.91%) with all vaccine peptides perfectly conserved across 4,690 geographically sampled SARS-CoV-2 genomes. Our proposed MHC class II vaccine formulations provide 97.21% predicted coverage with at least five vaccine peptide-HLA average hits per person with all peptides having an observed mutation probability of ≤ 0.001. We provide an open-source implementation of our design methods (OptiVax), vaccine evaluation tool (EvalVax), as well as the data used in our design efforts here: https://github.com/gifford-lab/optivax.

Robust computational design and evaluation of peptide vaccines for cellular immunity with application to SARS-CoV-2.

We present a combinatorial machine learning method to evaluate and optimize peptide vaccine formulations, and we find for SARS-CoV-2 that it provides superior predicted display of viral epitopes by MHC class I and MHC class II molecules over populations when compared to other candidate vaccines. Our method is robust to idiosyncratic errors in the prediction of MHC peptide display and considers target population HLA haplotype frequencies during optimization. To minimize clinical development time our methods validate vaccines with multiple peptide presentation algorithms to increase the probability that a vaccine will be effective. We optimize an objective function that is based on the presentation likelihood of a diverse set of vaccine peptides conditioned on a target population HLA haplotype distribution and expected epitope drift. We produce separate peptide formulations for MHC class I loci (HLA-A, HLA-B, and HLA-C) and class II loci (HLA-DP, HLA-DQ, and HLA-DR) to permit signal sequence based cell compartment targeting using nucleic acid based vaccine platforms. Our SARS-CoV-2 MHC class I vaccine formulations provide 93.21% predicted population coverage with at least five vaccine peptide-HLA hits on average in an individual (≥ 1 peptide 99.91%) with all vaccine peptides perfectly conserved across 4,690 geographically sampled SARS-CoV-2 genomes. Our MHC class II vaccine formulations provide 90.17% predicted coverage with at least five vaccine peptide-HLA hits on average in an individual with all peptides having observed mutation probability ≤ 0.001. We evaluate 29 previously published peptide vaccine designs with our evaluation tool with the requirement of having at least five vaccine peptide-HLA hits per individual, and they have a predicted maximum of 58.51% MHC class I coverage and 71.65% MHC class II coverage given haplotype based analysis. We provide an open source implementation of our design methods (OptiVax), vaccine evaluation tool (EvalVax), as well as the data used in our design efforts.

Critiquing Protein Family Classification Models Using Sufficient Input Subsets.

In many application domains, neural networks are highly accurate and have been deployed at large scale. However, users often do not have good tools for understanding how these models arrive at their predictions. This has hindered adoption in fields such as the life and medical sciences, where researchers require that models base their decisions on underlying biological phenomena rather than peculiarities of the dataset. We propose a set of methods for critiquing deep learning models and demonstrate their application for protein family classification, a task for which high-accuracy models have considerable potential impact. Our methods extend the Sufficient Input Subsets (SIS) technique, which we use to identify subsets of features in each protein sequence that are alone sufficient for classification. Our suite of tools analyzes these subsets to shed light on the decision-making criteria employed by models trained on this task. These tools show that while deep models may perform classification for biologically relevant reasons, their behavior varies considerably across the choice of network architecture and parameter initialization. While the techniques that we develop are specific to the protein sequence classification task, the approach taken generalizes to a broad set of scientific contexts in which model interpretability is essential.

Pharmaco-resistant Neonatal Seizures: Critical Mechanistic Insights from a Chemoconvulsant Model.

Neonatal seizures are harmful to the developing brain and are associated with mortality and long-term neurological comorbidities. Hypoxic-ischemic encephalopathy (HIE) seizures represent a significant proportion of such seizures. Phenobarbital (PB) remains the first line anti-seizure drug (ASD) treatment but fails ~50% of the time. Translational models of neonatal seizures are crucial to investigating mechanisms underlying PB-resistance. A model of PB-resistant ischemic seizures in postnatal day 7 (P7) CD-1 mice reported K-Cl cotransporter 2 (KCC2) degradation that has been shown to be due to activation of the TrkB pathway. We investigated PB-efficacy in a pentylenetetrazole (PTZ) model of neonatal seizures in the same strain and age using identical treatment protocols to gain insights into mechanisms underlying PB-resistance. A single dose of PTZ (80 mg/kg; IP) consistently induced repetitive seizures that did not progress to status epilepticus (SE). PB (25 mg/kg; IP, single dose) significantly suppressed the PTZ-induced seizures. This was associated with significant KCC2 upregulation and stable Na-K-Cl cotransporter 1 (NKCC1) expression at 24h. The TrkB pathway was not activated. PTZ seizure burdens were significantly higher than those reported for ischemic seizures, indicating seizure severity did not dictate the differences in PB-efficacy. Bumetanide (BTN) (0.1-0.2 mg/kg; IP) did not work as an anti-seizure agent, similar to the ischemic model. When investigating mechanisms underlying the emergence of PB-resistance in translational models, the method by which seizures are induced may dictate mechanisms underlying emergence of PB-resistance.

Clinical assessment and management of general surgery patients via synchronous telehealth.

Objective This paper describes how a clinical team at Landstuhl Regional Medical Center (LRMC) successfully integrated synchronous telehealth (TH) into their routine clinical practice. Methods and materials Synchronous TH encounters were performed using Polycom® software on surgeons' computers with high-definition (HD) cameras on monitors at distant sites and PolyCom HDX9000® Telehealth Practitioner Carts at originating sites. Patients provided consented and were presented to general surgeons by nurses and medical technicians at Army health clinics throughout the European Theater. Results In calendar year (CY) 2014, five general surgeons and two surgical physician assistants (PAs) at Landstuhl Regional Medical Center along with registered nurses (RNs) at six originating clinic sites throughout Europe completed 130 synchronous TH encounters for 101 general surgery patients resulting in 73 completed and 16 recommended surgeries. Eighty-eight percent of patients had a completed or recommended surgery. No surgeries or procedures planned after initial TH evaluation were cancelled. Originating site clinics ranged in distance from 68 miles to 517 miles. Acceptance by providers, patients and clinic staff was high. Conclusion Synchronous TH was effective and safe in evaluating common general surgical conditions. We excluded sensitive and complex conditions requiring a nuanced physical examination. The TH efforts of the general surgery staff have resulted in high-quality, seamless and predictable TH activities that continue to expand into other surgical and medical specialties beyond general surgery. Seven surgeons and two PAs use synchronous TH regularly serving patients over a broad geographic area.

Complete genome sequence of a mosaic bacteriophage, waukesha92.

In this study, we determined the complete genome sequence of a mosaic bacteriophage, Waukesha92, which was isolated from soil using Bacillus thuringiensis as the host organism. This temperate Myoviridae bacteriophage has similarities to phages SpaA1 and BceA1 and the Bacillus thuringiensis plasmid pBMB165.

Comparative genomics reveals surprising divergence of two closely related strains of uncultivated UCYN-A cyanobacteria.

Marine planktonic cyanobacteria capable of fixing molecular nitrogen (termed 'diazotrophs') are key in biogeochemical cycling, and the nitrogen fixed is one of the major external sources of nitrogen to the open ocean. Candidatus Atelocyanobacterium thalassa (UCYN-A) is a diazotrophic cyanobacterium known for its widespread geographic distribution in tropical and subtropical oligotrophic oceans, unusually reduced genome and symbiosis with a single-celled prymnesiophyte alga. Recently a novel strain of this organism was also detected in coastal waters sampled from the Scripps Institute of Oceanography pier. We analyzed the metagenome of this UCYN-A2 population by concentrating cells by flow cytometry. Phylogenomic analysis provided strong bootstrap support for the monophyly of UCYN-A (here called UCYN-A1) and UCYN-A2 within the marine Crocosphaera sp. and Cyanothece sp. clade. UCYN-A2 shares 1159 of the 1200 UCYN-A1 protein-coding genes (96.6%) with high synteny, yet the average amino-acid sequence identity between these orthologs is only 86%. UCYN-A2 lacks the same major pathways and proteins that are absent in UCYN-A1, suggesting that both strains can be grouped at the same functional and ecological level. Our results suggest that UCYN-A1 and UCYN-A2 had a common ancestor and diverged after genome reduction. These two variants may reflect adaptation of the host to different niches, which could be coastal and open ocean habitats.

Coupling FACS and genomic methods for the characterization of uncultivated symbionts.

Symbioses between microbes are likely widespread and functionally relevant in diverse biological systems; however, they are difficult to discover. Most microbes remain uncultivated, symbioses can be relatively rare or dynamic, and intercellular connections can be delicate. Thus, traditional methods such as microscopy are inadequate for efficient discovery and precise characterization of novel interactions, their metabolic basis, and the species involved. High-throughput metagenomic sequencing of entire microbial communities has revolutionized the field of microbial ecology; however, genomic signals from symbionts can get buried in sequences from abundant organisms and evidence for direct links between microbial species cannot be gained from bulk samples. Thus, a specialized approach to the characterization of symbioses between naturally occurring microbes is required. This chapter presents methods for combining fluorescence-activated cell sorting to isolate and separate uncultivated symbionts with molecular biology techniques for DNA amplification in order to characterize uncultivated symbionts through genomic and metagenomic techniques.