From specialization to broad adoption: Key trends in droplet microfluidic innovations enhancing accessibility to non-experts

Droplet-based microfluidics has emerged as a versatile and powerful tool for a wide range of high-impact applications, such as single-cell analysis,^1–4 high-throughput screening,^5–7 synthetic biology,^8,9 precision medicine,^10–12 and diagnostics.^13–15 In the decade following the first report of water-in-oil (w/o) droplets formed by microfluidics in 1997,¹⁶ most endeavors focused on the development of methods for high-throughput monodisperse droplet generation,^17,18 sample encapsulation,¹⁹ and droplet manipulations such as splitting,²⁰ reagent addition,²¹ incubation,²² and sorting.^17,23 However, at that time, the adoption of droplet microfluidics was largely confined to specialized technology labs, limiting its accessibility to researchers in other fields who could in fact benefit most from the technology. Collaboration with experts in microfluidics was often a prerequisite for utilizing this transformative technology effectively. In recent years, the landscape of droplet microfluidics has undergone a remarkable transformation. Although droplet manipulation technologies are continuously being developed and improved, the paradigm has shifted toward the implementation of groundbreaking biological or chemical assays inside the w/o droplets.

Driven by the requirements posed by scientists to translate their assays to droplet microfluidics, three noticeable developments to improve accessibility of droplet microfluidic workflows have emerged, which will be discussed in this Perspective article. First of all, increasing efforts are made on advancing robustness and automation of w/o droplet microfluidics, as well as on implementing multiple, complex manipulations in a single workflow to enable applications with highly demanding prerequisites. These efforts can be found both in academic literature as well as in the commercial field. Second, researchers in various scientific domains are gaining unprecedented access to w/o droplet microfluidic platforms, thanks to the commercialization of several droplet manipulation systems. These commercial systems have varying levels of automation and flexibility. Finally, alternatives to w/o droplets are arising, such as double emulsions and hydrogel capsules, which attempt to offer more user-friendly workflows by integrating standard laboratory equipment, while maintaining advantages of w/o droplet systems. We conclude the article with a future outlook, shedding light on the promising future of droplet microfluidics as a widely accessible and versatile tool in various scientific disciplines.

Thus, the aim of this article is to give a perspective on how the field of droplet microfluidics is shifting from a specialized domain toward broad adoption, fueled by advancements that improve the robustness and automation of droplet manipulations, along with the increasing availability of commercially accessible systems and alternatives to traditional w/o droplets. Unlike other reviews, this article provides only a brief description of the common droplet manipulations, focusing primarily on recent developments aimed at enhancing reliability and usability. The article serves a dual purpose: first, to inform non-experts about the potential and challenges of droplet microfluidics and to offer guidance on selecting the most suitable and accessible technologies for their needs. Second, it provides experts with insights into emerging trends in robust droplet manipulations, encouraging them to design solutions that prioritize end-user accessibility and practical application.

Since this article primarily focuses on droplet manipulation techniques, we refer the reader to other literature for in-depth discussions on chip fabrication techniques,^24,25 sample encapsulation in droplets,^26,27 and droplet read-out systems.^28,2 Moreover, this article focuses on channel-based droplet microfluidics, thereby omitting digital microfluidic systems using (optical) electro-wetting-on-dielectric^30,31 or droplet-digital microfluidic hybrid systems.^32–34 

The primary step in all droplet microfluidic systems is droplet generation, during which w/o emulsions are conventionally generated by passive methods (e.g., via T-junctions or flow-focusing structures).³⁵ Droplets with diameters of a few to hundreds of μm (corresponding to fl to nl volumes) can be generated at hundreds to ten-thousands per second, where the generation throughput is typically inversely related to the droplet size. Key factors to take into account during droplet generation are (1) the required droplet size, which is influenced by the microfluidic geometry, (2) the operational flow rates, and (3) the liquid properties (interfacial tension and viscosity). However, for given liquid properties and a particular microfluidic droplet generation chip, the user can tune the droplet dimensions only to a limited extent by adjusting flow rates. Hence, proper chip selection is required, which might be cumbersome for the non-expert user. Therefore, recent research focusses on developing a framework that enables researchers to make informed decisions on the channel dimensions and flow conditions, required to obtain droplets of the size of interest.³⁶ In this context, it is also of paramount importance that the generated droplets are monodisperse, for which a highly accurate flow control system is required (e.g., syringe- or pressure-based pumps). Although these systems typically enable generation of droplets with coefficients of variation between 2% and 5%, during prolonged experiments the droplet sizes can drift for a variety of reasons. To render the droplet generation process more robust, recent efforts focus on including image-based closed-loop feedback systems during droplet generation (Fig. 1), in which in-line droplet size monitoring and coupled adjustments to the pressures or flow rates ensure drastic improvement of monodispersity over prolonged running times.^37–39 This is exemplified by Crawford et al. who showed an almost 10-fold decrease (from 3.8% to 0.4%) in coefficient of variation (CV) in droplet volume between pressure pump-based and feedback system-based droplet generation over 45 min.³⁸ 

The most commonly used methods for reagent addition into w/o droplets are droplet merging and picoinjection. In droplet merging, two or more sequences of droplets, either formed on-chip or pre-formed and reintroduced, are paired and fused, with fusion occurring either actively under the influence of, e.g., an acoustic⁴⁰ or electric field^41–43 or passively due to hydrodynamic forces^44–46 at rates of up to several kHz. Various droplet merging techniques, encompassing single and combinatorial droplet merging, have been explored, but precise synchronization between different droplet sequences remains the key challenge in order to achieve robust one-on-one droplet pairing. As a result, the user typically needs to continuously observe the experiment and adjust flow parameters in order to maintain accurate droplet pairing. In this context, developments in automated, passive droplet pairing based on self-synchronization through precise channel designs are emerging, reaching over 99% one-on-one pairing efficiency at a rate of 111 (Ref. 42) and 200 Hz,⁴⁷ as compared to up to 95% efficiency by systems not implementing this design⁴⁷ (Fig. 2).

Picoinjection involves reagent addition through pressurized liquid in a side-channel that merges with passing droplets in the main channel under the influence of, e.g., an electric^48–50 or acoustic⁵¹ field and has been reported up to 10 kHz.⁴⁸ This method can employ either a single injector or multiple serial injectors for combinatorial reagent addition. The first-reported picoinjection approach utilized pressure to set the injector at the Laplace pressure, i.e., the pressure at which the water in the injector channel and the oil in the main channel are at a stable equilibrium, with an electric field triggering the injection.⁴⁸ However, due to pressure fluctuations, injection using the equilibrium pressure is not robust for longer experiment times, leading to failed injection, drift in injection volume, or the generation of droplets by the injector. As such, similarly, as for merging, the user needs to observe the experiment continuously to make adjustments to flow parameters. Therefore, over time, improvements have been sought, such as including the implementation of pressure stabilizers which enable a 10-fold reduction in internal pressure at the picoinjector during fluctuations.⁴⁹ Additionally, syringe pumps to control the injected volume by varying flow rates^52–55 can be used to fully eliminate the need to set the equilibrium pressure. While injection by flow rate control is robust and even allows for electrode-free systems,^56–58 precisely turning off picoinjection is challenging. A semi-automated method for combined flow rate and pressure control of up to three serial picoinjectors has recently emerged (Fig. 3), allowing them to inject varying volumes as well as to turn off injection, respectively, in an unsupervised way. This is achieved by performing a calibration of the equilibrium pressure, which is then automatically adjusted in real-time.⁵⁰ However, the pressure calibration needs to be performed manually and requires significant microfluidics expertise. Recently, triggered picoinjection^59–61 and droplet merging⁶² have also been reported, where reagents are added only to droplets of interest, allowing to truly control the content of individual droplets. For both droplet merging and picoinjection, implementing camera-based feedback, as previously performed for low throughput droplet manipulations,⁶³ could enhance the robustness and automation of these methods.²² 

Droplets can be split into two or more daughter droplets of equal or distinct sizes. Several active mechanisms have been described for splitting, including mechanical^64,65 and acoustic⁶⁶ systems. However, thanks to its simplicity, most splitting systems rely on passive mechanisms, by utilizing approaches such as standard flow-focusing geometries,^67,68 akin to those employed in droplet generation, or a combination of specific channel geometries (e.g., bifurcations) and hydrodynamic forces.^20,69–72 Passive droplet splitting systems, especially when carefully designing the microchannels, hold particular significance for robust automation and can operate at speeds of several kHz. One major challenge in the application of droplet splitting is the ability to link different daughter droplets, which can be of interest when there is a need to couple various readouts to a single mother droplet. In addressing this challenge, accurately designed delay lines (Fig. 4, see also Incubation section) have proven successful to implement readout on one daughter droplet and perform sorting on the other daughter droplet for downstream processing.^73,74

Washing is a relatively underexplored droplet manipulation, primarily due to its inherent complexity. Namely, droplet washing requires alternating between droplet splitting, during which the content of interest is retained in one of the daughter droplets, and reagent addition to the daughter droplet with the content of interest, using a combination of the splitting and addition techniques as described above.^58,75–79 To specifically retain cells or microparticles with immobilized molecules into a specific daughter droplet during splitting, either passive mechanical restrictions⁸⁰ or active forces, such as acoustophoresis,^78,79 dielectrophoresis,⁷⁶ and magnetism^{58,75,77,81,82} are used. Given the complexity of the workflows, most droplet washing protocols operate at only several Hz, with a few exceptions operating at 500 Hz⁸¹ or several kHz.⁸⁰ Although automating individual steps involved in washing should be feasible, the efficiency of retaining cells and microparticles will depend upon the sample. Additionally, the use of these washing techniques is closely linked to the concurrent development of microparticle-based bioassays. Altogether, these factors may present challenges in the creation of robust, user-friendly devices. Nevertheless, given the critical role of washing in many assays, we anticipate further developments in this area.

In some applications, droplet incubation is crucial to allow (bio)chemical reactions to occur inside the droplets. In this context, some microfluidic systems employ delay lines, featuring long, broad channels where droplets remain in continuous flow. Clever channel design can even maintain the sequential order of droplets, resulting in precise and known incubation times per droplet.²² While this approach is robust, it is only applicable to shorter incubation periods, typically from a few seconds up to 1 h.⁸³ When dealing with large amounts of droplets and extended incubation periods, a common approach is to collect the droplets within reservoirs (e.g., tubing, tubes, and syringes) located off the microfluidic chip.²⁵ After the necessary incubation period, these droplets are subsequently reintroduced into another chip for further analysis or manipulation. However, this off-chip method can be challenging to ensure robustness and automation as it often requires manual interventions, which lead to significant droplet loss and increased chance of spontaneous droplet coalescence. Therefore, to enable on-chip monitoring, arrays designed to trap droplets within microfluidic channels have been proposed,^84–86 though they may impose limitations on the total number of droplets that can be accommodated. Finally, advanced on-chip systems incorporating valves (Fig. 5) have been proposed, enabling the storage and manipulation of millions of droplets.^87,88 The latter approach offers a high potential for robustness, automation, and the integration of multiple droplet manipulations onto a single microfluidic chip.

Droplet sorting is a frequently used manipulation for the enrichment of droplet populations and is based on the sorting of droplets toward different outlets using passive or active methods. Passive methods rely on (1) hydrodynamic forces to sort droplets by size^89–92 or their viscoelastic properties⁹³ or (2) interfacial interactions to sort by interfacial tension.^94,95 While these methods offer automation and robustness, they are limited to applications where the physical properties of droplets are directly related to the assay and are, therefore, not broadly applicable.

Active methods, using mainly pneumatic,⁹⁶ acoustic,⁹⁷ and dielectrophoretic^4,98–100 (DEP)-based actuation, typically offer more versatility due to the potential for integration with a wide range of readout methods allowing for on-demand control. Here, we will focus on the more widely used DEP-based sorters as this actuation method allows highly robust sorting at high accuracies (typically, >99%) and high throughputs, while being compatible with a large range of droplet sizes.^101,102 Many research efforts have led to the advancement in sorting throughput by hardware developments and design improvements without compromising accuracy. The major limitation to ultrahigh sorting rates remains the droplet size, as sorting rates of 2^4,103,104 up to 30 kHz⁹⁸ have been demonstrated for droplets up to 25 μm in diameter, while droplet sorting of larger droplets of 267 and 337 μm in diameter was limited to 21 and 4 Hz, respectively.¹⁰⁵ The accessibility of these tools for non-experts has been improving thanks to the publication of several detailed protocols,^106–108 guiding non-trained users through building and operating the droplet sorting platforms. However, current methods mostly rely on manual setting of the operational parameters, which can lead to false sorting in long experiments due to flow perturbations. Therefore, in practice, the user needs to continuously monitor the experiment and adjust flow parameters where needed. In an effort to automate and enhance the robustness of droplet sorting, iSort¹⁰⁹ (Fig. 6) was recently introduced, which is based on a closed feedback system that measures the impedance in the channel before and after the sorting junction to obtain information (i.e., spacing, droplet width, and trajectory) on the droplets passing by. With this information, the sorting efficiency and throughput are optimized in real-time by adjusting different parameters affecting the electrical actuation and fluid flow, ultimately lowering the false positives by almost 17-fold.

Advances are also being made in the context of DEP-based multiplex droplet sorting with up to six outlets on one chip.^110–112 This has been achieved in two main ways. One approach arranges multiple sorting points in series, allowing sorting up to 3 (Ref. 110) and 700 Hz.¹¹³ However, this setup is susceptible to flow disturbances, leading to droplets arriving at the sorting points at different times, which reduces sorting accuracy. The other approach places sorters in parallel, allowing sorting up to 10 (Ref. 114), 100 (Ref. 115), and 200 Hz¹¹¹ but requires strong electric forces to move the droplets which sometimes lead to droplet splitting. To overcome these challenges, researchers have developed a specialized electrode setup that applies smaller electric forces in steps, thereby creating a larger displacement without the risk of droplet splitting. As such, the multiplex sorting throughput was increased up to 473 Hz in a fiveway configuration.^112,116,117 Additionally, a multiplex sorter based on the combination of parallel and serial sorters, allowing sorting up to 80 Hz, was introduced, thereby addressing the major limitations of the individual approaches.¹¹⁸ Ultimately, the combination of the iSort feedback system with multiplex sorters would provide a great opportunity to succeed in automated and robust multiplex droplet sorting.

For many applications, downstream processing of droplets is crucial. Although droplets are commonly recovered after bulk dispensing in a tube,¹¹⁹ certain applications, including specific single-cell studies, require droplet collection at the single droplet level.^119,120 The developments in this field are rather recent and require precise coordination between (1) the droplet release, which is mostly based on optical detection,^87,121–123 and (2) the substrate position, which is generally enabled by programming a mechanical stage.^121,124 In this context, several stand-alone droplet dispensers have been developed that allow to eject droplets into well plates using electrohydrodynamic actuation at rates of about 1 Hz.^122,123 However, these dispensers require manual intervention to transfer bulk-collected droplets to the dispenser. Several approaches with increased integration have recently implemented single-droplet dispensing in-line with a DEP-based droplet sorter, by directly printing the droplets onto a substrate with carrier oil at 4 Hz¹²⁴ or by using pressure-based ejection of the droplets into a tube up to 20 Hz.¹²¹ In this context, a recently developed workflow enables droplet sorting, collection of the sorted droplets in a collection chamber on-chip, followed by single-droplet imaging and dispensing in 96-well plates using pressure-based ejection, at a rate of 1 Hz with over 99% accuracy (Fig. 7).⁸⁷ Next to ensuring robustness in an integrated workflow, this approach provides users with visual evidence on individually dispensed droplets and as such facilitates meaningful post-processing of relevant droplets only (e.g., containing a single cell).

For many applications, further exemplified in Sec. V, different droplet manipulations can be combined into functional workflows tailored to specific applications. For non-expert users, integrating these manipulations into a single chip is highly desirable and several approaches to achieve this have been reported.^{2,83,87,88,125} Aside from reducing user intervention, another clear advantage of chip integration is minimizing droplet coalescence and losses, which often occur during off-chip incubation and reinjection steps. However, several challenges arise when attempting to integrate these workflows into a single chip.

First, delay lines are suitable only for relatively short incubation times (up to 1 h²²), while incubation chambers sealed with valves require larger chip footprints when working with extensive droplet libraries. Second, when using delay lines, it is critical to match the droplet speeds between consecutive manipulations, which is challenging because, for example, droplet sorting rates are often slower than droplet generation rates. One possible solution is to decouple these droplet manipulations by incorporating incubation chambers and on-chip valves^87,88 between each step. Finally, more complex workflows inherently pose a higher risk of errors. Therefore, real-time monitoring, using techniques like camera-^38,87 or impedance-based¹⁰⁹ measurements, might be necessary to ensure the robustness of these workflows. Despite these challenges, more integration is certainly achievable, and it will be exciting to observe how future innovations in microfluidic technology push these boundaries further.

In recent years, more droplet microfluidic platforms are becoming commercially available to the end users. First of all, microfluidic chips of various specified droplet size ranges can nowadays be purchased through companies such as Microfluidic ChipShop,¹²⁶ Micronit,¹²⁷ Atrandi Biosciences,¹²⁸ and Dolomite microfluidics,¹²⁹ either by selecting from their product portfolio or by requesting custom designs. These chips can then be used with commercial syringe pumps (e.g., from CETONI¹³⁰ and Harvard Apparatus¹³¹) or pressure-driven pumps (e.g., from Elveflow,¹³² Cellix Limited,¹³³ and Fluigent¹³⁴). Pressure-driven pumps reach equilibrium faster and are able to deliver a constant, non-pulsating flow,³⁷ while syringe pumps allow flow rate programming, which simplifies the microfluidic workflow as flow rates are typically held constant between runs of the same experiment. Pressure-driven pumps can often be integrated with a downstream flow sensor that feeds back to the pressure module, which is of high interest for accurate and responsive flow control. In addition to the pumps, setups for monitoring and readout are required (e.g., a microscope with a high-speed camera and photomultiplier tubes for fluorescent droplet detection).

Therefore, companies are providing more integrated solutions for droplet generation and/or manipulation, as detailed in Table I. For these technologies, there is typically a clear trade-off between flexibility vs robustness and user-friendliness.

For example, 10X Genomics¹³⁷ and Mission Bio¹³⁹ offer highly integrated and robust platforms, yet these are only capable of generating droplets of a fixed size and are highly tailored for application in single-cell omics studies, while Bio-Rad Laboratories¹⁴¹ offers a similarly integrated and robust platform specifically designed for droplet digital PCR.

In contrast, companies such as Dolomite Microfluidics,¹⁴² Fluigent,¹⁴⁶ Secoya,¹⁴⁷ LiveDrop,¹⁴⁸ and PreciGenome¹⁴⁵ are offering microfluidics application packs for microfluidic droplet generation, consisting of a set of pressure pumps, droplet generation chips, and a high-speed camera. These setups provide increased flexibility since they can be utilized for applications as defined by the user, yet a higher user-interference is required and the platforms are still limited to droplet generation only, which is too limited for many applications.

To provide even more flexibility, devices such as the Onyx¹⁵⁰ and Styx¹⁵² from Atrandi Biosciences and Modaflow¹⁵³ from LiveDrop have been developed. These devices allow multiple droplet manipulations and are highly tailorable to the application of interest. Although these platforms are aimed to provide user-friendly workflows, user-interference is still high since manual manipulations are required between different droplet manipulation steps. Therefore, for workflows with multiple subsequent droplet manipulations, skilled personnel are still required.

In this regard, the Cyto-Mine platforms^155,156 of Sphere Fluidics are much more integrated since they enable droplet generation, incubation, sorting, and dispensing on a single chip, without requiring any user interference. Additionally, these are currently the only platforms providing single-droplet dispensing. As a downside of this integrated workflow, it is only compatible with droplets of a specific size, and it is not possible to include any other droplet manipulations, such as reagent addition, that might be required for an application.

In conclusion, for commercially available devices, high automation and integration often limit flexibility, making them suitable for specific rather than widespread biological applications. On the other hand, highly integrated platforms enhance the robustness of droplet manipulations, while more flexible systems need constant monitoring and expertise, hindering their use in non-specialized labs. Given the wide variety of options available today, suited systems can be selected depending on the skills and application needs of the users. When searching for the ideal system for a specific application, it is important to look beyond performance metrics such as droplet generation frequency, sorting speed, and sorting accuracy. These metrics can vary significantly depending on factors such as droplet size, the type of bioassay being implemented, and, in some cases, the user's level of expertise. Therefore, it is essential to consider the broader context, including the adaptability of the system to the required workflow and the ease of operation. Whenever possible, arranging a demonstration or trial run of the system can provide valuable insights into its practical performance and help ensure it meets the requirements.

In addition to improving the robustness and user-friendliness of w/o droplet manipulations, ongoing research aims to provide alternatives that enable replacing some or all of the complex manipulations while maintaining several advantages of microfluidic w/o droplets (Fig. 8).

Some of these are water-in-oil-in-water (w/o/w) droplets and so-called double emulsions. These have one or more aqueous cores and a thin oil shell and are dispersed in an aqueous continuous phase. Generation of double emulsions conventionally is achieved by either two separate microfluidic devices¹⁵⁷ or one device with two consecutive droplet generation zones,^61,158 having, respectively, hydrophobic and hydrophilic properties for consecutive generation of the w/o and w/o/w emulsions. To overcome the need for the complex localized adjustment of surface wettability in the chip, and thereby render the w/o/w generation more robust, an alternative one-step emulsification procedure was developed, relying on a dual bore capillary system.¹⁵⁹ This was later on extended to multiple-bore systems to enable the robust generation of monodisperse double emulsions with up to four inner drops.¹⁶⁰ Thanks to the aqueous outer phase of double emulsions, these can be sorted and dispensed with standard, commercially available FACS¹⁶¹ or other dispensers,¹⁶² omitting the need for more complex on-chip droplet sorting. The compatibility of double emulsions with FACS implies certain levels of robustness, automation, and accessibility as FACS is commonly used in research environments.^157,161 This translates into several published protocols for non-expert users.^163,164 Nevertheless, it should be noted that commonly reported post-sort recoveries are low (40–70%),^161,165 with a maximal recovery of 81.5% being recently reported¹⁶⁶ only after (1) a specialized drop delay calibration procedure, which requires the preparation of dedicated double emulsions comprising the standard AccuDrop beads used for drop delay calibration, and (2) using double emulsions with a limited outer diameter (30 μm) to ensure compatibility with the FACS nozzle size, which limits their use for large mammalian cells or combinations of multiple cells, which need larger droplets for maintaining their cellular activities. In addition, other manipulations such as reagent addition, splitting, and washing are more complicated for double emulsions due to the additional liquid layer, and currently only a few proof-of-concepts of reagent addition^167–170 have been described. Nevertheless, double emulsions have been used in a variety of applications, including sorting of sequence-specific DNA molecules through double emulsion digital droplet PCR,¹⁶⁵ cellular heterogeneity studies and rare cell sorting,¹⁶³ and directed evolution of enzymes.¹⁵⁷ 

Other heavily researched alternatives are hydrogel-based particles, which can be generated by relying on the same principles as w/o droplets: first, droplets of a water-soluble hydrogel precursor are generated using standard droplet generation systems, followed by cross-linking the hydrogel by, e.g., illumination with light and subsequent removal of the oil phase, resulting in hydrogel-based particles resuspended in an aqueous phase. As such, three types of hydrogel-based particles can be generated (1) hydrogel beads, consisting completely of a hydrogel mesh,¹⁷¹ (2) hollow hydrogel capsules (also known as semi-permeable capsules),¹⁷² which are liquid cores surrounded by a hydrogel shell, and (3) crescent-shaped hydrogel nanoparticles with a cavity.^173–175 Similar to double emulsions, all hydrogel-based particles can be sorted and dispensed with FACS^176–179 or similar dispensers,¹⁶² although being hampered by the same recovery limitations mentioned above for double emulsions, linked to specialized calibration requirements and limited particle size. In addition, all hydrogel particle types have specific features, rendering them more or less suited for specific applications. For example, for cell studies, it should be noted that cells in hydrogel beads are in close contact with the hydrogel matrix, which is known to impact, e.g., cellular growth and other processes,¹⁸⁰ and which is not the case in hydrogel capsules or crescent particles. With respect to the ease of particle generation, hollow hydrogel capsules, and crescent hydrogel particles require precise tuning of the hydrogel precursor compositions, flow rates, and channel geometries to generate stable particles, as they rely on aqueous two phase separation.¹⁸¹ However, the crescent-shaped hydrogel particles can be fabricated in the absence of the sample as they enable subsequent sample encapsulation by mixing using standard lab equipment.¹⁷⁵ As such, the microfluidic workflow, required for particle fabrication, is completely decoupled from the biological implementation and can be performed months to years in advance,¹⁷⁵ hence obviating the need for any microfluidics expertise for the end user. In terms of reagent exchange and implementation in multi-step assays, all particle types allow the exchange of reagents through the hydrogel mesh or particle cavity by resuspending in alternating aqueous solutions through simple pipetting,^172,179 as such facilitating manipulations for the end user in a robust and user-friendly manner. However, with respect to molecule retention, hydrogel beads do not allow for selective retention of components because of the uniform mesh permeability and the lack of a hollow core.¹⁷² The retention potential of hollow hydrogel capsules, in turn, can be tuned by changing the shell permeability, but this requires precise control over the composition to not hamper capsule concentricity and stability.¹⁷² In addition, while retaining small molecules (e.g., enzymes and short DNA fragments) in capsules remains challenging, this would also result in an overall reduced reagent exchange through the shell, thereby limiting the applicability of these capsules in a range of molecular assays. Finally, retention of cells and molecules of interest in crescent hydrogel particles can only be ensured by capturing them on the particle surface or/and by particle-templated droplet generation.¹⁸¹ Whereas the latter ensures molecule confinement and avoids crosstalk between particles, it also introduces the need for additional de-emulsification steps to subsequently render the particles compatible with standard laboratory methods.^172,179

Based on their specifications and limitations, the different particles have been used for a wide range of applications. Hydrogel beads have, for example, been used for rare cell detection¹⁸² and as 3D culture platforms,¹⁸³ and are also at the core of the barcoding methods for single-cell sequencing in Drop-seq,¹⁸⁴ InDrop,¹⁸⁵ and Hydrop.¹⁸⁶ Alternatively, semi-permeable capsules have been demonstrated to enable complex molecular workflows including digital gene expression profiling of individual mammalian cells¹⁷⁹ and single genome amplification as well as clonal expansion of bacteria.¹⁷² The crescent-shaped particles in turn are commonly used for fluorescence immunoassays on proteins, secreted by individual cells.^175,181 However, considering the unique advantages of application of these hydrogel-based particles by non-microfluidic-experts, in combination with future technological advancements that will drastically expand the functionality of these particles in the coming years, they are expected to continue to appear in exciting applications.

Also, for these alternative technologies, commercial platforms are arising. For example, double emulsion and hydrogel bead formulation are supported by the InstaDrop platform of Secoya,¹⁴⁷ the Complex emulsion platform of Fluigent¹⁴⁶ and the μEncapsulator system by Dolomite Microfluidics.¹⁴² Double emulsions are also supported by the Xdrop^® Sort system of Samplix.¹⁸⁷ Semi-permeable capsules can be generated using the Onyx¹⁵⁰ and Flux¹⁸⁸ devices from Atrandi Biosciences. Finally, pre-made crescent nanoparticles can be purchased from Partillion Bioscience.¹⁸⁹ 

Many applications require multistep workflows, where several of the described droplet manipulations must be performed consecutively. In this section, we present three examples of applications that significantly benefit from droplet microfluidics or their alternatives, along with their associated workflows. As such, we aim to inspire researchers with potential workflow ideas for their applications.

The first key application is high-throughput enzyme engineering and discovery, which is extensively reviewed elsewhere.²⁵ Here, single variants from a large protein library (10⁵–10⁸ variants) are encapsulated to identify the best-performing ones, for which two approaches are commonly used. The first approach relies on protein-expressing cells. Here, individual cells expressing protein variants are encapsulated with assay reagents (e.g., fluorogenic substrate for an enzyme and/or lysis buffer) to identify and sort optimal variants. This workflow is most often performed using w/o droplet microfluidics, where droplets are either incubated and sorted on the same chip^83,125 or incubated off-chip for re-injection into a sorting chip,^{83,105,125,190,191} depending on the reaction time. Since the workflow requires only an encapsulation and sorting step, also double emulsions¹⁹² and hydrogels¹⁷⁸ in combination with FACS are suitable for this application. The second approach relies on gene variants that need to be transcribed and translated into proteins after encapsulation. Here, single gene variants are encapsulated in droplets together with amplification reagents. Multiple subsequent steps have to be performed, including the addition of in vitro transcription and translation reagents, the addition of assay reagents, and sorting.^6,193,194 Because many steps are required, current work relies fully on w/o droplet microfluidic workflows, in which the steps are typically performed on separate chips to accommodate lengthy incubation times. In both approaches, the sorted variants are either re-screened for further optimization or sequenced for in vitro production. As such, droplet microfluidics and its alternatives greatly accelerate the discovery of highly efficient enzymes, given the increased throughput compared to classical plate-based assays.

A second application is the field of single-cell omics, where droplet microfluidics is a key technology for single-cell RNA and DNA sequencing, achieving throughputs of thousands of cells.^195–198 Typically, single cells are encapsulated with barcoded beads and lysis and/or reverse transcription reagents, resulting in capturing of the cell-specific RNA and/or DNA for downstream analysis. On the microfluidics level, the most commonly used platforms are relatively straightforward since they rely solely on droplet generation. This has resulted in several open source^1,186 and commercially available^137,139 platforms that are highly specified for these applications. Recent research is delving into using multi-step workflows for increased functionality¹⁹⁹ or increased efficiency and sensitivity, for example, by (1) uncoupling different reaction steps using picoinjection²⁰⁰ or (2) processing only cell-containing droplets using either droplet sorting²⁰⁰ or selective merging of those droplets with the barcoded beads.^62,87,88,201

A third application is the field of monoclonal antibody discovery, where cells expressing antibodies are screened to identify the antibody-expressing cell of interest for single-cell sequencing and in vitro antibody production. In droplet microfluidic workflows, the cells are co-encapsulated in droplets together with bioassay reagents. After incubation, either on-chip^2,87,88 or off-chip,^3,88,201,202 the droplets are sorted. Different approaches have been applied to enable the subsequent analysis, for example, by single-droplet dispensing into a microtiter plate for sequencing^87,203 or reinjection of all positive cells into a single-cell sequencing chip.²⁰¹ After sequencing, the antibodies can be produced in vitro for functional testing. As such, large libraries of cells can be screened and rare cells of interest can be isolated, accelerating monoclonal antibody discovery. In this context, the commercially available Cyto-Mine^® devices^87,155,156 are specifically designed for single-cell phenotype screening applications, such as monoclonal antibody discovery, with as key advantage their capability for single-droplet dispensing, enabling immediate single-cell sequencing. Alternatively, these workflows can also be developed on more open systems like the Onyx¹⁵⁰ and Styx,¹⁵² or LiveDrop Modaflow¹⁵³ platforms, albeit with pooled collection of cells of interest. Finally, alternatives to droplet microfluidics have also proven effective for monoclonal antibody discovery, including the encapsulation of assay reagents and antibody-producing cells in full hydrogel beads²⁰⁴ or crescent-shaped hydrogel particles,¹⁷⁵ followed by single-particle sorting with FACS, thereby offering a more accessible approach.

We refer the reader to several in-depth reviews^{9,12,35,205–209} for more examples of possible high-impact applications of droplet microfluidics, such as in the fields of particle synthesis, organoids, synthetic biology, single-cell analysis, personalized medicine, and digital polymerase chain reaction (PCR).

In recent years, the field of droplet microfluidics is increasingly focusing on the implementation of groundbreaking biological or chemical assays inside w/o droplets. Therefore, to improve the reliability and widespread use of these techniques, significant efforts are being made to enhance the robustness of droplet microfluidics on different aspects such as minimizing user-intervention, droplet loss, spontaneous droplet coalescence, and flow instabilities. A clear trend in this context is the implementation of feedback systems, as is seen for, e.g., (1) monodisperse droplet generation by using camera-feedback and real-time adjustments of fluid flow, (2) accurate picoinjection by using oil pressure-feedback and real-time adjustments of the injector Laplace pressure, and (3) robust droplet sorting by impedance-feedback and real-time adjustments of the electrical actuation and fluid flow. Another trend is the integration of multiple droplet manipulations in a single chip to reduce the need for constant user intervention, thereby simplifying experimental workflows and reducing droplet loss. This is mostly obtained by implementing incubation steps in between different manipulations, using either passive delay lines or incubation chambers which can be actively opened or closed. A last aspect is the reliance on passive systems, which can be obtained by clever channel design and has shown suitable for, e.g., delay lines, droplet splitting, and droplet pairing prior to droplet merging. Since the droplet size is highly crucial for passive manipulations, a feedback system on the droplet size might need to be incorporated to ensure a robust workflow. In the future, we anticipate increased integration of microfluidic manipulations, relying on a combination of feedback mechanisms and passive systems.

The trends discussed above are not only observed and foreseen for droplet workflows developed during academic research but also for commercial devices. Namely, boosted by the great interest from a wide range of fields to use droplets, commercialization of w/o droplet generation and other manipulations are increasing, thereby leading to broader adoption of the technique by different end users. The commercial landscape is very broad, ranging from plug-and-play flexible systems that require a lot of user-interference, to highly integrated and robust devices that are tailored to a certain droplet manipulation sequence or even one specific application. This trade-off between robustness and flexibility has to be taken into account by the end user. Namely, when the application aligns with a commercial, integrated workflow, it is often the preferable option. However, when there is no such fit, end users are directed to more open platforms to sequentially implement different droplet manipulations. One of the major challenges in implementing such workflows is the precise design of microfluidic chips, mainly because designs for droplet generation and manipulation are tailored to specific droplet sizes, and certain manipulations, such as reagent addition or splitting, affect the droplet size. In this context, the increasing variety of commercially available chip designs and application packages is expected to enable mix-and-match capabilities for different manipulation units, facilitating a range of functional workflows in the near future. Other efforts to enable faster development of these customizable workflows would be increased sharing of microfluidic designs by the community and the further development of AI-based design tools.²¹⁰ Moreover, more companies are offering chip fabrication as a service, eliminating the need for microfabrication infrastructure and skills on the part of the end user. On a last note, despite having suitable microfluidic designs, chip operation will still be challenging. Therefore, it is valuable for the community to share common operational principles through protocol papers, which are already becoming more prevalent.

In the context of more straightforward handling by the end user, alternatives such as double emulsions and hydrogel-based particles are arising, enabling non-technical users to benefit from some of the advantages of w/o droplets (e.g., assay miniaturization or single-cell analysis), while using standard laboratory equipment. Given the high interest of end users, we expect further advances in these fields as well, such as increased manipulation possibilities of double emulsions or the use of new hydrogel formulations with different porosities.

As such, we foresee significant advances in both w/o droplets as well as their alternatives that will be mostly focused on the broad adoption of the technologies to further spread their impact throughout different scientific communities.

This work has received funding from KU Leuven (Nos. IDN/20/011, iBOF/23/005, C3/22/046, IOFM/14/002, VTI-24-00161, and C24E/20/035) and Research Foundation Flanders (FWO) (Nos. 12A4W25N, G082522N, S000825N, and G090120N).

The authors have no conflicts to disclose.

Jolien Breukers: Conceptualization (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Karen Ven: Conceptualization (equal); Investigation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Wannes Verbist: Conceptualization (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Iene Rutten: Conceptualization (equal); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal). Jeroen Lammertyn: Conceptualization (equal); Supervision (equal); Writing – review & editing (equal).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.