Characterizing effects of age, sex and psychosis symptoms on thalamocortical functional connectivity in youth

Abstract

The thalamus is composed of multiple nuclei densely connected with the cortex in an organized manner, forming parallel thalamocortical networks critical to sensory, motor, and cognitive functioning. Thalamocortical circuit dysfunction has been implicated in multiple neurodevelopmental disorders, including schizophrenia, which also often exhibit sex differences in prevalence, clinical characteristics, and neuropathology. However, very little is known about developmental and sex effects on thalamocortical networks in youth. The present study characterized the effects of age, sex and psychosis symptomatology in anatomically constrained thalamocortical networks in a large community sample of youth (n = 1100, aged 8–21) from the Philadelphia Neurodevelopmental Cohort (PNC). Cortical functional connectivity of seven anatomically defined thalamic nuclear groups were examined: anterior, mediodorsal, ventral lateral, ventral posterolateral, pulvinar, medial and lateral geniculate nuclear groups. Age and sex effects were characterized using complementary thalamic region-of-interest (ROI) to cortical ROI and voxel-wise analyses. Effects of clinical symptomatology were analyzed by separating youth into three groups based on their clinical symptoms; typically developing youth (n = 298), psychosis spectrum youth (n = 320), and youth with other psychopathologies (n = 482). As an exploratory analysis, association with PRIME scores were used as a dimensional measure of psychopathology. Age effects were broadly characterized by decreasing connectivity with sensory/motor cortical areas, and increasing connectivity with heteromodal prefrontal and parietal cortical areas. This pattern was most pronounced for thalamic motor and sensory nuclei. Females showed greater connectivity between multiple thalamic nuclear groups and the visual cortex compared to males, while males showed greater connectivity with the inferior frontal and orbitofrontal cortices. Youth with psychosis spectrum symptoms showed a subtle decrease in thalamic connectivity with the premotor and prefrontal cortices. Across all youth, greater PRIME scores were associated with lower connectivity between the prefrontal cortex and mediodorsal thalamus. By characterizing typical development in anatomically constrained thalamocortical networks, this study provides an anchor for conceptualizing disruptions to the integrity of these networks observed in neurodevelopmental disorders.

1. Introduction

The thalamus is composed of multiple nuclei densely connected with the cortex. Each nucleus has distinct neuroanatomical connections and functions, and can be broadly segregated into sensory/motor and association nuclei based on their neuroanatomical connections (Jones, 2007). Sensory/motor nuclei are connected with sensory organs and primary sensory/motor cortices, while association nuclei are connected to multiple association cortical areas, such as prefrontal and posterior parietal cortices. Historically the thalamus has been conceptualized as a passive relay center, based largely on the organization of sensory systems. However, this view has changed—there is growing evidence that the thalamus, particularly association nuclei, are critically involved in mechanisms of cognitive function, including working memory, cognitive flexibility, and attentional control (Bolkan et al., 2017; Parnaudeau et al., 2013, 2015; Rikhye et al., 2018; Saalmann et al., 2012; Schmitt et al., 2017; Wimmer et al., 2015; Zhou et al., 2016).

The thalamus has been implicated in several psychiatric illnesses, mostly notably schizophrenia. Thalamic dysfunction in schizophrenia is observed in markers of neuronal integrity, reduced thalamus volume, and altered thalamic function during cognitive tasks (Anticevic et al., 2014; Huang et al., 2019, 2020; Kraguljac et al., 2012; Minzenberg et al., 2009; Ragland et al., 2009; van Erp et al., 2016; Woodward et al., 2012). Abnormal functional connectivity of the thalamus, often characterized by a distinct pattern of prefrontal–thalamus hypoconnectivity and sensory/motor–thalamus hyperconnectivity, is one of the most consistent findings in functional neuroimaging studies of schizophrenia (Giraldo-Chica and Woodward, 2017; Ramsay, 2019). Thalamocortical dysconnectivity is present in both non-affective psychosis (i.e. schizophrenia spectrum disorders) and affective psychosis (i.e. bipolar disorder with psychotic features) (Anticevic et al., 2014; Woodward and Heckers, 2016), suggesting that it may be a neurobiological phenotype that is present across the psychosis spectrum.

Evidence of abnormal thalamocortical functional connectivity in early stage psychosis (Woodward and Heckers, 2016) and high-risk individuals (Anticevic et al., 2015; Colibazzi et al., 2017; Dean et al., 2018; Zhu et al., 2019) is consistent with a neurodevelopmental hypothesis of schizophrenia, which posits thalamic abnormalities result from atypical brain development (Andreasen et al., 1998; Cronenwett and Csernansky, 2010; Jones, 1997; Pergola et al., 2015; Sim et al., 2006; Steullet, 2019; Swerdlow, 2010). Many signs of brain dysfunction observed in psychosis manifest during childhood and early adolescence, years before the onset of florid psychosis, and identifying these neurobiological changes allows earlier interventions in those at risk. Youth showing subclinical psychosis symptoms are at elevated risk of converting to a full-blown psychotic disorder, and even those that do not convert demonstrate impaired cognition, elevated mood symptoms and anxiety, higher rates of substance abuse, and reduced global functioning (Marenco and Weinberger, 2000; Seidman et al., 2016). However, neurodevelopmental explanations for thalamocortical dysconnectivity face a significant challenge: relatively little is known about the normal development of thalamocortical functional connectivity. Based largely on an early study of 52 healthy individuals that found greater prefrontal–thalamus connectivity and lower sensory/motor–thalamus connectivity in adults compared to children (Fair et al., 2010), we hypothesized that the pattern of lower frontal and exaggerated somatomotor connectivity with the thalamus in schizophrenia was a consequence of disrupted maturation during adolescence (Woodward et al., 2012). However, a recent study of 95 individuals aged 5–25 years found that except for the cortical dorsal attention network, thalamus connectivity with the cortex generally decreases with age, or remains largely stable (Steiner et al., 2020). The somewhat discrepant findings across studies, which included relatively few individuals, underscore the need for more investigation of age and sex effects using relatively large sample sizes.

Our aims in this study were to: (1) chart age effects of thalamocortical functional connectivity from childhood to young adulthood; (2) characterize sex differences in thalamocortical functional connectivity in youth; and (3) determine whether youth with psychosis spectrum symptoms demonstrate abnormalities in thalamocortical functional connectivity compared to healthy youth and youth with other psychopathologies.

2. Methods

2.1. Study participants

Participants were part of the Philadelphia Neurodevelopmental Cohort (PNC) obtained from the database of Genotypes and Phenotypes (dbGaP). We used the most recent release of the PNC (Study Accession phs000607.v3.p2), consisting of 9498 youth (aged 8–21) including the full neuroimaging sample (n = 1601). Of the 1397 with completed resting state data, 57 were excluded for serious medical conditions, 58 for insufficient clinical data to reach a diagnosis, and 182 for scan quality (e.g. motion, structural image segmentation, normalization). See Table 1 for demographics of our final sample including 1100 youth. Using similar procedures as previously reported (Huang et al., 2020; Roalf et al., 2017), we classified youth into typically developing (n = 298), psychosis spectrum (n = 320), and youth with other psychopathologies (n = 482).

Table 1.

Demographics of study participants.

	Typically Developing (N = 298)		Psychosis Spectrum (N = 320)		Other Psychopathology (N = 482)		Statistics
							df	X2 /t/F	p	Post-Hoc
Gender (F:M)	149:149		179:141		267:215			2.78	.249	–
Ethnicity (W:AA:O)	168:99:29		99:179:41		241:198:41			47.74	< 0.001	–
	Mean	SD	Mean	SD	Mean	SD
Age	14.8	3.6	16.1	2.9	15.5	3.3	2,1097	11.92	< 0.001	PS > OP > TD
Education	7.7	3.6	8.4	2.6	8.5	3.2	2,1097	5.43	.004	PS,OP > TD
Parental Education Neuropsychological Functioning	14.5	2.4	13.6	2.2	14.2	2.3	2,1091	13.28	< 0.001	TD,OP > PS
Global Cognition (z-score)	0.12	0.51	0.05	0.51	0.16	0.50	2,1097	4.11	.017	OP > PS
WRAT (standard score)	105.6	15.9	99.2	17.0	102.9	16.1	2,1097	11.86	< 0.001	TD > OP > PS

Abbreviations: AA = African American; F = Female; M = Male; O =Other; W = White; WRAT = Wide Range Assessment Test.

Clinical assessments included structured interviews and rating scales previously described (Calkins et al., 2014). All participants underwent clinical assessment with the GOASSESS, a computerized structured clinical interview, collecting demographic data, a timeline of live events, medical history data and including an abbreviated version of the Schedule for Affective Disorders and Schizophrenia for School-Age Children (K-SADS), which assesses a range of psychopathology symptoms. Psychosis symptomatology was assessed with the PRIME Screen-Revised (PS-R), K-SADS psychosis screen, and select items from the Scale for Prodromal Syndromes (SOPS), assessing for negative or disorganized symptoms. Psychosis spectrum youth were identified based on (1) having an age-deviant total PRIME score ≥2 SD above age-matched peers, or rating 1 item at >6, or 3 items >5, (2) endorsing definite or possible hallucinations or delusions on the K-SADS and (3) having an age-deviant total SOPS score ≥2 SD above age-matched peers. Youth with other psychopathologies were identified as youth endorsing sub-threshold psychopathology symptoms and duration consistent with SCID-IV diagnosis on the K-SADS and reporting significant distress, but did not meet psychosis spectrum criteria. Typically developing youth were identified as youth who did not meet either psychosis spectrum or other psychopathologies criteria. See Supplemental Materials for full description of the clinical characteristics.

Study procedures were approved by the institutional review boards at the University of Pennsylvania and the Children’s Hospital of Philadelphia. Written informed consent was obtained from all adult participants and assent was obtained from minors as well as written informed consent from their parent or guardian.

2.2. Neuroimaging data and acquisition and preprocessing

Neuroimage data storage and processing took place on the Vanderbilt University Institute of Imaging Science Center for Computational Imaging XNAT (Harrigan et al., 2016; Huo et al., 2018). The processing pipelines are accessible through github (https://github.com/baxpr/freesurfer-singularity; https://github.com/baxpr/connprep; https://github.com/baxpr/fmriqa; https://github.com/baxpr/fsthalconnMNI-public) and were containerized using Singularity and built at SingularityHub (Sochat et al., 2017) (https://singularity-hub.org).

2.2.1. Neuroimaging acquisition

Structural scan: High resolution T1-weighted structural scans were acquired with the MPRAGE sequence on a Siemens Tim Trio 3T scanner with a 32-channel head coil with 0.93 × 0.93 × 1 mm voxels (160 slices, TR/TE = 1810/3.5, FOV = 180 × 240 × 160, matrix = 192 × 256 × 160; flip angle = 9°).

Resting state scan: A 6:18 min T2^∗-weighted resting state BOLD scan was acquired for each participant with a single-shot, interleaved, multi-slice, gradient-echo EPI sequence. 124 volumes were acquired with a voxel resolution 3mm³ isotropic (46 slices, TR/TE = 3000/32, FOV = 192 × 192, matrix = 64 × 64, flip angle = 90°).

2.2.2. Neuroimaging preprocessing

Anatomical images were segmented into gray matter, white matter and cerebrospinal fluid (CSF) with the Computational Anatomy Toolbox 12 (CAT12, version 12.5; http://www.neuro.uni-jena.de/cat/). Resting-state scans were preprocessed in SPM12 and were (1) realigned to a mean scan, (2) coregistered with the native space structural scan, then (3) underwent resting-state denoising procedures: bandpass filter (0.01–0.1 Hz), regression of CSF and white matter signal, regression of 12 motion parameters (6 translation and rotation parameters and their first derivative). All resting state scans went through a quality assurance procedure that included calculating framewise displacement (FD) and temporal signal to noise ratio (tSNR). Scans with a mean FD > 0.5 or a tSNR lower than the 5th percentile of the distribution of the entire sample were excluded from further analysis.

2.3. Regions-of-interest definition

Regions-of-interest (ROIs) used in the functional connectivity analyses described below were defined with FreeSurfer 6 development version including the thalamus segmentation module (Iglesias et al., 2018) using standard parameters (Dale et al., 1999; Fischl et al., 2002). Cortical ROIs were defined for each participant based on the Desikan-Killiany Atlas/DKT atlas. Seven cortical ROIs were created for each hemisphere: prefrontal, parietal, temporal, motor (including the premotor and supplementary motor regions), somatosensory, visual cortices and the hippocampus (see Supplementary Fig. 1 and Supplementary Table 2). Using the thalamus segmentation module from FreeSurfer, we created 7 nuclear groups for each hemisphere: anterior, mediodorsal, pulvinar, ventral lateral (including ventrolateral and ventral anterior nuclei), ventral posterolateral, medial and lateral geniculate nuclear groups (see Supplementary Fig. 2 and Supplementary Table 3). To complement the main analyses using structurally defined thalamic seeds, we used a set of functionally defined thalamic seeds based on a previous study with an independent developmental sample (Woodward et al., 2017).

2.4. Functional connectivity analyses

In order to enhance scientific rigor, comprehensively map thalamocortical connectivity at multiple levels (i.e. cortex, thalamus), and facilitate comparisons of our results with prior studies of thalamic functional connectivity development and abnormalities in psychotic disorders, two approaches were used to examine thalamocortical connectivity. First, to maximize statistical power and ensure we are examining thalamocortical networks with a strong anatomical connectivity basis, we performed ROI-to-ROI analyses that quantified functional connectivity between structurally defined thalamic nuclear groups and cortical regions that are their primary source of anatomical connections (e.g. the mediodorsal nucleus with the prefrontal cortex). We complemented our primary analysis with a voxel-wise analysis of the cortical connectivity pattern with each of our thalamic nuclear groups. These two methods are described in detail below.

2.4.1. ROI-to-ROI analyses

Seven predefined thalamocortical networks were created by pairing each thalamic nuclear group with their primary anatomically connected cortical region. These networks were (1) prefrontal–mediodorsal (2) motor–ventral lateral (3) somatosensory–ventral posterolateral (4) temporal–medial geniculate (5) parietal–pulvinar (6) occipital–lateral geniculate (7) hippocampus–anterior nuclear groups (Behrens et al., 2003; Johansen-Berg et al., 2005; Zhang and Li, 2010). Connectivity (measured as Fisher’s Z) for each of these networks were calculated as follows. Time-courses were extracted from the denoised individual subject space resting state scans for each thalamic and cortical ROI, separately for the left and right hemispheres. For each cortical ROI, connectivity with the thalamic ROI from the same hemisphere was calculated, then connectivity for each cortical ROI–thalamic ROI pair was averaged across left and right hemispheres to obtain a bilateral connectivity value. For validation purposes, we plotted the full connectivity between each cortical with each thalamic ROI (see Supplementary Fig. 3). Statistical analyses in predefined thalamocortical networks were conducted in SPSS (Version 27), as these analyses were repeated over 7 thalamocortical networks, critical alpha was set at a Bonferroni corrected alpha of 0.007.

2.4.2. Thalamic seed-based voxel-wise analyses

Seven separate seed-based voxel-wise functional connectivity analyses were conducted for each of the six thalamic ROIs in SPM12. For each thalamic ROI, the seed time-course was extracted from the thalamic ROI in the denoised individual subject space (native space) resting state scans. Deformation parameters from normalizing T1 structural scans were applied to the resting state scans. The seed time-course was then correlated with every voxel in the denoised normalized resting state scans. Smoothing (6 mm FWHM Gaussian filter) was applied to each resulting connectivity image. Finally, for each thalamic nuclear group, a bilateral connectivity map was created by averaging connectivity values of each voxel across the left and right thalamic ROI for that nuclear group. Cluster significance was determined at p < .05 familywise error (FWE) corrected clusters with a cluster forming threshold of p < .001.

2.5. Statistical analyses

2.5.1. Age effects

In ROI-to-ROI analyses, age effects in thalamocortical networks were modeled as linear and quadratic effects of age using the following linear regression models separately for each thalamocortical network, or thalamus seed:

Linear model

$$Connectivity={\beta }_0+{\beta }_{1}Age+{\beta }_{2}Sex+{\beta }_{3}motion+\epsilon$$ (1)

Quadratic model

$$Connectivity={\beta }_0+{\beta }_{1}Age+{\beta }_{2}Ag{e}^2+{\beta }_{3}Sex+{\beta }_{4}motion+\epsilon$$ (2)

For all models investigating the effects of age, sex and motion (FD) were included as covariates of no interest. For voxel-wise analyses, one model was applied per voxel, then corrected for multiple comparisons at the cluster level.

2.5.2. Sex effects

In ROI-to-ROI analyses, sex effects were investigated using an omnibus Analysis of Covariance (ANCOVA) with each thalamo-cortical network included as repeated measures, and sex as the independent variable, including age, age² and motion (FD) as covariates of no interest, followed by one-way univariate ANCOVAs for each thalamocortical network with sex as the independent variable and age, age² and motion (FD) as covariates of no interest.

ANCOVAs were conducted separately for whole-brain voxel-wise connectivity from each of the 7 thalamus seeds, with sex as an independent variable and age, age² and motion (FD) included as covariates of no interest, as described in the following equation.

$$Connectivity={\beta }_0+{\beta }_{1}Sex+{\beta }_{2}Age+{\beta }_{3}Ag{e}^2+{\beta }_{4}motion+\epsilon$$ (3)

2.5.3. Effects of clinical symptomatology

To investigate the association of clinical symptomatology with thalamocortical networks, we examined clinical symptomatology as a difference between diagnostic groups and also, as an exploratory analysis, in linear regressions including PRIME scores as a dimensional measure of psychopathology.

In ROI-to-ROI analyses, effects of clinical symptomatology were investigated after segregating individuals into diagnostic groups (typically developing, psychosis spectrum, other psychopathologies) using an omnibus ANCOVA with each thalamo-cortical network included as repeated measures, and diagnostic group as the independent variable, including age, age², sex and motion (FD) as covariates of no interest, followed by a one-way univariate ANCOVAs for each thalamocortical network with diagnostic group as the independent variable and age, age², sex and motion (FD) as covariates of no interest.

ANCOVAs were conducted separately for whole-brain voxel-wise connectivity from each of the 7 thalamus seeds, with diagnostic group as an independent variable and age, age², sex and motion (FD) included as covariates of no interest, as described in the following equation.

$$\begin{gathered} Connectivity={\beta }_0+{\beta }_{1}DiagnosticGroup+{\beta }_{2}Age+{\beta }_{3}Ag{e}^2+{\beta }_{4}Sex \\ +{\beta }_{5}motion+\epsilon \end{gathered}$$ (4)

As a dimensional measure of psychopathology, the sum of PRIME scores was included as a predictor in linear regression models predicting thalamocortical connectivity. In ROI-to-ROI analyses, effects of PRIME score on thalamocortical networks were modeled as linear effects with PRIME score as a predictor as shown in the regression model below.

$$\begin{gathered} Connectivity={\beta }_0+{\beta }_{1}PRIME+{\beta }_{2}Age+{\beta }_{3}Ag{e}^2+{\beta }_{4}Sex \\ +{\beta }_{5}motion+\epsilon \end{gathered}$$ (5)

For all models investigating the effects of PRIME score, age, age², sex and motion (FD) were included as covariates of no interest.

2.5.4. Interaction effects

Interaction effects between diagnostic group and sex were investigated using a two-way ANCOVA with group and sex as independent variables and age, age² and motion (FD) as covariates of no interest. Interaction effects between group and age, and sex and age were investigated using linear regression models, motion was also included as a covariate of no interest. Equations for all the interaction effects are presented in the Supplemental Materials.

3. Results

3.1. Thalamocortical functional connectivity: ROI-to-ROI analyses

3.1.1. Age effects

As shown in Fig. 1 A, somatosensory cortex–ventral posterolateral nucleus and occipital cortex–lateral geniculate nucleus connectivity showed significant negative linear associations with age (r_partial = −0.163, p < .001 and r_partial = −0.174, p < .001 respectively). Older youth showed less connectivity within these networks.

Fig. 1.

Age (A), Sex (B) and Group (C) effects in predefined thalamocortical networks across the whole sample. The seven networks tested were prefrontal cortex–mediodorsal nucleus (PFC–MD), motor cortex–ventrolateral nuclear group (MTR–VL), somatosensory cortex–ventral posterolateral nucleus (SS–VPL), temporal cortex–medial geniculate nucleus (TMP–MGN), hippocampus–anterior nuclear group (HIPP–AN), parietal cortex–pulvinar (PAR–PU), occipital cortex–lateral geniculate nucleus (OCC–LGN). † indicates p < .05, uncorrected.

Temporal cortex–medial geniculate nucleus connectivity showed a significant quadratic association with age (r_partial = 0.097, p = .001). Higher connectivity was observed in late childhood/early adolescence and adulthood, while late adolescence showed lower functional connectivity in this network. No other networks showed significant linear or quadratic effects of age on functional connectivity. See Supplementary Materials for results based on functionally defined thalamic ROIs.

3.1.2. Sex effects

An omnibus ANCOVA including all seven networks found no significant interaction between network and sex (F(6,6570) = 1.964, p = .067), main effect of sex (F(1,1095) = 0.114, p = .736) or network (F(6, 6570) = 1.199, p = .304). As seen in Fig. 1B, connectivity between the parietal cortex and pulvinar showed an effect of sex (F(1,1095) = 4.939, p = .026), though this effect did not survive Bonferroni correction (ɑ = 0.007). No other network showed an effect of sex. See Supplementary Materials for results based on functionally defined thalamic ROIs.

3.1.3. Effects of clinical symptomatology

To examine psychosis symptomatology by diagnostic group, an omnibus ANCOVA including all seven networks revealed a significant interaction between network and diagnostic group (F(12,6558) = 2.216, p = .009), but no significant main effect of diagnostic group (F(2,1093) = 1.418, p = .243) or network (F(6,6558) = 1.274, p = .266). Post-hoc ANCOVAs were conducted for each network including age, age², motion and sex as covariates of no interest. As seen in Fig. 1C, connectivity between the motor cortex and ventrolateral nucleus showed an effect of diagnostic group (F(2,1093) = 3.131, p = .044), though this did not survive Bonferroni correction (ɑ = 0.007). Typically developing youth showed greater connectivity than both the psychosis spectrum youth (p = .021) and youth with other psychopathologies (p = .036). No other network showed any effect of diagnostic group.

As there was a significant difference in age and sex between the three diagnostic groups, we conducted a sensitivity analysis by creating sub-samples of the psychosis spectrum and typically developing youth matched for age, sex and race. Matching was conducted in the MatchIt package (Version 3.0.2) (Ho et al., 2011) in R (Version 3.6.3; R Core Team, 2020) and resulted in a final sample of 83 psychosis spectrum and 99 typically developing youth. See the Supplemental Table 1 for demographics of the matched samples. The sensitivity analysis did not show any significant main effects of diagnostic group (F(1176) = 0.691, p = .407), network (F(6,1056) = 1.054, p = .489) or interaction between network and diagnostic group (F(6,1056) = 1.054, p = .489). None of the 7 predefined thalamocortical networks showed any significant effect of diagnostic group. See Supplemental Material for additional details and results based on functionally defined thalamic ROIs.

In an exploratory analysis of the dimensional effect of psychopathology, we examined the association between connectivity within thalamocortical networks and PRIME scores in a series of linear regressions. As seen in Supplemental Fig. 5, the prefrontal–mediodorsal nucleus network showed a significant negative association with PRIME scores, such that greater PRIME scores were associated with lower connectivity within this network.

3.1.4. Interaction effects of age, sex and diagnostic group

As seen in Supplementary Fig. 4, there was a significant interaction between linear age and sex in the somatosensory–ventral posterolateral nucleus network (r_partial = −0.083, p = .007), showing that males had a more pronounced negative association with age compared to females. No other network showed any significant interaction between linear age, sex and diagnostic group.

3.2. Thalamocortical functional connectivity: voxel-wise analyses

3.2.1. Age effects

As shown in Figs. 2 and 3, linear age effects were characterized by negative associations with sensory/motor cortical regions and anterior insula, together with positive associations with heteromodal prefrontal and parietal regions. These effects were more pronounced with sensory/motor thalamic nuclei.

Fig. 2.

Voxelwise age and sex effects in thalamic association nuclear groups. Freesurfer segmented mediodorsal, pulvinar and anterior nuclear groups were used in a seed-based voxelwise functional connectivity analyses. Linear age, quadratic age and sex effects were masked to only include cortical voxels and thresholded at a cluster level p = .05 FWE corrected, and voxelwise p = .001 uncorrected.

Fig. 3.

Voxelwise age and sex effects in thalamic sensory/motor nuclear groups. Freesurfer segmented ventral lateral, ventral posterolateral, medial geniculate and lateral geniculate nuclear groups were used in seed-based voxelwise functional connectivity analyses. Linear age, quadratic age and sex effects were masked to only include cortical voxels and thresholded at a cluster level p = .05 FWE corrected, and voxelwise p = .001 uncorrected.

As shown in Figs. 2 and 3, quadratic age effects were observed between multiple thalamic nuclei and primarily dorsal attention network regions (including premotor cortex and inferior and superior parietal lobules) characterized by a U-shaped association such that connectivity was higher in early in adolescence and early adulthood, and lower in middle to late adolescence. See Supplemental Figs. 7 and 8 for examples, and see Supplementary Materials for results based on functionally defined thalamic ROIs.

3.2.2. Sex effects

As seen in Figs. 2 and 3, the mediodorsal, pulvinar, ventrolateral and ventral posterolateral nuclei all showed greater connectivity with occipital and parietal visual regions in females compared to males. The pulvinar, ventrolateral and ventral posterolateral nuclei showed greater connectivity with the inferior frontal and orbitofrontal cortices in males compared to females. See Supplementary Materials for results based on functionally defined thalamic ROIs.

3.2.3. Effects of clinical symptomatology

As seen in Fig. 5, psychosis spectrum youth showed lower connectivity compared with typically developing youth in connectivity between somatomotor thalamic nuclei (ventral lateral and ventral posterolateral nuclear groups) and the premotor cortex (extending into the inferior frontal cortex in a sensitivity analysis matched for age, sex and race, as seen in Supplementary Fig. 12). Psychosis spectrum youth showed lower connectivity compared to youth with other psychopathologies in medial prefrontal cortex (with ventral lateral nuclear group), lateral prefrontal cortex (with ventral posterolateral nuclei), and as seen in Fig. 4, inferior temporal cortex (with the pulvinar). Youth with other psychopathologies showed greater connectivity with visual cortex (with the lateral geniculate nucleus) compared to psychosis spectrum youth. See Supplementary Materials for results based on functionally defined thalamic ROIs.

Fig. 5.

Effects of diagnostic groups in thalamic sensory/motor nuclear groups. FreeSurfer segmented ventral lateral, ventral posterolateral, medial geniculate and lateral geniculate nuclear groups were used in seed-based voxel-wise functional connectivity analyses. Connectivity in typically developing youth (TD) and youth with other psychopathologies (OP) were compared with psychosis spectrum youth (PS). Maps were restricted to the cortex and thresholded at a cluster level p = .05 FWE corrected, and voxelwise p = .001 uncorrected.

Fig. 4.

Effects of diagnostic groups in thalamic association nuclear groups. FreeSurfer segmented mediodorsal, pulvinar and anterior nuclear groups were used in seed-based voxelwise functional connectivity analyses. Connectivity in typically developing youth (TD) and youth with other psychopathologies (OP) were compared with psychosis spectrum youth (PS). Maps were restricted to the cortex and thresholded at a cluster level p = .05 FWE corrected, and voxelwise p = .001 uncorrected.

4. Discussion

Thalamocortical dysconnectivity has been implicated in multiple neurodevelopmental disorders, especially schizophrenia. However, development of these networks in typically developed youth and youth atrisk for psychosis remains largely unknown. To address this knowledge gap, we characterized the effects of age, sex and clinical symptomatology on functional connectivity in anatomically constrained thalamocortical circuits in a large sample of youth from the PNC data. Broadly, we found significant age effects in thalamic connectivity along with modest sex differences. Subtle associations with psychopathology were also observed, although these results did not always survive correction for multiple comparisons and sensitivity analyses.

4.1. Thalamocortical network development from childhood to early adulthood

Both ROI-to-ROI and voxel-wise analyses found that thalamic connectivity with sensory cortical areas decreased with age. In the ROI-to-ROI analyses, the ventral posterolateral nuclei and lateral geniculate nuclei demonstrated age-related decreases in connectivity with their primary cortical targets, somatosensory and occipital cortex respectively. Voxel-wise analyses recapitulated these results. These findings are largely consistent with prior resting-state studies by Fair et al. (2010) and Steiner et al. (2020), which, despite using different methods, found lower functional connectivity in sensory thalamocortical networks from late childhood to young adulthood. Interestingly, the human thalamus is dominated by functional connectivity with sensory/motor networks in neonates, though this connectivity decreases rapidly in the first year after birth. Instead, over the first two years after birth, thalamocortical connectivity becomes dominated by the salience network, which includes the anterior insula and dorsal anterior cingulate cortex (Alcauter et al., 2014). Our results show that thalamocortical functional connectivity with both sensory/motor and salience network regions, primarily the anterior insula, also decreases in late childhood and adolescence into early adulthood. This developmental pattern differs from their cortico-cortical counterparts, which show their adult pattern in the first two years of life and appear to be largely mature by late childhood/early adolescence (Gao et al., 2015a, 2015b; Lin et al., 2008; Marek et al., 2015).

In contrast to sensory nuclei, which demonstrated prominent decline in connectivity with their primary sensory cortical targets with age, connectivity between the mediodorsal, pulvinar and ventral lateral nuclear groups with their primary cortical targets were relatively stable. Our findings help clarify prior developmental studies that have produced mixed findings. An early study found that adults showed greater connectivity between the thalamus and prefrontal cortex compared with children (Fair et al., 2010). A recent study investigating thalamus connectivity with multiple cortical networks found no effects of age in executive networks, which are composed primarily of prefrontal and some parietal association regions (Steiner et al., 2020). These differences may be explained by differences in analysis methods. Both our study and the recent study by Steiner and colleagues included larger sample sizes and modelled age as a continuous variable, rather than a categorical variable. The frontal and parietal thalamocortical networks do not have prominent connectivity in the first two years of life (Alcauter et al., 2014), yet our results show that they are prominent and relatively stable in late childhood/early adolescence. This pattern is different from that observed in studies of cortico-cortical networks, which show that frontoparietal networks start developing in the first two years of life and continue to develop through adolescence into early adulthood (Gao et al., 2015a, 2015b; Lin et al., 2008; Marek et al., 2015).

Our voxel-wise analyses revealed decreased within-network and increased cross-network connectivity with age. In addition to lower connectivity with sensory/motor cortices, there was a positive age effect on connectivity between multiple thalamic nuclei, though predominantly sensory and motor nuclei, and regions belonging to the cingulo-opercular and default mode networks. The pattern of lower connectivity with sensory/motor cortices and greater connectivity with association cortices was reminiscent of a sensorimotor–association gradient that has been described in cortical connectivity (Margulies et al., 2016). In particular, development of cortical networks shift from a gradient dominated by the distinction between somatomotor and visual cortices, to a gradient dominated by the distinction between sensorimotor and association cortices in adolescence (Dong et al., 2021). These thalamocortical functional connectivity changes may reflect maturation of thalamocortical white matter microstructure. In this sample, diffusion weighted imaging found greater fractional anisotropy with age, indicating increased myelination, particularly in white matter tracts between the thalamus and association cortices (Avery et al., 2021 Submitted for publication). This shift may also reflect increased integration of sensory/motor thalamus with association cortical networks. In young adults, the thalamus functions as an integrative hub for cortical brain networks, with both sensory/motor and association thalami showing strong functional connectivity with multiple cortical networks, though the cingulo-opercular and default mode networks showed the most connectivity across thalamic nuclei (Hwang et al., 2017). In youth, thalamocortical functional connectivity with salience and default mode networks are positively associated with top down control functions such as selective attention and cognitive flexibility (Steiner et al., 2020). Together with these previous studies, our results suggest that connectivity between the thalamus and cingulo-opercular networks mature during adolescence, possibly reflecting maturation of top-down control processes (Marek et al., 2015).

Our voxel-wise analyses also showed a U-shaped age effect between multiple thalamic nuclear groups with dorsal attention network regions, though this effect was substantially reduced in a sample with stricter motion control. In a recent study examining thalamic connectivity with large cortical networks, the dorsal attention network was the only cortical network that showed a positive effect of age (Steiner et al., 2020), though they did not test whether there were any quadratic effects of age. The dorsal attention network is involved in cognitive functions that are disrupted in neurodevelopmental disorders such as schizophrenia, including the control of attention and working memory (Greene and Soto, 2014; Majerus et al., 2018; Ptak, 2012; Ptak et al., 2017; Spreng et al., 2010). The quadratic effect of age may indicate that there is reorganization of the dorsal attention network during adolescence, making it particularly sensitive to disruption.

4.2. Sex effects in thalamocortical functional networks during development

The ROI-to-ROI analysis showed a trend towards females having higher connectivity in the parietal–pulvinar network. This finding was confirmed in the voxel-wise analyses, which found a consistent pattern across multiple nuclei where females showed greater connectivity with visual regions in both the occipital and parietal cortices. Sex effects have been reported in various visual processes including contrast sensitivity, visual acuity, color vision and visual attention (Gur et al., 2000; Vanston and Strother, 2017; Weiss et al., 2003), though the direction of the effect is not always consistent. In contrast to the greater thalamocortical connectivity we observed, males tend to have larger visual cortices (Amunts et al., 2007; Brun et al., 2009). Previous imaging studies have also observed that males tend to have higher BOLD activation within the visual cortex to visual stimuli (Cowan et al., 2000; Levin et al., 1998), while females showed shorter latencies and larger amplitudes in visual evoked potentials (Dion et al., 2013). Due in part to the inconsistency surrounding sex differences in visual processing, the functional consequences of the thalamocortical functional connectivity differences with visual cortex are unknown and deserve further study.

The voxel-wise analyses further found greater connectivity between the pulvinar, ventral lateral and ventral posterolateral nuclear groups and the orbitofrontal and inferior frontal cortex in males compared with females, which was not tested for in the ROI-to-ROI analyses. The orbitofrontal cortex is implicated in affective decision-making processes, such as reward-motivated decisions (Padoa-Schioppa and Conen, 2017; Rolls, 2004). Males have a small, but consistent advantage on the orbitofrontal dependent Iowa Gambling Task, which measures people’s ability to make decisions that favor smaller rewards that lead to long-term gains over larger rewards that lead to long-term losses (van den Bos et al., 2013). Differences in this task have been attributed to the organizational effects gonadal hormones have on the orbitofrontal cortex during gestation (Overman, 2004). Whether males continue to show greater thalamus connectivity with the orbitofrontal cortex in adulthood or if this is a developmental phenomena deserves further study.

4.3. Thalamocortical functional connectivity in psychosis spectrum youth

As discussed earlier, thalamocortical functional dysconnectivity is a robust finding in schizophrenia and is often characterized by a combination of reduced prefrontal-thalamic connectivity and somatomotorthalamic hyper-connectivity (Giraldo-Chica and Woodward, 2017; Ramsay, 2019). Consistent with this, voxel-wise analyses revealed lower thalamic connectivity with regions of the frontal cortex in psychosis spectrum youth, though this effect was primarily with the somatomotor thalamic nuclei rather than with the mediodorsal nucleus, as has been previously reported in some clinical samples. In addition, our exploratory analysis showed PRIME scores were negatively associated with connectivity in the prefrontal–mediodorsal nucleus network, indicating a dimensional approach to psychopathology may be more sensitive to early dysconnectivity in thalamocortical networks in youth. A previous study in a clinical-high risk sample showed both greater sensory/motor connectivity and lower prefrontal connectivity; this dysconnectivity was greater in individuals that converted compared with those that did not (Anticevic et al., 2015). We find that maturation of thalamocortical functional connectivity is reflected in decreased connectivity with sensory/motor networks and increased connectivity with association networks. Our findings, together with findings from Anticevic and colleagues suggests that this process is arrested in psychosis, supporting a neurodevelopmental origin to thalamocortical dysconnectivity in psychotic disorders, though evidence in cortical functional networks have also shown signs of hypermaturation in psychosis spectrum youth (Truelove-Hill et al., 2020).

Our results differed from a previous study using the PNC sample that did not find thalamocortical connectivity differences in psychosis spectrum youth (Jacobs et al., 2019). Our methodology differed from their study in two major ways. First, we separated youth that show symptoms of other psychopathologies from typically developing youth, which is necessary as thalamus dysconnectivity has been observed in other neurodevelopmental disorders such as autism (Nair et al., 2013; Woodward et al., 2017) and attention deficit hyperactivity disorder (Clerkin et al., 2013; Mills et al., 2012). Second, we defined thalamocortical networks differently, examining both anatomically constrained networks and the cortical connectivity from anatomically defined thalamic nuclei. We found only subtle effects of psychosis symptoms in this sample, suggesting that the appropriate probe of thalamocortical networks is necessary to find effects.

4.4. Strengths and limitations

Strengths of our study include the relatively large, community-ascertained samples; inclusion of youth with other psychopathologies in addition to typically developing and psychosis spectrum groups; use of anatomically defined thalamic sub-regions as seeds for the functional connectivity analyses rather than using the whole thalamus as a seed, as is often done, and complementary ROI-to-ROI and voxel-wise analyses to comprehensively map thalamic connectivity. The use of cross-sectional data to examine age effects is the main limitation of our study. However, we are not aware of a comparable longitudinal dataset comprised of typically developing and psychosis spectrum youth, as well as youth with other psychopathologies. In the future, the on-going Adolescent Brain Cognitive Development (ABCD) study (Casey et al., 2018) may be ideal for examining typical and atypical development of thalamocortical circuitry. Another limitation of our study is that we conducted seven different seed-based analyses for each thalamus ROI. While this allows us to examine different thalamocortical networks, it does inflate the chance of false positives. Future studies that take a gradient approach to examining thalamocortical connectivity may shed greater light on thalamic contribution to the development of the sensorimotor–association axis.

5. Conclusions

By characterizing age-related thalamocortical connectivity, and sex-related differences in thalamocortical networks for the first time, this study provides a framework for neurodevelopmental theories of thalamocortical dysconnectivity in clinical populations. We showed that over development, thalamocortical functional networks appear to integrate less with sensory systems and more with higher-order cortical systems, such as the cingulo-opercular network. This reflects the shift from sensory guided responses to more top-down, cognitively guided processes with maturation. This process may be disrupted in neurodevelopmental disorders, presenting as the pattern of lower-prefrontal, higher-sensory thalamocortical dysconnectivity observed in clinical psychosis samples. We show that subtle signs of this pattern are already present in youth showing psychosis symptomatology who will never convert to a psychotic disorder.