Ultrahigh Throughput Evolution of Tryptophan Synthase in Droplets via an Aptamer Sensor

Tryptophan synthase catalyzes the synthesis of a wide array of noncanonical amino acids and is an attractive target for directed evolution. Droplet microfluidics offers an ultrahigh throughput approach to directed evolution (up to 107 experiments per day), enabling the search for biocatalysts in wider regions of sequence space with reagent consumption minimized to the picoliter volume (per library member). While the majority of screening campaigns in this format on record relied on an optically active reaction product, a new assay is needed for tryptophan synthase. Tryptophan is not fluorogenic in the visible light spectrum and thus falls outside the scope of conventional droplet microfluidic readouts, which are incompatible with UV light detection at high throughput. Here, we engineer a tryptophan DNA aptamer into a sensor to quantitatively report on tryptophan production in droplets. The utility of the sensor was validated by identifying five-fold improved tryptophan synthases from ∼100,000 protein variants. More generally, this work establishes the use of DNA-aptamer sensors with a fluorogenic read-out in widening the scope of droplet microfluidic evolution.

The biosynthesis, degradation, and function of cell wall β-xylosylated xyloglucan mirrors that of arabinoxyloglucan

Xyloglucan is an abundant polysaccharide in many primary cell walls and in the human diet. Decoration of its α-xylosyl sidechains with further sugars is critical for plant growth, even though the sugars themselves vary considerably between species. Plants in the Ericales order - prevalent in human diets - exhibit β1,2-linked xylosyl decorations. The biosynthetic enzymes responsible for adding these xylosyl decorations, as well as the hydrolases that remove them in the human gut, are unidentified. GT47 xyloglucan glycosyltransferase candidates were expressed in Arabidopsis and endo-xyloglucanase products from transgenic wall material were analysed by electrophoresis, mass spectrometry, and nuclear magnetic resonance (NMR) spectroscopy. The activities of gut bacterial hydrolases BoGH43A and BoGH43B on synthetic glycosides and xyloglucan oligosaccharides were measured by colorimetry and electrophoresis. CcXBT1 is a xyloglucan β-xylosyltransferase from coffee that can modify Arabidopsis xyloglucan and restore the growth of galactosyltransferase mutants. Related VmXST1 is a weakly active xyloglucan α-arabinofuranosyltransferase from cranberry. BoGH43A hydrolyses both α-arabinofuranosylated and β-xylosylated oligosaccharides. CcXBT1's presence in coffee and BoGH43A's promiscuity suggest that β-xylosylated xyloglucan is not only more widespread than thought, but might also nourish beneficial gut bacteria. The evolutionary instability of transferase specificity and lack of hydrolase specificity hint that, to enzymes, xylosides and arabinofuranosides are closely resemblant.

Chemoenzymatic Photoreforming: A Sustainable Approach for Solar Fuel Generation from Plastic Feedstocks

Plastic upcycling through catalytic transformations is an attractive concept to valorize waste, but the clean and energy-efficient production of high-value products from plastics remains challenging. Here, we introduce chemoenzymatic photoreforming as a process coupling enzymatic pretreatment and solar-driven reforming of polyester plastics under mild temperatures and pH to produce clean H2 and value-added chemicals. Chemoenzymatic photoreforming demonstrates versatility in upcycling polyester films and nanoplastics to produce H2 at high yields reaching ∼103-104 μmol gsub-1 and activities at >500 μmol gcat-1 h-1. Enzyme-treated plastics were also used as electron donors for photocatalytic CO2-to-syngas conversion with a phosphonated cobalt bis(terpyridine) catalyst immobilized on TiO2 nanoparticles (TiO2|CotpyP). Finally, techno-economic analyses reveal that the chemoenzymatic photoreforming approach has the potential to drastically reduce H2 production costs to levels comparable to market prices of H2 produced from fossil fuels while maintaining low CO2-equivalent emissions.

spinDrop: a droplet microfluidic platform to maximise single-cell sequencing information content

Droplet microfluidic methods have massively increased the throughput of single-cell sequencing campaigns. The benefit of scale-up is, however, accompanied by increased background noise when processing challenging samples and the overall RNA capture efficiency is lower. These drawbacks stem from the lack of strategies to enrich for high-quality material or specific cell types at the moment of cell encapsulation and the absence of implementable multi-step enzymatic processes that increase capture. Here we alleviate both bottlenecks using fluorescence-activated droplet sorting to enrich for droplets that contain single viable cells, intact nuclei, fixed cells or target cell types and use reagent addition to droplets by picoinjection to perform multi-step lysis and reverse transcription. Our methodology increases gene detection rates fivefold, while reducing background noise by up to half. We harness these properties to deliver a high-quality molecular atlas of mouse brain development, despite starting with highly damaged input material, and provide an atlas of nascent RNA transcription during mouse organogenesis. Our method is broadly applicable to other droplet-based workflows to deliver sensitive and accurate single-cell profiling at a reduced cost.

Versatile Product Detection via Coupled Assays for Ultrahigh-Throughput Screening of Carbohydrate-Active Enzymes in Microfluidic Droplets

Enzyme discovery and directed evolution are the two major contemporary approaches for the improvement of industrial processes by biocatalysis in various fields. Customization of catalysts for improvement of single enzyme reactions or de novo reaction development is often complex and tedious. The success of screening campaigns relies on the fraction of sequence space that can be sampled, whether for evolving a particular enzyme or screening metagenomes. Ultrahigh-throughput screening (uHTS) based on in vitro compartmentalization in water-in-oil emulsion of picoliter droplets generated in microfluidic systems allows screening rates >1 kHz (or >107 per day). Screening for carbohydrate-active enzymes (CAZymes) catalyzing biotechnologically valuable reactions in this format presents an additional challenge because the released carbohydrates are difficult to monitor in high throughput. Activated substrates with large optically active hydrophobic leaving groups provide a generic optical readout, but the molecular recognition properties of sugars will be altered by the incorporation of such fluoro- or chromophores and their typically higher reactivity, as leaving groups with lowered pKa values compared to native substrates make the observation of promiscuous reactions more likely. To overcome these issues, we designed microdroplet assays in which optically inactive carbohydrate products are made visible by specific cascades: the primary reaction of an unlabeled substrate leads to an optical signal downstream. Successfully implementing such assays at the picoliter droplet scale allowed us to detect glucose, xylose, glucuronic acid, and arabinose as final products of complex oligosaccharide degradation by glycoside hydrolases by absorbance measurements. Enabling the use of uHTS for screening CAZyme reactions that have been thus far elusive will chart a route toward faster and easier development of specific and efficient biocatalysts for biovalorization, directing enzyme discovery by challenging catalysts for reaction with natural rather than model substrates.

Plakoglobin is a mechanoresponsive regulator of naive pluripotency

Biomechanical cues are instrumental in guiding embryonic development and cell differentiation. Understanding how these physical stimuli translate into transcriptional programs will provide insight into mechanisms underlying mammalian pre-implantation development. Here, we explore this type of regulation by exerting microenvironmental control over mouse embryonic stem cells. Microfluidic encapsulation of mouse embryonic stem cells in agarose microgels stabilizes the naive pluripotency network and specifically induces expression of Plakoglobin (Jup), a vertebrate homolog of β-catenin. Overexpression of Plakoglobin is sufficient to fully re-establish the naive pluripotency gene regulatory network under metastable pluripotency conditions, as confirmed by single-cell transcriptome profiling. Finally, we find that, in the epiblast, Plakoglobin was exclusively expressed at the blastocyst stage in human and mouse embryos - further strengthening the link between Plakoglobin and naive pluripotency in vivo. Our work reveals Plakoglobin as a mechanosensitive regulator of naive pluripotency and provides a paradigm to interrogate the effects of volumetric confinement on cell-fate transitions.

Age-associated B cells predict impaired humoral immunity after COVID-19 vaccination in patients receiving immune checkpoint blockade

Age-associated B cells (ABC) accumulate with age and in individuals with different immunological disorders, including cancer patients treated with immune checkpoint blockade and those with inborn errors of immunity. Here, we investigate whether ABCs from different conditions are similar and how they impact the longitudinal level of the COVID-19 vaccine response. Single-cell RNA sequencing indicates that ABCs with distinct aetiologies have common transcriptional profiles and can be categorised according to their expression of immune genes, such as the autoimmune regulator (AIRE). Furthermore, higher baseline ABC frequency correlates with decreased levels of antigen-specific memory B cells and reduced neutralising capacity against SARS-CoV-2. ABCs express high levels of the inhibitory FcγRIIB receptor and are distinctive in their ability to bind immune complexes, which could contribute to diminish vaccine responses either directly, or indirectly via enhanced clearance of immune complexed-antigen. Expansion of ABCs may, therefore, serve as a biomarker identifying individuals at risk of suboptimal responses to vaccination.

Generation of liver mesenchyme and ductal cell organoid co-culture using cell self-aggregation and droplet microfluidics

Within the peri-portal region of the adult liver, portal fibroblasts exist in close proximity to epithelial ductal/cholangiocyte cells. However, the cellular interactions between them are poorly understood. Here, we provide two co-culture techniques to incorporate liver portal mesenchyme into ductal cell organoids, which recapitulate aspects of their cellular interactions in vitro. We integrate several techniques from mesenchyme isolation and expansion to co-culture by microfluidic cell co-encapsulation or 2D-Matrigel layer. The protocol is easily adaptable to other cells from other organs. For complete information on the generation and use of this protocol, please refer to Cordero-Espinoza et al.¹.

Ultrahigh-Throughput Enzyme Engineering and Discovery in In Vitro Compartments

Novel and improved biocatalysts are increasingly sourced from libraries via experimental screening. The success of such campaigns is crucially dependent on the number of candidates tested. Water-in-oil emulsion droplets can replace the classical test tube, to provide in vitro compartments as an alternative screening format, containing genotype and phenotype and enabling a readout of function. The scale-down to micrometer droplet diameters and picoliter volumes brings about a >10⁷-fold volume reduction compared to 96-well-plate screening. Droplets made in automated microfluidic devices can be integrated into modular workflows to set up multistep screening protocols involving various detection modes to sort >10⁷ variants a day with kHz frequencies. The repertoire of assays available for droplet screening covers all seven enzyme commission (EC) number classes, setting the stage for widespread use of droplet microfluidics in everyday biochemical experiments. We review the practicalities of adapting droplet screening for enzyme discovery and for detailed kinetic characterization. These new ways of working will not just accelerate discovery experiments currently limited by screening capacity but profoundly change the paradigms we can probe. By interfacing the results of ultrahigh-throughput droplet screening with next-generation sequencing and deep learning, strategies for directed evolution can be implemented, examined, and evaluated.

Ultra-High-Throughput Absorbance-Activated Droplet Sorting for Enzyme Screening at Kilohertz Frequencies

Droplet microfluidics is a valuable method to "beat the odds" in high throughput screening campaigns such as directed evolution, where valuable hits are infrequent and large library sizes are required. Absorbance-based sorting expands the range of enzyme families that can be subjected to droplet screening by expanding possible assays beyond fluorescence detection. However, absorbance-activated droplet sorting (AADS) is currently ∼10-fold slower than typical fluorescence-activated droplet sorting (FADS), meaning that, in comparison, a larger portion of sequence space is inaccessible due to throughput constraints. Here we improve AADS to reach kHz sorting speeds in an order of magnitude increase over previous designs, with close-to-ideal sorting accuracy. This is achieved by a combination of (i) the use of refractive index matching oil that improves signal quality by removal of side scattering (increasing the sensitivity of absorbance measurements); (ii) a sorting algorithm capable of sorting at this increased frequency with an Arduino Due; and (iii) a chip design that transmits product detection better into sorting decisions without false positives, namely a single-layered inlet to space droplets further apart and injections of "bias oil" providing a fluidic barrier preventing droplets from entering the incorrect sorting channel. The updated ultra-high-throughput absorbance-activated droplet sorter increases the effective sensitivity of absorbance measurements through better signal quality at a speed that matches the more established fluorescence-activated sorting devices.

Ultrahigh-Throughput Directed Evolution of a Metal-Free α/β-Hydrolase with a Cys-His-Asp Triad into an Efficient Phosphotriesterase

Finding new mechanistic solutions for biocatalytic challenges is key in the evolutionary adaptation of enzymes, as well as in devising new catalysts. The recent release of man-made substances into the environment provides a dynamic testing ground for observing biocatalytic innovation at play. Phosphate triesters, used as pesticides, have only recently been introduced into the environment, where they have no natural counterpart. Enzymes have rapidly evolved to hydrolyze phosphate triesters in response to this challenge, converging onto the same mechanistic solution, which requires bivalent cations as a cofactor for catalysis. In contrast, the previously identified metagenomic promiscuous hydrolase P91, a homologue of acetylcholinesterase, achieves slow phosphotriester hydrolysis mediated by a metal-independent Cys-His-Asp triad. Here, we probe the evolvability of this new catalytic motif by subjecting P91 to directed evolution. By combining a focused library approach with the ultrahigh throughput of droplet microfluidics, we increase P91's activity by a factor of ≈360 (to a k_cat/K_M of ≈7 × 10⁵ M^-1 s^-1) in only two rounds of evolution, rivaling the catalytic efficiencies of naturally evolved, metal-dependent phosphotriesterases. Unlike its homologue acetylcholinesterase, P91 does not suffer suicide inhibition; instead, fast dephosphorylation rates make the formation of the covalent adduct rather than its hydrolysis rate-limiting. This step is improved by directed evolution, with intermediate formation accelerated by 2 orders of magnitude. Combining focused, combinatorial libraries with the ultrahigh throughput of droplet microfluidics can be leveraged to identify and enhance mechanistic strategies that have not reached high efficiency in nature, resulting in alternative reagents with novel catalytic machineries.

Acoustic sorting of microfluidic droplets at kHz rates using optical absorbance

Droplet microfluidics allows one to address the ever-increasing demand to screen large libraries of biological samples. Absorbance spectroscopy complements the golden standard of fluorescence detection by label free target identification and providing more quantifiable data. However, this is limited by speed and sensitivity. In this paper we increase the speed of sorting by including acoustofluidics, achieving sorting rates of target droplets of 1 kHz. We improved the device design for detection of absorbance using fibre-based interrogation of samples with integrated lenses in the microfluidic PDMS device for focusing and collimation of light. This optical improvement reduces the scattering and refraction artefacts, improving the signal quality and sensitivity. The novel design allows us to overcome limitations based on dielectrophoresis sorting, such as droplet size dependency, material and dielectric properties of samples. Our acoustic activated absorbance sorter removes the need for offset dyes or matching oils and sorts about a magnitude faster than current absorbance sorters.

Microgel culture and spatial identity mapping elucidate the signalling requirements for primate epiblast and amnion formation

The early specification and rapid growth of extraembryonic membranes are distinctive hallmarks of primate embryogenesis. These complex tasks are resolved through an intricate combination of signals controlling the induction of extraembryonic lineages and, at the same time, safeguarding the pluripotent epiblast. Here, we delineate the signals orchestrating primate epiblast and amnion identity. We encapsulated marmoset pluripotent stem cells into agarose microgels and identified culture conditions for the development of epiblast- and amnion-spheroids. Spatial identity mapping authenticated spheroids generated in vitro by comparison with marmoset embryos in vivo. We leveraged the microgel system to functionally interrogate the signalling environment of the post-implantation primate embryo. Single-cell profiling of the resulting spheroids demonstrated that activin/nodal signalling is required for embryonic lineage identity. BMP4 promoted amnion formation and maturation, which was counteracted by FGF signalling. Our combination of microgel culture, single-cell profiling and spatial identity mapping provides a powerful approach to decipher the essential cues for embryonic and extraembryonic lineage formation in primate embryogenesis.

Embryo model completes gastrulation to neurulation and organogenesis

Embryonic stem (ES) cells can undergo many aspects of mammalian embryogenesis in vitro1-5, but their developmental potential is substantially extended by interactions with extraembryonic stem cells, including trophoblast stem (TS) cells, extraembryonic endoderm stem (XEN) cells and inducible XEN (iXEN) cells6-11. Here we assembled stem cell-derived embryos in vitro from mouse ES cells, TS cells and iXEN cells and showed that they recapitulate the development of whole natural mouse embryo in utero up to day 8.5 post-fertilization. Our embryo model displays headfolds with defined forebrain and midbrain regions and develops a beating heart-like structure, a trunk comprising a neural tube and somites, a tail bud containing neuromesodermal progenitors, a gut tube, and primordial germ cells. This complete embryo model develops within an extraembryonic yolk sac that initiates blood island development. Notably, we demonstrate that the neurulating embryo model assembled from Pax6-knockout ES cells aggregated with wild-type TS cells and iXEN cells recapitulates the ventral domain expansion of the neural tube that occurs in natural, ubiquitous Pax6-knockout embryos. Thus, these complete embryoids are a powerful in vitro model for dissecting the roles of diverse cell lineages and genes in development. Our results demonstrate the self-organization ability of ES cells and two types of extraembryonic stem cells to reconstitute mammalian development through and beyond gastrulation to neurulation and early organogenesis.

Vector redesign and in-droplet cell-growth improves enrichment and recovery in live Escherichia coli

Directed evolution (DE) is a widely used method for improving the function of biomolecules via multiple rounds of mutation and selection. Microfluidic droplets have emerged as an important means to screen the large libraries needed for DE, but this approach was so far partially limited by the need to lyse cells, recover DNA, and retransform into cells for the next round, necessitating the use of a high‐copy number plasmid or oversampling. The recently developed live cell recovery avoids some of these limitations by directly regrowing selected cells after sorting. However, repeated sorting cycles used to further enrich the most active variants ultimately resulted in unfavourable recovery of empty plasmid vector‐containing cells over those expressing the protein of interest. In this study, we found that engineering of the original expression vector solved the problem of false positives (i.e. plasmids lacking an insert) cells containing empty vectors. Five approaches to measure activity of cell‐displayed enzymes in microdroplets were compared. By comparing various cell treatment methods prior to droplet sorting two things were found. Substrate encapsulation from the start, that is prior to expression of enzyme, showed no disadvantage to post‐induction substrate addition by pico‐injection with respect to recovery of true positive variants. Furthermore in‐droplet cell growth prior to induction of enzyme production improves the total amount of cells retrieved (recovery) and proportion of true positive variants (enrichment) after droplet sorting.

Spatial profiling of early primate gastrulation in utero

Gastrulation controls the emergence of cellular diversity and axis patterning in the early embryo. In mammals, this transformation is orchestrated by dynamic signalling centres at the interface of embryonic and extraembryonic tissues1-3. Elucidating the molecular framework of axis formation in vivo is fundamental for our understanding of human development4-6 and to advance stem-cell-based regenerative approaches7. Here we illuminate early gastrulation of marmoset embryos in utero using spatial transcriptomics and stem-cell-based embryo models. Gaussian process regression-based 3D transcriptomes delineate the emergence of the anterior visceral endoderm, which is hallmarked by conserved (HHEX, LEFTY2, LHX1) and primate-specific (POSTN, SDC4, FZD5) factors. WNT signalling spatially coordinates the formation of the primitive streak in the embryonic disc and is counteracted by SFRP1 and SFRP2 to sustain pluripotency in the anterior domain. Amnion specification occurs at the boundaries of the embryonic disc through ID1, ID2 and ID3 in response to BMP signalling, providing a developmental rationale for amnion differentiation of primate pluripotent stem cells (PSCs). Spatial identity mapping demonstrates that primed marmoset PSCs exhibit the highest similarity to the anterior embryonic disc, whereas naive PSCs resemble the preimplantation epiblast. Our 3D transcriptome models reveal the molecular code of lineage specification in the primate embryo and provide an in vivo reference to decipher human development.

Stem cell-derived synthetic embryos self-assemble by exploiting cadherin codes and cortical tension

Mammalian embryos sequentially differentiate into trophectoderm and an inner cell mass, the latter of which differentiates into primitive endoderm and epiblast. Trophoblast stem (TS), extraembryonic endoderm (XEN) and embryonic stem (ES) cells derived from these three lineages can self-assemble into synthetic embryos, but the mechanisms remain unknown. Here, we show that a stem cell-specific cadherin code drives synthetic embryogenesis. The XEN cell cadherin code enables XEN cell sorting into a layer below ES cells, recapitulating the sorting of epiblast and primitive endoderm before implantation. The TS cell cadherin code enables TS cell sorting above ES cells, resembling extraembryonic ectoderm clustering above epiblast following implantation. Whereas differential cadherin expression drives initial cell sorting, cortical tension consolidates tissue organization. By optimizing cadherin code expression in different stem cell lines, we tripled the frequency of correctly formed synthetic embryos. Thus, by exploiting cadherin codes from different stages of development, lineage-specific stem cells bypass the preimplantation structure to directly assemble a postimplantation embryo.

Gigavalent Display of Proteins on Monodisperse Polyacrylamide Hydrogels as a Versatile Modular Platform for Functional Assays and Protein Engineering

The assembly of robust, modular biological components into complex functional systems is central to synthetic biology. Here, we apply modular "plug and play" design principles to a solid-phase protein display system that facilitates protein purification and functional assays. Specifically, we capture proteins on polyacrylamide hydrogel display beads (PHD beads) made in microfluidic droplet generators. These monodisperse PHD beads are decorated with predefined amounts of anchors, methacrylate-PEG-benzylguanine (BG) and methacrylate-PEG-chloroalkane (CA), that react covalently with SNAP-/Halo-tag fusion proteins, respectively, in a specific, orthogonal, and stable fashion. Anchors, and thus proteins, are distributed throughout the entire bead volume, allowing attachment of ∼10⁹ protein molecules per bead (⌀ 20 μm) -a higher density than achievable with commercial surface-modified beads. We showcase a diverse array of protein modules that enable the secondary capture of proteins, either noncovalently (IgG and SUMO-tag) or covalently (SpyCatcher, SpyTag, SnpCatcher, and SnpTag), in mono- and multivalent display formats. Solid-phase protein binding and enzymatic assays are carried out, and incorporating the photocleavable protein PhoCl enables the controlled release of modules via visible-light irradiation for functional assays in solution. We utilize photocleavage for valency engineering of an anti-TRAIL-R1 scFv, enhancing its apoptosis-inducing potency ∼50-fold through pentamerization.

Functional metagenomic screening identifies an unexpected β-glucuronidase

The abundance of recorded protein sequence data stands in contrast to the small number of experimentally verified functional annotation. Here we screened a million-membered metagenomic library at ultrahigh throughput in microfluidic droplets for β-glucuronidase activity. We identified SN243, a genuine β-glucuronidase with little homology to previously studied enzymes of this type, as a glycoside hydrolase 3 family member. This glycoside hydrolase family contains only one recently added β-glucuronidase, showing that a functional metagenomic approach can shed light on assignments that are currently 'unpredictable' by bioinformatics. Kinetic analyses of SN243 characterized it as a promiscuous catalyst and structural analysis suggests regions of divergence from homologous glycoside hydrolase 3 members creating a wide-open active site. With a screening throughput of >10⁷ library members per day, picolitre-volume microfluidic droplets enable functional assignments that complement current enzyme database dictionaries and provide bridgeheads for the annotation of unexplored sequence space.

Enzymatic N-Allylation of Primary and Secondary Amines Using Renewable Cinnamic Acids Enabled by Bacterial Reductive Aminases

Allylic amines are a versatile class of synthetic precursors of many valuable nitrogen-containing organic compounds, including pharmaceuticals. Enzymatic allylic amination methods provide a sustainable route to these compounds but are often restricted to allylic primary amines. We report a biocatalytic system for the reductive N-allylation of primary and secondary amines, using biomass-derivable cinnamic acids. The two-step one-pot system comprises an initial carboxylate reduction step catalyzed by a carboxylic acid reductase to generate the corresponding α,β-unsaturated aldehyde in situ. This is followed by reductive amination of the aldehyde catalyzed by a bacterial reductive aminase pIR23 or BacRedAm to yield the corresponding allylic amine. We exploited pIR23, a prototype bacterial reductive aminase, self-sufficient in catalyzing formal reductive amination of α,β-unsaturated aldehydes with various amines, generating a broad range of secondary and tertiary amines accessed in up to 94% conversion under mild reaction conditions. Analysis of products isolated from preparative reactions demonstrated that only selective hydrogenation of the C=N bond had occurred, preserving the adjacent alkene moiety. This process represents an environmentally benign and sustainable approach for the synthesis of secondary and tertiary allylic amine frameworks, using renewable allylating reagents and avoiding harsh reaction conditions. The selectivity of the system ensures that bis-allylation of the alkylamines and (over)reduction of the alkene moiety are avoided.

Droplet-based screening of phosphate transfer catalysis reveals how epistasis shapes MAP kinase interactions with substrates

The combination of ultrahigh-throughput screening and sequencing informs on function and intragenic epistasis within combinatorial protein mutant libraries. Establishing a droplet-based, in vitro compartmentalised approach for robust expression and screening of protein kinase cascades (>10⁷ variants/day) allowed us to dissect the intrinsic molecular features of the MKK-ERK signalling pathway, without interference from endogenous cellular components. In a six-residue combinatorial library of the MKK1 docking domain, we identified 29,563 sequence permutations that allow MKK1 to efficiently phosphorylate and activate its downstream target kinase ERK2. A flexibly placed hydrophobic sequence motif emerges which is defined by higher order epistatic interactions between six residues, suggesting synergy that enables high connectivity in the sequence landscape. Through positive epistasis, MKK1 maintains function during mutagenesis, establishing the importance of co-dependent residues in mammalian protein kinase-substrate interactions, and creating a scenario for the evolution of diverse human signalling networks.

Here, the authors use a droplet-based screen for phosphate transfer catalysis, testing variants of the human protein kinase MKK1 for its ability to activate its downstream target ERK2. Data reveal a flexible motif in the MKK1 docking domain that promotes efficient activation of ERK2, and suggest epistasis between the residues within that sequence.

High-throughput total RNA sequencing in single cells using VASA-seq

Most methods for single-cell transcriptome sequencing amplify the termini of polyadenylated transcripts, capturing only a small fraction of the total cellular transcriptome. This precludes the detection of many long non-coding, short non-coding and non-polyadenylated protein-coding transcripts and hinders alternative splicing analysis. We, therefore, developed VASA-seq to detect the total transcriptome in single cells, which is enabled by fragmenting and tailing all RNA molecules subsequent to cell lysis. The method is compatible with both plate-based formats and droplet microfluidics. We applied VASA-seq to more than 30,000 single cells in the developing mouse embryo during gastrulation and early organogenesis. Analyzing the dynamics of the total single-cell transcriptome, we discovered cell type markers, many based on non-coding RNA, and performed in vivo cell cycle analysis via detection of non-polyadenylated histone genes. RNA velocity characterization was improved, accurately retracing blood maturation trajectories. Moreover, our VASA-seq data provide a comprehensive analysis of alternative splicing during mammalian development, which highlighted substantial rearrangements during blood development and heart morphogenesis.

Improved RAD51 binders through motif shuffling based on the modularity of BRC repeats

Exchanges of protein sequence modules support leaps in function unavailable through point mutations during evolution. Here we study the role of the two RAD51-interacting modules within the eight binding BRC repeats of BRCA2. We created 64 chimeric repeats by shuffling these modules and measured their binding to RAD51. We found that certain shuffled module combinations were stronger binders than any of the module combinations in the natural repeats. Surprisingly, the contribution from the two modules was poorly correlated with affinities of natural repeats, with a weak BRC8 repeat containing the most effective N-terminal module. The binding of the strongest chimera, BRC8-2, to RAD51 was improved by -2.4 kCal/mol compared to the strongest natural repeat, BRC4. A crystal structure of RAD51:BRC8-2 complex shows an improved interface fit and an extended β-hairpin in this repeat. BRC8-2 was shown to function in human cells, preventing the formation of nuclear RAD51 foci after ionizing radiation.

Ultrahigh-Throughput Detection of Enzymatic Alcohol Dehydrogenase Activity in Microfluidic Droplets with a Direct Fluorogenic Assay

The exploration of large DNA libraries of metagenomic or synthetic origin is greatly facilitated by ultrahigh‐throughput assays that use monodisperse water‐in‐oil emulsion droplets as sequestered reaction compartments. Millions of samples can be generated and analysed in microfluidic devices at kHz speeds, requiring only micrograms of reagents. The scope of this powerful platform for the discovery of new sequence space is, however, hampered by the limited availability of assay substrates, restricting the functions and reaction types that can be investigated. Here, we broaden the scope of detectable biochemical transformations in droplet microfluidics by introducing the first fluorogenic assay for alcohol dehydrogenases (ADHs) in this format. We have synthesized substrates that release a pyranine fluorophore (8‐hydroxy‐1,3,6‐pyrenetrisulfonic acid, HPTS) when enzymatic turnover occurs. Pyranine is well retained in droplets for >6 weeks (i. e. 14‐times longer than fluorescein), avoiding product leakage and ensuring excellent assay sensitivity. Product concentrations as low as 100 nM were successfully detected, corresponding to less than one turnover per enzyme molecule on average. The potential of our substrate design was demonstrated by efficient recovery of a bona fide ADH with an >800‐fold enrichment. The repertoire of droplet screening is enlarged by this sensitive and direct fluorogenic assay to identify dehydrogenases for biocatalytic applications.

A Titratable Cell Lysis-on-Demand System for Droplet-Compartmentalized Ultrahigh-Throughput Screening in Functional Metagenomics and Directed Evolution

Water-in-oil emulsion droplets are an attractive format for ultrahigh-throughput screening in functional metagenomics and directed evolution applications that allow libraries with more than 10⁷ members to be characterized in a day. Single library members are compartmentalized in droplets that are generated in microfluidic devices and tested for the presence of target biocatalysts. The target proteins can be produced intracellularly, for example, in bacterial hosts in-droplet cell lysis is therefore necessary to allow the enzymes to encounter the substrate to initiate an activity assay. Here, we present a titratable lysis-on-demand (LoD) system enabling the control of the cell lysis rate in Escherichia coli. We demonstrate that the rate of cell lysis can be controlled by adjusting the externally added inducer concentration. This LoD system is evaluated both at the population level (by optical density measurements) and at the single-cell level (on single-cell arrays and in alginate microbeads). Additionally, we validate the LoD system by droplet screening of a phosphotriesterase expressed from E. coli, with cell lysis triggered by inducer concentrations in the μM range. The LoD system yields sufficient release of the intracellularly produced enzymes to bring about a detectable quantity of product (measured by fluorescence in flow cytometry of double emulsions), while leaving viable cells for the downstream recovery of the genetic material.

Dynamic cell contacts between periportal mesenchyme and ductal epithelium act as a rheostat for liver cell proliferation

In the liver, ductal cells rarely proliferate during homeostasis but do so transiently after tissue injury. These cells can be expanded as organoids that recapitulate several of the cell-autonomous mechanisms of regeneration but lack the stromal interactions of the native tissue. Here, using organoid co-cultures that recapitulate the ductal-to-mesenchymal cell architecture of the portal tract, we demonstrate that a subpopulation of mouse periportal mesenchymal cells exerts dual control on proliferation of the epithelium. Ductal cell proliferation is either induced and sustained or, conversely, completely abolished, depending on the number of direct mesenchymal cell contacts, through a mechanism mediated, at least in part, by Notch signaling. Our findings expand the concept of the cellular niche in epithelial tissues, whereby not only soluble factors but also cell-cell contacts are the key regulatory cues involved in the control of cellular behaviors, suggesting a critical role for cell-cell contacts during regeneration.

Engineering the protein dynamics of an ancestral luciferase

Protein dynamics are often invoked in explanations of enzyme catalysis, but their design has proven elusive. Here we track the role of dynamics in evolution, starting from the evolvable and thermostable ancestral protein AncHLD-RLuc which catalyses both dehalogenase and luciferase reactions. Insertion-deletion (InDel) backbone mutagenesis of AncHLD-RLuc challenged the scaffold dynamics. Screening for both activities reveals InDel mutations localized in three distinct regions that lead to altered protein dynamics (based on crystallographic B-factors, hydrogen exchange, and molecular dynamics simulations). An anisotropic network model highlights the importance of the conformational flexibility of a loop-helix fragment of Renilla luciferases for ligand binding. Transplantation of this dynamic fragment leads to lower product inhibition and highly stable glow-type bioluminescence. The success of our approach suggests that a strategy comprising (i) constructing a stable and evolvable template, (ii) mapping functional regions by backbone mutagenesis, and (iii) transplantation of dynamic features, can lead to functionally innovative proteins.

Agarose microgel culture delineates lumenogenesis in naive and primed human pluripotent stem cells

Human periimplantation development requires the transformation of the naive pluripotent epiblast into a polarized epithelium. Lumenogenesis plays a critical role in this process, as the epiblast undergoes rosette formation and lumen expansion to form the amniotic cavity. Here, we present a high-throughput in vitro model for epiblast morphogenesis. We established a microfluidic workflow to encapsulate human pluripotent stem cells (hPSCs) into monodisperse agarose microgels. Strikingly, hPSCs self-organized into polarized epiblast spheroids that could be maintained in self-renewing and differentiating conditions. Encapsulated primed hPSCs required Rho-associated kinase inhibition, in contrast to naive hPSCs. We applied microgel suspension culture to examine the lumen-forming capacity of hPSCs and reveal an increase in lumenogenesis during the naive-to-primed transition. Finally, we demonstrate the feasibility of co-encapsulating cell types across different lineages and species. Our work provides a foundation for stem cell-based embryo models to interrogate the critical components of human epiblast self-organization and morphogenesis.

Controlled Ligand Exchange Between Ruthenium Organometallic Cofactor Precursors and a Naïve Protein Scaffold Generates Artificial Metalloenzymes Catalysing Transfer Hydrogenation

Many natural metalloenzymes assemble from proteins and biosynthesised complexes, generating potent catalysts by changing metal coordination. Here we adopt the same strategy to generate artificial metalloenzymes (ArMs) using ligand exchange to unmask catalytic activity. By systematically testing RuII (η6 -arene)(bipyridine) complexes designed to facilitate the displacement of functionalised bipyridines, we develop a fast and robust procedure for generating new enzymes via ligand exchange in a protein that has not evolved to bind such a complex. The resulting metal cofactors form peptidic coordination bonds but also retain a non-biological ligand. Tandem mass spectrometry and 19 F NMR spectroscopy were used to characterise the organometallic cofactors and identify the protein-derived ligands. By introduction of ruthenium cofactors into a 4-helical bundle, transfer hydrogenation catalysts were generated that displayed a 35-fold rate increase when compared to the respective small molecule reaction in solution.

"NAD-display": Ultrahigh-Throughput in Vitro Screening of NAD(H) Dehydrogenases Using Bead Display and Flow Cytometry

NAD(H)‐utiliing enzymes have been the subject of directed evolution campaigns to improve their function. To enable access to a larger swath of sequence space, we demonstrate the utility of a cell‐free, ultrahigh‐throughput directed evolution platform for dehydrogenases. Microbeads (1.5 million per sample) carrying both variant DNA and an immobilised analogue of NAD⁺ were compartmentalised in water‐in‐oil emulsion droplets, together with cell‐free expression mixture and enzyme substrate, resulting in the recording of the phenotype on each bead. The beads’ phenotype could be read out and sorted for on a flow cytometer by using a highly sensitive fluorescent protein‐based sensor of the NAD⁺:NADH ratio. Integration of this “NAD‐display” approach with our previously described Split & Mix (SpliMLiB) method for generating large site‐saturation libraries allowed straightforward screening of fully balanced site saturation libraries of formate dehydrogenase, with diversities of 2×10⁴. Based on modular design principles of synthetic biology NAD‐display offers access to sophisticated in vitro selections, avoiding complex technology platforms.

Error-Free Synthetic DNA by Molecular Dictation

Synthetic DNA is the linchpin of the rapidly accelerating biotechnological era and is perhaps the most promising candidate for long-term digital data storage. Despite huge advances, manufacturing error-free DNA at low cost and high throughput remains challenging. Borrowing from well-established sequencing-by-synthesis technologies, we describe a new solution for DNA error correction.

Cell-free Directed Evolution of a Protease in Microdroplets at Ultrahigh Throughput

Compartmentalization of single genes in water-in-oil emulsion droplets is a powerful approach to create millions of reactors for enzyme library selections. When these droplets are formed at ultrahigh throughput in microfluidic devices, their perfect monodispersity allows quantitative enzyme assays with a high precision readout. However, despite its potential for high quality cell-free screening experiments, previous demonstrations of enrichment have never been successfully followed up by actual enzyme library selections in monodisperse microfluidic droplets. Here we develop a three-step workflow separating three previously incompatible steps that thus far could not be carried out at once: first droplet-compartmentalized DNA is amplified by rolling circle amplification; only after completion of this step are reagents for in vitro protein expression and, finally, substrate added via picoinjection. The segmented workflow is robust enough to allow the first in vitro evolution in droplets, improving the protease Savinase that is toxic to E. coli for higher activity and identifying a 5-fold faster enzyme.

Multiplexed Affinity Characterization of Protein Binders Directly from a Crude Cell Lysate by Covalent Capture on Suspension Bead Arrays

The precise determination of affinity and specificity is a crucial step in the development of new protein reagents for therapy and diagnostics. Paradoxically, the selection of protein binders, e.g., antibody fragments, from large combinatorial repertoires is a rapid process compared to the subsequent characterization of selected clones. Here we demonstrate the use of suspension bead arrays (SBA) in combination with flow cytometry to facilitate the post-selection analysis of binder affinities. The array is designed to capture the proteins of interest (POIs) covalently on the surface of superparamagnetic color-coded microbeads directly from expression cell lysate, based on SpyTag-SpyCatcher coupling by isopeptide bond formation. This concept was validated by analyzing the affinities of a typical phage display output, i.e., clones consisting of single-chain variable fragment antibodies (scFvs), as SpyCatcher fusions in 12- and 24-plex SBA formats using a standard three-laser flow cytometer. We demonstrate that the equilibrium dissociation constants (K_d) obtained from multiplexed SBA assays correlate well with experiments performed on a larger scale, while the antigen consumption was reduced >100-fold compared to the conventional 96-well plate format. Protein screening and characterization by SBAs is a rapid and reagent-saving analytical format for combinatorial protein engineering to address specificity maturation and cross-reactivity profiling of antibodies.

Growth amplification in ultrahigh-throughput microdroplet screening increases sensitivity of clonal enzyme assays and minimizes phenotypic variation

Microfluidic ultrahigh-throughput screening of enzyme activities provides information on libraries with millions of variants in a day. Each individual library member is represented by a recombinant single cell, compartmentalised in an emulsion droplet, in which an activity assay is carried out. Key to the success of this approach is the precision and sensitivity of the assay. Assay quality is most profoundly challenged when initially weak, promiscuous activities are to be enhanced in early rounds of directed evolution or when entirely novel catalysts are to be identified from metagenomic sources. Implementation of measures to widen the dynamic range of clonal assays would increase the chances of finding and generating new biocatalysts. Here, we demonstrate that the assay sensitivity and DNA recovery can be improved by orders of magnitude by growth of initially singly compartmentalised cells in microdroplets. Homogeneous cell growth is achieved by continuous oxygenation and recombinant protein expression is regulated by diffusion of an inducer from the oil phase. Reaction conditions are adjusted by directed droplet coalescence to enable full control of buffer composition and kinetic incubation time, creating level playing field conditions for library selections. The clonal amplification multiplies the product readout because more enzyme is produced per compartment. At the same time, phenotypic variation is reduced by measuring monoclonal populations rather than single cells and recovery efficiency is increased. Consequently, this workflow increases the efficiency of lysate-based microfluidic enzyme assays and will make it easier for protein engineers to identify or evolve new enzymes for applications in synthetic and chemical biology.

Inducible Stem-Cell-Derived Embryos Capture Mouse Morphogenetic Events In Vitro

The development of mouse embryos can be partially recapitulated by combining embryonic stem cells (ESCs), trophoblast stem cells (TS), and extra-embryonic endoderm (XEN) stem cells to generate embryo-like structures called ETX embryos. Although ETX embryos transcriptionally capture the mouse gastrula, their ability to recapitulate complex morphogenic events such as gastrulation is limited, possibly due to the limited potential of XEN cells. To address this, we generated ESCs transiently expressing transcription factor Gata4, which drives the extra-embryonic endoderm fate, and combined them with ESCs and TS cells to generate induced ETX embryos (iETX embryos). We show that iETX embryos establish a robust anterior signaling center that migrates unilaterally to break embryo symmetry. Furthermore, iETX embryos gastrulate generating embryonic and extra-embryonic mesoderm and definitive endoderm. Our findings reveal that replacement of XEN cells with ESCs transiently expressing Gata4 endows iETX embryos with greater developmental potential, thus enabling the study of the establishment of anterior-posterior patterning and gastrulation in an in vitro system.

•

Stem cells generate mouse-embryo-like structures with improved potential

•

These structures undertake anterior visceral endoderm formation and gastrulation

•

Single-cell sequencing shows improved resemblance to mouse embryo

UMI-linked consensus sequencing enables phylogenetic analysis of directed evolution

The success of protein evolution campaigns is strongly dependent on the sequence context in which mutations are introduced, stemming from pervasive non-additive interactions between a protein's amino acids ('intra-gene epistasis'). Our limited understanding of such epistasis hinders the correct prediction of the functional contributions and adaptive potential of mutations. Here we present a straightforward unique molecular identifier (UMI)-linked consensus sequencing workflow (UMIC-seq) that simplifies mapping of evolutionary trajectories based on full-length sequences. Attaching UMIs to gene variants allows accurate consensus generation for closely related genes with nanopore sequencing. We exemplify the utility of this approach by reconstructing the artificial phylogeny emerging in three rounds of directed evolution of an amine dehydrogenase biocatalyst via ultrahigh throughput droplet screening. Uniquely, we are able to identify lineages and their founding variant, as well as non-additive interactions between mutations within a full gene showing sign epistasis. Access to deep and accurate long reads will facilitate prediction of key beneficial mutations and adaptive potential based on in silico analysis of large sequence datasets.

Investigating host-microbiome interactions by droplet based microfluidics

BACKGROUND

Despite the importance of the mucosal interface between microbiota and the host in gut homeostasis, little is known about the mechanisms of bacterial gut colonization, involving foraging for glycans produced by epithelial cells. The slow pace of progress toward understanding the underlying molecular mechanisms is largely due to the lack of efficient discovery tools, especially those targeting the uncultured fraction of the microbiota.

RESULTS

Here, we introduce an ultra-high-throughput metagenomic approach based on droplet microfluidics, to screen fosmid libraries. Thousands of bacterial genomes can be covered in 1 h of work, with less than ten micrograms of substrate. Applied to the screening of the mucosal microbiota for β-N-acetylgalactosaminidase activity, this approach allowed the identification of pathways involved in the degradation of human gangliosides and milk oligosaccharides, the structural homologs of intestinal mucin glycans. These pathways, whose prevalence is associated with inflammatory bowel diseases, could be the result of horizontal gene transfers with Bacteroides species. Such pathways represent novel targets to study the microbiota-host interactions in the context of inflammatory bowel diseases, in which the integrity of the mucosal barrier is impaired.

CONCLUSION

By compartmentalizing experiments inside microfluidic droplets, this method speeds up and miniaturizes by several orders of magnitude the screening process compared to conventional approaches, to capture entire metabolic pathways from metagenomic libraries. The method is compatible with all types of (meta)genomic libraries, and employs a commercially available flow cytometer instead of a custom-made sorting system to detect intracellular or extracellular enzyme activities. This versatile and generic workflow will accelerate experimental exploration campaigns in functional metagenomics and holobiomics studies, to further decipher host-microbiota relationships. Video Abstract.

Ultrahigh throughput screening for enzyme function in droplets

Water-in-oil droplets, made and handled in microfluidic devices, provide a new experimental format, in which ultrahigh throughput experiments can be conducted faster and with minimal reagent consumption. An increasing number of studies have emerged that applied this approach to directed evolution and metagenomic screening of enzyme catalysts. Here, we review the considerations necessary to implement robust workflows, based on choices of device design, detection modes, emulsion formulations and substrates, and scope out which enzyme classes have become amenable to droplet screening.

Microfluidic platform for 3D cell culture with live imaging and clone retrieval

Combining live imaging with the ability to retrieve individual cells of interest remains a technical challenge. Combining imaging with precise cell retrieval is of particular interest when studying highly dynamic or transient, asynchronous, or heterogeneous cell biological and developmental processes. Here, we present a method to encapsulate live cells in a 3D hydrogel matrix, via hydrogel bead compartmentalisation. Using a small-scale screen, we optimised matrix conditions for the culture and multilineage differentiation of mouse embryonic stem cells. Moreover, we designed a custom microfluidic platform that is compatible with live imaging. With this platform we can long-term culture and subsequently extract individual cells-in-beads by media flow only, obviating the need for enzymatic cell removal from the platform. Specific beads may be extracted from the platform in isolation, without disrupting the adjacent beads. We show that we can differentiate mouse embryonic stem cells, monitor reporter expression by live imaging, and retrieve individual beads for functional assays, correlating reporter expression with functional response. Overall, we present a highly flexible 3D cell encapsulation and microfluidic platform that enables both monitoring of cellular dynamics and retrieval for molecular and functional assays.

Accessing unexplored regions of sequence space in directed enzyme evolution via insertion/deletion mutagenesis

Insertions and deletions (InDels) are frequently observed in natural protein evolution, yet their potential remains untapped in laboratory evolution. Here we introduce a transposon-based mutagenesis approach (TRIAD) to generate libraries of random variants with short in-frame InDels, and screen TRIAD libraries to evolve a promiscuous arylesterase activity in a phosphotriesterase. The evolution exhibits features that differ from previous point mutagenesis campaigns: while the average activity of TRIAD variants is more compromised, a larger proportion has successfully adapted for the activity. Different functional profiles emerge: (i) both strong and weak trade-off between activities are observed; (ii) trade-off is more severe (20- to 35-fold increased kcat/KM in arylesterase with 60-400-fold decreases in phosphotriesterase activity) and (iii) improvements are present in kcat rather than just in KM, suggesting adaptive solutions. These distinct features make TRIAD an alternative to widely used point mutagenesis, accessing functional innovations and traversing unexplored fitness landscape regions.

Split & mix assembly of DNA libraries for ultrahigh throughput on-bead screening of functional proteins

Site-saturation libraries reduce protein screening effort in directed evolution campaigns by focusing on a limited number of rationally chosen residues. However, uneven library synthesis efficiency leads to amino acid bias, remedied at high cost by expensive custom synthesis of oligonucleotides, or through use of proprietary library synthesis platforms. To address these shortcomings, we have devised a method where DNA libraries are constructed on the surface of microbeads by ligating dsDNA fragments onto growing, surface-immobilised DNA, in iterative split-and-mix cycles. This method-termed SpliMLiB for Split-and-Mix Library on Beads-was applied towards the directed evolution of an anti-IgE Affibody (ZIgE), generating a 160,000-membered, 4-site, saturation library on the surface of 8 million monoclonal beads. Deep sequencing confirmed excellent library balance (5.1% ± 0.77 per amino acid) and coverage (99.3%). As SpliMLiB beads are monoclonal, they were amenable to direct functional screening in water-in-oil emulsion droplets with cell-free expression. A FACS-based sorting of the library beads allowed recovery of hits improved in Kd over wild-type ZIgE by up to 3.5-fold, while a consensus mutant of the best hits provided a 10-fold improvement. With SpliMLiB, directed evolution workflows are accelerated by integrating high-quality DNA library generation with an ultra-high throughput protein screening platform.

Self-Organization of Mouse Stem Cells into an Extended Potential Blastoid

Mammalian blastocysts comprise three distinct cell lineages essential for development beyond implantation: the pluripotent epiblast, which generates the future embryo, and surrounding it the extra-embryonic primitive endoderm and the trophectoderm tissues. Embryonic stem cells can reintegrate into embryogenesis but contribute primarily to epiblast lineages. Here, we show that mouse embryonic stem cells cultured under extended pluripotent conditions (EPSCs) can be partnered with trophoblast stem cells to self-organize into blastocyst-like structures with all three embryonic and extra-embryonic lineages. Morphogenetic and transcriptome profiling analyses reveal that these blastocyst-like structures show distinct embryonic-abembryonic axes and primitive endoderm differentiation and can initiate the transition from the pre- to post-implantation egg cylinder morphology in vitro.

Global fitness landscapes of the Shine-Dalgarno sequence

Shine-Dalgarno sequences (SD) in prokaryotic mRNA facilitate protein translation by pairing with rRNA in ribosomes. Although conventionally defined as AG-rich motifs, recent genomic surveys reveal great sequence diversity, questioning how SD functions. Here, we determined the molecular fitness (i.e., translation efficiency) of 4⁹ synthetic 9-nt SD genotypes in three distinct mRNA contexts in Escherichia coli. We uncovered generic principles governing the SD fitness landscapes: (1) Guanine contents, rather than canonical SD motifs, best predict the fitness of both synthetic and endogenous SD; (2) the genotype-fitness correlation of SD promotes its evolvability by steadily supplying beneficial mutations across fitness landscapes; and (3) the frequency and magnitude of deleterious mutations increase with background fitness, and adjacent nucleotides in SD show stronger epistasis. Epistasis results from disruption of the continuous base pairing between SD and rRNA. This “chain-breaking” epistasis creates sinkholes in SD fitness landscapes and may profoundly impact the evolution and function of prokaryotic translation initiation and other RNA-mediated processes. Collectively, our work yields functional insights into the SD sequence variation in prokaryotic genomes, identifies a simple design principle to guide bioengineering and bioinformatic analysis of SD, and illuminates the fundamentals of fitness landscapes and molecular evolution.

Deep learning guided image-based droplet sorting for on-demand selection and analysis of single cells and 3D cell cultures

Uncovering the heterogeneity of cellular populations and multicellular constructs is a long-standing goal in fields ranging from antimicrobial resistance to cancer research. Emerging technology platforms such as droplet microfluidics hold the promise to decipher such heterogeneities at ultra-high-throughput. However, there is a lack of methods able to rapidly identify and isolate single cells or 3D cell cultures. Here we demonstrate that deep neural networks can accurately classify single droplet images in real-time based on the presence and number of micro-objects including single mammalian cells and multicellular spheroids. This approach also enables the identification of specific objects within mixtures of objects of different types and sizes. The training sets for the neural networks consisted of a few hundred images manually picked and augmented to up to thousands of images per training class. Training required less than 10 minutes using a single GPU, and yielded accuracies of over 90% for single mammalian cell identification. Crucially, the same model could be used to classify different types of objects such as polystyrene spheres, polyacrylamide beads and MCF-7 cells. We applied the developed method for the selection of 3D cell cultures generated with Hek293FT cells encapsulated in agarose gel beads, highlighting the potential of the technology for the selection of objects with a high diversity of visual appearances. The real-time sorting of single droplets was in-line with droplet generation and occurred at rates up to 40 per second independently of image size up to 480 × 480 pixels. The presented microfluidic device also enabled storage of sorted droplets to allow for downstream analyses.