Critical transitions in chronic disease: transferring concepts from ecology to systems medicine

Critical

transitions

in

chronic

disease:

transferring

concepts

from

ecology

to

systems

medicine

Christophe

Trefois

,

Paul

MA

Antony

,

Jorge

Goncalves

,

Alexander

Skupin

and

Rudi

Balling

Ecosystemsandbiologicalsystemsareknowntobeinherently complexandtoexhibitnonlineardynamics.Diseasessuchas microbiomedysregulationordepressioncanbeseenas complexsystemsaswellandwereshowntoexhibitpatternsof nonlinearityintheirresponsetoperturbations.These nonlinearitiescanberevealedbyasuddenshiftinsystem states,forinstancefromhealthtodisease.Theidentification andcharacterizationofearlywarningsignalswhichcould predictupcomingcriticaltransitionsisofprimordialinterestas preventionofdiseaseonsetisamajoraiminhealthcare.Inthis review,wefocusonrecentevidenceforcriticaltransitionsin diseasesanddiscussthepotentialofsuchstudiesfor therapeuticapplications.

Addresses

1_{ExperimentalNeurobiologyGroup,LuxembourgCentreforSystems}

Biomedicine(LCSB),UniversityofLuxembourg,CampusBelval, 7AvenuedesHauts-Fourneaux,L-4362Esch-sur-Alzette,Luxembourg

2_{SystemsControlGroup,LuxembourgCentreforSystemsBiomedicine}

(LCSB),UniversityofLuxembourg,CampusBelval,7Avenuedes Hauts-Fourneaux,L-4362Esch-sur-Alzette,Luxembourg

3_{NationalCenterforMicroscopyandImagingResearch,Universityof}

CaliforniaSanDiego,9500GilmanDrive,LaJolla,CA,UnitedStates

4_{IntegrativeCellSignallingGroup,LuxembourgCentreforSystems}

Biomedicine(LCSB),UniversityofLuxembourg,CampusBelval, 7AvenuedesHauts-Fourneaux,L-4362Esch-sur-Alzette,Luxembourg Correspondingauthor:Balling,Rudi([email protected])

CurrentOpinioninBiotechnology2015,34:48–55 ThisreviewcomesfromathemedissueonSystemsbiology EditedbySarahMariaFendtandCostasDMaranas

http://dx.doi.org/10.1016/j.copbio.2014.11.020

0958-1669/#2014TheAuthors.PublishedbyElsevierLtd.Thisisan openaccessarticleundertheCCBY-NC-SAlicense(http:// creativecommons.org/licenses/by-nc-sa/3.0/).

Critical

transitions

in

complex

systems

Biologicalsystemsarecomplex,characterizedbyemerging

behaviorand oftenobeynonlinear dynamics(Box1).In

manycasesthisnonlinearbehaviorof biologicalsystems

leadstotippingpointswheretheequilibriumstate(Box1)

of thesystemabruptlychanges fromonestable stateto

another(Box1).Thischangeisalsocalledcriticaltransition

(Box1) or regimeshift[1].Although suddenstate

tran-sitions suchas phasetransitions or exothermalreactions

are established concepts in physics and chemistry,

increasing awareness rises that also complex biological

systems can exhibit abrupt changes in their dynamics.

The idea that natural systems might exhibit sudden

changesintheirdynamicalstatesoriginatedmostlyfrom

theoreticalmodelsover50yearsago[2–4]andarebasedon

mathematical catastrophic bifurcation theory [5]. These

studies,motivated bydescriptionsof magneticsystems,

laidtheconceptthatnaturalsystemsmighthave

alterna-tivestablestatesandthereby undergocriticaltransitions

(Box 1), typically without obvious warning signals

(Figure 1). At that time, these theories lacked robust

empiricalevidence[6].

Onlyinthelastdecade,severalstudiesprovidedevidence

for the existence of critical transitions in natural and

societal systems. In ecology [7], there are examples of

thedesertificationof Mediterraneanaridecosystems [8]

orthetreeabundanceintropicalforestandsavannah[9].

Critical transitions have also been associated with the

eutrophicationoflakes [10],collapseof fishpopulations

duetooverfishing[11]andalgaeovergrowthinCaribbean

coralreefs[12].Onlargerscalesuchastheearth’sclimate

system,reductionof Greenlandicesheetsormelting of

arctic sea-ice hasbeenassociated witha potential

tran-sitionintheglobalclimatesystem,whichmayormaynot

bereversible[13].Similartoecologicalsystems,humans

have complex traits. When considered as complex

sys-tems,bothpresentpositiveandnegativefeedbackloops,

inherentnonlinearity andhysteresis (Box1)[14].Here,

weshowhowconceptsfirstdevelopedinphysicsarenow

increasingly used to describe complex systems in the

contextof healthand disease.

Critical

transitions

in

medicine

Recently the concept of criticaltransitions and tipping

pointshasbeenappliedtoclinicalquestionsandsystems

medicine.Weareconvincedthatadetailed

understand-ingofcriticaltransitionsindiseaseonsetandprogression

willprovidebroadapplicationsinhealthcare.The

identi-ficationofearlywarningsignals(Box1)forexample,can

be expected to leverage prevention strategies. Already

identifiedsuddentransitionsinthemedicalcontexthave

been associated with gut microbiome dysregulation

[15], pulmonary disease [16], depression [17], type

1 and 2 diabetes [18,19], inflammation [20], start [21]

and termination [22] of epileptic seizures [23], cancer

[24], and cardiovascular events [25]. Examples will be

elaborated in more details in the section Examples of

Early

warning

signals

to

detect

upcoming

critical

transitions

A critical transitionis usually detectableafter the

tran-sition [26] and difficult to anticipate [27]. Before the

criticalforwardtransition,thesystem’sequilibriumstate

mightstayrelativelyunchangeduntiltheforwardtipping

point is reached (Figure 2c) [28]. Consequently, static

observations might not provide enough information to

detect upcoming abrupt transitions [29]. By contrast,

changingsystemdynamicshavebeensuggestedasearly

warning signals (EWS) for critical transitions [30]. A

general challengefor dataanalysis istheintrinsic noise

inbiologicalsystemswhichoriginatesfromthestochastic

nature of molecular interactions and heterogeneity of

individualentitieslikecellsor organisms.Abrief

expla-nationforthemostcommonlyusedearlywarningsignals

canbefoundin Table1.

Theinfluenceoftherandombehaviorisamplifiedinthe

vicinityoftippingpointsbecausesmall perturbationsin

thevulnerableregimeofthesystemcanhavelargeeffects

(Figure 2). Due to this amplified heterogeneity, an

increase in variance [10,31] or coefficient of variation

[32] has been associated with upcoming critical

tran-sitions. Further, an increase or decrease of lag-1

auto-correlationmayindicatetheunfoldingofanabruptshift

[6,33]. An increase of flickering activity[34] has been

identified as EWS for critical transitions in lake

eutro-phication[35,36].Changingskewnessinthedistribution

of time-series climate data, could be used as a robust

indicator forsome complexnatural systems[37].

Dyna-micalnetworkbiomarkersareanewapproachtopredict

upcoming transitions and showed promising results for

livercancer[38].Criticalslowingdownisfoundinsome

ecological systems when approaching a tipping point

[39–41,42]. A transition from vegetationto

desertifica-tionwasprecededbychangesinthespatialdistributionof

vegetativepatches[8].Finally,significant

heteroscedas-ticity [43]was observed one yearbefore acritical

tran-sition in a lake [44]. This multitude of early warning

signals to detect criticaltransitionsshows thatthe

non-linearityindifferentsystemsarenotalwaysaccompanied

bythesameEWS.

Examples

of

clinical

relevance

in

the

context

of

critical

transitions

Arecentexampleofalternativestablestateswasfoundin

the context of microbiome dysregulation in human

intestines. A highly diverse and dynamically evolving

microbial ecosystem, mainly including bacteria from

the Firmicutes, Actinobacteria and Bacteroidetes phyla, is

living in thehuman gutand its dysregulationcan have

stark consequences on health [45]. It is thought that

certain diseasessuchas obesityand irritablebowel

syn-dromemightunfoldduetotransitionsinmicrobial

com-position [46]. Recently, the question whether these

transitionsare linearorrather abrupt andnonlinearwas

raised[47].Alternativestable stateswithhighresilience

wereidentifiedinhumanindividualsafterrepeated

pro-longed exposure to an antibiotic (Figure 2) [47]. Such

alternative stable states of bacterial ecosystems were

confirmedin astudycovering1000individuals[15].

Acute asthmaattacksarecharacterizedbyaconstriction

ofthebronchioleswhichultimatelyleadstopatchinessin

lungventilation[16]anddifficultiestobreathe[48].Such

patch clusters canpotentially leadto criticaltransitions

via an interaction of feedback mechanisms [16]. One

study developed amodelof abronchial treeand

simu-latedincrementalairwaysmoothmusclestimulation[16].

At a stimulation threshold, the system underwent a

criticaltransition andshowedsevereventilation defects

[48]. In environmental epidemiology, early warning

signals in theformsof changing varianceand skewness

were found for the deterioration of lung activity in

humansafterexposureto ozone[49].

Clinical depression is characterized by a wide array of

symptoms such as inability to sleep, low mood, loss of

interestandsuicidaltendencies.Onsetandremissionof

clinical depression can occur suddenly. A recent study

suggests that critical slowing down could be an early

warning signal for onset and termination of depression

[17].Duringthestudy,subjectswereloggingtheirmood

statesbyself-assessmentonanemotionalscaleatrandom

intervals duringtheday.Infollow-upassessments,

sub-jects werere-evaluated using thesame scales.

Interest-ingly, existence of critical slowing down based on the

collectedmoodstateswasconfirmedandwasindicativeof

Box1Stablestate—Astablestateofadynamical(phenotype) systemdoesnotchangeitsaveragephenotypictraitwhenbeing exposedtosmallrandomperturbations.

Alternativestablestates—Distinctstablestatesofadynamical systemforthesamesetofenvironmentalconditions.These alternativestablestatesareseparatedbyametastablestate. Equilibriumstate—Thestateinwhichasystemstayswhenno additionalexternalperturbationsareapplied.

Criticaltransition—Suddenshiftfromonestablestatetoan alternativeonewheretheactualtransitioncanbetriggeredbysmall perturbations.

Tippingpoint—Athresholdpointatwhichasystemwillundergoa criticaltransitionwhenexposedtoperturbations.

Hysteresis—Currentsystemstateisdependedonboththeinput andthehistoryofthesystem.

Nonlinearity—Adynamicalsystemisnonlinearifthesetofunderlying differentialequationsexhibitproductsofvariablesinatleastone equation.Notethatsystemscanobeylineardynamicsevenifthe outcomehasanonlinearformlikeitisthecaseofexponentialgrowth. Earlywarningsignal—Anobservablevariablewhosedynamics changeconsiderablybeforeacriticaltransitionoccurs.

upcomingtransitionsfromanormaltoadepressedstateor

viceversa.

Diabetesmellitushastwomajorsub-types.Intype1

dia-betes,patientscannotproduceenoughinsulinandintype

2 diabetes the produced insulin is not used efficiently

[50]. Both sub-types can be categorized into multiple

diseasestagesrangingfrompre-diseasetofullonsetwith

clinicaldiagnosisoftenoccurringonlyatthelatterstage.

Dynamical network biomarkers werefound to beearly

warningsignalsfor criticaltransitionsfromapre-disease

to a disease state for type 1 [18,51] and type 2 [19]

diabetes in mice. On the scale of cellular interactions,

critical transitions were identified in insulin-producing

pancreatic-cellislets[52].Inthissystem,acriticalamount

ofb-celldeathatwhichpancreaticisletsloseconnectivity

intheisletnetworkresultsin systemicfailure.

Dynamics of certain pro-inflammatory cytokines playa

pivotal role in the unfolding of inflammation [53]. For

instance, Interleukin-6 (IL-6) dynamics in pig blood

revealedthatvariationsofthiscytokineovertimecould

potentiallyserve as anearly warningsignal[20].Scheff

etal.showedonamathematicalmodelthatunder

gradu-allyincreasing inflammatoryconditions, thestate ofthe

systemremainsstableuntilacriticalthresholdatwhichit

evolves fromahealthy to aninflammatory state[54].A

mathematicalmodelwascreatedfromexperimentaldata

where humans underwent an endotoxin application to

simulate chronic inflammation. More generally, it has

been proposed that early warning signals could predict

upcoming epidemic infectious inflammatory disease

occurrencesin populations[55].

Epileptic seizures start and terminate suddenly and

diminish quality of life in patients [56]. Increased

var-ianceinspikingpatternsofindividualneuronshasbeen

proposedasanearlywarningsignaltodetecttheonsetofa

sudden epilepsy seizure [57]. Further, the mechanism

leadingto theunfolding ofa seizure,whenconsidering

groups rather than single neurons, was attributable to

a Hopf bifurcation. Recently, Kramer and colleagues

Figure1 Poincaré 1854-1912: Geometric approaches to nonlinear dynamics 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 Applications of nonlinear oscillators in physics and engineering

Inventions: radio, radar, and laser

Theory of Hopf bifurcations

From theory to critical transitions in engineering, biology, and medicine

Identification of potential critical transitions in ecosystems:

Lewontin, Holling, and May

Empirical evidence for critical transitions in biological systems Breakthrough discovery

of nonlinear dynamics in giant squid axon

Hodgkin-Huxley model Nobel prize in 1963

Desertification and spatial vegetation patterns

Self-termination of epileptic seizures

Early warning signals for onset and termination of depression

Theoretical foundations

Applications in physics and engineering

Hypothetical critical transitions in biological systems

Empirical evidence Historical milestones for the understanding of critical transitions in biomedicine

Extracellular Medium Intracellular Medium

Current Opinion in Biotechnology

presentedevidenceoftheexistenceofacriticaltransition

betweentheictalandpost-ictalstatesofaseizure,

repre-sentingalternative stablestates[22].Thiscritical

tran-sition waspreceded bycriticalslowing down, increased

autocorrelationand flickering.

Outlook:

cross-sectional

and

longitudinal

monitoring

of

health

states

A potentialapplication of using some of the described

early warning signals for upcoming critical transitions

could be in the generation of a population reference

profileandthelongitudinalmonitoringofhealthypeople

and patients.Similartothehumanreferencegenome,a

populationreferenceprofilewouldrepresentthevariation

ofsomeoftheearlywarningsignals,intimeandspecific

conditionsfromarepresentativehealthypopulation.Due

to the expected variability between individuals,

longi-tudinal studies are expected to provide clearer signals

than studies across populations. Such unique patient

referenceprofilescould,forinstance,beobtainedduring

regular visits at the clinic. At each point of sample

extraction, the current profile could be compared to

thereferenceprofile.Thedifferencebetweentheprofiles

would then be indicative of disease progression. At a

certain threshold, the clinician could take appropriate

measures to prevent a criticaltransition from ahealthy

to adiseasestateinindividual patients.

Recently,wearabledevicessuchasFitbit1devices[58],

andautomatedvoicerecordingdevices[59,60]havebeen

proposed to be able to provide easy-use monitoring of

specific activities in patients. For instance, symptoms

such as tiredness, anxietyor speech difficulties,

contri-buting towearing-offinParkinson’sdiseasepatientsare

difficulttoassessduringbriefclinicvisits[61,62].

Wear-ing-off meansthatspecificsymptomsmakea

re-appear-ance before the next scheduled administration of the

therapeutic drugs. Wearing-off could become one way

toassesscriticaltransitionsinParkinson’sdiseasefroma

clinicalpointofview.Onecouldcomputevariablessuch

asthenumberofwearing-offoccurrencesduringtheday

orthetimeintervalbetweenthebeginningofwearing-off

and the next scheduled drug intake. Consequently,

indicators as discussed throughoutthisreview could be

Figure2 Condition Phenotype Health Disease Parameter value Output (a) (b) (c) Change in variability Change in variability Critical slowing down Change in phenotype distribution Flickering Hysteresis Emerging properties

Temporal warning signals

Conditi onal warning signals FTP RTP Cond ition Phentoype

Current Opinion in Biotechnology

determined and used to assess the progression of the

disease and classify the patient’s disease states. Such

predictors could allow clinicians to modify treatments

accordinglyandtherebyincreasepatients’qualityoflife.

Conclusions

Althoughalargenumberoftheoreticalstudiesand

com-putationalsimulations havebeen carried outto provide

evidenceforcatastrophicshifts,therearestillonlyafew

biologicalandmedicalstudiesexperimentallyor

empiri-callyvalidatingthepredictionsandtherelevanceofthe

criticaltransitionconceptfor medicalapplications.

Furthermore,errorrateestimationsforfalsepositiveand

falsenegativeratesofthemodelsunderlyingthe

predic-tions are rarely available. In order to prevent or revert

critical system states this will become more important

[63]. In some disciplines, replicate experiments are

often not possible, for example, when analyzing the

occurrenceofice-ages,thesuddencollapseof

fish-popu-lationsorcatastrophictransitionsinthefinancialmarkets,

ashasbeenwitnessedafewyearsago.Biologicalsystems

and disease pathogenesis however might bestudied in

depthinanimalmodels,patientswithidenticalorclosely

relatedclinicalsymptomsoradverseresponsestospecific

drugtreatments.Duetothisfeasibilityandtheir

import-ance,biologicalandbiomedicalapplicationsmightplaya

driverroleinourattemptstoexperimentallydissectand

understandcomplexnonlinearnaturalsystems.The

de-velopmentof system specific mathematical modelsand

machinelearningtools,integratingawiderangeofomics

andclinicaldata,andpriorknowledge,forexample,from

literature, public databases and modern media, will

becomeacentral domaininsystemsmedicine.

Anoteofcaution

Critical slowingdown mightonly occur in specific

situ-ations.Catastrophiccollapsecanoccurwithoutpriorearly

warning signals in autocorrelation or variance [64].

Thereforetheseearlywarningsignalsarenotalways

glob-allyapplicableanditwassuggestedthatinfacteachsystem

mighthaveacharacteristicsubsetofearlywarningsignals

[65].Thisindividualityisbasedontheintricateinterplayof

theunderlyingdynamicswhicharetypicallynonlinearand

includeintrinsicrandomforces originating,for example,

fromthestochasticnatureofmolecularinteractions.The

existence or occurrence of multiple stable states

corre-spondstoinduciblesystemsmanyofwhichexhibit

excit-able dynamics. Many natural systems exhibit such

excitabledynamicslikeneuronspiking[66]orlaserpulsing

[67].Drivingsuchnonlineardynamicsbyrandom

pertur-bationscaninducestatetransitionswithamathematically

strictlydefinedtransitionrate[68]andleadtonon-trivial

effectslikestochasticandcoherenceresonance[69,70].

Table1

Potentialindicatorsforearlywarning.

Earlywarningsignal Definition Observedsystemvariable

Variance Scatterofdata Phosphorusconcentrationinlake[10]

Pollutantacrossmultipleregions[31] Interleukin-6levels[20]

Spikingpatternsinneurons[57]

Coefficientofvariation Standarddeviationnormalized

bythemean

Populationdensity[32]

Spatialheterogeneityofventilation[16] Infectiouspopulationdynamics[55] Ozonelevels[49_]

Lag-1autocorrelation Correlationofdatawithitselfshifted byonetimepoint

Rateofresourceharvesting[33_] Connectedpopulationdensity[32]

Flickering Systemstatesaredrivenbackand

forthbetweenalternatestablestates byintrinsicnoise

Sedimentdiatomcomposition[35] Iceconductivity[34]

Phosphorusdynamics[36]

Invasiveelectrocorticogramrecordings[22_]

Skewness Thirdstandardizedmomentofthe

distributionofsystemstates

Vegetationbiomass[37] Phosphorusdensity[37] Dynamicalnetworkbiomarkers Evolutionovertimeofdifferencein

molecularnetworks

Geneexpressionprofiles[18,19,38,51]

Criticalslowingdown Recoveryratestendtozeroafter

smallexternalperturbation

Calciumcarbonatelevels[41] Cyanobacteriapopulationdensity[42] Nutrientcyclinginlakes[39]

Macrophytecover[39] Vegetationgrowth[40] Mooddynamics[17_]

Spatialdistribution Non-randomdistributionofelements

inabiologicalentity

Vegetationpatchiness[8] Conditionalheteroscedasticity Varianceisconditionalonpasttimepoints E.colipopulationgrowth[28]

Stochasticresonancedescribesthescenariowhereaweak

periodicinputsignalof anexcitablenonlinearsystemis

amplified by intrinsic noise. Although first used to

describe ice-age periodicity [71] stochastic resonance

was subsequently foundin avariety of naturaland

bio-logical systems [72]. In general, stochastic resonance is

characterizedbyamaximuminthesignal-to-noiseratioin

dependenceonthenoiseintensityandcorresponds toa

minimuminthecoefficientofvariation.Similarly,

coher-enceresonanceoccursinexcitablesystemsonlydrivenby

inherentnoiseandisalsocharacterizedbyminimalvalues

of coefficient of variations for optimal random

pertur-bations [70]. Thesemechanisms can interfere with the

general assumptions that critical transitions are

accom-panied byanincrease ofvariability.

Theseprominentexamplesdemonstratethatnoisy

non-linear systems like those found in biology can obey

unintuitive dynamics. The resulting complex behavior

including noise induced effects question the general

applicationoftheearlywarningsignalsmentionedabove

but emphasize the need to complement experimental

investigations with mechanistic theoretical models to

fully characterizecriticaltransitions.

Acknowledgments

WewouldliketothankLindaWampachforherhelpwithgraphical elementsinthisarticle.Theauthorsalsoacknowledgethe‘FondsNational delaRechercheLuxembourg(FNR)’forfinancialsupportofCTthrough anAFRgrant(3118186).ASreceivedfundingfromtheproject‘plan TechnologiesdelaSante´ parleGouvernementduGrand-Duche´ de Luxembourg’throughtheLuxembourgCentreforSystemsBiomedicine (LCSB),UniversityofLuxembourg.Allauthorshavecontributedtothe editingofthisarticle.

References

and

recommended

reading

Papersofparticularinterest,publishedwithintheperiodofreview, havebeenhighlightedas:

 ofspecialinterest ofoutstandinginterest

1. SchefferM:CriticalTransitionsinNatureandSociety.Princeton UniversityPress;2009.

2. LewontinRC:Themeaningofstability.BrookhavenSympBiol 1969,22:13-24.

3. HollingCS:Resilienceandstabilityofecologicalsystems.Annu RevEcolSyst1973,4:1-23.

4. MayRM:Thresholdsandbreakpointsinecosystemswitha multiplicityofstablestates.Nature1977,269:471-477. 5. StrogatzStevenH:NonlinearDynamicsAndChaos.Westview

Press;2001.

6. SchefferM,BascompteJ,BrockWA,BrovkinV,CarpenterSR, DakosV,HeldH,vanNesEH,RietkerkM,SugiharaG: Early-warningsignalsforcriticaltransitions.Nature2009,461:53-59. 7. SchefferM,CarpenterS,FoleyJA,FolkeC,WalkerB:

Catastrophicshiftsinecosystems.Nature2001,413:591-596. 8. Ke´fiS,RietkerkM,AladosCL,PueyoY,PapanastasisVP,ElaichA,

deRuiterPC:Spatialvegetationpatternsandimminent desertificationinMediterraneanaridecosystems.Nature2007, 449:213-217.

9. HirotaM,HolmgrenM,VanNesEH,SchefferM:Globalresilience oftropicalforestandsavannatocriticaltransitions.Science 2011,334:232-235.

10. CarpenterSR,BrockWA:Risingvariance:aleadingindicatorof ecologicaltransition.EcolLett2006,9:311-318.

11. JacksonJB,KirbyMX,BergerWH,BjorndalKA,BotsfordLW, BourqueBJ,BradburyRH,CookeR,ErlandsonJ,EstesJAetal.: Historicaloverfishingandtherecentcollapseofcoastal ecosystems.Science2001,293:629-637.

12. MumbyPJ,HastingsA,EdwardsHJ:Thresholdsand theresilienceofCaribbeancoralreefs.Nature2007, 450:98-101.

13. LentonTM,HeldH,KrieglerE,HallJW,LuchtW,RahmstorfS, SchellnhuberHJ:TippingelementsintheEarth’sclimate system.ProcNatlAcadSciUSA2008,105:1786-1793. 14. HigginsJP:Nonlinearsystemsinmedicine.YaleJBiolMed

2003,75:247-260. 15.



LahtiL,Saloja¨rviJ,SalonenA,SchefferM,deVosWM:Tipping elementsinthehumanintestinalecosystem.NatCommun 2014,5:1-10.

In this study to identify nonlinear dynamics in the microbiome of 1000 adults, the authorsidentified signs of alternative stable states betweensubgroupsofbacterialphyla.

16. VenegasJG,WinklerT,MuschG,VidalMeloMF,LayfieldD, TgavalekosN,FischmanAJ,CallahanRJ,BellaniG,HarrisRS: Self-organizedpatchinessinasthmaasapreludeto catastrophicshifts.Nature2005,434:777-782. 17.



VandeLeemputIa,WichersM,CramerAOJ,BorsboomD, TuerlinckxF,KuppensP,vanNesEH,ViechtbauerW,GiltayEJ, AggenSHetal.:Criticalslowingdownasearlywarningforthe onsetandterminationofdepression.ProcNatlAcadSciUSA 2014,111:87-92.

In this medical study, theexistence of alternativestable states and associatedtippingpointswasshownusingmoodinformation. Interest-ingly, the authorsalso providean attempt to predict theonset and terminationofdepressionbasedonthemoodinformation.

18. ZengT,ZhangC,ZhangW,LiuR,LiuJ,ChenL:Decipheringearly developmentofcomplexdiseasesbyprogressivemodule network.Methods2014,67:334-343.

19. LiM,ZengT,LiuR,ChenL:Detectingtissue-specificearly warningsignalsforcomplexdiseasesbasedondynamical networkbiomarkers:studyoftype2diabetesbycross-tissue analysis.BriefBioinform2014,15:229-243.

20. TambuyzerT,DeWaeleT,ChiersK,BerckmansD,GoddeerisBM, AertsJ-M:Interleukin-6dynamicsasabasisforan early-warningmonitorforsepsisandinflammationinindividual pigs.ResVetSci2014,96:460-463.

21. LittB,EstellerR,EchauzJ,D’AlessandroM,ShorR,HenryT, PennellP,EpsteinC,BakayR,DichterMetal.:Epilepticseizures maybeginhoursinadvanceofclinicalonset.Neuron2001, 30:51-64.

22. 

KramerMA,TruccoloW,EdenUT,LepageKQ,HochbergLR, EskandarEN,MadsenJR,LeeJW,MaheshwariA,HalgrenEetal.: Humanseizuresself-terminateacrossspatialscalesviaa criticaltransition.ProcNatlAcadSciUSA2012, 109:21116-21121.

This studyhastwo major outcomes. Firstit wasshown that critical transitions can bestudied in acellular systemand second that the termination of epilepsy seizures can be associated to such critical transitions.

23. McSharryPE,SmithLa,TarassenkoL:Predictionofepileptic seizures:arenonlinearmethodsrelevant? NatMed2003, 9:241-242.

24. KianercyA,VeltriR,PientaKJ:Criticaltransitionsinagame theoreticmodeloftumourmetabolism.InterfaceFocus2014, 4:20140014.

26. BoettigerC,HastingsA:Quantifyinglimitstodetectionofearly warningforcriticaltransitions.JRSocInterface2012, 9:2527-2539.

27. ClarkJS,CarpenterSR,BarberM,CollinsS,DobsonA,FoleyJA, LodgeDM,PascualM,PielkeR,PizerWetal.:Ecological forecasts:anemergingimperative.Science2001,293:657-660. 28. TakimotoG:Earlywarningsignalsofdemographicregime

shiftsininvadingpopulations.PopulEcol2009,51:419-426. 29. CarpenterSR,BrockWA,ColeJJ,KitchellML,PaceJF:Leading

indicatorsoftrophiccascades.EcolLett2008,11:128-138. 30.



DakosV,CarpenterSR,BrockWA,EllisonAM,GuttalV,IvesAR, Ke´fiS,LivinaV,SeekellDA,vanNesEHetal.:Methodsfor detectingearlywarningsofcriticaltransitionsintimeseries illustratedusingsimulatedecologicaldata.PLoSOne2012, 7:e41010.

Inthisarticle,theauthorsprovideapracticalguideonhowtodetectEWS incriticaltransitionsfromtimeseriesdataandmostimportantlymade availableatoolboxtoanalyzesuchdata.

31. BrockWA,CarpenterSR:Varianceasaleadingindicatorof regimeshiftinecosystemservices.EcolSoc2006:11. 32. DaiL,KorolevKS,GoreJ:Slowerrecoveryinspacebefore

collapseofconnectedpopulations.Nature2013,496:355-358. 33.



DakosV,vanNesEH,D’OdoricoP,SchefferM:Robustnessof varianceandautocorrelationasindicatorsofcriticalslowing down.Ecology2012,93:264-271.

Inthisworktheauthorsshowthatanincreaseaswellasadecreasein variancecanbeindicativeofanupcomingcriticaltransition.Importantly, theyshowthatvariancecanbeunderestimatedinsparsedatasets. 34. TaylorKC,LamoreyGW,DoyleGA,AlleyRB,GrootesPM,

MayewskiPA,WhiteJWC,BarlowLK:Theflickeringswitchof latePleistoceneclimatechange.Nature1993,361:432-436. 35.



WangR,DearingJA,LangdonPG,ZhangE,YangX,DakosV, SchefferM:Flickeringgivesearlywarningsignalsofacritical transitiontoaeutrophiclakestate.Nature2012,492:419-422. Wangetal.show,usingacombinationofexperimentaldataandamodel, thatflickeringcanoccurinanecologicalsystemreflectedbybimodality, anincreaseinvariance,andadecreaseinautocorrelationandskewness. 36. DakosV,NesEH,SchefferM:Flickeringasanearlywarning

signal.TheorEcol2013,6:309-317.

37. GuttalV,JayaprakashC:Changingskewness:anearlywarning signalofregimeshiftsinecosystems.EcolLett2008, 11:450-460.

38. WuX,ChenL,WangX:Networkbiomarkers,interaction networksanddynamicalnetworkbiomarkersinrespiratory diseases.ClinTranslMed2014,3:16.

39. VanNesEH,SchefferM:Slowrecoveryfromperturbationsasa genericindicatorofanearbycatastrophicshift.AmNat2007, 169:738-747.

40. Ke´fiS,DakosV,SchefferM,VanNesEH,RietkerkM:Early warningsignalsalsoprecedenon-catastrophictransitions. Oikos2013,122:641-648.

41. DakosV,SchefferM,vanNesEH,BrovkinV,PetoukhovV,HeldH: Slowingdownasanearlywarningsignalforabruptclimate change.ProcNatlAcadSciUSA2008,105:14308-14312. 42.



VeraartAJ,FaassenEJ,DakosV,vanNesEH,Lu¨rlingM, SchefferM:Recoveryratesreflectdistancetoatippingpointin alivingsystem.Nature2012,481:357-359.

In this fully controllable experimental study, Veraart and colleagues present evidence that critical slowing down occurs before a critical transitioninapopulationofcyanobacteria.Theconclusionisthatcritical slowingdownmayoccurinrealcomplexsystems.

43. SeekellDA,CarpenterSR,ClineTJ,PaceML:Conditional heteroskedasticityforecastsregimeshiftina whole-ecosystemexperiment.Ecosystems2012,15:741-747. 44. SeekellDA,CarpenterSR,PaceML:Conditional

heteroscedasticityasaleadingindicatorofecologicalregime shifts.AmNat2011,178:442-451.

45. DeVosWM,deVosEAJ:Roleoftheintestinalmicrobiomein healthanddisease:fromcorrelationtocausation.NutrRev 2012,70(Suppl1):S45-S56.

46. LozuponeCA,StombaughJI,GordonJI,JanssonJK,KnightR: Diversity,stabilityandresilienceofthehumangutmicrobiota. Nature2012,489:220-230.

47. DethlefsenL,RelmanDA:Incompleterecoveryand

individualizedresponsesofthehumandistalgutmicrobiotato repeatedantibioticperturbation.ProcNatlAcadSciUSA2011, 108(Suppl):4554-4561.

48. WinklerT,VenegasJG,HarrisRS:Mathematicalmodelingof ventilationdefectsinasthma.DrugDiscovTodayDisModel 2014http://dx.doi.org/10.1016/j.ddmod.2014.02.008. 49.



HsiehN-H,ChengY-H,LiaoC-M:Changingvarianceand skewnessasleadingindicatorsfordetectingozone exposure-associatedlungfunctiondecrement.StochEnvironResRisk Assess2014http://dx.doi.org/10.1007/s00477-014-0887-2. Forthiswork,adultsubjectswereexposedtovariouslevelsofozone.The authorsconcludethatstatisticalearlywarningsignalscanbeusedto detectrespiratoryproblemsandproposeariskassessmentframeworkto potentiallyallowforanearlyidentificationoftheshiftfromhealthyto disease.

50. PolonskyKS:Thepast200yearsindiabetes.NEnglJMed2012, 367:1332-1340.

51. LiuX,LiuR,ZhaoX-M,ChenL:Detectingearly-warningsignals oftype1diabetesanditsleadingbiomolecularnetworksby dynamicalnetworkbiomarkers.BMCMedGenomics2013, 6(Suppl2):S8.

52. StamperIJ,JacksonE,WangX:Phasetransitionsinpancreatic isletcellularnetworksandimplicationsfortype-1diabetes. PhysRevE2014,89:012719.

53. AshleyNT,WeilZM,NelsonRJ:Inflammation:mechanisms, costs,andnaturalvariation.AnnuRevEcolEvolSyst2012, 43:385-406.

54. ScheffJD,CalvanoSE,AndroulakisIP:Predictingcritical transitionsinamodelofsystemicinflammation.JTheorBiol 2013,338:9-15.

55. O’ReganSM,DrakeJM:Theoryofearlywarningsignalsof diseaseemergenceandleadingindicatorsofelimination. TheorEcol2013,6:333-357.

56. MormannF,AndrzejakRG,ElgerCE,LehnertzK:Seizure prediction:thelongandwindingroad.Brain2007,130:314-333. 57. MeiselC,KuehnC:Scalingeffectsandspatio-temporal

multileveldynamicsinepilepticseizures.PLoSOne2012, 7:e30371.

58. CookDJ,ThompsonJE,PrinsenSK,DearaniJA,DeschampsC: Functionalrecoveryintheelderlyaftermajorsurgery: assessmentofmobilityrecoveryusingwirelesstechnology. AnnThoracSurg2013,96:1057-1061.

59. TsanasA,LittleMA,McSharryPE,RamigLO:Accurate telemonitoringofParkinson’sdiseaseprogressionby noninvasivespeechtests.IEEETransBiomedEng2010, 57:884-893.

60. OllerDK,NiyogiP,GrayS,RichardsJa,GilkersonJ,XuD, YapanelU,WarrenSF:Automatedvocalanalysisofnaturalistic recordingsfromchildrenwithautism,languagedelay,and typicaldevelopment.ProcNatlAcadSciUSA2010, 107:13354-13359.

61. StorchA,SchneiderCB,WolzM,Stu¨rwaldY,NebeA,OdinP, MahlerA,FuchsG,JostWH,ChaudhuriKRetal.:Nonmotor fluctuationsinParkinsondisease:severityandcorrelation withmotorcomplications.Neurology2013,80:800-809. 62. StacyM,BowronA,GuttmanM,HauserR,HughesK,LarsenJP,

LeWittP,OertelW,QuinnN,SethiKetal.:Identificationofmotor andnonmotorwearing-offinParkinson’sdisease:

comparisonofapatientquestionnaireversusaclinician assessment.MovDisord2005,20:726-733.

63. 

This critical review looks atcases where critical transitions are not precededbyspecificearlywarningsignalssuchascriticalslowingdown andshowspublishedworkwithcontradictingconclusionontheuseof EWS.Thisarticlehighlightsthegrowingobservationsthattheremightbe noglobal indicators butrather thatcertainclassesofsystemsshare commondynamics.

64. 

BoerlijstMC,OudmanT,deRoosAM:Catastrophiccollapsecan occurwithoutearlywarning:examplesofsilentcatastrophes instructuredecologicalmodels.PLoSOne2013,8:e62033. Inthisoriginalarticle,Boerlijstshowsthatcriticaltransitionscanoccur withnoearlywarningsignals.Theauthorsconcludethatmoretheoretical workneedstobeperformedasthecurrenthypothesis,thatgenericearly warningsignalsexistinallnaturalsystems,canberefuted.

65. HastingsA,WyshamDB:Regimeshiftsinecologicalsystems canoccurwithnowarning.EcolLett2010,13:464-472. 66. LindnerB:Effectsofnoiseinexcitablesystems.PhysRep2004,

392:321-424.

67. GiacomelliG,GiudiciM,BalleS,TredicceJ:Experimental evidenceofcoherenceresonanceinanopticalsystem.Phys RevLett2000,84:3298-3301.

68. KramersHA:Brownianmotioninafieldofforceandthe diffusionmodelofchemicalreactions.Physica1940, 7:284-304.

69. BenziR,SuteraA,VulpianiA:Themechanismofstochastic resonance.JPhysAMathGen1981,14:L453-L457.

70. PikovskyA,KurthsJ:Coherenceresonanceinanoise-driven excitablesystem.PhysRevLett1997,78:775-778.

71. BenziR,ParisiG,SuteraA,VulpianiA:Atheoryofstochastic resonanceinclimaticchange.SIAMJApplMath1983, 43:565-578.