A Tunable Diffusion-Consumption Mechanism of Cytokine Propagation Enables Plasticity in Cell-To- Cell Communication in the Immune System

Article

A Tunable Diffusion-Consumption Mechanism of Cytokine Propagation Enables Plasticity in Cell-to- Cell Communication in the Immune System

Graphical Abstract Authors Alon Oyler-Yaniv, Jennifer Oyler-Yaniv, Benjamin M. Whitlock, ..., Morgan Huse, Gre´ goire Altan-Bonnet, Oleg Krichevsky

Correspondence [email protected] (G.A.-B.), [email protected] (O.K.)

In Brief Cytokine-mediated communication allows immune cells to achieve a context- appropriate response, but the distance over which this communication happens is unclear. Oyler-Yaniv et al. (2017) show that a simple diffusion-consumption mechanism quantitatively describes the spatial spread of cytokines in vivo and results in localized niches of high cytokine concentrations that contribute to cell-to-cell variability.

Highlights d Cytokine penetration in tissues is governed by a diffusion- consumption mechanism d Spherical cytokine niches are generated around cytokine- producing cells d The characteristic niche size depends on the density of cytokine consumers d Cytokine niches are a source of variability in otherwise identical cells

Oyler-Yaniv et al., 2017, Immunity 46, 609–620 April 18, 2017 Published by Elsevier Inc. http://dx.doi.org/10.1016/j.immuni.2017.03.011 Immunity Article

A Tunable Diffusion-Consumption Mechanism of Cytokine Propagation Enables Plasticity in Cell-to-Cell Communication in the Immune System

Alon Oyler-Yaniv,1,2,3,4 Jennifer Oyler-Yaniv,2,3,4,5 Benjamin M. Whitlock,3,4,6 Zhiduo Liu,7 Ronald N. Germain,7 Morgan Huse,4 Gre´ goire Altan-Bonnet,2,3,4,9,* and Oleg Krichevsky1,3,4,8,* 1Physics Department, Ben Gurion University of the Negev, Beer-Sheva 84105, Israel 2ImmunoDynamics Group, Cancer and Inﬂammation Program, Center for Cancer Research, National Cancer Institute, NIH, Bethesda, MD 21701, USA 3Computational Biology Program, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA 4Immunology Program, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA 5Immunology and Microbial Pathogenesis Program, Weill Cornell Graduate School of Medical Sciences, New York, NY 10065, USA 6Biochemistry and Molecular Biology Graduate Program, Weill-Cornell Medical College, New York 10065, USA 7Lymphocyte Biology Section, Laboratory of Systems Biology, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Bethesda, MD 20892, USA 8Ilse Kats Center for Nanoscience, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel 9Lead Contact *Correspondence: [email protected] (G.A.-B.), [email protected] (O.K.) http://dx.doi.org/10.1016/j.immuni.2017.03.011

SUMMARY and doses of stimuli. Diversity is often achieved through spatial gradients in the concentration of diffusible ligands. A cell’s Immune cells communicate by exchanging cytokines position within the gradient is translated to downstream fate to achieve a context-appropriate response, but the decisions. distances over which such communication happens Gradients of soluble molecules could arise by pure diffusion are not known. Here, we used theoretical consid- from a source, by diffusion coupled to biochemical degradation, erations and experimental models of immune re- or by diffusion coupled to consumption of the molecule. In the sponses in vitro and in vivo to quantify the spatial latter two cases, gradients will have a characteristic length- scale: the typical distance over which cytokine interactions extent of cytokine communications in dense tissues. persist, which is determined by a balance between diffusion We established that competition between cytokine and degradation or consumption. Specifically, when consump- diffusion and consumption generated spatial niches tion is the predominant mechanism of removing molecules of high cytokine concentrations with sharp bound- from the system, we expect this length scale to depend on the aries. The size of these self-assembled niches abundance and efficiency of consumers. Alternatively, if mole- scaled with the density of cytokine-consuming cells, cules are removed primarily via molecular degradation, this a parameter that gets tuned during immune re- length scale will be independent of cellular composition. Indeed, sponses. In vivo, we measured interactions on length it has been shown that a model of diffusion coupled to consump- scales of 80–120 mm, which resulted in a high degree tion is the mode of morphogen spreading during development of cell-to-cell variance in cytokine exposure. Such (Wartlick et al., 2009). heterogeneous distributions of cytokines were a Immune cells rely on a network of diffusible cytokine mediators that enable cell-to-cell communication. Cytokines broadcast source of non-genetic cell-to-cell variability that is the magnitude and nature of pathogenic insults, scale the immune often overlooked in single-cell studies. Our findings response, and determine lymphocyte fate through transcription thus provide a basis for understanding variability in factor activation (Paul and Seder, 1994). A vast array of different the patterning of immune responses by diffusible cytokines exist that bind strongly to their cognate receptors, often factors. with characteristic binding affinities in the nano- or pico-molar range. An effective immune system must be flexible enough to combat various types and doses of pathogen while minimizing INTRODUCTION collateral tissue damage caused by inflammation. A key difference between the case of the developing embryo The evolution of multi-cellular organisms is made possible by a and that of the immune system is that the latter requires plas- division of labor across different cell types. Differentiation of pre- ticity. Development necessitates accuracy and precision, so cursors into diverse and specialized cell types is often mediated morphogen gradients must contribute to robustness in the spatial by diffusible extracellular stimuli, such as morphogens, growth pattern of gene expression across individuals (Houchmandzadeh factors, or cytokines. To generate diverse fates, cells must et al., 2002). In contrast, an effective immune system must be occupy diverse environments characterized by variable types flexible enough to combat various types and doses of pathogen.

Immunity 46, 609–620, April 18, 2017 Published by Elsevier Inc. 609 This implies that the immune system must have a tunable length (Snijder and Pelkmans, 2011). Next, we presented measurements scale of cell-to-cell communication that determines the extent of of the spatial extent of cytokine-mediated interaction in vivo. We its action in response to a given threat. tested our theory in vivo by increasing the density of cytokine- It is critical to consider how the length scales of communica- consuming cells and measuring the decrease in the size of the tion compare to the overall size of an organ. Communications niches during an immune response. Finally, we examined an inde- that happen over length scales that are comparable to or larger pendent dataset involving resting lymph nodes in which the source than the organ result in homogeneous cytokine fields (Perona- and spread of IL-2 could be determined. The experimental data Wright et al., 2010). In these conditions, the system is approxi- were in agreement with the theoretical predictions on the shape mately well-mixed, with cells responding uniformly. In contrast, of the cytokine field and its spatial propagation extent around a communications on length scales much smaller than the size secreting cell. Overall, we observed that the density of cytokine re- of the organ will result in heterogeneity in cytokine exposure ceptors dynamically regulates how far a cytokine spreads from its and localized domains of high cytokine concentrations (Maldo- site of production during an immune reaction. nado et al., 2004; Pangault et al., 2010; Sabatos et al., 2008; Thurley et al., 2015), i.e., niches. This heterogeneity in cytokine RESULTS exposure translates to variability in cellular response (e.g., transcription factor activation), which results in divergent paths of The Size of the Regulatory T Cell Pool Is Dynamic and differentiation and/or proliferation for individual cells (Snijder Dependent on the System’s State of Activation and Pelkmans, 2011). It has long been appreciated that the size of the Treg cell pool Quantifying how far cytokines spread from their source, and and the expression level of IL-2Ra can be tuned by endoge- the gradients they form, is a prerequisite for explaining how these nously-produced and exogenously-administered IL-2 (Amado cytokines generate the immense phenotypic and functional het- et al., 2013; Boyman et al., 2006a; Spangler et al., 2015; Webster erogeneity observed in immune cell populations (Busse et al., et al., 2009). For example, administration of particular IL-2-aIL-2 2010; Feinerman et al., 2010; Ho¨ fer et al., 2012; Muller€ et al., complexes augments the Treg cell pool, increasing the number 2012; Thurley et al., 2015). In this study, we used theoretical con- of cells and their IL-2Ra expression level (Boyman et al., siderations and direct experimental testing in vitro and in vivo to 2006a; Spangler et al., 2015). Given that one key function of quantify cytokine tissue penetration. We show that the spatial Treg cells is to scavenge IL-2 (Fontenot et al., 2005; Ho¨ fer extent of cytokine signaling is a dynamic parameter, which is et al., 2012; Pandiyan et al., 2007), this system presents an ideal tuned by the state and extent of lymphocyte activation. model to learn how the length scale of cytokine signaling During an immune response, activated CD4+ effector T (Teff) changes based on the density of cytokine-consuming cells. cells produce a variety of cytokines. In particular, interleukin-2 To determine the plasticity in the size of the Treg cell compart- (IL-2) is produced transiently for several hours following activation ment during an actual immune response, TCR transgenic CD4+ through the T cell receptor (Figure S1, Helmstetter et al., 2015; T cells specific for moth cytochrome c peptide (5C.C7) were Huang et al., 2013; Sojka et al., 2004) and serves a critical role adoptively transferred (AdTr) into B10.A wild-type recipient as a differentiation and proliferation factor (Heltemes-Harris mice. Mice were then immunized with cognate K5 peptide (a et al., 2011; Paul and Seder, 1994). IL-2 is consumed by regulatory superagonist variant of moth cytochrome c peptide) and bacte- T (Treg) cells, which comprise 5%–15% of helper T cells (Duprez rial lipopolysaccharide (LPS) as an adjuvant. The lymph nodes et al., 1991; Feinerman et al., 2010; Liu et al., 2015; Paul and Seder, and spleen were harvested 48 hr post-immunization, to measure 1994), and by Teff cells during later stages of activation (Feinerman the abundance of Treg cells in the T cell population (Figure 1). et al., 2010; Paul and Seder, 1994; Sojka et al., 2004). IL-2- The size of the Treg cell pool was increased by 50% in the lymph consuming T cells express a high level of the high-affinity IL-2 nodes and doubled in the spleen. Such a state was shown to receptor a chain (IL-2Ra), and do not produce the cytokine (Feiner- enhance suppression and mitigate autoimmune disorders (Boy- man et al., 2010; Sojka et al., 2004; Tkach et al., 2014). Given the man et al., 2006b), as well as to increase the consumption rate of source-sink relationship between IL-2-producing Teff cells and IL-2 (Feinerman et al., 2010). In summary, we demonstrated the IL-2-consuming T cells (Fontenot et al., 2005; Ho¨ fer et al., 2012; dynamic nature of the system by showing that the Treg popula- Pandiyan et al., 2007), IL-2 is an ideal model to quantify the range tion increases during an immune challenge. and heterogeneity of cytokine communications. First, we showed how the size of the Treg cell population The Extent of Cytokine Interactions Is Explained by a changes as a result of an immune challenge. We then developed Diffusion-Consumption Mechanism with Freely a theoretical framework to investigate how far cytokines diffuse Diffusing Molecules away from their source, based on the density of cytokine- Given the well-defined role of Treg cells as scavengers of IL-2 consuming cells, and the resulting heterogeneity in cellular (Fontenot et al., 2005; Ho¨ fer et al., 2012; Pandiyan et al., 2007; response to the cytokine. We validated this theory experimentally Thurley et al., 2015), we set out to quantify how changes in the in vitro, in multiple settings approximating tissue, and with a size of the Treg cell compartment impacts the length scale of variety of different cytokines. We proceeded to directly image cell-cell communication. We use the theoretical framework of the response of IL-2 consumers to IL-2 producers in their Diffusion-Consumption kinetics (Figure 2). Cytokines originate neighborhood and showed the emergence of niches of cellular from a source and are assumed to diffuse freely between the response whose dimensions depended on the density of cytokine cells. Eventually a cytokine molecule binds to a receptor on a consumers. These localized interactions represented a source of consuming cell and is endocytosed. In this model, the amount of cell-to-cell variability that is largely overlooked in single cell studies consumed cytokine linearly depends on the density of consuming

610 Immunity 46, 609–620, April 18, 2017 settings, a cytokine producing cell might be able to communi- of CD4 + cate only with its nearest neighbor (lniche<10 mm), or with every cell in the tissue ðlniche/NÞ. For IL-2 in physiological conditions Immune activated (Table S2), we expect lniche to have a value of z8–14 cell diam- Naive eters (80–140 mm, throughout, 1 cell diameter = 10 mm), much smaller than the organ diameter. Hence, cytokine diffusion and ** ** consumption would be a sufﬁcient mechanism to generate IL-2 niches in lymphoid organs. In the Supplemental Information, 20 we present a complete mathematical derivation that yields the proﬁles of cytokine concentrations around cytokine secreting cells. To conclude, our simple formalism yields a generalizable

% Treg cells formula relating the size and shape of cytokine domains to the 10 tissue properties (cytokine diffusion and consumption rate).

In Vitro Measurements of Cytokine Penetration Are Consistent with Theoretical Predictions 0 Conventional tissue culture plates are unsuitable for measure- Spleen Lymph ment of cytokine communication because cells are at low den- Nodes sity relative to that which exists in peripheral lymphoid organs. In addition, diffusing molecules rapidly mix in medium, elimi- nating localized cytokine gradients. To culture cells in an Figure 1. The Treg Cell Compartment Is Dynamic during an Immune environment that better approximates in vivo conditions, we Challenge designed and fabricated a simple device we call a clusterwell Recipient mice were adoptively transferred with TCR transgenic T cells and immunized. Animals were euthanized after 2 days and the size of the Treg cell plate, which confines cells to a dense cluster, with simple cylin- compartment was measured. Bars show the relative size of the Treg cell drical geometry (Figures S2A and S2B). The clusterwell plate was compartment in naive and in immune-activated mice, in the lymph nodes and modeled similarly to a conventional 96-well plate, making it in the spleen. Results are from experiment using 5 naive mice and 4 immune- compatible with most multi-channel pipettes and centrifuge activated mice. Data are shown as mean ± SEM. plate-holders. A cell suspension was loaded into wells, and then the plate was centrifuged to sediment the cells in a cluster at the bottom of the well. These conditions better approximated cells - nconsumers. A consuming cell that is proximal to the source the densely-packed nature of tissue compared with conven- has a greater probability of capturing cytokine and depleting the tional tissue-culture plates. Cells could be added sequentially, total pool of diffusing cytokine so that cells distal to the source creating a layered stack with different cellular compositions. sense a lower concentration. The cumulative effect of the con- Clusterwell plates were designed to enable easy multiplexing. sumers leads to the formation of a gradient in the concentration Exposure of T cells to IL-2 was assessed with single-cell field that can be detected by measuring downstream signaling re- resolution by fluorescent labeling of phosphorylated STAT5 ðl Þ sponses. One can estimate a characteristic length scale niche for (pSTAT5)—the transcription factor immediately downstream of t the spatial extent of cytokine communications. Briefly, diffusion,the IL-2-receptor engagement. Figure 3A shows the typical response mean time a molecule would diffuse freely before getting of activated CD4+IL-2Ra+ consuming T cells to varying concentra- captured is inversely proportional to (nconsumers kconsumption), where tions of IL-2 in well-mixed conditions. After the system reached nconsumers is the density of consumers and kconsumption the kinetic steady state, intracellular pSTAT5 was assessed by flow cytome- rate of consumption per cell. During that time the molecule would try. For analysis, we digitized our measurement by setting a l f t ½ travel on average a distance of niche (D diffusion) , where D is threshold for exposure at 20% of the max pSTAT5 mean fluores- the molecular diffusion coefficient. The signaling length scale is cence intensity (MFI) (Figure 3, dashed black line. See section 3.1 therefore expected to be: of the Supplemental Information for more details). A cell was qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi + counted as pSTAT5 if its fluorescence is above this threshold. l = ð Þ niche D nconsumerskconsumption We probed how the density of cytokine-consuming cells affected the extent of cytokine penetration. CD4+IL-2Ra+ IL-2 Cytokine producers are not assumed to be inert; they in fact consuming cells were mixed at different proportions with CD4- consume a small fraction of the produced cytokine via an auto- depleted splenocytes, which are inert to IL-2. To mimic the crine loop (Berg and Purcell, 1977), which slightly decreases high cell density of tissue, we loaded cells into the clusterwell the concentration at the source. However, this length scale plate (Figure 3B). Exogenous IL-2 was added on top. Under does not depend on the concentration of the source, but instead these conditions, a steady-state was reached in the level of reflects the mean displacement of a single molecule. signaling within 30 minutes (data not shown). Cells were there- Within lymphoid tissues, the density of consuming cells can fore incubated for 1 hr and then fixed and collected. Typical dis- vary from a few cells per organ to essentially packed cells tributions of pSTAT5 within the CD4+IL-2Ra+cells were bimodal, 9 3 15 3 (10 m < nconsumers <10 m ). Based on this variability, we as shown in Figure 3C. As the density of consumers decreases, a anticipate large differences in the signaling length scale for in vivo larger fraction of cells was exposed to IL-2 and became communications by different cytokines. Depending on the tissue pSTAT5+. In contrast, when identical cell preparations were

Immunity 46, 609–620, April 18, 2017 611 Figure 2. Diagram of Simple Diffusion-Con- sumption Kinetics Cytokines are secreted by a producing cell and freely diffuse between cells. Upon binding to a receptor the cytokine is consumed. This creates a gradient of localized cytokine niche with a typical

length scale lniche. Increasing consumption, either by increasing the density of consuming cells,

nconsumers, or by enhancing the single-cell con-

sumption rate, kconsumption, will lead to a decrease in lniche and vice versa. When lniche is small compared to the size of the system, we expect an increase in cell-to-cell variability.

We validated our theory with other cytokines by making clusters of cells that responded to different cytokines. As before, CD4+IL-2Ra+ T cells responded to IL-2, CH12 murine lymphoma cell lines responded to IL-4, and B16 murine melanoma cell lines responded to interferon-g (IFN-g)(Fig- ure S3B). After preparing the cell columns, we added the appropriate cytokine at different concentrations. After the system reached steady-state, cells were fixed and permeabilized and the relevant transcription factor (TF), pSTAT5, pSTAT6, and pSTAT1, respectively, was measured by flow cytometry (Figure S3C). We again observed that cytokine consumption limited its penetration into the cell column. The fraction of consuming cells followed the theoretically predicted expression derived from a Hill Function type consumption term (Figure S3D, see Supplemental Informa- tion for further details). mixed with IL-2 and cultured in a 96-well plate, we observed a We concluded that a simple model based on cytokine diffusion unimodal distribution of pSTAT5+ cells (Figure S2E). and consumption was sufficient to explain quantitatively the var- To verify that the heterogeneity in the pSTAT5 response re- ied extent of cytokine mediated cell-cell communication in this flected a spatial gradient of IL-2, where the concentration de- simplified in vitro system. We observed a decay in cytokine con- cayed as a function of distance into the column of cells, we first centration caused by consumption over the length of the cluster- added a thin layer of fluorescently-labeled CD4+IL-2Ra+ IL-2 well cell columns (roughly 2 mm). Given that the volume of mouse consuming cells to the bottom of the clusterwells. These cells lymph nodes is approximately 1-4 mm3 (Economopoulos et al., acted as sensors of the cytokine concentration at the bottom 2011), these gradients were formed over biologically relevant of the cluster. We reasoned that if a concentration gradient ex- length-scales. It is important to stress that the broad distributions isted, cells at the bottom would respond to IL-2 only in conditions of cellular response we observe in the clusterwells did not reflect of saturation—when all of the cells above them were also any a priori difference between the cells (Figures 3A, S2,andS3). exposed to cytokine. Figures S2C and S2D demonstrates that Rather, we demonstrate here how differential exposure to stimuli, this was indeed the case. Additionally, the dependence of cyto- due to spatial localization, generates cell-to-cell variability. kine penetration on the density of consumers indicates that consumption, rather than molecular degradation, is the predominant Imaging Cells Reveals Cytokine Niches around mode of cytokine removal from the system. Producers and Allows Direct Measurement of the + As predicted by our theoretical model, the percent of pSTAT5 Signaling Length Scale lniche + + cells as a function of CD4 IL-2Ra cell density nIL-2Ra+ (Fig- Next, we extended our results by observing the formation of + ½ ure 3D), followed the scaling relation %pSTAT5 f(nIL-2Ra+) cytokine gradients using a more realistic system of cytokine over a wide range of densities. This type of scaling could only producers and consumers, without an exogenous cytokine be achieved by a diffusion-consumption mechanism. source. Specifically, we directly imaged the response of cultured

612 Immunity 46, 609–620, April 18, 2017 A Figure 3. Cytokine Penetration in Dense B Conditions Is Explained by a Diffusion-Con- sumption Mechanism (A) The pSTAT5 response of CD4+IL-2Ra+ consuming cells to different doses of IL-2 in well mixed conditions. The dashed line represents a threshold set at 20% of the maximum MFI response. Cells with levels of pSTAT5 higher than the threshold are considered exposed to cytokine and counted at pSTAT5+. (B) Diagram of experimental design. (C) Distributions of pSTAT5 in IL-2Ra+ cells in D clusterwells for different ratios of consumers to inert cells. (D) The percent pSTAT5+ cells of IL-2 consumers in C the clusters, and the theoretically predicted scaling. Data shown as (mean ± SEM), results from an experiment using three experimental replicates. The data are representative of at least three additional independent experiments. See also Figures S2 and S3.

of cells. We used these parameters to infer the radial profile of IL-2 concentration around producing cells (Figure 4D). In conclusion, we developed a method for imaging dense cell populations in vitro and directly measured the interac- CD4+IL-2Ra+ cells to IL-2 produced by CD4+ T cells in dense tion length-scale of IL-2. We found sharp cytokine gradients and conditions. discrete, spherically symmetric niches of high cytokine concen- For this we developed a new assay for imaging densely trations around cytokine producers. The size of these niches packed cells, which we call the PlaneView imaging device (see varied between 30 and 150 microns, depending on the density section 4.4.2 for Extended Methods). Briefly, IL-2 consuming of IL-2 consumers, and consistently with a theoretical model of T cells, or a combination of 10% consuming T cells and 90% free diffusion and consumption for cytokines. IL-2-inert splenocytes were mixed with 0.1% IL-2 producing T cells and deposited in a monolayer on a glass slide. Then, 10 Altering the Treg Cell Compartment In Vivo Changes the more layers of cells containing no producers were added on IL-2 Signaling Length Scale during an Immune Response top, forming a three dimensional array with producing cells We showed that the Treg cell compartment naturally grows dur- dispersed on the bottom. A semipermeable membrane was ing the course of an immune response (Figure 1). This suggests placed on the cells to preserve their positions. Cells were then that the characteristic size of IL-2 niches might expand and incubated to reach steady state, rapidly fixed in situ, and stained shrink in vivo. To test this, we used sequential injections of IL- for pSTAT5, IL-2Ra, and DAPI as a nuclear counterstain. 2-aIL-2 immunocomplexes (so-called IL-2 i.c. boost) to expand Imaging pSTAT5 revealed spherically symmetric micro do- the Treg cell compartment (Figure 5A, Figure S5B) (Boyman mains of cells responding to IL-2 around producers (Figure 4A, et al., 2006a). This regimen has been shown to enhance immune Figure S4A). The apparent spherical symmetry of these domains suppression and mitigate autoimmune disorders (Boyman suggested that directional secretion did not play a role in our et al., 2006a, 2006b; Spangler et al., 2015; Webster et al., setting. In conditions with higher consumer cell densities, these 2009). Importantly, the IL-2 i.c. boost increased the Treg cell micro domains appeared significantly smaller representing a compartment by a similar amount as that resulting from systemic more localized concentration field and a smaller signaling length immune activation (Figure 1, Figure S5B). We verified that the scale. abundant endogenous CD4+IL-2Ra+ lymphocytes present in We fit the observed pSTAT5 profiles to the theoretically ex- IL-2 boosted mice are bona fide Treg cells by confirming that pected expression with excellent agreement (Figure 4B; A full they also express the lineage specifying transcription factor description of the fitting procedure is given in section 3.2 of the FoxP3 (Figure S5A). Supplemental Information) and obtained a direct measurement Either 5 3 106 or 5 3 105 naive, CD4+ 5C.C7 T cells were adop- for the niche’s characteristic length scale lniche: 3.5 ± 0.2 cell di- tively transferred (AdTr) into IL-2-boosted or wild-type B10.A ameters for 100% consumers, and 13.5 ± 1.6 cell diameters for recipient mice (Figure 5A). Recipients were then immunized 10% consumers, in agreement with estimations based on known with antigenic K5 peptide and LPS as an adjuvant. This induces microscopic parameters (Table S2). These measurements again rapid production of IL-2 from the AdTr T cells (Producers)(Sojka demonstrated how spatial heterogeneity in stimuli could result in et al., 2004). Six hours post-immunization, mice were euthanized downstream variability in an otherwise homogeneous population by cervical dislocation to preserve the signaling environment.

Immunity 46, 609–620, April 18, 2017 613 A 100% Consumers 10% Consumers Figure 4. Micro Domains of Signaling Cells Are Generated around Cytokine Sources 100 m 100 m (A) Immunofluorescence staining of cell preparations containing either 100% IL-2Ra+ consuming cells or 10% consuming cells and 90% IL-2Ra inert cells, and a small number (< 0.01%) of IL-2 producing T cells in a PlaneView imaging device. (See Figure S4 for more examples.) (B) A profile of the pSTAT5 response was generated as a function of distance from the central producer by segmenting individual cells and measuring pSTAT5 on a single cell level. The profiles were then fit with the theoretical solution of the 3D diffusion-consumption equation. The signaling BC Distance from producer ( m) length-scale l was extracted as a fitting Inferred average IL−2 profile niche 10 100 200 300 400 2 parameter. The data shown is aggregated from 34 10 2 micro domains for the 100% condition and 17 for 10 10% Consumers 10% Consumers 100% Consumers 100% Consumers the 10%. Data shown as mean ± SEM. 1 l 10 (C) Fitted signaling length-scales niche are 3.5 ± 0.2 cell diameters for 100% consumers and 13.5 ± 1.6 for 10% consumers. Black dashed line are the 1 0 10 2 (pM) 10 predicted values based on average parameters, −

IL gray dashed lines are extreme values given the

pSTAT5 (a.u.) −1 10 biological range of parameters. = 13.5 1.6 cd 10% (D) The average concentration of IL-2 at a given 0 = 3.4 0.2 cd 10 100% −2 distance from a producing cell was inferred from 10 1 10 20 30 40 10 100 200 300 400 the pSTAT5 proﬁles for different densities of con- Distance from producer (cell diameters) Distance from producer ( m) sumers. See section 3.2.1 of the Supplemental Information for full details of the procedure. See also Figures S4.

Lymph nodes and spleen were rapidly dissected and reserved For intracellular flow cytometric analysis of pSTAT5, fractions for (1) IL-2 secretion assay (Figure S6), (2) intracellular pSTAT5 of the lymph nodes and spleen were rapidly harvested directly flow cytometric analysis (Figures 5B and S5), or (3) immunofluo- into cold fixative and then permeabilized. Similar to what was rescence (I.F.) microscopy (Figure 6). As positive and negative observed in vitro, we saw heterogeneity and bimodality in the controls for pSTAT5, a sample of splenocytes was exposed pSTAT5 distributions among Treg cells in the lymph nodes and ex vivo to either a saturating dose of IL-2, or the JAK1/2 inhibitor spleen (Figures 3C and 5B). Rather than invoking some under- AZD1480 before fixation and permeabilization (Figure S5C). lying source of intrinsic variability, our theoretical model can The pSTAT5 levels in Treg cells from non-immunized mice explain the observed bimodality by the fact that some Treg cells were very low for both groups of mice (Figure S5C), as compared have access to IL-2, whereas others do not. In addition, the frac- to immunized mice (Figures 5B and 5C). To ensure that the IL-2 tion of pSTAT5+ Treg cells is uniformly greater in the spleen rela- produced by the adoptively transferred 5C.C7 T cells induces tive to the lymph nodes, reflecting the greater proportion of acti- the measured pSTAT5, an additional group of wild-type B10.A vated, IL-2-producing effector T cells in the spleen (Figures 5B mice received 3 3 106 5C.C7 Rag2/Il2/ CD4+ T cells. In and 5C, Figure S6). these mice, pSTAT5 levels in Treg cells were similar to back- For both numbers of adoptively transferred T cells in either the ground levels (Figure S5C). We concluded that the observed lymph nodes or spleen, we observed a reduced proportion of pSTAT5 signals were primarily a result of IL-2 produced by acti- pSTAT5+ Treg cells in IL-2 boosted versus wild-type mice (Fig- vated adoptively transferred 5C.C7 T cells. ure 5B and 5C), in agreement with our in vitro results (Figure 3). Although we immunized mice with a large dose of the supera- The reduced fraction of pSTAT5+ Treg cells meant fewer cells gonist K5 peptide, we needed to rule out that boosting the Treg were exposed to IL-2 in the boosted mice. Combined with the cell compartment prior to immunization would dampen the demonstrated dynamic nature of the Treg cell compartment dur- effector T cell response, resulting in reduced IL-2 production ing an infection (Figure 1), this showed that the dimensions of by these cells. For this, we measured IL-2 production by adop- cytokine niches could change rapidly and depend on the nature tively transferred 5C.C7 T cells. Six hours post-immunization, and timing of the immune response. fractions of the lymph nodes and spleen were harvested directly For immunofluorescence microscopy, fractions of the lymph into culture media, and IL-2 secretion assays were immediately nodes and spleen segments were harvested and placed directly performed ex vivo. There were no differences in IL-2 production into cold fixative, dehydrated in ethanol, and sectioned in observed between IL-2 boosted and wild-type mice for any num- paraffin wax. Sections were stained sequentially for FoxP3 ber of adoptively transferred T cells (Figure S6). Hence, IL-2 pro- and pSTAT5 using a machine-based, multiplexed Tyramide duction by the adoptively transferred T cells was not affected by Signal Amplification technique (Yarilin et al., 2015). For analysis, the IL-2 boost. We also verified equivalent number of adoptively we focused on tissues collected from mice injected with a transferred T cells across mice (Figure S6). lower number of producing cells, where the organs are not

614 Immunity 46, 609–620, April 18, 2017 A Figure 5. Boosting the Treg Cell Compart- ment Results in Decreased Fractions of IL- 2-Responding Cells after Immunization (A) Diagram of experimental design: B10.A recipient mice were boosted as described in the Experimental Procedures. Boosted and wild-type mice were adoptively transferred with either 5 3 105 or 5 3 106 5C.C7 CD4+ T cells. Transferred mice were immunized, and then rapidly euthanized 6 hr later. Spleens and lymph nodes were reserved for either ﬂow cytometric pSTAT5 analysis or I.F. B Wild type Boosted microscopy. (B) Representative histograms of pSTAT5 among Lymph Nodes Spleen Treg cells from ﬂow cytometry. pSTAT5+ cells were 6 5 6 5 gated based on the dashed threshold line drawn. 5 10 AdTr 5 10 AdTr 5 10 AdTr 5 10 AdTr + 100 (C) The fraction of pSTAT5 Treg cells is lower for IL-2 i.c. boosted mice. Bar plots of pSTAT5+ Treg 80 cells are from an experiment using three mice per 60 group. The data are representative of three independent experiments. See also Figures S5 and S6. 40 % of Max 20

0 1 10 100 0 1 10 100 0 1 10 100 0 1 10 100 pSTAT5 (a.u.) tion function yields unreliable results. We performed this analysis on lymph nodes, C Lymph Nodes Spleen where T cell zones are large and relatively homogeneous. 6 5 6 5 5 10 AdTr 5 10 AdTr 5 10 AdTr 5 10 AdTr We found a good agreement with the 100 * *** ** ** predicted exponential decay for the 80 + pSTAT5 autocorrelation function. Fitted 60 length scales were 11.4 ± 0.7 cell diame- 40 ters for the wild-type mice and 8.2 ± 0.7 for the IL-2 i.c. boosted mice. The volume % pSTAT5

(of Treg cells) 20 of IL-2 niches decreased by 63% ± 4%

due to the increase in the regulatory %pSTAT5 + = 29 %pSTAT5 + = 20 %pSTAT5 + = 28 %pSTAT5 + = 48 T cell population. Importantly, these results suggest a mechanism by which Treg cells dictate the magnitude of an saturated with cytokine (see section 3.3 of Supplemental Infor- immune response and prevent autoimmunity: by setting an mation for details). Imaging revealed spatially distinct niches of inherent length scale for inter-cellular communication, the pSTAT5+ cells among Treg cells in the proximity of 5C.C7 system isolates rare autoreactive proto-effector T cells (Liu T cells (Figure 6). et al., 2015) and prevents systemic autoimmune responses at Our ﬂow cytometry measurements suggested a transition homeostasis. from a cytokine perfused state for the wild-type mice spleen, During homeostasis, pSTAT5+ Treg cells have been shown to to a spatially localized state for the spleens of IL-2 boosted exist in discrete clusters in the vicinity of self-reactive IL-2 mice (Figures 5B and 5C). Indeed, we saw this transition in Fig- secreting proto-effector T cells in vivo (Liu et al., 2015). This ure 6A. In both sites, we saw that the pSTAT5 signal became study was done using inguinal lymph nodes from Foxp3-eGFP restricted around the IL-2 producing cells as we increased the mice that were shown to respond to transiently-secreted IL-2 number of IL-2 consuming Treg cells. at homeostasis. Directly proﬁling the spatial distribution of pSTAT5, as done To estimate the extent of IL-2 signaling in Foxp3-eGFP mice in vitro, is impossible here: First, only a fraction of AdTr 5C.C7 at homeostasis, we re-analyzed the images presented in (Liu cells get activated (Figure S5). In addition, Treg cells might be re- et al., 2015) and calculated the pSTAT5 autocorrelation function sponding to an ‘‘invisible’’ secreting cell outside of the 5 mm tis- in lymph node sections collected from these mice (Figure 6D). sue section. Thus, to further quantify the extent of signaling we We note that the density of Treg cells in these animals was calculated G(r), the pSTAT5 autocorrelation function for the slightly elevated compared to the B10.A wild-type mice. Within spatial distribution of pSTAT5 (Figure 6D). Our theory predicts these organs, Treg cells constitute approximately 15% of the that this function should be a decaying exponential with a char- T cell zone. We found that in agreement with our theory, the auto- acteristic length-scale lniche, independent of the production rate correlation function for the spatial distribution of pSTAT5 de- of the AdTr 5C.C7 T cells (see Section 2 of Supplemental Infor- cayed exponentially. The signaling length scale in homeostatic mation). Due to the irregular structures and variable sizes of pSTAT5+ clusters decreased slightly to 9.5 ± 0.2 cell diameters the T cell zones in spleen nodules, calculating the autocorrela- as was expected by the difference in Treg cells content. These

Immunity 46, 609–620, April 18, 2017 615 FoxP3 pSTAT5 DAPI Producers A Wild type IL−2 i.c. Boosted

100 m 100 m

Spleen Spleen B FoxP3−eGFP(Homeostasis) Wild type IL−2 i.c. Boosted

100 m 100 m 100 m

Lymph node Lymph node Lymph node

CDr (microns) %IL−2R + (of CD4+) 50 100 150 10 100

1 = 11.4 0.7 cd WT G(r) = 8.2 0.7 cd − B 10 = 9.5 0.2 cd F

0.5 Wild type

Wild type (cell diameter) IL−2 i.c. Boosted IL−2 i.c. Boosted FoxP3−GFP FoxP3−GFP in vitro pSTAT5 autocorrelation 0.2 1 5 10 15 0.1 1 r (cell diameter) IL−2R + cell density (per 103 m3)

Figure 6. The Size and Shape of Signaling Micro Domains In Vivo, Are Explained by a Tunable Diffusion-Consumption Mechanism (A and B) Immunofluorescence staining of spleen nodule (A) and lymph nodes (B) sections from wild-type (middle) and IL-2 i.c. boosted (right) mice after peptide immunization, and of lymph node section from Foxp3-eGFP mouse at homeostasis (left). (C) pSTAT5 autocorrelation functions were calculated for the immunized IL-2 i.c. boosted and wild-type B10.A mice, and for Foxp3-eGFP B6 mice at homeostasis, and fit to a decaying exponential (See section 3.3 of Supplemental Information for details). The correlations decay coefficient is the signaling length scale lniche. Results quantify an experiment using three mice per group and 3–4 lymph nodes per mouse. Whenever errors do not appear they are smaller than markers. (D) Measured signaling length scales for the three different mouse groups, and for in vitro samples (Figure 4), and estimations based on known microscopic parameters. Black dashed line are the predicted values based on average parameters, gray dashed lines are extreme values given the biological range of parameters. See also Figures S5 and S6.

616 Immunity 46, 609–620, April 18, 2017 measurements are consistent with our theory and our measure- must occur, implying perfusion of the cytokine throughout the ments for conditions with immunized T cells (Figure 6D). Hence, lymphoid organ. We observed that the density of cytokine cytokine interactions appear to be of the same nature at homeo- consuming cells changes during an immune response. As a stasis and during an immune response. consequence, cytokine could spread over many cell diameters, We concluded that a simple diffusion-consumption mecha- or signal only to a producer’s closest neighbors. nism is sufficient to account for the formation of cytokine niches IL-4 and IFN-g are known to participate in positive feedback in dense tissues. We find quantitative agreement between the loops. Exposure of activated T cells to IL-4 or IFN-g can induce shapes of cytokine niches in vivo and in vitro and our param- production of either cytokine, respectively (Zhu et al., 2010). This eter-free theoretical predictions (Figure 6D). ‘‘cytokine relay’’ could increase the fraction of producing T cells in the tissue, resulting in more cells with cytokine-producing DISCUSSION neighbors. This could explain why both IL-4 and IFN-g were shown to perfuse the reactive lymph nodes during the potent In this study, we demonstrated that the dense cell conditions that immune responses elicited by either helminths or protozoa exist in vivo limit the extent of cytokine communication through a (Perona-Wright et al., 2010). Furthermore cytokines can also tunable diffusion-consumption mechanism. The mechanism is influence the expression of specific surface receptors. For general and its effects do not appear to be significantly affected example, IL-2 has been shown to increase expression of both by different complexities of the living system such as the motion the IL-4Ra and IL-12Rb chains, while decreasing both of the of cells, directionality in cytokine secretion, lymphatic flow, etc. IL-6 receptor chains (Liao et al., 2008, 2011). Therefore cytokines Before discussing some ramifications of this work, we will briefly themselves could dynamically alter the extent of signaling for summarize our findings. either themselves or others by controlling cytokine receptor Our over-arching conclusion is that the length-scale of cyto- expression. kine communication is dynamic, predictable, and can be ex- One distinctive feature of T cell differentiation is the spectac- plained by a simple diffusion-consumption mechanism. The ular phenotypic and functional diversity generated following acti- characteristic length-scale depends on the density of cytokine vation (Gerlach et al., 2013; Ma et al., 2011; Newell et al., 2012; ½ consumers as lniche f nconsumers and is typically on the order Zhu et al., 2010). However, it is unknown exactly how this hetero- of 30–150 microns. In a three-dimensional system, this results in geneity is achieved. Our study suggests that T cells spontane- discrete, self-assembled, spherical, cytokine niches around ously self-organize into dynamic micro domains of variable cytokine-producing cells. Our results indicate that the size of cytokine content during the course of an immune response. such cytokine niches depends on the density of cytokine- This constitutes a fundamental mechanism leading to pheno- consuming cells. Given that populations of consuming cells typic variability in cell differentiation as observed in recent sin- vary during an immune response, this length-scale will also swell gle-cell measurements (Shalek et al., 2013, 2014). and shrink with time. Our study illustrates that in designing and interpreting experi- Previous studies have yielded conflicting results regarding ments, special consideration should be given to the physical how far cytokines spread from their source during an immune structure of the tissue where immune reactions take place. response. One set of experiments showed that the cytokines Indeed, the use of conventional tissue culture plates is often IL-4 and IFN-g completely perfuse the reactive lymph node dur- not an appropriate representation of in vivo physiological ing infection, while others have observed spatial heterogeneity in conditions. cytokine spread (Huse et al., 2006; Maldonado et al., 2004; Pan- There is a body of work demonstrating how differential expo- gault et al., 2010; Perona-Wright et al., 2010; Sabatos et al., sure to cytokines or nutrients generates phenotypically different 2008). Our work reconciles these divergent results (Huse et al., immune cells (Antebi et al., 2013; Fleming and Paige, 2001; 2006; Maldonado et al., 2004; Pangault et al., 2010; Perona- Marshall et al., 1998). These studies have been limited so far to Wright et al., 2010; Sabatos et al., 2008; Thurley et al., 2015) either in vitro conditions (Antebi et al., 2013), or to circumstantial by revealing that the size of cytokine niches is not a static feature in vivo evidence (Fleming and Paige, 2001; Marshall et al., 1998). of the system, but is instead dynamic, such that both signaling Our study provides a description of how this differential expo- outcomes (perfusion versus heterogeneity) are possible based sure is generated and opens up the possibility of directly map- on the state of the system. Reconciling these conflicting results ping positional information onto phenotypic diversity. Further- carries with it important downstream biological implications. On more, we predict that local variations in access to cytokines, one hand, spatial heterogeneity in cytokine signaling generates nutrients, or growth factors, provide an important source of the functional and phenotypic variability that exists between non-genetic gene-expression heterogeneity in dense tissues, immune cells (Shalek et al., 2014; Snijder and Pelkmans, such as lymphoid organs and tumors (Carmona-Fontaine et al., 2011). After antigen stimulation, both CD4+ and CD8+ T cells 2013). Future work should focus on translating cell-to-cell vari- undergo differentiation into diverse subpopulations with vastly ability in gene expression to cytokine or growth factor access. different functional capabilities (Ma et al., 2011; Newell et al., 2012; Zhu et al., 2010). This diversity confers flexibility to the immune system and maximizes an individuals’ ability to respond EXPERIMENTAL PROCEDURES effectively to different pathogens and doses. The mechanisms Mice that generate this functional and phenotypic diversity are still C57BL/6, 5C.C7 TCR-transgenic Rag2/, 5C.C7 TCR-transgenic Rag2/ active areas of investigation. On the other hand, to orchestrate Il2/ mice were obtained from Taconic Laboratories. B10.A mice were ob- the collective response of many individual cells, communication tained from The Jackson Laboratory. Mice were maintained in SPF conditions

Immunity 46, 609–620, April 18, 2017 617 at an Association for Assessment and Accreditation of Laboratory Animal cells were labeled using the Cell Trace Far-Red Cell Proliferation Kit (Molecular Care-accredited animal facilities. 10- to 14-week-old mice were used for all Probes). Labeled cells were transferred by intravenous injection at 5 3 106 or experiments and were randomly allocated into treatment groups. Mice were 5 3 105 CD4+ T cells per recipient. 18 hr after transfer, recipient mice were always sex-matched for in vivo experiments. The Institutional Animal Care immunized via intraperitonal injection with 100 mg per mouse K5 peptide and Use Committee of Memorial Sloan Kettering Cancer Center and Ben- (ANERADLIAYFKAATKF, GenScript). 10 mg per mouse of LPS (Invivogen) was Gurion University of the Negev approved animal experiments. used as an adjuvant. Six hours post-immunization mice were rapidly euthanized by cervical dislocation. Lymph nodes and spleens were rapidly dissected and Media and Culture Conditions prepared for immunofluorescence microscopy and flow cytometry. Primary T cells, B16-F10 melanoma cells (ATCC CRL-6475), and CH12 B cell lymphoma cells were maintained in, and all experiments were performed Immunohistochemistry with RPMI 1640 media supplemented with heat-inactivated 10% fetal calf Lymph nodes and spleen segments were harvested and immediately fixed us- serum, 2 mM L-glutamine, 10 mM HEPES, 0.1 mM non-essential amino acids, ing fixation buffer containing 4% PFA for 4 hr and then dehydrated in 70% 1 mM sodium pyruvate, 100 mg/ml of penicillin, 100 mg/ml of streptomycin, and EtOH before embedding in paraffin wax. 5 mm sections were dewaxed and 50 mM b-mercaptoethanol. blocked for 30 min in blocking buffer containing 2% bovine serum albumin C57BL/6, B10.A primary cells were harvested from the lymph nodes and and 10% normal goat serum. Sections were stained sequentially for FoxP3 spleen. T cells were activated using 10 ng/ml PMA and 500 ng/ml Ionomycin and pSTAT5 using a machine-based, multiplexed Tyramide Signal Amplifica- and cultured for 3 days. Dead cells were removed using Ficoll-Paque plus tion technique (Yarilin et al., 2015). and subsequently cultured in RPMI supplemented with 2 nM recombinant hu- The following reagents were used for staining: anti-FoxP3 (FJK-16 s, eBio- man IL-2 for up to 10 days. science), anti-pSTAT5 (C71E5, Cell Signaling Technologies), biotinylated For in vitro imaging experiments, IL-2 production was induced by restimulat- goat anti-rabbit IgG (Vector labs), biotinylated goat anti-rat IgG (Vector labs), ing cultured T cells with 5 ng/ml PMA and 500 ng/ml Ionomycin for 4–6 hr. Streptavidin-HRP D (Ventana Medical Systems), Tyramide-Alexa Fluor 488 For in vivo experiments, lymphocytes were harvested from the lymph (Invitrogen), Tyramide-Alexa Fluor 568 (Invitrogen), DAPI. nodes and spleen of TCR transgenic 5C.C7 Rag2/ mice and 5C.C7 The procedures were performed at the Molecular Cytology Core Facility Rag2/Il-2/ mice. of Memorial Sloan Kettering Cancer Center using Discovery XT processor (Ventana Medical Systems). Flow Cytometry Images were collected on a Pannoramic 250 Flash Whole slide scanner (3D T cells were fixed for 10 min in 1.6% PFA on ice and permeabilized in 90% meth- Histech). Images from the (Liu et al., 2015) paper and additional unpublished anol for at least 30 min at 20C. Cells were stained with anti-CD4 (RM4-5: BD data were collected as reported in (Liu et al., 2015) and analyzed by the Bioscience), anti-IL-2Ra (PC61.5: BD Bioscience), and anti-pSTAT5 (C11C5: same methods as in the new in vivo immunization studies presented here. Cell Signaling Technologies). Cells were labeled with Cell Trace Violet (CTV) Analysis was done using the OpenSlide package (Goode et al., 2013) and or Cell Trace Far-Red (CTFR) (Molecular Probes) in accordance with experi- MATLAB software (Mathworks). mental design. B16-F10 and CH12 cells were fixed as above and stained with anti-pSTAT1 (58D6: Cell Signaling Technologies) and anti-pSTAT6 Statistical Analysis and Data Presentation (#9361: Cell Signaling Technologies), respectively. For FoxP3 staining, cells All relevant data are shown as (mean ± SEM). Statistical analyses were per- were treated with FoxP3 fixation/permeabilization kit (eBioscience) and stained formed using MATLAB software. Statistical tests were selected based on with anti-FoxP3 (MF23: Biolegend). IL-2 production was measured using the appropriate assumptions with respect to data distribution and variance char- Mouse IL-2 Secretion Assay Kit (Miltenyi Biotec). acteristics. Student’s t test (two-tailed) was used for the statistical analysis of Flow cytometric data were collected on an LSR II (BD Biosciences). Analysis differences between two groups. In Figures 3D, 3E, 4C, and 6E, cell density is was done with FlowJo software (TreeStar), and with in-house designed tools calculated based on a calibration of cell fraction to spatial density (Figure S2D) written in MATLAB (Mathworks). Codes are available upon request. in in vivo conditions. The absolute density in our in vitro assays is comparable to in vivo conditions (data not shown). In Vitro Immunofluorescence Staining and Imaging using PlaneView Device SUPPLEMENTAL INFORMATION Cell pellets were incubated on a poly-l-lysine coated glass slide under a semipermeable membrane (as described in section 3.2, see Supplemental Exper- Supplemental Information includes six figures, two tables, Mathematical imental Procedures). Cells were fixed, without removing the membrane, in 4% Framework, Theory and Data Analysis, and Supplemental Experimental warm PFA for 15 m at room temperature, and permeabilized in 90% methanol Procedures and can be found with this article online at http://dx.doi.org/10. for longer than 30 m at 20C. The membrane was then removed and slides 1016/j.immuni.2017.03.011. blocked for 1 hr with a blocking buffer containing 2% bovine serum albumin and 0.3% Triton X-100. Slides were stained with anti-IL-2Ra (7D4, Miltenyi Bio- AUTHOR CONTRIBUTIONS tec) and anti-pSTAT5 (C71E5: Cell Signaling Technologies), and then with anti- rabbit IgG (Jackson ImmunoResearch) for 1 hr at room temperature in a dark, A.O.-Y., J.O.-Y., G.A.-B., and O.K. designed the experiments. A.O.-Y., J.O.-Y., humidified chamber. Finally, DAPI counterstaining was performed for 5 min and G.A.-B. performed the experiments and collected data. A.O.-Y., G.A.-B., and slides were mounted in Fluoromount Aqueous Mounting Medium (Sigma). and O.K. analyzed the data. Z.L. and R.N.G. provided data from in vivo homeo- Images were collected on a Zeiss Axiovert 200M microscope using home- stasis experiments. B.M.W. and M.H. performed single cell IL-2 secretion made acquisition software programmed in LabView (National Instruments). dynamics measurements. A.O.-Y., J.O.-Y., G.A.-B., and O.K. wrote the Image analysis was done using MATLAB (Mathworks). manuscript.

Boosting Treg Cell Compartment and Adoptive Transfer ACKNOWLEDGMENTS The protocol for boosting the Treg cell compartment in mice was adapted from Boyman et al. (Boyman et al., 2006a). To form immunocomplexes, 1 mg We thank the Molecular Cytology Core Facility at MSKCC for help with histol- mouse-IL-2 (Biolegend) and 5 mg anti-mouse-IL-2 (JES6-1A12, BioXCell), ogy and imaging. A.O.-Y. and O.K. are grateful to Professor Noah Isakov, Mar- per mouse, were mixed and incubated at room temperature for 10 min before galit Krup, and Gad Afek for help in setting up experiments at Ben-Gurion Uni- being diluted in PBS to a volume of 100 ml. Three doses were administered via versity. This work was supported in part by the U.S.-Israel Binational Science intraperitoneal injection every 24 hr to B10.A mice. Mice were then rested for Foundation (#2012327 to G.A.-B. and O.K.), by the US National Institutes of 1 day before proceeding with adoptive transfer and immunization. Health (R01-AI083408 to G.A.-B. and M.H., and R01-AI087644 to M.H.), by The lymph nodes and spleen were collected from TCR transgenic Rag2/ the Intramural Research programs of the Center for Cancer Research, NCI, 5C.C7 or Rag2/Il2/ TCR transgenic 5C.C7 mice. After red blood cell lysis, and by the Intramural Research programs of the NIAID, NIH.

618 Immunity 46, 609–620, April 18, 2017 Received: September 17, 2016 Houchmandzadeh, B., Wieschaus, E., and Leibler, S. (2002). Establishment of Revised: January 23, 2017 developmental precision and proportions in the early Drosophila embryo. Accepted: February 16, 2017 Nature 415, 798–802. Published: April 4, 2017 Huang, J., Brameshuber, M., Zeng, X., Xie, J., Li, Q.-J., Chien, Y.-H., Valitutti, S., and Davis, M.M. (2013). A Single Peptide-Major Histocompatibility Complex Ligand Triggers Digital Cytokine Secretion in CD4 (T Cells. Immunity). REFERENCES Huse, M., Lillemeier, B.F., Kuhns, M.S., Chen, D.S., and Davis, M.M. (2006). T Amado, I.F., Berges, J., Luther, R.J., Mailhe´ , M.-P., Garcia, S., Bandeira, A., cells use two directionally distinct pathways for cytokine secretion. Nat. Weaver, C., Liston, A., and Freitas, A.A. (2013). IL-2 coordinates IL-2-produc- Immunol. 7, 247–255. ing and regulatory T cell interplay. J. Exp. Med. 210, 2707–2720. Liao, W., Schones, D.E., Oh, J., Cui, Y., Cui, K., Roh, T.-Y., Zhao, K., and Leonard, W.J. (2008). Priming for T helper type 2 differentiation by interleukin Antebi, Y.E., Reich-Zeliger, S., Hart, Y., Mayo, A., Eizenberg, I., Rimer, J., 2-mediated induction of interleukin 4 receptor alpha-chain expression. Nat. Putheti, P., Pe’er, D., and Friedman, N. (2013). Mapping differentiation under Immunol. 9, 1288–1296. mixed culture conditions reveals a tunable continuum of T cell fates. PLoS Biol. 11, e1001616. Liao, W., Lin, J.-X., Wang, L., Li, P., and Leonard, W.J. (2011). Modulation of cytokine receptors by IL-2 broadly regulates differentiation into helper T cell Berg, H.C., and Purcell, E.M. (1977). Physics of chemoreception. Biophys. J. lineages. Nat. Immunol. 12, 551–559. 20, 193–219. Liu, Z., Gerner, M.Y., Van Panhuys, N., Levine, A.G., Rudensky, A.Y., and Boyman, O., Kovar, M., Rubinstein, M.P., Surh, C.D., and Sprent, J. (2006a). Germain, R.N. (2015). Immune homeostasis enforced by co-localized effector Selective stimulation of T cell subsets with antibody-cytokine immune com- and regulatory T cells. Nature 528, 225–230. plexes. Science 311, 1924–1927. Ma, C., Fan, R., Ahmad, H., Shi, Q., Comin-Anduix, B., Chodon, T., Koya, R.C., Boyman, O., Surh, C.D., and Sprent, J. (2006b). Potential use of IL-2/anti-IL-2 Liu, C.-C., Kwong, G.A., Radu, C.G., et al. (2011). A clinical microchip for eval- antibody immune complexes for the treatment of cancer and autoimmune dis- uation of single immune cells reveals high functional heterogeneity in pheno- ease. Expert Opin. Biol. Ther. 6, 1323–1331. typically similar T cells. Nat. Med. 17, 738–743. Busse, D., de la Rosa, M., Hobiger, K., Thurley, K., Flossdorf, M., Scheffold, A., Maldonado, R.A., Irvine, D.J., Schreiber, R., and Glimcher, L.H. (2004). A role and Ho¨ fer, T. (2010). Competing feedback loops shape IL-2 signaling between for the immunological synapse in lineage commitment of CD4 lymphocytes. helper and regulatory T lymphocytes in cellular microenvironments. Proc. Natl. Nature 431, 527–532. Acad. Sci. USA 107, 3058–3063. Marshall, A.J., Fleming, H.E., Wu, G.E., and Paige, C.J. (1998). Modulation of Carmona-Fontaine, C., Bucci, V., Akkari, L., Deforet, M., Joyce, J.A., and the IL-7 dose-response threshold during pro-B cell differentiation is dependent Xavier, J.B. (2013). Emergence of spatial structure in the tumor microenviron- on pre-B cell receptor expression. J. Immunol. 161, 6038–6045. ment due to the Warburg effect. Proc. Natl. Acad. Sci. USA 110, 19402–19407. Muller,€ A.J., Filipe-Santos, O., Eberl, G., Aebischer, T., Spath,€ G.F., and Duprez, V., Ferrer, M., Cornet, V., Olive, D., and Dautry-Varsat, A. (1991). Bousso, P. (2012). CD4+ T cells rely on a cytokine gradient to control intracel- Modulation of interleukin 2 internalization and interleukin 2-dependent cell lular pathogens beyond sites of antigen presentation. Immunity 37, 147–157. growth by antireceptor antibodies. J. Biol. Chem. 266, 1497–1501. Newell, E.W., Sigal, N., Bendall, S.C., Nolan, G.P., and Davis, M.M. (2012). Economopoulos, V., Noad, J.C., Krishnamoorthy, S., Rutt, B.K., and Foster, Cytometry by time-of-flight shows combinatorial cytokine expression and P.J. (2011). Comparing the MRI appearance of the lymph nodes and spleen virus-specific cell niches within a continuum of CD8+ T cell phenotypes. in wild-type and immuno-deficient mouse strains. PLoS ONE 6, e27508. Immunity 36, 142–152. Feinerman, O., Jentsch, G., Tkach, K.E., Coward, J.W., Hathorn, M.M., Pandiyan, P., Zheng, L., Ishihara, S., Reed, J., and Lenardo, M.J. (2007). Sneddon, M.W., Emonet, T., Smith, K.A., and Altan-Bonnet, G. (2010). CD4+CD25+Foxp3+ regulatory T cells induce cytokine deprivation-mediated Single-cell quantification of IL-2 response by effector and regulatory T cells apoptosis of effector CD4+ T cells. Nat. Immunol. 8, 1353–1362. reveals critical plasticity in immune response. Mol. Syst. Biol. 6, 437. Pangault, C., Ame´ -Thomas, P., Ruminy, P., Rossille, D., Caron, G., Baia, M., De Vos, J., Roussel, M., Monvoisin, C., Lamy, T., et al. (2010). Follicular lym- Fleming, H.E., and Paige, C.J. (2001). Pre-B cell receptor signaling mediates phoma cell niche: identification of a preeminent IL-4-dependent T(FH)-B cell selective response to IL-7 at the pro-B to pre-B cell transition via an ERK/ axis. Leukemia 24, 2080–2089. MAP kinase-dependent pathway. Immunity 15, 521–531. Paul, W.E., and Seder, R.A. (1994). Lymphocyte responses and cytokines. Cell Fontenot, J.D., Rasmussen, J.P., Gavin, M.A., and Rudensky, A.Y. (2005). 76, 241–251. A function for interleukin 2 in Foxp3-expressing regulatory T cells. Nat. Perona-Wright, G., Mohrs, K., and Mohrs, M. (2010). Sustained signaling by Immunol. 6, 1142–1151. canonical helper T cell cytokines throughout the reactive lymph node. Nat. Gerlach, C., Rohr, J.C., Perie´ , L., van Rooij, N., van Heijst, J.W.J., Velds, A., Immunol. 11, 520–526. Urbanus, J., Naik, S.H., Jacobs, H., Beltman, J.B., et al. (2013). Sabatos, C.A., Doh, J., Chakravarti, S., Friedman, R.S., Pandurangi, P.G., Heterogeneous differentiation patterns of individual CD8+ T cells. Science Tooley, A.J., and Krummel, M.F. (2008). A synaptic basis for paracrine inter- 340, 635–639. leukin-2 signaling during homotypic T cell interaction. Immunity 29, 238–248. Goode, A., Gilbert, B., Harkes, J., Jukic, D., and Satyanarayanan, M. (2013). Shalek, A.K., Satija, R., Adiconis, X., Gertner, R.S., Gaublomme, J.T., OpenSlide: A vendor-neutral software foundation for digital pathology. Raychowdhury, R., Schwartz, S., Yosef, N., Malboeuf, C., Lu, D., et al. J. Pathol. Inform. 4,27. (2013). Single-cell transcriptomics reveals bimodality in expression and Helmstetter, C., Flossdorf, M., Peine, M., Kupz, A., Zhu, J., Hegazy, A.N., splicing in immune cells. Nature 498, 236–240. Duque-Correa, M.A., Zhang, Q., Vainshtein, Y., Radbruch, A., et al. (2015). Shalek, A.K., Satija, R., Shuga, J., Trombetta, J.J., Gennert, D., Lu, D., Chen, Individual T helper cells have a quantitative cytokine memory. Immunity 42, P., Gertner, R.S., Gaublomme, J.T., Yosef, N., et al. (2014). Single-cell RNA- 108–122. seq reveals dynamic paracrine control of cellular variation. Nature 510, Heltemes-Harris, L.M., Willette, M.J.L., Vang, K.B., and Farrar, M.A. (2011). 363–369. The role of STAT5 in the development, function, and transformation of B and Snijder, B., and Pelkmans, L. (2011). Origins of regulated cell-to-cell variability. T lymphocytes. Ann. N Y Acad. Sci. 1217, 18–31. Nat. Rev. Mol. Cell Biol. 12, 119–125. Ho¨ fer, T., Krichevsky, O., and Altan-Bonnet, G. (2012). Competition for IL-2 Sojka, D.K., Bruniquel, D., Schwartz, R.H., and Singh, N.J. (2004). IL-2 secre- between Regulatory and Effector T Cells to Chisel Immune Responses. tion by CD4+ T cells in vivo is rapid, transient, and influenced by TCR-specific Front. Immunol. 3, 268. competition. J. Immunol. 172, 6136–6143.

Immunity 46, 609–620, April 18, 2017 619 Spangler, J.B., Tomala, J., Luca, V.C., Jude, K.M., Dong, S., Ring, A.M., Wartlick, O., Kicheva, A., and Gonza´ lez-Gaita´ n, M. (2009). Morphogen Votavova, P., Pepper, M., Kovar, M., and Garcia, K.C.C. (2015). Antibodies gradient formation. Cold Spring Harb. Perspect. Biol. 1, a001255. to Interleukin-2 Elicit Selective T Cell Subset Potentiation through Distinct Webster, K.E., Walters, S., Kohler, R.E., Mrkvan, T., Boyman, O., Surh, C.D., Conformational Mechanisms. Immunity 42, 815–825. Grey, S.T., and Sprent, J. (2009). In vivo expansion of T reg cells with IL-2- Thurley, K., Gerecht, D., Friedmann, E., and Ho¨ fer, T. (2015). Three- mAb complexes: induction of resistance to EAE and long-term acceptance Dimensional Gradients of Cytokine Signaling between T Cells. PLoS of islet allografts without immunosuppression. J. Exp. Med. 206, 751–760. Comput. Biol. 11, e1004206. Yarilin, D., Xu, K., Turkekul, M., Fan, N., Romin, Y., Fijisawa, S., Barlas, A., and Manova-Todorova, K. (2015). Machine-based method for multiplex in situ Tkach, K.E., Barik, D., Voisinne, G., Malandro, N., Hathorn, M.M., Cotari, molecular characterization of tissues by immunoﬂuorescence detection. Sci. J.W., Vogel, R., Merghoub, T., Wolchok, J., Krichevsky, O., and Altan- Rep. 5, 9534. Bonnet, G. (2014). T cells translate individual, quantal activation into collective, analog cytokine responses via time-integrated feedbacks. eLife 3, Zhu, J., Yamane, H., and Paul, W.E. (2010). Differentiation of effector CD4 e01944. T cell populations (*). Annu. Rev. Immunol. 28, 445–489.

620 Immunity 46, 609–620, April 18, 2017