Sensorimotor network alterations in children and youth with prenatal alcohol exposure

Abstract

Children with prenatal alcohol exposure (PAE) often have impaired sensorimotor function. While altered brain structure has been noted in sensorimotor areas, the functional brain alterations remain unclear. This study aims to investigate sensorimotor brain networks in children and youth with PAE using resting‐state functional magnetic resonance imaging (rs‐fMRI). A parcellation‐based network analysis was performed to identify brain networks related to hand/lower limb and face/upper limb function in 59 children and youth with PAE and 50 typically developing controls. Participants with PAE and controls had similar organization of the hand and face areas within the primary sensorimotor cortex, but participants with PAE had altered functional connectivity (FC) between the sensorimotor regions and the rest of the brain. The sensorimotor regions in the PAE group showed less connectivity to certain hubs of the default mode network and more connectivity to areas of the salience network. Overall, our results show that despite similar patterns of organization in the sensorimotor network, subjects with PAE have increased FC between this network and other brain areas, perhaps suggesting overcompensation. These alterations in the sensorimotor network lay the foundation for future studies to evaluate interventions and treatments to improve motor function in children with PAE.

1. INTRODUCTION

Children and adults with prenatal alcohol exposure (PAE) often have impaired sensorimotor function including gross and fine motor problems, poor motor coordination, and sensory processing deficits (Barr, Streissguth, Darby, & Sampson, 1990; Carr, Agnihotri, & Keightley, 2010; Doney et al., 2014; Duval‐White, Jirikowic, Rios, Deitz, & Olson, 2013; Franklin, Deitz, Jirikowic, & Astley, 2008; Jirikowic, Olson, & Kartin, 2008; Jirikowic et al., 2013; Lucas et al., 2014; Mattson, Riley, Gramling, Delis, & Jones, 1998; Riley and McGee, 2005). Previous neuroimaging studies have found abnormal brain structure in the sensorimotor cortex (Archibald et al., 2001; Sowell et al., 2008, 2002; Yang et al., 2012) and its associated white matter connections (Lebel et al., 2008; Wozniak et al., 2009), and altered brain development trajectories of cortical volumes and thickness within the sensorimotor cortex (Lebel et al., 2012; Treit et al., 2014).

Despite the motor deficits and structural brain alterations, however, little is known about sensorimotor brain function in individuals with PAE. Animal studies show that PAE leads to altered limb representation in the somatosensory cortex (Margret et al., 2006), and increased functional connectivity (FC) between the somatosensory cortex and cerebellum (Rodriguez, Davies, Calhoun, Savage, & Hamilton, 2016). In humans, a functional magnetic resonance imaging (fMRI) study found children with PAE showed lower blood oxygen level dependent (BOLD) activation in the cerebellum compared to healthy controls during a finger tapping task, and implied that PAE is associated with deficits in sensorimotor processing (Du Plessis et al., 2015). In the only resting‐state fMRI (rs‐fMRI) study examining sensorimotor networks in PAE to date, 13 infants with PAE showed increased functional connectivity (FC) between the sensorimotor brain network and the striatum and brainstem compared to 14 controls (Donald et al., 2016), suggesting that PAE might lead to altered development of the sensorimotor system. The animal work and preliminary human study suggest altered sensorimotor organization and connectivity in children with PAE, but the specific nature of the alterations remain unclear. Furthermore, as previous studies have shown links between facial dysmorphology in children with PAE and brain structure (Coles & Li, 2011; Lebel et al., 2012; Roussotte et al., 2012; Yang et al., 2012), investigating links between sensorimotor brain function, particularly of the facial region, and facial dysmorphology in children with PAE is of interest.

Using connectivity‐based parcellation and functional connectivity analysis, this study aimed to investigate sensorimotor functional brain networks in children and youth with PAE using rs‐fMRI. We hypothesized that individuals with PAE have altered sensorimotor FC patterns, and that more abnormal patterns would be related to more severe facial dysmorphology. A better understanding of sensorimotor network function in children and youth with PAE will provide insight into the neurological basis of the observed motor deficits, and may provide brain targets for treatments and interventions.

2. METHODS & MATERIALS

2.1. Participants

A total of 186 participants aged 5–18 years were recruited as part of the Kids Brain Health Network (KBHN, previously NeuroDevNet) fetal alcohol spectrum disorder (FASD) study (Reynolds et al., 2011) from four research sites across Canada: University of Alberta (UA), Queen's University (QU), University of Manitoba (UM), and University of British Columbia (UBC). Seventy‐seven participants were excluded due to severe head motion (Section 2.3) and/or poor data quality, such as initial scan duration <5 min or lack of whole brain coverage. Thus, the final sample included 109 participants, with 50 healthy controls and 59 participants with PAE (27 with alcohol‐related neurodevelopmental disorder (ARND), 15 partial fetal alcohol syndrome (pFAS) or full FAS, and 17 unspecified diagnoses with confirmed PAE (Table 1)). Children were diagnosed at their local FASD clinic using the Canadian Guidelines for FASD diagnosis (Chudley et al., 2005) and 51 participants with PAE were assigned a 4‐Digit Diagnostic Code (Astley & Clarren, 2000). All control participants were free of diagnosed developmental disorders and FASD. This sample overlaps with previously published studies of structural brain abnormalities (Paolozza, Treit, Beaulieu, & Reynolds, 2014, 2017; Treit et al., 2016). Informed consent was obtained from a parent or guardian before scanning. This study was approved by the local health research ethics committee for each research facility.

Table 1.

Demographics of the participants included in this study

	University of Alberta		Queen's University		University of Manitoba		University of British Columbia
	Controls (n = 14)	PAE (n = 15)	Controls (n = 14)	PAE (n = 19)	Controls (n = 8)	PAE (n = 10)	Controls (n = 14)	PAE (n = 15)
Age (years)	11.7 ± 3.1	12.9 ± 3.2	14.1 ± 4.1	12.3 ± 3.5	11.5 ± 2.8	14.0 ± 1.5(*)	13.8 ± 2.4	12.3 ± 4.7
Sex (M/F)	6/8	6/9	8/6	7/12	3/5	5/5	4/10	7/8
Handedness (R/L)	12/2	10/5	13/1	17/2	7/1	9/1	13/1	12/3
Ethnicity:
First Nations/Caucasian/Other	0/12/2	8/3/4(**)	0/14/0	4/11/4(*)	0/8/0	4/1/5(**)	0/11/3	11/3/1(**)
Head motion (mm)	0.06 ± 0.03	0.07 ± 0.03	0.07 ± 0.04	0.06 ± 0.03	0.05 ± 0.02	0.07 ± 0.03	0.06 ± 0.02	0.07 ± 0.02

Head motion is given as relative root‐mean‐square frame‐wise displacement (see Section 2.3). Asterisks indicate significant group differences (PAE vs controls) within each site. *p < .05; **p < .01. There were no significant differences in demographics between sites. Ethnicity, but not age or head motion, was significantly different between PAE and control groups across all sites.

To address data consistency across sites, 8 healthy adults acted as traveling phantoms, and were scanned twice at each of the four sites.

2.2. MRI acquisition

MRI data were acquired at each site with protocols matched as closely as possible. Scanners were: 1.5 T Siemens Sonata at UA, 3 T Siemens Trio at QU and UM, and 3 T Philips Intera at UBC. See Supplementary Table 1 for full scanning parameters. The T1‐weighted structural images were acquired using axial MPRAGE with 160 slices and voxel size = 1 × 1 × 1 mm³ fMRI data was acquired using a gradient‐echo echo‐planar imaging (EPI) sequence with, TR = 2500 ms, TE = 30 ms (40 ms for 1.5T Siemens at UA), 40 axial‐oblique slices, voxel size = 3 × 3 × 3 mm³ and 140 volumes. See Supplementary Table 1 for full scanning parameters for each site. Participants were instructed to close their eyes and not think of anything particular during the rs‐fMRI scan.

2.3. Data preprocessing

All T1 images were skull stripped and segmented into grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using AFNI and FSL (Cox, 1996; Jenkinson, Beckmann, Behrens, Woolrich, & Smith, 2012), then co‐registered to the individual fMRI data space and registered to a Montreal Neurological Institute (MNI) standard pediatric brain template from the Imaging Research Center at Cincinnati Children's Hospital Medical Center (CCHMC), version 2 (9/2002), based on the age range 5–18 years (Jenkinson et al., 2012; Wilke, Schmithorst, & Holland, 2003). For the resting‐state fMRI data, the first 8 volumes were deleted to allow for MR signal stabilization and adaptation of participants. The following preprocessed procedures were used: slice acquisition timing correction, head motion correction, estimation of the co‐registration parameters to T1 image, and signal linear de‐trending, conducted in FSL and AFNI (Cox, 1996). The relative root‐mean‐square frame‐wise displacement (FD) and its mean were calculated using the FSL command mcflirt (Jenkinson, Bannister, Brady, & Smith, 2002). Then, the preprocessed fMRI signals were put into a head motion regression analysis. A 36‐parameter model was generated from the averaged signals from the individual whole brain, CSF mask, WM mask, the 6 head motion parameters, their temporal derivatives, and quadratic term signals (Ciric et al., 2016; Satterthwaite et al., 2013). Spike volumes were identified by high relative FD (>0.25 mm) and a spike volume vector was created for each fMRI dataset (Power et al., 2014; Satterthwaite et al., 2013). Then, the 36‐parameter model combined with the spike volume vector was regressed out of the preprocessed fMRI signals in AFNI. Participants with high mean relative FD (>0.25 mm) or spike volumes long enough to leave signals shorter than 5 min were excluded from further analysis (Long, Benischek, Dewey, & Lebel, 2017; Satterthwaite et al., 2013). Finally, the processed fMRI signals were band‐pass filtered (0.009–0.08 Hz), transformed to MNI standard pediatric template space and spatially smoothed with 4 mm full‐width at half‐maximum of the Gaussian kernel by AFNI and FSL. A GM mask (including 56528 voxels) was created with the combination of the consensus whole brain mask and the GM structures of the pediatric T1 image template.

2.4. Connectivity‐based parcellation and FC analysis

Brain regions related to sensorimotor function of specific body parts were identified using connectivity‐based parcellation (Eickhoff, Thirion, Varoquaux, & Bzdok, 2015), and used as seed regions for further brain network analysis. Connectivity‐based parcellation analysis allows researchers to identify functionally meaningful brain regions without asking participants to perform specific tasks (Cloutman and Lambon Ralph, 2012; Eickhoff et al., 2015), and has been used in previous studies to identify somatotopic organization within the human sensorimotor cortex (Gorbach et al., 2011; Gordon et al., 2014; Long, Goltz, Margulies, Nierhaus, & Villringer, 2014; Nebel et al., 2012; Power et al., 2011; Roca et al., 2010; Schubotz, Anwander, Knösche, Cramon, & Von Tittgemeyer, 2010; Yeo et al., 2011).

2.5. Experimental design and statistical analysis

Left and right sensorimotor cortex were extracted from a published brain template named the Brainnetome Atlas (Fan et al., 2016). Specifically, for each hemisphere, both Area 4 (head and face region) and Areas 1/2/3 (upper limb, head and face region) from the atlas were selected and combined as a single region‐of‐interest (ROI). Then left and right regions‐of‐interest (ROIs) were registered to the pediatric template in MNI space and resampled to the same resolution as the processed fMRI dataset (685/615 voxels within the left/right sensorimotor ROI). For each ROI, spatial correlation maps between the signal of each voxel within the ROI and all other voxels in the GM mask were calculated and transformed with Fisher's r‐to‐z transformation (Buckner, Krienen, & Yeo, 2013). An eta² value (range 0–1), which measures the similarity of two images by the differences of their voxels’ values, was calculated between each voxel pair's functional connectivity maps to create a similarity matrix for each individual (Cohen et al., 2008). An averaged similarity matrix was constructed across all participants. Then this matrix was classified by spectral K‐means algorithm into two clusters with 1000 iterations using the Spectral Clustering Toolbox (Verma, Verma, & Meila, 2003). The dorsal–lateral cluster was considered to be the hand/upper limb area; the ventral–lateral cluster was considered to be the face/lower limb area (Long et al., 2014; Power et al., 2011; Yeo et al., 2011).

Then, FC analysis was conducted separately for the hand/upper limb area and face/lower limb area ROIs. For each ROI, spatial correlation maps were generated between the averaged time courses within each ROI and time courses of other voxels within the GM mask. All spatial correlation maps were transformed to Fisher's z score maps. Therefore, four FC maps were created for each participant: left and right hand‐area FC maps (HFC) and face‐area FC maps (FFC). Random‐effects two sample t tests were performed across those four networks between the control and PAE groups (Fig. 1). Age, sex, site, handedness, and mean FD were included as covariates. Finally, the statistic map for each comparison was corrected for multiple comparisons at p = .05 with degrees of freedom (df) is 107 (i.e., voxel‐wise t(107) = 1.985, p = .0497, two sample t test, cluster size > 2889 mm³) using 3dClustSim with the estimated smoothing parameters by 3dFWHMx in AFNI (Version AFNI_16.2.12). The network comparisons were also performed individually for each site with age, sex, handedness, and mean FD as covariates. All results were displayed on a standard surface map by MATLAB‐based BrainNet Viewer toolbox (Xia, Wang, & He, 2013).

Figure 1.

The parcellation‐based network analysis. An atlas (a) was used to select the sensorimotor region (b, in yellow), and functional connectivity maps were calculated for each participant for each voxel in the sensorimotor region (c). From these, an individual similarity matrix is calculated (d), and the hand (yellow) and face (blue) areas within the sensorimotor area are separated based on the data (e). These are then used to produce functional connectivity maps with the rest of the cortex for each individual (f, g) that are then used for group‐level comparisons between the prenatal alcohol exposure (PAE) and healthy control groups (h). Red and blue colors represent positive and negative correlation coefficients with the seed region, respectively, in (c), (f), and (g) [Color figure can be viewed at http://wileyonlinelibrary.com]

Within all identified clusters, correlations were examined between averaged FC and the facial dysmorphology score from the FASD 4‐Digit Diagnostic Code in the PAE group. The facial dysmorphology score ranges from 1 to 4, where 1 represents no dysmorphology and 4 represents the presence of all 3 distinguishing facial features (smooth philtrum, thin upper lip, and short palpebral fissures) (Astley & Clarren, 2000). Correlation coefficients between face codes and FC were calculated using a robust correlation paradigm with the Robust Correlation Toolbox (Pernet, Wilcox, & Rousselet, 2013). Age, sex, site, handedness, and mean FD were regressed out of the FC value as covariates of noninterest. A similar correlation analysis was performed between age and average FC in regions with group differences; sex, site, handedness, and mean FD were included covariates of noninterest and regressed out of the FC value.

3. RESULTS

3.1. Data quality across sites

The temporal signal‐to‐noise ratio (tSNR) was measured using the mean divided by the standard deviation of the preprocessed time courses within a self‐defined mask (Murphy, Bodurka, & Bandettini, 2007), and was evaluated across all four sites. The tSNR mask is a 10‐mm‐radius sphere located in the medial frontal lobe in MNI standard space. There was no significant difference in tSNR between controls and PAE at the whole group level (t(107) = 0.41, p = .68, two sample t test), or for individual sites (UA: t(27) = −0.31, p = .76, two sample t test; QU: t(31) = −0.16, p = .87, two sample t test; UM: t(16) = 0.04, p = .97, two sample t test; UBC: t(27) = 1.47, p = .15, two sample t test).

Group‐level parcellation of the sensorimotor area in the traveling phantom data showed overlap of 72% for the left hemisphere and 84% for the right hemisphere across sites. The global average intraclass correlation for functional connectivity maps across sites was low (0.08–0.1), similar to previous studies (Kristo et al., 2014; Fiecas et al., 2013).

3.2. Somatotopic organization

The subregion organization within the sensorimotor cortex, derived from data‐driven clustering, was similar across controls, subjects with PAE, and the total group (Fig. 2). The overlap ratio, calculated by summing the number of overlapped voxels and dividing by total number of voxels within the sensorimotor ROI between controls and the PAE group‐level parcellation maps (Long et al., 2014; Nebel et al., 2012), was 95.2% and 93.7% for the left and right sensorimotor ROI parcellations, respectively. For both groups, the entire sensorimotor ROI (both parts) was contained within the post/precentral regions from the structural AAL brain template.

Figure 2.

The parcellation of the sensorimotor cortex in control subjects (top), participants with prenatal alcohol exposure (PAE; middle), and across all participants (bottom). Yellow/red corresponds to the hand/lower limb area, while blue corresponds to the face/upper limb area. Parcellations look very similar across both groups and the combined group; overlap was 94% and 95% for the right and left parcellations, respectively [Color figure can be viewed at http://wileyonlinelibrary.com]

3.3. Sensorimotor FC differences

Group differences were seen for all 4 seed regions (Fig. 3 and Table 2). The left face ROI had moderate positive connectivity to the left precuneus in controls, but a strong negative correlation in the PAE group. The left hand ROI had a strong positive correlation with the left superior temporal gyrus (STG) in controls, but negative correlations in the PAE group (Fig. 3).

Figure 3.

Significant differences of FC between PAE and healthy controls. Group differences were observed between the seed regions and frontal, parietal, and temporal areas. Warm colors indicate PAE > controls and cold colors indicate PAE < controls. The bar plots show the mean FC z value and its standard error within the detected brain regions. “All” means the group combined across four sites. * indicates p = .05, corrected. FFC = face functional connectivity; HFC = hand functional connectivity; STG = superior temporal gyrus; MFG = middle frontal gyrus; ACC = anterior cingulate cortex; POCG = postcentral gyrus [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 2.

Details of the significant regions shown in Figure 3, p = .05, corrected

	Left/right	Peak x, y, z (MNI)	Volume (mm3)	Peak T value
Left FFC seed
Precuneus	L	−8, −54, 48	4536	−3.77
Left HFC seed
STG	L	−62, −39, 6	3780	−4.87
Right FFC seed
Insula	R	34, 0, 15	4779	3.73
Right HFC seed
ACC	bilateral	13, 36, 21	5670	4.11
MFG	L	−38, 48, 27	3186	3.94
POCG	R	43, −42, 63	3186	4.01

Abbreviations: ACC = anterior cingulate cortex; FFC = face functional connectivity; HFC = hand functional connectivity; MFG = middle frontal gyrus; MNI = Montreal Neurological Institute coordinates space; POCG = postcentral gyrus; STG = superior temporal gyrus.

Positive t values indicate PAE > controls, and negative vice versa.

In the right face ROI, controls had reduced positive FC to the right insula compared to the PAE group. The right hand ROI had negative correlations with the left middle frontal gyrus (MFG) and bilateral anterior cingulate cortex (ACC) in controls, but positive correlations in the PAE group. The right postcentral gyrus (POCG) had positive FC with the right hand‐ROI in both groups, but less connectivity in controls (Fig. 3). The results were consistent across sites, with a few exceptions for the UM data (Fig. 3).

Handedness was included as a covariate in the analyses. However, to further ensure that handedness is not biasing our results, we conducted a supplementary analysis with the left‐handers removed. Results were very similar; only the right postcentral gyrus and the left superior temporal gyrus no longer met significance in group comparisons. Left handers represented 15% of the sample, so the lack of significance in these regions is likely due to reduced power.

Given the asymmetry of connectivity differences in the group results (PAE had lower connectivity from left ROIs and higher connectivity from right ROIs), a follow‐up analysis was conducted to investigate further. Patterns of connectivity from the left and right homologous regions were similar, even though group differences only met the significance threshold for one side (Supporting Information, Figures 1 and 2).

3.4. Alternate corrections for false‐positive rates

The latest version of AFNI (17.3.03) contains additional options for controlling false positive rates. Using the autocorrelation function (ACF) correction with voxel‐wise p < .01 and cluster size > 29 voxels, we found similar results in connectivity to the precuneus, superior temporal gyrus, middle frontal gyrus, and postcentral gyrus. Group differences in connectivity to the right insula and left anterior cingulate were no longer significant. Using the nonparametric randomization analysis (voxel‐wise p < .01, cluster size > 89 voxels), no group differences were significant.

3.5. Relationships between FC and facial dysmorphology or age

Among the participants with PAE, FC between the left face ROI and the left precuneus (where subjects with PAE had reduced FC compared to controls) was significantly negatively correlated with facial dysmorphology scores from the 4‐digit diagnostic code (p = .05, effect size = 0.10, Figure 4), indicating that subjects with worse dysmorphology had lower connectivity. No other regions had significant correlations with the face code.

Figure 4.

Correlation between the face code and FC between the left precuneus and the left face ROI in the PAE group only. A significant negative correlation was observed where children with worse facial dysmorphology (higher face code) had lower connectivity between the left face ROI and the left precuneus. CI = confidence interval of the bootstrap tests at p = .05. The prefix “r‐” of the y‐axis means the values were residual after the covariates regression. Scatter solid dots depict individual values and trend lines (bold red lines) [Color figure can be viewed at http://wileyonlinelibrary.com]

Only one region had significant age‐group interactions (controls: p = .05 at bootstrap, effect size = 0.15, PAE: effect size = 0.0004, Figure 5). FC between the right face ROI and the right insula showed significant increases with age in controls, but no age‐related changes in subjects with PAE.

Figure 5.

Correlations between age and FC in controls (green) and participants with PAE (yellow). Significant age‐group difference was observed in FC between the right face ROI and the right insula. CI = confidence interval of the bootstrap tests at p = .05. The prefix “r‐” of the y‐axis means the values were residual after the covariates regression. Green round shape dots with a bold green line are controls and yellow round shape dots with a bold mustard line are PAE participants [Color figure can be viewed at http://wileyonlinelibrary.com]

4. DISCUSSION

Our study investigated the organization and functional connectivity of the sensorimotor network in children and youth with PAE using multisite rs‐fMRI data. The PAE group had more widespread functional connectivity from the right sensorimotor area to other brain regions than controls, although the PAE group had isolated regions of reduced connectivity between left sensorimotor regions and certain default mode network regions. FC between the right insula and the sensorimotor subregion showed differential age‐related changes in controls and individuals with PAE. Furthermore, connectivity to the left precuneus was related to facial dysmorphology in the individuals with PAE.

4.1. Somatotopic organization within the sensorimotor cortex

In this study, we employed FC‐based parcellation analysis to identify subregions within the sensorimotor cortex (Figure 2). FC patterns of motor function are among the earliest identifiable patterns by rs‐fMRI are found to be similar to those seen during task activation (Biswal, Yetkin, Haughton, & Hyde, 1995), and are in line with known somatotopic organization (Heuvel & Pol, 2010). Our parcellation results are consistent with the previous task/rest findings, and are assumed to be related to the sensorimotor function of hand/lower limb and face/upper limb, respectively (Long et al., 2014; Meier et al., 2008; Yeo et al., 2011). The somatotopic organization of the hand/lower limb and face/upper limb was similar between controls and the PAE group, suggesting no major disruptions to the sensorimotor organization in individuals with PAE, despite deficits in a wide range of motor skills (Doney et al., 2014). In contrast to the preserved organization observed here, an animal study found altered somatotopic organization (reduced forepaw representation) within the sensorimotor cortex of rats with PAE using electrophysiology (Margret et al., 2006). The different results may be due to more severe alcohol exposure in the rats than is typical in human PAE participants, and/or subtle differences that are not apparent on functional MRI.

4.2. Sensorimotor connectivity in PAE

The PAE group had higher FC from sensorimotor regions to areas of the sensorimotor network (POCG), and parts of the salience network (insula, anterior cingulate) (Sridharan, Levitin, & Menon, 2008; Uddin, 2016). The PAE group also had higher connectivity to the MFG, which is typically associated with the executive network (Raichle, 2015; Young et al., 2017). Decreased FC in the PAE group compared to controls was observed between sensorimotor seed regions and parts of the default mode network, specifically the STG and the precuneus. Though there seemed to be some asymmetry in the results, further analysis showed that brain regions had similar trends in left and right subregions, suggesting that the results are more about regional variation in the network rather than striking asymmetry.

The only other study to investigate sensorimotor networks in PAE found increased FC between bilateral sensorimotor networks and other brain networks in infants (Donald et al., 2016). The increased FC observed in our study, and the previous study in infants, suggests that individuals with PAE may recruit additional brain regions for sensorimotor processing, leading to lower network efficiency. In support of this, reduced global network efficiency has been previously observed in children with PAE (Wozniak et al., 2013). The motor network generally shows decreased FC in children with other neurodevelopmental disorders, such as autism spectrum disorder (Mostofsky et al., 2009), developmental coordination disorder, and attention‐deficit/hyperactivity disorder (McLeod, Langevin, Goodyear, & Dewey, 2014; McLeod, Langevin, Dewey, & Goodyear, 2016), suggesting that reduced FC is associated with deficits in motor control, regulation, and planning. The contrasting results in our study suggest that PAE may impact motor networks via different mechanisms than other neurodevelopmental disorders, leading to increased, rather than decreased functional connectivity.

The regions highlighted here are linked to cognitive processes that are impaired in PAE. Children with PAE often have arithmetic processing difficulties (Riley and McGee, 2005), and previous studies have shown altered activity during calculation tasks in the precuneus, right insula, right MFG, and POCG in children and adults with PAE (Meintjes et al., 2010; Santhanam, Li, Hu, Lynch, & Coles, 2009; Santhanam et al., 2011; Woods, Meintjes, Molteno, Jacobson, & Jacobson, 2015), all regions observed here to have altered connectivity with the sensorimotor regions. The ACC showed higher FC to the right hand sensorimotor region in PAE, also shows higher activation during inhibition tasks (O'Brien et al., 2013), and altered structure in children and adolescents with PAE compared to controls (Bjorkquist, Fryer, Reiss, Mattson, & Riley, 2010; Migliorini et al., 2015). The insula and STG are involved in speech production and language performance (Behroozmand et al., 2015; Oh, Duerden, & Pang, 2014), which are known to be impaired in children with FASD (McGee, Bjorkquist, Riley, & Mattson, 2009; Wyper and Rasmussen, 2011).

The precuneus and insula are hubs of the default mode network and the salience network, which are involved in motor function, cognitive control, emotions, and consciousness (Cavanna, 2007; Fransson and Marrelec, 2008; Margulies et al., 2009; Menon and Uddin, 2010; Utevsky, Smith, & Huettel, 2014). Altered FC between those hubs and the face/upper limb sensorimotor area may indicate abnormal interactions between cognitive and motor function (Wenderoth, Debaere, Sunaert, & Swinnen, 2005) in individuals with PAE, which may underlie deficits in speech, social cognition, and sensation modulation (Cone‐Wesson, 2005; Kerns, Siklos, Baker, & M?Ller, 2016; Oberlander et al., 2010). The precuneus is also associated with face processing related to emotion and memory (Fusar‐Poli et al., 2009; Gobbini and Haxby, 2007; Utevsky et al., 2014), a function that is known to be impaired in individuals with PAE (Greenbaum, Stevens, Nash, Koren, & Rovet, 2009; Wheeler, Stevens, Sheard, & Rovet, 2011).

Several of the regions observed here are also important for sensorimotor processing. The precuneus is involved in motor imagery and manual coordination (Cavanna and Trimble, 2006), while the STG is part of the auditory cortex; both of these had reduced FC to sensorimotor regions in participants with PAE. The POCG is a key part of the sensorimotor cortex, and was observed here to have higher FC to sensorimotor ROIs in the PAE group.

Our current findings were robust across research sites, providing additional confidence in our results. Multisite neuroimaging studies in FASD are becoming more common as a way to increase sample sizes and improve generalizability of findings (Mattson et al., 2010; Reynolds et al., 2011), but it is important to ensure that the additional variability does not bias results.

4.3. Relationship between FC and facial dysmorphology

Previous studies found worse facial dysmorphology is associated with more severe structural brain abnormalities (Lebel et al., 2012; Roussotte et al., 2012; Yang et al., 2012), suggesting that the face is a “window” to the human brain (Fryer, 2012). Facial dysmorphology is also associated with other symptoms, including intelligence and language ability, in individuals with PAE (Astley & Clarren, 2001). Here, we observed that FC between the left face sensorimotor area and the left precuneus had a significant negative correlation with facial dysmorphology (Figure 4). Previous findings report relationships between facial dysmorphology and cortical volume in the precuneus (Lebel et al., 2012), which may underlie relationships between dysmorphology and functional connectivity. Interestingly, facial dysmorphology was not associated with functional connectivity from the hand region, suggesting some specificity for predicting facial sensorimotor connectivity within the brain.

4.4. Altered sensorimotor FC development in PAE

FASD is a neurodevelopmental disorder that impairs the development of cognition, emotion, and motor abilities in children (Jacobson and Jaconson, 2002; Kalberg et al., 2006; Riley et al., 2011). Several studies have found altered development trajectories of brain structure in children and adolescents with PAE compared to healthy controls (Gautam et al., 2015; Lebel et al., 2012, 2008; Treit et al., 2016, 2014), including in the sensorimotor areas studied here. In our study, FC between the facial sensorimotor region and the right insula showed different age‐related changes in PAE compared to healthy controls: FC increased in controls but not in those with PAE. The insula's FC to brain frontal regions increases from childhood to adolescence (Hwang, Hallquist, & Luna, 2013), which is in line with the present finding in controls. The altered development of FC in individuals with PAE may underlay their altered cognitive‐motor development (Adnams et al., 2001). Furthermore, due to the high neuroplasticity of facial sensorimotor function (Avivi‐Arber, Lee, Yao, Adachi, & Sessle, 2010; Avivi‐Arber, Martin, Lee, & Sessle, 2011), this FC might be an potential biomarker to evaluate brain development and interventions to improve sensorimotor function in the FASD population.

4.5. Limitations

AFNI and other fMRI analysis software can be prone to high rates of false‐positives (Cox, Chen, Glen, Reynolds, & Taylor, 2017; Eklund, Nichols, & Knutsson, 2016). While we used a recent version of AFNI that corrects for a previous bug that inflated false positives, some of these results may still be false positives. When even more stringent correction for multiple comparisons was used, some (using ACF) or all (using nonparametric randomization) of the group differences were no longer significant. These stringent corrections are much stricter than used in the vast majority of previous literature, and trade lower rates of false positives for higher rates of false negatives. Therefore, future studies will need to replicate our findings to determine which ones are most robust. Furthermore, correlations with facial dysmorphology and age were only found in limited areas, and could be spurious findings. Future studies will help clarify the nature of relationships between these functional brain alterations and dysmorphology and age in children and adolescents with PAE.

We did not collect a measure of sensorimotor function in participants, so the relationship between alterations in the sensorimotor network of children and youth with PAE and participants’ motor or sensory deficits is still unclear. Future studies incorporating measures of sensorimotor function could provide more information on behavior–brain relationships. We extracted our measure of facial dysmorphology from the 4‐digit code, which only includes 4 severity levels (Astley & Clarren, 2000). More specific metrics, including lipometer and philtrum scales, and direct measures of palpebral fissure length, could potentially be more sensitive to links between brain function and dysmorphology. Finally, the control and PAE groups were significantly different in their ethnicity distributions, which may introduce a confound to the data analysis.

5. CONCLUSION

In conclusion, children and youth with PAE had similar organization of the hand and face regions within the primary sensorimotor cortex compared to controls, but functional connectivity from the sensorimotor cortex to other brain regions was altered. The PAE group had higher functional connectivity from sensorimotor seed regions to specific areas of the sensorimotor network, salience network, and executive network, suggesting recruitment of additional brain areas for sensorimotor function in individuals with PAE. The PAE group also had lower functional connectivity to some areas of the default mode network. Our study highlights the brain abnormalities underlying altered sensorimotor function in children and youth with PAE, and lays a foundation for future studies of novel interventions and treatments designed to improve sensorimotor function for children with PAE.

DISCLOSURES

CL's spouse is an employee of General Electric Healthcare; other authors report no conflict of interest.

Supporting information

Additional Supporting Information may be found online in the supporting information tab for this article.

Supporting Information Figure 1

Supporting Information Figure 2

Supporting Information Figure Legends

Supporting Information Table 1

Long X, Little G, Beaulieu C, Lebel C. Sensorimotor network alterations in children and youth with prenatal alcohol exposure. Hum Brain Mapp. 2018;39:2258–2268. 10.1002/hbm.24004