Droplet flow cytometry for single-cell analysis

The interrogation of single cells has revolutionised biology and medicine by providing crucial unparalleled insights into cell-to-cell heterogeneity. Flow cytometry (including fluorescence-activated cell sorting) is one of the most versatile and high-throughput approaches for single-cell analysis by detecting multiple fluorescence parameters of individual cells in aqueous suspension as they flow past through a focus of excitation lasers. However, this approach relies on the expression of cell surface and intracellular biomarkers, which inevitably lacks spatial and temporal phenotypes and activities of cells, such as secreted proteins, extracellular metabolite production, and proliferation. Droplet microfluidics has recently emerged as a powerful tool for the encapsulation and manipulation of thousands to millions of individual cells within pico-litre microdroplets. Integrating flow cytometry with microdroplet architectures surrounded by aqueous solutions (e.g., water-in-oil-in-water (W/O/W) double emulsion and hydrogel droplets) opens avenues for new cellular assays linking cell phenotypes to genotypes at the single-cell level. In this review, we discuss the capabilities and applications of droplet flow cytometry (DFC). This unique technique uses standard commercially available flow cytometry instruments to characterise or select individual microdroplets containing single cells of interest. We explore current challenges associated with DFC and present our visions for future development.


Introduction
Single-cell analysis plays an essential role in revealing the differences of individual cells in structure and function. 1,2 The adaption of single-cell techniques has highlighted critical cellto-cell heterogeneity, identied rare sub-populations of functional importance and discovered unique characteristics of individual cells. 3,4 Flow cytometry (FC) has been widely used as a versatile and powerful technique to interrogate single cells, offering automated, quantitative, high-throughput and multiparameter characterisation of cell phenotypes and activities. 5,6 In this technique, individual cells suspended in a uid ow are directed to pass through a laser beam one by one and then measured based on the resulting scatter or emission of light energy from uorescence molecules. By adding cell sorting functionality, uorescence-activated cell sorting (FACS, a specialised type of FC) 7,8 can selectively isolate individual cells of particular interest from heterogeneous populations into separate containers (e.g., tubes, well plates or other collection containers), allowing subsequent experiments on the selected cells. Single-cell analysis by FC (including FACS) has enabled signicant advances in life and biomedical sciences, clinical diagnosis and drug development. [9][10][11] However, FC-based single-cell analysis is exclusively limited to readouts from uorescent molecules tagged either within the cells or on the cell surface. The attributes of individual cells are directly quantied and assessed by the light emitted by the uorophore, which indicates the number of antibodies present within or on the cell. Unfortunately, this largely excludes important extracellular biomarkers and cell phenotypes and activities that are spatially and temporally important, such as secreted proteins, extracellular metabolite production, and proliferation. For example, the proteins secreted by cancer cells (i.e., secretomes) could represent putative tumour biomarkers or therapeutic targets. 12 Metabolites produced by industrial microorganisms have great values in nutrition, sustainable agriculture and healthcare sectors; 13 real-time cell proliferation assays are essential in drug discovery and clinical evaluation. 14,15 Therefore, there is a need to develop strategies that enable FC to make comprehensive measurements of cells, including extracellular, spatial and temporal biomarkers of biological, industrial and clinical signicance.
Droplet microuidics 16,17 has emerged as an incredible tool in this effort, which can compartmentalise individual cells in pico-litre microdroplets containing a highly reproducible volume of a reaction mixture. Unlike conventional water-in-oil (W/O) droplets, two main architectures of droplets suspended in aqueous solutions are compatible with FC: water-in-oil-inwater (W/O/W) 18 double emulsion (DE) droplets and hydrogel droplets. 19 The pairing of FC and droplets allows a complete characterisation of individual cells beyond the conventional surface and intracellular biomarkers, featuring high throughput, sensitivity and dynamic range, and low cost. Moreover, it enables tandem genomic, epigenomic, or transcriptomic analyses on isolated cells, 20 allowing integrative single-cell analysis techniques. This integrated approach has helped a wide range of discoveries, including directed evolution of enzymes and proteins, [21][22][23][24] probing cellular heterogeneity in response to drug treatment, 25,26 recognition of rare cells in microbial communities, 25,[27][28][29][30] determination of antibiotic resistance genes, 31,32 and identication of biomarkers linked to diseases. 33,34 Although uorescence-activated droplet sorting (FADS) 35,36 and other variants of the technique (e.g., sequentially addressable dielectrophoretic array (SADA) 37,38 and printed droplet microuidics (PDM) 39 ) allow on-chip screening and sorting of W/O droplets based on uorescent readouts of in-droplet assays, the high complexity of custom devices and instruments involved limit their widespread adoptions. 35,40 Moreover, they can only simultaneously measure 1 or 2 uorescent parameters and requires relatively slow rates (i.e., 0.1-2 kHz, two orders of magnitude slower than FACS) for high-accuracy sorting. 40 Besides uorescence, other optical detection techniques, such as adsorption spectroscopy 41 and Raman spectroscopy, 42 have been applied to characterise and select droplets encapsulated with individual cells in a microuidic chip. However, these techniques suffer from low specicity (adsorption spectroscopy) and low sensitivity (Raman spectroscopy), making them unpopular in high-throughput droplet screening workows.
The article reviews the current achievements in use of FC (including FACS) to screen and isolate microdroplets encapsulated with single cells of interest. It should be noted that we do not highlight advances in FADS and other techniques for screening and sorting of cell-laden W/O droplets on a micro-uidic chip. As shown in Fig. 1, different types of single cells (e.g., mammalian, bacterial and yeast cells) are encapsulated individually within microdroplets (either W/O/W DE droplets or hydrogel droplets), which can be screened and sorted by commercial FC machines, and further analysed at the downstream (e.g., sequencing). This integrated technology, so called droplet ow cytometry (DFC), allows a variety of biochemical assays at the single-cell level, such as single-cell cultivation, molecular evolution, single-cell detection and cell-to-cell interactions, due to droplet monodispersity, cell compartmentalisation and high-throughput processing. Also, this opens avenues for broad applications in the elds of drug discovery, metabolic engineering, medical diagnosis and cell biology.
Specically, this review is composed of the following sections. First, Section 2 introduces two main architectures of emulsion microdroplets compatible with FC, namely W/O/W and hydrogel droplets. Next, Section 3 summarised the main capabilities of integrating emulsion microdroplets and FC techniques (so-called DFC) for high-throughput single-cell analysis. Also, Section 4 discusses critical eld applications of DFC. Furthermore, Section 5 discusses current challenges confronted by DFC with an eye towards future development. Finally, Section 6 concludes the review.
2. Architectures of emulsion microdroplets that are compatible with flow cytometry An FC system is typically equipped with a uidic system for cell handling, consisting of sheath uid (e.g., a buffered saline solution) to deliver and focus cells one by one in a narrow uniform stream. It enables cells to be equally illuminated, detected and measured as they pass through the laser intercept or interrogation point. Since FC requires cells to be suspended in an aqueous sheath uid, it is not compatible with conventional W/O droplets having an insulating oil surrounding the aqueous core. Fortuitously, two main formats of droplets suspended in aqueous solutions, W/O/W DE droplets and hydrogel droplets, provide alternate droplet architectures compatible with FC. Individual cells can be encapsulated, cultured, subjected to biochemical assays in droplets; the cell-laden microdroplets can be detected as individual events by the cytometers and then screened or selected. Flow cytometric analysis of single-cell laden microdroplets allows a detailed study of cellular characteristics at the single-cell level by combining high throughput and rapid analysis with a relatively low reagent consumption and a high dynamic detection range.

Water-in-oil-in-water (W/O/W) double emulsion (DE) droplets
Conventional emulsion droplets can be either water-in-oil (W/ O) or oil-in-water (O/W) single emulsions. There also can be more complex systems, such as W/O/W DE droplets, in which a W/O emulsion is dispersed in a second continuous aqueous phase. 43,44 Therefore, W/O/W DEs can be viewed as two aqueous phases separated by a thin oil layer, stabilised by a monolayer of molecules of emulsier or emulsifying agents at the interface between water and oil. Wettability plays a vital role in the types of emulsion droplets generated, as hydrophobic and hydrophilic surfaces are required to generate W/O and O/W emulsions, respectively. [45][46][47] By selectively controlling the wettability of microchannels, microuidic devices enabled a hierarchical generation of monodisperse W/O/W DEs encapsulated with individual cells. [47][48][49][50] There are two main approaches to generate W/O/W DEs: the two-step batch method and the one-step continuous ow method using microuidic devices. 51 In the rst approach, water phase and oil phase are mixed by stirring to form W/O droplets, which can then be encapsulated by another water phase via constantly adding more water under the stirring condition. However, DE droplets produced by this approach have large variations in size. 52 The one-step continuous ow method is more widely used for single-cell encapsulation. DE Fig. 1 Droplet flow cytometry enables a wide range of highthroughput single-cell analysis, by integrating emulsion microdroplets (e.g., water-in-oil-water double emulsion and hydrogel droplets) with commercially available flow cytometry. production by this approach always requires a relatively high rate of oil phase ow to cut water phase with a low rate to generate W/O droplets, and thereaer a second aqueous ow with a much higher rate is used to cut the oil-water layer ow. Unlike the batch method, it ensures that microdroplets can be consecutively made with a uniform size.
In the DE, the aqueous internal phase can be composed of a reaction mixture (e.g., substrates, buffers, uorescent dyes and antibodies), cell suspensions or a gene library, providing a suitable environment for different cellular assays. 49,53,54 The oil shell is bioinert and non-adhesive to hydrophilic activities and cells. The usage of uorinated oil (e.g., HFE-7500 and FC-40) acting as a selective barrier can reduce hydrophilic mass transfer and communication between DEs. 49,55,56 The aqueous external phase allows the suspension of DEs in nutrient medium for long-term cell culture and the compatibility with commercial cytometers for high-throughput assays 18,49,57,58 (with the sorting speed >10 7 events per hour 31,32,56 ).
The integration of FACS and W/O/W DE droplets have already found various applications, such as analysis of cellsecreted molecules and enzymatic turnover. Moreover, the isolated cells from heterogeneous populations allow downstream bioinformatics analysis (e.g., tandem genomic, epigenomic and transcriptomic analyses) of the same target, revealing the biological functions of massive genetic pathways. 48-50

Hydrogel droplets
Hydrogels are three-dimensional cross-linked hydrophilic polymer networks that are capable of swelling or de-swelling reversibly in water. By using monomer or polymer solutions as the aqueous dispersed phase and oil with a surfactant as the continuous phase, hydrogel droplets in oil phase can rapidly generate by either laboratory-customised 2D microuidic devices 59 or 3D-printed microdroplet generators 60 for various biomedical assays. Unlike W/O droplets, the generated hydrogel droplets can be transferred from an oil suspension into an aqueous solution (e.g., buffer or cell culture medium) aer gelation for use in FC. The hydrogel microdroplets can offer a biologically friendly environment (mimic the extracellular environment by a three-dimensional matrix for cell culture). A microcapsule remains the activities of living cells [61][62][63][64][65] and their genes, 66-69 secreted proteins [70][71][72][73] and other ingredients (e.g., drugs [74][75][76][77] ). Due to their biocompatibility, mechanical and chemical stability in aqueous media, sufficient permeability of porous structure and tunable polymer properties, hydrogels (both natural and synthetic) have been widely used in droplet microuidics for single-cell assays.
Natural hydrogels used for emulsion microdroplets are either polysaccharides (e.g., agarose, alginate, and chitosan) or proteins (e.g., collagen and gelatin). For example, agarose is a linear neutral polysaccharide extracted from specic red seaweed, which undergoes sol-gel transition depending on temperature. 78 Alginate is an anionic polysaccharide typically obtained from brown seaweed and bacteria, which chelates with alkaline earth metal ions, such as calcium (Ca 2+ ), barium (Ba 2+ ) and strontium (Sr 2+ ) to form hydrogels. Gelatin is a protein derived from collagen, gels via a cold-setting mechanism and remains gel-like upon heating up to physiological temperature. It allows the liquid form to remain during the cultivation period of cells (i.e., at 20-27 C) and cell recovery under mild conditions (i.e., at 35 C) without cell damage. 79 Besides, synthetic hydrogels, such as poly(ethylene glycol), poly(acrylic acid), poly(vinyl alcohol), poly(acrylamide) and their derivatives, offer desirable, controllable and reproducible properties (e.g., molecular weights, structures and crosslinking density), which translate into tunable chemical, mechanical, and degradation properties of hydrogels. Typically, the aqueous monomer solutions are mixed with initiator and cross-linker, followed by polymerisation or crosslinking via different methods, such as photoirradiation. For example, poly(ethylene glycol) (PEG) can be chemically modied to generate ultraviolet (UV)-sensitive PEG-diacrylate that allows photoinitiated polymerisation. 80,81

Capabilities
By encapsulating individual cells from a heterogeneous population within monodisperse microdroplets (e.g., W/O/W DE and hydrogel droplets) and analysing cell-laden microdroplets as individual events by FC (including FACS), DFC has enabled new functionalities and diversied applications for single-cell analysis. This section aims to summarise and discuss four main capabilities of DFC, namely single-cell cultivation, molecular evolution, single-cell detection, and cell-to-cell interaction (see Fig. 1), which allow a broad range of applications in biological, biomedical and environmental elds.

Single-cell cultivation
Microdroplets can act as miniaturised bioreactors to facilitate cell-based biological studies. This is because they mimic the micro-scale environment where biochemical events transpire and signicantly enhance the efficiency and sensitivity of highthroughput cellular studies by reducing reagent consumption and improving detection signals. 82,83 Flow cytometric analysis of single cells encapsulated in microdroplets was reported as a powerful tool for high-throughput screening of cell growth kinetics and metabolite accumulation. 31,32,56 Y. Zhang et al. 56 proposed ow cytometric characterisation of bacterial growth in DE droplets acting as programmable bioreactors. Unlike single emulsion droplets that are incapable of continuous supply of nutrient molecules, the oil shell in the DEs has demonstrated the capability to allow selective transport of small nutrient molecules and chemical inducers into the aqueous core. This study demonstrated the capability of tracking the growth of individual bacterial cells both qualitatively and quantitatively, where DEs enable more control over microbial cell growth conditions. Moreover, it shows the potential of using DEs for the study of complex biological processes requiring communication with external environments. M. Li et al. 79 quantitatively tracked the growth and metabolite (i.e., chlorophyll and lipid) accumulation of two microalgal species, Euglena gracilis (E. gracilis) and Chlamydomonas reinhardtii (C. reinhardtii), in gelatin hydrogel microdroplets using FACS (see Fig. 2A). The porous hydrogel structures enabled the infusion of uorescent reporter molecules into the hydrogel matrices, and cells exhibiting desired properties (e.g., high biomass and metabolite production) can be recovered aer sorting for re-culture. In comparison to previous works, this research enables quantitative tracking of not only the growth of individual cells but also their metabolite production. The proposed technology can be easily integrated into directed cellular evolution pipelines for microbial strain improvement.
The integration of emulsion microdroplets with FACS has also enabled antibiotic susceptibility testing, the discovery of antibiotic resistance genes and the quantication of microbiome diversity. Y. J. Eun et al. 84 encapsulated E. coli cells in agarose droplets and determined the minimum inhibitory concentration (MIC) of rifampicin using FACS. Spontaneous mutants resistant to antibiotic were isolated and further characterized by DNA sequencing, which identied b-subunit of RNA polymerase, RpoB, as the target. S. S. Terekhov et al. 31 reported the culture of individual cells from a heterogeneous bacterial population in DE droplets to investigate the efficacy of an antibiotic, amicoumacin A (Ami), toward different microbiomes (see Fig. 2B). Unlike bulk cultures where fast-growing species can quickly dominate a population, single-cell cultivation in droplet compartments allows researchers to assess minor microbiota species representing <0.1% of the initial population. Later, potential antibiotic producers were selected using FACS, and a unique mechanism of self-resistance was discovered using proteomics and heterologous expression. This outcome shows that DFC could be efficiently applied to analyse complex microbial communities, and potentially expanded to functional proling of eukaryotic cells.
In addition to microorganisms (e.g., bacterial and microalgal cells), mammalian cells have been successfully encapsulated and cultured in DE droplets for high-throughput screening. K. K. Brower et al. 18 performed the phenotyping of individual cells from four mammalian cells lines, including mouse embryonic stem cells, human T lymphocytes, mouse macrophage cells, human embryonic kidney cells, and mouse embryonic broblasts having different morphologies and sizes (5-20 mm), in DEs (with an inner diameter of 35 mm). Using a live-cell labelling dye, Calcein Blue AM, as an indicator of cell-laden DEs, FACS measurement results successfully discriminated empty droplets from those containing cells, which showed good agreement with theoretical predictions by Poisson distribution. This highthroughput screening of DEs encapsulated with different types of mammalian cells opens up new avenues for various biomedical applications, such as disease diagnostics, drug discovery, and rare mutation detection.

Molecular evolution
Microbial production of industrial enzymes and other valueadded bioproducts is attractive due to their advantages compared to conventional chemical syntheses, such as sustainability, reduction in environmental pollution, and cost--effectiveness. 6 Although agar plate screening and microtiter plate assays enable automatic screening of enzymatic activities, they are costly and low-throughput, limiting the number of strain variants that can be evaluated (typically in the range of 10 3 -10 4 ). 53,54,85 Single-cell FACS has already demonstrated its capability to select individual cells exhibiting desired properties from a heterogeneous cell population, signicantly increasing the screening throughput (with sorting speed >10 7 events per hour). 86 However, FACS for enzyme selection is limited to the cases where signalling molecules have restricted diffusion out of the cells or can be entrapped on the cell surface or antibodyconjugated microbeads. This limitation can be overcome by encapsulating single cells within emulsion microdroplets, which restrict the diffusion of products by compartmentalization, and the uorescence intensity of the whole droplet can be detected by FC, which is also accurately correlated with the activity of encapsulated individual cells.
A. Aharoni et al. 54 rst reported using the DE-FACS technique and achieved high-throughput screening of enzyme libraries (see Fig. 3A). A variant gene library was transformed and cloned into E. coli, and the encoded protein, serum paraoxonase (PON1), was allowed to translate in the cytoplasm or on the surface of the E. coli cells. Aer encapsulation and cultivation of individual E. coli cells in DEs, compartments containing the uorescent products, encapsulated cell variants, and the gene encoding enzyme were sorted by FACS. The isolated PON1 variants showed 100 fold improvements in TBLase catalytic efficiency. These works demonstrated the use of DE-FACS for directed enzyme evolution by directly monitoring the actual levels of endogenous cellular enzymes (rather expression levels of a reporter protein). Moreover, it shows the capability to identify rare phenotypes and genotypes for analysing large populations at a single-cell level.
To date, DE-FACS has been widely used for the directed evolution of different types of enzymes, including thiolactonase, 54 b-glucosidase, 23 arysulfatase 50 , G-type nerve agent hydrolase, 24 protease, 22 cutinase, 87 cellulase, 88 glucose oxidase, 89 esterase, 90,91 and polymerase, 92 expressed in different microbial strains, such as S. cerevisiae, E. coli and Bacillus subtilis (B. subtilis). Moreover, hydrogel droplets with either core-shell (e.g., agarose-alginate and polyacrylamide-agarose) or single emulsion (e.g., PEG and alginate) structures have enabled the directed evolution of phosphotriesterase, 93 hydrolytic enzymes 94 for high-throughput cell cultivation and enzymatic analysis 95 by integrating with FACS. For example, M. Fischlechner et al. 93 utilised gel-shell microbeads with a polyelectrolyte shell to encapsulate an enzyme, its encoding DNA and uorescent reaction product. FACS was able to identify active clones within the compartments at rates of higher than 10 7 beads per hour (see Fig. 3B). This work allows the establishment of gel-shell microbeads as biomimetic compartments in which the content can be modied by directed evolution. The presence of the hydrogel shell allows more effective retention of target molecules (e.g., cell-secreted molecules and cleaved products from the substrate) within the droplet compartments.
Besides the directed evolution of enzymes, microorganisms exhibiting high yields of other industrially essential metabolites, e.g., organic acids and vitamins, have also been screened and selected using DE-FACS. J. M. Wagner et al. 96 successfully isolated Yarrowia lipolytica (Y. lipolytica) mutants, resulting in a 54 fold increase in Riboavin (vitamin B2) production parent strain. Since Riboavin is an innate uorescent product, no extra uorescent substrates were used. Also, the isolated mutants can produce more extracellular products and the total amount of products than those separated by single-cell FACS, as conrmed by absorbance assays and high-performance liquid chromatography (HPLC). Similarly, X. D. Zhu et al. 97 selected a Bacillus coagulans (B. coagulans) mutant yielding 52% higher lactic acid production than its parent strain. A pH-sensitive uorescent sensor was adopted to indicate the level of lactic acid production inside the DEs, allowing the selection of DEs containing high-yielding mutants based on uorescence.
DE-FACS has been widely used for directed molecular evolution by isolating single cells exhibiting a desirable phenotype. By combing with sequencing of the isolated variants, it provides valuable insights into the underlying mechanisms of molecular evolution. It is also obvious that this technology can be applied to other areas where analysing and sorting single cells in a high-throughput manner is needed, such as functional genomics and advanced protein engineering.

Single-cell detection
Sensitive and selective detection of single molecules (e.g., DNA, RNA and proteins) and single cells are essential for probing cellular heterogeneity and discovering the unbiased diversity of cells in biology, chemistry and biomedicine. However, due to the limited initial numbers of single cells, traditional single-cell sequencing approaches are time-consuming and lowthroughput. In contrast, droplet microuidics is a promising technique for encapsulating and processing individual cells for whole-transcriptome or genomic analysis in a massively parallel manner with minimal reagent consumption. By taking advantage of FC (and FACS), ultra-high throughput and multiparameter genetic analysis of a large heterogeneous population of cells at the single-cell level can be achieved to discover stochastic variations in a complex biological system.
The integration of agarose microdroplets and FACS allows the identication and sorting of rare pathogens using polymerase chain reaction (PCR) signals from individual cells. A notorious pathogen causing many food-borne illnesses, E. coli O157:H7, was used as a target analyte, and the typical E. coli K12 was used as background noise. 27 Massively parallel singlemolecule PCR, singleplex and multiplex single-cell PCR were achieved in agarose droplets, and FACS determined the uorescence of droplets (see Fig. 4A). This technique enabled the sensitive and quantitative analysis of single E. coli O157:H7 cells in the high background of 100 000 excess normal E. coli K12 cells. A similar work using agarose microdroplets for singlemolecule Reverse Transcription Polymerase Chain Reaction (RT-PCR) was reported by H. Zhang et al. 28 Agarose droplets demonstrated the capability to carry RT-PCR products aer amplication at the single-molecule level. As a result, single-cell transcriptome analysis revealed a clear difference in the expression level of a cancer biomarker gene (i.e., EpCAM) between different types of cancer cells (Kato III and MDA-MB-231 cells) at the single-cell level.
The detection of a single gene of individual microbial genomes from a mixture of microbial cells was achieved using hydrogel droplets with core-shell structures. 29 Polyacrylamide droplets were used to encapsulate individual microbial cells and trap their genomes aer cell lysis, which were further converted into agarose pico-reactors for subsequent multiple displacement amplication (MDA) reaction. Aer PCR, the polyacrylamide-agarose hydrogel droplets containing genomic DNA with target genes were uorescently labelled with SYBR Green and analysed by FC. E. coli XL1 genomes containing a tetracycline resistance gene present 0.1% of E. coli MC1061 genomes without the tetracycline resistance gene can be differentiated.
The proling of the cellular heterogeneity in cytokine secretion by cancer cells 12,98 and immune cells 26 at the single- cell level has been demonstrated. V. Turcanu et al. 30 encapsulated lymphocytes in agarose microdroplets, which were functionalized with avidin to capture biotin-labelled cytokine-specic antibodies. Cytokines secreted by the encapsulated cells can be detected using a second uorescent antibody specic to the cytokine. The analysis of several key cytokines, such as interleukin (IL)-4, transforming growth factor (TGF)-b, interferon-g (IFN-g) and IL-10 secreted by human peripheral blood mononuclear cells (PBMCs) and a rat keratinocyte cell line transfected with the human TGF-b gene was achieved. Also, polystyrene beads functionalized with specic antibodies have been used to capture cytokines secreted by immune cells entrapped in the droplet compartments. V. Chokkalingam et al. 25 encapsulated single activated Jurkat Tet cells in monodisperse agarose droplets with functionalized cytokine-capture nanobeads. Aer incubation and phase transfer, the agarose beads were stained with uorophore-labelled antibodies that bind to the cytokines, allowing ow cytometric detection of secreted cytokines (IL-2, IFN-g, TNF-a) from single Jurkat T cells (see Fig. 4B). Besides hydrogel functionalisation and use of cytokine-capture nanobeads.
In addition to gene expression and cytokine, proteins, 98 metabolites, 99 and exosomes 100 can be used as biomarkers for single-cell detection. For example, X. Yang et al. 101 performed on-cell target detection and phenotypic probing of live cancer cells at the single-cell level by integrating droplet compartments with a uorescence microscope. A biomarker-designated enzymatic uorescent assay was carried out within the droplets, which was based on a DNA-antibody conjugate to the surface of live cancer cells. This approach shows great potential for the detection of different biomarkers at the single-cell level by modifying the antibody-DNA constructs with other antibodies of interest.
The single-cell agarose droplet ow cytometry method has tremendous implications in detecting rare pathogenic cells, analysing single-cell protein secretion and proling of cellular heterogeneity at the single-molecule level and enumerating rare cancer cells, and shows great potential for numerous applications (e.g., environmental science, drug discovery, diagnostics, cancer research, regenerative medicine, and synthetic biology).

Cell-to-cell interaction
Microbial interactions play a pivotal role in the dynamics and functions of microbial communities. However, it is difficult to reveal this intertwined phenomenon due to limitations in parallel culture and absolute abundance quantication of community members across environments and replicates. 102 DFC has enabled rapid and quantitative investigation of the interaction networks between community members by encapsulating microbial cells into the micro-droplets that are screened and selected in parallel.
S. S. Terekhov et al. 32 investigated different types of bacterial cell-to-cell interactions using DE-FACS (see Fig. 5A). A common pathogen, Staphylococcus aureus (S. aureus), was co-cultured with a mate (E. coli), which led to the growth of both bacteria in droplets, and an antibiotic producer, Streptomyces venezuelae (S. venezuelae), which inhibited the growth of S. aureus as a killer. Since all the bacteria used in the experiment have different uorescence reporters, killers and mates can be distinguished, and droplets encapsulating with killers can be selected. This technique was then used to identify rare S. aureus killers from human oral microbiota by selecting the droplets containing a high concentration of killers and a low concentration of S. aureus. Slow-growing oral microbiota species that inhibit the growth of S. aureus were isolated, and genera associating with inhibitory activity was revealed. Using the same platform, S. S. Terekhov et al. 31 performed environmental microbiota communities proling (i.e., a Siberian bear oral microbial community) to discover secondary microbial metabolites. The target pathogen, S. aureus, was co-encapsulated with individual cells from the heterogeneous community to identify rare antimicrobial strains. As a result, bacterial clones with inhibitory activity, especially 'unculturable bacteria' that cannot be recognized by standard bulk methods, were selected and identied.
In addition to DEs, hydrogel (agarose) microdroplets have also been used to uncover yeast-bacterium interactions for antibiotic drug discovery, 33 where bacterial pathogens were cocultured with recombinant host microorganisms capable of secreting biocatalytic antibiotics and/or secondary metabolites (see Fig. 5B). S. cerevisiae strains, including both the positive control (with the secretion of bacteriolytic enzyme lysostaphin) and the negative control (with the secretion of inactive lytic hydrolase enzyme), were co-encapsulated with a target pathogen S. aureus, respectively. Following host S. cerevisiae growth and protein induction, agarose microdroplets were stained with a uorogenic viability probe (SyTox Orange) that selectively stains killed target pathogens, and then analysed by FACS. Rare positive clones producing a lytic hydrolase specic for S. aureus can be isolated with an enrichment factor of up to 10 000 fold relative to the initial mixture.
Moreover, the interactions between microalgal and bacterial cells have been investigated. J. Ohan et al. 34 utilised agarose gel droplets to co-encapsulate individual bacterial cells from a complex microbial community with Chlorella sorokiniana (C. sorokiniana), a candidate algal strain for biofuel production. Using FACS, viable bacteria species exhibiting the desired phenotype (i.e., improved algal growth rate) were isolated and recovered. Researchers have also reported the use of DFC for the screening of co-cultures of mammalian cells and yeast cells. Individual yeast secretors (e.g., S. cerevisiae 103 and Pichia pastoris 104 ) and mammalian reporter cells were co-encapsulated and cultivated in agarose microdroplets, which is followed by FACS to select clones secreting bioactive recombinant proteins. These studies demonstrated the feasibility of DFC for screening and isolation of protein variants produced by microbes, which would enable and accelerate the discovery of new functional biologics, such as cytokines, antibodies, and soluble receptors.
Using the same technique, E. Tumarkin et al. 105 uncovered mammalian cell interaction networks. MBA2 cell line (secreting interleukin-3 (IL-3)) and the human megakaryoblastic leukemia cell line, M07e (depending on IL-3 for proliferation and survival), were co-encapsulated within agarose hydrogel droplets at different ratios. Flow cytometric analysis results showed that local paracrine secretion could modulate the viability of the factor-dependent M07e cells. Besides, IL-3 secreting MBA2 cells were co-encapsulated with umbilical cord blood (UCB) cells to determine the impact of IL-3 on a heterogeneous cell population, which revealed subpopulations that are primarily dependent on locally delivered IL-3 (i.e., CD14 + cells).
Mammalian cell-to-cell interactions are crucial in cell development, tissue and organ homeostasis, and immune interactions in disease. 106,107 Researchers have achieved in vitro co-culture of different types of mammalian cells to investigate the underlying metabolic processes and interactions. 108 Unfortunately, despite the biological and biomedical signicance of mammalian cellular interactions, there are still quite limited reports in literature regarding the use of the DFC in the study of mammalian cell communications. With further advances in droplet compartments and cell co-culture models, we can expect that wider communities will adopt DFC for cell-to-cell interaction analysis.

Applications
By pairing FC with emulsion droplets (i.e., W/O/W DE and hydrogel droplets), it enables the screening and isolation of microdroplets encapsulated with rare cells exhibiting properties of interest and characterization of unexpected cell-to-cell heterogeneity within the population. Table 1 summarised the capabilities of DFC for high-throughput single-cell analysis. This integrated technology is critical for various applications, including but not limited to drug screening, medical diagnosis, metabolic engineering and cell biology.

Drug discovery
Due to the emergence and spread of drug resistance and new diseases, there is an urgent need for new drug development. An effective technique for high-throughput drug screening plays a critical role in nding a new drug against the target of particular diseases, where diverse compound libraries are tested under different conditions in parallel with the drug target. 109 Cell-based assays are always involved in this process, which provides substantive information on various cellular responses upon compound exposure, reducing the need for animal models. DFC allows high-throughput monitoring and screening of cultures of isolated single cells (e.g., bacteria cells) within droplet compartments in response to a specic drug or a combination of multiple drugs. This reveals a heterogenic reaction that cannot be obtained by traditional drug screening methods (which are based on the average of the entire population). In addition, the use of emulsion droplets enables the analysis of a single cell's anti-drug response (e.g., antibiotic resistance 31,32,84 ). Microdroplets are also used to analyse the drug response of a cancer cell line with different phenotypes, showing a clear distinction in drug response as a function of a specic phenotype. This ability to reveal different cell line responses to the same compound allows scientists to gain important information for new drug discovery and development.

Medical diagnosis
The presence and the stage of a particular disease can be indicated by sensitive and reliable detection of specic biomarkers, bacteria and viruses. For example, the proteins secreted by cancer cells (i.e., secretomes) are a fundamental source of biomarkers. 98 Although conventional immunoassays, in particular, enzyme-linked immunosorbent assay (ELISA) 110,111 have been widely used to quantify protein biomarkers in bulk, the detectable concentrations of proteins are limited to generally above the picomolar range, making them unable to detect cancer biomarkers that exist in biological uids at concentrations in the range of 10 À12 to 10 À16 M. 112 Moreover, it is quite challenging to perform characterisation of immune responses in many cases, as it requires the isolation of individual immune cells. For instance, a diversity of T cells can be produced by human immune system, but only a fraction of them are capable of being recognised and distinguished from those infected and mutant cells. Therefore, DFC has been applied to detect lowabundance proteins secreted from single cells and to quantify single-cell immune signatures. 113 Tumour heterogeneity can confound the detection and diagnosis of cancer patients, which has led to high-throughput single-cell analysis platforms to aid in the determination of cancer malignancy. [114][115][116][117] Also, coencapsulation of distinct cell types in microdroplets holds great potential for future cancer research by probing heterogeneous cell-cell interactions (e.g., cancer-immune cell interactions in cancer surveillance).

Metabolic engineering
Due to the limited knowledge of complicated cellular networks, directed evolution has played an essential role in improving the performance and functionality of industrially important microbes. 118 High-throughput and cost-effective screening tools are required in the selection of super variants. Emulsion droplets provide compartmentalization of individual cells from a massive variant library containing genetic diversity. Each cellladen droplet in aqueous suspension can be considered a distinct event in the cytometers. Selection of improved variants from the library can be carried out based on the level of phenotypes (e.g., secreted protein production), allowing to establish the linkages of phenotypes and genotypes for directed evolution of functional proteins. This mimics the process of natural evolution in a laboratory setting, which relies on iterative cycles of genetic diversity generation and phenotypic selection to isolate evolved industrial strains with desired traits. DFC has been providing an efficient screening avenue to improve microbial strains for enhanced enzymatic activity, and improved production of value-added bioproducts (e.g., vitamins and organic acids). 96

Cell biology
DFC opens new opportunities in cell biology by improving our understanding of how the behaviours of individual cells shape biological processes. Microbial cells from a complex community can be randomly encapsulated into microdroplets that can be used as miniaturized compartments and screened or sorted in parallel by ow cytometers to study how the community members interact. This high-throughput approach can rapidly resolve microbial interaction networks across different initial community states, population sizes, and environments, enabling a better understanding of the critical parameters shaping the structures and functions of microbial communities. 102 For example, co-encapsulation of two bacterial strains allows for discrimination between viable and non-viable cells, identifying the conditions that support the growth of unculturable cells and improved understanding of mechanisms underlying bacterial inhibitory activity. 31 Moreover, co-culture of a microalgal strain with a pool of environmentally sourced bacterial cells allows the isolation of the ones that can induce a desired phenotype (i.e., improvement in algae growth rate). 34

Current challenges and perspectives
As discussed above, the ability to screen and select microdroplets encapsulated with single cells by FC brings in new possibilities in various elds. Compared with traditional well plate assays, this integrated technique has impressive advantages, such as high throughput and low cost. However, there are still issues that limit DFC as a common technique for highthroughput single-cell analysis.

Molecule leakage
The rst area that needs improvement is the transport and leakage of the molecules encapsulated within emulsion droplets. To ensure accurate and effective in-drop assays, it is essential to maintain all the contents within the droplet compartments during the process (e.g., incubation, relocation and manipulation) without any cross-contamination or analyte loss. 22 Unfortunately, there is a possible leakage of small molecules from the droplets into the surrounding oil and neighbouring droplets, which may be due to the hydrophobicity and solubility of uorophores, formation, and transport micellular or vascular structures, or media pH. 50 This results in the reduction of sensitivity and resolution of biomechanical assays performed in microdroplets.
Although several approaches have been reported to increase the containment of small molecules within droplets, such as the use of uorinated oils, 119,120 decrease in the concentration of surfactants, 121 and addition of bovine serum albumin (BSA) or sugar additives into the inner aqueous phase, 122 some uorophores of widely used substrates are hydrophobic and suffer rapid diffusion or exchange between droplets. One interesting approach is nanoparticles that can interface with target hydrophobic molecules to prevent their leakage. 123 Another approach that may circumvent this limitation is the chemical or enzymatic modication of the hydrophobic products into hydrophilic, charged molecules that cannot diffuse into the oil shell. 55 There is still a need for further development of new uorescence dyes and metabolic biomarkers that can be conned within the droplets and new biochemical technologies that can prevent the leakage of small molecules from droplets.

Access to droplet microuidic technology
The second important challenge is the restricted access to droplet microuidic technology. Currently, microuidic systems are mainly designed, fabricated and operated by specialized engineering laboratories, and the applications in life and biomedical sciences achieved are based on the collaborations with engineering research groups. Therefore, more efforts should be put into commercialising simple and costeffective instruments and surfactants to facilitate wider adoptions, allowing users without expertise in microuidics to carry out the wide range of applications summarised above.
We note that a few examples of droplet microuidic technology have been successfully commercialized, e.g., DropSeq and InDrops by 10Â Genomics, which enables single-molecule detection by measuring thousands of independent amplication events within a single sample. Moreover, there are many companies provide a wide catalogue of standard microuidic chips for various applications, and help with design, prototyping and fabrication of full custom microuidic chips that meet specic needs. It would be helpful to push some relatively basic instruments for individual applications (e.g., enzymatic assays and PCR) into the market. Also, it looks great to make the technology instruments capable of integration with the multiwell plate format, so that highly parallelized downstream procedures can be carried out with existing liquid handling robots. We foresee that the accessibility of basic microuidic technology will be expanded to non-specialists in the broader research and clinical communities.

Optical interrogation
Another challenge is that the screening and analysis of droplet bioreactors have mainly relied on the detection of uorescent products. Although uorescent labelling allows multiple readouts simultaneously with high sensitivity, it is oen challenging to incorporate them into biochemical assays. The native target is not uorescently active, and adding a uorescent dye may cause cellular cytotoxicity, photobleaching, and interface with cellular metabolism. 124 Also, there are limited uorescent dyes available to probe complex cell activities in microdroplets. Therefore, developing a label-free optical detection technology for sensitive detection and characterization of droplet bioreactors would be of great promise by expanding the available assays in droplet screening and sorting. Different optical techniques have been used to analyse droplets beyond uorescence detection. For example, absorbance and mass-based detection 41,125 has been used for enzymatic activities in droplets, but it suffers from relatively low specicity. Although Raman spectroscopy has been applied to quantify single-cell metabolite and enzyme production in microdroplets, 42,126 the relatively high level of noise and low sensitivity limit its comprehensive implementation in highthroughput assays. We note that recent advances in coherent Raman spectroscopy [127][128][129][130] have enabled high-throughput (i.e., 2 kHz) and label-free analysis of heterogeneous populations at the single-cell level, which holds great promise for highaccuracy cell classication.
Moreover, imaging ow cytometry (IFC) has become an attractive tool for single-cell analysis in diverse biomedical elds, and a two-dimensional image signal has been employed for real-time classication and sorting of droplets encapsulating with cancer 131,132 and planktonic cells. 133 Combining with high-speed image-based sorting, 134-139 this will eventually be used in the process of isolating individual cells exhibiting properties of interest. With further advances in sensitivity and specicity of label-free optical detection techniques, we expect that droplet encapsulation and compartmentalization will expand into a more versatile platform technology for highthroughput single-cell analysis.

Conclusions
The ability to encapsulate, cultivate, monitor and select single cells in a large number of droplet bioreactors provides new opportunities in tackling numerous problems in fundamental and applied biology and health. New toolsets and new uorescent toolsets may deliver breakthroughs in tackling the challenges regarding transport and leakage of small molecules in droplets. The availability of droplet technology has a chance to increase via collaborations between engineers and biologists rapidly by commercializing standard instruments and chemicals (e.g., surfactants) for droplet generation and manipulation. We envision that further development of droplet systems and optical detection techniques should allow DFC to expand into a more typical and versatile platform technology for highthroughput single-cell analysis.

Conflicts of interest
There are no conicts to declare.