s Muinonen-Martin, A. J. et al. (2014) Melanoma cells break down LPA to establish local gradients that drive chemotactic dispersal. PLoS Biology, 12 (10). e1001966. ISSN 1544-9173 Copyright © 2014 The Authors http://eprints.gla.ac.uk/101796 Deposited on: 27 January 2015 Enlighten – Research publications by members of the University of Glasgow http://eprints.gla.ac.uk Melanoma Cells Break Down LPA to Establish Local Gradients That Drive Chemotactic Dispersal AndrewJ.Muinonen-Martin1,2,3,OliviaSusanto1,QifengZhang4,ElizabethSmethurst4,WilliamJ.Faller1, Douwe M. Veltman1, Gabriela Kalna1, Colin Lindsay1,5, Dorothy C. Bennett6, Owen J. Sansom1, Robert Herd7, Robert Jones1,5, Laura M. Machesky1, Michael J. O. Wakelam4, David A. Knecht8, Robert H. Insall1* 1CRUKBeatsonInstitute,Glasgow,UnitedKingdom,2YorkTeachingHospitalNHSFoundationTrust,York,UnitedKingdom,3TheLeedsTeachingHospitalsNHSTrust, Leeds,UnitedKingdom,4TheBabrahamInstitute,Cambridge,UnitedKingdom,5BeatsonWestofScotlandCancerCentre,Glasgow,UnitedKingdom,6MolecularCell SciencesResearchCentre,St.George’s,UniversityofLondon,London,UnitedKingdom,7AlanLyellCentreforDermatology,Glasgow,UnitedKingdom,8Departmentof MolecularandCellBiology,UniversityofConnecticut,Storrs,Connecticut,UnitedStatesofAmerica Abstract Thehighmortalityofmelanomaiscausedbyrapidspreadofcancercells,whichoccursunusuallyearlyintumourevolution. Unlike most solid tumours, thickness rather than cytological markers or differentiation is the best guide to metastatic potential.Multiplestimulithatdrivemelanomacellmigrationhavebeendescribed,butitisnotclearwhichareresponsible forinvasion,norifchemotacticgradientsexistinrealtumours.Inachamber-basedassayformelanomadispersal,wefind that cells migrate efficiently away from one another, even in initially homogeneous medium. This dispersal is driven by positivechemotaxisratherthanchemorepulsionorcontactinhibition.Theprincipalchemoattractant,unexpectedlyactive across all tumour stages, is the lipid agonist lysophosphatidic acid (LPA) acting through the LPA receptor LPAR1. LPA induceschemotaxisofremarkableaccuracy,andisbothnecessaryandsufficientforchemotaxisandinvasionin2-Dand3-D assays. Growth factors, often described as tumour attractants, cause negligible chemotaxis themselves, but potentiate chemotaxis to LPA. Cells rapidly break down LPA present at substantial levels in culture medium and normal skin to generate outward-facing gradients. We measure LPA gradients across the margins of melanomas in vivo, confirming the physiological importance of our results. We conclude that LPA chemotaxis provides a strong drive for melanoma cells to invadeoutwards.Cellscreatetheirowngradientsbyactingasasink,breakingdownlocallypresentLPA,andthusforminga gradientthatislowinthetumourandhighinthesurrounding areas.Thekeystepisnotacquisitionofsensitivitytothe chemoattractant, but rather the tumour growing to break down enough LPA to form a gradient. Thus the stimulus that drivescelldispersal is notthe presenceofLPAitself, butthe self-generated,outward-directed gradient. Citation:Muinonen-MartinAJ,SusantoO,ZhangQ,SmethurstE,FallerWJ,etal.(2014)MelanomaCellsBreakDownLPAtoEstablishLocalGradientsThatDrive ChemotacticDispersal.PLoSBiol12(10):e1001966.doi:10.1371/journal.pbio.1001966 AcademicEditor:AndreLevchenko,YaleUniversity,UnitedStatesofAmerica ReceivedFebruary24,2014;AcceptedSeptember5,2014;PublishedOctober14,2014 Copyright:(cid:2)2014Muinonen-Martinetal.Thisisanopen-accessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense,which permitsunrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalauthorandsourcearecredited. Funding:ThisresearchwasfundedbyCancerResearchUKcoregrantstoRI,LM,andOS;bytheWellcomeTrust(fellowship095186/Z/10/ZtoAMMand programmegranttoDB);andbytheBBSRC(coregranttoMJOW).Thefundershadnoroleinstudydesign,datacollectionandanalysis,decisiontopublish,or preparationofthemanuscript. CompetingInterests:Theauthorshavedeclaredthatnocompetinginterestsexist. Abbreviations:EGF,epidermalgrowthfactor;FBS,fetalbovineserum;LPA,lysophosphatidicacid;PDGF,platelet-derivedgrowthfactor;RGP,radialgrowth phase;SCF,stemcellfactor;SEM,standarderrorofthemean;VGP,verticalgrowthphase. *Email:[email protected] Introduction size alone. Metastasis is therefore an important, and under- medicated, potential target forcancer therapy[4,5]. Melanoma Metastasis Melanoma is an unusually aggressive cancer, which often Melanocyte Migration during Development metastasizesearlyduringtumourdevelopment[1].Tumoursthat One principal reason behind the aggressiveness of melanoma have not clinically metastasized are frequently curable, but derives from the developmental history of melanocytes, the patients arefarlesslikelytosurvive iftumourshavemetastasized pigment producing cells in the skin that mutate to form beforetheyaresurgicallyremoved,andmetastasisistheprincipal melanomas. During mammalian development melanoblasts, the cause of cancer mortality [2]. The most influential prognostic melanocyte precursors, emerge from a restricted location at the factor in predicting metastasis and survival is the thickness of the neural crest, and migrate rapidly from there throughout the tumour(the‘‘Breslowdepth’’)[3].Thereisadramaticincreasein developing dermis, before maturing into melanocytes on the the risk of metastasis with only millimeter increases in Breslow basementmembraneoftheepidermis[6].Thusasubstantiallevel depth [3]. This characteristic is unlike most solid tumours, in of cell migration is required for even skin pigmentation. Even in which the cytological morphology of the tumour cells and the adults—forexamplefollowingtreatmentforvitiligo—melanocytes individual genes mutated in the cancer are more important than can spread significant distances from the hair follicles to PLOSBiology | www.plosbiology.org 1 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal [14]).Potentialattractantsformextremelysteepandrathershort- Author Summary lived concentration gradients, unlike the physiological conditions Melanoma is feared because it spreads very rapidly when theassayaimstoreproduce.Moreseriouslystill,conditionseither tumours are relatively small. It is not known why this side of the filter may be discretely different; cells may grow, metastasisissoefficientandaggressive.Inparticular, itis survive, or adhere better on one side of the filter than the other, notknownwhatdrivesmelanomacellstostarttomigrate giving changes in the numbers of cells that can be artifactually outfromthetumour.Here,wehavestudiedthechemical interpreted as chemotaxis. Direct viewing chambers, such as signals that guide the migration of melanoma cells. We Dunn,Zigmond,orInsallchambers,aremorelaborioustousebut find that a component of serum, lysophosphatidic acid yielda farhigherquality ofdata, with fewerartifacts[15–17]. In (LPA),functionsasaremarkablystrongattractantforallof work described here, we use direct-viewing chambers to identify the melanoma cells that we examined. We also observe lysophosphatidicacid(LPA)asafarmorepotentchemoattractant thatmelanomacellsrapidlybreakdownLPA.Weconclude formelanomacellsthanotherpreviouslydescribedattractants.We that melanomas create their own gradients of LPA, with have developed and refined two direct-viewing assays to assess low LPA in the tumour and high LPA outside. Since mechanisms of cell dispersal and chemotaxis, allowing us to melanoma cells are attracted by LPA, this LPA gradient distinguish chemotactic from chemokinetic and contact-driven around the melanomas serves as a signal that drives the tumour cells out into the surrounding skin and blood responses under defined conditions that minimize artifacts. vessels. Finally, we show that such gradients exist in a Furthermore, the use of direct-viewing chambers makes compar- mousemodelofmelanoma.Self-generatedLPAgradients isonof attractants’ relative efficienciespractical. are therefore an intriguing new driver for melanoma dispersal. The Source of Attractant Gradients In Vivo The suggested role of chemoattractants in cancer dispersal— whether growth factors, chemokines, or LPA—raises the crucial repopulate the surrounding skin. The melanocyte lineage is thus question of how gradients are generated. Chemotaxis will only inherently migratory. work with signals that are presented as gradients—homogeneous However, several questions about melanoma progression signals contain no directional information—and the steeper the remainunanswered.Thefirstiswhatdrivesmelanomastochange gradient, the more efficient the chemotaxis. Chemical gradients from the relatively benign radial growth phase (RGP) to the far are typically effective over distances of less than a millimetre— moreinvasiveverticalgrowthphase(VGP)(seeschematicdiagram limits on the efficiency of diffusion make larger gradients inFigure 1A).In RGP melanomas, cellsonly spread horizontally impractical [18]. Thus for a gradient to be formed there must along the basement membrane, compared to VGP melanoma bea gradient sourcethat isclosetothetumour. cells, which are also capable of spreading both upwards into the Alternatively, local gradients may be formed from signals that epidermis(Pagetoidspread)anddownwards,intoandthroughthe are widely produced, but are absorbed or broken down locally. dermis(invasion).Thisspreadraisestherelatedquestion,ofwhat This local depletion mechanism is potentially just as effective as drives cells to migrate away from the primary tumour. Simple, local production, but less often invoked. In the cancer literature, random migration is an extremely inefficient way of dispersing onlylocalisedsourcesaretypicallyinvoked,forexampleindividual cellsandalsounlikelytodrivecellstoinvadethroughmatrixand macrophageswithinthevasculatureattracting cancercellswithin basement membranes. Chemotaxis—cell migration directed by thetumour [10]. gradients of soluble signalling molecules—is implicated as an Ifcellsthatarerespondingtoastimulusarealsoresponsiblefor importantdriverofmetastasisbyawiderangeofdata[7,8],andis breaking it down, the result is a self-generated gradient. Under considered necessary to drive efficient invasion. In breast cancer, these conditions the gradient is always oriented away from the forexample,sometumourcellsmigratetowardsepidermalgrowth current location of the cells. One such example has been shown factor(EGF)[9].However,EGFgradientshaveonlybeeninferred during the development of the zebrafish lateral line primordium invivo,nevermeasured,andtheirsourcesareusuallyunclear.In [19–21], in which a dummy receptor locally absorbs an SDF-1 the case of breast cancer, the EGF is thought to be secreted by stimulus to set up a gradient that is detected by a different macrophagesrecruitedinaparacrineloopbythetumour[10],but receptor. In this work we find that melanoma cells self-generate for other attractants and cell types the sources of chemotactic chemotactic gradients from unlocalised, exogenous LPA. These signals are not known. gradients tend to direct cells to disperse outwards from tumours, In the melanoma literature, most chemotaxis is attributed to thusdirectlypromotingmetastasis.Furthermore,wemeasureLPA growthfactorssuchasplatelet-derivedgrowthfactor(PDGF)and gradients across real melanomas in vivo. Since melanomas of EGF [11] and the CXCR4 ligand SDF-1 [12], though a wide sufficientsizebothgeneratetheirownLPAgradientsandrespond varietyofpotentialattractantshavebeendiscussed[13].Gradients to them, chemotaxis-steered spread of melanomas is almost of growth factor or SDF-1 have not been identified in vivo, they inevitable. can only be inferred from the cells’ behaviour or pattern of responses invitro. Results Chemotaxis and Invasion Assays Density-Dependent Outward Migration of Tumour Cell Chemotaxis assays are typically performed in transwell cham- Toexaminethesignalsthatdrivethespreadofmelanomacells, bers,inwhichcellsaregrownononesideofamembranefilterand wesetup2-Dassaysfortumourcellspreadusingadirect-viewing potential attractants are added to the other side. Chemotaxis is chemotaxischamberthatallowsdetailedanalysisofcellmigration assayedbythenumberofcellsobservedonthefarsideofthefilter [15].Thechambercontainstwowells,connectedbyabridgethat after a fixed interval. These assays are subject to a wide range of allows diffusion of attractants but not flow. Both cells were artifacts. Cells’ behaviour during chemotaxis cannot be studied, homogeneously filled with complete medium, but cultured which makes it extremely difficult todistinguish chemotaxis from melanoma cells [22] were only seeded in one well, at a range of directionless changes in migratory behaviour (i.e., chemokinesis different densities. PLOSBiology | www.plosbiology.org 2 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal PLOSBiology | www.plosbiology.org 3 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal Figure1.Density-dependentdispersalofmelanomacells.(A)Schematicshowingthestagesofmelanomaspread.(B)WM239Ametastatic melanomacellsdispersinginuniformmedium.26104cellswereintroducedintoonereservoirofanInsallchambercontainingcompletemedium with10%FBSthroughout,andobservedbytime-lapsephasecontrastmicroscopy.SeeMovieS1.Theleftsideofeachimageshowsthereservoir containingcells,whiletherightsideistheviewingbridgeofthechamber.(C–D)Migrationisdensity-dependent.WM1158metastaticmelanomacells wereseededatdifferentdensitiesinfullmediumwith10%FBS,andobservedasbefore.At26104cells/wellandabove,peakmigrationdistances increasesharply,asconfirmedbythedistanceat17hours(D;graphshowsmean6SEM).(E)Migrationisnotdrivenbyproductionofarepellent. 26104WM1158cellswereintroducedintoachamberinminimalmediumwithoutserumandobservedat17hoursasbefore.Cellssurviveand adhere,butdonotdisperse.(F)Migrationisnotdrivenbyproductionofaserum-derivedrepellent.26104WM1158cellswereintroducedintoa chamberinminimalmediumwithoutserumandobservedat17hoursasbefore.Cellsdisperselessefficientlyinconditionedmediumthaninfresh medium.(G)Migrationmediatedbychemotaxisupaserumgradientissimilartodensity-inducedmigration.Leftpanel:26104WM1158cellswere introducedintoachamberinthepresenceofagradientfrom0%FBSaroundthecellsto10%intheoppositereservoir[15].Thecellsrapidlymigrate towardsthewellcontainingserum.Rightpanel:similarassaywith10%seruminbothreservoirs.PanelstakenfromMoviesS3andS1,respectively. doi:10.1371/journal.pbio.1001966.g001 Our initial results were surprising: Cells consistently spread create a gradient that consistently directs migration away from outwards from the well in which they started, even in uniform themselves. We considered that homogeneous attractants would mediumwithoutanexternallyappliedgradient(Figure 1B;Movie mostlikelyderivefromtheserumaddedtofullmedium.Tofindif S1). This effect was density-dependent; cells plated at 26103 or dispersalusedarepellentoraconsumedattractant,wecompared 66103cells/wellbarelymigrated,while26104cells/wellmigrated celldispersalinserum-freeandnormalmedium.Cellsinserum-free up to 350mm in 24hours (Figure 1C and 1D). This behaviour medium are healthy and motile in control movies, but do not strikinglyresemblesthebehaviourofrealmelanomas,inwhichthe migrateawayfromoneanother(Figure1E),demonstratingthatthe chanceofmetastasisismorecorrelatedwithtumourthicknessthan cells do not secrete chemorepellents. We also compared cells any otherparameter [3]. movingoutoffreshmediumintoserum-freeandfullmedium.Cells This type of density-dependent spreading requires individual dispersed far more efficiently into the rich medium (Figure1F), cells (or small clusters of cells) to migrate away from the bulk implyingthattheyaredrivenbyattractantsinfreshmediumrather population. This dispersal occurred in our assays; cells moved thananinhibitorwhoseproductiondependsonserum. directly away from the well they resided in with unprecedented To test whether consumption of a component of serum accuracy (Movie S1). This directed, non-random migration can producesapositivechemotaxisresponse,wecomparedmigration onlyoccurifthemovingcellsperceiveadirectionalcuefromthe in uniform serum to an assay in which cells are exposed to a bulk populationof thecells to spread. We therefore analyzed the gradientbetweenserum-freemediumandmediumsupplemented nature of the signal that was directing cells away from the bulk with10%serum(MovieS3).Wefoundthatbothassaysproduced population.Themostprobablesignallingmechanismsarecontact similar directed migratory responses; cells migrated towards the inhibition of migration [23] or chemotaxis. We therefore opposite well with or without a preformed serum gradient examined these potential mechanismsinturn. (Figure 1G). This finding further supports the concept that the Contact inhibition (of migration, as opposed to the more outward migration is driven by positive chemotaxis, most likely frequently described contact inhibition of growth) is an effective towards a chemoattractant globally present in the serum but mechanism for short-range dispersal in which cell:cell contact depleted around thecells. directs cells away from one another. It has been shown in other We tested this hypothesis using a more traditional chemotaxis neural crest-derived cell types [24]. However we found no assay,inwhichcellsarespreadhomogeneouslyoverthefieldatthe evidence to suggest it drives cell dispersal in our assays. Movie start of the assay, giving the cells the opportunity to move in any S2showsoneexampleinwhichcellsspreadbothindividuallyand direction [14]. We loaded cells into the chamber in complete whilecontactingoneanother.Somecellssteeraccuratelyoutwards mediumthathadbeenconditionedbymelanomacellsfor48hours, through multiple cycles of new pseudopods independently of thenreplacedthemediuminonewellwithfreshmediumcontaining cell:cell contact. Others continue to migrate outwards when 10% serum. The cells migrated towards the well containing fresh contactingthecellinfront,wherecontactinhibitionpredictsthese medium very efficiently (Figure2A and 2B), showing that an cells should reverse into the space behind them. Analysis of the attractantinfreshmediumisconsumedbythemelanomacells. paths of individual cells (Figure S1) shows that cell-cell contact is We confirmed that chemoattractants are present in normal notsteeringcells;thepathsofcellsthatarecontactingothers,have serumbyexposingmelanomacells—againhomogeneouslyseeded recently contacted others, and are not in contact are strikingly inthechemotaxischamber—toexogenousgradientsofserum.In similar. The one apparent example of contact inhibition (Movie homogeneous serum-free medium the cells were healthy, and S2, cell 2) changed the cell’s direction but did not improve its migrated, but randomly (Figure 2C). When a gradient of serum outwardaccuracy.Thuswhilethesecellsmayexperiencecontact wasapplied,thecellsmigratedtowardsthehigherconcentrations inhibition,weconsideredchemotaxisasthemostlikelymechanism with unprecedented precision (Figure2D); their paths are steering themaway fromthemainpopulation. overwhelmingly oriented up-gradient, in a manner more usually Cellscouldgeneratechemotacticgradientstodrivedispersalby associated with neutrophils and Dictyostelium [28] than cancer either of two mechanisms. They could secrete an autocrine cells, which typically chemotax less accurately [29]. The high chemorepellent and migrate away from it. We have previously chemotacticindexwasmaintainedthroughoutasustainedperiod, shownthistobeakeydriverofEntamoebapathogenesis[25],in with narrow and accurate confidence interval, and strongly which chemotaxis away from ethanol generated by the amoebas significantRayleightest[30]fordirectionalmigration(Figure2E). themselves causescells tomigrate fromthelumen ofthe gutinto Thus serum contains a remarkably potent chemoattractant for the walls of the gut and eventually the liver of the patient. melanoma cells. Alternatively, the melanoma cells could locally break down or We therefore conclude that melanoma dispersal across the consume a chemoattractant that is produced externally, but chamber is driven by positive chemotaxis towards an attractant spatially homogeneously [26,27], as seen in the zebrafish lateral thatispresentinserum.Theattractantisbrokendownbythecells lineprimordium[19,21].Ineithercase,densepopulationsofcells themselves into agradient that efficiently disperses cells. PLOSBiology | www.plosbiology.org 4 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal Figure 2. Dispersal is due to a chemoattractant present in serum. All panels show data from melanoma cells migrating in chemotaxis chambers as described [15]. (A–B) Cells migrate from conditioned medium towards fresh medium. WM1158 cells were randomly attached to a coverslipandassembledinachamberin48hourWM1158cellconditionedmedium.Themediuminonechamberwasreplacedwithfreshmedium, whiletheotherwasleftalone.Tracksofindividualcellsareshownascolouredlines(A).Cellsmovetowardsthefreshmedium,asshownbythespider plot(B)showingallcelltracks.(C–D)ExampleimagesshowingWM239Ametastaticmelanomacellsafter21hoursinserum-freemedium(C)anda 0%–10%FBSgradient(D).Colouredpathsshowcentroidtracksfromtime0.(E)Quantitativeanalysisofchemotacticresponses.‘‘Spider’’plots(large panels), rose plots, mean chemotactic index, and Rayleigh test for directionality are shown for cells in serum-free medium and a 0%–10% FBS gradient(n.100cellsinthreeindependentexperimentsforbothconditions).SpiderplotsshowstrongchemotaxisinFBSgradients;inserum-free mediumonlyrandommovementisseen.Roseplotsshowoverallmovementfrom6–12hours;theproportionoftotalcellsineachsectorisshownon a log scale, with red lines representing the 95% confidence interval. The majority of cells in the FBS gradient move in the direction of the chemoattractant.Rayleightestsstatisticallyconfirmedthishighlysignificantunimodaldirectionality.Graphsofchemotacticindexweregenerated fromthesamedata. doi:10.1371/journal.pbio.1001966.g002 Chemotaxis during Tumour Progression physiologically appropriate behaviour (Figure 3A) [22]. Surpris- Onepotentialexplanationforcancercellsbecomingmetastatic ingly, all the lines we examined responded chemotactically to is that they evolve chemotactic competence as the tumours serum gradients (Figure3B). Cells from metastases were more develop [13,31,32], and thus move from unsteered to steered motile than cells from earlier stages (Figure 3C); highly invasive migration. We therefore examined the ability of a panel of cell (VGP)cellswereslightlymoreaccurate,butnotsignificantlyfaster lines isolated from different tumour stages and selected for thanthebiologicallyearlier,RGPcells.Cellsfrommoreadvanced PLOSBiology | www.plosbiology.org 5 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal tumours responded more robustly, but the progression from LPA Is the Dominant Attractant in Serum in 2-D and 3-D nonmetastatic to metastatic was not marked by the cells newly Assays acquiring responsiveness—all lines examined were chemotactic To examine whether LPA was the principal attractive enough to spread away from the tumour efficiently in the component of serum, we assayed chemotaxis in the presence of presenceofanappropriategradient.Severallinesofdatasuggest theantagonistKi16425,whichspecificallyinhibitsbindingtoLPA that genetic and epigenetic changes during progression from receptors 1 and 3 [38]. The effects were again remarkably clear. RGPtoVGPincreasecells’abilitytosurvive[33];ourdataimply 10mM Ki16425 blocked cell spread in our original, density- that it is cell survival, rather than chemotactic sensitivity, that dependent assay (Movie S4) and chemotaxis towards 10% serum defines the difference. The increase in migratory ability could (Figure 5A;MovieS5),reducingthechemotacticindexfrommore modulatecells’abilitytoescapefromaprimarytumour,butour than +0.4 to zero (Figure 5B). Ki16425-treated cells were principal conclusion is that melanoma cells from all stages are obviously healthy, and moved similarly to untreated cells, with chemotactic. similar track lengths, showing that the treatment was not making thecellsnonspecificallysickornon-motile.KnockdownofLPAR1 Identifying the Chemoattractant in Serum bysiRNAhadasimilareffect(FigureS2A),showingthatLPAR1 There are multiple reports of chemotaxis driving metastasis of is the key receptor for this process, and 10mM Ki16425 also melanoma and other tumour cells, in particular breast cancer. blocked chemotaxis towards pure LPA (Figure S2B). Again, LPA Published accounts of chemotactic invasion most often describe chemotaxis is not tumour stage-specific; Ki16425 blocked growth factors as the attractants—for example EGF for solid chemotaxis in all cell lines from all stages of cancer progression tumours [34], and EGF, hepatocyte growth factor (HGF), and (Figure 5C). RGP and VGP cell lines were completely inhibited, stem cell factor (SCF)/KitL for melanoma [13]. However these andthehighlymotilemetastaticlinesweresubstantiallyinhibited. attractants were often identified in transwell chambers, which as The residual chemotaxis in the presence of inhibitor could earlier discussed are subject to a range of artifacts, in particular represent either incomplete inhibition by the antagonist, or a false positive. For example, the positive well might promote smallamountofchemotaxistoanotheragent.Fromthesedata,we survival, growth, or adhesion of cells that move randomly across concludethatLPAisoverwhelminglythedominantchemoattrac- the membrane. Our direct-viewing chambers provide a far more tant inserum forall linesexamined. rigorousanalysis.Wethereforetestedabroadrangeofattractants While chamber-based assays are optimized to allow accurate in our assays. To our surprise, no growth factor acted as an anddetailedrecording,theyprovidea2-Dviewofaprocessthat attractanttoanymeasurabledegree(Figure 4A);steeporshallow more often happens in 3-D tissues in vivo [39]. We therefore gradients gave no obvious movement upgradient, and no examined the role of LPA in a widely used organotypic tumour significant chemotactic index towards any growth factor tested cellinvasionmodel[40].Inthissystemmelanomacellsareadded (Figure 4B). We therefore conclude that the chemotaxis towards to the top of a plug of collagen in which fibroblasts are growing, serumweobservedwasunlikelytobetowardsgrowthfactors.This and over time they migrate vertically downwards into the 3-D does not, of course, demonstrate that melanoma cells are never matrix.Duringthecourseoftheassay,thecollagenplugissetso chemotactic towards growth factors; but it clearly shows the only its bottom face contacts the medium, at which point surprisingandefficientchemotaxistowardsserumobservedearlier malignant melanoma cells invade downwards [41]. We hypoth- ismediated byanother molecule. esized that the melanoma cells were driven by a self-generated EGFandPDGFdidincreasecells’speed(Figure 4C),butthey LPA gradient as in Figure 1B, once fresh LPA could only be did not provide directional specificity. They therefore acted as supplied fromthebottom. Thishypothesis issupported by assays chemokines, regulating overall cell behaviour, rather than as inwhichthecollagenplugsremainsubmerged,andnoinvasionis chemoattractants that could steer thecells. seen (Figure S3), further rejecting contact inhibition of migration The striking accuracy of chemotaxis demonstrated by melano- as a mechanism of invasion. When the gels were treated with ma cells towards serum was more reminiscent of neutrophil Ki16425, themelanomacells did not invadedownwards intothe chemotaxis towards formyl peptides, or Dictyostelium towards gel (despite comparable numbers of cells at the end, showing no cAMP, which signal through G-protein coupled receptors changeingrowthorsurvival).Quantitativeanalysisconfirmsthat (GPCRs) rather than growth factor receptors like EGFR and Ki16425 strongly inhibited invasion in both cell lines that were PDGFR. We therefore investigated SDF-1, the ligand for the invasiveinthisassay(Figure 5Dand5E).ThusLPAisadominant GPCR CXCR4, which has been associated with poor prognosis steering system for 3-D organotypic assays, as well as for 2-D and malignancy of melanoma [35]; but again, it was not chamber assays. measurably attractive to cells in our assays (Figure 4B, compare with strongresponse toserum). Melanoma Cells Break down LPA to Form Outward- However, LPA, another well-known component of serum that signals through GPCRs, was strikingly attractive to melanoma Facing Gradients cells. A gradient from 0 to 1 mM LPA across the chamber Ourearlierdata(Figures1Band2A,inparticular)showedthat (consistent with the approximate concentration of LPA in serum; melanoma cells disperse by depleting a chemoattractant from see below) induced chemotaxis almost as effectively as 0%–10% serum. We therefore tested whether melanoma cells are able to serum (Figure 4D), yielding a comparable chemotactic index deplete LPA from their surroundings. Full medium with and (Figure 4E). This was a surprise: LPA is more typically described withoutserumwasincubatedwithdifferentdensitiesofmelanoma as an inflammatory mitogen, acting on haematopoietic cells such cells for different times, then LPA was extracted from the asmacrophages.Itappearsfrequentlyinthecancerliterature,but conditioned medium and analyzed by mass spectrometry [42]. moreoftenasamitogenandchemokineforcancercells,actingvia This confirms that the melanoma cells effectively break down autotaxin, whichcatalyzes theproduction of LPA fromlysophos- LPA; the conditioned medium was depleted in a density- phatidylcholine [36]. However in our assays the chemotaxis of dependent manner (Figure 6A) and in a timescale that correlates melanomatoLPAwasagainremarkablyaccuratecomparedwith withthemediumconditioningexperiments inFigure 2Aand2B. theweaker chemotaxistypically seen incancer cells [37]. Oneadvantageofusingmassspectrometryistheidentificationof PLOSBiology | www.plosbiology.org 6 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal Figure3.Chemotaxisofcellsfromdifferentmelanomastages.(A)Chemotaxisofapanelofsixcelllinesfromdifferentmelanomastages (RGP,green;VGP,purple;metastatic,red)upa0%–10%FBSgradientwasmeasuredasabove(n$45cellspercellline).(B)Chemotacticindexofcells fromdifferentstages.Datafrom(A)werecollatedbymelanomastage.Chemotaxisimprovesasthestageofmelanomaprogresses,althougheven theearliestRGPcellsshowclearchemotaxis.(C)Speedsofcellsfromdifferentstages.Datafrom(A)werecollatedbymelanomastage.Metastatic lines are conspicuously faster (p-values from unpaired t-tests), although again the speed of RGP and VGP cells is still relatively high for non- haematopoieticcells. doi:10.1371/journal.pbio.1001966.g003 PLOSBiology | www.plosbiology.org 7 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal PLOSBiology | www.plosbiology.org 8 October2014 | Volume 12 | Issue 10 | e1001966 Self-generatedLPAGradientsDriveMelanomaDispersal Figure4. Identification of LPA,ratherthan growthfactors, as the principalattractantin serum.(A) WM239Acells wereexposed to gradientsoflow(light)andhigh(dark)concentrationsofseveralgrowthfactorsandthechemokineSDF-1incombinationwithSFM.SpiderandRose plotswithRayleightestsareshown(n.40cellsforeachcondition).ConcentrationstestedwereEGF(6.25and25ng/ml),PDGF(25and100ng/ml), HGF(10and30ng/ml),SCF(10and100ng/ml),andSDF-1(100and300ng/ml).Noneshowsobviouschemotaxis.(B)Quantificationofdatafrom(A). Serumgradientspromotestrongchemotaxis(p,0.0001,unpairedt-test),butgradientsofallgrowthfactorstestedshownosignificantchemotactic index (p$0.40). (C) Growth factors enhance cell speed. Data quantitated from the cells in Figure3A. Directionless cell speed was measured by totallingthedistancemovedbetweentimepoints.EGFandPDGFstimulatecellsinserum-freeminimalmediumtospeedscomparablewithserum gradients. Single asterisk: Different from SFM alone, p,0.001, unpaired t-test; double asterisk: p,0.0001). (D) LPA and serum drive comparably efficientchemotaxis.WM239Acellswereexaminedinachamberrespondingto0%–10%FBSand0–1mMLPA.Thespiderplotshowssimilarcellular responses to the two gradients. (E) Quantitative analysis of chemotaxis towards LPA and serum. Chemotactic index was calculated from three experimentsincludingthatshownin(D).Cellsrespondcomparablytobothconditions.BarsshowSEM. doi:10.1371/journal.pbio.1001966.g004 molecular subspecies. The biological activity of LPA is known to still in the tissues immediately outside the tumour (Figure 8D). vary with its structure [43,44]. In particular, there is a strong Cells at the edges of the tumour are therefore experiencing an correlation between biological activity and the degree of poly- outward-oriented LPA gradient tending to drive them out into unsaturation, and also acyl chain length [45]. Melanoma cells surrounding tissues andvasculature. broke down the biologically active species more rapidly than the WefurtherexaminedtheLPAspeciesinthetissue.Formsthat others (Figure 6B), ensuring that the most active species also arestronglyassociatedwithsignalling,inparticular18:2-LPAand formed thesteepest gradients. 20:4-LPA [48], formed the steepest gradients (Figure 8E), while gradients of non-signalling forms such as 18:0-LPA were flatter. The Role of Growth Factors This finding further supports the idea that the gradients of LPA The results we have obtained conflict with the established are specifically produced assignals targeted atLPA receptors. dogma that growth factors are primary melanoma chemoattrac- This study is, to our knowledge, the first time a chemotactic tants [13]. To reconcile these accounts with our data, we gradient has been directly measured around tumours in vivo. examined the role of growth factors during chemotaxis towards Thereareanumberofsituationswherethepresenceofagradient LPA. As shown previously (Figure4C), EGF and (particularly) hasbeeninferredfromcellularbehaviour,mostprominentlyinthe PDGF increased the basal speed of cells. Gradients of EGF and paracrine loops shown by Segall and others [10]. However, such PDGF, and mixtures of both, enhanced the accuracy of gradientsmustbydefinitionbelocalandtendtobetransient.The chemotaxis to LPA (Figure 7); LPA, EGF, and PDGF together gradients we observe in melanomas are clear, large-scale, and in serum-free minimal medium were as effective as 10% serum. provide a convincing driver for cell dispersal, and one highly Mosttellingly,however,whencellswerepresentedwithLPAand plausibleexplanationofwhymelanomasaboveacertainsize,and growth factor gradients oriented in opposite directions, they hence Breslowthickness, always tendtobemetastatic. chemotaxed towards the LPA not the growth factors; if anything they migrated towards the LPA with enhanced efficiency Discussion (Figure 7B, bottom two lines). Thus when examined in the high levels of detail afforded by our chambers, the growth factors are In this work, we have shown that LPA is a potent chemoat- potentiallyimportantaccessoryfactorsthatincreasecellspeedand tractantformelanomacellsingeneral,andthatoutward-oriented efficiency of chemotaxis, but they do not themselves act as gradients of LPA are self-generated by melanoma cells. Because chemoattractants. self-generated gradients are always oriented away from tumours, Theseresultsarereminiscentofobservationsofdevelopmentin this combination provides a plausible mechanism for driving vivo,inwhichthegrowthfactorSCFpromotesmigrationbutnot tumourcelldispersal.Wedonotexcludeothermechanisms;ithas direction of melanoblast migration [46]. It is possible that the for example been proposed that LPA regulates cadherin levels melanoma chemotaxis to growth factors observed in other work [49], which would not be visible in our assays. Growth factor [13] is due to changes in speed alone, which as discussed earlier chemotaxis may be visible under the appropriate conditions can cause a false positive in transwell assays. It has also been (though, as discussed previously, many data are from transwell shown that growth factors can cause cancer cells to secrete LPA assays, which are artifact-prone and unreliable). Likewise, we do [47],whichcould alsoprovideanelement ofindirectchemotaxis notexcludeothermechanismsthanchemotaxis.Contactinhibition in manytypesof assay. ofmigrationoccursinmanycelltypesderivedfromtheneuralcrest andsoisprobablyfoundinmelanoma,anddefectsincellgrowth LPA Gradients in Tumours In Vivo and survival in inappropriate locations are of course important factors. But the mechanism we have found that overwhelmingly WehaveclearlyshownthatLPAisapotentchemoattractantfor dominates the dispersal in our assays is robust and is apparently melanomacellsofallbiologicalstages.Todeterminewhetherthis chemotaxis was an important driver of melanoma chemotaxis in activeinahighproportionofmelanomas.Itisthereforelikelytobe vivo, we investigated whether the tissue surrounding real a particularly important mediator of tumour cell dispersal. We melanomas contained LPA gradients that would direct cells out hypothesize that similar mechanisms will be common in cancer of tumours. Mice that are heterozygotes for the driver mutation metastasis. BrafV600E (the most prevalent driver of human melanomas) and deletion of the tumour suppressor PTEN develop sporadic The Source of LPA melanomas (Figure 8A) genetically and cytologically comparable The source of LPA around melanomas is unknown. In many tohumantumours(Figure 8B).Wetookpunchbiopsiesfromthe tumours, including melanoma, expression of autotaxin and thus tissue in and across melanomas (Figure8C) from several mice, autocrine production of LPA has been associated with tumour extracted total lipids, and examined LPA levels using mass progression[50].ThisLPAproductionappearstobeamechanism spectrometry.Inallnon-ulceratedmelanomasweexamined,LPA forpromoting melanomagrowth,rather than drivingchemotaxis levelswerelowinsidethetumour,higherattheedges,andhigher andinvasion.LPAgeneratedbythetumouritselfwouldbefound PLOSBiology | www.plosbiology.org 9 October2014 | Volume 12 | Issue 10 | e1001966