Performance ramifications of abnormal functional connectivity of ventral posterior lateral thalamus with cerebellum in abstinent individuals with Alcohol Use Disorder

Abstract

The extant literature supports the involvement of the thalamus in the cognitive and motor impairment associated with chronic alcohol consumption, but clear structure/function relationships remain elusive. Alcohol effects on specific nuclei rather than the entire thalamus may provide the basis for differential cognitive and motor decline in Alcohol Use Disorder (AUD). This functional MRI (fMRI) study was conducted in 23 abstinent individuals with AUD and 27 healthy controls to test the hypothesis that functional connectivity between anterior thalamus and hippocampus would be compromised in those with an AUD diagnosis and related to mnemonic deficits. Functional connectivity between 7 thalamic structures [5 thalamic nuclei: anterior ventral (AV), mediodorsal (MD), pulvinar (Pul), ventral lateral posterior (VLP), and ventral posterior lateral (VPL); ventral thalamus; the entire thalamus] and 14 “functional regions” was evaluated. Relative to controls, the AUD group exhibited different VPL-based functional connectivity: an anticorrelation between VPL and a bilateral middle temporal lobe region observed in controls became a positive correlation in the AUD group; an anticorrelation between the VPL and the cerebellum was stronger in the AUD than control group. AUD-associated altered connectivity between anterior thalamus and hippocampus as a substrate of memory compromise was not supported; instead, connectivity differences from controls selective to VPL and cerebellum demonstrated a relationship with impaired balance. These preliminary findings support substructure-level evaluation in future studies focused on discerning the role of the thalamus in AUD-associated cognitive and motor deficits.

1. Introduction

The thalamus is a significant node in neural circuits compromised by Alcohol Use Disorder (AUD) (Pitel et al., 2015; Segobin et al., 2019). Global thalamic volume shrinkage in AUD (Cardenas et al., 2007; Chanraud et al., 2007; Mechtcheriakov et al., 2007; Pitel et al., 2012; Sullivan, 2003) is associated with deficits in episodic (Fama et al., 2014; Sullivan et al., 2003) and working (Chanraud et al., 2010) memory. Clinical (e.g., stroke) and preclinical (e.g., lesion) studies implicate the anterior thalamus as part of a corticolimbic (Papez) circuit comprising hippocampus, fornix, and mamillary bodies and as a substrate of episodic and working memory (Aggleton and Brown, 1999; Bubb et al., 2017; Fama and Sullivan, 2015; Tanaka et al., 2020). Postmortem neuropathological studies suggest that the anterior thalamus is preferentially compromised in AUD with (Harding et al., 2000) or without (Belzunegui et al., 1995) amnesia. The aim of the current study was to use resting-state functional Magnetic Resonance Imaging (rs-fMRI) [i.e., fluctuations of low frequency blood oxygenation level-dependent (BOLD) signals synchronized among functionally related brain regions] data to evaluate whether altered functional connectivity between the anterior thalamus and hippocampus underlies AUD-associated deficits in working memory.

fMRI studies in AUD support a role for the thalamus in measures of craving and AUD severity (e.g., George et al., 2001), but not working memory. For example, an fMRI study in alcoholic-dependent relative to healthy control individuals showed higher working memory-related activation of the dorsal anterior cingulate cortex; activation of the thalamus, however, was associated with higher scores on the Obsessive-Compulsive Drinking Scale (Vollstadt-Klein et al., 2010). Similarly, during exposure to alcohol or neutral cues, bilateral thalamic responses to cues in nondependent drinkers correlated with scores on the Alcohol Use Disorders Identification Test (AUDIT) (Zhornitsky et al., 2019) (also see Ide et al., 2018; Zhornitsky et al., 2018). By contrast, non-dependent binge drinkers relative to matched controls showed greater activation in cerebellum, thalamus, and insula while performing a working-memory task (Campanella et al., 2013)

Alcohol effects on specific nuclei rather than the entire thalamus may explain a lack of consensus on the role of the thalamus in AUD. The thalamus has been parcellated using high-resolution structural MRI (Iglesias et al., 2018; Liu et al., 2019; Su et al., 2019), diffusion tensor imaging (DTI) (Behrens et al., 2003; Duan et al., 2007; Jakab et al., 2012; Johansen-Berg et al., 2005; Kumar et al., 2015; Mang et al., 2012; O’Muircheartaigh et al., 2015; Stough et al., 2014; Wiegell et al., 2003; Ziyan et al., 2006), and functional connectivity measures derived from rs-fMRI. Small, in vivo rs-fMRI studies in healthy human subjects have demonstrated functional connectivity between thalamus and subcortical structures such as hippocampus (Stein et al., 2000) and basal ganglia (Lenglet et al., 2012). To date, however, functional connectivity strength between anterior thalamus and hippocampus as a substrate for working memory in AUD has not been evaluated. We therefore hypothesized that functional connectivity between anterior thalamus and hippocampus would be compromised in AUD relative to healthy controls and related to performance on tests of working memory.

2. Materials and Methods

2.1. Participants

The Institutional Review Boards of Stanford University and SRI International approved this study. In accordance with the Declaration of Helsinki, all participants provided written informed consent by signing relevant documents in the presence of appropriately trained staff. Study participants were 23 individuals diagnosed with AUD (6 women) and 27 healthy controls (12 women) (Table 1). Individuals with AUD were referred from local treatment centers or, like the healthy control participants, were recruited from the local community by referrals and flyers. Specifically, of 23 AUD participants, 9 were recruited from treatment centers or shelters (e.g., Free at Last, East Palo Alto; Project 90, San Mateo; WeHOPE shelter, East Palo Alto), 6 were referred by community members (e.g., friend or other study participant), 4 were referred by Palo Alto VA physicians, 1 responded to a flyer, and 3 were recruited via unknown sources. AUD and excluding diagnoses were determined using the Structured Clinical Interview for DSM-5 (American Psychiatric Association, 2013); a semi-structured timeline follow-back interview quantified lifetime alcohol consumption (Skinner and Sheu, 1982); the Clinical Institute Withdrawal Assessment of Alcohol (CIWA) scale was also administered. Upon initial assessment, subjects were excluded if they had a significant history of medical (e.g., epilepsy, stroke, multiple sclerosis, uncontrolled diabetes, or loss of consciousness >30 minutes), psychiatric (i.e., schizophrenia or bipolar I disorder), or neurological (e.g., Parkinson’s) disease. Table 1 summarizes the demographic information of the two groups. The AUD group had drunk an average of 34.5±43.9kg of ethanol in the past year. Of 23, 2 AUD subjects had moderate and 20 had severe DSM-5 symptoms (data unavailable for 1 AUD participant); 4 AUD individuals had been to urgent care at least once for detoxification and 18 had been arrested at least twice. Liver enzymes were all in normal range [mean±SD U/L (Quest reference range): alkaline phosphatase (ALP) 90.4±37.6 (40–115); alanine aminotransferase (ALT) 22.7±15.0 (9–46); aspartate aminotransferase (AST) 25.0±14.8 (10–35); and gamma-glutamyl transferase (GGT) 55.8±108.2 (3–70).

Table 1.

Characteristics of the study groups: mean±SD / frequency count

	Control (n=27)	AUD (n=23)	p-value*
N (men/women)	15/12	17/6	n.s.
Age (years)	50.6±10.1	52.8±7.9	n.s.
Handedness (Right/Left)	26/1	18/5	n.s.
Ethnicity1	15/5/7	13/7/3	n.s.
Body Mass Index (BMI)	24.8±3.4	28.8±5.5	0.005
Education (years)	16.2±2.4	12.5±1.5	<.0001
Socioeconomic Status (SES)2	21.9±17.7	45.5±12.6	<.0001
WTAR IQ (predicted full scale)	104.7±9.9^	94.6±10.0+	0.002
Beck Depression Index (BDI) scores	3.8±4.1^	12.9±9.6	0.0003
Smoker (never/past/current)	24 / 1 / 2	5 / 4 / 14	<.0001
Global Assessment of Functioning (GAF)	87.6±5.3^	65.2±9.3	<.0001
Hepatitis C Virus (HCV, positive/negative)	2/25	4/19	n.s.
Lifetime alcohol consumption (kg)	54.0±80.7#	1842.5±1662.4	<.0001
AUD onset age	-	21.7±6.7	n.a.
days since last drink	-	111.7±93.0	n.a.
CIWA3 scale	-	6.2±16.7	n.a.

^*t-tests used on continuous variables (e.g., age);

χ2 used on nominal variables (e.g., handedness);

¹self-defined: Caucasian / African American / Other (Asian, Native American, Islander);

²lower score = higher status;

³Clinical Institute Withdrawal Assessment of Alcohol;

^#n=19,

^{^}n=18,

⁺n=21

2.2. Cognitive and Motor Testing

Participants completed a comprehensive neuropsychological battery to assess working memory, memory & learning, visuospatial abilities, and executive functions (cf., Zahr et al., 2019). Raw scores on individual neuropsychological tests (listed below) were statistically corrected for age of the control group [mean and standard deviation for control group = 0 ± 1], allowing averaging across tests. Composite scores were then calculated as the mean of Z-scores of tests comprising each of the functional domains. Working Memory: Wechsler Memory Scale-Revised (WMS-R) block forward total; WMS-R block forward span. Memory & Learning: Wechsler Adult Intelligence Scale (WAIS) digit symbol incidental recall of symbols; WAIS digit symbol incidental recall of numbers; California Verbal Learning Test (CVLT) short delay free recall; CVLT long delay free recall; Montreal Cognitive Assessment (MOCA) delayed recall; WMS-R logical memory story A raw score; WMS-R logical memory story B raw score. Visuospatial Abilities: Rey-Osterrieth copy raw score; WMS-R visual reproduction item 1 raw score; WMS-R visual reproduction item 2 raw score. Executive Functions: Controlled Oral Word Association Test (F+A+S total); Semantic fluency (inanimate objects, animals); WAIS digit symbol total time to complete set; WAIS digit symbol standard score at 90s; MOCA abstraction score.

Additionally, standing balance was assessed using an ataxia battery (stand heel to toe, walk 10 steps on a line, balance on one leg) in eyes open and closed conditions (Fregly, 1968). Age-corrected Z-scores were summed for the eyes open and closed conditions separately (Sullivan et al., 2000).

2.3. MRI Acquisition

Scanning was performed on a 3.0 Tesla MRI scanner [MR750, General Electric (GE) Healthcare, Waukesha, WI] with a 32-channel receive array head coil. A T1-weighted inversion-recovery prepared spoiled gradient-recalled (SPGR) sequence [repetition time (TR) = 5.904 – 6.148ms, echo time (TE) = 1.932 – 1.984ms, inversion time (TI) = 300 – 400ms, matrix = 256×256, thickness = 1.25mm, skip = 0mm, 124 slices, spatial resolution 0.7 mm × 0.7 mm × 1.0 mm] was used for structural scans. Resting-state functional MRI (rs-fMRI) scans (eyes open, 8:55min acquisition time, 1.71×1.71×3mm spatial resolution, 200 time points, TE = 30ms, dwell time 0.28ms) used different TRs ranging from 2.4 to 2.86s [2.4s n=1 control; 2.648s n=23 control + n=16 AUD participants; 2.754s for n=2 control + n=2 AUD participants; 2.86s n=1 control + 5 AUD participants]. As the proportion of scans acquired using the four different TRs was similar between control and AUD groups (χ²=4.6, p = 0.20) and as preliminary statistical analysis showed no significant effects of differing relaxation times, this variable was not further considered in analysis.

2.4. Structural MRI Processing

Structural T1-weighted MRI images were denoised and skull stripped (Coupe et al., 2008). The skull-stripping brain mask was generated by performing majority voting across segmentations generated by FSL (v5.0.6) BET (Smith, 2002), AFNI (v16.1.15) 3dSkullStrip (Cox, 1996), and Robust Brain Extraction (ROBEX v1.2)(Iglesias et al., 2011). T1-weighted images were corrected for field inhomogeneity using ANTS (v2.1.0) N4ITK (Avants et al., 2014; Tustison et al., 2010). The brain mask was further refined by applying the described segmentation methods and FreeSurfer (v5.3.0) mri_gcut (Sadananthan et al., 2010) to the inhomogeneity-corrected images and performing majority voting (Rohlfing et al., 2004). After skull stripping, images were segmented into 3 tissue types [grey matter, white matter, and cerebrospinal fluid (CSF)] using ANTS atropos (Avants et al., 2014; Avants et al., 2011).

2.5. Functional MRI Processing

Each acquired rs-fMRI brain volume was up-sampled to 2.5mm isotropic spatial resolution via ITK-SNAP (v1.0.0) C3D (Yushkevich et al., 2006) and motion-corrected using FSL MCFLIRT (Jenkinson et al., 2002; Jenkinson et al., 2012). Volumes were rejected from further analysis if their average frame-to-frame, in-scanner motion was greater than 0.75mm or if their scan duration (after removing volumes corrupted by frame-to-frame motion greater than 0.3mm) was below 180s (average duration = 6:24min). Mean blood oxygen level dependent (BOLD) images were non-rigidly registered to SRI24 atlas space (1.5mm isotropic resolution)(Rohlfing et al., 2010) using ANTS symmetric diffeomorphic non-rigid registration (Avants et al., 2008) as suggested (Calhoun et al., 2017; Dohmatob et al., 2018). BOLD images passing quality control were further processed using Nipype (v11.0) resting-state specific analysis (Gorgolewski et al., 2011). Specifically, a linear regressor was constructed that combined the motion outliers detected by Nipype rapidart with detrending parameters (normalized threshold=0.3; intensity Z-threshold=5). To interpolate the removed frames (i.e., motion outliers), the regressor was defined as a general linear model (GLM) via FSL. The pipeline then estimated and corrected for physiological noise via Nipype CompCor (Behzadi et al., 2007) and FSL GLM. The series were processed by a discrete Fourier transform bandpass filter [low pass frequency: 0.1, high pass frequency: 0.01; Numpy v1.16.2 (http://www.numpy.org/)] for temporal smoothing. The corrected BOLD images were non-rigidly aligned to SRI24 atlas space (Rohlfing et al., 2010) by applying the previously-computed transformation between the mean BOLD images and atlas space. Finally, the aligned BOLD images were spatially smoothed with a Gaussian filter of 5.0mm full-width half maximum (FWHM, (Mikl et al., 2008) implemented by FSL fslmaths (Jenkinson et al., 2012).

2.6. Thalamus Atlas

A high-resolution atlas of the thalamus (Saranathan et al., 2019; Saranathan et al., 2020) was converted from Montreal Neurological Institute (MNI) to SRI24 (Rohlfing et al., 2010) atlas space using ANTS symmetric diffeomorphic non-rigid registration (Avants et al., 2008; Avants et al., 2014). For the whole thalamus, the ventral thalamus, and 5 thalamic nuclei [anterior ventral (AV), mediodorsal (MD), pulvinar (Pul), ventral lateral posterior (VLP), ventral posterior lateral (VPL)], the functional connectivity to the other brain regions were determined. The 5 nuclei are shown in Figure 1.

Figure 1.

Thalamus atlas registered to SRI24 atlas at 1.5mm isotropic resolution showing 5 thalamic nuclei: anteroventral (AV, total volume: 525 mm³), mediodorsal (MD, total volume: 1933 mm³), pulvinar (Pul, total volume: 3826 mm³), ventral lateral posterior (VLP, total volume: 2519 mm³), ventral posterior lateral (VPL, total volume: 857 mm³).

2.7. Functional Connectivity

The SRI24 atlas defines 111 cortical and subcortical (54 in each hemisphere and three bilateral cerebellar) gray matter regions of interest (ROIs) (Rohlfing et al., 2010). For each ROI, the average BOLD signal across the entire brain was regressed out from the average regional BOLD signal and the resulting time series was normalized with a mean=0 and a standard deviation=1. The functional connectivity of each participant was then encoded by a correlation matrix, whose entries were defined by inner products between the normalized time series of two regions, refined by Oracle Approximating Shrinkage (Chen et al., 2010) and Fisher z-transformed (Fisher, 1915). Given that the study consisted of just 50 participants, evaluation of functional connectivity between 7 thalamic regions and 111 ROIs with appropriate statistical corrections for multiple comparisons would mask potentially significant AUD effects. Thus, the 111 ROIs were aggregated into larger “functional regions” by combining ROIs with strong functional correlations. Specifically, the average functional connectivity matrix (across the 111 ROIs) across all participants was computed. The positive entries of this average matrix were considered similarity measures and used as inputs to the Louvain method for community detection to determine “functional regions”. The Louvain method for community detection is a greedy graph clustering method that aggregates network nodes according to their connectivity to create a partition of the network into densely connected communities. The output of this algorithm depends on a single parameter that influences the size (and therefore the number) of communities determined by the method (Blondel et al., 2008; Yeo et al., 2011). For each “functional region” and subject, the average BOLD time series was calculated, then correlated to the average bold signal of each thalamic structure; correlations were then Fisher z-transformed (Fisher, 1915).

2.8. Statistical Analysis

For each thalamic structure and “functional region” pair, a linear model predicting connectivity based on AUD diagnosis, age, or sex was fit using python (v6.1) statsmodels api (Seabold and Perktold, 2010). Coefficients with p<0.00357 after correcting for multiple comparisons [i.e., false discovery rate (FDR), p=.05/14 regions, see Results for details] were considered significant (Benjamini and Hochberg, 1995). Post-hoc analysis – considering only connectivity measures significantly affected by AUD – used connectivity corrected for age and sex [i.e., residuals with effects removed via the robust linear regression function of the MASS library in R v3.3.0 (R Core Team, 2013) and the MM estimation method (Hampel, 1986)]. Variables distinguishing the two groups were evaluated for their contribution to residual connectivity values using AIC stepwise regressions. Follow up analyses also included multiple regressions.

3. Results

3.1. Functional Connectivity

Louvain-based aggregation of the 111 SRI24 ROIs resulted in 10 parcellations with a varying number of regions included in each parcellation (Supplementary Tables 1 and 2, Supplementary Figure 1). The parcellation including 14 regions optimized connectivity measures affected by AUD (Supplementary Table 1); too few regions resulted in noise in the connectivity measures, too many required strict FDR-corrected p-values. Subsequent analyses are thus based on 14 “functional regions”. Population (across all 50 participants) average connectivity between the 7 thalamic structures and 14 functional regions are presented in Figure 2A. An AUD diagnosis was significantly associated with three connectivity changes. Specifically, an AUD diagnosis changed a weak and non-significant anticorrelation among the population average to a positive correlation between the entire thalamus and a bilateral middle temporal lobe region (i.e., area 13: middle temporal left and right, inferior temporal left) and between VPL thalamus and the same bilateral middle temporal lobe region (Figure 2B–C, 3, 4). An AUD diagnosis also altered connectivity between VPL and the cerebellum (i.e., area 14: left and right cerebellar crus 1; left and right cerebellar crus 2; left and right cerebellar regions 3–10, cerebellar vermis 1–3; Figure 2B–C, 3, 4) but in this case strengthened the existing (non-significant) anticorrelation among the population average. Connectivity of the remaining thalamic structures were not significantly affected by an AUD diagnosis. Functional connections were also not significantly affected by age or sex (Figure 2D–E).

Figure 2.

A) Population average (AUD + control participants) connectivity between thalamic structures and 14 functional regions; red=correlated, blue=anticorrelated. B) Coefficients of linear regression accounting for AUD effects on connectivity between thalamic structures and 14 functional regions; significant AUD effects surviving FDR correction requiring p=.00357 (p=.05/14 regions) are highlighted in C) presenting only AUD effects passing FDR correction for 14 regions. Population average (AUD + control participants) connectivity between thalamic structures and 14 functional regions as a function of D) age and E) sex. Neither age nor sex significantly affected connectivity.

Figure 3.

Altered connectivity between control and AUD groups demonstrating a disrupted anticorrelation between entire thalamus (A) and VPL thalamus (B) to area 13 (middle temporal lobes); and (C) a strengthened anticorrelation between VPL and area 14 (cerebellum).

Figure 4.

Regions affected: A) entire thalamus (dark green) and middle temporal regions (orange, area 13); B) VPL thalamus (green) and middle temporal regions (orange, area 13) and cerebellum (yellow, area 14).

Support for these findings comes from a comprehensive functional connectivity analysis across the Louvain-based parcellations. Significant AUD effects on the functional connectivity between the entire thalamus and a “functional region” including bilateral middle temporal lobes was evident among 3 of the 10 Louvain-based parcellations. Similarly, AUD altered the connectivity between VPL and a bilateral middle temporal lobe region among 4 of the parcellations and between VPL and the cerebellum among 9 of the 10 parcellations. These findings are summarized in Supplementary Figure 2 and Supplementary Table 3.

3.2. Functional Connectivity: Relationships with Relevant Variables

Further analyses were conducted in the AUD group only and evaluated relationships between variables differentiating the two diagnostic groups and connectivity measures corrected for age and sex (i.e., residuals). Of the 7 thalamic volumes evaluated, VLP volume was uniquely sensitive to alcoholism (smaller in AUD, t=3.1, p=.004). The entire thalamus (t=−2.1, p=.04) and VPL (t=−2.2, p=.03) were also smaller in AUD relative to healthy controls, but neither comparison survived a Bonferroni correction (i.e., 0.05/7 regions would require p=.007). Nevertheless, a follow-up AIC stepwise regression to determine whether functional connectivity measures were related to structural volumes included volumes of the VLP, VPL, and entire thalamus. Neither of the VPL connectivity measures correlated with volume (all p-values >.23); connectivity strength between the entire thalamus and temporal regions, however, was correlated with VLP volume (r=−0.34, p=.11; ρ=−0.42, p=.05).

Next, a stepwise regression was conducted for each of the 3 connectivity measures considering demographic variables that distinguished the 2 groups (from Table 1: BMI, Education, SES, WTAR IQ, BDI scores, smoking status, and GAF). Connectivity of the entire thalamus with temporal regions was not modulated by these variables. VPL connectivity measures were only affected by SES: VPL and temporal regions (r=.42, p=.05; ρ=.43, p=.04); VPL and cerebellar regions (r=−.40, p=.06; ρ=−0.48, p=.02). Follow-up stepwise regressions demonstrated that diagnosis (but not age, sex, or SES) consistently predicted strength of the VPL connections [i.e., VPL thalamus and temporal region (F_2,48=7.3, p=.009); VPL and cerebellum (F_2,48=9.5, p=.003)]. In a multiple regression, age, sex, diagnosis, and SES explained 14.2% of the variance in the connectivity strength between VPL and temporal regions (F_4,49=1.9, p=.13). Although none of the variables significantly contributed, diagnosis (p=.14) had the lowest p-value. Removing diagnosis reduced the variance explained by the model (F_3,49=1.7, p=.18) to 10.0% and SES became significant (p=.03); removing SES reduced the variance explained by the remaining 3 variables to only 13.4% and diagnosis became significant (p=.01). Similarly, the 4 variables explained 18.3% of the variance in connectivity between VPL and cerebellum (F_4,49=2.5, p=.06); removing diagnosis reduced the variability explained by the model (F_3,49=2.1, p=.11) to 12.1% and SES became significant (p=.02); removing SES from model (F_3,49=3.2, p=.03) reduced the variance explained by the remaining 3 variables to only 17.5% and diagnosis became significant (p=.003).

Finally, relationships with behavioral measures were evaluated using AIC stepwise regressions. Connectivity between the entire thalamus and temporal regions was related to performance on the working memory composite (r=−.41, p=.06; ρ=−.43, p=.04, Figure 5a); between VPL and cerebellar regions to ataxia with eyes closed (r=.37, p=.10; ρ=.43, p=.05, Figure 5b).

Figure 5.

Among the AUD group only, relations between connectivity (i.e., residuals after removing age and sex) A) for the entire thalamus and middle temporal regions and the working memory composite score; B) the VPL and the cerebellum and ataxia with eyes closed scores.

4. Discussion

Imaging methods including DTI (Behrens et al., 2003; Duan et al., 2007; Jakab et al., 2012; Johansen-Berg et al., 2005; Kumar et al., 2015; Mang et al., 2012; O’Muircheartaigh et al., 2015; Wiegell et al., 2003; Ziyan et al., 2006) and fMRI (Hale et al., 2015; Kim et al., 2013; Kumar et al., 2017; Zhang et al., 2008; Zhang and Li, 2017; Zou et al.) have been used to parcellate the thalamus and assess thalamocortical connectivity patterns. Of the various thalamic nuclei, the anterior thalamus has been implicated by postmortem neuropathological studies as preferentially compromised in AUD (Belzunegui et al., 1995; Harding et al., 2000). Based on the extant literature in AUD including DTI reports (e.g., Harris et al., 2008; Pfefferbaum et al., 2009; Schulte et al., 2012; Trivedi et al., 2013), we expected altered connectivity between the anterior thalamus and Papez-related regions such as the hippocampus. Alternatively, because the MD is strongly and reliably connected with the prefrontal cortex (He et al., 2014; Metzger et al., 2010; Pergola et al., 2013; Walter et al., 2008), which is particularly vulnerable to alcoholism (Jung et al., 2012; Pfefferbaum et al., 2001; Pfefferbaum et al., 1998), we might also have expected that MD to prefrontal cortex connectivity strength would be altered in individuals with AUD relative to controls. However, our previous in vivo work comprising a larger sample (49 healthy controls; 41 individuals with AUD) and using a high resolution structural MRI based parcellation technique (Thomas et al., 2017) identified only the ventral lateral posterior (VLP) nucleus as showing volume deficits in AUD relative to healthy controls (Zahr et al., 2020) – a finding that was replicated in this smaller sample. Our findings that AUD is associated with volume deficits (VLP) and altered connectivity (VPL) in ventral thalamus are consistent with recent claims that alcoholics demonstrate cognitive and motor deficits related to corticocerebellar circuit dysfunction that involves the ventral thalamus (Ide and Li, 2011; Ide et al., 2018; Pitel et al., 2015). Thalamic nuclei AV and MD may have been preserved (both structurally and functionally) in this sample of alcoholics because they are thiamine replete; AV and MD pathology may be related to thiamine depletion (cf., Harding et al., 2000).

The current results demonstrate that an AUD diagnosis disrupts an anticorrelation between the entire thalamus or VPL with bilateral middle temporal lobe regions (changed to a positive correlation) and strengthened an anticorrelation between the VPL and the cerebellum. Statistics demonstrated that these relationships were modulated by SES. However, comparisons between models with and without SES indicated that diagnosis better explains functional connectivity variability than SES, and that most of the variability explained by SES overlaps with the variability explained by diagnosis. This comports with epidemiological data demonstrating a strong correlation between alcohol abuse and low SES (World Health Organization, 2018); conversely, high SES is protective against AUD (Calling et al., 2019).

AUD-altered connectivity between the entire thalamus and bilateral middle temporal regions was associated with worse performance on the working memory composite score. Worse connectivity between the VPL and cerebellar regions was related to worse performance on ataxia with eyes closed. The ventral posterior (VP) thalamus processes sensory information and supports alertness and arousal (Diamond et al., 1992; Krause et al., 2012; Nicolelis and Chapin, 1994). The VPL – part of the somatosensory thalamus – receives neuronal input from the medial lemniscus and spinothalamic tracts and projects to the somatosensory cortex (Behrens et al., 2003; Johansen-Berg et al., 2005; Mai and Forutan, 2012; Schmidt and Willis, 2007); functions include perception of touch, pain, temperature, itch, body position, taste, and arousal.

Previous functional connectivity studies often report correlations between VPL and somatosensory cortices (Fan et al., 2015; Hale et al., 2015; Ji et al., 2016; Kumar et al., 2017; Zhang et al., 2008; Zhang and Li, 2017) confirming rodent histology tracing studies (e.g., Kumar et al., 2017; but see Lenglet et al., 2012); a few studies also report VPL to motor cortex connections (e.g., Hale et al., 2015; Zhornitsky et al., 2018). In vivo results, however, can depend on the analysis methods used (e.g., Hale et al., 2015). Indeed, the current results instead comport with a rs-fMRI study using independent component analysis that reported for VPL positive correlations with temporal lobe regions including the hippocampus and negative connections to regions such as the cerebellum (Zhang and Li, 2017). Connectivity between the entire thalamus and temporal lobe region subserving working memory has a relatively strong basis in the literature including results from fMRI studies in temporal lobe epilepsy (Chen et al., 2015; Englot et al., 2017; González et al., 2019). There is only indirect support for a multisensory connection between VPL and temporal lobes (Campi et al., 2010; Hackett et al., 2007). With respect to the cerebellum, a task-based, finger tapping fMRI study showed activity in cerebellum, ventrolateral thalamus, and sensorimotor cortex (Lutz et al., 2000). Connections between the ventral thalamus and the cerebellum have also been reported in histology studies conducted in rats (Aumann et al., 1994) and macaques (Darian-Smith et al., 1990; Sakai et al., 1999, 2002), though questions remain with respect to the precise destination of cerebellar afferents to the thalamus (i.e., motor or sensory thalamus Aumann et al., 1996; Mackel and Miyashita, 1992).

Limitations of this study include a small sample size, which may have precluded identifying alterations to circuitry involving anterior thalamus. Another limitation is that our previous thalamic subregion study identified the VLP as sensitive to volume deficits in AUD (cf., Zahr et al., 2020), while the current functional study found altered connectivity in AUD unique to VPL. One possible explanation for these differences is that the previous study included a larger sample (n=49 healthy controls and n=41 individuals with AUD) because both 8-channel and 32-channel structural images were adequate for thalamic substructural volume determination (Zahr et al., 2020). Further, the previous structural study used 1mm isotropic resolution while the current work on functional images required 2.5mm smoothing, effectively reducing contrast and resolution (cf., Caparelli et al., 2019) which may have blurred the boundaries between ventral nuclei. This interpretation is supported by the fact that although these 2 datasets (i.e., structural and functional) were analyzed independently, they both revealed vulnerability of the ventral thalamus to AUD. Another explanation is that structural atrophy and connectivity changes do not necessarily manifest simultaneously (cf., Hafkemeijer et al., 2013; He et al., 2012; Li et al., 2017). Alternatively, motor structures such as the VLP may have more compensation to maintain connectivity than somatosensory structures such as the VPL. Another potential limitation is that the Louvain method is a greedy algorithm attempting to solve a NP-hard optimization program: as a result, the method can sometimes produce suboptimal communities (cf., Traag et al., 2019). The great inter-hemispheric symmetry displayed by the parcellations generated in this work suggests, however, that they were at least plausible and meaningful, if not mathematically optimal. In conclusion, the present results demonstrated that AUD is associated with altered connectivity between the entire thalamus and middle temporal regions that was predictive of working memory performance and between VPL and the cerebellum that was predictive of postural instability.

Highlights.

• High-resolution functional connectivity achieved using a 32-channel coil.

• Thalamic nuclei connectivity evaluated in alcoholics relative to controls.

• Altered connectivity in AUD unique to ventral posterior lateral thalamus.

• Compromised connectivity with cerebellum associated with worse ataxia performance.