Preterm birth leads to impaired rich-club organization and fronto-paralimbic/limbic structural connectivity in newborns

(1)

Article

Reference

Preterm birth leads to impaired rich-club organization and fronto-paralimbic/limbic structural connectivity in newborns

ALVES SA DE ALMEIDA, Joana Rita, et al.

Abstract

Prematurity disrupts brain development during a critical period of brain growth and organization and is known to be associated with an increased risk of neurodevelopmental impairments. Investigating whole-brain structural connectivity alterations accompanying preterm birth may provide a better comprehension of the neurobiological mechanisms related to the later neurocognitive deficits observed in this population. Using a connectome approach, we aimed to study the impact of prematurity on neonatal whole-brain structural network organization at term-equivalent age. In this cohort study, twenty-four very preterm infants at term-equivalent age (VPT-TEA) and fourteen full-term (FT) newborns underwent a brain MRI exam at term age, comprising T2-weighted imaging and diffusion MRI, used to reconstruct brain connectomes by applying probabilistic constrained spherical deconvolution whole-brain tractography. The topological properties of brain networks were quantified through a graph-theoretical approach. Furthermore, edge-wise connectivity strength was compared between groups. Overall, VPT-TEA infants' brain networks evidenced [...]

ALVES SA DE ALMEIDA, Joana Rita, et al . Preterm birth leads to impaired rich-club

organization and fronto-paralimbic/limbic structural connectivity in newborns. NeuroImage , 2021, vol. 225, p. 117440

DOI : 10.1016/j.neuroimage.2020.117440 PMID : 33039621

Available at:

http://archive-ouverte.unige.ch/unige:155481

Disclaimer: layout of this document may differ from the published version.

1 / 1

(2)

ContentslistsavailableatScienceDirect

NeuroImage

journalhomepage:www.elsevier.com/locate/neuroimage

Preterm birth leads to impaired rich-club organization and fronto-paralimbic/limbic structural connectivity in newborns

Joana Sa de Almeida

^a

, Djalel-Eddine Meskaldji

^a^,^b

, Serafeim Loukas

^a^,^c

, Lara Lordier

^a

, Laura Gui

^d

, François Lazeyras

^d

, Petra S. Hüppi

^a^,^∗

aDivision of Development and Growth, Department of Woman, Child and Adolescent, University Hospitals of Geneva, Geneva, Switzerland

bInstitute of Mathematics, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

cInstitute of Bioengineering, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

dDepartment of Radiology and Medical Informatics, Center of BioMedical Imaging (CIBM), University of Geneva, Geneva, Switzerland

a r t i c le i n f o

Keywords:

Diﬀusion magnetic resonance imaging Connectomics

Graph-theory

Human brain development Preterm birth

a b s t r a ct

Prematuritydisruptsbraindevelopmentduringacriticalperiodofbraingrowthandorganizationandisknown tobeassociatedwithanincreasedriskofneurodevelopmentalimpairments.Investigatingwhole-brainstructural connectivityalterationsaccompanyingpretermbirthmayprovideabettercomprehensionoftheneurobiological mechanismsrelatedtothelaterneurocognitivedeﬁcitsobservedinthispopulation.

Usingaconnectomeapproach,weaimedtostudytheimpactofprematurityonneonatalwhole-brainstruc- turalnetworkorganizationatterm-equivalentage. Inthiscohortstudy,twenty-fourverypreterminfantsat term-equivalentage(VPT-TEA)andfourteenfull-term(FT)newbornsunderwentabrainMRIexam atterm age,comprisingT2-weightedimaginganddiﬀusionMRI,usedtoreconstructbrainconnectomesbyapplying probabilisticconstrainedsphericaldeconvolutionwhole-braintractography.Thetopologicalpropertiesofbrain networkswerequantiﬁedthroughagraph-theoreticalapproach.Furthermore,edge-wiseconnectivitystrength wascomparedbetweengroups.

Overall,VPT-TEAinfants’brainnetworksevidencedincreasedsegregationanddecreasedintegrationcapac- ity,revealedbyanincreasedclusteringcoefficient,increasedmodularity,increasedcharacteristicpathlength, decreasedglobalefficiencyanddiminishedrich-clubcoefficient.Furthermore,incomparisontoFT,VPT-TEA infantshaddecreasedconnectivitystrengthinvariouscortico-cortical,cortico-subcorticalandintra-subcortical networks,themajorityofthembeingintra-hemisphericfronto-paralimbicandfronto-limbic.Inter-hemispheric connectivitywasalsodecreasedinVPT-TEAinfants,namelythroughconnectionslinkingtotheleftprecuneus orleftdorsalcingulategyrus– tworegionsthatwerefoundtobehubsinFTbutnotinVPT-TEAinfants.More- over,posteriorregionsfromDefault-Mode-Network(DMN),namelyprecuneusandposteriorcingulategyrus,had decreasedstructuralconnectivityinVPT-TEAgroup.

OurfindingthatVPT-TEAinfants’brainnetworksdisplayedincreasedmodularity,weakenedrich-clubcon- nectivityanddiminishedglobalefficiencycomparedtoFTinfantssuggestsadelayedtransitionfromalocal architecture,focusedonshort-rangeconnections,toamoredistributedarchitecturewithefficientlong-rangecon- nectionsinthoseinfants.Thedisruptionofconnectivityinfronto-paralimbic/limbicandposteriorDMNregions mightunderliethebehavioralandsocialcognitiondifficultiespreviouslyreportedinthepretermpopulation.

1. Introduction

Dynamicmacrostructural and microstructuralchanges take place fromthemid-fetalstagetobirth.Indeed,thethirdtrimesterofgesta-

Abbreviations:CSD,constrainedsphericaldeconvolution;DCG,dorsalcingulategyrus;DGM,deepgray-matter;dMRI,diﬀusionmagneticresonanceimaging;

FA,fractionalanisotropy;fMRI,functionalmagneticresonanceimaging;FT,full-term;GM,gray-matter;OFC,orbito-frontalcortex;PCG,posteriorcingulategyrus;

PCUN,precuneus;RC,rich-club;SC,streamlinecounting;SW,small-worldindex,TEA,term-equivalentage;TPO,temporalpole;VPT-TEA,verypreterminfantsat term-equivalentage;WM,whitematter.

∗Correspondingauthor.

E-mailaddress:petra.huppi@hcuge.ch(P.S.Hüppi).

tionischaracterizedbyarapidbraingrowth,withgray-matter(GM) andwhitematter(WM)volume expansiondrivenbyincreasingden- sity, diameter and myelination of both dendritic and axonal struc- tures(Duboisetal.,2008;InnocentiandPrice,2005;Kissetal.,2014;

https://doi.org/10.1016/j.neuroimage.2020.117440

Received4May2020;Receivedinrevisedform8September2020;Accepted5October2020 Availableonline8October2020

(3)

Mills etal., 2016; Nossin-Manor et al., 2013),as well asthe estab- lishmentoffunctionalthalamocorticalandotherlong-rangebraincon- nections(Batalleetal.,2017; KostovicandJovanov-Milosevic,2006; KostovicandJudas,2010;Kostovicetal.,2014;vandenHeuveletal., 2015).

Pretermbirthdisruptsbrainmaturationduringacriticalperiodof fetalbraingrowth(Kissetal.,2014;Radley andMorrison,2005).It impactstypicalbraindevelopmentalprocesses,asaresultofcomplex environmentalfactors(Simmonsetal.,2010),leadingconsequentlyto structuralbrainalterations.Thesealterationsmightbeattheoriginof thelaterneurodevelopmentalimpairmentsobservedinpreterminfants, comprisingmotor, cognitive,behavioral andsocio-emotional deﬁcits (Anjarietal.,2007;Balletal.,2012;Cismaruetal.,2016;Huppietal., 1998;Inderetal.,2005;MontagnaandNosarti,2016;Thompsonetal., 2007,2013).

Magneticresonanceimaging(MRI)hasbeenemployedtostudythe earlybraindevelopmentalanomaliesobservedinpreterminfants.Such anomalieshave beenshowntobe associatedwithspeciﬁcneuropsy- chologicaldeﬁcitsafterpretermbirthandincluderegionalcorticaland subcorticalvolumetricchanges(Balletal.,2017;Cismaruetal.,2016; HuppiandDubois,2006;Mentetal.,2009;Nosarti,2013;Padillaetal., 2015;Petersonetal.,2000),aswellasalterationsofbrainWMnetworks andtheirmicrostructuralcharacteristics(Balletal.,2014;Batalleetal., 2017;Fischi-Gomezetal.,2015;Mullenetal.,2011;Panneketal.,2018; Pechevaetal.,2018;Rogersetal.,2012;SadeAlmeidaetal.,2019).

Previousstudieshavefocused mostlyonregional brainstructural diﬀerencesbetweenverypreterminfantsatterm-equivalentage(VPT- TEA)andfull-termborninfants(FT).Todate,thereisstilllimitedknowl- edgeregardingwhole-brainstructuralalterationsatterm-equivalentage (TEA)followingpretermbirth.

Themacroscopicwhole-brainstructuralnetworkorganizationcanbe investigatedusingaconnectomicanalysis,wherebrainnetworksarede- rivedbywhole-braintractography,throughreconstructionofWMcon- nectionsbetweenbrainregions usingbraindiﬀusionMRI(dMRI).In theresultingconnectome,brainregionsarerepresentedbynodesand thestructuralconnectionsbetweenbrainregionsbyedges.Theedge weightscanbeeitherthenumberofstreamlinesconnectingtworegions (SC)orcanbedeﬁnedbymeasuresofmicrostructuralorganization,such asfractionalanisotropy(FA).

Graph-basedmeasuresallowustostudybrainnetworktopology, revealingmeaningfulinformationregardingsmall-worldness,integration,segregation,modularorganization,thepresenceofhubsandrich- clubconnectivity(Bullmore andSporns,2009; HeandEvans, 2010; Meunieretal.,2010;vandenHeuvelandSporns,2013).Thesenetwork measureshavebeenshowntobereliablebiomarkersfordiscriminat- ingnormalandabnormalbrainnetworks(Loetal.,2010;Owenetal., 2013;Shuetal.,2011).Moreover,alterationsinstructuralconnectivity havebeenrelatedtoarangeofdevelopmentaldisorders(Crossleyetal., 2015;vandenHeuveletal.,2013).

Todate,onlythreestudieshaveusedaconnectomicanalysistore- constructwhole-brainnetworksofVPT-TEAandFTinfantsandcompare theirtopologicalorganizationattermageusinggraph-analysis.Inpar- ticular,Balletal.,usingunweightedbrainconnectomes,havefocused mainlyonrich-clubconnectivityandhavefoundthat,althoughitwas intactfollowingpretermbirth,VPT-TEAhadagreaterproportionofdi- rectcortico-corticalconnectionswithin therich-clubthanFTinfants, whatwasinterpretedasameasureofaltereddeepgray-matter(DGM) connectivity.Inaddition,brainnetworksofVPT-TEAinfantsevidenced increasedlocalconnectivitybetweenperipheralnodes,correlatingwith increasednetworksegregation(Balletal.,2014).Leeetal.andBrown etal.,usingweightedbrainconnectomes,havealsoaddressedthisques- tionusinggraphtheoreticalmeasures(Brownetal.,2014;Leeetal., 2019).Leeetal.havefoundadecreasednetworkglobaleﬃciencyin VPT-TEAinfantsincomparisontoFTnewborns(Leeetal.,2019).On theotherhand,Brownetal.havefoundanincreasedglobalnetwork eﬃciencyandmodularityinVPT-TEAinfants.However,theyhavecom-

pared theirowncohort of VPT-TEAinfants toanexternalcohort of FTinfants froma previouslypublishedstudy(Yapetal., 2011) and havestatedthattheirfindingswereresultingfromdifferencesin im- ageacquisitionandconnectomeconstructionpipelines(Brownetal., 2014).Therefore,literatureisstillnotcompletelyclearregardingthis matterandlackingamorecompletegraph-basedanalysisbetweenthe brainnetworksofVPT-TEAandFTinfants.Additionally,thereareno reportsregardinghubdifferencesandthereisonlyonestudyevaluating whole-brainedge-wiseconnectivitystrengthdifferencesbetweenthese groupsattermage(Panneketal.,2013).Furtherinformationonthis topicwouldthusbeofgreatimportanceforunderstandingtheimpact ofpretermbirthonglobalbrainnetworkorganizationandconnectivity atTEA.

Inthisstudy,weconstructedweightedstructuralconnectomesofa cohortofVPT-TEAandFTinfantsusingdMRIwhole-brainprobabilis- ticconstrainedsphericaldeconvolution(CSD)-basedtractography.We aimedtostudywhole-brainnetworktopologicalalterationsrelatedto prematuritybyemployinggraph-basedanalysisonweightedstructural networksandcomparisonofbrainnetworkhubsandedge-wiseconnec- tivitydiﬀerencesbetweenthesetwogroupsofinfants.Capitalizingon thelatestmethodologicaladvances,weintendedtocontributetothe describedgapintheliteratureregardingthissubject.

Wehypothesizethat,incomparisontoFTneonates,VPT-TEAinfants’

structuralconnectomeswillrevealmacroscalebrainnetworkdiﬀerences relatedtothedisruptionofearlybraindevelopmentbypretermbirth.

2. Materialsandmethods 2.1. Subjects

Acohortofthirty-nineverypreterm(VPT)infants(gestationalage (GA)atbirth<32weeks)andtwenty-fourFTinfantswasrecruitedat theneonatalunitoftheUniversityHospitalsofGeneva(HUG),Switzer- land,from2013to2016,inordertoevaluatetheimpactofprematurity onbraindevelopmentandclinicalneurodevelopment.Thestudywas approvedbythelocalResearchEthicsCommitteeandwrittenparental consentwasobtainedbeforetheinfants’participationtothestudy.All subjectsunderwentanMRIexaminationatTEA(37–42weeksGA).In- fantswhoseMRIprotocolacquisitionwasincomplete,notcomprising aT2-weightedimageand/ordMRIsequence,orthatpresentedmajor focalbrainlesionswereexcludedfromtheanalysis.

Thefinalsampleusedinthiscase-controlstudywasof24VPTand 14 FTinfants. Asexpected,VPTinfants’ GAat birth,weight, height andheadcircumferenceat birth,aswell asAPGARscores weresig- nificantly inferiortothoseofFTnewborns.Additionally,presenceof bronchopulmonarydysplasiawassignificantlysuperiorinVPTinfants.

NosigniﬁcantdiﬀerencesbetweenVPTandFTgroupswerefoundinthe followingdemographicandperinatalvariables:sex,intrauterinegrowth restriction,intraventricularhemorrhage(gradeI-IV),neonatalasphyxia, bloodculturepositivesepsis,gestationalageatMRIandsocioeconomic parentalstatus(Largoetal.,1989)(Table1).

2.2. MRIacquisition

Allinfantswerescannedafterreceivingbreastorformulafeeding, duringnaturalsleep(nosedationused)andusingavacuummattress forimmobilization.MR-compatibleheadphoneswereused(MRconfon, Magdeburg,Germany)toprotectinfantsfromthescanner’snoise.All infantsweremonitoredforheartrateandoxygensaturationduringthe entirescanningtime.

MRI acquisition was performed on 3.0T Siemens MR scanners (Siemens, Erlangen, Germany):Siemens TIMTrio witha 32-channel headcoilandSiemensPrismawitha64-channelheadcoil.Chi-squared test revealed no signiﬁcant diﬀerence regarding the distribution of scanner/coils between groups, 𝜒²(1)= 1.057, p = 0.304 (TIMTrio- 32channel:FT=11,VPT=15;Prisma-64channel:FT=3,VPT=9).

(4)

Table1.

Clinicalcharacteristicsoftheinfants.

Clinical characteristics FT n = 14 VPT-TEA n = 24 p -value ^∗FT vs. VPT-TEA Gestational age at birth, weeks, mean (SD) 39.2 ( ± 1.3) 28.3 ( ± 2.5) 0.0001 ^∗

Gestational age at birth, weeks, range 37 ^2/7– 41 ^5/7 23 ^6/7– 32 ^0/7

Sex: female (%)/male (%) 6(15.8)/8(21.1) 10(26.3)/14(36.8) 0.943 Birth weight, gram, mean (SD) 3223.6 ( ± 440.5) 1077.5 ( ± 328.7) 0.0001 ^∗ Birth height, centimetre, mean (SD) 49.8 ( ± 1.7) 36.2 ( ± 3.6) 0.0001 ^∗ Birth head circumference (cm), mean (SD) 34.3 ( ± 1.1) 25.8 ( ± 2.6) 0.0001 ^∗ APGAR score, 1 min (SD) 8.9 ( ± 0.5) 4.8 ( ± 3.1) 0.0001 ^∗ APGAR score, 5 min (SD) 9.5 ( ± 0.9) 7.1 ( ± 1.9) 0.0001 ^∗ Intrauterine Growth Restriction, n (%) 1 (2.6) 3 (7.9) 0.604

Bronchopulmonary dysplasia (%) 0 (0.0) 7 (18.4) 0.025 ^∗

Intraventricular Hemorrhage grade I-II (%) 0 (0.0) 5 (13.2) 0.067 Intraventricular Hemorrhage grade III-IV (%) 0 (0.0) 0 (0.0)

Neonatal asphyxia, n (%) 0 (0.0) 0 (0.0)

Blood culture positive Sepsis (%) 0 (0.0) 5 (13.2) 0.067 Gestational age at MRI scan, weeks, mean (SD) 39.6 ( ± 1.2) 40.3 ( ± 0.6) 0.073 Gestational age at MRI scan, weeks, range 37 ^2/7– 42 ^0/7 39 ^1/7– 41 ^1/7

Socioeconomic score (range 2–12), mean (SD) 4.10 ( ± 2.5) 5.67 ( ± 3.1) 0.169

∗Group-characteristicswerecomparedusingindependentsamplesT-testforcontinuousvariablesandchi-squared testforcategoricalvariables.

Table2.

Group-wisecomparisonofdMRImotionmetrics.

Motion metrics FT n = 14 VPT-TEA n = 24 p -value ^∗FT vs. VPT-TEA Number of outliers, mean percentage (SD) 3.24 ( ± 1.68) 3.05 ( ± 2.33) 0.789

Number of outliers, percentage, range 0.87 – 6.22 0.22 – 9.67

Absolute between volumes motion, mm (SD) 1.40 (0.46) 1.39 (0.49) 0.950 Absolute between volumes motion, mm, range 0.60 – 2.30 0.64 – 2.41

Relative between volumes motion, mm (SD) 0.66 ( ± 0.30) 0.76 ( ± 0.38) 0.375 Relative between volumes motion, mm, range 0.16 – 1.21 0.25 – 1.88

∗MotionmetricswerecomparedbetweengroupsusingindependentsamplesT-test.

T2-weightedimageswereacquiredusingthefollowingparameters:

113coronalslices,TR=4990ms,TE=151ms,flipangle=150°,matrix size=256×164;voxelsize=0.4×0.4×1.2mm³, totalscantimeof 6:01minutes.dMRIwasacquiredwithasingle-shotspinecho-planar imaging(SE-EPI)Stejksal-Tannersequence(TE=84ms,TR=7400ms, acquisitionmatrix128×128mm,reconstructionmatrix128×128mm, 60 slices, voxel size2 ×2× 2mm³). Imageswere acquiredin the axialplane,in anterior-posterior(AP)phaseencoding(PE)direction, withdiffusiongradientsappliedin30 non-collineardirectionswitha b-valueof1000s/mm²andonenon-diffusionweightedimage(b=0), withatotalscantimeof4:12minutes.dMRImotionmetricsestimates per subject,includingpercentageof outliers,as wellas relativeand absolute between volumes motion, werecalculated using EDDY QC toolsfromdiffusiontoolboxoftheFMRIBSoftwareLibrary,FSLv6.0.3, https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/(Behrensetal.,2003;Smithetal., 2004). No statistically significant differences between groups were foundforanyofthemotionmetrics(Table2,SupplementaryFig.S1).

2.3. Pre-processing

T2-weighted brainvolumeswerebias-corrected,skull-striped and tissue-segmentedinto WM,GM, DGM, cerebrospinalfluid (CSF)and cerebellum using an automatic neonatal-specific segmentation algorithm (Gui et al., 2012). dMRI data were preprocessed using the diffusion toolbox of the FMRIB Software Library, FSL v5.0.10, https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/(Behrensetal.,2003;Smithetal., 2004). Brain extraction was performed using “BET”. Eddy current- induceddistortionsandgrosssubjectmovementwerecorrectedusing theFSL“EDDY” tool(AnderssonandSotiropoulos,2016)optimizedfor neonataldMRIdata.This commandis partoftheneonataldMRIau- tomatedpipelinefromthedevelopingHumanConnectomeProjectand correctsdistortionscausedbymotion-inducedsignaldropoutandintra-

volumesubjectmovement(Anderssonetal.,2016,2017;Bastianietal., 2019).

2.4. Networknodesdeﬁnition:brainparcellation

Parcellationintocorticalandsubcorticalregionswasbasedonthe UNCneonatalbrainatlas(Shietal., 2011),consistingof90 regions, whichrepresentedthenodesusedinthenetworkanalysis.

TheT2UNCneonatalbrainatlas(Shietal.,2011)wasregisteredto eachsubject’sT2-weightedstructuralimagebymeansofanon-linear registration using the advanced normalization tools (ANTs)package (Avantsetal.,2011)adaptedforneonataldata(Bastianietal.,2019).

Subject’sdMRIdatawasregisteredtotheT2-weightedstructuralimage bymeansofFSLFLIRT(Jenkinsonetal., 2002).Alltheregistrations wereindividuallyvisuallyinspectedinordertoguaranteethequalityof theprocess.

Wethencombinedthelineartransformation(subject’sdMRItoT2 structural)withthenon-linearwarpregistration(subject’sT2structural toT2atlas)toobtainthetransformationsneededtobringtheT2atlas tosubjects’dMRIspace.Therefore,theanatomicalregions(labels)from theUNCatlaswerepropagatedtoeachsubject’sdMRIspacebyapplying thiscombinedtransformationusingnearestneighborinterpolation,and usedasnodesfornetworkanalysis.

2.5. Networkedgesdeﬁnition:tractography

InordertoreconstructWMfiberstodefinenetworkedges,whole- brain tractography was performed in native diffusion space using probabilisticsingle-shell,singletissueconstrainedsphericaldeconvo- lution (SSST-CSD)tracking. It was performed by meansof MRtrix3 (Tournier etal., 2019), using iFOD2, anatomically constrained tractography(ACT)andbacktrackre-trackingalgorithms,withstreamlines seededdynamicallythroughouttheWM,producing100millionstream-

(5)

linespersubject.Examplesofthetheoverlapbetweenthemrtrix5TT tissuesegmentationimageandthesubject’snativediﬀusionspace(dif- fusionb0image)inrepresentativesubjectsfrombothpretermandfull- terminfants’groupscanbefoundinSupplementaryFig.S2.Therecon- structeddatawasthenfurtherprocessedusingspherical-deconvolution informedﬁlteringoftractograms(SIFT)algorithm,producing10million streamlinesforeachsubject,whichwereusedasbrainnetworkedges.

2.6. Networkconstructionandglobalnetworkanalysis

InordertoconstructtheindividualWMstructuralbrainnetworks, the reconstructed set of streamlines were combined with each subject’sanatomicalparcellation.Thisresultedina90×90interregional connectivitymatrix,whereeachanatomicalregionisconsideredasa node(SupplementaryTableS1),andeachpairofnodesisconnected byanedge,representingthestrengthofconnectivitybetweenthecor- respondingregions.Self-connections wereexcluded.Foreachpairof nodes/regions,we computedthenumber ofﬁltered streamlines(SC) connectingbothregions,aswellasthemeanfractionalanisotropy(FA) along streamlines connectingthe regions.In order toremove spuri- ousconnectionsintheSC-weighted(SCw)connectomes,allconnections whereFAwasinferiororequalto0.1wereconsideredasnoiseﬂoorand thereforeremoved,accordingtoJonesandBasser(2004).Thisthresh- oldwaschosenduetothelowanisotropyofthepreterminfant’sbrain andhasbeenappliedinpreviousstudiesperformingstructuralnetwork analysisinthepretermandneonatalbrain(Brownetal.,2014;vanden Heuveletal.,2015;Zhaoetal.,2019).

Foranalysingresultsfromgraphnetworkmeasures,ﬁrst,foreach subject,network weights werenormalized by the maximum weight valueofthatnetwork,inordertonormalizealledgeweightspersub- ject.TheedgeweightsinSCwnetworkswillrepresentthereforerelative connectivity.

Graph theoretical analysis of SCw connective matrices was performed by using the Brain Connectivity Toolbox (BCT) for Matlab (RubinovandSporns,2010).Themetricsevaluatedincluded:

1 Density(d):computestheobservedconnectionsasapercentageof allpossibleconnections.

2 Totalstrength (S):computed asthe sumof thestrength (sum of weights)acrossallthenodesinthenetwork.

3 Characteristicpathlength(L):computedastheaverageofshortest pathlengthsbetweenallpairsofnodesinthenetwork(Wattsand Strogatz,1998).

4 Globaleﬃciency (GEﬀ): computed asthe averageinverse of the shortestpathlengths(RubinovandSporns,2010).

5 Averageclustering coefficient (C):clustering coefficient was first computedasthefractionof anode’sneighborsthatarealsocon- nectedtoeachother.These fractionsateachnodewereaveraged overall the nodesof thenetworktogive theoverallnetworkC (WattsandStrogatz,1998).

6 Localefficiency(LEff):computedastheaverageoftheglobaleffi- ciencyofeachnode’sneighborhoodsub‐graph,averagedoverallthe nodesofthenetwork(LatoraandMarchiori,2001).

7 Small-worldindex(SW):computedastheratioofthenormalizedC andthenormalizedL(whichwereobtainedastheratiobetweenthe metricsderivedfromeachsubjectnetworkanditsmatchedrandom- izednetworkmetrics).

8 Modularityindex(Q):Louvainalgorithmforcommunitydetection wasusedtoidentifymodulesinthenetwork(Luetal.2015)(Lu.Q wascalculatedbycomputingtheratiobetweenthenumberofcon- nections(sumofedgeweights)withinthemodulesandthenumber ofconnectionsexitingthesamemodules,andtakingthemaximum oftheseratiosacrossallpossiblemodules(Newman,2004).

9 Rich-club (RC) coeﬃcient (𝜙(𝜅)): computed as proposed by Op- sahl etal. for weightednetworks (Opsahlet al., 2008).Given a nodalstrength𝜅,weidentiﬁedthesubsetofnodesE_>𝜅withanodal

strength>𝜅.Theweighted𝜙(𝜅)wascomputedastheratiobetween thesumofweightsoftheconnectionsbetweenthenodespresentin

E_>𝜅,andthemaximalpossibleweightedconnectednesswithinE_>𝜅.

Thechosen𝜅 corresponded tothe9thhighestnodalstrengthper subject,inordertoincludethetop10%nodes.

Inordertoremove biasrelatedtothefractionofmeasure’svalue thatmayarrivebychance,wegenerated10matched randomequiv- alent networksfor eachsubject(where eachedgewasrewired1000 times),preservingthesamenumberofnodesanddegreedistributions asintheSCwnetworks.Thenetworkmeasuresofeachofthe10random networkswerethenaveragedpersubjectandusedfornormalization.

Normalized(calibrated)C(nC),LEff (nLEff),L(nL),GEff (nGEff)and Q(nQ)wereestimatedastheratiobetweenthemetricsderivedfrom eachsubjectnetworkanditsmatchedrandomizednetworkmetrics.The normalized𝜙(𝜅)(𝜙norm)wasobtainingbymultiplying𝜙(𝜅)bytheratio betweenthenumberofnodesincludedintotheRCpoolE>𝜅(varyingas afunctionofRCstrength𝜅)andthetotalnumberofnetworknodes.This computationismotivatedbythefactthat𝜙(𝜅)isinverselyproportional tothenumberofnodesincludedintheRCpool(Karolisetal.,2016).

Group-averageweightedstructuralconnectomesandmatchedran- domizednetworkstructuralconnectomeswerecomputedforbothVPT- TEAandFTinfants’groups.𝜙(𝜅)wascomputedasafunctionof𝜅per groupforbothnetworks.𝜙(𝜅)_normwascomputedhastheratiobetween the𝜙(𝜅),forthediﬀerentlevelsof𝜅,ofthegroup-averageweighted structural connectomeandthe𝜙(𝜅),forthesamelevelsof𝜅,of the matchedrandomizednetworkstructuralconnectomes.

Forthedetectionofhubswithineachgroup’sbrainnetworks,group- averagebinarystructuralconnectomes werecomputed forbothVPT- TEAandFTinfants’groups,byincludingthestructuralconnectionsthat werepresentinatleast70%ofthetotalgroup.Weusedthenodaldegree (thenumberofconnectionsofanode)asameasuretodeﬁnethehubs.

UsingBCT, nodaldegreewas calculatedforallnodesof eachbinary groupnetwork.Hubswereidentiﬁedforeachgroupasthenodeswith anodaldegreesuperiortotheaveragenodaldegreeplus1standard deviation.

2.7. Edge-wiseconnectivitystrengthanalysis

To evaluate edge-wise connectivity strength diﬀerences between groupsbasedonSCwstructuralnetworks,weperformedt-testsatthe connectionlevelwithfalsediscoveryrate(FDR)control(signiﬁcance valueof0.05and100.000randompermutations).Tothisend,weused theNBStoolboximplementedinMatlabbyZaleskyetal.(Benjaminiand Hochberg,1995;Zaleskyetal.,2010).

Connectomic data was visualized using the Circos software (Krzywinskietal.,2009).

Statisticallysigniﬁcantlydiﬀerentconnectionsbetweengroupswere categorized into: (1) cortico-cortical; (2) cortico-subcortical and (3) subcortico-subcortical.

2.8. Statisticalanalysis

Neonatal and demographic data, categorical variables (sex, intrauterine growth restriction,bronchopulmonary dysplasia,intraventricular hemorrhage (grade I-IV), blood culture positive sepsis and neonatalasphyxia)wereanalysedusingchi-squaredtest,whereascon- tinuousvariableswerecomparedusingindependentsamplest-test,with groupasindependentvariableandthefollowingdependentvariables:

GAatbirth,GAatMRI,birthweight,birthheight,birthheadcircum- ference,APGAR scoreat 1and5minafterbirthandsocioeconomic parentalstatus.dMRImotionmetricswerecomparedbetweengroups usingindependentsamplest-test,withgroupasindependentvariable andthefollowingdependentvariables:percentageofnumberofoutliers, absolutebetweenvolumesmotion,relativebetweenvolumesmotion.

Group-wise diﬀerences in network graph-theoretical metrics and number of thalamic connections were investigated with one-way

(6)

Table3.

Globalnetworkmeasurescomparisonbetweengroups.

Global network measures VPT-TEA FT p -value

d 0.435 ( ± 0.031) 0.452 ( ± 0.026) 0.135 S 1.158 ( ± 0.231) 1.170 ( ± 0.251) 0.892 nL 2.657 ( ± 0.675) 2.175 ( ± 0.391) 0.049 nGEff 0.701 ( ± 0.068) 0.775 ( ± 0.039) 0.045 nC 2.827 ( ± 0.155) 2.618 ( ± 0.135) 0.001 nLEff 1.522 ( ± 0.076) 1.416 ( ± 0.042) 0.0001 SW 1.117 ( ± 0.232) 1.157 ( ± 0.115) 0.677 nQ 1.017 ( ± 0.023) 0.998 ( ± 0.018) 0.006 One-wayANCOVA,controllingforGAatMRIandsex,wasperformedto determinediﬀerencesbetweengroups.Numbersinboldindicatesigniﬁcant resultsbetweengroups(p<0.05).

between-subjectsANCOVA,controllingforGA atMRIandsex,using IBMSPSSStatisticsversion25(IBMCorp.,Armonk,N.Y.,USA).One- waybetween-subjectsMANCOVA,controllingforGAatMRIandsex, wasperformedfirst,comprisingallgraph-theoreticalmetricsanalysed, namelynetworkstrength,networkdensity,nC,nLEff,nL,nGEff,nQand 𝜙norm,tocorrectforthemultiplicityofglobaltests.

Independent-samplest-testwasusedtocompareQfromSCwnet- workstoQfromitsmatchedrandomnetworks.

P-valuesthat resultedfrom edge-wisecomparison of connectivity matricesstatisticswerecorrectedusingBenjamini-HochbergFDRcon- trolprocedureatlevelalphaof5%,implementedintheNBStoolbox (BenjaminiandHochberg,1995;Zaleskyetal.,2010).

2.9. Dataandcodeavailability

Alldata wereacquiredinthe contextofthe researchprojectap- provedbytheethicalcommitteein2011.Thepatientconsentformdid notincludeanyclauseforreuseorshareofdata.Itstatedexplicitlythat alldata(clinic,biologicandimaging)wouldnotbeusedwithanyother aimapartfromthepresentresearchstudyandwouldnotbesharedwith thirdparties.

Software and code used in this study are publicly available as part of FSL v5.0.10 (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/), MRtrix3 (Tournieretal.,2019),ANTs(Avantsetal.,2011),BCT(Rubinovand Sporns,2010)andNBS(BenjaminiandHochberg,1995;Zaleskyetal., 2010)softwarepackages. dMRIdatawerepre-processed usingEDDY commandadaptedforneonatalmotion,whichispartfromtheneonatal dMRIautomatedpipelinefromdevelopingHumanConnectomeProject (dHCP,http://www.developingconnectome.org),andcan befoundat thislink:https://git.fmrib.ox.ac.uk/matteob/dHCP_neo_dMRI_pipeline_

release(Bastianietal.2019).

3. Results

3.1. Pretermneonatalstructuralbrainnetworktopologicalorganization

3.1.1. Integration,segregationandsmall-worldproperties

Structuralnetworktopology wascompared atTEAbetween VPT- TEA infants and FT newborns using metrics derived from graph- analysis.One-wayMANCOVAmultivariatetestyieldedasignificantdif- ferencebetweenthegroupcontrasts,[Wilks’𝜆=0.362,F(8,26)=4.182, p=0.005],whenconsideringallgraphmetrics.Giventhesignificance oftheoveralltest,theunivariatemaineffectsofeachmetricwereex- amined.

No signiﬁcant diﬀerences were found in mean network density (p=0.135)ortotalnetworkstrength(p=0.892)betweenFTnewborns andVPT-TEAinfants’SCwnetworks(Table3).

Weexplored thedifferencesinbrain networksintegration capac- itybetweenFTandVPT-TEA infantsusing characteristicpathlength andglobalefficiency.IncomparisontoFTnewborns,VPT-TEAinfants showed a significantly higher normalized characteristic path length

(p=0.049)andsignificantlydiminishednormalizedglobalefficiency (p= 0.045) (Fig.1A, B,Table 3).Although a similar tendencywas observedinnetworksbeforenormalizationwithmatchedrandomnet- works,thedifferenceswerenotstatisticallysignificant,p>0.05(Sup- plementaryTableS2).

Networksegregationwasmeasuredbymeansofaverageclustercoef- ficient,localefficiencyandmodularity.IncomparisontoFTnewborns, VPT-TEAinfantsshowedasignificantlyhighernormalizedaverageclus- teringcoefficient,(p=0.001)andsignificantlyhighernormalizedlocal efficiency(Fig.1C,D,Table3).Asimilartendencywasobservedinnet- worksbefore normalizationwithmatched randomnetworks,butthe differenceswerenotstatisticallysignificant,p>0.05(Supplementary TableS2).

BothFTandVPT-TEAinfantspresentedamodulararchitecturein theirbrainnetworks.ThisisreﬂectedbyQ>0.3inbothgroups(Sup- plementary TableS2), whichis indicativeof a non-randommodular structure(NewmanandGirvan,2004).VPT-TEAinfantshadasignif- icantlyhighernQ(p=0.032)andQ(p=0.007)comparedtoFTinfants (Fig.1D,Table3,SupplementaryTableS2).

Thebalancebetweensegregationandintegrationwasquantifiedby thesmall-worldnetworktopology.BothFTandVPT-TEAinfantsbrain networkshadaSW>1,withoutstatisticallysignificantdifferencesbe- tweengroups(Table3),provingthesmall-worldorganizationof the neonatalbraininbothgroups.

3.1.2. Hubsandrich-cluborganization

BothFTandVPT-TEAinfants’structuralbrainnetworksrevealeda right-taileddistributionofnodaldegree,suggestiveoftheexistenceof denselyconnectednodes,hubs(Fig.2AandB).

We found9hubs in FTinfants withanaverage nodaldegree of 51.67,whereasVPT-TEAinfantspresented ahighernumberofhubs, namely16,butwithaninferioraveragenumberofconnections(aver- agenodaldegreeof45.38).Thehubregionsinbothgroupsconsistedof mainlysubcortical(thalamus,caudateandputamen)andlimbic(dorsal cingulategyrus,insulaandhippocampus)nodes.Noticeably,FTnew- bornshadaparietalhub(leftprecuneus)notpresentinVPT-TEAin- fants,whereasVPT-TEAinfantshadtemporalhubsnotpresentinFT newborns(bilateralsuperiortemporalgyrusandleftHeschlgyrus),as wellasbothamygdalaeashubs(Fig.3,SupplementaryTableS3).The followinghubswerediscoveredinVPT-TEAinfants:bilateralthalamus, bilateralcaudate,bilateralputamen,bilateralhippocampus,bilateralin- sula,bilateralamygdala,rightdorsalcingulategyrus,leftHeschlgyrus andbilateralsuperiortemporalgyrus.Thefollowinghubswerepresent inFTnewborns:bilateralthalamus,leftcaudate,rightputamen,right insula,righthippocampus,bilateraldorsalcingulategyrusandleftpre- cuneus(SupplementaryTableS3).

Forbothgroups,leftandrightthalamuswerethetworegionswith highestnodaldegree(SupplementaryTableS3).Whencomparingthe numberofconnectionsofthesetworegions,combined,betweenVPT- TEA andFTinfants, theVPT-TEAshowedastatisticallysigniﬁcantly lowernumberofconnectionsoftheleftandrightthalamusincompari- sontoFTinfants(p=0.025)(Fig.2C).

Normalizedrich-clubcurves,representativeofbothFTandVPT-TEA groups,showanincreasing𝜙normwiththeincreaseof𝜅forbothgroups.

ThisincreaseissuperiorforFTinfants,incomparisontoVPT-TEAin- fants,atthestrongestnetworknodes(from𝜅=45to𝜅=60)(Fig.2E).

BoththeFTandVPT-TEAinfants’brainnetworksdisplayedRCorganiza- tion,with𝜙norm(𝜅)>1for𝜅≥9.Consideringthetop10%strongestnetworknodes,FTinfantspresentedbothastatisticallysigniﬁcantlyhigher 𝜙norm(p=0.007)and𝜙(p=0.003)incomparisontoVPT-TEAinfants (Fig.2D,SupplementaryTableS2).

3.2. Prematurityandtheconnectivitystrengthofstructuralnetworks IncomparisontoFT,theVPT-TEAinfantsbrainnetworkspresented 90 connections (out of the total 4005 unique possible connections)

(7)

Fig.1. GlobalnetworkmeasuresdiﬀerencesbetweenVPT-TEAandFTinfants.BoxplotsofglobalnetworkmeasuresinVPT-TEA(inred)andFTinfants(ingreen):

normalizedcharacteristicpathlength,nL(A),normalizedglobalefficiency,nGEff (B),normalizedclustercoefficient,nC(C),normalizedlocalefficiency,nLEff (D), andnormalizedmodularityindex,nQ(E).One-wayANCOVA,controllingforGAatMRIandsex,wasperformedtodeterminedifferencesbetweengroups.Meanand standarddeviationareillustratedinyellowoneachboxplot.Linesindicatesignificantdifferencesbetweengroups(^∗p<0.05,^∗^∗p<0.01,^∗^∗^∗p<0.001).

Fig.2. Hubs,thalamicconnectivityandrich-cluborganization.AandB:StructuralbrainhubsofFT(A)andVPT-TEA(B)infants.Hubs(inred)weredefinedasnodes whosedegreewas ≥ 1standarddeviationabovetheaveragenodaldegree.CandD:Boxplotsofmeanthalamicconnections(C)andnormalizedrich-clubcoefficient ofthetop10%strongestnetworknodes(𝜙normTop10%)(D)inVPT-TEAandFTinfants.One-wayANCOVA,controllingforGAatMRIandsex,wasperformedto determinedifferencesbetweengroups.Meanandstandarddeviationareillustratedinyellowoneachboxplot.Linesindicatestatisticallysignificantdifferences betweengroups(^∗p<0.05,^∗^∗p<0.01).E:Normalizedrich-clubcurves(𝜙norm)ofFTandVPT-TEAinfant’srepresentativeweightednetworks.

(8)

Fig.3. Brainnetworkconnectionswithsigniﬁcantdiminishedstructuralconnectivityinprematurelybornneonates.A.Representationofbrainnetworkconnectionswitha statisticallysigniﬁcantlydecreasedstructuralconnectivitystrengthinVPT-TEAincomparisontoFTinfants.Intra-hemisphericconnectionsarerepresentedinblue whileinter-hemisphericinpink.VisualizationmadeusingtheCircossoftware(Krzywinskietal.,2009).B.Bargraphrepresentationoftherelativepercentualdistri- butionofcorticalnodesparticipatinginthecortico-corticalandcortico-subcorticalconnectionswithadiminishedconnectivitystrengthinVPT-TEAincomparison toFTinfants.C.Bargraphrepresentingtherelativepercentualdistributionofcortico-corticalconnectionswithadiminishedconnectivitystrengthinVPT-TEAin comparisontoFTinfants.

withastatisticallysigniﬁcantdecreasedstructuralconnectivitystrength (p<0.05,FDRcorrected).Theseincludedbothintra-hemispheric(88%, outofwhich51%intheleftand37%inrighthemisphere)andinter- hemispheric(12%)connections(Fig.3A).Acompletedescriptionofall connectionswithdecreasedstructuralconnectivitystrengthinVPT-TEA canbefoundinSupplementaryTableS4.

Inbothcortico-corticalandcortico-subcorticalconnections,thecor- ticalnodesparticipatingintheconnectionswithdecreasedconnectivity strengthinVPT-TEAinfantsbelongedmostlytothefrontalcortex.In cortico-corticalconnections,thesewerefollowedbytemporal,limbic, parietalandoccipitalcortices,andincortico-subcorticalconnections, bythetemporal,occipitalandlimbiccortices(Fig.3B).

Amongthefronto-corticalandfronto-subcorticalconnectionswith decreasedconnectivitystrengthinVPT-TEAinfants,theorbito-frontal cortex(OFC)wasthefrontalnodethemostfrequentlyinvolved(partic- ipatingin44%ofallfronto-cortical/subcorticalconnections)(Supple- mentaryTableS5andS6).

Fronto-temporalconnectionswerethemostfrequentlyaﬀected,not onlyamongfronto-cortical,butamongallcortico-corticalconnections (Fig.3C,SupplementaryTableS5).Outofthose,93%wereconnecting tothetemporalpole(TPO).Indeed,connectionsfromtheOFCtothe TPOwerethemostfrequent(50%)amongallfrontal-temporalconnec- tionswithdecreasedconnectivitystrengthinVPT-TEA,aswellasamong allfronto-corticalconnections(almost20%of these)(Supplementary TablesS4andS5).Noticeably,theTPOwastheregionmostfrequently involvedinallconnectionswithdecreasedconnectivitystrengthinVPT- TEAinfants,followedbytheOFC(SupplementaryTablesS7).

Thefronto-limbicconnectionswerethesecondmostaﬀectedcortico- corticalconnectionsafterfronto-temporalconnections.Thecingulate gyruswasthelimbicregionmostfrequentlypresent,notonlyintheaf-

fectedfronto-limbicconnections,butamongalllimbicconnectionswith decreasedconnectivitystrength inVPT-TEAinfants,in particularthe posteriorcingulategyrus(PCG)(Fig.3C,SupplementaryTablesS4–S7).

Indeed,thePCGwasthethirdmostfrequentregionamongallconnec- tionswithdecreasedconnectivitystrengthinVPT-TEAinfants,afterthe TPOandOFC(SupplementaryTableS7).

Ofnotice,theprecuneus(PCUN),thefourthmostfrequent region among allconnections withdecreased connectivity strengthin VPT- TEAinfants,wasthemostfrequentregionparticipatinginallparieto- corticalaﬀectedconnections(Supplementary TablesS6andS7).Fur- thermore,theleftPCUNwasthenodethemostfrequentlyinvolvedin inter-hemisphericconnectionswithdecreasedconnectivitystrengthin VPT-TEAinfants,followedbytheleftdorsalcingulategyrus(DCG)(Sup- plementaryTableS4).

4. Discussion

Inthis study, using aconnectomeapproach, we exploredwhole- brainnetworktopologicalalterationsusing graph-based analysisand edge-wise connectivitystrength in ordertocomparebrainstructural connectivityalterationsonVPT-TEAinfantsandFTnewbornsandthus assesstheimpactofprematurityonwhole-brainstructuralorganization atTEA.

Themajorfindings fromthisstudyare:1)VPT-TEAinfants’brain networkspresentedanincreasedcharacteristicpathlength,decreased globalefficiency,increasedclusteringcoefficientandincreasedmodu- larityindex,reflectingincreasedsegregationanddecreasedintegration incomparisontoFTnewborns;2)Alteredconnectivityoftypicalneona- talnetworkhubs,inparticularthethalamus,leftPCUNandleftDCG,as wellasadiminishedrich-clubcoefficientinVPT-TEAinfants’brainnet-

(9)

works;3)Decreasedconnectivityoffronto-paralimbicandfronto-limbic connections,inparticularoftheorbitofrontal-temporalpolecircuitry,as wellasofDMNposteriorregions,suchasthePCUNandPCG,inVPT- TEAinfants’brainnetworksincomparisontoFTinfants.

4.1. Prematurityisassociatedwithanimmaturebrainstructuralnetwork topologicalorganizationatTEA

4.1.1. Increasedsegregationanddecreasedintegrationofbrainnetworksin prematurelybornneonates

Thetopological organizationof thebrain structural networkhas beeninvestigatedinfetalandpreterminfantsthroughoutdevelopment.

FromthesecondtrimestertoTEA,variousstudieshaveshownanin- creaseinnetworkstrength andintegration capacity,withadecrease ofcharacteristicpathlengthandincreaseofnetworkglobalefficiency, bothduringnormalfetal(Songetal.,2017)aswellaspretermbrainde- velopment(Balletal.,2014;Batalle etal.,2017;Brownetal.,2014; van denHeuvel et al., 2015; Zhao et al.,2019). Characteristicpath lengthandnetworkglobalefficiencyrefertotheabilityofcommuni- cationbetweendistributednodesandthusreflecttheglobalinforma- tiontransferefficiency(RubinovandSporns,2010).Indeed,duringfetal braindevelopment,thereisadramaticgrowthofmajorlongassociation WMfibersfrom20to40weeksGA,whichisthoughttounderliethedra- maticincreasesofnetworkstrengthandefficiency(Songetal.,2017).

Sincepretermbirthoccursduringthiscriticalperiodofbraindevelop- ment,itmightimpactnormaltopologicalnetworkorganization,namely bydisruptingtheestablishmentof long-rangestructural connections.

Thiscouldexplainthesignificantlyhighercharacteristicpathlengthand significantlydiminishedglobalefficiencythatwefoundattermagein thebrainnetworksofVPT-TEAinfantsincomparisontoFTnewborns, whatisinlinewithpreviousliteraturefindings(Balletal.,2014;Fischi- Gomezetal.,2015;Leeetal.,2019;Thompsonetal.,2016).

Literaturehassuggested thatthetypicaldevelopmentalcourseof thebrainstructuralnetworkischaracterizedbyaprogressivematura- tionalshiftfromshort-rangetolong-distanceconnections,withacontin- uouslyincreasingintegrationanddecreasingsegregationwithage,fa- voringthusanintegrativetopology(Hagmannetal.,2010;Huangetal., 2015;Tymofiyevaetal.,2013;Yapetal.,2011).Networksegregation relatestoitsorganizationintoacollectionofsub-networksandcanbe measuredbymeansofaverage clusteringcoefficient,local efficiency andmodularity.Duringearlydevelopment,adecreasingclusteringco- efficienthasbeenobservedbothinfetaldevelopment,from20to35 weeksGA(Songetal.,2017),aswellasinpretermdevelopmentbe- tween30and40weeksGA(Balletal.,2014).Inthisstudy,wefound thatVPT-TEAinfantsshowed asignificantlyhigherclustering coeffi- cientandlocalefficiencyincomparisontoFTnewborns,whichconfirms previousreportedfindings(Balletal.,2014)andsuggeststhattheex- pecteddecreaseinbrainnetworksegregationdescribedduringnormal developmentisaffectedbypretermbirth.Innon-normalisednetworks,a tendencyforanincreasedcharacteristicpathlength,diminishedglobal efficiencyandincreasedclusteringcoefficientandlocalefficiencywas alsoobservedin FTnewborns,incomparisontoVPT-TEAinfants,al- thoughnotstatisticallysignificant,whatcanberelatedtothereduced studysamplesize.

Additionally,modularityindexrevealedthatbothVPT-TEAandFT groupshadamodularnetworkorganization,inwhichnodesarehighly connectedtooneanotherwithinthesamecommunityandsparselycon- nectedtonodesinothercommunities(Newman,2004).However,we foundanincreasedmodularityindexinVPT-TEAinfantscomparedto FTnewborns.Modularityhasbeenshowntodecreasealongdevelop- ment,withanincreaseofcommunicationsbetweendiﬀerentmodules frombirthtoadolescence(Huangetal.,2015).Ourresultsindicatefor theﬁrsttimethatprematurityleadstoanincreaseofstructuralbrain networkmodularityatTEA,withnetworknodesholdingmoreconnec- tionswithinnodesfromthesamemodulethanwithnodesfromother

modules,whatsupportstheincreasedsegregationanddecreasedinte- grationobservedinVPT-TEAinfants’brainnetworks.

Anoptimalbalancebetweensegregationandintegrationisidealto allowhighcapacityforlocalandglobalinformationtransfer.Thisas- pectwasquantifiedbythesmall-worldnetworktopology.Nodifferences werefoundbetweenVPT-TEAandFTinfantsregardingtheSWindex, whichwas>1inbothcases,suggestingagenerallyefficientlocaland globalorganizationinbothgroups.

Summing up, ourdata reveals that thebrain of a VPTinfant at TEA,incomparisontoaFTnewborn,evidencesatopologicalorgani- zationcorresponding toa typicalearlier stageof development,with decreased integration and increased segregation properties. Integra- tion ofdistributedneuronal activityacrossspecializedbrain systems is required for higher brainfunction andsupports diverse cognitive processes, suchaslanguage(FriedericiandGierhan,2013),emotion (Pessoa,2012),cognitivecontrol(PowerandPetersen,2013)andlearn- ing(Bassettetal.,2011,2015),skillsknowntobeimpairedininfants thatwerebornpreterm(AndersonandDoyle,2003;Bhuttaetal.,2002; Marlow,2004;MontagnaandNosarti,2016;Spittleetal.,2009,2011; Wittetal.,2014).ThefactthatVPT-TEAinfants’structuralbrainnet- workswerefoundtobemoremodularandlessgloballyefficientthan inFTnewbornsmightresultfromdifferentfactorspotentiallyassoci- atedwithpretermbirth,suchasadisruptionoftheactivity-dependent synapticpruning,ofthegrowthandmaturationofmajorlong-rangeas- sociationWMfibers,oranimpairedreinforcementofthematurationof WMconnections.Suchfactorscouldexplaintheobserveddelayedshift fromalocalemphasis,basedonshort-rangeconnections,toamoredis- tributedbrainnetworkarchitecture,basedonlong-rangeconnections.

Topological characteristics of structural brainnetworks observed earlyin development,includingnetworkintegration andsegregation measures,suchasglobalefficiencyandclusteringcoefficient,havebeen found, respectively,tobe significantlycorrelatedwithcognitiveper- formance(Keunenetal.,2017)andinternalizingandexternalizingbe- havioursassessedinearlychildhood(Weeetal.,2017).Thesefindings supportthehypothesisthatthealteredglobalbrainnetworkorganiza- tionobservedinVPT-TEAinfantsmightunderlietheneurodevelopmen- talimpairmentsreportedinpreterm-borninfantsatlaterages.

4.1.2. Brainnetworkhubsdiﬀerencesandimpairedrich-cluborganization inprematurelybornneonates

Connections within theconnectomecan be distributed unevenly, such that certain nodes possess a larger number of connections than others,which makes them networkhubs (van den Heuvel and Sporns,2013).Brainnetworkhubsplayacentralroleinfacilitatingthe integrationofdisparateneuralsystems,namelybyforminglong-range connections(vandenHeuvelandSporns,2011;vandenHeuveletal., 2012).Structuralhubshavebeenshowntobepresentalreadyduring theprenatalperiodandtohaveaconsistentspatialtopographythrough- outdevelopment(Balletal.,2014;Chenetal.,2013;Hagmannetal., 2010).

Ourresultsconﬁrmthatbytermage,bothFTandVPT-TEAinfants’

brainnetworksdemonstratethepresenceofnetworkhubs,comprising mainlysubcortical(thalamus,caudateandputamen)andlimbic(dorsal cingulategyrus,insulaandhippocampus)regionsinbothgroups.Inter- estingly,theleftPCUNwasdetectedasahubonlyinFTinfants,whereas thebilateralsuperiortemporalgyrus,leftHeschlgyrusandbothamyg- dalaewereidentiﬁedashubsonlyinVPT-TEAinfants.Mostofthesere- gionshavealreadybeenreportedasnetworkhubsinneonatesandmany ofthemhavebeenshowntopersistthroughoutdevelopment(Balletal., 2014;Chenetal.,2013;Huangetal.,2015;Panditetal.,2014;vanden HeuvelandSporns,2011;vandenHeuveletal.,2015).

Fromalldetectednetworkhubs,theleftandrightthalamuswerethe tworegionswiththehighestnodaldegreeinbothVPT-TEAandFTnew- borns.Ofnotice,VPT-TEAinfantspresentedasigniﬁcantlydiminished numberofthalamicconnectionsincomparisontoFTinfants,whichis in agreementwithpreviousliteratureevidencing diminishedthalam-

(10)

ocorticalconnectivityandreduced networkcapacityofDGM followingpretermbirth(Balletal.,2013,2014,2015).Moreover,thisdimin- ishedconnectivitymightcontributetothereducedintegrationcapacity observedinVPT-TEAinfants’brainnetworks.Thalamocorticalconnec- tionsareestablishedduringthelatesecondandthirdtrimesterofdevel- opment,whenpretermbirthoccurs(Kostovicetal.,2002,2014),and thereforeprematurityanditsassociatedevents,suchasinﬂammation, hypoxia/ischemia,stressandimportantenvironmentalchanges,mayaf- fecttheestablishmentoftheseconnections.Thethalamus,asasubcor- ticalhub,hasbeenshowntopresentarich-clubconnectivitypattern,to beinvolvedinlarge-scaleintegrationofneuralsignals(Alexanderetal., 1986; van den Heuvel and Sporns, 2011) and in important higher- ordercognitivefunctions(Bressler,1995;BresslerandMenon,2010).

In particular, animpaired thalamocortical structural connectivity at TEA,followingpreterm birth,hasbeen shown tocorrelatewithim- pairedcognitiveperformanceinprematureinfantsattwoyearsofage (Balletal.,2015).Additionally,atschoolage,alterationsinthecortico- basalganglia-thalamo-corticalloop(namelyprefrontal-subcorticalcir- cuits),followingpretermbirth,wereassociatedwithpoorersocialbe- havior,recognitionofsocialcontextandsimultaneousinformationpro- cessingatschoolage(Fischi-Gomezetal.,2015).

TheleftPCUN,ahubthatweobservedinFTbutnotinVPT-TEA infants,hasalreadybeenshowntobeanetworkhubinthenewborn brain(Huangetal.,2015;Yapetal.,2011).ThePCUNisabrainre- gionsituatedinastrategiclocationwithwide-spreadconnectionsand issuggestedtobeamajorassociationarea,supportingavarietyofbe- haviouralfunctions(CavannaandTrimble,2006),includingmemory retrieval(Lundstrometal.,2005),rewardmonitoring(Haydenetal., 2008),emotionalprocessing(Maddocketal.,2003)andmediationbe- tweentaskandreststates(Lietal.,2019;Utevskyetal.,2014).Italso playsanimportantrolein“defaultmodenetwork” (DMN),functionally linkedtoself-consciousness(CavannaandTrimble,2006;Franssonand Marrelec, 2008),social cognitive functions(Mars et al., 2012),reg- ulation of attentional states andcognition (Leech and Sharp, 2014; McKiernanetal.,2003),andwhichisknowntoappearlaterinthedevel- opmentofpreterminfants(Smyseretal.,2010).Interestingly,resting- statefunctionalMRI (rs-fMRI)hasshownthattheleft PCUNplays a centralroleinintegrationinnewbornsandalsobelongstothenodes hierarchicallymoreimportantforthepropagationofneuralactivityto otherbrainregionsinFTbutnotinVPTinfants,whenat10yearsold (Padillaetal., 2020).Additionally,atTEA,rs-functionalconnectivity betweenthesaliencenetworkandthePCUNhasbeenshowntobesig- nificantlydecreasedinVPT-TEAinfantsincomparisontoFTnewborns (Lordieretal.,2019).Structuralbrainnetworkstudieshaveshownthat thePCUNhasanincreasedefficiencyassociatedwithlongergestational agewhenevaluatedatpreadolescentage(Kimetal.,2014),supporting theimpactofprematurityinthisnodeefficiency,anditpresentsacon- sistentlyhighandincreasingcentralityfromearlyinfancytoadulthood (Chenetal.,2013;Hagmannetal.,2010;Huangetal.,2015).Onthe otherhand,thesuperiortemporalgyrusandtheHeschlgyrus,which weredetectedashubs inVPT-TEA butnot in FTinfants,have been showntohaveadecreasedcentralitywithage,fromneonataltopre- adolescenceperiod(Huangetal.,2015).Suchfindingsarethoughttobe relatedtoaselectivestrengtheningofthegrowingstructuralnetworks, throughbothmicrostructuralenhancementofmajorWMtractsandaxon pruning,basedonageneticsandepigeneticsbackground(Huangetal., 2015).Ourresultssupportthehypothesisofadelayedstructuralmatu- rationinVPT-TEAstructuralbrainnetworks,sincethePCUN,ahubwith increasedcentralityduringdevelopmentfromtermbirthtoadulthood, wasfoundinFTnewbornsbutnotinVPT-TEAinfants,whiletemporal regionsthattend tohaveadecreasedcentralityduringnormalbrain developmentwerestillidentifiedashubsinVPT-TEA,butnotinFTin- fants.AnotherimportantreflectionisthatVPTinfantsareknowntobe exposedtomanynewdifferentacousticsolicitationsfromtheexternal world,incomparisontoFTnewbornsthatremaininsidetheirmother’s wombuntiltermage.Thedifferencesintheacousticenvironmentmight

leadtochangesinactivity-dependentsynapticplasticityandcouldjus- tifythefactthattemporalregionsimportantforsoundprocessing,such astheHeschlgyrusandthesuperiortemporalgyrus(Fulfordetal.,2004; Jardrietal.,2008),werefoundashubsonlyinVPT-TEAinfants.

Remarkably,VPT-TEApresentedasuperiornumberofhubs,with aninferioraveragenumberofconnections,incomparisontoFTnew- borns.Thisﬁndingwasconﬁrmedandcomplementedbythereduced

“rich-club” coefficientobservedinVPT-TEAinfants.A“rich-club” orga- nizationreferstothepresenceofsomehubregionsthatareparticularly denselyconnectedtoeachother(morethanwhatwouldbe expected by chance)andisthought tobe animportantcharacteristicforeffi- cientglobalinformationintegration(vandenHeuvelandSporns,2011; vandenHeuveletal.,2012).Literaturehasshownthatthedevelopmen- talwindowbetween30and40weeksGA,whenpretermbirthoccurs, correspondstoanessentialperiodfortheestablishmentofconnections between RCregionsandtherestofthecortex,contributingtoanin- creaseinRCconnectivity(Balletal.,2014).Ourdatademonstrates,for thefirsttimeinliterature,thatVPT-TEAinfants’structuralbrainnet- workshaveahighernumberofhubs,butpresentadiminishedweighted RCcoefficientincomparisontoFTinfants,revealingthusasignificantly diminishedconnectivityofhighlyconnectedandhighlycentralbrainre- gionsinVPT-TEAvsFTinfants,inagreementwithourfindingthatin VPT-TEAbrainnetworksaremoresegregatedandlessintegratedthan inFTinfants.Balletal.hadfoundnodifferencebetweenVPT-TEAand FTinfantsinthenodaldegreewithintheRCdomain,whenconsidering onlytheconnectionsamongtheRCnodes(Balletal.,2014),butthey haveevaluatedunweightedstructuralbrainnetworks.UsingfMRI,ithas alsobeenshownthatVPT-TEAinfantsexhibitasignificantlyreduced functionalconnectivitywithin RCnodesincomparison toFTinfants (Scheinostetal.,2016).Literaturehasalsoshownthatlongergestation preferentiallyenhancesRCconnectionsandincreasesglobalnetworkef- ficiencyinthepreadolescentbrain(Kimetal.,2014).Additionally,an atypicalRCorganizationhasbeenfoundinvariousclinicaldisordersof neuronalconnectivity,comprisingattention-deficit/hyperactivitydisor- der(ADHD)andautismspectrumdisorders(ASD),frequentlyobserved inthepretermpopulation(Rayetal.,2014).Ourfindingssupportthus notonlythehypothesisthatVPT-TEAinfants’brainmightpresentade- layinearlydevelopment,withanorganizationwhichistypicalofan earlierdevelopmentalstageincomparisontoFTinfants,butalsothat prematurityanditsassociatedeventsmightdisruptearlystructuralcon- nectivityandnetworkorganization.

4.2. Prematurityisassociatedwithdecreasedconnectivitystrengthin structuralbrainnetworksatTEA

In our study, VPT-TEA infants evidenced decreased connectivity strength in a set of interand intra-hemisphericcortico-cortical connections, as well as in a set of intra-hemisphericcortico-subcortical andsubcortico-subcorticalconnectionsincomparisontoFTnewborns.

Theseﬁndingsareinagreementwithotherstructuralconnectivityin- vestigationsinpreterminfants(Balletal.,2014;Batalleetal.,2017).

Amongallcorticalregions,thefrontalcortexwastheoneexhibit- ingthemajorityofconnectionswithdecreasedconnectivitystrengthin VPT-TEAinfants,withtheOFCbeingthefrontalregionthemostfre- quentlyinvolved.Decreasedconnectivityoftheorbitofrontalnetwork hasalreadybeenshowninextremelypreterminfantsat6yearsofage (Fischi-Gomezetal.,2015).LiteraturehasalsoshownthatVPTinfants presentvolumetricreductions,corticalimmaturityandcorticalgyrifica- tionabnormalitiesintheOFCincomparisontoFTinfants(Ganellaetal., 2015; Rogersetal., 2012;Thompsonetal., 2007).These couldbea consequenceofthealteredWMconnectivity,toandfromtheOFC,that wefindin VPTinfantsatTEA,andmightbe attheorigin oftheor- bitofrontalspecificcognitiveandexecutivefunctioningdeficitsreported inpretermpopulation(Luuetal.,2011;Mulderetal.,2009;Taylorand Clark,2016).

(11)

From all the affected connections, the majority were intra- hemisphericcortico-corticaland,fromthese,thefronto-temporalcon- nectionswerethemostfrequent,inparticularthoseconnectingtheOFC totheTPO.Thesetworegionswerealsotheonesparticipatingthemost frequentlyintheconnectionswithdecreasedconnectivitystrengthin VPT-TEA. Boththe OFCand theTPO havebeen considered as par- alimbicregions,based ontheiranatomicallocationandconnectivity (Duvernoy,1999;Hofetal.,1995;Mesulam,2000).Oneofthemost importantWMconnectionsbetweentheOFCandtheTPOistheunci- natefasciculus(Oishietal.,2015;VonDerHeideetal.,2013),whichhas beenproventobeinvolvedinbehavioralandsocio-emotionalprocessing(GhashghaeiandBarbas,2002;Kringelbach,2005;Wildgruberetal., 2005) and to present a decreased maturation in prematurely born neonates,adolescentsandadults(Constableetal.,2008;Mullenetal., 2011; Rimoletal., 2019; Sa deAlmeidaet al.,2019). Afterfronto- temporalconnections,thefronto-limbicconnectionswerethesecond mostaffectedcortico-corticalconnections.Indeed,theseresultsarein agreementwith previous studies, which have identifiedan immatu- rityofcorticalfronto-temporalconnectivitybetweenVPT-TEAandFT infants(Panneket al., 2013),as well asalterations in frontal-basalganglia-thalamocorticalandlimbicnetworksin6year-oldinfantsborn extremelyprematurely,correlatingwithbehaviorandsocio-emotional evaluationscores(Fischi-Gomezetal.,2016).Thus,ourfindingssup- portthehypothesisthatthediminishedconnectivitystrengthoffrontal, paralimbicandlimbicnetworksisalreadypresentinpreterminfantsby term-ageandmightbeassociatedwiththebehavioralsocio-emotional difficultiesobservedinprematurelyborninfantslaterindevelopment (Bhutta et al., 2002; Fischi-Gomez et al., 2015; Hille et al., 2001; Hughesetal.,2002;Lejeuneetal.,2015;MontagnaandNosarti,2016; Spittleetal.,2009;Wittetal.,2014).

AftertheOFCandTPO, thePCG gyruswas theregionthemost frequently involved in the connections with decreased connectivity strength inVPT-TEA, aswell asthemost frequent amongall limbic nodesparticipatinginsuchconnections.ItwasfollowedbythePCUN, whichwasalsothemostfrequentamongallparietalnodesparticipat- ingintheconnectionswithdecreasedconnectivitystrengthinVPT-TEA infants.ThesetworegionsarepartofDMN,whichisassociatedwith cognitivedevelopment(Palaciosetal.,2013;Raichleetal.,2001)and isinvolvedin avarietyofhigh-levelfunctions,suchasattentionand inhibitorycontrol(Fairetal.,2008;Whitﬁeld-GabrieliandFord,2012), socialcognition(Iacobonietal.,2004),episodicmemory(Greiciusand Menon,2004),andself-relatedprocesses(Gusnardetal.,2001).Anal- teredDMNconnectivityhasbeenfoundtobeassociatedwithdevelop- mentaldisorders,suchasADHD(Castellanosetal.,2009;Fairetal., 2010) andASD(Assafetal., 2010; Nielsenet al.,2014), whichare commonlyreportedinpretermpopulation(JohnsonandMarlow,2011; Nosartietal.,2012;Treyvaudetal.,2013).DMNisknowntobeincom- plete/fragmentedandwithaposteriorpredominanceinpreterminfants, evenatTEA,appearinglaterintheirdevelopment(Doriaetal.,2010; Franssonetal.,2007;Lordieretal.,2019;Smyseretal.,2010).Ourre- sultsrevealthatcomponentsfromposteriorDMN,namelythePCGand PCUN,haveadecreased structuralconnectivitystrengthin VPT-TEA infants,incomparisontoFTinfants.Thisdiminishedstructuralconnec- tivitymightunderliethediminishedfunctionalconnectivityobserved inDMNinpreterminfantsatTEAandlaterinlife(Lordieretal.,2019; Smyseretal.,2010;Whiteetal.,2014),aswellasthedeﬁcitsinbe- haviorandsocialcognitionobservedinthispopulation(Andersonand Doyle,2003;ArpiandFerrari,2013;Bhuttaetal.,2002;Marlow,2004; MontagnaandNosarti,2016;Spittleetal.,2009;Wittetal.,2014).

Additionally,theleftPCUNwasthenodethemostfrequently involved in inter-hemispheric connectionswith decreased connectivity strengthin VPT-TEAinfants,followedbytheleftDCG. Interestingly, thesetworegionswerebothdetectedashubsinFTinfants,evidencing thusahighcentralityinthisgroup,butnotinVP-TEAinfants.Inter- hemisphericconnectionsareknowntoplayanimportantroleinlong- rangeintegration.Thefactthattheinter-hemisphericconnectionsthat

passthroughtheleftPCUNandleftDCGhadadecreasedconnectivity strengthinVPT-TEAinfants,incomparisontoFTinfants,mightexplain whythesetworegionsweredetectedashubsonlyinFTinfants,aswell asthedecreasedintegrationcapacitydetectedinVPT-TEAbrainnet- worksincomparisontoFTinfants.

4.3. Limitationsandfuturedirections

Thisstudyhaslimitationsthathavetobeconsidered.First,thesam- plesizeismodest.Thislimitationisrelatedtothedifficultyofrecruit- mentofnewbornsandfamiliesintoaresearchprojectduringastressful timeintheirlife,inparticularforhealthyFTnewborns,whereanMRI hasonlyalimitedclinicalrole.Wearenowrecruitinganewcohortof VPTandFTinfantsforasecondlongitudinalstudy,whichwillallowto furthervalidateourresults.Second,althoughchi-squaredtestrevealed nosignificantdifferenceregardingthedistributionofscanner/coilsbe- tweengroups,this isstillalimitationtobe takenintoaccount.Fur- thermore,topupcorrectionforEPIdistortionswasnotappliedin the presentdataset,giventheacquisitionprotocol.Suchisalimitationthat shouldbeconsideredwheninterpretingtheresultsfromACTtractogra- phyalgorithm.Nevertheless,aswearecomparingtwogroupsofinfants withthesameageatMRI,thepossibleerrorsarrivingfromtractography wouldbesimilarbetweengroupsandthusdatawouldbecomparable.

Additionally,wehaveoptedtouseSIFTforstreamlinesﬁltering,instead ofthenewavailableSIFT2algorithm,mainlyforthefollowingreasons:

1)SIFThasbeenconsideredaseffectiveinachievingitsgoalandhas beenusedinpreviousrecentneonatalconnectomestudies(Batalleetal., 2017;Blesaetal.,2019);2)SIFToffersmoreflexibilityinstreamlines visualization;3)SIFTallowsmoreflexibilityregardingsubsequentanal- ysis,sinceSIFT2requiresanyfollowingprocessingtobecompatiblewith thedefinitionofacross-sectionalmultiplierforeachstreamline(Smith etal.,2015).

Whileourstructuralbrainresultssupportpublisheddataonspecific neurocognitiveandneurobehavioraldeficitsinpreterminfants,theclin- icalsignificanceofthesespecificbrainstructuralalterationsremainsto be furtherevaluated byinvestigatingneurodevelopmentalandcogni- tiveoutcomesinthisstudypopulation.Theseinfantsarecurrentlybeing followed-upinourchilddevelopmentcenterwithdetailedneurodevel- opmentalassessmentsandthefirstresultsareexpectedwithinoneyear.

GiventhespeciﬁcalterationsinnetworkcharacteristicsfoundinVPT infantsatTEA,comparedtoFTinfants,thequestionremainsopenasto theoriginofthesealterations. Inourcurrentlongitudinaldesign,we areinvestigatingifthesebrainnetworkalterationsarepresentinVPT infantsalready atbirthoriftheydevelopthroughtheﬁrstweeksof lifeduringNICU(neonatalintensivecareunit)stay.Thisinformation willbecrucialforthedevelopmentofpotentialinterventionsaimingto preventbrainnetworkalterationsinpreterminfants.

5. Conclusion

A whole-brainconnectomeanalysis was usedtostudybrain networks’diﬀerencesbetweenFTandVPT-TEAinfants.

Overall, ourresults suggestthat, although bothFTandVPT-TEA groups evidencea small-world, modular andrich-club organization, VPT-TEA infants’ brainnetworks display atypically more immature topology,comparedtotheexpecteddevelopmentattermage.Inpar- ticular,theyevidenceincreasedsegregationanddecreasedintegration, andthusadelayedshiftfromalocalemphasistoamoredistributed networkarchitecture.Theimpairedintegrationmightbeexplainedby thedetectedincreasedmodularityindex,diminishedrich-clubcoeﬃ- cientand,inparticular,bytheobserveddecreasedconnectivityoftyp- icalneonatalhubs,suchasbilateralthalamus,leftPCUNandleftDCG.

Fronto-paralimbicandfronto-limbicwerethemostfrequentlyaﬀected connectionsinVPT-TEA.Thenoticeabledecreasedconnectivitystrength of theOFC-TPOcircuitrymightbeat theorigin ofthebehaviorand socio-emotionaldiﬃcultiesassociatedwithprematurityandthusearly