In the past decade, microfluidics has made remarkable advancements in exploring temporal cellular behaviors (Irimia, 2010; Gao et al., 2012; Kim et al., 2014; Sinha et al., 2018). Microfluidic devices can replicate in vivo biological environments with great accuracy and enable high-content analysis of cells. Microfluidics technology offers precise automation and control of analytical functions, enabling high-resolution manipulation of cells and their microenvironments. With these properties, we can modulate cellular signaling pathways to gain insights into mechanisms underlying cell activation, migration, and intercellular communication. Recent studies investigating temporal signaling dynamics using microfluidics-based global input generators are summarized in Table 1.

A typical microfluidic platform for studying cellular signaling consists of a microfluidic device, a custom software control system, a pressure pump, solenoid valves, and a live-cell imaging microscope (Figure 2Aa) (Yang et al., 2022). The low cost and biocompatibility of polydimethylsiloxane (PDMS) make it ideal for rapid fabrication of microfluidic devices using soft lithography, which has led to the development of various microfluidic designs (Sia and Whitesides, 2003). PDMS is permeable to gases, allowing replication of artificial cellular microenvironments in vitro, and its flexibility enables easy integration of membrane valves and pumps to create intricate networks of microchannels (Thorsen et al., 2002). This enables full automation of protocols using programming software (White and Streets, 2018; Kehl et al., 2021). The PDMS microfluidic device consists of a bottom flow layer for sample loading and a top control layer for valve actuation (Figure 2Ad). The membrane valves can be pneumatically/hydraulically actuated using a pressure pump and solenoid valves (Brower et al., 2018; Watson and Senyo, 2019). This precise control allows for cell seeding, medium exchange and input delivery for studying cellular signaling (Figure 2Ab).

Various input profiles can be defined and implemented with high precision (Table 1). Different input types can be achieved by controlling input amplitude and duration through opening and closing the embedded membrane valves (the layer in red in Figure 2Ad). The typical input modes of cytokine interferon γ (IFNγ), such as pulse and continuous (Figure 2Ac), were applied to perturb the activity of transcription factor STAT1 in single fibroblasts or populations (Yang et al., 2022). Distinct STAT1 activation dynamics were observed between one-pulse and continuous IFNγ treatment. This indicates that STAT1 activation can be temporally modulated by introducing different temporal stimulation profiles. Another transcription factor, NF-κB, displayed activation and oscillation dynamics when subjected to a continuous cytokine input of tumor necrosis factor a (TNFα). Applying a stepwise ramping input of TNFα or interleukine-1β (IL-1β) to fibroblasts revealed that the activity of the NF-κB signaling pathway correlated with the rate of change in cytokine concentrations rather than the absolute cytokine concentrations. In addition, the implementation of sinusoidal inputs was realized using a multiple-layer PDMS device with eight triangular converging valves (Figure 2B). Fibroblasts stimulated with sinusoidal TNF inputs showed characteristic NF-κB nucleocytoplasmic oscillations with great heterogeneity in single-cell responses (Piehler et al., 2017). While the duration and amplitude of inputs can be readily controlled in membrane valve-embedded PDMS microfluidic devices, implementing ramping analog inputs (Song et al., 2018; Son et al., 2021) in the PDMS devices presents a challenge. Recently, a gravity-driven flow has been achieved in a microfluidic device with high-aspect-ratio channels controlled by a gravity pump (Figure 2C) (Mokashi et al., 2019). This fully analog system is capable of producing arbitrarily complex patterns of input signals. Ramping input of TNFα led to increased NF-κB dynamics in a fraction of cells compared to those showing qualitatively different NF-kB responses to continuous stimulation. These observations demonstrate the ability of microfluidic systems to create various defined input types that can induce distinct cellular responses, which is crucial for discovering underlying mechanisms of temporal cellular signaling.

Microfluidic devices integrated with cell traps have significantly advanced research by providing opportunities to study single cells and gain insights into their signaling dynamics. These devices allow the isolation of individual cells, which is often challenging with other technologies. The designs for single-cell analysis typically utilize unique geometric structures, such as pillar-like (Junkin et al., 2016; Sinha et al., 2022) and V-type (Rho et al., 2016) valves. A microfluidic device with pillar-like traps was developed for quantitative analysis of single-cell immune dynamics (Junkin et al., 2016). With these traps, single macrophages were isolated and exposed to different input types, including a single-pulse, continuous, and repeated pulses of lipopolysaccharide (LPS), separately. The dynamics of TNFα secretion in single macrophages was found highly heterogeneous and surprisingly uncorrelated with the dynamics of NF-κB, the transcription factor that controls TNFα production. Additionally, simulation analysis revealed that a trap with an optimal geometric structure can achieve single-cell trapping with high precision (Sinha et al., 2022). These global signal generators have facilitated the exploration of temporal signaling dynamics (such as transcription factor dynamics) in single cells encountering defined environmental perturbations, enriching our understanding of how extracellular signal inputs were interpreted by single cells, and how these dynamics affect their downstream signaling events, such as cytokine secretion.

While microfluidic systems can be used to deliver global input signals to cells, they may result in a delay of seconds for inputs among different locations within cell culture, thereby posing a challenge for investigating fast signaling events. The activity of global input molecules can be suppressed and controllably activated by stimuli, such as light irradiation (Klan et al., 2013). This occurs because every signal is transmitted into cells upon binding of input molecules to specific receptors. Recently, several photoactivation-based methods have been reported (Ryu et al., 2014; Mogaki et al., 2019; Perdue et al., 2020), which allows for spatial control of signal molecules using light. Due to the simplicity and ease of light irradiation, efforts have been made to develop different strategies for small molecules and proteins.

For small molecule inputs, they can be caged by a photocleavable group, such as the 2-nitrobenzyl group, to inhibit their activity. Two small molecules, Imiquimod (R837) and Resiquimod (R848), which are agonists of Toll-like receptor 7 (TLR7) and TLR7/8, respectively, were conjugated with the photo-protecting group carbamate of 2-(2-nitrophenyl)-propyloxycarbonyl (NPPOC) to suppress their spatial activity. Irradiation with 360 nm UV light deprotected these small-molecule agonists, triggering signal transmission and NF-κB pathway activation in cells (Figure 3A) (Ryu et al., 2014). Similarly, a TLR4 agonist, pyrimido [5,4-b]indole, was photocaged at a position critical for receptor binding by protecting the indole nitrogen with 6-nitroveratryloxycarbonyl (NVOC). Upon exposure to UV light, the agonist was uncaged and activated NF-κB (Stutts and Esser-Kahn, 2015). In addition to photocaging methods, photoresponsive conformational switches of small molecule inputs can also reversibly change their activity. A photoswitchable Pam₃CS₄ derivative–P10 was synthesized to control the activation of the TLR1/2 signaling pathway. The ground-state trans-P10 can activate antigen-presenting cells (APCs) by promoting TLR1/2 heterodimerization. In the presence of UV irradiation, trans-P10 is converted to cis-P10, which reduces the activities of APCs by impeding the TLR1/2 heterodimerization (Hu et al., 2020). These methods offer the potential to regulate immune activation and inflammation.

For macromolecule inputs, such as growth factors and cytokines, it is challenging to directly modify them with photo-protecting groups. A dendritic molecular glue, ^PCGlue-NBD, carries multiple guanidinium ion (Gu⁺) pendants. This functional molecular glue can strongly adhere to the target protein, hepatocyte growth factor (HGF), and cover the region for protein-protein interactions (PPIs) on their surfaces. The PPIs are inactive, suppressing cellular signaling. Upon irradiation with UV light, ^PCGlue-NBD is photocleaved, reducing the multivalency for the adhesion. Consequently, uncaged HGFs regains its intrinsic PPI affinity toward c-Met, leading to pathway activation and cell migration (Figure 3B) (Mogaki et al., 2019). This study demonstrates a universal strategy for suppressing the activity of macromolecule inputs, holding great promise for controlling protein input-mediated signaling.

Another strategy that can reversibly suppress protein activity is chemical modification with photolabile polymers (Perdue et al., 2020). Cytokines such as human interleukin-2 (IL-2), IL-15, and mouse scIL-12 were caged with polyethylene glycol (PEG) -conjugated with a 2-nitrobenzyl linker. UV irradiation photocleaved the 2-nitrobenzyl linkers, causing PEG to detach and thus restoring the activity of cytokines (Figure 3C). The magnitude and the duration of cytokine signaling can be tuned on demand, with high spatial resolution. This approach is also applicable to a range of additional cytokine or chemokine proteins. Although the activity of proteins is initially suppressed, cells still initiate a delayed response to the photocaged cytokine proteins. In contrast, the uncaged cytokine proteins induce a significantly faster response. These findings demonstrate the controllability of cytokine signaling latency using light. Although this strategy does not directly control the activation, continuous efforts may lead to improvements for this purpose.

Several recent light-controlled global signal generators are summarized in Table 2. The effectiveness of photoactivation methods relies on the photoresponsive groups or linkers used. Current methods are based on the use of short-wavelength light, such as UV irradiation. The should be noted that pathways sensitive to UV light may be activated or interfered with (Devary et al., 1993; Li and Karin, 1998; O'Dea et al., 2008). Due to the strong phototoxicity of UV light, cells may not survive prolonged exposures. Exploring alternative photocleavable groups responsive to long wavelengths of light can be a promising direction for controlling cellular signaling (Wegener et al., 2017).

Living cell senders serve as natural local signal sources due to their ability to secrete signals within the physiological range. A critical step is the activation of sender cells. This section will discuss recent strategies for the controlled activation of living sender cells, including pre-stimulation, microfluidics-assisted stimulation, photocaged global input, and optogenetic activation. Recent studies on living cell-based local input generators are summarized in Table 3.

To construct a local signaling model, sender cells can be pre-stimulated before cocultured with receiver cells. The activated sender cells become local signal sources, secreting input signals in limited areas. Recently, a diffusion-consumption model has been created using the pre-stimulated T cells as the living senders to produce IL-2, which stimulates surrounding T cell receivers (Oyler-Yaniv et al., 2017). Immunofluorescence staining revealed the generation of microdomains of STAT5-activated T cells around local IL-2 sources. Although this method is operationally simple and can be easily applied to investigate the activation status of cells within local regions, it is challenging to measure the signaling dynamics of receiver cells. The challenge lies in precisely controlling the secretion of cytokine from living sender cells, which makes it difficult to track the origin of signal propagation and transduction in receiver cells.

Valve-integrated microfluidic devices enable the coculture of a single sender cell (e.g., macrophage) and a population of receiver cells (e.g., fibroblasts), as well as the control of signal propagation (Frank and Tay, 2015; Son et al., 2022; Watson et al., 2022). These devices consist of connected channels with a separation valve between macrophages and fibroblasts, (Figure 4Ab), creating a 1D signaling axis (Frank and Tay, 2015; Son et al., 2022). Dynamic LPS inputs can be delivered to single macrophages, initiating NF-κB pathway activation and TNFα secretion, which becomes a local TNFα source. By opening the separation valve, TNFα transmits along the channels in a wave-like propagation, initiating temporal and spatial responses of NF-κB in the cocultured fibroblasts (Figures 4Aa–c). This model enables control of local signal flow by opening and closing the separation valve. A microfluidic device facilitating unidirectional intercellular communication can avoid crosstalk and interference between sender and receiver cells (Fang et al., 2021). The device consists of two separated half-ellipse-shaped chambers for different cell cultures, which are mixed in Matrigel and loaded into the left and right chambers, respectively. Matrigel and physical barriers restrict the medium flow to form a unidirectional signal flow from sender to receiver cells. Additionally, the device allows the analysis of functional signals secreted by sender cells via a signal-blocking inlet.

Although 1D signaling models have been realized for spatiotemporal signaling studies, the output information is still limited because in vivo local signaling patterns are typically 2D or 3D. A 2D model of developmental signaling center has been created in microfluidic device (Manfrin et al., 2019). Localized morphogen signaling sources were generated upon treatment with an input signal of bone morphogenetic protein 4 (BMP4), resulting in the formation of morphogen gradients along human pluripotent stem cell (hPSC) colonies. The hPSCs exhibited spatially differential expression of MIXL1, T, SOX17 and CDX2 genes, demonstrating spatiotemporally controlled morphogen signaling gradients. This study shows the possibility of constructing a 2D signaling model in a microfluidic device, provided that a global signal does not stimulate receiver cells.

Photocaging and photoactivation strategies can also be used to control the activation of sender cells in a 2D signaling model. For example, a light-controlled immunostimulant probe that can photosensitize immune cells was synthesized to control the origin of inflammation (Mancini et al., 2015). This probe, a photocaged TLR agonist modified with a 2-nitrobenzyl linker, can tag and remotely induce a guided immune response (TRIGIR) (Figure 4B). With light irradiation, the TRIGIR probe is uncaged after the photocleavage of the 2-nitrobenzyl linker, functioning as a photoactive immunopotentiator to activate TLR signaling and inflammation pathways. In a coculture environment, the photoactivated TRIGIR probe controllably activates bone marrow-derived dendritic cells (BMDCs) as a global input. The cocultured fibroblasts are further activated upon receiving local input signals from the activated BMDCs, thereby initiating TNF secretion.

Synthetic biology offers powerful tools such as chemogenetic (Keifer et al., 2020; Tsai et al., 2021; Raper and Galvan, 2022) and optogenetic (Tischer and Weiner, 2014; Zhang and Cu, 2015; Leopold et al., 2018; Hongdusit et al., 2020; Farahani et al., 2021; Kramer et al., 2021) techniques to control the activity of signaling proteins in living cells. By leveraging these tools, it is possible to control the activation of living sender cells that have been transfected with chemogenetic or optogenetic response elements. An example involved the production of morphogen Sonic Hedgehog (SHH) by a sender cell line under the control of the chemical 4-hydroxytamoxifen (4-OHT) (Figure 4Ca) (Li et al., 2018). The SHH signaling gradients resulted in radial and linear activation geometries in neighboring receiver cells, as evidenced by the expression of nuclear-localized Histone 2B (H2B)-Citrine fluorescent protein (Figure 4Cb). This finding showcases the ability to achieve localized signal sources through the utilization of synthetic circuits controlled by chemical inputs. The optogenetic approach allows the construction of an intercellular communication model in which local signal generation can be precisely controlled through light irradiation on living sender cells. The design of an opto-SOS system enabled the controlled initiation of signaling protein Ras activation, nuclear translocation of Erk2, and secretion of IL-6 family cytokines (Toettcher et al., 2013). The observation of STAT3 nuclear translocation in receiver cells confirmed the propagation of IL-6 from the signaling senders. Notably, 2 hours of light irradiation on sender cells led to STAT3 activation in the receivers, whereas two separate 1-h light irradiations with an interval did not produce the same effect. In another study, an optogenetic intercellular system was implemented in the budding yeast Saccharomyces cerevisiae. This system involved controlling the production of α-factor pheromone through blue light irradiation of Optogenetic Sender (OS) strains, subsequently leading to luciferase induction in the Receiver (R) strains (Figure 4D) (Rojas and Larrondo, 2022). These studies demonstrate the versatility of optogenetic tools in the introducing different types of local inputs that lead to distinct fate decisions in receiver cells.

Artificial cells, also known as synthetic protocells, are designed to replicate the structures and functions of living cells. These cell mimics provide a valuable tool for studying intercellular communications with minimal interference from cellular complexity, such as diverse secretion levels and rates of signal molecules. Additionally, artificial cells offer advantages in controlling the release of local input signals compared to living cells, which opens up new possibilities for various applications. In the context of local signal generation, artificial cells can be engineered to replace living cells as local signal sources. While several studies have investigated communication between artificial cells (Niederholtmeyer et al., 2018; Aufinger and Simmel, 2019; Joesaar et al., 2019; Karoui et al., 2022), the interaction between artificial and living cells has received less attention. This section will introduce three types of signaling models, namely, paracrine, contact-dependent, and embedded signaling, in the context of artificial-living cell communities (Figure 5A).

Paracrine signaling involves the transmission of signals over short distances, eliciting diverse responses in receiver cells. Artificial cells with biocompatibility can be cocultured with living cells to deliver local input signals via paracrine signaling. A recent development induces an artificial cell system that integrated a brain-derived neurotrophic factor (BDNF) and perfringolysin O (PFO) gene expression construct (Toparlak et al., 2020). This system allows for the controlled activation of both genes using N-3-oxohexanoyl homoserine lactone (3OC6 HSL). In the presence of 3OC6 HSL, both PFO and BDNF are produced, and BDNF is released through formed PFO pores. In a coculture system, the artificial cells responded to 3OC6 HSL, releasing BDNF that subsequently drives the differentiation of mouse embryonic stem cell-derived neural stem (mNS) cells (Figure 5B). Communication between artificial cells and engineered HEK293T cells has also been established through the addition of 3OC6 HSL. The released BDNF induces GFP expression in the HEK293T cells. These results demonstrate the suitability of artificial cells in delivering paracrine signals as substitutes for biological cells.

Nested (or embedded) architectures involving artificial and living cells provide non-native signaling configurations (Figure 5Ab). In this construct, artificial cells are embedded within living cells, allowing for the exploration of signaling events initiated inside the system, such as antiviral innate immune signaling (Seth et al., 2006). Although this approach has been relatively less explored in current studies, it holds promise for future applications. For example, micron- or submicron-sized artificial cells loaded with viral DNA/RNA can be endocytosed by living cells, mimicking nested communication during viral infection. The viral DNA/RNA can then be released, triggering the retinoic acid-inducible gene I (RIG-I) and melanoma differentiation-associated gene 5 (MDA5) pathways, as well as NF-κB pathway (Brisse and Ly, 2019; Rehwinkel and Gack, 2020; Onomoto et al., 2021). Another possible configuration involves living cells embedded within an artificial cell, enabling a cellular bionics approach where living cells can function as organelle-like modules. A recent study presented a cellular bionic system consisting of a single host lipid vesicle-based artificial cell encapsulating colon carcinoma cells, and established an embedded glucose oxidase (GOx)/horseradish peroxidase (HRP) enzyme cascade (Figure 5C) (Elani et al., 2018). This localized communication was initiated by the production of glucose (Glc) upon stimulation of the cancer cells with lactose pre-loaded in the artificial cell. This innovative approach has the potential to uncover more signaling mechanisms underlying nested communication.