network crosstalk dynamically changes during neutrophil polarization

Evolving cross-talk in the neutrophil polarity network

Chin-Jen Ku1,*, Yanqin Wang1,*, Orion Weiner2, Steven J. Altschuler1, and Lani F. Wu1

1Department of Pharmacology, Green Center for Systems Biology, Simmons Cancer Center,University of Texas Southwestern Medical Center, Dallas TX 75390, USA2Cardiovascular Research Institute and Department of Biochemistry, University of California, SanFrancisco, San Francisco CA 94158, USA

SUMMARYHow complex signaling networks shape highly-coordinated, multistep cellular responses is poorlyunderstood. Here, we made use of a network-perturbation approach to investigate causalinfluences, or “cross-talk,” among signaling modules involved in the cytoskeletal response ofneutrophils to chemoattractant. We quantified the intensity and polarity of cytoskeletal markerproteins over time to characterize stereotyped cellular responses. Analyzing the effects of networkdisruptions revealed that not only does cross-talk evolve rapidly during polarization but alsointensity and polarity responses are influenced by different patterns of cross-talk. Interestingly,persistent cross-talk is arranged in a surprisingly simple circuit: a linear cascade from front to backto microtubules influences intensities, and a feed-forward network in the reverse directioninfluences polarity. Our approach provided a rational strategy for decomposing a complex,dynamically evolving, signaling system and revealed evolving paths of causal influence that shapethe neutrophil polarization response.

INTRODUCTIONA central challenge in biology is to determine how signals are dynamically transducedthrough signaling networks (Amit et al., 2009; Cheong et al., 2011; Levine et al., 2006;Muzzey et al., 2009; Sachs et al., 2005; Yu et al., 2008). Signaling networks are often cast ashighly complex, static structures, containing many components and interactions inferred bycombining results from diverse assays. One approach for studying how these networksprocess information is to combine detailed biochemical measurements with mathematicalanalyses and modeling (Kalir and Alon, 2004; Kentner and Sourjik, 2009; Lee et al., 2003;Marco et al., 2007; Oleksiuk et al., 2011). Such studies illuminate the contribution of eachcomponent in shaping the responses of the other components. However, obtaining sufficientbiochemical constants required for accurate, physical models of complex networks can bedifficult. Complete sets of measurements typically do not exist, and it is often not obviouswhich biochemical properties of the individual components have the strongest influence onthe output of the system.

© 2012 Elsevier Inc. All rights reserved.

Correspondence and requests for materials should be addressed to S.J.A ([email protected]) or L.F.W.([email protected]).*These authors contributed equally to the work.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to ourcustomers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review ofthe resulting proof before it is published in its final citable form. Please note that during the production process errors may bediscovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

NIH Public AccessAuthor ManuscriptCell. Author manuscript; available in PMC 2013 May 25.

Published in final edited form as:Cell. 2012 May 25; 149(5): 1073–1083. doi:10.1016/j.cell.2012.03.044.

NIH

-PA Author Manuscript

NIH


NIH


An alternative approach for characterizing signal processing is to quantify the effects ofperturbations to network components (Janes et al., 2006; Muzzey et al., 2009; Natarajan etal., 2006; Sachs et al., 2005; Schneider and Haugh, 2006; Tkachenko et al., 2011).Perturbation analyses have enabled inference of causal network structure, including thereconstruction of a static, causal protein-signaling network in T-cells (Sachs et al., 2005) andthe characterization of dynamic cross-talk induced by co-stimulation with multiple growthfactors in epithelial cells (Janes et al., 2006). As with epistasis experiments in genetics,causal interactions can be uncovered without requiring detailed biochemical mechanism orproximity within the network of components.

Neutrophils, innate immune cells that detect, hunt and kill bacteria, offer an ideal system forstudying dynamic signal processing. The polarity network of neutrophils responds rapidly tochemoattractant. Upon stimulation with fMLP, neutrophils undergo a rapid, highlystereotyped progression through distinct stages: cells initiate polarization within seconds,develop polarization within 2 to 3 minutes, and then maintain their polarized state for about10 minutes before adapting (Zigmond et al., 1981; Zigmond and Sullivan, 1979). Transitionsamong these different stages could reflect changes in the underlying composition of thepolarity network. For example, different phases of the cellular differentiation or cell cycleare due, in part, to the synthesis and degradation of key regulatory components (deLichtenberg et al., 2005; Fraser and Germain, 2009; Loose et al., 2007; Luscombe et al.,2004). However, for the rapid time scale of neutrophil polarization, this multiphasicresponse is unlikely to be guided simply by the appearance or disappearance of networkcomponents.

Many components and interactions involved in this neutrophil polarity network have beenidentified and placed within a small number of spatially and molecularly distinctcytoskeletal “modules” (Niggli, 2003; Servant et al., 2000; Small et al., 2002; Srinivasan etal., 2003; Van Keymeulen et al., 2006; Weiner et al., 2002; Weiner et al., 2006; Wong et al.,2006; Wong et al., 2007; Xu et al., 2003; Xu et al., 2005). These include: a front module thatpromotes membrane protrusion in the direction of the chemoattractant through F-actinassembly at the leading edge of the cell; a back module that regulates cell contractionthrough activation of myosin light chain 2 (MLC2) at the rear end; and a module thatconsists of microtubules and associated proteins that is thought to act both as a sink anddelivery system for several polarity components (Figure 1A; Figure S1). These modulesprovide a tractable starting point for exploring causal interactions, or “cross-talk,” betweendistinct biochemical networks that produce a complex behavioral phenotype.

Numerous routes of communication between these cytoskeletal modules have beenidentified (front to back, front to microtubules, microtubule to back, and back to front). Thefront recruits the SCAR/WAVE complex, which not only shapes the actin architecture at theleading edge but also acts as a scaffold for inhibitors of the Rho/myosin back program(Weiner et al., 2006). Likewise, the back program opposes the actin-based membraneprotrusions necessary to support leading edge activities (Weiner et al., 2007). Themicrotubule module delivers Rho-family GEFs to the membrane to locally activate the Rho/myosin back program (Krendel et al., 2002; Odell and Foe, 2008; Xu et al., 2005), and actinpolymerized at the leading edge locally excludes microtubules (Eddy et al., 2002). How thiscross-talk occurs in space and time to give rise to a rapid, multiphasic response inneutrophils is poorly understood.

Here, we investigated paths of communication between front, back, and microtubulemodules in stimulated, primary human neutrophils using a network-perturbation approach.We made use of quantitative microscopy and fixed-cell assays over multiple time points tocapture the spatial-temporal dynamics of ~105 polarizing neutrophils. We labeled cells with

Ku et al.

Cell. Author manuscript; available in PMC 2013 May 25.

NIH


NIH


NIH


three cytoskeletal marker proteins that served as proxies (or direct readouts) for the states ofthe front, back, and microtubule modules. For each marker, we measured two phenotypes tocapture its activity levels or spatial distribution. To identify cross-talk among the modules,populations of neutrophils were treated before stimulation with pharmacologicalperturbations. Six mechanistically distinct drugs were selected, with pairs of these drugstargeting each of the three modules with “opposite” effects. This approach allowed us tocapture cross-talk between modules by quantifying the degree to which perturbed responsecurves for marker activity levels and spatial distributions deviated from control. Our analysisrevealed that cross-talk interactions differ across different phases of polarization. Further,cross-talk is different for activity and spatial distribution of the signaling markers, withpersistent cross-talk arranged in a surprisingly simple circuit: a linear cascade from front toback to microtubule modules influences intensities, and a feed-forward network in thereverse direction influences polarity.

RESULTSMeasurement of polarization responses in primary human neutrophils

We chose to analyze polarization responses in primary human neutrophils due to theirphysiological relevance and lower cell-to-cell variability compared with cultured cell lines.Given the experimental challenges of measuring live readouts of signaling activities in theseshort-lived cells, we inferred the phenotypic responses of neutrophils based on statisticscollected from populations of cells fixed at different time points. Freshly harvested humanneutrophils were seeded onto 96-well plates followed by 30 min treatment of drug or controlvehicle (DMSO). After uniform stimulation of chemoattractant f-Met-Leu-Phe (fMLP,10nM), cells were fixed at eleven time points ranging from 0 to 600 seconds. The timepoints were chosen non-uniformly to capture different phases of the polarization process(Figure 1B, top; Supplemental Information).

We co-stained cells for F-actin, monophosphorylated myosin light chain 2 (p-MLC2) and α-tubulin to obtain integrated signaling readouts of front, back and microtubule modules,respectively (Figure 1B, bottom). F-actin assembly at the front is the engine for leading edgeadvance and represents a proxy for upstream signals (like Rac activation) that controlprotrusion. Phosphorylation of MLC2 activates the contractile ability of myosin and is aproxy for upstream signals (like Rho activation) that activate trailing edge contraction. Themicrotubule module is both a sink and delivery system for polarity components; the massand spatial distribution of microtubules are thought to regulate neutrophil polarity. Inaddition to these primary readouts, we also stained neutrophil nuclei (Hoechst) and acquiredbright field microscopy images to perform image segmentation and image quality control(Ku et al., 2010).

To identify individual cellular regions from microscopy images, we made use of ourpreviously developed image segmentation algorithm (Ku et al., 2010). After intensitynormalization (Supplemental Information, Figure S2A), we extracted two phenotypes foreach marker on a cell-by-cell basis. First, we measured the average intensity of each markerto capture an integrated activity level of its associated module. For example, a high value ofF-actin intensity indicated high levels of front activity. Second, we measured the polarity ofeach marker to capture the spatial extent of the most active signaling region. Morespecifically, polarity measured how tightly packed the brightest pixels were inside the cell(Supplemental Information) (Ku et al., 2010). For example, uniform actin assembly gives alow value for F-actin polarity whereas tight concentration of actin at the leading edge gives ahigh value for F-actin polarity. Similarly, the higher the value of polarity for p-MLC2 ormicrotubules, the tighter is their spatial distribution.

Ku et al.


NIH


NIH


NIH


To perturb each module in the neutrophil polarity network, we made use of pairs ofpharmacological compounds with “opposite” effects (Figures 1A–B): the F-actindepolymerizer latrunculin A (LatA) or stabilizer jasplakinoide (Jas) to perturb the frontmodule, the microtubule depolymerizer nocodazole (Noco) or stabilizer taxol (Taxol) toperturb the microtubule module, and the myosin phosphorylation inhibitor Y27632 oractivator calpeptin (Calp) to perturb the back module. Drug concentrations were chosen byserial titration assays to be strong enough to induce a noticeable effect on primary targetactivity, yet sufficiently small to minimize cytotoxic effects (Figures S1B–D). Analysis ofthe intensity response of the drug-targeted readouts clearly showed the expected opposingeffects of these drug perturbations (Figure 1C).

Together, these experiments captured a collection of data points far too large to examine byeye. To identify circuits controlling cellular behavior, we designed an analysis approach toquantify and summarize this data. We next describe our steps for: constructing dynamicresponse curves from the time course data, constructing deviation profiles by quantifyinghow perturbed response curves varied from control, and constructing cross-talk diagramsand causal networks from the deviation profiles.

Dynamic response curvesWe first developed a robust approach for summarizing the dynamics of the polarizationresponse observed over large numbers of neutrophils. A population median was estimatedfor each replicate (independent experiment), condition (control or perturbation), marker (F-actin, p-MLC2, or microtubule), phenotype (intensity or polarity), and time point(Supplemental Information). Next, by interpolating these medians across the 11 time points,we obtained response curves for each replicate, condition, marker, and phenotype(Supplemental Information). Finally, representative response curves were obtained by takingmedians across all replicates. Together, these response curves provided temporalcharacterizations of how the intensity or polarity phenotypes of different module readoutschanged upon fMLP stimulation in control or perturbed conditions (Figure 2A and FigureS2C).

As multiple batches of primary human neutrophils had been assayed over a period greaterthan a year, a concern was whether the time-varying responses of neutrophil polarizationhad remained consistent and reproducible. To assess this variability, we first investigated thecontrol responses of neutrophils to fMLP from 20 replicates taken across all batches ofexperiments. We focused first on analyzing the F-actin intensity response and morphologicalelongation, which are classic measurements of neutrophil responses to chemoattractant. Theresulting response curves showed expected increases in actin polymerization andmorphological elongation after stimulation. Importantly, the collection of response curvesshowed remarkably consistent trends from experiment to experiment (Figure S2B). Next, foreach phenotype and marker, we examined variability among our 20 replicate controlresponse curves as measured by standard deviation (Figures S2C; light gray shaded regions)around the mean response (Figure S2C; dark gray lines). Again, we observed a high degreeof agreement among replicate response curves for each phenotype and marker. Takentogether, these results demonstrated that our experimental and computational approachproduced consistent and reproducible phenotypes from different batches of polarizingprimary human neutrophils. Our control response curves provided a baseline for subsequentinvestigation of the time-varying effects of perturbing the polarity network.

Response dynamics depends on phenotypeAs was apparent from examining the response curves, the intensity and polarity phenotypescould have dramatically different dynamic behaviors, even for the same marker. For

Ku et al.


NIH


NIH


NIH


example, the control response curves for front and back intensity returned to pre-stimulationlevels, while those for polarity did not. The intensity of microtubules continued to increaseover the duration of the 600 sec observation (consistent with classic studies (Schliwa et al.,1982), while polarity stabilized after only 60 sec (Figure 2A).

Next, we observed that some perturbations yielded different phenotypic consequences forthe intensity versus polarity of the same marker (Figure 2B). For example, Calp perturbationto the back module strongly affected the polarity but not the intensity of the front (Figure2B, left column; dark green drug perturbation curve outside gray control region).Conversely, LatA and Jas perturbations to the front module strongly affected the intensitybut not the polarity of the back (Figure 2B, right column). The difference in response curvesbetween intensity and polarity suggested that cross-talk between modules may be phenotypedependent.

Deviation profilesWe next quantified the degree to which perturbed response curves deviated from control.First, to begin summarizing our large dataset, we partitioned our 10 minute observation timeinto five time intervals: a first time interval of one minute to capture initial fast responsesand four subsequent time intervals spaced approximately 2 minutes apart to capture laterresponses (Figure 3A). Second, as a robust measure of the deviation of a phenotypicresponse, for each time interval we measured the area difference between a drug-perturbedresponse curve and its corresponding control response curves from the same experiment (i.e.96-well plate). Third, to determine statistical significance, we turned this area difference intoa dimensionless z-score by comparing the drug-induced deviation to variation amongreplicate controls (Figure S2D; Supplemental Information) (Natarajan et al., 2006; Perlmanet al., 2004). A positive or negative z-score signified an increased or decreased drugresponse (respectively). Fourth, we combined z-scores over different time intervals toconstruct a deviation “profile.” These profiles provided a compact summary of when orwhether a perturbed response curve deviated from and/or returned to the control response(Figure 3A).

We observed a spectrum of deviation dynamics, including response curves that never oralways deviated, as well as curves that deviated only at early, middle or late times fromcontrol response (Figure 3B; Supplemental Information). As expected, most deviationprofiles associated with directly targeted modules (e.g. LatA/Jas on F-actin; Figure 3B blackdots at right, and Figure 2B response curves) showed no recovery, with deviations thatlargely persisted throughout the process of polarization (9 out of 12 direct targets). On theother hand, the majority of deviation profiles that showed recovery were associated to non-directly targeted modules (15 out of 18 deviation profiles). Interestingly, the majority ofpositive deviation profiles (in red) were transient, showing deviation only at early times,while the majority of negative deviation profiles (in green) showed no recovery, though to alesser degree. We further analyzed the distribution of deviation profiles based on theirability to recover at the end of the observation time. We found that the front intensity hadthe highest recovery rate, while microtubule intensity had the least recovery rate.Interestingly, this trend was reversed for polarity, with microtubule polarity having thehighest recovery rate (Figure 3C). These results suggested that front intensity andmicrotubule polarity are the least likely to be persistently affected by perturbations to theother modules, while the situation is reversed for front polarity and microtubule intensity.

From deviation profiles to network cross-talkTo capture inter-module influence on shaping response curves, we next made use of thedeviation profiles to determine when the phenotypic response curve of a module deviated

Ku et al.


NIH


NIH


NIH


significantly from control. Network cross-talk was visualized by combining all the deviationprofiles associated with each perturbation (Figure 4). For each perturbation and timeinterval, we created a diagram in which modules were colored by their deviation z-scores.Arrows from directly to non-directly targeted modules were drawn and colored (gray scale)to indicate cross-talk if significant deviations of intensity or polarity were found(Supplemental Information). That is, a link is drawn from module 1 to module 2 at time t ifstimulation with fMLP in the presence of a perturbation to module 1 causes a change to aspecified phenotype of module 2 at time t. This approach is similar in spirit to epistasisstudies in genetics, in which phenotypes are measured in the presence or absence of a pre-existing perturbation.

The network diagrams revealed time- and phenotype-dependent cross-talk. For example, theback inhibitor Y27632 affected front polarity at early but not later times yet affectedmicrotubule intensity at later but not early times. Satisfyingly, when and where both of thepaired perturbations (i.e. LatA/Jas, Noco/Taxol or Y27632/Calp) induced significantcrosstalk, the observed deviations to non-directly targeted modules were largely consistent.For example, both microtubule drugs induced negative deviations to back polarity. Theconsistency of perturbation effects, despite using drugs of different mechanisms, supportedthe assumption that observed cross-talk was due to perturbation of the targeted module(Supplemental Information). Our network diagrams provided an intuitive, perturbation-based and module-centric visualization of how cross-talk evolved in time and varied byphenotype.

Intriguingly, for several of the cross-talk links, either activation or inhibition of one moduledecreased activation of another module. For example, both inhibition of actinpolymerization (with latrunculin A) and increased actin polymerization (with jasplakinolide)resulted in decreased back intensity. This suggested that the front is at an “optimal” level ofactivity to trigger maximal back activation, potentially because low levels of front arenecessary to stimulate back but high levels inhibit back. This is consistent with previousexperiments showing that leading edge activities are globally required for back activation(Pestonjamasp et al., 2006; Van Keymeulen et al., 2006) but also inhibit back activation(Weiner et al., 2006; Xu et al., 2003). These dual roles are analogous to the IP3 receptorcalcium channel, which is activated at low levels of calcium but inhibited by high levels ofcalcium (Bezprozvanny et al., 1991; Finch et al., 1991). For the IP3 receptor, these dualinputs play an important role in regulating the spatial and temporal dynamics of calciumwaves, and such dual interactions could serve a similar function in sculpting the back ofpolarized neutrophils.

From network cross-talk to causal networksFinally, we constructed “causal networks” (Sachs et al., 2005) to reveal cross-talk duringdifferent phases of the polarization process. For each phenotype, we combined all cross-talkdiagrams from individual perturbations to obtain causal networks for each of the 5 timeintervals. Here, we combined perturbations via maximal projection to capture all observedinfluences between modules (Supplemental Information, Figure S3). We next chose tomerge several of these 5 time intervals to align our analysis with three commonly observedphases of the polarization process: “initiation” (0–1min), during which time the corticalburst of F-actin occurs; “establishment” (1–5min), during which time front/back polarity isbeing established; and “maintenance” (5–10min), during which time polarity is largelystabilized. The resulting three causal networks were obtained by combining the causalnetworks in the appropriate time intervals via averaging to reduce noise. (Other approachesfor constructing these causal networks, such as obtaining deviation profiles by combiningreplicate data via medians or means, determining cross-talk links via different z-scorethresholds, or combining causal networks from different time intervals via an average or

Ku et al.


NIH


NIH


NIH


minimum of the z-scores, also gave similar results; Supplemental Information; Figures S3–4). The resulting causal networks revealed striking changes in how intensity and polaritycross-talk evolved over time (Figure 5A).

When examining all potential pairs of interactions among modules, we observed thatthroughout the polarization process, front-to-back cross-talk persisted for intensity but notfor polarity phenotypes. In contrast, back-to-front cross-talk persisted for polarity but not forintensity phenotypes (consistent with Figure 2B). Microtubule-to-front/back cross-talk wasrelatively static for polarity, but evolved dynamically for intensity (microtubuleperturbations affected F-actin early on, then switched to p-MLC2 during establishment, andfinally disappeared at the maintenance phase). Finally, persistent front/back-to-microtubulecrosstalk was observed for intensity (though mostly via the back module) but was absent bythe maintenance phase for polarity. We next investigated two properties of the causalnetworks: first, some modules appeared more “insulated” from cross-talk than others;second, some cross-talk links were more persistent than others.

Insulation in the causal networkThe direction of cross-talk in these causal networks and our previous observations of thedeviation profiles (Figure 3C) suggested that intensity phenotypes of the front module afterthe initiation phase, or polarity phenotypes of the microtubule module during themaintenance phase, should be insulated from (i.e. not significantly affected by) perturbationsto the other modules (Figures 5A, S4, blue squares). To test this prediction, we used double-drug perturbations to disrupt simultaneously both non-insulated modules (Figure 5B, bluesquares). We only used combinations of inhibitors/destabilizers, as their indirect effectswere more pronounced than enhancers/stabilizers (Figure 4). As predicted by the causalnetwork diagrams based on single-drug perturbation studies, front intensity was insulatedfrom indirect and combined perturbations after the initiation phase (Figure 5B, top leftpanel). For microtubules, polarity was also insulated from combined front and backperturbations during the maintenance phase (Figure 5B, bottom right panel). However, thisinsulation was also observed during the initial and establishment phases, which could be dueto synergistic effects of combined front/back disruptions. In contrast, and as would beexpected from the causal networks (Figure 5A), microtubule intensity and front polaritywere vulnerable to combined perturbations of the other modules (Figure 5B, bottom left andtop right panels). These cross-talk patterns were consistently observed for different choicesof significance thresholds on the deviation z-scores (Figure S4B). These results wereconsistent with predictions of the directionality of cross-talk in the neutrophil polaritynetwork.

Persistent cross-talkWe next identified cross-talk links in the causal networks that persisted over time (Figure5C). Surprisingly, cross-talk moved in opposite directions for intensity and polarityphenotypes, particularly during the maintenance phase. Cross-talk persistently originatedfrom the front module for intensity and from the microtubule module for polarity. Thepersistent links revealed an underlying simplicity of cross-talk influence in the network,namely a linear cascade for intensity and a feed-forward network (Milo et al., 2004) in theopposite direction for polarity (Figure 5C). These two simple, persistent motifs wereobscured within the completely connected network obtained by combining links from alltimes and phenotypes.

Ku et al.


NIH


NIH


NIH


DISCUSSIONCurrent efforts to understand signal transduction are focused on two complementaryapproaches: building extensive and comprehensive catalogues of cellular networks, andunderstanding how complex behaviors arise from these networks. A major challenge is howto connect these two approaches. Here we show that current knowledge of a network can beleveraged to design perturbation analyses that allow the dynamics of signal transduction tobe inferred directly without a complete biochemical understanding of the entire network.

While previous studies did not focus on phenotype-specific cross-talk, many of the links thatwe identified were in agreement with previously identified interactions in the neutrophilpolarity network (front to back, front to microtubules, microtubule to back, and back tofront). Our analyses also uncovered previously uncharacterized cross-talk interactions in theneutrophil polarity cascade (back to microtubules, microtubules to front) that representpotential starting points for future biochemical investigations. Our approach could beapplied in future studies to characterize an expanded set of cross-talk by including additionalmodules or subdividing the current ones.

Rapid signaling events are typically conceptualized as arising from cross-talk interactionsthat are unchanging in time. However, we found that this assumption does not hold true forthe well-studied process of neutrophil polarization. Cross-talk evolves rapidly during the600 second period after neutrophil stimulation. What potential mechanisms might underliethese changes? One possibility is that the concentration of components could be changingduring the polarization response due to synthesis/degradation, similar to the proteomicchanges during different phases of the cell cycle. Alternatively, even static biochemicalnetworks could give rise to changing information flow during different phases of theresponse due to mechanisms such as time delays or feedbacks. For example, if component Aactivates component B, and B in turn amplifies its own activity through positive feedback,then initial changes to A should affect the activity of B, but later changes may not. Finally,cross-talk among the modules may be changing due to mechanisms such as post-translational modifications. Future experiments will be necessary to discriminate amongdifferent mechanisms of modulating cross-talk.

Intriguingly, the causal network topologies differ depending on the phenotypes forneutrophils. Identifying links that were persistently present for each phenotype revealed asurprising simplicity: a linear cascade underlies the levels of signaling activities, while afeed-forward network in the opposite direction underlies the spatial distribution of signalingactivities. These phenotype-dependent paths of cross-talk could help explain neutrophils’remarkable chemotactic abilities. For efficient directional migration in the face of subtleexternal gradients, cells need a mechanism of integrating signals over time to combat thenoisy instantaneous receptor binding events that occur at any given instant of time. Onemechanism of integration is for cells to respond to not just the instantaneous receptoroccupancy, but also their recent history–for instance by having cell response biased byexisting polarity (Zigmond et al., 1981). For neutrophils, one of the primary activitiesactivated immediately downstream of receptor activation is actin assembly. Based on ourexperiments, actin regulates myosin and microtubule intensities in a linear cascade that iswell suited to read out the current external ligand distribution. To force integration overlonger time periods, the microtubule and back modules then set the spatial polarity ofleading edge activity in a feed-forward loop that is well suited to filter noise and ensure thatthe leading edge signaling responds to only the persistent distribution of myosin andmicrotubules.

Ku et al.


NIH


NIH


NIH


How cells process signals is a central question in biology (Amit et al., 2009; Muzzey et al.,2009; Pe’er and Hacohen, 2011). Simple network motifs that represent core causalinfluences can lay hidden beneath comprehensive, “everything-connects-to-everything”networks obtained by combining links from many times and phenotypes. Attempts tosuperimpose diverse experimental results onto a single topology could lead to overlycomplex network diagrams and incorrect interpretation of epistatic or feedback relations forcomplex signaling processes like cell polarization. Our approach for investigating time andphenotype-specific organization of signaling networks through perturbation analysis isgeneral and can be applied more broadly to investigate signal transduction networks beyondour current focus on leukocytes.

EXPERIMENTAL PROCEDURESDetailed methods can also be found in Supplemental Information.

Isolation of primary neutrophils from human bloodHuman neutrophils were isolated as described in (Boyum, 1968). In brief, neutrophils fromvenous blood of a single healthy donor were purified by dextran sedimentation and density-gradient centrifugation with Ficoll (GE health care, #17-5442-02). Contaminating red bloodcells were removed by hypotonic lysis.

Choice of drug perturbationsTo perturb different modules in the neutrophil polarity network, we made use of a set ofpharmacological compounds that targeted Front (jasplakinolide, latrunculin), Microtubules(taxol, nocodazole) or Back (calpeptin, Y27632) modules “positively” or “negatively”(Figure 1A). Drug concentrations were chosen by serial titration assays to be strong enoughto induce a noticeable effect on primary target activity, yet sufficiently small to minimizecytotoxic effects. The drug concentrations were chosen largely to be in the same ranges asvalues reported in the literature (Figures S1B–D; Supplemental Information). The sameindividual drug concentrations were used when we combined pairs of drugs to test theirsimultaneous effects on neutrophil polarity, i.e. front and back modules (LatA&Y27632)and microtubules and back module (Noco&Y27632).

Chemotactic assay for drug-treated cellsPurified human neutrophils were plated into 96-well Nunc glass plate (Fisher, #12-566-35),precoated with fibronectin (BD Bioscience, #354008) at a density of ~10,000 cells per well.Cells were incubated at 37°C with 5% CO2 for 20 minutes before adding drugs. Theconcentrations for each drug were as follows: 50nM for latrunculin A (LatA) (Sigma,L5163), 0.4μM for jasplakinolide (Jas) (Sigma, C5231), 9μM for nocodazole (Noco)(Sigma, M1404), 5μM for taxol (Taxol) (Sigma, T1912), 25μM for Y27632 (Y27632)(Sigma, Y0503) and 2μg/μl for calpeptin (Calp) (Cytoskeleton, CN01-A) (Figure S1C).Each drug experiment had 2 or more repeats that were performed on at least two differentdays (Noco n = 6; Taxol n = 3; LatA n = 3; Jas n = 3, Y27632 n = 3; Calp n = 2, LatA &Y27632 n = 3, Noco & Y27632 n = 2; controls n = 20). Each repeat experiment consisted oftwo technical replicates in separate wells (that were pooled together for subsequentanalysis). After incubation with drugs or control (DMSO) for 30 minutes at roomtemperature (RT), cells were simultaneously stimulated with uniform fMLP (10nM) at 37°Cbefore formaldehyde fixation at different time points ranging from 0 to 600 seconds (e.g. 0,15, 30, 45, 60, 90, 120, 180, 300, 450 and 600 sec).

Ku et al.


NIH


NIH


NIH


Immunofluorescence assayHuman neutrophils were fixed and permeabilized after fMLP stimulation. Markermultiplexing was performed as follows. The primary antibodies, anti-p-MLC2 (Cellsignaling technology, #3675) and anti-alpha tubulin (Cell signaling technology, #2144) wereadded to each well for overnight incubation at 4°C. After three washes, cells were incubatedwith secondary antibodies conjugated with Alexa 488 (Invitrogen, A11055) and Alexa 546(Invitrogen, A10040) for 2 hrs at room temperature (RT) to fluorescently label p-MLC2 andalpha tubulin, respectively. To label F-actin and DNA, cells were incubated with Alexa 647conjugated phalloidin (Invitrogen, A22287) and Hoechst 33342 (Invitrogen, H1399),respectively, for 30 minutes at RT followed by three washes.

Image acquisition for fixed cells assayAll fluorescence images were acquired using a BD Pathway 855 Bioimager (BDBiosciences) equipped with laser auto-focus system, Olympus 40x objective lens, and highresolution Hamamatsu ORCA ER CCD camera using 1×1 camera binning. Imageacquisition was controlled by AttoVision v1.5 (BD Biosciences).

Data AnalysisImage quality control—We manually inspected all fluorescence images and discardedthose presenting obvious anomalies (e.g. focus issues and abnormal fluorescence staining).Images with poorly segmented cells were re-segmented with manually optimizedsegmentation parameters.

Image preprocessing and cellular identification—Image background subtractionwas performed using the National Institute of Health ImageJ software (Rasband, 1997–2011). Identification of cellular regions and intra- and inter-plate intensity normalizationprocedures are discussed in Supplemental Information (see also (Ku et al., 2010)). The totalnumbers of cells analyzed in each of the drug conditions are given in Table S1.

Cellular feature extraction—For each segmented cellular region, we extractedmorphological elongation. Also, for each of the fluorescence readouts (i.e. F-actin forfrontness, microtubules, and p-MLC2 for backness), we extracted: 1) the averagefluorescence intensity (total intensity in the cellular region divided by the cell area) as anintegrated readout of the signaling activity of the associated signaling module; and 2) thespatial polarity phenotype, known as “compactness”, to capture the degree to which thefluorescent readout was spatially concentrated within the cell. Both steps are described inSupplemental Information.

Establishment of phenotypic response curves—For each phenotype, we establishedthe response curve based on cubic interpolation of the population median values collected atdifferent fixation time points. The response curves were smoothed using the functionfastsmooth.m downloaded at Matlab Central (http://www.mathworks.com/matlabcentral/fileexchange/19998-fast-smoothing-function). We estimated the variation of the controlresponse curve by the standard deviation among the control median response curves acrossreplicates experiments (Figure S2C, light gray shaded region; Figure S2D, top row). For thedrug treated conditions, variation of the median response curve at each fixation time pointwas estimated by the standard error of the mean among replicate response curves (FigureS2C, second to fourth rows; Figure S2D, bottom).

Quantification of drug-induced phenotypic deviation over time—To estimate thedrug-induced deviation of a phenotypic response curve during polarization, for each

Ku et al.


NIH


NIH


NIH


http://www.mathworks.com/matlabcentral/fileexchange/19998-fast-smoothing-function

http://www.mathworks.com/matlabcentral/fileexchange/19998-fast-smoothing-function

phenotype f, drug response d, and time interval T (E, EI, I, IL, or L), dimensionless z-scoreswere computed as given in Supplemental Information. The vector of z-scores, z(f,d),describing temporal evolution of the deviation of response curve was termed a “deviationprofile”:

Categorization of deviation profiles—To categorize deviation profiles with differentpatterns, we binarized their entries by threshold operation. The magnitude of the deviation z-scores were compared against a numerical threshold τ (τ = 1 in our study) and mapped to 0(1) if their absolute values were smaller (larger) than τ, respectively. Upon transformation ofthe deviation z-scores, we categorized the deviation profiles based on their dynamic patternsand further grouped them according to their ability to return to the control level at the end ofthe polarization process (Figure 3B; Supplemental Information).

Visualization of network cross-talk—For a given drug perturbation d and phenotype f,we considered that a cross-talk link existed from the directly targeted module m1 to a non-directly targeted module m2 during the time interval t if the deviation z-score zt

(f,d) ofmodule m2 exceeded a specified threshold τ (i.e. |zt

(f,d)|>τ) (Figure 4).

Construction of causal networks—We combined cross-talk observed across multipledrug perturbations using maximum projection: that is, the maximum amplitude of cross-talkbetween any pair of signaling modules was kept as the final cross-talk strength (Figure S3;Supplemental Information). To merge cross-talk diagrams associated to different timeintervals, we computed average cross-talk strength (Figures 5A–B; Figure S4; SupplementalInformation).

Supplementary MaterialRefer to Web version on PubMed Central for supplementary material.

AcknowledgmentsWe thank Neal Alto, Al Gilman, Sasha Jilkine, David Mangelsdorf, Benjamin Pavie, Rama Ranganathan, MikeRosen, and all other members of the Altschuler, Weiner, and Wu labs. ODW was supported by NIH GM084040and the Searle Scholars Foundation. LFW and SJA were supported by the Endowed Scholars program at UTSWMedical Center. LFW was supported by NIH GM081549 and the Welch Foundation I-1619. SJA was supported byNIH GM071794, the Welch Foundation I-1644 (S.J.A.), and the Rita Allen Foundation.

ReferencesAmit I, Garber M, Chevrier N, Leite AP, Donner Y, Eisenhaure T, Guttman M, Grenier JK, Li W, Zuk

O, et al. Unbiased reconstruction of a mammalian transcriptional network mediating pathogenresponses. Science. 2009; 326:257–263. [PubMed: 19729616]

Bezprozvanny I, Watras J, Ehrlich BE. Bell-shaped calcium-response curves of Ins(1,4,5)P3- andcalcium-gated channels from endoplasmic reticulum of cerebellum. Nature. 1991; 351:751–754.[PubMed: 1648178]

Boyum A. Isolation of mononuclear cells and granulocytes from human blood. Isolation of monuclearcells by one centrifugation, and of granulocytes by combining centrifugation and sedimentation at 1g. Scand J Clin Lab Invest Suppl. 1968; 97:77–89. [PubMed: 4179068]

Cheong R, Rhee A, Wang CJ, Nemenman I, Levchenko A. Information transduction capacity of noisybiochemical signaling networks. Science. 2011; 334:354–358. [PubMed: 21921160]

Ku et al.


NIH


NIH


NIH


de Lichtenberg U, Jensen LJ, Brunak S, Bork P. Dynamic complex formation during the yeast cellcycle. Science. 2005; 307:724–727. [PubMed: 15692050]

Eddy RJ, Pierini LM, Maxfield FR. Microtubule asymmetry during neutrophil polarization andmigration. Molecular Biology of the Cell. 2002; 13:4470–4483. [PubMed: 12475966]

Finch EA, Turner TJ, Goldin SM. Calcium as a coagonist of inositol 1,4,5-trisphosphate-inducedcalcium release. Science. 1991; 252:443–446. [PubMed: 2017683]

Fraser ID, Germain RN. Navigating the network: signaling cross-talk in hematopoietic cells. NatImmunol. 2009; 10:327–331. [PubMed: 19295628]

Janes KA, Gaudet S, Albeck JG, Nielsen UB, Lauffenburger DA, Sorger PK. The response of humanepithelial cells to TNF involves an inducible autocrine cascade. Cell. 2006; 124:1225–1239.[PubMed: 16564013]

Kalir S, Alon U. Using a quantitative blueprint to reprogram the dynamics of the flagella genenetwork. Cell. 2004; 117:713–720. [PubMed: 15186773]

Kentner D, Sourjik V. Dynamic map of protein interactions in the Escherichia coli chemotaxispathway. Mol Syst Biol. 2009; 5:238. [PubMed: 19156130]

Krendel M, Zenke FT, Bokoch GM. Nucleotide exchange factor GEF-H1 mediates cross-talk betweenmicrotubules and the actin cytoskeleton. Nature Cell Biology. 2002; 4:294–301.

Ku CJ, Wang Y, Pavie B, Altschuler SJ, Wu LF. On Identifying Information from Image-BasedSpatial Polarity Phenotypes in Neutrophils. Proc IEEE Int Symp Biomed Imaging. 2010; 14–17:1029–1032.

Lee E, Salic A, Kruger R, Heinrich R, Kirschner MW. The roles of APC and Axin derived fromexperimental and theoretical analysis of the Wnt pathway. PLoS Biol. 2003; 1:E10. [PubMed:14551908]

Levine H, Kessler DA, Rappel WJ. Directional sensing in eukaryotic chemotaxis: a balancedinactivation model. Proc Natl Acad Sci U S A. 2006; 103:9761–9766. [PubMed: 16782813]

Loose M, Swiers G, Patient R. Transcriptional networks regulating hematopoietic cell fate decisions.Curr Opin Hematol. 2007; 14:307–314. [PubMed: 17534154]

Luscombe NM, Babu MM, Yu H, Snyder M, Teichmann SA, Gerstein M. Genomic analysis ofregulatory network dynamics reveals large topological changes. Nature. 2004; 431:308–312.[PubMed: 15372033]

Marco E, Wedlich-Soldner R, Li R, Altschuler SJ, Wu LF. Endocytosis optimizes the dynamiclocalization of membrane proteins that regulate cortical polarity. Cell. 2007; 129:411–422.[PubMed: 17448998]

Milo R, Itzkovitz S, Kashtan N, Levitt R, Shen-Orr S, Ayzenshtat I, Sheffer M, Alon U. Superfamiliesof evolved and designed networks. Science. 2004; 303:1538–1542. [PubMed: 15001784]

Muzzey D, Gomez-Uribe CA, Mettetal JT, van Oudenaarden A. A systems-level analysis of perfectadaptation in yeast osmoregulation. Cell. 2009; 138:160–171. [PubMed: 19596242]

Natarajan M, Lin KM, Hsueh RC, Sternweis PC, Ranganathan R. A global analysis of cross-talk in amammalian cellular signalling network. Nat Cell Biol. 2006; 8:571–580. [PubMed: 16699502]

Niggli V. Microtubule-disruption-induced and chemotactic-peptide-induced migration of humanneutrophils: implications for differential sets of signalling pathways. J Cell Sci. 2003; 116:813–822. [PubMed: 12571279]

Odell GM, Foe VE. An agent-based model contrasts opposite effects of dynamic and stablemicrotubules on cleavage furrow positioning. J Cell Biol. 2008; 183:471–483. [PubMed:18955556]

Oleksiuk O, Jakovljevic V, Vladimirov N, Carvalho R, Paster E, Ryu WS, Meir Y, Wingreen NS,Kollmann M, Sourjik V. Thermal robustness of signaling in bacterial chemotaxis. Cell. 2011;145:312–321. [PubMed: 21496648]

Pe’er D, Hacohen N. Principles and strategies for developing network models in cancer. Cell. 2011;144:864–873. [PubMed: 21414479]

Perlman ZE, Slack MD, Feng Y, Mitchison TJ, Wu LF, Altschuler SJ. Multidimensional drug profilingby automated microscopy. Science. 2004; 306:1194–1198. [PubMed: 15539606]

Ku et al.


NIH


NIH


NIH


Pestonjamasp KN, Forster C, Sun CX, Gardiner EM, Bohl B, Weiner O, Bokoch GM, Glogauer M.Rac1 links leading edge and uropod events through Rho and myosin activation during chemotaxis.Blood. 2006; 108:2814–2820. [PubMed: 16809619]

Rasband, WS. ImageJ. National Institutes of Health; Bethesda,, and Maryland: 1997–2011.

Sachs K, Perez O, Pe’er D, Lauffenburger DA, Nolan GP. Causal protein-signaling networks derivedfrom multiparameter single-cell data. Science. 2005; 308:523–529. [PubMed: 15845847]

Schliwa M, Pryzwansky KB, Euteneuer U. Centrosome splitting in neutrophils: an unusualphenomenon related to cell activation and motility. Cell. 1982; 31:705–717. [PubMed: 7159931]

Schneider IC, Haugh JM. Mechanisms of gradient sensing and chemotaxis - Conserved pathways,diverse regulation. Cell Cycle. 2006; 5:1130–1134. [PubMed: 16760661]

Servant G, Weiner OD, Herzmark P, Balla T, Sedat JW, Bourne HR. Polarization of chemoattractantreceptor signaling during neutrophil chemotaxis. Science. 2000; 287:1037–1040. [PubMed:10669415]

Small JV, Geiger B, Kaverina I, Bershadsky A. How do microtubules guide migrating cells? Nat RevMol Cell Biol. 2002; 3:957–964. [PubMed: 12461561]

Srinivasan S, Wang F, Glavas S, Ott A, Hofmann F, Aktories K, Kalman D, Bourne HR. Rac andCdc42 play distinct roles in regulating PI(3,4,5)P3 and polarity during neutrophil chemotaxis. JCell Biol. 2003; 160:375–385. [PubMed: 12551955]

Tkachenko E, Sabouri-Ghomi M, Pertz O, Kim C, Gutierrez E, Machacek M, Groisman A, Danuser G,Ginsberg MH. Protein kinase A governs a RhoA-RhoGDI protrusion-retraction pacemaker inmigrating cells. Nat Cell Biol. 2011; 13:660–667. [PubMed: 21572420]

Van Keymeulen A, Wong K, Knight ZA, Govaerts C, Hahn KM, Shokat KM, Bourne HR. To stabilizeneutrophil polarity, PIP3 and Cdc42 augment RhoA activity at the back as well as signals at thefront. Journal of Cell Biology. 2006; 174:437–445. [PubMed: 16864657]

Weiner OD, Marganski WA, Wu LF, Altschuler SJ, Kirschner MW. An actin-based wave generatororganizes cell motility. PLoS Biol. 2007; 5:e221. [PubMed: 17696648]

Weiner OD, Neilsen PO, Prestwich GD, Kirschner MW, Cantley LC, Bourne HR. A PtdInsP(3)- andRho GTPase-mediated positive feedback loop regulates neutrophil polarity. Nat Cell Biol. 2002;4:509–513. [PubMed: 12080346]

Weiner OD, Rentel MC, Ott A, Brown GE, Jedrychowski M, Yaffe MB, Gygi SP, Cantley LC, BourneHR, Kirschner MW. Hem-1 complexes are essential for Rac activation, actin polymerization, andmyosin regulation during neutrophil chemotaxis. PLoS Biol. 2006; 4:e38. [PubMed: 16417406]

Wong K, Pertz O, Hahn K, Bourne H. Neutrophil polarization: spatiotemporal dynamics of RhoAactivity support a self-organizing mechanism. Proc Natl Acad Sci U S A. 2006; 103:3639–3644.[PubMed: 16537448]

Wong K, Van Keymeulen A, Bourne HR. PDZRhoGEF and myosin II localize RhoA activity to theback of polarizing neutrophil-like cells. J Cell Biol. 2007; 179:1141–1148. [PubMed: 18086913]

Xu JS, Wang F, Van Keymeulen A, Herzmark P, Straight A, Kelly K, Takuwa Y, Sugimoto N,Mitchison T, Bourne HR. Divergent signals and cytoskeletal assemblies regulate self-organizingpolarity in neutrophils. Cell. 2003; 114:201–214. [PubMed: 12887922]

Xu JS, Wang F, Van Keymeulen A, Rentel M, Bourne HR. Neutrophil microtubules suppress polarityand enhance directional migration. Proceedings of the National Academy of Sciences of theUnited States of America. 2005; 102:6884–6889. [PubMed: 15860582]

Yu RC, Pesce CG, Colman-Lerner A, Lok L, Pincus D, Serra E, Holl M, Benjamin K, Gordon A,Brent R. Negative feedback that improves information transmission in yeast signalling. Nature.2008; 456:755–761. [PubMed: 19079053]

Zigmond SH, Levitsky HI, Kreel BJ. Cell polarity: an examination of its behavioral expression and itsconsequences for polymorphonuclear leukocyte chemotaxis. J Cell Biol. 1981; 89:585–592.[PubMed: 7251666]

Zigmond SH, Sullivan SJ. Sensory adaptation of leukocytes to chemotactic peptides. J Cell Biol. 1979;82:517–527. [PubMed: 479314]

Ku et al.


NIH


NIH


NIH


Figure 1. Pharmacological perturbations of key modules in the human neutrophil polaritynetworkA. Simplified schema of the polarity signaling network in primary human neutrophilscontaining the front (F), back (B), and microtubule (M) modules. Two opposing drugperturbations were chosen to disrupt each of the three modules: latrunculin A (LatA) andjasplakinolide (Jas) for the front; nocodazole (Noco) and taxol (Taxol) for the microtubules;and Y27632 (Y27632) and calpeptin (Calp) for the back. B. Cells were treated for 30minutes with drugs, stimulated with 10nM fMLP, and fixed at multiple time points. Shownare representative images of human neutrophils at different time points after fMLPstimulation. Color: red (F-actin), blue (microtubules), and green (p-MLC2). Scale bar:10μm. C. Quantification of opposing drug effects on their targeted modules. Bar graphsshow fold-change of overall response to drug perturbation (mean ± standard error acrossreplicate experiments; Supplemental Information).See Figure S1.

Ku et al.


NIH


NIH


NIH


Figure 2. Dynamic responses of neutrophils to drug perturbationsA. Neutrophil response curves for front (left), back (middle), and microtubules (right)intensity (top row) and polarity (bottom row) phenotypes. Dark gray curve: mean of medianresponse across replicates (n = 20). Gray band: one standard deviation above and below themean values. B. Response of front module to perturbations of the back (left) or response ofback module to perturbations of the front (right) for intensity (top) and polarity (bottom)phenotypes. Color curves: mean of population median responses (red/brown: frontperturbations; bright/dark green: back perturbations). Error bars: standard error of the meandrug responses at different time points.See Figure S2.

Ku et al.


NIH


NIH


NIH


Figure 3. Deviation profiles summarized dynamics of perturbation effectsA. Illustration of how deviation profiles were generated by quantifying drug-induceddeviations to control response curves at different time intervals (0–1min; 1–3min; 3–5min;5–7.5min; 7.5–10min) (Supplemental Information). Color map: red/white/green(significantly increased/unchanged/decreased feature value). B. Deviation profiles acrossphenotypes and perturbations (heatmap) showed various temporal patterns: deviation at notime, early time, middle time, late time, or all times (cartoon illustrations at left;Supplemental Information). Deviation profiles were further grouped depending on whetherthe deviation disappeared by (i.e. recovery) or persisted through (i.e. no recovery) the end ofthe polarization process. Numbers next to different labels represented the number ofdeviation profiles exhibiting the corresponding patterns. C. Distribution of deviation profilesof front, microtubule, and back modules based on their ability to recover (recovery: white;non recovery: black). Top: intensity phenotype. Bottom: polarity phenotype.

Ku et al.


NIH


NIH


NIH


Figure 4. Dynamic cross-talk in perturbed neutrophil polarity signaling networks“Cross-talk diagrams” were constructed by analyzing deviations to each of the signalingmodules (F: front module; M: microtubules; B: back module) for each drug perturbation,phenotype, and time interval. Color map for the signaling modules: red (increased intensity/polarity), white (unchanged), green (decreased intensity/polarity). Diamond arrows: drugperturbation to the targeted module. Square arrows: cross-talk from directly to non-directlytargeted modules. Square arrows were colored to reflect the strength of drug effects (bright/dark gray: weak/strong deviation). Left column: intensity phenotype. Right column: polarityphenotype. Cross-talk links with small deviations (|z-score| < 1) are not shown.See Figure S3.

Ku et al.


NIH


NIH


NIH


Figure 5. Causal networks revealed dynamic, phenotype-dependent cross-talk with persistentinteractions arranged in a simple configurationA. Causal networks across different stages of polarization (initiation, establishment, andmaintenance) were constructed by combining cross-talk diagrams across perturbations andtime intervals (Supplemental Information). Square arrows: cross-talk from directly to non-directly targeted modules. As in Figure 4, cross-talk with small deviations (|z-score| < 1) arenot shown. Blue square: insulated module. B. Cross-talk diagrams obtained from doubledrug perturbation assays. Double dashed lines indicate both direct drug targets. Top row:Noco & Y27632 and bottom row: LatA & Y27632. Arrows and color map are as in Figure4. C. Summary of all cross-talk links in the causal networks for intensity (left) and polarity(right) phenotypes. Solid/dotted link: persistent/transient cross-talk. Network motifsobtained from persistent cross-talk between modules are shown to the right of thesummarized cross-talk diagrams.See Figure S4.

Ku et al.


NIH


NIH


NIH


network crosstalk dynamically changes during neutrophil polarization

Documents