Immune quorum sensing dictates IFN‐I response dynamics in human plasmacytoid dendritic cells

Type I interferons (IFN‐Is) are key in fighting viral infections, but also serve major roles beyond antiviral immunity. Crucial is the tight regulation of IFN‐I responses, while excessive levels are harmful to the cells. In essence, immune responses are generated by single cells making their own decisions, which are based on the signals they perceive. Additionally, immune cells must anticipate the future state of their environment, thereby weighing the costs and benefits of each possible outcome, in the presence of other potentially competitive decision makers (i.e., IFN‐I producing cells). A rather new cellular communication mechanism called quorum sensing describes the effect of cell density on cellular secretory behaviors, which fits well with matching the right amount of IFN‐Is produced to fight an infection. More competitive decision makers must contribute relatively less and vice versa. Intrigued by this concept, we assessed the effects of immune quorum sensing in pDCs, specialized immune cells known for their ability to mass produce IFN‐Is. Using conventional microwell assays and droplet‐based microfluidics assays, we were able the characterize the effect of quorum sensing in human primary immune cells in vitro. These insights open new avenues to manipulate IFN‐I response dynamics in pathological conditions affected by aberrant IFN‐I signaling.


Introduction
Cytokine-mediated immune responses are crucial in establishing systemic protection against a wide array of pathogens.Accordingly, type I interferons (IFN-Is) are crucial in fighting viral infections, but also serve major roles beyond antiviral immunity [1,2].Their power to fight invaders comes at the cost of risking IFN-I-related pathologies, such as observed during autoimmune diseases, during which IFN-I responses are dysregulated [3].Therefore, IFN-I responses must be carefully regulated and precisely matched with the potential thread.Underlying this crucial regulation lies a myriad of clever cellular mechanisms, of which the majority could only quite recently get explored with the rise of novel single-cell technologies (e.g., single-cell RNA sequencing, single molecule RNA fluorescence in situ hybridization, singlecell quantitative PCR, etc.) [4].Together with the technological advances, the understanding on the fundamentals of (single) cellular decision-making is expanding [5].In essence, whether considering single immune cell behaviors or collective populationwide immune cell behaviors, immune responses are generated by single cells making their own decisions.These decisions (e.g., whether to produce IFN-Is) are based on the signals they perceive on their cell surface, which must be processed intracellularly, thereby distinguishing noise from actual cues [6].Additionally, immune cells must anticipate the future state of their environment, thereby weighing the costs and benefits of each possible outcome, in the presence of other potentially competitive decision makers [5].In short, (single) cellular decision-making is fundamentally complex.
To lead immune responses into the right direction, individual immune cells rely on rather basic (cytokine-mediated) communication modes [7,8].Besides the already well-known and wellcharacterized autocrine and paracrine signaling, a rather new mode of communication is making its way into the conceptualization of cellular decision-making.Established in the field of microbiology, the phenomenon of quorum sensing is more and more related to immune cell function [7,9].On the one hand, this mode of communication is similar to autocrine signaling regarding its molecular parts.On the other hand, similarities with paracrine signaling lie in the fact that quorum sensing is a mode of communication aimed at neighboring cells, and therefore considered different, in a cell density-dependent and collective manner.Therefore, once a certain density threshold is reached, cells make the collective decision to adjust their secretory behaviors accordingly.In fact, the density threshold is not directly sensed, but indirectly by the amount of signaling molecules (also referred to as autoinducers) present and sensed [10].Although in bacterial quorum this often leads to the total population of cells responding, in immune quorum sensing it often only induces a change of fraction of responding cells.Therefore, in the context of immune cells, the term quorum licensing has been used instead [11].Relating to the presence of other potentially competitive decision makers as mentioned earlier, it is thought that quorum sensing allows for a better response anticipation while immune cells can adjust their secretory behaviors according to the amount of competitive decision makers present [6].
In this study, we set out to investigate the effects of quorum sensing on IFN-I production in human primary plasmacytoid dendritic cells (pDCs), as this specialized cell type is known for its ability to produce massive amounts of IFN-Is and serves major roles in early antiviral immunity and systemic inflammation [12][13][14][15].Besides, an earlier study provided evidence for quorum sensing dictating IFN-I responsiveness in a murine reporter cell model, and therefore we are now exploring whether these findings translate to human primary immune cells too [16].Across cell types, IFN-I response dynamics are already well characterized, involving three distinct cellular fates, characterized both in vitro and in vivo [17][18][19][20][21]. IFN-I production is initiated in only a fraction of 1-3% of the total population, referred to as the first responders, upon detection of the initial stimulus (i.e., often a pathogen).Subsequently, IFN-I production gets enhanced upon autocrine and paracrine signaling, thereby activating a much larger fraction of IFN-I-producing cells, referred to as the second responders.Although the fraction of first responders seems rather solid across cell types and experimental conditions, the fraction of second responders seem to be much more flexible, and therefore adjustable to meet the desired amount of IFN-I needed [19,18].The nonresponders comprise the majority of cells and will not produce IFN-Is during the course of infection, despite being infected and/or activated via paracrine signaling.To investigate whether quorum sensing is dictating responsiveness in human primary pDCs, both conventional bulk analyses in microwells as well as single-cell analyses in microfluidic droplets were executed.Here, we demonstrate that the principles of immune quorum sensing apply to human primary pDCs in vitro, while low cell densities induce higher fractions of IFN-I-producing cells and vice versa.The quantity of IFN-I produced per cell is only affected when cultured at varying densities for a prolonged amount of time, arguing that IFN-I production is primarily based on all-or-nothing cellular decision-making, like described for other cytokine secretion systems in innate immunity [22,23].

Limiting population-wide IFN-I signaling to first responders
During IFN-I response initiation, first responders play a crucial role in orchestrating population-wide IFN-I production (validated in earlier studies using ELISAs) and could therefore be well affected by quorum sensing [19,18].Before, first responders were only studied using either (droplet-based) microfluidics to allow for single-cell activation and prevention of paracrine signaling, or by using KO mice (e.g., IFN receptor KOs) [18,24,19,21].However, as quorum sensing is affected by cell density, we aimed to establish a methodology to limit IFN-I responses successfully and reliably to the first responders in a rather conventional bulk setting.In theory, by blocking IFN-I receptors (IFNARs) by using JAK inhibitors (JAKi; upadacitinib), both autocrine and paracrine signaling gets halted, thereby limiting the population-wide responses to only the first responders (Fig. 1A) [17].Of note, upadacitinib is a selective JAK1 inhibitor, which is not specific to the IFNAR receptor.Instead, JAK1 is part of numerous type I and II cytokine receptors.Therefore, upadacitinib most likely affects numerous cytokines in this experimental setting (e.g., IL-2, IL-6, IL-10, IFNγ, etc.).However, while pDCs dedicate over 50% of their transcriptome to IFN-I production upon activation, we assume that IFN-Is are the main target [25].Therefore, we sought to investigate and optimize how to successfully utilize JAKi to limit IFN-I responses to the first responders for further characterization of potential immune quorum sensing effects.
Human primary pDCs were isolated from healthy donors, rested overnight with IL-3, and incubated with varying JAKi concentrations upon activation with TLR7/8 ligand R848 (Fig. 1B).The viability of the cells upon increasing concentrations of JAKi was carefully monitored throughout (Supporting information Fig. S1).Over the first 4 hours, IFN-α production was assessed using an intracellular IFN-α staining.After 2 hours, which is considered the peak response time, around 25% of de pDCs were IFN-α positive, which agrees with literature (Fig. 1C) [18,19].
Upon treatment with 10 μM JAKi, around 10% of the cells produced IFN-α, which is still considerably higher than the expected percentage of first responders, which is around 1-3% of the total population [18][19][20].Also, the MFI values of IFN-α-positive events only slightly decreased, arguing that positive autocrine feedback loops could still have been initiated, leading us to conclude that 10 μM JAKi was not sufficient to block all receptors present on the cells (Fig. 1D).Upon treatment with 100 μM JAKi, the percentage of positive cells dropped to around 3%, which corresponds with the expected percentage of first responders.Theoretically, this proves that a concentration of 100 μM allows for blocking all IFNARs, thereby limiting the IFN-I responses to only the first responders.Accordingly, the corresponding dot plots clearly show that the ∼3% of positive events only become slightly positive, quantified by MFI values, upon treatment with 100 μM JAKi (Fig. 1E).This is in line with the effects of blocking IFNARs on halting the autocrine feedback loops, meaning that the first responders are not able to enhance their own production upon IFN-α secretion and subsequent IFNAR signaling.In contrast, the corresponding dot plots obtained from cells treated with 10 μM JAKi still displayed events with a relatively high fluorescent intensity (MFI of 693, compared to 519 for 10 and 100 μM, respectively), reflecting rather high levels of intracellular IFN-α signal, arguing that these cells were still able to enhance their IFN-I production via autocrine signaling.
To conclude, 100 μM JAKi was sufficiently and reliably able to limit the population-wide IFN-I signaling to only the first responders, thereby allowing their further characterization in immune quorum sensing experiments.

Immune quorum sensing induced upon varying cell densities in microwells
After we have established a reliable methodology to limit population-wide IFN-I signaling to first responders only, we next aimed to assess the effects of quorum sensing induced upon varying cell densities in rather conventional bulk experiments.human primary pDCs were isolated, rested, and seeded at different densities (i.e., 1.000, 5.000, and 25.000 cells per microwell) overnight (Fig. 2A).At a low cell density, the individual cells were sparsely distributed over the bottom of the well, whereas at a high cell density the individual cells were in close proximity to one another (Fig. 2B).JAKi was used to characterize the density-induced quorum sensing effects on first responders, specifically.
Over the course of the first 4 hours upon activation, pDCs were monitored for intracellular IFN-α at varying cell densities.In untreated conditions, meaning no JAKi was administered, regular population-wide IFN-I signaling mediated by both autocrine and paracrine signaling was induced, allowing both the first and second responders to start producing IFN-I upon activation.Strikingly, biological replicates at a low cell density (1.000 cells per well) resulted in an increased percentage of IFN-α-producing pDCs compared to biological replicates at high cell densities (25.000 cells per well), which are in this case mostly represented by the second responders as paracrine signaling is still allowed (Fig. 2C).Upon JAKi treatment, thereby halting both autocrine and paracrine signaling, a similar phenomenon was observed (Fig. 2D).For these conditions, the percentages of IFN-α-positive cells are, in theory, only represented by the first responders.The MFIs, reflecting the quantitative amount of secretion per IFN-α producer, did not display significant differences induced by the immune quorum sensing effects (Supporting information Fig. S2A  and B).Additionally, similar experiments were performed with density seeding executed just prior to cell activation, instead of the day before activation, which showed similar trends (Supporting information Fig. S3A-F).However, the effects on the MFI were significant, meaning that not only the percentage of IFN-I-producing cells increased, but also the quantities they produced (Supporting information Fig. S3D and F).
As quorum sensing is thought to operate through varying concentrations of autoinducer as a reflection of cell density, we sought to look for correlations between available volume per cell and percentage of responsiveness.Accordingly, a high volume per cells translates to a low cell number per well, reflecting a relatively low level of autoinducer, and vice versa.Interestingly, these two parameters correlated fairly well (R 2 = 0.56, 0.23 for second and first responders, respectively) (Fig. 2E).This implies that the less autoinducer amounts get perceived, reflecting a lower cell density, the more IFN-α is being produced (Fig. 2F).Additionally, the difference in slope, with the slope of the second responders being over twice as large (slope = 0.20, 0.09 for second and first responders, respectively), argues that the second responders are better capable of adjusting their responsiveness to the autoinducer levels, meaning the effects of quorum sensing are bigger compared to its effects on first responders.
Overall, the effect of cell density influences the percentage of IFN-α-producing pDCs.Therefore, we conclude that immune quorum sensing drives IFN-I production, affecting both the first and second responders.

Immune quorum sensing induced upon varying autoinducer levels
Following up on the results obtained from the experiments in microwells, which indicated an inverse correlation between cell density and fractions of IFN-I producers, we hypothesized that cells estimate the density based on autoinducer levels, as described for bacterial quorum sensing [7].Generally, bacterial systems that are dictated by quorum sensing depend on three basic criteria.First, the community (in our case immune cells) produces soluble signaling molecules, called autoinducers.Second, these signaling molecules are detected by receptors expressed on or in the individual community members.Third, the signaling molecules can regulate gene expression.This will allow for cooperative behaviors, especially as the production of autoinducers typically induces a feed-forward loop to enhance autoinducer production, thereby promoting synchrony across communities [26].Given these criteria, we hypothesized that IFN-Is themselves could be potent autoinducers, while a vast fraction of the total population is able to produce them, all cells express the IFN-I receptor, and IFN-Is are known to regulate gene expression [27].Moreover, feed-forward loops are well established for IFN-I signaling [28].
To assess whether IFN-Is are the autoinducers dictating the quorum sensing effects observed earlier, we first administered varying levels of IFN-Is in microwells.By administering brefeldin A (BrefA), a drug that inhibits cytokine secretion, we could control and maintain the same cellular inputs (i.e., cytokine levels) across experiments, in theory [29].In practice, we could on average halt 90.31% of the total cytokine secretion (Supporting information Fig. S4A and B).In other words, experimental conditions without BrefA include both the administered autoinducer levels, as well as the autocrine/paracrine-induced cytokine levels produced by the cells themselves.In contrast, the conditions treated with BrefA had the administered autoinducer levels as main input, with only minor levels of autocrine/paracrine induced cytokine levels.
pDCs were isolated, rested, and density seeded as described before.BrefA and varying levels of IFN-β were administered together with R848.Two hours post activation (response peak), the cells were assessed for intracellular IFN-α (Fig. 3A).Cells were incubated at cell densities ranging from 1.000 to 25.000 cells per well, incubated with IFN-β ranging from 5 to 500 U/mL (Fig. 3B).Interestingly, the varying levels of IFN-β were not influencing the percentage of IFN-α-positive cells for most conditions (Fig. 3C).The only experimental condition that showed a correlation between administered IFN-β and the percentage of IFNα-positive cells was the condition with 25.000 cells per well, treated with BrefA (R 2 = 0.41).The other conditions remained stable over the different IFN-β concentrations; however, they all resulted in response rates of either around 10% or 20%.Another remarkable finding we could extract from the data was that the conditions treated with BrefA always showed an approximate twofold increase in responsiveness compared to their corresponding untreated conditions.This finding is in line with the hypothesis that BrefA-treated cells perceive less cytokines, thereby enhancing their own cytokine production.However, this is not in line with what was observed across treatment conditions, with in both cases (BrefA treated vs. untreated) the conditions with 1.000 and 5.000 cells giving lower response numbers compared to the conditions with 25.000 cells.In other words, IFN-β seems not to be the autoinducer to explain the observed immune quorum sensing in pDCs.Instead, another, yet unknown mechanisms seem to dictate the percentages of IFN-I producers in these experiments.
To eliminate all the uncertain factors from the experiments performed in microwells, we wondered whether this could be overcome by performing similar experiments in microfluidic droplets.Accordingly, we isolated pDCs, coated them with cytokine catch reagent for IFN-α, and encapsulated individual cells in microfluidic droplets with varying IFN-β concentrations (Fig. 4D and E).The results again showed no clear inverse correlation between responsiveness and IFN-β concentration, however, an inverse trend seemed to arise when increasing the concentration up to 5.000 U/mL, resulting in lower percentages of IFN-I producing cells (not significant).Whether this could be an immune quorum sensing effect, responsible for the earlier observed phenomena, or whether this is due to enhanced negative feedback regulation must be further explored before drawing further conclusions.
Altogether, even though IFN-β meets all the requirements to be an autoinducer, our results were insufficient to conclude that IFNβ is an autoinducer responsible for the immune quorum sensing effects observed in pDCs.

Immune quorum sensing induced upon varying cell densities in microfluidic droplets
In addition to the immune quorum sensing effects observed in microwells, we found another elegant way to further unravel it, this time on a much smaller scale, namely, in microfluidic droplets.The advantage over this approach is that droplet-based microfluidics allows for well-tunable and controllable microenvironments, in which the cells receive precise inputs to make decisions, physically isolated from one another [19].
Freshly isolated pDCs were encapsulated in microfluidic droplets upon varying cell densities to allow for both single-cell and multicell encapsulation (Fig. 4A).The results obtained from the single-cell encapsulation served an important role, while it reveals the varying, donor-dependent fraction of first responders present in the total population.Using those fractions, one can calculate the probability of a first responder present in a multi-cell droplet, customized per donor, which increases when the number of cells per droplet increases.For example, given that the fraction of first responders is 3% for a given donor, the probability of at least one first responder being present in a given droplet containing one cells is obviously 0.03, while for a droplet with two cells the probability turns 0.059, and 0.087 for a droplet with three cells (see Materials and Methods section for more details on the probability calculations).While the probability of at least one first responder present can be predicted relatively easily and surely, the effect this cell has on the potential other cells in that given droplet is very uncertain.In fact, the fractions of second responders can widely vary, even within the same donor [19,20].To overcome this uncertainty, we chose an experimental approach in which one single first responder will turn all potential other cells within the same droplet positive too by coating all cells with cytokine catch reagents.Therefore, once the first responder starts producing IFN-Is, these will not only bind to the producer itself, but also saturate the remaining cells in the same droplet [19].
Upon the production of microfluidic droplets, the encapsulation of cells follows a Poison distribution, meaning that each batch of multicell droplets consists of predictable fractions of droplets containing 0, 1, 2, etc., cells (Fig. 4B) [30].In theory, if the cell concentration for droplet production is similar across experiments, this distribution would be exactly equal across experiments, given that the droplet volumes are also kept constant.In practice, the actual distribution can deviate slightly from the predicted distribution.Besides, the number of isolated pDCs can deviate quite drastically between donors (ranging from 0.5 to 2 million per buffy coat).Therefore, we decided to monitor the Poison distributions for each experiment manually, as for the probability calculations this distribution is crucial to make good predictions on possible immune quorum sensing effects (Fig. 4C).Next, the predicted values will be compared to the obtained values using flow cytometry.Of note, upon droplet breaking for flow cytometry analysis, the information of which cells were coencapsulated with which other cells is lost, which is overcome by using probability calculations.
To assess the effect of quorum sensing in microfluidic droplets, we hypothesized that the experiment-specific stochastic predictions (obtained from the probability calculations) are higher than the actual experimental outcomes.This would be in line with the results obtained from the quorum sensing experiments in microwells, where higher cell densities resulted in less cells producing IFN-Is.As indicated, each donor was characterized by a slight difference in fraction of first responders (Fig. 4D).Accordingly, having performed the experiment on six donors, we observed indeed a difference between the stochastic predictions and the corresponding experimental outcomes, with the stochastic predictions being always higher than the experimental outcomes (Fig. 4E).This made us conclude that we were able to replicate the immune quorum sensing effects that we observed earlier, where fewer cells induce higher fractions of IFN-I-producing cells, and vice versa, in microfluidic droplets.
While the differences showed quite some variation, we reasoned that for the experiments in which relatively many droplets contained multiple cells, the expected the immune quorum sensing effects could have been more potent, resulting in larger differences, and vice versa.To put this to the test, we quantified the amount of multi-cell encapsulation, simply by taking the sum of the percentages of droplets times the corresponding cell counts (see Materials and Methods section), which we called the Poisson score.Interestingly, the Poisson scores correlated fairly well (R 2 = 0.301) with the corresponding difference between the stochastic prediction and experimental outcome, emphasizing the increasing effect of immune quorum sensing upon increasing density, as observed in the microwells (Fig. 4F).
Altogether, by using a different experimental approach, namely, droplet-based microfluidics, we able to study a similar phenomenon as observed earlier in microwells.Therefore, we are confident to conclude that immune quorum sensing is dictating IFN-I production in pDCs in vitro.Its effects in vivo need to be characterized in future studies to further assess the relevance of our findings.

Discussion
Cellular decision-making exhibits different levels of heterogeneity, originating from variations in genome architecture [31,32], in concert with regulatory signaling events [33,34], through intrinsic noise in stochastic processes and extrinsic differences between cells [35][36][37].In other words, single-cell gene expression is inherently variable, but how this variability is controlled on the population level remains elusive.Intuitively, many hands make light work, which may hold true for immune cells too.In other words, in the presence of many other potential IFN-I producers, less individual immune cells must contribute to fight a given pathogen.This concept is starting to get more and more appreciated in the field of immunology [7].While we are the first to prove the phenomenon for pDCs, immune quorum sensing has already been established for monocyte-derived cells in vitro and dendritic cells in vivo [10,38].Interestingly, the immune quorum sensing effects described for DCs in vivo are similar to the classical effect of quorum sensing in bacteria, while a high density of cells was required to assure optimal DC activation in the lymph nodes.While optimal DCs activation relies on IFN-I signaling, these results are easy to comprehend.Regarding the pDCs, which are able to mass-produce IFN-Is, we hypothesize that the possible immune quorum sensing effects operate very differently, using density as a measure for the desired fraction of IFN-I producing cells to prevent overshooting.As indicated, additional studies should verify whether this hypothesis holds true in vivo.
While IFN-β seemed a logical candidate to be assigned as autoinducer, it brings up the question what exactly defines an autoinducer.If our results indicated that IFN-β could serve the role of autoinducer, it might as well have been due to its effects on manipulating positive/negative feedback loops that determine IFN-I production, rather than giving the cells an indication on cell density.The difference between the two might seem minor in practice; however, the current conceptual understanding of the two scenarios cover two sides of a spectrum.Similarly, the differences between the effects of priming and trained immunity seem small; however, the underlying mechanisms are very different [39].Therefore, it might be extremely challenging to characterize autoinducers for immune quorum sensing, while most immune signaling systems have considerable overlap with one another, with multiple common transcription factors and signaling intermediates (e.g., NF-kB and STAT proteins).
Up till recently, local cellular competition for signaling molecules and spatiotemporal limitations, posed by physical tissue architecture, have been considered as being the major mechanisms by which immune cell homeostasis is shaped [40].With increasing numbers of studies reporting quorum sensing principles in immune cell homeostasis, this rather new phenomenon gets more and more established across immunology.Immune quorum sensing allows for immune cell coordination through synchronization, conferring robustness to perturbations, thereby ensuring stability.However, as discussed above, various mechanisms, including ones far beyond quorum sensing principles, can work together to maintain immune cell homeostasis, and are likely to share signaling molecules and pathways, making it extremely difficult to disentangling immune quorum sensing mechanisms from already well-established modes of cellular regulation and communication.An example to illustrate this paradox entails the function of interleukin 2 (IL-2).T-cell homeostasis is mainly driven by IL-2, whereas IL-2 signaling is regulated both through its consumption and removal by target cells, as well as through IL-2-producing cell density-dependent availability [7,41].Computational simulations and mathematical modeling could help identify new quorum sensing mechanisms in immune cells, thereby disentangling immune quorum sensing from other known modes of immune cell regulation, as it served its role elegantly for characterizing immune quorum sensing mechanisms in T cells, and more recently, in macrophages [11,42].

Data limitations and perspectives
The main limitation of this study is that only in vitro assays were used to study immune quorum sensing in pDCs.The effects of immune quorum sensing in pDCs in vivo have yet to be investigated.Before that, the relevance of our findings is limited.

Soft lithography and microfluidic setup
Microfluidic devices were fabricated with polydimethylsiloxane base and curing agent at a ratio of 10:1 (Sylgard 184; Sigma-Aldrich, 101697).The polydimethylsiloxane mix was poured onto a master silicon wafer and cured at 65°C for 3 h.Both the surface of the devices and the glass slides were OH-terminated by exposure to plasma (Emitech K1050X) and bonded to yield closed microchannels.Finally, channels were treated with 2% silane in fluorinated HFE-7500 3M Novec (Fluorochem, 051243).Liquids were dispensed from syringes driven by computer-controlled pumps (Nemesys, Cetoni GmbH).Note that 2.5 v/v% Pico-Surf surfactant (Sphere Fluidics, C024) in fluorinated HFE-7500 3M Novec was used for the oil inlet, whereas mineral oil was used for the two aqueous phases.The cell suspension and stimulus suspension were loaded onto the microfluidic device by using the Tip-Loading method, as described elsewhere [30].

Bulk activation assay in microwells
Freshly isolated PBMCs or pDCs were incubated in 100 μL per 10 6 cells PBA containing the TNF-α and IFN-α Cytokine Catch Reagent (Miltenyi Biotec, 130-092-605) at 4°C for 20 min.Next, cells were washed and resuspended X-Vivo 15 cell culture medium (Lonza), supplemented with 2% human serum (pooled; Sanquin), 1% antibiotics (penicillin-streptomycin), at 25 cells per 100 μL in U-bottom microwell plates.Regarding all experiments in which cytokine production was assessed by intracellular cytokine stain-ings, cells were not pre-incubated with Cytokine Catch Reagent, but directly transferred to the microwells upon isolation.

Single-cell activation assay in picoliter droplets
Single-cell encapsulation was achieved at a cell concentration of 2.6 × 10 6 cells/mL in 92 pL droplets on average, as described elsewhere [18,19].Droplets were produced at flow rates of 900 μL/h for the oil phase and 300 μL/h for the aqueous phases.Single-cell encapsulation and droplet production were carefully monitored using a microscope (Nikon) at 10× magnification and a high-speed camera.The droplet emulsion was collected in Eppendorf tubes with punched holes to allow gas exchange, covered with culture medium to protect the emulsion from evaporation, and incubated at 37°C and 5% CO 2 .After 18 h of incubation, the droplets were de-emulsified by adding 100 μL 20 v/v% 1H,1H,2H,2H-perfluoro-1-octanol (Sigma Aldrich, 370533) in HFE-7500.

Antibody staining
Cells were washed with PBS and dead cells were stained with Zombie Green fixable viability dye (BioLegend, 423111), 1:10.000 in PBS, 100 μL) at 4°C for 20 min.Next, cells were washed and incubated with antibodies against surface proteins in 50 μL PBA at 4°Celsius for 20 min.Regarding the timecourse experiments, after each timepoint, cells were fixed with Cytofix/Cytoperm solution (BD Biosciences, 554714) at 4°C for 20 min and kept at 4°C upon measuring.

Flow cytometry
Acquisition was performed in PBA on FACS Aria (BD Biosciences).Flow cytometry data were analyzed using FlowJo X (Tree Star).FMO stainings served as controls for gating strategy.

ELISA analysis
Samples were collected at various time points to quantify TNFα and IFN-α production by ELISA (BioLegend 446404, 430204) according to the manufacturer's instructions.

Data analysis and statistics
Analysis and data visualization was performed using PRISM for windows version 9 (GraphPad).For statistical analysis, Mann-Whitney test and Wilcoxon signed-rank test were performed.

Probability calculations and Poisson score
Predicted percentages of IFN-α-positive cells were calculated for each donor specifically, using the corresponding number of first responders measured upon single-cell encapsulation.Additionally, the corresponding actual Poisson distribution of the multi-cell droplets was taken into account, while this varied across experiments due to limiting pDC numbers.The predicted probability of at least one responder present was calculated based on the following formula: P 1 ≤ first responder present = 1 − P no responders For multicell droplets, the calculation was extended to 1 − P no responders number of cells per droplet The corresponding probability was multiplied by the percentage of droplets.Finally, we assumed that at least one responder is able to turn all remaining cells in the droplet, for which we corrected the probability calculation accordingly.The Poisson score was calculated by the sum of the absolute percentage of droplets with a given cell number times that same cell number.

Figure 1 .
Figure 1.Limiting population-wide IFN-I signaling to first responders by blocking IFNAR using JAKi.(A) Graphical representation of regular IFN-I dynamics, dictated by first responders (FR) and second responders (SR), and JAKi-induced IFN-I dynamics, limiting the IFN-I production to only FR. (B) Schematic representation of experimental overview.(C) Percentages of IFN-α-positive cells upon varying JAKi concentrations, 2-h post activation with 5 μg/mL R848; mean ± SEM, n = 5 donors, Student's t-test *p < 0.05, ****p < 0.001.(D) IFN-α MFI, depicted in fold change compared to corresponding R848 condition; n = 5 donors, Wilcoxon test on absolute numbers *p < 0.05.(E) Dot plots of representative donor, indicating both the percentage of IFN-α-positive cells and corresponding MFI value.

Figure 2 .
Figure 2. Quorum sensing induced upon varying cell densities.(A) Schematic representation of experimental overview.(B) Microscopy images of pDCs seeded at varying cell densities in microwells, in a total volume of 100 μL medium; scale bar = 500 μm.(C) IFN-α response dynamics upon varying cell densities; mean ± SEM, n = 7 donors, Mann-Whitney test.(D) IFN-α response dynamics upon varying cell densities, JAKi treated; mean ± SEM, n = 7 donors, Mann-Whitney test.(C) Percentages of IFN-α-positive cells at 2-h post activation, as in (C) and (D), plotted against the available volume per cell in pL.Mean ± SEM, n = 7 donors.(F) Graphical representation of quorum sensing mechanism based on varying cell density and autoinducer levels.

Figure 4 .
Figure 4. Quorum sensing in microfluidic droplets.(A) Schematic representation of experimental approach with example probability calculations.Red dots in the response initiation phase represent IFN-I-producing pDCs.Red dots in the response propagation phase represent IFN-I positive cells.Flow cytometry analysis is based on the percentage of IFN-I positive cells, corresponding with pDCs coated with IFN-α coating antibody, IFN-α protein, and IFN-α detection antibody.(B) Microscopy image of microfluidic droplets with multiple cells encapsulated.(C) Poisson distributions with corresponding Poisson scores, n = 6 donors.(D) Percentages of IFN-α-positive cells upon activation in microfluidic droplets.(E) Percentages of IFN-α-positive cells of both the stochastic prediction and the corresponding experimental outcome, connected with dashed lines.(F) Scatter plot of delta IFN-α-positive cells (experimental outcome-stochastic prediction), plotted against corresponding Poisson score.Diagonal dashed line represents linear regression model.