Intracortical myelin in individuals with alcohol use disorder: An initial proof‐of‐concept study

Abstract Introduction Disruption of cortical gray matter and white matter tracts are well‐established markers of alcohol use disorder (AUD), but less is known about whether similar differences are present in intracortical myelin (ICM, i.e., highly myelinated gray matter in deeper cortical layers). The goal of this study was to provide initial proof‐of‐concept for using an optimized structural magnetic resonance imaging (MRI) sequence to detect differences in ICM in individuals with AUD compared to control participants reporting drinking within recommended guidelines. Methods This study used an optimized 3T MRI sequence for high intracortical contrast to examine ICM‐related MRI signal in 30 individuals with AUD and 33 healthy social drinkers. Surface‐based analytic techniques were used to quantify ICM‐related MRI signal in 20 bilateral a priori regions of interest based on prior cortical thickness studies, and exploratory vertex‐wise analyses were examined using Cohen's d effect size. Results The global distribution of ICM‐related signal was largely comparable between groups. Region of interest analysis indicated that AUD group exhibited greater ICM‐related MRI signal in precuneus, ventromedial prefrontal cortex, posterior cingulate, middle anterior cingulate, middle/posterior insula, and dorsolateral prefrontal cortex (Cohen's ds = 0.50–0.75). Four regions (right precuneus, ventromedial prefrontal cortex, posterior cingulate and left dorsolateral prefrontal cortex) remained significant (p < .05) after covarying for smoking status. Conclusion These findings provide initial evidence of ICM differences in a moderately sized sample of individuals with AUD compared to controls, although the inflation of type 1 error rate necessitates caution in drawing conclusions. Robustly establishing these differences in larger samples is necessary. The cross‐sectional design cannot address whether the observed differences predate AUD or are consequences of heavy alcohol consumption.


INTRODUCTION
Structural and functional disruptions of the cerebral cortex are wellestablished indicators of alcohol use disorder (AUD) (Fritz et al., 2022;Sullivan et al., 2010). People who meet criteria for AUD have been found to exhibit thinner cerebral cortices (Durazzo et al., 2011;Fortier et al., 2011), smaller gray matter volume (Bühler & Mann, 2011), as well as smaller volume in the frontal cortex and other cortical and subcortical regions (Durazzo et al., 2011;Yang et al., 2016) relative to people without AUD. These neuroanatomical differences were confirmed in a mega-analysis of previously published MRI data pooled from 23 laboratories including 3240 individuals (898 of whom were diagnosed with AUD) (Mackey et al., 2019). Of all substances examined (alcohol, nicotine, cocaine, methamphetamine, and cannabis), differences between AUD and control groups revealed the largest effect sizes for gray matter volume and cortical thickness in 27 cortical regions spanning frontal, temporal, parietal, and insular cortices. Other studies specifically focusing on cortical thickness in AUD have reported that decreased cortical thickness is related to a number of clinical indicators in AUD, including quantity/frequency of alcohol use and severity of alcohol-related problems (Fortier et al., 2011;Mashhoon et al., 2014), poor treatment outcomes (Durazzo et al., 2011), and impaired inhibitory control (Pennington et al., 2015).
There is considerable interest in alcohol-related myelin reductions as these may directly contribute to dysfunction in the central nervous system in people diagnosed with AUD. White matter is particularly susceptible to damage from alcohol and other drugs due to its low level of blood supply and the extensive energy requirements of oligodendrocyte function (Filley et al., 2017;Haroutunian et al., 2014).
White matter atrophy and effects of alcohol on myelin are considered hallmarks of AUD-related neuropathology (for a review, see Rice & Gu, 2019). Global deficits in white matter volume have been reported in numerous studies, as confirmed by a meta-analysis of 19 studies with over 1300 participants (Monnig et al., 2013). A coordinate-based meta-analysis of 18 studies reported macro-and microstructural WM alterations in patients with AUD that were assigned to the genu and body of the corpus callosum, anterior and posterior cingulum, fornix, and the posterior limb of the internal capsule (Spindler et al., 2022).
Evidence for disrupted myelin in AUD includes postmortem studies reporting demyelination in humans with a history of AUD and irregular folding and splitting of myelin membranes (Rice & Gu, 2019). Although primarily concentrated in the white matter, myelinated axons are also found in gray matter on local inhibitory interneurons and excitatory projection neurons in deep layers of the cortex (Nieuwenhuys, 2013;Rice & Gu, 2019). These deeper, myelinated layers of the cortex (layers IV-VI) are commonly referred to as intracortical myelin (ICM). The functional role of ICM is not completely understood, though it is involved in synchronizing and speeding of neuronal signals through the cortex in support of cognitive processing. Decreased ICM in posterior cingulate cortex is correlated with decreased error processing and cognitive control in healthy individuals (Grydeland et al., 2015), and reduced ICM in insula and superior temporal gyrus is associated with increased performance variability (Grydeland et al., 2013). Grydeland et al. (2013) also found that developmental trajectories of ICM showed correspondence with other imaging indices of myelin structure (e.g., mean diffusivity on intracortical diffusion tensor imaging scans).
The extent to which ICM is altered in people with AUD remains unknown, although animal models suggest that ICM disruptions may accompany excessive alcohol exposure. Preclinical studies in rats have reported increased vulnerability of myelin in medial prefrontal cortex to neurotoxic effects of ethanol (Vargas et al., 2014). These differences may normalize following prolonged alcohol abstinence (Navarro & Mandyam, 2015). Medial prefrontal cortex is critical for higher order learning and social behavior and continues to develop into adulthood. In their review of myelin regulation in AUD, Rice and Gu (2019) conclude that animal studies show that alcohol-mediated effects on myelination resemble some aspects of AUD in humans, and they emphasized that the extent to which myelinated gray matter is disrupted in AUD remains unknown and more extensive investigations are needed.
Advances in structural magnetic resonance imaging (MRI) have allowed for in vivo imaging of ICM content in human participants (Bock et al., 2013;Glasser & Van Essen, 2011). In the protocol developed by Bock et al. (2013), the T 1 -weighted MRI contrast is optimized for differentiating cortical regions with low and high myelin content. This T 1 image is commonly divided by a proton-density-weighted image to generate a ratio image that is strongly T 1 -weighted with high intracortical contrast. Various studies have cross-validated these MRI measurements of ICM with histological samples of nonhuman primate brain tissue (Bock et al., 2009), as well as postmortem human brain tissue (Fracasso et al., 2016). Together, these studies suggest that MRIgenerated ICM maps provide an accurate representation of underlying cortical myeloarchitecture.
The MRI protocol developed by Bock et al. (2013) has been used in studies of ICM in healthy participants (e.g., Rowley et al., 2015) and a limited number of studies of psychiatric disorders (Rowley et al., 2015;Sehmbi et al., 2018Sehmbi et al., , 2019. These studies provide precedent for studying ICM in clinical disorders, but no neuroimaging studies to date have characterized ICM in addictive disorders. The ability to image ICMrelated signal in vivo affords a unique opportunity to explore effects of AUD in cortical subunits. This research may ultimately clarify the extent to which reductions in cortical thickness seen across studies are attributable to disruptions in ICM. To this end, the current study examined ICM in adults with AUD compared to social drinkers without AUD using the optimized ICM imaging protocol (Bock et al., 2013). The focus of this exploratory study was to provide initial proof-of-concept of the technique in an AUD sample. Based on extensive evidence that people with AUD show disruptions in myelination and animal models suggesting that myelin in gray matter is also disrupted by alcohol exposure (Rice & Gu, 2019), we hypothesized that participants meeting criteria for AUD would show disruption in ICM relative to participants who do not have AUD. Although a reduction in ICM is the most plausible disruption, we were hesitant to make a definitive prediction given the novelty of the ICM imaging approach and since all participants in this study were at minimum social drinkers. The latter point is important to consider because any amount of excessive alcohol use over prolonged periods of time could impact ICM as shown in studies of adult heavy drinkers (Topiwala et al., 2017).

Procedures
Prospective participants completed an initial screening by telephone or web survey to determine eligibility. Eligible individuals were scheduled for an in-person screening which included a battery of self-report

Measures
Participants completed several clinical and individual differences measures during the first session. A subset of these measures was analyzed in the current study, as detailed below.

Structured Clinical Interview DSM-5
Diagnoses of current (last 12 months) AUD and SUD were obtained via the Structured Clinical Interview for DSM-5 (SCID-5) (American Psychiatric Association, 2015). All participants completed the AUD module, and participants who reported using drugs also completed the SUD module for their most frequent substance.

Alcohol-related problems
The Alcohol Use Disorders Identification Test (AUDIT) (Saunders et al., 1993) was used as a self-report measure of drinking behavior and alcohol-related problems. The AUDIT is a 10-item scale with total scores ranging from 0 to 40; scores of 8 or greater are indicative of hazardous alcohol drinking.

Alcohol and substance use frequency
The timeline follow-back interview was used to assess participants' daily alcohol consumption for the past 30 days (Sobell & Sobell, 1992 Heatherton et al., 1991), a validated self-report assessment of cigarette consumption (e.g., cigarettes/day) and nicotine dependence severity.

Demographics
A demographic questionnaire assessed age, sex assigned at birth, ethnicity, years of education, income, and handedness.

MRI scanning protocol
MRI images were acquired on a 3T General Electric Discovery system using a 32-channel receive-only radio frequency coil for the head and a transmit radio frequency body coil. All images were acquired with 1 mm isotropic resolution using a scan protocol developed by Bock et al.(2013). For the anatomical reference image for registration, we acquired a 3D T 1 -weighted anatomical image of the whole head

Image preprocessing and data analysis
Image processing was performed predominately in FSL (v5.0.9), To account for head motion between scans, the T 1 with high intracortical contrast and proton-density weighted images were registered to the reference T 1 -weighted anatomical image using a six-parameter linear affine registration using FLIRT. The T 1 -weighted anatomical image was then linearly registered to the MNI 152 T 1 1 mm brain atlas and the corresponding transform applied to the other two images; this resulted in a single subject's images being in register with each other and each subject's images being in the same orientation as the MNI 152 atlas. The proton density-weighted image was filtered using a 5 × 5 × 5 mm 3D median filter and a strongly T 1 -weighted image (ratio) created by dividing the T 1 -weighted high intracortical contrast image by the filtered proton-density-weighted image. This removed radiofrequency receive field (B 1 −) inhomogeneities, some radiofrequency transmit field (B 1 +) inhomogeneities, and T 2 *-weighting arising from the gradient echo readout in the inversion-recovery image. The ratio image still contained some B 1 + inhomogeneity from the magnetization preparation portion of the inversion-recovery pulse sequence.
The T 1 -weighted anatomical image was segmented in Freesurfer using recon-all, with manual corrections for gray matter/pial boundary errors to yield surfaces corresponding to the cerebral-spinal-fluid/pial boundary (Freesurfer's lh.pial/rh.pial surfaces) and the pial/white matter boundary (lh.white/rh.white surfaces).
The remaining processing steps generated a surface at a middle depth in the cortex and mapped the intensity from the ratio image onto this surface to represent myelination in the cortex. They closely followed the Human Connectome Project's minimal processing pipeline using custom scripts for myelin mapping adapted from F I G U R E 1 Anatomical locations of a priori regions of interest (Rgenerated from the multi-modal parcellation atlas. Twenty ROIs were examined in each hemisphere. Region numbers correspond to numeric labels in Table 2.
the HCP scripts (Glasser et al., 2013)  Group comparisons were performed using an a priori anatomical region of interest (ROI) approach and a whole-brain vertex-wise analysis. ROIs were chosen based on theoretical models (Crews & Boettiger, 2009;Koob & Volkow, 2010;Longo et al., 2016), previous cortical thickness findings in AUD (Durazzo et al., 2011;Fortier et al., 2011;Lawyer et al., 2010;Mashhoon et al., 2014;Momenan et al., 2012;Morris et al., 2019), and studies of neurocognitive correlates of ICM (Grydeland et al., 2013(Grydeland et al., , 2015. Specific ROIs were used from the multi-modal parcellation atlas (MMP) (Glasser et al., 2017) in HCP 32k_LR space. We consolidated neighboring ROIs into composite regions corresponding to our a priori ROIs. The 20 bilateral ROIs examined are provided in Table 2, and anatomical locations are visualized in Figure 1. Second, we conducted an exploratory vertex-wise analysis by calculating effect size (Cohen's d) for differences between AUD and control groups at each vertex. Since this is the first study to characterize ICM in AUD, we chose to be comprehensive and report statistical significance at p < .05. Anonymized imaging and self-report data from this study are available upon request from the corresponding author.

Sample characteristics
The AUD group did not differ significantly from the control group with regard to sex, age, race, or handedness (

ICM group maps
Surface visualization maps depicting ICM-related ratio signal for the AUD and the control groups are shown in Figure 2a

Region of interest analysis
Mean ICM-related ratio signal was extracted from the a priori ROIs, and group differences were examined. A complete table of ICM values by group and brain region (including group means, mean differences, and 95% confidence intervals) is provided in Table S1. The AUD group exhibited greater ICM-related ratio signal in 10 regions at p < .05 (uncorrected for multiple comparisons). See Figure 3 for ICM signal by group and Table 3  prefrontal cortex (BA8) in the right hemisphere. Importantly, none of these group differences remained significant after FDR correction.

Exploratory whole-brain analysis
Two exploratory analyses were conducted to examine differences in ICM across the entire cortex. First, we compared a total ICM index generated by summing ICM values across all vertices in the MMP atlas.
Group means and 95% confidence intervals are presented in Table S2.
There were no statistically significant differences between groups for the left hemisphere total ICM (F(1,61) = 1.95, p = .168, d = 0.41). For the right hemisphere, the difference between groups was marginally significant (F(1,61) = 3.73, p = .058, d = 0.46) potentially indicating greater total right hemisphere ICM for the AUD group compared to the control group. We also characterized group differences outside of the a priori ROIs by computing a whole-brain vertex-wise map of effect sizes (Cohen's d) for the differences between the AUD and control groups (see Figure 2c). Here, the largest effect sizes where AUD F I G U R E 2 Group average intracortical myelin (ICM) maps for alcohol use disorder (AUD) (top) and control (middle) groups. Bottom panel depicts vertex-wise effect size maps (Cohen's d) reflecting differences between AUD and controls. In each panel, data are projected onto the middle-depth cortical surface, shown in lateral, medial, and superior views for left and right hemispheres separately.
participants exhibited greater ICM signal compared to control participants were in anterior aspects of the inferior, middle, and superior temporal gyri, the bilateral ACC, and various medial and lateral frontal regions. With respect to regions showing less ICM signal in AUD compared to control, the largest effect sizes were found in left preand postcentral gyrus.

DISCUSSION
This study is the first to use an optimized MRI pulse sequence to characterize ICM in participants with AUD compared to social drinkers.
Our results indicated that the overall cortical distribution of ICM appears to be similar between people with AUD and social drinkers although trend-level group differences in right hemisphere total ICM may be indicative of greater ICM in AUD compared to control participants. When we conducted a finer-grained ROI analysis, we found evidence of localized differences between groups with smallto-medium effect sizes. Comparing the two groups, we found greater The direction of ICM differences was counter to our hypotheses.
Based on prior imaging and postmortem studies with human partic-F I G U R E 3 Extracted intracortical myelin (ICM)-related ratio signal values for each a priori region of interest (ROI) for left hemisphere (Panel A) and right hemisphere (Panel B). Alcohol use disorder (AUD) group shown in black; control group shown in gray. Bars reflect mean +1 standard error. *p < .05. Region abbreviations are provided in Table 2.
ipants and animal models of alcohol neurotoxicity, it is reasonable to expect that ICM would be reduced in AUD. Instead, we observed increased ICM in the AUD participants relative to controls in several regions. Unfortunately, the current study was not able to fully evaluate reasons for this result, and future research using these imaging techniques is necessary. Although speculative, one potential explanation for why the AUD participants had increased ICM signal relative to controls may be stem from myelin-related compensatory mechanisms or other inflammatory responses to neurotoxic effects of chronic, excessive alcohol consumption. Alcohol has neurotoxic and inflammatory effects in both humans  and rodents (Pascual et al., 2014). As well, neuroinflammation can lead to the initiation of myelin repair (Glezer et al., 2007), and oligodendrocyte precursor cells may differentiate into remyelinating oligodendrocytes when demyelination is present in the case of an injury or lesion (Setzu et al., 2006).
The biproducts of neuroinflammation-such as loss of tissue, demyelination, swelling-also can affect T 1 relaxation times during MRI image acquisition (Albrecht et al., 2016) which may contribute to greater ICM signal in the AUD group. Importantly, since much of this work is in animal models, caution is needed when translating findings to humans. It would be informative for future imaging studies to examine changes in ICM while also collecting systemic and central markers of neuroinflammation .
Comparing our results with previous ICM studies in healthy and clinical populations reveals some overlap in brain regions. We observed differences in regions of medial prefrontal cortex and anterior cingulate, which is consistent with preclinical and human studies reviewed in Rice and Gu (2019). The PCC region in the current study is in the same general location as the Grydeland et al.'s study of ICM in healthy participants that reported associations between ICM in PCC and error processing (Grydeland et al., 2015). The current study was not designed to test associations between ICM and neurocognitive performance; exploring whether increased ICM in PCC in people with AUD is related to neurocognitive performance is an important next step. Differences in DLPFC/Brodmann Area 46 have also been reported in previous studies with patients with bipolar disorder (e.g., Sehmbi et al., 2018), and DLPFC is consistently implicated in the neuropathology of AUD (e.g., Mackey et al., 2019).
Future studies of ICM in AUD should recruit a larger sample of participants with a wider range of alcohol involvement (e.g., across the continuum from social drinking through mild to severe AUD). As a proof-of-concept study, our intent was to compare a moderate/severe AUD group with social drinkers, but a dimensional study will help to further characterize ICM variation in relation to the full range of AUD  (Durazzo et al., 2011). Many of the AUD participants in the current study were actively engaged in treatment and were not currently drinking. Therefore, it is unclear whether the lack of significant group differences may have resulted from short-term recovery of ICM in abstinence. Finally, it will be informative to examine ICM in conjunction with other imaging methods for assessing white matter structural integrity, such as intracortical diffusion tensor imaging (e.g., Grydeland et al., 2013). Doing so may provide a more comprehensive account of myelin disruption in AUD.
The current results should be considered in the context of the study's limitations. As noted above, the cross-sectional design prevents testing of temporal or causal effects. The moderate sample size may have been underpowered to detect significant differences and precluded analyses of sex differences. The AUD and control groups also differed on several demographic and substance-related variables, including education, income, cigarette smoking, and cannabis use. Of these variables, only current smoking was associated with ICM, and sensitivity analyses revealed that several differences remained even after controlling for smoking status. Finally, many participants in the AUD group were treatment-seeking and were not currently drinking. Whether larger differences in ICM would emerge in participants who were not actively reducing their drinking is not known.
In conclusion, this is the first study to our knowledge to examine ICM-related MRI signal in people meeting criteria for AUD. The distribution of ICM across the cortex was largely similar between the AUD and control groups, although a more precise analysis revealed several areas of potentially greater ICM signal in the AUD participants even after controlling for cigarette smoking. These initial findings offer proof-of-concept for studying ICM in addiction samples and provide a foundation for future studies to unpack the clinical and neurocognitive significance of the differences observed, neurobiological mechanisms (e.g., compensatory or neuroinflammatory changes), and potential for changes in ICM following treatment.

CONFLICT OF INTEREST
James MacKillop is a principal in BEAM Diagnostics Inc., but no BEAM products were used in this study. No other authors have conflicts of interest to declare.

DATA AVAILABILITY STATEMENT
Anonymized imaging and self-report data from this study are available upon request from the corresponding author.

NOTE
i Our processing scripts differed from the standard HCP pipeline scripts in three ways. First, the resolution of our data was 1.0 mm isotropic, not 0.7 mm isotropic. Second, our scanning protocol did not include a T 2 -weighted image along with the T 1 -weighted image that is used to improve cortical surface segmentation. Instead, we performed these corrections manually. Third, we are mapping our T 1 -weighted ratio image instead of a T 1 / T 2 contrast.