Neural Network Functional Interactions Mediate or Suppress WM-Emotional Behavior Relationships in Infants

Abstract

Background:

Elucidating the neural basis of infant positive emotionality (PE) and negative emotionality (NE) can identify biomarkers of pathophysiological risk. Our goal was to determine how functional interactions among large-scale networks supporting emotional regulation influence WM microstructural-emotional behavior relationships in 3-month-old infants. We hypothesized that microstructural-emotional behavior relationships would be differentially mediated or suppressed by underlying resting-state functional connectivity (rsFC), particularly between Default Mode Network (DMN) and Central Executive Network (CEN) structures.

Methods:

The analytic sample comprised primary caregiver-infant dyads [52 infants (42% female, mean age at scan=15.10 weeks)], with infant neuroimaging and emotional behavior assessments at 3 months. Infant WM and rsFC were assessed by diffusion-weighted imaging/tractography and resting-state magnetic resonance imaging (MRI) during natural, non-sedated sleep. The Infant Behavior Questionnaire-R provided measures of infant PE and NE.

Results:

Following significant WM-emotional behavior relationships, multimodal analyses were performed using whole-brain voxelwise mediation. Results revealed that greater cingulum bundle volume was significantly associated with lower infant PE (ß = −0.263, p = 0.031); however, a pattern of lower rsFC between CEN and DMN structures suppressed this otherwise negative relationship. Greater uncinate fasciculus volume was significantly associated with lower infant NE (ß = −0.296, p = 0.022); however, lower orbitofrontal cortex (OFC)-amygdala rsFC, suppressed this otherwise negative relationship, while greater OFC-CEN rsFC mediated this relationship.

Conclusions:

Functional interactions among neural networks have an important influence on WM microstructural-emotional behavior relationships in infancy. These relationships can elucidate neural mechanisms contributing to future behavioral and emotional problems in childhood.

Introduction

Temperamental features/emotional behaviors (i.e., low positive emotionality(PE), high negative emotionality(NE)), are early risk factors for subsequent behavioral and emotional problems(1–5). Elucidating the neural basis of infant emotional behaviors can identify objective biomarkers of risk that reflect underlying pathophysiological processes(6–8) and provide targets for interventions(9). Employing multimodal neuroimaging provides a deeper understanding of neural mechanisms underlying emotional behavior and has greater potential to reveal biomarkers of psychopathology than reliance on any single modality(10, 11).

In early infancy, the human brain undergoes complex and coordinated development of the neural architecture (12). Neuritic growth and myelination progress rapidly within the first postnatal year(12–16), and infant white matter (WM) development is associated with future cognitive processes and behavioral problems(17–19). The cingulum bundle(CB) and uncinate fasciculus(UF) connect key intrahemispheric regions implicated in emotional regulation, particularly anterior temporal structures, and anterior and posterior cingulate and orbitofrontal cortices, respectively(20–23). The forceps minor(FM) of the corpus callosum connects prefrontal and orbitofrontal regions and is important for interhemispheric communication(24). Although these structures reach full maturity in adulthood(25), they are identifiable in neonates(26); thus, a focus of infant multimodal neuroimaging research is to examine how WM development shapes clinically-relevant emotional behaviors. Diffusion-weighted imaging(DWI) provides proxy measures of WM structure, including structural integrity and fiber collinearity(e.g., fractional anisotropy, FA) and morphology(e.g., tract volume). We demonstrated that lower UF structural integrity and/or greater UF volume at 3 months predicted greater 9-month NE, which might reflect greater fiber arborization, potentially compromising emotional regulation capacity(27).

Neural network WM and function have a reciprocal relationship, and integration between neural network structure and functional connectivity allows global and integrative brain processes(28–30). Thus, another focus of infant multimodal neuroimaging is to examine relationships among large-scale network functional connectivity and measures of early emotionality. Resting-state functional connectivity(rsFC) reflects endogenous functioning and studies in neonates have shown that higher-order networks [e.g., default mode network(DMN), central executive network(CEN)] are topologically incomplete but become highly synchronized by age one, with increasing network-level integration(31, 32) and continued maturation beyond the first postnatal year(33–35). These higher-order, large-scale networks support emotional regulation processes(36, 37). The DMN(or medial frontoparietal network(38)) supports stimuli and perception value coding and self-monitoring; anchoring structures are medial prefrontal cortex, posterior cingulate cortex(PCC), and posterior inferior parietal lobule(36, 38, 39). The “limbic network,” comprising the orbitofrontal cortex(OFC) and anterior temporal lobes, is encompassed by the DMN(38). The CEN(or lateral frontoparietal network(38)) supports executive, goal-directed functions(38); anchoring structures are lateral prefrontal cortex(including dorsolateral prefrontal cortex, dlPFC) and the anterior inferior parietal lobule(36, 38, 40, 41). DMN-CEN connectivity is important for internally-directed attention(42), where DMN deactivation, greater CEN activation, and inverse CEN-DMN functional connectivity(or anticorrelation) during executive function(43, 44) reflects an adaptive reduction in self-monitoring processes(45–48). Aberrant positive(or reduced inverse) CEN-DMN connectivity is also linked to lower emotional regulation capacity and greater psychopathology risk(49–57).

We demonstrated significant positive relationships among infant smiling and amygdala rsFC, predominantly with CEN regions; and that caregiver affect-infant emotional behavior relationships were mediated by patterns of amygdala rsFC associated with lower infant smiling: greater amygdala-DMN rsFC and lower amygdala-CEN rsFC(58). The few other infant rsFC studies have also primarily focused on amygdala rsFC-emotional behavior relationships and reported that greater amygdala rsFC is associated with greater NE(i.e., fear and/or sadness)(59, 60) and less change in infant NE over time(61); however, none focused on rsFC between large-scale networks or on infant PE.

Critically, the CB, UF and FM connect key regions within these large-scale networks. The CB and UF form connecting pathways within the DMN; the CB connects temporal-parietal-frontal areas and the UF connects the temporal pole with the inferior frontal lobe(62) and posterior orbitofrontal areas(63). The CB also carries fibers connecting CEN and DMN anchors (dlPFC and PCC)(23). The FM provides lateral and medial PFC connectivity(64) across CEN and DMN, respectively. Given the integrative nature of WM structure and functional connectivity(28–30), it is possible that neural network function mediates the relationship between WM structure and behavior. Yet, few infant studies employed multimodal neuroimaging to examine intrinsic relationships among microstructural measures and endogenous network function(65–67). One study explored structure-function associations underlying noxious stimulus-evoked responses(65); others examined effects of prenatal conditions on multimodal neural outcomes(66, 67). Furthermore, most infant neuroimaging studies were performed at birth and 12 months(68). Yet, emerging emotional regulation(69) and neural functional specialization for processing negative emotion(70) occur around 3 months suggesting that this is an important developmental window for examining neural markers of early emotional behavior.

Neurodevelopment is influenced by the pre- and postnatal environment, including caregiver affect and socioeconomic status(SES)(17, 58, 71–74). Few studies of infant WM have examined emotion-related outcomes; one study showed that infant WM moderated the maternal psychopathology-infant NE relationship(75). Neonatal and infant rsFC is also sensitive to the pre- and postnatal environment(67, 72, 73, 76–79) and can predict later emotionality, including affective symptoms(e.g. internalizing)(59, 60, 80–82).

The goal of the present study was to employ multimodal neuroimaging to examine how functional interactions between large-scale neural networks supporting emotional regulation influence WM microstructural-emotional behavior relationships in 3-month-old infants. We first examined relationships between WM microstructure and infant PE and NE, focusing on three key WM tracts interconnecting regions in the DMN and CEN: CB, UF and FM. In order to integrate structural and functional measures, we used mediation models, where endogenous rsFC within these large-scale networks mediated relationships between WM and infant PE and NE. Measures of caregiver affect and socioeconomic status(SES) are included as additional predictors in our analyses.

Based on our prior work indicating a relationship between greater UF volume and higher later NE(27), and previous research showing relationships between greater CEN-DMN rsFC and lower emotional regulation capacity(49–57), we hypothesized that greater CEN-DMN connectivity would mediate(facilitate) relationships between greater WM volume and lower PE/higher NE, while lower CEN-DMN connectivity would suppress(reduce) those relationships.

Methods and Materials

Study Design and Participants

Overview

The study sample comprised caregiver-infant dyads, with both neuroimaging of the infant and caregiver report of infant emotional behavior at 3 months. Neuroimaging was completed at Children’s Hospital of Pittsburgh(CHP) between September 2018 and March 2020 (no caregiver or infant underwent study procedures during the COVID-19 shutdown period). All procedures were approved by the University of Pittsburgh Institutional Review Board.

Infants were healthy at birth with the following exclusion criteria: premature birth(prior to 37 weeks), low birthweight(<5.5lb; caregiver report, medical records), abnormal morphometry(i.e., small occipitofrontal circumference(<32cm)), abnormal APGAR scores(<7 at 5 mins) and extended hospitalization for physical health problems, including neurological illness. Infant MRI metal exclusion criteria were pacemakers, aneurysm clips and non-removable ferromagnetic material. Caregivers were excluded if they had prenatal/concurrent illicit substance use/substance use disorder(obstetric records/self-report), respectively, <2 hours/day care of their infant or were <18 years old(unable to give informed written consent).

Participants

Participants comprised 79 caregivers and their 3-month-old infants recruited from the community, including hospital postnatal wards, pediatric practices and a university research registry. Fifty-seven infants had usable DWI data; one was subsequently excluded for unreported low birthweight, and four others had incomplete behavioral data, leaving 52 caregiver-infant dyads(caregivers: 98% biological mothers, ages 20–42, mean=31.54 years, SD=4.09; infants: 42% female, ages 10–22 weeks, mean=15.10 weeks, SD=2.93; Supplement, Table S1.)

Magnetic Resonance Imaging (MRI) Procedures

At 3 months of age, infants underwent an MRI scan on a Siemens 3 Tesla Skyra system using a 32-channel head coil during natural, non-sedated sleep using the feed-and-wrap technique(83, 84).

DWI Acquisition and Preprocessing

DWI data were acquired using a multi-shell diffusion scheme with 2mm slice thickness; two sequences were acquired, one with posterior to anterior phase encoding with 10 b0 volumes(TR=2500 ms, TE=80 ms, voxel size=2.0× 2.0×2.0 mm³, FoV=200 mm), and another with posterior to anterior phase encoding with 159 volumes with b-values of 750 (50 directions) and 2000 (100 directions) s/mm² (TR=3000 ms, TE=97 ms, voxel size = 2.0×2.0×2.0 mm³, FoV=200 mm).

DWI data were preprocessed using fMRI Software Library(FSL)’s TOPUP and EDDY tools. Using described methods(85), FSL’s TOPUP tool corrected for susceptibility-induced distortions. FSL’s BET removed non-brain tissue(86) and EDDY then corrected for eddy current distortions and subject motion(87). Images with artifact, signal loss or distortion were excluded from further analyses(n=22/79 infants; Supplement).

DWI Tractography

Using DSI Studio software, DWI data were reconstructed using a native-space generalized q-sampling imaging(GQI) approach(88). Tractography was performed for each targeted tract (e.g., left and right CB, left and right UF and FM) using the Automated Fiber Tracking(AutoTrack) function with consistent parameters across infants(Supplement, Table S2, Fig. S1). Only the frontoparietal/“standard” cingulum(22) was included in the tractography. Mean fractional anisotropy(FA)(89) and tract volume were extracted from each tract and averaged across hemispheres for CB and UF. Extracted axial and radial diffusivity, and intracranial volume (ICV, calculated from the brain mask of the b0 image) were used for post-hoc, exploratory analyses.

RsFC Acquisition and Preprocessing

To enhance the likelihood of acquiring usable data, two, five-minute resting-state imaging acquisitions were conducted. Resting-state data were acquired using echo-planar imaging (EPI)-blood-oxygenation-level dependent(BOLD), voxel size=3 mm³ isotropic, TR=800 ms, TE=32 ms, simultaneous multi-slice(SMS) factor=4,380 total frames acquired.

Previous methods(90) were used to minimize the risk of spurious resting-state correlations due to participant motion(Supplement). Five infants with usable DWI and behavioral data did not have usable rsFC data, leaving 47 infants for multimodal DWI and rsFC analyses.

Behavioral Assessments

Infant emotional behavior and caregiver (e.g., affect and sociodemographic) questionnaires were completed at an initial in-person, at home visit. The Edinburgh Postnatal Depression Scale(EPDS) was readministered if the scan visit fell beyond two weeks following the home visit.

Infant Emotional Behavior

Infant emotional behavior was measured via caregiver report at 3 months using the Infant Behavior Questionnaire Revised(IBQ-R)(91). PE and NE composites were calculated using the raw IBQ-R data with a mean of the items from each of the relevant subscales (i.e., PE - smiling, laughter and high pleasure; NE - sadness, distress, fear and falling reactivity subscales(58, 91)).

Caregiver Affect and Sociodemographic Covariates

As caregiver affect and SES are also important for neurodevelopment(17, 58, 71–74), we included these measures as additional predictors in analyses. Caregiver depressed mood and affective instability were assessed at 3 months using the EPDS(92) and the Personality Assessment Inventory-Borderline Features Scale(PAI-BOR(93)), respectively. A summary measure of public assistance was used as a SES proxy(Supplement, Table S1). (Supplement Table S3-correlations among behavioral and imaging variables.)

Statistical Analyses

WM-Emotional Behavior Relationships

Elastic net regressions using GLMNET in R (v.3.6.0) with cross validation(94) were performed for variable selection. We included 6 variables representing WM microstructure (FA and volume from each of the 3 tracts, UF, CB, FM), infant sex and age at scan(weeks), and caregiver age, SES, and EPDS and PAI-BOR scores at the time of the infant scan. Outcome variables included PE and NE evaluated in two separate models(Supplement, Figs. S2 & S3).

General Linear Models(GLM) were performed with elastic net-selected variables using a robust estimator(all variables and outcomes were Z-scored). Bootstrapped 95% confidence intervals(CIs) were generated using 1000 samples(bias corrected accelerated) and used to assess statistical significance. Post-hoc, exploratory GLMs substituting axial(AD) or radial diffusivity(RD) for significant WM volume variables were performed, as well as analyses including ICV as an additional predictor.

Mediation Analyses

For an overview of Mediation/Suppression analyses, see the Supplement. For rsFC, we included representative seed regions within the CEN(dlPFC) and DMN(PCC) with a whole-brain target. The medial OFC, a key UF region (21), was an additional seed(Supplement). The Shi et al. neonatal parcellation atlas provided seeds(95).

We then performed whole-brain voxelwise mediation analyses(n=47), with WM measures as the independent variables, rsFC as mediators, and NE and PE as separate outcome variables, focusing on the significant WM-emotional behavior relationships. The same covariates were used as in the elastic net regressions examining WM-emotional behavior relationships, adding a variable for resting-state image quality(square root of the number of retained functional MRI frames). Coefficients were extracted for a and b arms of the indirect effect, reflecting the WM-rsFC and rsFC-emotional behavior relationships, respectively; and the total effect, c, and direct effect, c’(Supplement). As there are not established directional relationships between WM structure, rsFC and behavior, we also performed secondary reverse mediations in which WM mediated the relationships between rsFC and infant emotionality(PE/NE) using the same seed regions for rsFC as in the main analyses.

Results

WM-Emotional Behavior Relationships

Positive Emotionality

Elastic net selected: FM FA (elastic net coefficients, 2.99), infant sex (0.33), 3-month infant age (0.08), caregiver postnatal depression (−0.04), CB volume (−1.85E-04) and FM volume (1.68E-04) as predicting PE at 3 months.

Post-hoc GLM showed that CB volume was the only significant predictor of PE (ß=−0.263, p=0.031, CI: −0.526,−0.049; Fig. 1, Table 1A). [EPDS approached significance (ß=−0.206, p=0.051), although not indicated by CIs.] Post-hoc, exploratory GLM substituting CB AD or RD for CB volume into the model did not reveal significant effects of CB AD or RD on PE (Supplement, Table S4&5). When ICV was included as an additional predictor in the PE model, averaged CB volume remains significant (ß=−0.252, p=0.035, Supplement, Table S6A).

Figure 1.

Lower CEN-DMN Resting-state Functional Connectivity Suppressed the Relationship between Cingulum Bundle Volume and Positive Emotionality. The plot displays the significant, negative relationship between cingulum bundle (CB) volume and positive emotionality (ß=−0.263, p=0.031), i.e., the direct effect (c’). Schematic depictions of the mediation analyses with resting-state functional connectivity (rsFC) of the dorsolateral prefrontal cortex (dlPFC/Anterior Central Executive Network, CEN) seed (Top) and the posterior cingulate cortex (PCC/Posterior Default Mode Network, DMN) seed (Bottom) depict the negative relationships (red arrows) of the a (CB volume-rsFC) and b (rsFC-PE) arms of the mediations. Images depict suppression effects (light blue) of dlPFC/Anterior CEN connectivity with predominantly DMN regions: middle temporal gyrus (MTG), angular gyrus (AG) and precuneus (PC), and suppression effects of PCC/Posterior DMN connectivity with a CEN region, the middle frontal gyrus (MFG). Thus, lower CEN-DMN/DMN-CEN connectivity suppressed the otherwise negative relationship between CB volume and infant PE (p_{FWE-corrected} < 0.05).

Table 1.

General Linear Model Results: White Matter-Infant Emotionality Relationships

A. Infant Positive Emotionality
	Beta	Std. Error	p
Infant Sex	.186	.1070	.083
Infant Age (weeks, 3m)	.230	.1272	.070
Postnatal Depression (EPDS)	−.206	.1056	.051
Forceps Minor FA	.142	.1204	.237
Cingulum Bundle Volume	−.263	.1216	.031
Forceps Minor Volume	.189	.1197	.114
B. Infant Negative Emotionality
Affective Instability (PAI-BOR)	.331	.1488	.026
Forceps Minor FA	.175	.1457	.229
Uncinate Fasciculus Volume	−.296	.1295	.022

Postnatal Depression (Edinburgh Postnatal Depression Scale, EPDS); Fractional Anisotropy (FA); Affective Instability (Personality Assessment Inventory– Borderline Features Scale, PAI-BOR)

Negative Emotionality

Elastic net selected: FM FA (elastic net coefficients, 0.47), caregiver affective instability (0.02) and UF volume (−4.12E−05) as predicting NE.

Post-hoc GLM showed that both caregiver affective instability (ß=0.331, p=0.026, CI: 0.017, 0.710) and UF volume (ß=−0.296, p=0.022, CI: −0.533,−0.087; Fig. 2) were significant predictors of NE (Table 1B). Post-hoc, exploratory GLM substituting UF AD or RD for UF volume into the model did not reveal significant effects of UF AD or RD on NE (Supplement, Table S4&5). When ICV was included as an additional predictor in the NE model, averaged UF volume was no longer significant (ß=−0.180, p=0.169, Supplement, Table 6B&C).

Figure 2.

Lower OFC-Amygdala and OFC-DMN Resting-state Functional Connectivity Suppressed the Relationship between Uncinate Fasciculus Volume and Negative Emotionality While Greater OFC-DMN and OFC-CEN Connectivity Mediated This Relationship. The plot displays the significant, negative relationship between uncinate fasciculus (UF) volume and negative emotionality (ß=−0.296, p=0.022), i.e., the direct effect (c’). Schematic depictions of the mediation analyses with resting-state functional connectivity (rsFC) of the orbitofrontal cortex (OFC) seed depict the negative (Top, red arrow) and positive relationships (Bottom, green arrow) of the a (UF volume-OFC rsFC) arm, and the negative relationships (red arrows) of the b (OFC rsFC-NE) arms of the mediations. Top: Images depict suppression effects (blue) of OFC connectivity with the amygdala and DMN regions, including inferior frontal gyrus (IFG) and middle temporal gyrus (MTG). Thus, lower OFC-amygdala and OFC-DMN connectivity suppressed the otherwise negative relationship between UF volume and infant NE. Bottom: Images depict mediation effects (light green) of OFC connectivity with DMN regions: superior frontal gyrus (SFG) and angular gyrus (AG), and CEN regions: supramarginal gyrus (SMG) and middle frontal gyrus (MFG). Thus, greater OFC-DMN and OFC-CEN connectivity mediated the negative relationship between UF volume and infant NE (p_{FWE-corrected} < 0.05).

RsFC Mediates or Suppresses WM-Emotional Behavior Relationships

CB Volume, dlPFC rsFC & PE

The WM-emotional behavior relationship was a negative association between CB volume and infant PE. Mediation analyses (abbreviated results described in text, all p_{FWE-corrected} < 0.05) revealed significant positive indirect relationships (suppression). Here, greater CB volume was associated with lower dlPFC (anterior CEN) connectivity with predominantly DMN regions: middle temporal gyrus, angular gyrus and precuneus; and lower dlPFC-DMN connectivity was associated with greater PE (Fig. 1, Fig. S4). Thus, lower CEN-DMN connectivity suppressed the otherwise negative relationship between CB volume and infant PE (Table 2A).

Table 2.

DlPFC & PCC Resting-state Connectivity Suppressed the Negative Relationship between Cingulum Bundle Volume and Infant Positive Emotionality

A. Resting-state Connectivity Seed: Dorsolateral Prefrontal Cortex (dlPFC) – Anterior CEN
Regions	Coordinates	k	a	b	Direct	Total	Indirect	Indirect CIs	M/S
Middle temporal gyrus left	−33, −36, 10	64	−3.50E-04	−7.72E-01	−4.38E-04	−1.68E-04	2.70E-04	0.000121, 0.001450	S
Middle temporal gyrus right	32, −40, 11	68	−2.90E-04	−7.00E-01	−3.71E-04	−1.68E-04	2.03E-04	0.000043, 0.000653	S
Angular gyrus right	29, −40, 22	78	−2.92E-04	−5.80E-01	−3.38E-04	−1.68E-04	1.69E-04	0.000036, 0.000500	S
Angular gyrus left	−29, −39, 22	50	−2.65E-04	−4.52E-01	−2.88E-04	−1.68E-04	1.20E-04	0.000013, 0.000657	S
Precuneus right	8, −42, 26	68	−3.30E-04	−3.60E-01	−2.87E-04	−1.68E-04	1.19E-04	0.000017, 0.000640	S
Postcentral gyrus left	−25, −20, 39	56	4.66E-04	2.37E-01	−2.79E-04	−1.68E-04	1.10E-04	0.000018, 0.000648	S
Superior parietal gyrus left	−15, −39, 31	11	−2.59E-04	−4.17E-01	−2.76E-04	−1.68E-04	1.08E-04	0.000005, 0.000429	S
Superior temporal gyrus right	35, −30, 15	24	−2.88E-04	−3.44E-01	−2.67E−04	−1.68E-04	9.90E-05	0.000007, 0.000635	S
Paracentral lobule left	−8, −22, 47	28	3.54E-04	2.67E-01	−2.63E−04	−1.68E-04	9.46E-05	0.000016, 0.000548	S
Middle occipital gyrus left	−22, −43, 23	14	−2.05E-04	−4.40E-01	−2.58E−04	−1.68E-04	9.02E-05	−0.000017, 0.000360	S
Supramarginal gyrus right	35, −31, 20	14	−1.86E-04	−2.87E-01	−2.22E−04	−1.68E-04	5.34E-05	−0.000007, 0.000597	S
Precuneus left	−9, −37, 29	41	−1.80E-04	−2.92E-01	−2.21E−04	−1.68E-04	5.26E-05	−0.000012, 0.000456	S
Superior occipital gyrus right	14, −47, 25	20	−2.31E-04	−2.13E-01	−2.17E−04	−1.68E-04	4.92E-05	−0.000007, 0.000361	S
Cuneus right	10, −48, 23	18	−1.95E-04	−1.63E-01	−2.00E−04	−1.68E-04	3.18E-05	−0.000013, 0.000478	S
Inferior parietal lobule left	−27, −29, 30	17	2.23E-04	−1.04E-02	−1.66E−04	−1.68E-04	−2.33E-06	−0.000187, 0.000100	M
B. Resting-state Connectivity Seed: Posterior Cingulate Cortex (PCC) – Posterior DMN
Middle frontal gyrus left	−23, 16, 22	69	−2.34E-04	−6.02E-01	−3.09E−04	−1.68E-04	1.41E-04	0.000005, 0.000566	S

k – number of voxels within a cluster; a – independent variable-to-mediator relationship; b - mediator-to-dependent variable relationship; M/S – Mediation/Suppression.

CB Volume, PCC rsFC & PE

Mediation analyses revealed a significant positive indirect relationship (suppression). Here, greater CB volume was associated with lower PCC (posterior DMN) connectivity with a CEN region, the middle frontal gyrus; and lower PCC-middle frontal gyrus connectivity was associated with greater PE (Fig. 1, Fig. S5). Thus, lower DMN-CEN connectivity suppressed the otherwise negative relationship between CB volume and infant PE (Table 2B).

UF Volume, OFC rsFC & NE

The WM-emotional behavior relationship was a negative association between UF volume and infant NE. Mediation analyses revealed significant positive indirect relationships (suppression). Here, greater UF volume was associated with lower OFC connectivity with the amygdala (salience network) and DMN regions, including inferior frontal gyrus, middle temporal gyrus and intra-OFC (inferior) connectivity; and lower OFC-amygdala and OFC-DMN connectivity patterns were associated with greater NE (Fig. 2, Fig. S6). Thus, lower OFC-amygdala and OFC-DMN connectivity suppressed the otherwise negative relationship between UF volume and infant NE (Table 3).

Table 3.

Orbitofrontal Cortex Resting-state Connectivity Suppressed and Mediated the Negative Relationship between Uncinate Fasciculus Volume and Infant Negative Emotionality

Resting-state Connectivity Seed: Orbitofrontal Cortex (OFC) – Limbic, DMN
Coordinates	Regions	k	a	b	Direct	Total	Indirect	Indirect CIs	M/S
Amygdala left	−18, −5, −10	12	−4.53E-04	−3.29E-01	−3.91E-04	−2.42E-04	1.49E-04	0.000042, 0.000418	S
Inferior frontal gyrus triangular left	−30, 18, 7	86	−4.69E-04	−2.64E-01	−3.65E-04	−2.42E-04	1.24E-04	0.000050, 0.000388	S
Middle temporal gyrus left	−34, −28, 0	60	−4.58E-04	−2.03E-01	−3.35E-04	−2.42E-04	9.30E-05	0.000022, 0.000323	S
Orbitofrontal cortex inferior left	−25, 21, −2	13	−4.63E-04	−8.02E-02	−2.79E-04	−2.42E-04	3.71E-05	0.000001, 0.000291	S
Superior frontal gyrus dorsal left	−10, 20, 30	39	4.36E-04	−1.43E-01	−1.79E-04	−2.42E-04	−6.26E-05	−0.000321, −0.000009	M
Superior frontal gyrus medial right	5, 25, 25	10	3.57E-04	−1.55E-01	−1.86E-04	−2.42E-04	−5.52E-05	−0.000432, −0.000006	M
Supramarginal gyrus left	−34, −30, 22	18	2.65E-04	−1.80E-01	−1.94E-04	−2.42E-04	−4.76E-05	−0.000221, 0.000019	M
Superior temporal gyrus left	−30, −27, 14	15	1.91E-04	−2.01E-01	−2.03E-04	−2.42E-04	−3.85E-05	−0.000185, 0.000021	M
Middle frontal gyrus left	−19, 12, 31	32	3.39E-04	−1.00E-01	−2.08E-04	−2.42E-04	−3.39E−05	−0.000279, 0.000001	M
Angular gyrus left	−33, −37, 23	31	2.40E-04	−1.34E-01	−2.09E-04	−2.42E-04	−3.22E-05	−0.000320, 0.000001	M
Inferior parietal lobule left	−35, −30, 29	18	1.95E-04	−1.35E-01	−2.15E-04	−2.42E-04	−2.64E-05	−0.000248, 0.000006	M
Superior frontal gyrus medial left	−2, 25, 26	26	8.80E-05	−2.09E-02	−2.40E-04	−2.42E-04	−1.84E-06	−0.000405, 0.000013	M

k – number of voxels within a cluster; a – independent variable-to-mediator relationship; b - mediator-to-dependent variable relationship; M/S – Mediation/Suppression.

Mediation analyses also revealed significant negative indirect relationships (mediation). Here, greater UF volume was associated with greater OFC connectivity with DMN regions: superior frontal gyrus (dorsal, medial) and angular gyrus, and CEN regions: supramarginal gyrus, middle frontal gyrus, and inferior parietal lobule; and greater OFC-DMN and OFC-CEN connectivity was associated with lower NE (Fig. 2, Fig. S6). Thus, greater OFC-DMN and OFC-CEN connectivity mediated the negative relationship between UF volume and infant NE (Table 3).

Including ICV as a covariate in this model produced largely similar network connectivity patterns of suppression and mediation (Supplement, Table S8, Fig. S7).

Reverse Mediations: WM Mediates or Suppresses rsFC-Emotional Behavior Relationships

Reverse mediations focused on the significant mediations described above; these analyses used dlPFC, PCC or OFC seed to whole-brain target rsFC as the independent variable, either PE or NE as the dependent variable and either CB volume or UF volume as the mediator. These reverse-direction mediation analyses largely confirmed the nature of the tripartite relationships among WM, rsFC and PE and NE shown in our main findings(Supplement).

Discussion

Dynamic reconfiguration of structural and functional connectivity across large-scale brain networks occurs throughout development, with associations between developmental changes in rsFC and WM structural connectivity(11). The present study extends this work to infancy, demonstrating that relationships between the morphology of key WM tracts within large-scale networks important for emotional regulation and early emotional behavior are influenced by underlying rsFC. We examined the CB, UF and FM, which link structures within the CEN and DMN(62). Our primary findings were that greater CB volume was significantly associated with lower PE; however, an adaptive pattern of lower rsFC between CEN and DMN structures suppressed this relationship, and was associated with greater PE. Greater UF volume was significantly associated with lower NE; however, lower OFC-amygdala and OFC-DMN rsFC suppressed this relationship, while greater OFC-CEN rsFC mediated this relationship.

The CB is an association pathway connecting frontal, parietal and temporal areas(23). We focused on the frontoparietal segment, which connects specific CEN (dlPFC) and DMN (PCC) anchors(23). The CB is implicated in emotion, memory and attention-related processes and has the longest developmental trajectory of the major WM bundles(25). Thus, early differences in CB microstructure may have long-lasting consequences for emotional regulation. Our finding that greater CB volume was associated with lower PE in 3-month-old infants might reflect differences in WM or other microstructure, e.g., greater extracellular volume and/or greater dispersion of axons within the tract(96), that diminish emotional regulation capacity. Alternatively, greater volume may indicate inefficient synaptic pruning contributing to dysfunction(97).

The negative relationship between CB volume and PE was suppressed by lower CEN-DMN rsFC(i.e., dlPFC connectivity with DMN structures, and PCC connectivity with MFG). Thus, lower CEN-DMN connectivity was associated with greater PE in the context of greater CB volume. Deactivation in DMN regions(43, 44), alongside greater CEN activity and inverse CEN-DMN rsFC (or anticorrelation) during executive function and working memory tasks(43–45) is shown in adults(45, 46) and youth(47, 48), and is thought to reduce interference from self-monitoring processes during cognitive tasks(46, 98–100). Aberrant positive (or reduced inverse) CEN-DMN rsFC is associated with lower emotional regulation capacity and greater psychopathology risk(49–57), with emotional regulation deficits in preschool and school-age children(101), and worsening depressive symptom severity in school-aged youth at familial risk for affective disorders(102). Thus, our findings indicate that infants with greater CB volume but lower CEN-DMN rsFC might have more efficient switching capacity between these networks; these findings support a growing literature linking lower CEN-DMN rsFC with more adaptive emotional behavior, and suggest that these relationships emerge early in infancy.

The UF is also a long-range association pathway(25), connecting cortical nuclei of the amygdala and the temporal pole/anterior temporal lobe with posterior orbitofrontal areas(63). The UF connects DMN structures (temporal pole and inferior frontal lobe)(62) and the limbic network (including OFC and amygdala(41)) encompassed by the DMN(38), and supports formations of associations that motivate behavior and socio-emotional processing(21). Our findings indicate that in 3-month-old infants, larger UF volume is associated with lower NE, indicating that larger UF volume might yield greater opportunity for efficient pruning and ultimately greater connectivity, facilitating greater emotional regulation and lower NE. Our previous preliminary findings indicated that lower structural integrity (normalized quantitative anisotropy) and greater volume within the UF at 3 months were associated with greater NE prospectively at 9 months(27), however, potentially reflecting lower organization and/or greater dispersion of axons in this tract, yielding lower emotional regulation capacity. Together, these findings suggest that the association between early UF volume and infant NE changes across development.

The negative relationship between UF volume and NE was suppressed by patterns of OFC-amygdala and OFC-DMN(inferior frontal gyrus(IFG) and middle temporal gyrus) rsFC, such that lower OFC-amygdala and OFC-DMN rsFC were associated with greater NE. The OFC supports emotional processing and regulating and maintaining emotional responses(103); and the amygdala transmits salience and valence information to the anterior temporal lobe and OFC via the UF to guide cognitive processes, such as decision making(21, 104). Thus, lower OFC-amygdala rsFC might be associated with diminished coupling of valence information transmitted to the OFC, leading to emotional dysregulation and NE in the context of greater UF volume. The left IFG supports empathy, semantic processing and executive processes, such as working memory(105); and evaluation of actions and forming internal representations(105). Lower OFC-IFG rsFC in infancy might thus be associated with diminished capacity to integrate emotional and cognitive processing, contributing to greater NE. By contrast, greater OFC-DMN(superior frontal gyrus(SFG)) and OFC-CEN rsFC were associated with lower NE in the context of greater UF volume. The role of the CEN in executive function(36, 38), and that of the SFG in working memory(106) and cognitive control(43) suggest that greater connectivity between the OFC and these regions enhances cognitive processes important for the expression and regulation of NE.

We primarily focused on how WM-emotionality relationships were mediated/suppressed by rsFC. While it is well known that relationships between brain structure and function are reciprocal and important for global and integrative brain processes(28–30), the neurodevelopmental directionality of these relationships is less well understood and difficult to disentangle. Notably, our reverse-direction mediation analyses with WM volume as the mediator largely confirmed the nature of the tripartite relationships among WM, rsFC and PE and NE shown in our main findings.

Our initial GLMs indicated more robust effects of WM tract volume (i.e., CB volume-PE and UF volume-NE) than FA. Greater volume may indicate inefficient synaptic pruning(97), contributing to a greater density of collinear fibers and thus, more efficient – i.e., stronger - functional coupling (i.e., connectivity) among regions connected by the tract, or a greater degree of dendritic arborization, leading to greater fiber dispersion and more inefficient, dysfunctional – i.e., weaker - functional coupling among connected regions. By contrast, FA reflects a combination of fiber collinearity(AD) and dispersion(RD), which would mask the differential patterns of relationships between collinearity-based and dispersion-based volume and NE/PE; and neither AD- nor RD-based analyses alone would capture the complexity of these relationships(Supplement).

There were some limitations to this study; the sample size is modest, although post-hoc power analyses for the GLMs were robust, there was insufficient statistical power for testing moderating effects of sex or caregiver affect. While postnatal depression and affective instability were included as additional predictors, we did not account for effects of lifetime maternal psychiatric history on infant emotionality(107, 108). This sample was comprised of healthy, term-born infants; future studies should examine relationships among brain structure, function and infant emotionality in samples born pre-term or populations with greater levels of higher-risk prenatal exposures (e.g., poverty). Further, future studies can examine prospective relationships between neural measures and emotional behaviors, and how the development of the CB and UF predicts emotional behavior in infancy and later childhood. We also employed caregiver report of emotional behavior; future studies can examine more objective measures of infant behavior(e.g., independent observations). PE and NE were examined without accounting for other developmental processes(e.g., orienting, processing speed), thus, a potential limitation is the specificity of the observed brain-behavior associations(109, 110). Another limitation is the use of DWI-derived tract volume, which can be subject to distortions and rely on specific tractography parameters; these limitations are partially mitigated by corrections and consistent parameters, however, future replication is necessary. Further, neurite orientation dispersion and density imaging(NODDI)-derived measures are outside the scope of the current analysis, however, future work will include these to delve into differences in intracellular and extracellular compartments(111).

To our knowledge, these findings are the first to demonstrate that in 3-month-old infants, relationships between WM and emotional behavior are differentially suppressed or mediated by underlying rsFC within large-scale networks. Examining relationships among neural network WM microstructure and endogenous network function is thus important for identifying neural markers of emotional dysregulation, and can guide the development of interventions that modulate these structure-function relationships in order to reduce risk for future behavioral health problems later in childhood and adolescence.

KEY RESOURCES TABLE

Resource Type	Specific Reagent or Resource	Source or Reference	Identifiers	Additional Information
Add additional rows as needed for each resource type	Include species and sex when applicable.	Include name of manufacturer, company, repository, individual, or research lab. Include PMID or DOI for references; use “this paper” if new.	Include catalog numbers, stock numbers, database IDs or accession numbers, and/or RRIDs. RRIDs are highly encouraged; search for RRIDs at https://scicrunch.org/resources.	Include any additional information or notes if necessary.
Antibody
Bacterial or Viral Strain
Biological Sample
Cell Line
Chemical Compound or Drug
Commercial Assay Or Kit
Deposited Data; Public Database
Genetic Reagent
Organism/Strain
Peptide, Recombinant Protein
Recombinant DNA
Sequence-Based Reagent
Software; Algorithm	Mediation code available by request
Transfected Construct
Other