Specialization and integration of functionalthalamocortical connectivity in the human infantHilary Toulmina, Christian F. Beckmannb,c,d, Jonathan O’Muircheartaigha,e, Gareth Balla, Pumza Nongenaf,Antonios Makropoulosa,f, Ashraf Ederiesf, Serena J. Counsella, Nigel Kenneag, Tomoki Arichia,h, Nora Tusora,Mary A. Rutherforda, Denis Azzopardia, Nuria Gonzalez-Cincaf, Joseph V. Hajnala, and A. David Edwardsa,h,1

aCentre for the Developing Brain, Division of Imaging Sciences and Biomedical Engineering, King’s College London, St. Thomas’ Hospital, London SE1 7EH,United Kingdom; bDonders Institute for Brain, Cognition and Behaviour, Radboud University, 6500 HC, Nijmegen, The Netherlands; cDepartment of ClinicalNeuroscience, Radboud University Medical Centre, 6500 HB, Nijmegen, The Netherlands; dOxford Centre for Functional Magnetic Resonance Imaging of theBrain (FMRIB), University of Oxford, Oxford OX3 9DU, United Kingdom; eDepartment of Neuroimaging Sciences, Institute of Psychiatry, King’s CollegeLondon, London SE5 8AF, United Kingdom; fDepartment of Paediatrics, Hammersmith Hospital, Imperial College Healthcare Trust, LondonW12 0HS, UnitedKingdom; hDepartment of Bioengineering, Imperial College London, London SW7 2AZ, United Kingdom; and gNeonatal Unit, St. George’s UniversityHospitals NHS Foundation Trust, London SW17 0NN, United Kingdom

Edited by Marcus E. Raichle, Washington University in St. Louis, St. Louis, MO, and approved April 7, 2015 (received for review December 10, 2014)

Connections between the thalamus and cortex develop rapidlybefore birth, and aberrant cerebral maturation during this periodmay underlie a number of neurodevelopmental disorders. Todefine functional thalamocortical connectivity at the normal timeof birth, we used functional MRI (fMRI) to measure blood oxygenlevel-dependent (BOLD) signals in 66 infants, 47 of whom were athigh risk of neurocognitive impairment because of birth before33 wk of gestation and 19 of whom were term infants. We seg-mented the thalamus based on correlation with functionally definedcortical components using independent component analysis (ICA)and seed-based correlations. After parcellating the cortex usingICA and segmenting the thalamus based on dominant connectionswith cortical parcellations, we observed a near-facsimile of the adultfunctional parcellation. Additional analysis revealed that BOLDsignal in heteromodal association cortex typically had more wide-spread and overlapping thalamic representations than primarysensory cortex. Notably, more extreme prematurity was associatedwith increased functional connectivity between thalamus and lateralprimary sensory cortex but reduced connectivity between thalamusand cortex in the prefrontal, insular and anterior cingulate regions.This work suggests that, in early infancy, functional integrationthrough thalamocortical connections depends on significant func-tional overlap in the topographic organization of the thalamusand that the experience of premature extrauterine life modulatesnetwork development, altering thematuration of networks thoughtto support salience, executive, integrative, and cognitive functions.

resting-state fMRI | thalamus | preterm | functional connectivity | cortex

The formation of topographically organized neural connec-tions between cerebral cortex and thalamus is necessary for

normal cortical morphogenesis (1), and development of theseconnections requires thalamocortical projections to synapse tran-siently in the temporary cortical subplate before penetrating thecortical plate (2–4). In humans, the subplate is at maximal extentin the last trimester of gestation (5), a time of rapid growth forthalamocortical fibers and the cortical dendritic tree, particularlyin heteromodal cortex (6, 7). This process has been shown to bedisrupted by preterm birth (8). Premature delivery is associatedwith increased risk of neurocognitive impairment, and it is widelyhypothesized that abnormal development of brain structure duringthis period is the cause of these problems and may also underliethe development of autistic spectrum disorders and attentiondeficit disorders in genetically predisposed individuals.During the last trimester of pregnancy, functional MRI (fMRI)

detects the emergence of coordinated, spontaneous fluctuationsin the blood oxygen level-dependent (BOLD) signals, which areclosely linked with the development of electroencephalographicactivity (9–11) and develop into a near-facsimile of the matureadult resting-state network architecture by the normal age of birth

at 38–42 wk gestational age (12). However, little is known aboutthe growth of functional connectivity between the thalamus andcortex during this period.Anatomical studies in animals and postmortem adult human

subjects have defined the thalamic microstructure and describeda corticotopic parcellation of the thalamus with precise connec-tivity to specific cortical regions (13, 14). Diffusion tensor imagingstudies have described a similar pattern of structural thalamo-cortical connectivity (15, 16), with evidence in adults that somethalamocortical circuits share common thalamic territory, givingthe potential for integrative functions (17). Functional connec-tivity MRI analysis between the thalamus and the cortex has alsoshown corticotopic organization in the thalamus (18, 19).It is not known, however, when this thalamocortical mapping

develops or how it might be disrupted during development. We,therefore, used connectivity fMRI to address a series of ques-tions. First, is the pattern of dominant thalamocortical connec-tivity at the time of normal birth already similar to the matureadult pattern? Second, in addition to the dominant thalamo-cortical correlations, is there a pattern of overlapping corticalrepresentations in the neonatal thalamus that might reflect de-veloping integration of functional cortical regions? Third, does

Significance

We investigated the way in which the human thalamus andcortex are functionally connected at the time of normal birth.We found the functional parcellation of the thalamus to bea good facsimile of that found in adult studies. However,although primary cortical regions were almost entirely con-nected to specific thalamic regions, heteromodal cortex wasmore widely connected to multiple thalamic regions, giving thepotential for an integrative role for these circuits. Developmentseemed to have been modulated by the experience of pre-mature extrauterine life, with an increase in connectivity toprimary sensory cortex, but reduced connectivity between areasof the thalamus and heteromodal cortex known to support highercognitive functions.

Author contributions: H.T., C.F.B., S.J.C., and A.D.E. designed research; H.T., C.F.B., P.N.,A.E., S.J.C., N.K., T.A., N.T., M.A.R., D.A., N.G.-C., and J.V.H. performed research; H.T.,C.F.B., A.M., and J.V.H. contributed new reagents/analytic tools; H.T., J.O., and A.M. an-alyzed data; and H.T., C.F.B., J.O., G.B., P.N., A.M., A.E., S.J.C., N.K., T.A., N.T., M.A.R., D.A.,N.G.-C., J.V.H., and A.D.E. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. Email: ad.edwards@kcl.ac.uk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1422638112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1422638112 PNAS | May 19, 2015 | vol. 112 | no. 20 | 6485–6490

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Oct

ober

4, 2

021

the experience of preterm delivery and premature extrauterinelife affect the development of thalamocortical connectivity, andis the effect more marked in rapidly developing heteromodalcortex than in more mature primary cortex?

ResultsThe spatial distribution of the whole-brain BOLD signal in thegroup, assessed in all subjects at the time of normal birth (ges-tational age of 38–42 wk) using independent component analysis(ICA), was similar to that described previously (20, 21), includinga combined sensory motor component, auditory, visual, and sub-cortical components, and a fragmented default mode network. Itdid not differ according to gestational age of the infant at birth(Fig. S1). This finding is in accord with previous data (12, 21–23).

Functional Parcellation of the Thalamus at the Time of Normal BirthPresents a Near-Facsimile of Known Organization in the Adult. Hard-thresholding the functional connectivity estimates of nine func-tionally defined cortical areas selected from the group ICA onneuroanatomical grounds (Table 1) revealed a predominantly

symmetrical topographical representation of these cortical re-gions in the thalamus. Fig. 1 shows this topographical organi-zation of the thalamus defined by the cortical component withwhich each voxel was most highly correlated (Movies S1–S9).The primary sensory motor component had the largest terri-

tory of thalamic dominance and provided the dominant corticalconnectivity for the lateral portion of the thalamus with someposterior extension and an extension toward the medial thala-mus. The anterior lateral portion of the thalamus was mosthighly correlated with sensory motor association areas encom-passing the premotor, supplementary motor, and posterior pa-rietal areas and the frontal operculum. The anterior medialthalamus was most strongly connected with the anterior cingu-late. The inferior thalamus at this location was dominantlyconnected with the prefrontal component. Inferiorly, the centralportion of the thalamus was dominantly correlated with thefrontoparietal insular cortex. The medial thalami were domi-nated by connections to prefrontal cortex. The posterior medialextent of the thalamus was most correlated with the temporal

Table 1. Nine functional cortical components defined using independent component analysis (ICA) and suggested regions of dominantthalamic connectivity (the full list of 25 components from ICA is in Fig. S1)

Fig. 1. Dominant thalamocortical correlations from nine functionally defined cortical regions using maximum partial correlation coefficient. (A) Midthalamicview. (B) Posterior view. (C) Inferior view. Lateral parietal component does not have any dominant territory. Table 1 has anatomical descriptions and colors.Major cortical projection areas based on ref. 14. D shows cortical targets from specific thalamic nuclei (14). Images are displayed as per radiological convention.VA, ventral anterior; VL, ventral lateral; VPL, ventral posterior lateral.

6486 | www.pnas.org/cgi/doi/10.1073/pnas.1422638112 Toulmin et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

4, 2

021

cortical component composed of the fusiform gyrus (posteriorpart) and the medial and inferior temporal gyri. The infer-oposterior aspect of the thalamus was most strongly correlatedwith the anterior cingulate component. The posterolateral thal-amus was most strongly correlated with the primary auditorycortex bilaterally and with the primary visual component on theright side only. The lateral parietal component, including thepostcentral gyrus, was not found to have an area of dominantconnectivity with the thalamus.

Specialized and Integrated Thalamocortical Connectivity at the NormalTime of Birth.Using maps of dominant connectivity to characterizethalamic organization underestimates the complexity of thalamo-cortical connectivity by discarding information about shared orintegrated thalamic targets. Consideration of the full distributionof pairwise correlations reveals that, by the time of normal birth,the large-scale neuronal dynamics of the cortex share substantialparts of the anatomical infrastructure of the thalamus (24). Fig. 2shows each cortical component, and its territory of significantthalamic correlations thresholded at a significance of P < 0.05.Visual inspection of these maps shows that thalamocortical con-nectivity naturally divides into components with widespreadconnectivity throughout the thalamus (primary sensory motor,temporal, medial prefrontal cortex, anterior cingulate, andfrontoparietal insular cortex) and components with connectivitythat was mainly limited to their areas of dominance (primaryauditory, primary visual, and lateral parietal) (Fig. 2). Thalamicconnectivity with the visual cortex is included in Fig. 2 but onlyreaches a significance of P < 0.06, possibly because of its location

on the edge of the thalamus. Fig. 3 is a summary of Fig. 2 andshows, for every thalamic voxel, the number of components withsignificantly correlated BOLD activation. The central portion ofthe thalamus has significant connectivity with only two corticalcomponents (sensory motor and frontoparietal insular) joined bya third (anterior cingulate component) in the posteroinferiorportion. The medial territories of both thalami are significantlyconnected with six of nine studied cortical components, with nosignificant correlation with primary visual, primary auditory, orlateral parietal components.

Thalamocortical Connectivity Is Affected by Premature Birth.Of ninecortical areas examined, four showed a difference in correlationwith the thalamus according to the degree of prematurity expe-rienced by the subject. A decrease in magnitude of thalamo-cortical connectivity (P < 0.05) was found in premature infants inthree of nine components investigated (Fig. 4, i–iii). This de-creased connectivity was found between the frontoparietal insularcomponent and a widespread area of the thalamus; reducedconnectivity with prematurity was also found between thalamusand the anterior cingulate and prefrontal components (P < 0.05)(Fig. 4). The difference in connectivity between the anteriorcingulate component and the thalamus was observed within thearea of its dominant thalamic connectivity, whereas the differ-ence between the medial prefrontal component and the thala-mus was detected on the right side only and outside its dominantterritory. Thalamocortical connectivity involving the frontopar-ietal insular cortex, however, was affected by prematurity both inthe territory of its dominant connectivity and also outside it inareas dominantly connected with sensory motor and anteriorcingulate components. The only cortical area where connectivityincreased with prematurity was the lateral parietal component.This small area of the thalamus with increased connectivity isidentified from the segmentation in Fig. 1 as being most con-nected with the sensory motor cortex. There was no relationshipbetween gestational age at birth and motion (r = 0.055, df = 64,P > 0.5) or age at scan and motion (r = 0.064, df = 64, P > 0.5).

DiscussionThis study confirms that, at the time of normal birth, the infantbrain has robust, predominantly symmetrical network architec-ture. We show for the first time, to our knowledge, that, alreadyby the time of normal birth, the topographical organization offunctional thalamocortical connectivity defined by strongestconnectivity is consistent with current information on the adultbrain using tracer methods (25–30), diffusion studies (15, 16),and functional imaging in adult subjects (18). These results addto the growing evidence of the maturity of the human brain bythe time of normal birth (12, 31, 32).Although the topography shown in Fig. 1 is commensurate

with adult thalamic topography derived using other imagingmodalities, there are subtle differences with the adult functionalparcellations produced by Zhang et al. (18) and the tractography

Fig. 2. Areas of thalamus significantly correlated with cortical areas. Partialcorrelations are thresholded at a significance of P < 0.05 (family-wise error cor-rected), except for the primary visual network (lilac), which is shown at P < 0.06(FWE-corrected; in the text). Colors indicate primary sensory motor (pale blue),primary auditory (red), sensory motor association (yellow), primary visual (lilac),temporal (dark blue), prefrontal (orange), lateral parietal (pink), frontoparietalinsular (dark pink), and anterior cingulate (green). Images are displayed as perradiological convention.

Fig. 3. Number of components significantly correlated with activity inthalamic voxels at the time of normal birth. Red (arrow) shows significantconnectivity with two cortical areas (primary sensory motor and frontopar-ietal insular). Orange (♦) connected with three cortical components (primarysensory motor, frontoparietal insular, and anterior cingulate). *Yellow me-dial thalamus shows territory with connectivity to six cortical components.

Toulmin et al. PNAS | May 19, 2015 | vol. 112 | no. 20 | 6487

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Oct

ober

4, 2

021

by Behrens et al. (15), possibly because the initial parcellations inthis study are defined functionally rather than with cortical lobesand combinations of lobes. Studies of longitudinal development ofthalamocortical connectivity have used adult-defined cortical re-gions of interest and analyzed thalamic correlations to these areasin neonates and children rather than using areas of cortical activitythat are coherent at the time of normal birth as in this study. Thismethodological difference, perhaps, accounts for the difference inresults found (33).Although the neonatal thalamus does not have sufficient con-

trast to identify thalamic nuclei from high-resolution structuralMRI and acknowledging that the thalamus changes shape some-what during development, making direct registration of infantresults onto an adult map difficult, we have suggested anatomicallocations and therefore, thalamic nuclei by reference to an adultthalamic atlas (34).A more detailed analysis of thalamic connectivity revealed

overlapping connectivity profiles that share links with multiplecortical territories. The thalamus was differentiated betweenregions with connectivity to multiple cortical components andregions with connectivity to only one or two areas. The thalamicareas connected to multiple cortical regions were not connectedto primary visual, primary auditory, or lateral parietal compo-nents. These different connectivity profiles seem to reflect thefunctional role of the respective thalamocortical units. With theexception of primary sensory motor connectivity, thalamocorticalunits involving primary cortex seemed more restricted than thosewith connections to heteromodal cortex. Given the age of thesubjects, this result may represent a developmental stage, or maybe a feature of mature thalamocortical connectivity. It is notable,however, that analysis of adult data shows that human associa-tion cortices participate in multiple networks and provide func-tionally specialized and flexible regions, whereas somatomotorand early visual cortices participate in single networks (35, 36).Cortical regions with reduced thalamic connections in preterm

infants were areas receiving higher-order thalamic input: circuitsinvolving corticothalamic cortical information. These corticalregions are less mature but develop rapidly during the pretermperiod (37), suggesting their increased vulnerability to prematureextrauterine life and the importance of this developmental

window for the establishment of thalamocortical connections.The regions of frontoparietal, prefrontal, and anterior cingulatecortex affected form the basis of the salience network describedusing task-based functional imaging in adults (38), which may haverelevance to the high incidence of difficulties experienced bychildren born preterm (39), especially with regards to inattentiveattention deficit (40, 41), anxiety, and autistic spectrum disorders(42, 43) that persist into adulthood (44). Areas where we do notsee a difference with prematurity are thalamic areas highly cor-related with primary cortex described as first-order thalamic re-lays, because they are reported to be innervated exclusively andhomogeneously by subcortical drivers, receiving large excitatoryinputs from no secondary sources (45).We also found increased connectivity with prematurity between

the thalamus and a single cortical component, the lateral pa-rietal component, which in adults is involved in processingsignals from face, lips, jaw, tongue, and throat (46). This findingraises the hypothesis that premature exposure to activities, such asbreastfeeding and bottle feeding, may serve to increase functionalconnectivity to regions of cortex with more mature microstructure.This anatomical distribution of areas affected by prematurity

may reflect known developmental programming: microstructuralmaturation of primary sensory cortex seems to occur earlier thanheteromodal cortex (37), and growth association protein, whichis present in growing axons but lost when stable connections aremade (47), is absent from axonal connections to first-order relays(44) but persists in regions of heteromodal cortex, where it maymediate experience-dependent plasticity (48). Cortical areas re-ceiving inputs from first-order thalamic relays also mature ear-liest with regards to myelin formation and cortical thickness (49),whereas thalamocortical units found in this study to be affectedby prematurity involve cortex which takes the longest to reachpeak cortical thickness (50). In older children and young adults,Fair et al. (19) found that the development in motor and sensorythalamocortical interactions showed limited change over timecompared with other thalamic regions. This finding suggests thatprimary thalamocortical interactions are already well-establishedby early childhood.Because the BOLD imaging used in this study is very motion-

sensitive (51), only datasets displaying very minimal motion were

Fig. 4. Thalamocortical components significantly altered by preterm birth. Reduced gestational age at birth was associated with reduced connectivity be-tween thalamus and (i) frontoparietal insular, (ii) anterior cingulate, and (iii) prefrontal cortex and (iv) increased connectivity with lateral parietal cortex.Significance testing using randomization is shown at P < 0.05 (family-wise error corrected) using a general linear model to determine the relationship be-tween gestational age at birth and the partial correlation scores between each cortical component and the thalamus by the time of normal birth. A,Upper, shows cortical components, and B, Lower, shows corresponding areas of different connectivity with the thalamus below. (B) Scatterplots showingthe association between thalamocortical connectivity (vertical axis) for each component and gestational age at birth in weeks (horizontal axis). Imagesare displayed as per radiological convention.

6488 | www.pnas.org/cgi/doi/10.1073/pnas.1422638112 Toulmin et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

4, 2

021

chosen for inclusion in this study. This choice reflects the per-ceived need for lack of motion to accurately study the synchronyof BOLD signals between cortical areas and a structure as smallas the neonatal thalamus. In addition, the premature infantsstudied were without brain lesions to exclude those infants in whomsevere motor impairment could already be predicted from structuralMRI scans. We find effects of prematurity on thalamocorticalconnectivity in our analysis, while acknowledging that our studygroup has been selected in this way.

Materials and MethodsSubjects. The study was reviewed and approved by the National ResearchEthics Service, and all infants were studied with written consent from theirparents. All 66 infants were scanned once at the estimated time of completedgestation (defined as 38–42 wk from the last menstrual period); 47 infantshad been born prematurely (mean = 30 wk, range 24–32 wk), and 19 infantshad been born at full term (mean = 40 wk, range = 36–42 wk). Additionaldetails are in Table S1. All MRI studies were supervised by an experiencedpediatrician, and pulse oximetry, temperature, and electrocardiographydata were monitored throughout; ear protection of silicone-based puttyplaced in the external ear (President Putty, Coltene; Whaledent) and Mini-muffs (Natus Medical Inc.) was used. Parental consent was obtained, andchloral hydrate sedation (25–50 mg/kg) was administered in all but one term-born infant.

High-resolution anatomical scans (T1- and T2-weighted MRI scans) werereviewed by an expert in perinatal MRI: none had major focal destructiveparenchymal lesions, nine of the infants born prematurely had small punctatelesions, which are common in preterm infants and of uncertain significance(52), and one infant had a single small white matter cyst.

Imaging Methods. All images were acquired on a 3-T Philips Achieva MRIScanner. Whole-brain functional imaging was performed using T2* gradientecho planar image acquisition (sequence parameters: repetition time = 1.5 s;echo time = 45 ms; flip angle = 90°; 256 volumes; slice thickness = 3.25 mm; in-plane resolution = 2.5 mm2; 22 slices; scan duration = 6.4 min) with an eight-channel phased array head coil. T2-weighted fast-spin echo MRI was acquiredusing TR = 8,670 ms, TE = 160 ms, flip angle = 90°, slice thickness = 2 mm, fieldof view = 220 mm, matrix = 256 × 256 (voxel size = 0.86 × 0.86 × 1 mm).

Data Selection. Acknowledging the sensitivity of functional data to motion(reviews are in refs. 51 and 53) and with the aim of investigating a smallstructure such as the thalamus, only datasets with very low motion wereeligible for inclusion. Within the premature cohort, 150 datasets were ex-amined, of which 47 met the criteria for inclusion in this study. After headmotion correction with FSL MCFLIRT (54) (and exclusion of the first six vol-umes in every subject), only scans with 200 contiguous volumes with motionof ≤0.08 mm relative mean displacement were included. ICA was then per-formed on individual datasets using FSL MELODIC (55), because an ICA-based identification of artifacts, including head motion, has been shown tobe a very sensitive approach (56, 57). In addition to 47 included subjects, 5additional subjects who met the motion criteria of ≤0.08 mm relative meandisplacement were excluded at this point, because it was found in thesesubjects that there was still significant motion as defined by either compo-nent with spectra in the high-frequency range and/or spatial representationsaround the edge of the brain (58). Nineteen term subjects were selectedusing the same motion criteria, providing a dataset with minimal observedhead motion. There was no difference in motion between the infants bornpreterm and those born at the normal time. Motion parameters based onthe relative mean displacement were 47 infants born pretermmean = 0.052 mm(range = 0.03–0.08 mm) and 19 infants born at the normal time of birthmean = 0.050 mm (range = 0.03–0.08 mm).

Data Analysis.Preprocessing. After removal of nonbrain structures from the T2-weightedstructural image using a neonatal tissue segmentation algorithm (59),fMRI preprocessing was carried out using tools from the Oxford Centre forFunctional Magnetic Resonance Imaging of the Brain Software Library (60).Prestatistical processing consisted of removal of the first six functionalvolumes, correction for head motion (54), spatial smoothing by using aGaussian kernel of FWHM of 5 mm, and high-pass temporal filtering (200 s).Functional volumes were registered to the subject’s T2-weighted structuralimage (54), with boundary-based registration (61) optimized for neonataltissue contrasts. It is not possible to reliably identify single voxels of cere-brospinal fluid from the echo planar image of an infant at the time of

normal birth, which is done in adult studies to model time course regressors,because ventricles are too small to avoid partial volume effects. Therefore,cerebrospinal fluid was identified on the subject’s T2 structural image,and data from voxels corresponding to these areas were discarded. Theremaining data were then transformed to a population-based neonataltemplate (62) using nonlinear registration (60).

To allow for plasticity and acknowledging that cortical areas cannot bedefined from task-based paradigms at this age, cortical regions of interestwere defined using components defined by ICA. Preprocessed functional datacontaining 200 time points per subject were temporally concatenated acrosssubjects to produce a single 4D dataset, and resting-state componentscommon to the group were defined using MELODIC (55) with a fixed di-mensionality of 25, which achieved a good balance between interpretabilityand robustness, similar to that reported in adults (63). ICA maps werethresholded using an alternative hypothesis test based on fitting a Gaussian/γ-mixture model to the distribution of voxel intensities within spatial mapsand controlling the local false discovery rate at P < 0.05 (55). The resultingmaps and full ICA decomposition are shown in Fig. S1.Cortical component selection. Nonoverlapping cortical component masks werecreated by assigning each cortical voxel to a specific resting-state componentdepending onwhich network had the highest z score at that voxel. For analysisof thalamocortical connectivity, we selected nine bilateral cortical areas basedon prior anatomical knowledge and previous work in adults (63), term infants(12), and animals (Table 1). From 25 ICA components, we discarded subcorticalcomponents (thalamus, cerebellum, brainstem, and basal ganglia), compo-nents where the power spectra (in the frequency domain) in individuals was inthe high-frequency range in more than 10% of subjects, and unilateral com-ponents. The exception was the primary visual component: the anatomicalposition of the primary visual cortex leaves it vulnerable to noise caused bydense vasculature and motion, and this component was included, despitebeing characterized by high-frequency power spectra in 7 of 66 subjects(10.6%). Of the remaining components, spatial correlation with adult net-works reported in the work by Smith et al. (63) was tested using cross-corre-lation (Table S2). Visual and auditory components were represented morethan one time: in this case, we selected the network best corresponding withthe anatomical position of primary visual and auditory cortex.

Within a group-defined cortical functional area, there is likely to be someheterogeneity at the subject level. For each individual subject, each com-ponent identified at the group level was mapped back to each subject’sdataset through a spatial regression of the group ICA maps on the individualfMRI dataset followed by a regression of the resulting time series on thesame dataset (64). To ensure that the first Eigen time series at the subject-specific level best represented the function determined by the group anal-ysis rather than another functional area within the same group-definedcortical region, for each subject, the components were thresholded at Z =1.96, and the remaining voxels inside the group-defined mask were used asthe cortical target from which the first Eigen time series was taken. Theresulting component maps in individual subjects derived using this dual-re-gression approach were used as regions of interest, and partial correlationscores were calculated with the time series of each thalamic voxel (65).Group analysis. Correlation scores for each component were combined usingfixed effects analysis, and the results were used to parcellate the thalamus(Fig. 1 and Movies S1–S9), assigning thalamic voxel membership according tothe cortical component with which it had the highest average correlationscore in the group (18). In addition to this fixed effects model to showdominant connectivity, significance (P < 0.05 corrected for multiple com-parisons after threshold-free cluster enhancement) was assessed usingnonparametric permutation testing (66), and results are shown in Fig. 2. Toillustrate which regions of the thalamus are significantly connected withmore than one cortical area, the resulting nine significant thalamic maps(Fig. 2) were binarized and summed (Fig. 3). Finally, using a general linearmodel of gestational age at birth, the thalamocortical correlation mapsshown in Fig. 2 were tested voxelwise for statistically significant associationswith gestational age at birth using nonparametric permutation testing (66).The results were spatial maps characterizing the effect of prematurity onthalamocortical connectivity (Fig. 4A, Upper) and scatterplots showing thecorrelations coefficients in each infant according to gestational age at birth(Fig. 4B, Lower).

ACKNOWLEDGMENTS. The authors thank colleagues in hospitals for recruit-ing subjects for the ePrime Project, neonatal staff and radiographers at QueenCharlotte and Chelsea Hospital Neonatal Intensive Care Unit, Dr. EnitanOgundipe, and the parents and infants participating in this study. We alsoacknowledge the support of the Biomedical Research Centers at ImperialCollege London and King’s College London and Stephen Telfer for technicalhelp. This paper summarizes independent research funded by National

Toulmin et al. PNAS | May 19, 2015 | vol. 112 | no. 20 | 6489

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Oct

ober

4, 2

021

Insitute for Health Research (NIHR) Programme Grants for Applied ResearchProgramme Grant RP-PG-0707-10154 and Medical Research Council StrategicGrant Pre-Term Brain Injury Grant MR/K006355/1. H.T. is funded by Well-come Trust Research Training Fellowship 096039. J.O. is supported by SirHenry Wellcome Postdoctoral Fellowship 096195. The authors acknowledgefinancial support from the Department of Health through an NIHR Compre-hensive Biomedical Research Centre Award (to Guy’s and St. Thomas’

National Health Service (NHS) Foundation Trust in partnership with King’sCollege London and King’s College Hospital NHS Foundation Trust). Adultdata used in Table S2 (63) were made available by Washington University,University of Minnesota and Oxford University (the WU-Minn Human Con-nectome Project consortium) 1U54MH091657 funded by 16 National Instit-utes of Health Institutes and Centers that support the NIH Blueprint forNeuroscience Research.

1. Constantinople CM, Bruno RM (2013) Deep cortical layers are activated directly bythalamus. Science 340(6140):1591–1594.

2. Allendoerfer KL, Shatz CJ (1994) The subplate, a transient neocortical structure: Itsrole in the development of connections between thalamus and cortex. Annu RevNeurosci 17:185–218.

3. Bystron I, Blakemore C, Rakic P (2008) Development of the human cerebral cortex:Boulder Committee revisited. Nat Rev Neurosci 9(2):110–122.

4. Kostovi�c I, Judas M (2010) The development of the subplate and thalamocorticalconnections in the human foetal brain. Acta Paediatr 99(8):1119–1127.

5. Kostovi�c I, Judaš M, Sedmak G (2011) Developmental history of the subplate zone,subplate neurons and interstitial white matter neurons: Relevance for schizophrenia.Int J Dev Neurosci 29(3):193–205.

6. Huttenlocher PR, Dabholkar AS (1997) Regional differences in synaptogenesis inhuman cerebral cortex. J Comp Neurol 387(2):167–178.

7. Travis K, Ford K, Jacobs B (2005) Regional dendritic variation in neonatal humancortex: A quantitative Golgi study. Dev Neurosci 27(5):277–287.

8. Malik S, et al. (2013) Neurogenesis continues in the third trimester of pregnancy andis suppressed by premature birth. J Neurosci 33(2):411–423.

9. Goldman RI, Stern JM, Engel J, Jr, Cohen MS (2002) Simultaneous EEG and fMRI of thealpha rhythm. Neuroreport 13(18):2487–2492.

10. Omidvarnia A, Fransson P, Metsäranta M, Vanhatalo S (2014) Functional bimodalityin the brain networks of preterm and term human newborns. Cereb Cortex 24(10):2657–2668.

11. Li H, et al. (2013) Laminar and columnar development of barrel cortex relies onthalamocortical neurotransmission. Neuron 79(5):970–986.

12. Doria V, et al. (2010) Emergence of resting state networks in the preterm humanbrain. Proc Natl Acad Sci USA 107(46):20015–20020.

13. Jones EG (2007) The Thalamus (Cambridge Univ Press, Cambridge, UK), 2nd Ed.14. Sherman SM, Guillery R (2013) Functional Connections of Cortical Areas: A New View

from the Thalamus (MIT Press, Cambridge, MA).15. Behrens TEJ, et al. (2003) Non-invasive mapping of connections between human

thalamus and cortex using diffusion imaging. Nat Neurosci 6(7):750–757.16. Counsell SJ, et al. (2007) Thalamo-cortical connectivity in children born preterm

mapped using probabilistic magnetic resonance tractography. Neuroimage 34(3):896–904.

17. Draganski B, et al. (2008) Evidence for segregated and integrative connectivity pat-terns in the human Basal Ganglia. J Neurosci 28(28):7143–7152.

18. Zhang D, et al. (2008) Intrinsic functional relations between human cerebral cortexand thalamus. J Neurophysiol 100(4):1740–1748.

19. Fair DA, et al. (2010) Maturing thalamocortical functional connectivity acrossdevelopment. Front Syst Neurosci 4:10.

20. Smyser CD, Snyder AZ, Neil JJ (2011) Functional connectivity MRI in infants: Explo-ration of the functional organization of the developing brain. Neuroimage 56(3):1437–1452.

21. Fransson P, Aden U, Blennow M, Lagercrantz H (2011) The functional architecture ofthe infant brain as revealed by resting-state fMRI. Cereb Cortex 21(1):145–154.

22. Damaraju E, et al. (2010) Resting-state functional connectivity differences in pre-mature children. Front Syst Neurosci 4:4.

23. Smyser CD, et al. (2010) Longitudinal analysis of neural network development inpreterm infants. Cereb Cortex 20(12):2852–2862.

24. Friston KJ (1998) Modes or models: A critique on independent component analysis forfMRI. Trends Cogn Sci 2(10):373–375.

25. Goldman-Rakic PS, Porrino LJ (1985) The primate mediodorsal (MD) nucleus and itsprojection to the frontal lobe. J Comp Neurol 242(4):535–560.

26. Morel A, Magnin M, Jeanmonod D (1997) Multiarchitectonic and stereotactic atlas ofthe human thalamus. J Comp Neurol 387(4):588–630.

27. Rouiller EM, Tanne J, Moret V, Boussaoud D (1999) Origin of thalamic inputs to theprimary, premotor, and supplementary motor cortical areas and to area 46 in ma-caque monkeys: A multiple retrograde tracing study. J Comp Neurol 409(1):131–152.

28. Niemann K, Mennicken VR, Jeanmonod D, Morel A (2000) The Morel stereotactic atlasof the human thalamus: Atlas-to-MR registration of internally consistent canonicalmodel. Neuroimage 12(6):601–616.

29. López-Bendito G, Molnár Z (2003) Thalamocortical development: How are we goingto get there? Nat Rev Neurosci 4(4):276–289.

30. Price DJ, et al. (2006) The development of cortical connections. Eur J Neurosci 23(4):910–920.

31. Ball G, et al. (2014) Rich-club organization of the newborn human brain. Proc NatlAcad Sci USA 111(20):7456–7461.

32. van den Heuvel MP, et al. (May 15, 2014) The neonatal connectome during pretermbrain development. Cereb Cortex, 10.1093/cercor/bhu095.

33. Alcauter S, et al. (2014) Development of thalamocortical connectivity during infancyand its cognitive correlations. J Neurosci 34(27):9067–9075.

34. Krauth A, et al. (2010) A mean three-dimensional atlas of the human thalamus:Generation from multiple histological data. Neuroimage 49(3):2053–2062.

35. Yeo BTT, Krienen FM, Chee MWL, Buckner RL (2013) Estimates of segregation andoverlap of functional connectivity networks in the human cerebral cortex. Neuro-image 88C:212–227.

36. Yeo BT, et al. (September 23, 2014) Functional specialization and flexibility in humanassociation cortex. Cereb Cortex, 10.1093/cercor/bhu217.

37. Ball G, et al. (2013) Development of cortical microstructure in the preterm humanbrain. Proc Natl Acad Sci USA 110(23):9541–9546.

38. Seeley WW, et al. (2007) Dissociable intrinsic connectivity networks for salience pro-cessing and executive control. J Neurosci 27(9):2349–2356.

39. Bhutta AT, Cleves MA, Casey PH, Cradock MM, Anand KJS (2002) Cognitive and be-havioral outcomes of school-aged children who were born preterm: A meta-analysis.JAMA 288(6):728–737.

40. Anderson PJ, et al.; Victorian Infant Collaborative Study Group (2011) Attentionproblems in a representative sample of extremely preterm/extremely low birthweight children. Dev Neuropsychol 36(1):57–73.

41. Brogan E, et al. (2014) Inattention in very preterm children: Implications for screeningand detection. Arch Dis Child 99(9):834–839.

42. Treyvaud K, et al. (2013) Psychiatric outcomes at age seven for very preterm children:Rates and predictors. J Child Psychol Psychiatry 54(7):772–779.

43. Kuzniewicz MW, et al. (2014) Prevalence and neonatal factors associated with autismspectrum disorders in preterm infants. J Pediatr 164(1):20–25.

44. Aarnoudse-Moens CSH, Weisglas-Kuperus N, van Goudoever JB, Oosterlaan J (2009)Meta-analysis of neurobehavioral outcomes in very preterm and/or very low birthweight children. Pediatrics 124(2):717–728.

45. Rovó Z, Ulbert I, Acsády L (2012) Drivers of the primate thalamus. J Neurosci 32(49):17894–17908.

46. Purves D, et al., eds (2011) Neuroscience (Sinauer, Sunderland, MA), 5th Ed.47. Feig SL (2005) The differential distribution of the growth-associated protein-43 in first

and higher order thalamic nuclei of the adult rat. Neuroscience 136(4):1147–1157.48. Benowitz LI, Routtenberg A (1997) GAP-43: An intrinsic determinant of neuronal

development and plasticity. Trends Neurosci 20(2):84–91.49. Gogtay N, et al. (2004) Dynamic mapping of human cortical development during

childhood through early adulthood. Proc Natl Acad Sci USA 101(21):8174–8179.50. Shaw P, et al. (2006) Longitudinal mapping of cortical thickness and clinical outcome

in children and adolescents with attention-deficit/hyperactivity disorder. Arch GenPsychiatry 63(5):540–549.

51. Murphy K, Birn RM, Bandettini PA (2013) Resting-state fMRI confounds and cleanup.Neuroimage 80:349–359.

52. Rutherford MA, et al. (2010) Magnetic resonance imaging of white matter diseases ofprematurity. Neuroradiology 52(6):505–521.

53. Van Dijk KRA, Sabuncu MR, Buckner RL (2012) The influence of head motion onintrinsic functional connectivity MRI. Neuroimage 59(1):431–438.

54. Jenkinson M, Bannister P, Brady M, Smith S (2002) Improved optimization for therobust and accurate linear registration and motion correction of brain images.Neuroimage 17(2):825–841.

55. Beckmann CF, DeLuca M, Devlin JT, Smith SM (2005) Investigations into resting-stateconnectivity using independent component analysis. Philos Trans R Soc Lond B Biol Sci360(1457):1001–1013.

56. Salimi-Khorshidi G, et al. (2014) Automatic denoising of functional MRI data:Combining independent component analysis and hierarchical fusion of classi-fiers. Neuroimage 90:449–468.

57. Griffanti L, et al. (2014) ICA-based artefact removal and accelerated fMRI acquisitionfor improved resting state network imaging. Neuroimage 95:232–247.

58. Beckmann CF (2012) Modelling with independent components. Neuroimage 62(2):891–901.

59. Makropoulos A, et al. (2014) Automatic whole brain MRI segmentation of thedeveloping neonatal brain. IEEE Trans Med Imaging 33(9):1818–1831.

60. Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW, Smith SM (2012) FSL.Neuroimage 62(2):782–790.

61. Greve DN, Fischl B (2009) Accurate and robust brain image alignment using boundary-based registration. Neuroimage 48(1):63–72.

62. Kuklisova-Murgasova M, et al. (2011) A dynamic 4D probabilistic atlas of thedeveloping brain. Neuroimage 54(4):2750–2763.

63. Smith SM, et al. (2012) Temporally-independent functional modes of spontaneousbrain activity. Proc Natl Acad Sci USA 109(8):3131–3136.

64. Filippini N, et al. (2009) Distinct patterns of brain activity in young carriers of theAPOE-epsilon4 allele. Proc Natl Acad Sci USA 106(17):7209–7214.

65. O’Reilly JX, Beckmann CF, Tomassini V, Ramnani N, Johansen-Berg H (2010) Distinctand overlapping functional zones in the cerebellum defined by resting state func-tional connectivity. Cereb Cortex 20(4):953–965.

66. Nichols TE, Holmes AP (2002) Nonparametric permutation tests for functionalneuroimaging: A primer with examples. Hum Brain Mapp 15(1):1–25.

6490 | www.pnas.org/cgi/doi/10.1073/pnas.1422638112 Toulmin et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

4, 2

021