Design automation of microfluidic single and double emulsion droplets with machine learning.

Droplet microfluidics enables kHz screening of picoliter samples at a fraction of the cost of other high-throughput approaches. However, generating stable droplets with desired characteristics typically requires labor-intensive empirical optimization of device designs and flow conditions that limit adoption to specialist labs. Here, we compile a comprehensive droplet dataset and use it to train machine learning models capable of accurately predicting device geometries and flow conditions required to generate stable aqueous-in-oil and oil-in-aqueous single and double emulsions from 15 to 250 µm at rates up to 12000 Hz for different fluids commonly used in life sciences. Blind predictions by our models for as-yet-unseen fluids, geometries, and device materials yield accurate results, establishing their generalizability. Finally, we generate an easy-to-use design automation tool that yield droplets within 3 µm (<8%) of the desired diameter, facilitating tailored droplet-based platforms and accelerating their utility in life sciences.

Versatility and stability optimization of flow-focusing droplet generators via quality metric-driven design automation.

Droplet generation is a fundamental component of droplet microfluidics, compartmentalizing biological or chemical systems within a water-in-oil emulsion. As adoption of droplet microfluidics expands beyond expert labs or integrated devices, quality metrics are needed to contextualize the performance capabilities, improving the reproducibility and efficiency of operation. Here, we present two quality metrics for droplet generation: performance versatility, the operating range of a single device, and stability, the distance of a single operating point from a regime change. Both metrics were characterized in silico and validated experimentally using machine learning and rapid prototyping. These metrics were integrated into a design automation workflow, DAFD 2.0, which provides users with droplet generators of a desired performance that are versatile or flow stable. Versatile droplet generators with stable operating points accelerate the development of sophisticated devices by facilitating integration of other microfluidic components and improving the accuracy of design automation tools.

Phosphatase specificity principles uncovered by MRBLE:Dephos and global substrate identification.

Phosphoprotein phosphatases (PPPs) regulate major signaling pathways, but the determinants of phosphatase specificity are poorly understood. This is because methods to investigate this at scale are lacking. Here, we develop a novel in vitro assay, MRBLE:Dephos, that allows multiplexing of dephosphorylation reactions to determine phosphatase preferences. Using MRBLE:Dephos, we establish amino acid preferences of the residues surrounding the dephosphorylation site for PP1 and PP2A-B55, which reveals common and unique preferences. To compare the MRBLE:Dephos results to cellular substrates, we focused on mitotic exit that requires extensive dephosphorylation by PP1 and PP2A-B55. We use specific inhibition of PP1 and PP2A-B55 in mitotic exit lysates coupled with phosphoproteomics to identify more than 2,000 regulated sites. Importantly, the sites dephosphorylated during mitotic exit reveal key signatures that are consistent with MRBLE:Dephos. Furthermore, integration of our phosphoproteomic data with mitotic interactomes of PP1 and PP2A-B55 provides insight into how binding of phosphatases to substrates shapes dephosphorylation. Collectively, we develop novel approaches to investigate protein phosphatases that provide insight into mitotic exit regulation.

Short tandem repeats bind transcription factors to tune eukaryotic gene expression.

Short tandem repeats (STRs) are enriched in eukaryotic cis-regulatory elements and alter gene expression, yet how they regulate transcription remains unknown. We found that STRs modulate transcription factor (TF)-DNA affinities and apparent on-rates by about 70-fold by directly binding TF DNA-binding domains, with energetic impacts exceeding many consensus motif mutations. STRs maximize the number of weakly preferred microstates near target sites, thereby increasing TF density, with impacts well predicted by statistical mechanics. Confirming that STRs also affect TF binding in cells, neural networks trained only on in vivo occupancies predicted effects identical to those observed in vitro. Approximately 90% of TFs preferentially bound STRs that need not resemble known motifs, providing a cis-regulatory mechanism to target TFs to genomic sites.

A microwell platform for high-throughput longitudinal phenotyping and selective retrieval of organoids.

Organoids are powerful experimental models for studying the ontogeny and progression of various diseases including cancer. Organoids are conventionally cultured in bulk using an extracellular matrix mimic. However, bulk-cultured organoids physically overlap, making it impossible to track the growth of individual organoids over time in high throughput. Moreover, local spatial variations in bulk matrix properties make it difficult to assess whether observed phenotypic heterogeneity between organoids results from intrinsic cell differences or differences in the microenvironment. Here, we developed a microwell-based method that enables high-throughput quantification of image-based parameters for organoids grown from single cells, which can further be retrieved from their microwells for molecular profiling. Coupled with a deep learning image-processing pipeline, we characterized phenotypic traits including growth rates, cellular movement, and apical-basal polarity in two CRISPR-engineered human gastric organoid models, identifying genomic changes associated with increased growth rate and changes in accessibility and expression correlated with apical-basal polarity. A record of this paper's transparent peer review process is included in the supplemental information.

Decoupling of catalysis and transition state analog binding from mutations throughout a phosphatase revealed by high-throughput enzymology.

Using high-throughput microfluidic enzyme kinetics (HT-MEK), we measured over 9,000 inhibition curves detailing impacts of 1,004 single-site mutations throughout the alkaline phosphatase PafA on binding affinity for two transition state analogs (TSAs), vanadate and tungstate. As predicted by catalytic models invoking transition state complementary, mutations to active site and active-site-contacting residues had highly similar impacts on catalysis and TSA binding. Unexpectedly, most mutations to more distal residues that reduced catalysis had little or no impact on TSA binding and many even increased tungstate affinity. These disparate effects can be accounted for by a model in which distal mutations alter the enzyme's conformational landscape, increasing the occupancy of microstates that are catalytically less effective but better able to accommodate larger transition state analogs. In support of this ensemble model, glycine substitutions (rather than valine) were more likely to increase tungstate affinity (but not more likely to impact catalysis), presumably due to increased conformational flexibility that allows previously disfavored microstates to increase in occupancy. These results indicate that residues throughout an enzyme provide specificity for the transition state and discriminate against analogs that are larger only by tenths of an Ångström. Thus, engineering enzymes that rival the most powerful natural enzymes will likely require consideration of distal residues that shape the enzyme's conformational landscape and fine-tune active-site residues. Biologically, the evolution of extensive communication between the active site and remote residues to aid catalysis may have provided the foundation for allostery to make it a highly evolvable trait.

Double emulsions as a high-throughput enrichment and isolation platform for slower-growing microbes.

Our understanding of in situ microbial physiology is primarily based on physiological characterization of fast-growing and readily-isolatable microbes. Microbial enrichments to obtain novel isolates with slower growth rates or physiologies adapted to low nutrient environments are plagued by intrinsic biases for fastest-growing species when using standard laboratory isolation protocols. New cultivation tools to minimize these biases and enrich for less well-studied taxa are needed. In this study, we developed a high-throughput bacterial enrichment platform based on single cell encapsulation and growth within double emulsions (GrowMiDE). We showed that GrowMiDE can cultivate many different microorganisms and enrich for underrepresented taxa that are never observed in traditional batch enrichments. For example, preventing dominance of the enrichment by fast-growing microbes due to nutrient privatization within the double emulsion droplets allowed cultivation of slower-growing Negativicutes and Methanobacteria from stool samples in rich media enrichment cultures. In competition experiments between growth rate and growth yield specialist strains, GrowMiDE enrichments prevented competition for shared nutrient pools and enriched for slower-growing but more efficient strains. Finally, we demonstrated the compatibility of GrowMiDE with commercial fluorescence-activated cell sorting (FACS) to obtain isolates from GrowMiDE enrichments. Together, GrowMiDE + DE-FACS is a promising new high-throughput enrichment platform that can be easily applied to diverse microbial enrichments or screens.

De novo distillation of thermodynamic affinity from deep learning regulatory sequence models of in vivo protein-DNA binding.

Transcription factors (TF) are proteins that bind DNA in a sequence-specific manner to regulate gene transcription. Despite their unique intrinsic sequence preferences, in vivo genomic occupancy profiles of TFs differ across cellular contexts. Hence, deciphering the sequence determinants of TF binding, both intrinsic and context-specific, is essential to understand gene regulation and the impact of regulatory, non-coding genetic variation. Biophysical models trained on in vitro TF binding assays can estimate intrinsic affinity landscapes and predict occupancy based on TF concentration and affinity. However, these models cannot adequately explain context-specific, in vivo binding profiles. Conversely, deep learning models, trained on in vivo TF binding assays, effectively predict and explain genomic occupancy profiles as a function of complex regulatory sequence syntax, albeit without a clear biophysical interpretation. To reconcile these complementary models of in vitro and in vivo TF binding, we developed Affinity Distillation (AD), a method that extracts thermodynamic affinities de-novo from deep learning models of TF chromatin immunoprecipitation (ChIP) experiments by marginalizing away the influence of genomic sequence context. Applied to neural networks modeling diverse classes of yeast and mammalian TFs, AD predicts energetic impacts of sequence variation within and surrounding motifs on TF binding as measured by diverse in vitro assays with superior dynamic range and accuracy compared to motif-based methods. Furthermore, AD can accurately discern affinities of TF paralogs. Our results highlight thermodynamic affinity as a key determinant of in vivo binding, suggest that deep learning models of in vivo binding implicitly learn high-resolution affinity landscapes, and show that these affinities can be successfully distilled using AD. This new biophysical interpretation of deep learning models enables high-throughput in silico experiments to explore the influence of sequence context and variation on both intrinsic affinity and in vivo occupancy.

Large-scale mapping and mutagenesis of human transcriptional effector domains.

Human gene expression is regulated by more than 2,000 transcription factors and chromatin regulators1,2. Effector domains within these proteins can activate or repress transcription. However, for many of these regulators we do not know what type of effector domains they contain, their location in the protein, their activation and repression strengths, and the sequences that are necessary for their functions. Here, we systematically measure the effector activity of more than 100,000 protein fragments tiling across most chromatin regulators and transcription factors in human cells (2,047 proteins). By testing the effect they have when recruited at reporter genes, we annotate 374 activation domains and 715 repression domains, roughly 80% of which are new and have not been previously annotated3-5. Rational mutagenesis and deletion scans across all the effector domains reveal aromatic and/or leucine residues interspersed with acidic, proline, serine and/or glutamine residues are necessary for activation domain activity. Furthermore, most repression domain sequences contain sites for small ubiquitin-like modifier (SUMO)ylation, short interaction motifs for recruiting corepressors or are structured binding domains for recruiting other repressive proteins. We discover bifunctional domains that can both activate and repress, some of which dynamically split a cell population into high- and low-expression subpopulations. Our systematic annotation and characterization of effector domains provide a rich resource for understanding the function of human transcription factors and chromatin regulators, engineering compact tools for controlling gene expression and refining predictive models of effector domain function.

Microfluidic encapsulation of photosynthetic cyanobacteria in hydrogel microparticles augments oxygen delivery to rescue ischemic myocardium.

Cardiovascular disease, primarily caused by coronary artery disease, is the leading cause of death in the United States. While standard clinical interventions have improved patient outcomes, mortality rates associated with eventual heart failure still represent a clinical challenge. Macrorevascularization techniques inadequately address the microvascular perfusion deficits that persist beyond primary and secondary interventions. In this work, we investigate a photosynthetic oxygen delivery system that rescues the myocardium following acute ischemia. Using a simple microfluidic system, we encapsulated Synechococcus elongatus into alginate hydrogel microparticles (HMPs), which photosynthetically deliver oxygen to ischemic tissue in the absence of blood flow. We demonstrate that HMPs improve the viability of S. elongatus during the injection process and allow for simple oxygen diffusion. Adult male Wistar rats (n = 45) underwent sham surgery, acute ischemia reperfusion surgery, or a chronic ischemia reperfusion surgery, followed by injection of phosphate buffered saline (PBS), S. elongatus suspended in PBS, HMPs, or S. elongatus encapsulated in HMPs. Treatment with S. elongatus-HMPs mitigated cellular apoptosis and improved left ventricular function. Thus, delivery of S. elongatus encapsulated in HMPs improves clinical translation by utilizing a minimally invasive delivery platform that improves S. elongatus viability and enhances the therapeutic benefit of a novel photosynthetic system for the treatment of myocardial ischemia.

On the dependent recognition of some long zinc finger proteins.

The human genome contains about 800 C2H2 zinc finger proteins (ZFPs), and most of them are composed of long arrays of zinc fingers. Standard ZFP recognition model asserts longer finger arrays should recognize longer DNA-binding sites. However, recent experimental efforts to identify in vivo ZFP binding sites contradict this assumption, with many exhibiting short motifs. Here we use ZFY, CTCF, ZIM3, and ZNF343 as examples to address three closely related questions: What are the reasons that impede current motif discovery methods? What are the functions of those seemingly unused fingers and how can we improve the motif discovery algorithms based on long ZFPs' biophysical properties? Using ZFY, we employed a variety of methods and find evidence for 'dependent recognition' where downstream fingers can recognize some previously undiscovered motifs only in the presence of an intact core site. For CTCF, high-throughput measurements revealed its upstream specificity profile depends on the strength of its core. Moreover, the binding strength of the upstream site modulates CTCF's sensitivity to different epigenetic modifications within the core, providing new insight into how the previously identified intellectual disability-causing and cancer-related mutant R567W disrupts upstream recognition and deregulates the epigenetic control by CTCF. Our results establish that, because of irregular motif structures, variable spacing and dependent recognition between sub-motifs, the specificities of long ZFPs are significantly underestimated, so we developed an algorithm, ModeMap, to infer the motifs and recognition models of ZIM3 and ZNF343, which facilitates high-confidence identification of specific binding sites, including repeats-derived elements. With revised concept, technique, and algorithm, we can discover the overlooked specificities and functions of those 'extra' fingers, and therefore decipher their broader roles in human biology and diseases.

Alleviating Cell Lysate-Induced Inhibition to Enable RT-PCR from Single Cells in Picoliter-Volume Double Emulsion Droplets.

Microfluidic droplet assays enable single-cell polymerase chain reaction (PCR) and sequencing analyses at unprecedented scales, with most methods encapsulating cells within nanoliter-sized single emulsion droplets (water-in-oil). Encapsulating cells within picoliter double emulsion (DE) (water-in-oil-in-water) allows sorting droplets with commercially available fluorescence-activated cell sorter (FACS) machines, making it possible to isolate single cells based on phenotypes of interest for downstream analyses. However, sorting DE droplets with standard cytometers requires small droplets that can pass FACS nozzles. This poses challenges for molecular biology, as prior reports suggest that reverse transcription (RT) and PCR amplification cannot proceed efficiently at volumes below 1 nL due to cell lysate-induced inhibition. To overcome this limitation, we used a plate-based RT-PCR assay designed to mimic reactions in picoliter droplets to systematically quantify and ameliorate the inhibition. We find that RT-PCR is blocked by lysate-induced cleavage of nucleic acid probes and primers, which can be efficiently alleviated through heat lysis. We further show that the magnitude of inhibition depends on the cell type, but that RT-PCR can proceed in low-picoscale reaction volumes for most mouse and human cell lines tested. Finally, we demonstrate one-step RT-PCR from single cells in 20 pL DE droplets with fluorescence quantifiable via FACS. These results open up new avenues for improving picoscale droplet RT-PCR reactions and expanding microfluidic droplet-based single-cell analysis technologies.

A bead-based method for high-throughput mapping of the sequence- and force-dependence of T cell activation.

Adaptive immunity relies on T lymphocytes that use αß T cell receptors (TCRs) to discriminate among peptides presented by major histocompatibility complex molecules (pMHCs). Identifying pMHCs capable of inducing robust T cell responses will not only enable a deeper understanding of the mechanisms governing immune responses but could also have broad applications in diagnosis and treatment. T cell recognition of sparse antigenic pMHCs in vivo relies on biomechanical forces. However, in vitro screening methods test potential pMHCs without force and often at high (nonphysiological) pMHC densities and thus fail to predict potent agonists in vivo. Here, we present a technology termed BATTLES (biomechanically assisted T cell triggering for large-scale exogenous-pMHC screening) that uses biomechanical force to initiate T cell triggering for peptides and cells in parallel. BATTLES displays candidate pMHCs on spectrally encoded beads composed of a thermo-responsive polymer capable of applying shear loads to T cells, facilitating exploration of the force- and sequence-dependent landscape of T cell responses. BATTLES can be used to explore basic T cell mechanobiology and T cell-based immunotherapies.

Metasurface optofluidics for dynamic control of light fields.

The ability to manipulate light and liquids on integrated optofluidics chips has spurred a myriad of important developments in biology, medicine, chemistry and display technologies. Here we show how the convergence of optofluidics and metasurface optics can lead to conceptually new platforms for the dynamic control of light fields. We first demonstrate metasurface building blocks that display an extreme sensitivity in their scattering properties to their dielectric environment. These blocks are then used to create metasurface-based flat optics inside microfluidic channels where liquids with different refractive indices can be directed to manipulate their optical behaviour. We demonstrate the intensity and spectral tuning of metasurface colour pixels as well as on-demand optical elements. We finally demonstrate automated control in an integrated meta-optofluidic platform to open up new display functions. Combined with large-scale microfluidic integration, our dynamic-metasurface flat-optics platform could open up the possibility of dynamic display, imaging, holography and sensing applications.

Machine learning for microfluidic design and control.

Microfluidics has developed into a mature field with applications across science and engineering, having particular commercial success in molecular diagnostics, next-generation sequencing, and bench-top analysis. Despite its ubiquity, the complexity of designing and controlling custom microfluidic devices present major barriers to adoption, requiring intuitive knowledge gained from years of experience. If these barriers were overcome, microfluidics could miniaturize biological and chemical research for non-experts through fully-automated platform development and operation. The intuition of microfluidic experts can be captured through machine learning, where complex statistical models are trained for pattern recognition and subsequently used for event prediction. Integration of machine learning with microfluidics could significantly expand its adoption and impact. Here, we present the current state of machine learning for the design and control of microfluidic devices, its possible applications, and current limitations.

Systematic characterization of effect of flow rates and buffer compositions on double emulsion droplet volumes and stability.

Double emulsion droplets (DEs) are water/oil/water droplets that can be sorted via fluorescence-activated cell sorting (FACS), allowing for new opportunities in high-throughput cellular analysis, enzymatic screening, and synthetic biology. These applications require stable, uniform droplets with predictable microreactor volumes. However, predicting DE droplet size, shell thickness, and stability as a function of flow rate has remained challenging for monodisperse single core droplets and those containing biologically-relevant buffers, which influence bulk and interfacial properties. As a result, developing novel DE-based bioassays has typically required extensive initial optimization of flow rates to find conditions that produce stable droplets of the desired size and shell thickness. To address this challenge, we conducted systematic size parameterization quantifying how differences in flow rates and buffer properties (viscosity and interfacial tension at water/oil interfaces) alter droplet size and stability, across 6 inner aqueous buffers used across applications such as cellular lysis, microbial growth, and drug delivery, quantifying the size and shell thickness of >22 000 droplets overall. We restricted our study to stable single core droplets generated in a 2-step dripping-dripping formation regime in a straightforward PDMS device. Using data from 138 unique conditions (flow rates and buffer composition), we also demonstrated that a recent physically-derived size law of Wang et al. can accurately predict double emulsion shell thickness for >95% of observations. Finally, we validated the utility of this size law by using it to accurately predict droplet sizes for a novel bioassay that requires encapsulating growth media for bacteria in droplets. This work has the potential to enable new screening-based biological applications by simplifying novel DE bioassay development.

Tuning T cell receptor sensitivity through catch bond engineering.

Adoptive cell therapy using engineered T cell receptors (TCRs) is a promising approach for targeting cancer antigens, but tumor-reactive TCRs are often weakly responsive to their target ligands, peptide-major histocompatibility complexes (pMHCs). Affinity-matured TCRs can enhance the efficacy of TCR-T cell therapy but can also cross-react with off-target antigens, resulting in organ immunopathology. We developed an alternative strategy to isolate TCR mutants that exhibited high activation signals coupled with low-affinity pMHC binding through the acquisition of catch bonds. Engineered analogs of a tumor antigen MAGE-A3-specific TCR maintained physiological affinities while exhibiting enhanced target killing potency and undetectable cross-reactivity, compared with a high-affinity clinically tested TCR that exhibited lethal cross-reactivity with a cardiac antigen. Catch bond engineering is a biophysically based strategy to tune high-sensitivity TCRs for T cell therapy with reduced potential for adverse cross-reactivity.

uPIC-M: Efficient and Scalable Preparation of Clonal Single Mutant Libraries for High-Throughput Protein Biochemistry.

New high-throughput biochemistry techniques complement selection-based approaches and provide quantitative kinetic and thermodynamic data for thousands of protein variants in parallel. With these advances, library generation rather than data collection has become rate-limiting. Unlike pooled selection approaches, high-throughput biochemistry requires mutant libraries in which individual sequences are rationally designed, efficiently recovered, sequence-validated, and separated from one another, but current strategies are unable to produce these libraries at the needed scale and specificity at reasonable cost. Here, we present a scalable, rapid, and inexpensive approach for creating User-designed Physically Isolated Clonal-Mutant (uPIC-M) libraries that utilizes recent advances in oligo synthesis, high-throughput sample preparation, and next-generation sequencing. To demonstrate uPIC-M, we created a scanning mutant library of SpAP, a 541 amino acid alkaline phosphatase, and recovered 94% of desired mutants in a single iteration. uPIC-M uses commonly available equipment and freely downloadable custom software and can produce a 5000 mutant library at 1/3 the cost and 1/5 the time of traditional techniques.

Fundamentals to function: Quantitative and scalable approaches for measuring protein stability.

Folding a linear chain of amino acids into a three-dimensional protein is a complex physical process that ultimately confers an impressive range of diverse functions. Although recent advances have driven significant progress in predicting three-dimensional protein structures from sequence, proteins are not static molecules. Rather, they exist as complex conformational ensembles defined by energy landscapes spanning the space of sequence and conditions. Quantitatively mapping the physical parameters that dictate these landscapes and protein stability is therefore critical to develop models that are capable of predicting how mutations alter function of proteins in disease and informing the design of proteins with desired functions. Here, we review the approaches that are used to quantify protein stability at a variety of scales, from returning multiple thermodynamic and kinetic measurements for a single protein sequence to yielding indirect insights into folding across a vast sequence space. The physical parameters derived from these approaches will provide a foundation for models that extend beyond the structural prediction to capture the complexity of conformational ensembles and, ultimately, their function.

MRBLE-pep Measurements Reveal Accurate Binding Affinities for B56, a PP2A Regulatory Subunit.

Signal transduction pathways rely on dynamic interactions between protein globular domains and short linear motifs (SLiMs). The weak affinities of these interactions are essential to allow fast rewiring of signaling pathways and downstream responses but also pose technical challenges for interaction detection and measurement. We recently developed a technique (MRBLE-pep) that leverages spectrally encoded hydrogel beads to measure binding affinities between a single protein of interest and 48 different peptide sequences in a single small volume. In prior work, we applied it to map the binding specificity landscape between calcineurin and the PxIxIT SLiM (Nguyen, H. Q. et al. Elife 2019, 8). Here, using peptide sequences known to bind the PP2A regulatory subunit B56α, we systematically compare affinities measured by MRBLE-pep or isothermal calorimetry (ITC) and confirm that MRBLE-pep accurately quantifies relative affinity over a wide dynamic range while using a fraction of the material required for traditional methods such as ITC.