OUP user menu

Cerebral white matter recovery in abstinent alcoholics—a multimodality magnetic resonance study

Stefan Gazdzinski, Timothy C. Durazzo, Anderson Mon, Ping-Hong Yeh, Dieter J. Meyerhoff
DOI: http://dx.doi.org/10.1093/brain/awp343 1043-1053 First published online: 4 February 2010


Most previous neuroimaging studies of alcohol-induced brain injury and recovery thereof during abstinence from alcohol used a single imaging modality. They have demonstrated widespread microstructural, macrostructural or metabolite abnormalities that were partially reversible with abstinence, with the cigarette smoking potentially modulating these processes. The goals of this study were to evaluate white matter injury and recovery thereof, simultaneously with diffusion tensor imaging, magnetic resonance imaging and spectroscopy in the same cohort; and to evaluate the relationships between outcome measures of similar regions. We scanned 16 non-smoking and 20 smoking alcohol-dependent individuals at 1 week of abstinence from alcohol and 22 non-smoking light drinkers using a 1.5 T magnetic resonance scanner. Ten non-smoking alcohol-dependent individuals and 11 smoking alcohol-dependent individuals were re-scanned at 1 month of abstinence. All regional diffusion tensor imaging, magnetic resonance imaging and spectroscopic outcome measures were calculated over comparable volumes of frontal, temporal, parietal and occipital white matter. At 1 week of abstinence and relative to non-smoking light drinkers, non-smoking alcohol-dependent individuals had higher mean diffusivity in frontal, temporal and parietal white matter (all P < 0.008), whereas smoking alcohol-dependent individuals had elevated mean diffusivity only in frontal white matter (P = 0.03). Smoking alcohol-dependent individuals demonstrated lower concentrations of N-acetyl-aspartate (a marker of neuronal viability) in frontal white matter (P = 0.03), whereas non-smoking alcohol-dependent individuals had lower N-acetyl-aspartate in parietal white matter (P = 0.05). These abnormalities were not accompanied by detectable white matter atrophy. However, the patterns of white matter recovery were different between non-smoking alcohol-dependent individuals and smoking alcohol-dependent individuals. In non-smoking alcohol-dependent individuals, the increase in fractional anisotropy of temporal white matter (P = 0.003) was accompanied by a pattern of decreases mean diffusivity in all regions over 1 month of abstinence; no corresponding changes were observed in smoking alcohol-dependent individuals. In contrast, a pattern of white matter volume increase in frontal and temporal lobes was apparent in smoking alcohol-dependent individuals but not in non-smoking alcohol-dependent individuals. These results were not accompanied by significant changes in metabolite concentrations. Finally, there were no consistent patterns of association between measures obtained with different imaging modalities, either cross-sectionally or longitudinally. These data demonstrate significant white matter improvements with abstinence from alcohol, reflected either as microstructural recovery or volumetric increases that depend on the smoking status of the participants. We believe our results to be important, as they demonstrate that use of a single imaging modality provides an incomplete picture of neurobiological processes associated with alcohol-induced brain injury and recovery thereof that may even lead to improper interpretation of results.

  • White matter
  • diffusion tensor
  • spectroscopy
  • cigarette smoking
  • alcohol dependence


The majority of in vivo human neuroimaging studies investigating the recovery of alcohol-induced neurobiological injury during abstinence from alcohol in alcohol-dependent individuals have used a single imaging modality. Cross-sectional studies reported morphological abnormalities in both white matter and grey matter (for review see Sullivan, 2000). Proton magnetic resonance spectroscopy (1H MRS) generally demonstrated lower concentrations of N-acetyl-aspartate (NAA, a marker of neuronal viability, which may reflect neuronal loss, lower neuronal density, atrophied dendrites and axons, and/or deranged neuronal metabolism) and choline-containing compounds (Cho; involved in membrane turnover/synthesis; see Ross and Bluml, 2001), most notably in the frontal lobes, medial temporal lobes and cerebellum (Bendszus et al., 2001; Parks et al., 2002; Gazdzinski et al., 2008a). Cross-sectional diffusion tensor imaging (DTI) studies in alcohol-dependent individuals have indicated decreased fractional anisotropy (FA) and increased mean diffusivity (MD) in the genu, body and splenium of the corpus callosum, as well as in the centrum semiovale, which suggest compromised axonal/myelin integrity (e.g. Pfefferbaum et al., 2005a). FA quantifies the directionality of water diffusion and is an indicator of white matter coherence and white matter organization within fibre tracts. FA reductions are attributed to degradation of myelin sheaths and axonal membranes (Pierpaoli et al., 2001), abnormalities of myelin with sparing of the axonal fibres (e.g. Song et al., 2005) or reduced density of axonal fibres (Takahashi et al., 2002). MD quantifies the average magnitude of microscopic water diffusion, which is likely to reflect cellular density and extracellular fluid volume (Sotak, 2004), and relates to the volume fraction of the interstitial space (Sotak, 2004; Song et al., 2005).

Longitudinal MRI studies of alcohol-dependent individuals during abstinence from alcohol report increases in white matter volume (Shear et al., 1994; Agartz et al., 2003), though one study did not find any significant change in white matter volume (Pfefferbaum et al., 1995). Longitudinal MRS studies of alcohol-dependent individuals demonstrated regional increases in concentrations of NAA and Cho in cerebral white matter and cerebellum (Bendszus et al., 2001; Parks et al., 2002; Ende et al., 2005; Durazzo et al., 2006; Bartsch et al., 2007). However, to date, there are no longitudinal DTI studies with abstinent alcohol-dependent individuals. Furthermore, it is not clear to what extent the results obtained by different imaging modalities are interrelated and yield complementary findings or whether various imaging modalities yield results that are unique and non-convergent with other modalities.

Up to now, only a few studies have reported interrelations between results obtained with different imaging modalities within the same cohort of alcohol-dependent individuals. Pfefferbaum and colleagues (2005a) reported that volume reduction in the corpus callosum of alcohol-dependent individuals was accompanied by FA and MD abnormalities, suggesting disruption in the structural constituents of local white matter. Bartsch et al. (2007) demonstrated that global white matter volume increases correlated with cerebellar and fronto-mesial Cho increases in non-smoking alcoholics over 6–7 weeks of abstinence from alcohol. Our own data suggest that white matter regions of lower FA spatially coincide with regions of NAA abnormalities in alcoholics (Wang et al., 2009).

Additionally, co-morbid chronic cigarette smoking, which occurs in ∼60–80% of alcohol-dependent individuals (Romberger and Grant, 2004), is associated with greater cross-sectional NAA and Cho abnormalities in multiple brain regions of alcohol-dependent individuals (Durazzo et al., 2004), as well as diminished regional brain metabolite and neurocognitive recovery with abstinence from alcohol (Durazzo et al., 2006, 2007b; Gazdzinski et al., 2008a).

The goals of this study were: (i) to measure white matter volumes, MD, FA and concentrations of major 1H MRS detectable metabolites from comparable white matter regions in the same cohort of alcohol-dependent individuals at 1 week of abstinence from alcohol; and (ii) to assess changes in these measures over 1 month of abstinence from alcohol. We tested the following hypotheses: (i) at baseline, soon after last alcohol consumption, smoking alcohol-dependent individuals (sADIs) demonstrate the highest MD and lowest FA, followed by non-smoking alcohol-dependent individuals (nsADIs) and non-smoking light drinkers (nsLD) in all major lobes; (ii) over 1 month of abstinence from alcohol, nsADIs have greater increases in regional white matter FA, volumes and concentrations of NAA and Cho than sADIs, as well as greater decreases of regional white matter MD; (iii) at baseline, higher MD is associated with lower FA, smaller white matter volumes and lower concentrations of NAA; and finally (iv) over 1 month of abstinence, decreasing lobar white matter MD is associated with increases of FA, white matter volumes and NAA concentrations, and increasing Cho is associated with increasing white matter volumes.

Material and methods


Sixteen nsADIs (51.5 ± 10.3 years, two female) and 20 sADIs (47.6 ± 9.5 years, one female) were part of a longitudinal study that assessed the co-morbid effects of alcohol dependence and chronic cigarette smoking on neurobiological and neurocognitive abnormalities in abstinent alcoholics (Durazzo et al., 2007a). The number of participants available for this analysis was limited by the availability of DTI data. nsADIs and sADIs were primarily recruited from the San Francisco Veterans Affairs Medical Centre Substance Abuse Day Hospital and secondarily from the San Francisco Kaiser Permanente Chemical Dependence Recovery Program. They were first studied at baseline 4.6 ± 2.6 and 4.5 ± 2.1 days after their last alcoholic drink, respectively. Ten nsADIs and 11 sADIs were re-scanned after 37.5 ± 9.6 and 31.2 ± 6.9 days of abstinence, respectively (P = 0.10). Twenty-two healthy, age-matched nsLDs (48.3 ± 8.4 years, two females) were recruited from the San Francisco Bay Area community and nine were re-scanned after ∼1 year. Only 8 of the 36 alcohol-dependent individuals from this study contributed to our previous reports.

The inclusion and exclusion criteria are fully described in Durazzo et al. (2004). In short, all alcohol-dependent individuals met the Diagnostic and Statistical Manual of Mental Disorders (fourth edition) criteria for alcohol dependence with physiological dependence and consumed more than 150 standard alcoholic drinks per month (80 for females) for at least 8 years prior to enrolment into the study. A standard drink contains 13.6 g of pure ethanol, equivalent of 12 oz beer, 5 oz wine or 1.5 oz liquor. All participants were free of general medical, neurological and psychiatric conditions, except unipolar mood disorders, hypertension (medication controlled) and hepatitis C in alcohol-dependent individuals. These co-morbidities were not exclusionary in alcohol-dependent individuals, due to their high prevalence among alcohol-dependent individuals (Hasin et al., 2007) and chronic cigarette smokers (Fergusson et al., 2003). Six sADIs and two nsADIs met the Diagnostic and Statistical Manual of Mental Disorders (fourth edition) criteria for substance-induced (alcohol) mood disorder with depressive features. Three sADIs and two nsADIs were diagnosed with recurrent major depression, whereas two nsADIs were diagnosed with major depression, which were in partial or sustained full remission. One sADI had alcohol-induced psychotic disorder with hallucinations. Two nsADIs met criteria for past cannabis abuse in sustained full remission and their last use was 1 and 2 years before enrolment, whereas one sADI met criteria for past cocaine dependence and was in sustained full remission with last use 8 years before enrolment. One sADI and one nsADI with a unipolar mood disorder were taking fluoxetine. Six nsADIs and nine sADIs had medically controlled hypertension.

Alcohol consumption and smoking behaviour over lifetime were assessed via the Lifetime Drinking History (Skinner and Sheu, 1982; Sobell et al., 1988; Sobell and Sobell, 1992) and the Fagerstrom Tolerance Test for Nicotine Dependence (Fagerstrom et al., 1991), respectively. All sADIs continued to smoke at their baseline levels over the assessment interval, except for one individual, who stopped smoking and used a nicotine patch; all his magnetic resonance measures improved after baseline, but were within the range of the sADIs. Eight nsADIs never smoked during their lifetime and eight quit smoking between 2 and 30 years prior to enrolment. The DTI indices, volumes and metabolite concentrations were not statistically different between those who never smoked and former smokers. Of the 21 alcohol-dependent individuals, 19 who had a follow-up scan participated in continued out-patient substance abuse treatment programs at the San Francisco Veteran Affairs Medical Centre for the study duration and were randomly tested for alcohol consumption and given weekly drug screens to ensure abstinence.

Clinical laboratory measures, obtained within 1 day of the magnetic resonance studies, assessed for hepatocellular injury, red blood cell status and nutritional status (serum pre-albumin; Weinrebe et al., 2002). Participants were allowed to smoke ad libitum prior to magnetic resonance scans. Alcohol withdrawal (Sullivan et al., 1989), depressive (Beck, 1978) and anxiety symptomatology (Spielberger et al., 1977) were assessed within 1 day of the scanning session. No participant had clinically significant withdrawal symptoms. The Institutional Review Boards of the University of California San Francisco and the San Francisco Veteran Affairs Medical Centre approved all procedures, and written informed consent was obtained from all participants prior to study.

Data acquisition

All magnetic resonance data were obtained on a standard 1.5 T MRI system (Siemens Vision, Iselin, NJ, USA). DTI was performed with a single-shot double-refocused spin-echo echo-planar imaging sequence (repetition time/echo time = 5000/100 ms, 2.4 × 2.4 × 5 mm3, 20 contiguous slices covering supratentorial white matter, 3 min scan time), with diffusion sensitizing gradients of b = 0, 160, 360, 640 and 1000 s/mm2 applied along six independent directions and double refocusing diffusion gradients to remove eddy current-related geometrical image distortions in DTI (Reese et al., 2003). 3D T1-weighted images were acquired with a standard magnetization prepared rapid gradient echo sequence (repetition time/echo time/inversion time = 10/7/300 ms, 15° flip angle, 1 × 1 × 1.5 mm3, 7 min), for segmentation. Additionally, axial-oblique double spin-echo (repetition time/echo time1/echo time2 = 2500/20/80 ms, 1 × 1 × 3 mm3, 12 min) proton density and T2-weighted images were acquired and used to assess white matter signal hyperintensity qualitatively. Finally, metabolite spectra were acquired with proton multislice short-echo time 1H MRS imaging (repetition time/echo time/inversion time = 1800/25/165 ms, 30 min) in three parallel, oblique-axial slices, each 15 mm thick and 6 mm apart and covering the major cerebral lobes (Fig. 1). All data were carefully reviewed; motion artefacts and overall poor quality were exclusionary.

Figure 1

Brain regions scanned with our magnetic resonance spectroscopic sequence. Mean absolute atrophy-corrected metabolite concentrations were calculated over anatomically defined white matter lobar regions. The DTI and MRI sequences covered basically all lobar white matter.

Data processing

As the lack of full brain coverage of our MRS imaging sequence precluded voxel-wise analyses, we compared measures obtained with these modalities within comparable white matter regions.

Probability maps of grey matter, white matter and CSF in frontal, parietal, temporal and occipital white matter were obtained by combining Expectation–Maximization Segmentation (Van Leemput et al., 1999) with an atlas-based deformable registration method that was used to identify regions of interest in the brain automatically, as previously described (Cardenas et al., 2005). These maps were also used to calculate white matter volumes. To account for individual variation in head size, absolute volumes of identified structures were divided by intracranial volume, defined as the sum of white matter, grey matter and CSF.

For the DTI analyses, MD and FA were calculated in every voxel using a simple least squares fit of the tensor model using all five b-values. We aligned and interpolated the 3D T1-weighted images and corresponding lobar white matter probability maps to corresponding (b0) diffusion scans. Regions affected by susceptibility artefact (mostly orbito-frontal white matter) were manually removed from FA and MD images. The spatial extents of analysed DTI voxels included frontal, parietal, temporal and occipital lobes, and were comparable across participants and virtually the same in the longitudinal scans.

To minimize the effects of partial volumes, small mis-registrations and white matter hyperintensities (that generally segment as grey matter or CSF) on the analyses, we used only voxels containing >95% of white matter and with FA > 0.2. Thus, this study effectively compares normal appearing white matter between groups. Then, we calculated median MD and median FA within similar white matter regions as used for structural and spectroscopic analyses (similar to Lim et al., 1999; Fig. 1). Arithmetic means were not calculated, as the distributions of MD and FA were not Gaussian. The lobar MD and FA were calculated on ∼4700 voxels per individual for frontal white matter, 1700 voxels for temporal white matter, 2100 voxels for parietal white matter and 700 voxels for occipital white matter, with no significant differences in voxel counts between groups. The voxel counts were also similar in the smaller sample used for longitudinal analyses and did not differ significantly between baseline and follow-up scans. This approach was valid, as our preliminary voxel-wise DTI analyses determined that microstructural abnormalities in alcohol-dependent individuals, who were part of our cohort, are relatively widespread throughout white matter in all lobes (Yeh et al., 2008).

Processing details for spectroscopic data were described in Meyerhoff et al. (2004). The final MRS imaging outcome measures were tissue-specific, atrophy corrected, absolute, mean metabolite concentrations in institutional units over similar regions as used in DTI and structural analyses.

Study design and statistical analyses

The cross-sectional analyses evaluated for differences between groups in median MD, median FA, with Generalized Linear Model (Wald χ2), separate for each white matter region, due to heterogeneous variances in these measures. Lobar white matter volumes and individual metabolite concentrations over frontal, temporal, parietal and occipital white matter were compared between groups with one-way multivariate analyses of co-variance (MANCOVA; Wilks’ lambda), followed by univariate analyses of co-variance and pairwise one-tailed t-tests. Although age was not significantly different between groups, the cohort spanned a large age range (28–66 years) and therefore age was used as a covariate in all analyses to account for the potential effects of ageing (e.g. Bartzokis et al., 2001; Schuff et al., 2002; Pfefferbaum et al., 2005b). For NAA and Cho comparisons, body mass index (BMI, calculated as body mass in kilograms divided by height in metres squared) was used as a second covariate, because we observed associations between BMI and metabolite concentrations in a cohort of healthy middle-aged individuals (Gazdzinski et al., 2008b). The BMI did not correlate with MD, FA or white matter volumes (P > 0.20, corrected for age) in nsLDs and was not used as a covariate in volumetric and diffusion analyses. To correct for experiment-wise error rate, we created the following families of outcome measures: (i) lobar white matter diffusion measures—MD, FA; (ii) lobar white matter volumes; and (iii) lobar metabolite concentrations (NAA, Cho). When MANCOVA was significant, the P-values of the t-tests were multiplied by the number of magnetic resonance parameters within each family. Otherwise, if the MANCOVA was not significant, the significance levels of t-tests were adjusted by multiplying the P-values by 4 (number of evaluated white matter regions) and by the number of magnetic resonance parameters within each family. Finally, to account for differences in drinking history between nsADIs and sADIs, these groups were compared with additional covariates (age of onset of heavy drinking, total lifetime consumption of ethanol and average number of drinks per month over lifetime) in follow-up analyses. Cohen’s d was used to estimate effect sizes (ES).

Longitudinal analyses comparing longitudinal changes between nsADIs and sADIs utilized doubly multivariate analysis of repeated dependent measures multivariate analysis of variance (dMANOVA; Tabachnick and Fidell, 2001), separately for each imaging parameter. This procedure allows for the simultaneous evaluation of change in multiple dependent measures obtained on different occasions (i.e. changes in FA and MD), while protecting against the increased family-wise error rate that occurs when separately evaluating multiple dependent measures (the conventional approach). Additionally, to account for individual differences in scan intervals, we calculated monthly rates of change as: Embedded Image and compared change rates between nsADIs and sADIs with directional independent t-tests. Pearson’s correlations evaluated relationships between outcome measures and their longitudinal changes with estimates of drinking and smoking severity. Linear regression analyses were used to ensure that cigarette smoking was not a significant factor modulating these correlations. The associations among FA, MD, volumes and metabolite concentrations were considered separate scientific questions and no corrections were made for the number of magnetic resonance outcome measures. To correct the significance levels for multiplicity, we used the same families as described above. Additionally, the significance levels for correlation analyses within the same white matter region were corrected for experiment-wise error rate by multiplying the P-values by 4 (number of evaluated regions), whereas the P-values were multiplied by 16 when the evaluated measures were obtained from different regions. For example, the significance level of correlation between frontal white matter volume and frontal white matter FA was multiplied by 8 (four regions, two measures in the family of diffusion measures), whereas the significance level for correlation between frontal white matter NAA and parietal white matter FA was multiplied by 64 (16 × 2 × 2). Corrected P < 0.05 was considered statistically significant. All statistical tests were conducted with the Statistical Package for the Social Sciences-16.0 for Windows (SPSS; Chicago, IL, USA).


Participant characterization

Table 1 lists alcohol consumption measures and other demographic and clinical variables. All groups were equivalent on age [F(2,55) = 0.83, P = 0.44] but not on education [F(2,55) = 12.1, P = 0.001], with nsLDs having more education than both nsADIs and sADIs (both P < 0.003). nsADIs and sADIs consumed a similar average number of alcoholic drinks per month over 1 and 3 years prior to enrolment (P > 0.48). However, nsADIs began drinking at heavy levels (i.e. >100 drinks/month) at older age (30.4 ± 11.7 versus 19.5 ± 3.5, P = 0.001) and had fewer drinks per month over lifetime than sADIs (162 ± 86 versus 271 ± 168, P = 0.02). nsADIs did not differ from sADIs on self-reported measures of depressive, anxiety and withdrawal symptomatology, haemoglobin, haematocrit and red blood cell counts (all P > 0.34). In both alcohol-dependent individual groups, the liver enzymes γ-glutamyltransferase and aspartate-aminotransferase levels were elevated at baseline but normalized before follow-up, whereas their red blood cell counts were below normal at both assessments. Pre-albumin was within normal range for both nsADIs and sADIs, suggesting no gross malnourishment.

View this table:
Table 1

Demographics, alcohol consumption and clinical variables (mean ± SD)

Cross-sectional sampleSmaller longitudinal sample
ParameternsLD n = 22 (2F)nsADI n = 16 (2F)sADI n = 20 (1F)nsADI n = 10 (1F)sADI n = 11 (0F)
Age (years)48.3 ± 8.451.5 ± 10.347.7 ± 9.550.3 ± 10.350.9 ± 10.9
Education (years)17.1 ± 2.714.3 ± 2.213.8 ± 2.014.8 ± 2.413.4 ± 1.4
AMNART117 ± 6114 ± 9119 ± 6115 ± 8
BDI2.8 ± 3.411.1 ± 8.615.5 ± 8.97.1 ± 7.311.6 ± 5.3
1-year average (drinks/month)19 ± 17359 ± 172372 ± 146375 ± 211385 ± 153
Lifetime average (drinks/month)17 ± 10162 ± 86*277 ± 168*166 ± 97324 ± 206
Total lifetime consumption of ethanol (kg)891 ± 5161459 ± 1032866 ± 531**1825 ± 1242**
Age at onset of heavy drinking (years)30.4 ± 11.7**19.5 ± 3.4**30.2 ± 10.3**20.5 ± 2.6**
Major depressive disorder (n)04321
Substance induced mood disorder (n)02604
Hypertension (n)06945
Hepatitis-C [n]00202
GGT (institutional units)26 ± 17151 ± 15998 ± 53100 ± 9772 ± 17
AST (institutional units)27 ± 745 ± 2451 ± 3941 ± 2071 ± 54
RBC (M/mm3)5.02 ± 0.334.38 ± 0.444.49 ± 0.554.48 ± 0.384.55 ± 0.71
BMI (kg/m2)25.8 ± 5.529.8 ± 4.7*25.7 ± 6.3*30.5 ± 5.2**23.7 ± 4.1**
  • Values are mean ± SD.

  • AMNART = American National Adult Reading Test; BDI = Beck Depression Inventory; 1-year average = number of drinks per month over 1-year prior to study; lifetime average = number of drinks per month over lifetime; age at onset of heavy drinking = age when alcohol consumption exceeded 100 drinks per month; GGT = gamma-glutamyltransferase (local normal range = 7–64 institutional units); AST = aspartate aminotransferase (local normal range = 5–35 institutional units); RBC = red blood cell count (local normal range = 4.7–6.1 M/mm3).

  • *P < 0.05; **P < 0.01.

The sADI Fagerstrom score was 5.2 ± 2.2, indicating medium to high levels of nicotine dependence. The sADI participants smoked on an average 20.6 ± 11.6 cigarettes per day for 20.8 ± 12.4 years, resulting in 24 ± 22 pack-years. Most nsADIs were overweight or obese (BMI > 25), whereas sADIs were generally at normal weight or overweight (20 < BMI < 30; Table 1). These group characteristics were similar in the smaller cohort used for longitudinal analyses.

According to a clinical neuroradiologist’s review of study MRI, five nsADIs (31%) and nine sADIs (45%) demonstrated white matter signal hyperintensity (χ2 = 0.71, P = 0.40). Specifically, three nsADIs and two sADIs had punctate foci, two nsADIs and four sADIs had early confluence of white matter signal hyperintensities, and three sADIs (but no nsADIs) had large confluent areas of white matter signal hyperintensities. The presence of these white matter signal hyperintensities did not affect our DTI results as they segmented as grey matter or CSF, and thus, did not contribute to our white matter MD and FA measures. The smaller sample used in longitudinal analyses had demographics, clinical indices and white matter signal hyperintensity distribution similar to the larger cross-sectional cohort at baseline.

Cross-sectional group comparisons at baseline

This analysis tested the first hypothesis that MD is highest and FA is lowest in sADIs, followed by nsADIs and nsLDs.

Comparison of frontal white matter MD using Generalized Linear Model indicated significant group differences [χ2(2) = 10.45, P = 0.03]. Follow-up group comparisons revealed that nsADIs had 3.9% higher (P = 0.005, ES = 0.96) and sADIs 2% higher (P = 0.05, ES = 0.62) frontal white matter MD than nsLDs, with a trend for 1.9% higher MD in nsADIs than in sADIs (P = 0.10, ES = 0.44). Among the hypothesized group differences, nsADIs demonstrated higher MD than nsLDs in the temporal (3.1%, P = 0.005, ES = 0.88) and parietal lobes (3.3%, P = 0.003, ES = 0.96), but sADIs were not significantly different from nsLDs (ES < 0.4). No group differences were apparent for any of the lobar FA measures (P > 0.20); the effect sizes were consistently larger for comparisons between nsADIs and nsLDs than for comparisons between sADIs and nsLDs, consistent with MD results.

Lobar white matter volumes were insignificantly different between the three groups (<3%, P > 0.09, ES < 0.3). Among metabolites and after using age and BMI as covariates, the MANCOVA comparing NAA between groups was marginally significant [F(8,80) = 1.81, P = 0.09]. However, among the hypothesized contrasts, NAA was 9% lower in frontal white matter of sADIs than in nsLDs (P = 0.03, ES = 0.85) and 10% lower in parietal white matter of nsADIs as compared with nsLDs (P = 0.05, ES = 0.87), with a trend for 8% lower NAA in parietal white matter of sADIs relative to nsLDs (P = 0.08, ES = 0.71). Also, lobar white matter Cho, creatine and m-inositol concentrations did not differ significantly between groups (all P > 0.44, ES < 0.35).

Finally, all the differences between nsADIs and sADIs reported above were not explained by group differences in alcohol consumption. Furthermore, the results reported in this paragraph were not appreciably affected by excluding participants with co-morbid depressive disorders, cardiovascular disease or history of drug abuse/dependence. Same patterns of group differences were observed within the subsample re-scanned at follow-up.

Changes over 1 month of abstinence from alcohol

This analysis tested the second hypothesis that reductions of MD and increases in FA, volumes and increases in concentrations of NAA and Cho with abstinence from alcohol will be more pronounced in nsADIs than in sADIs. The results are depicted in Fig. 3.

Figure 2

MD (A) and FA (B) in frontal white matter of nsLDs, nsADIs and sADIs. The statistical significance of MD was not driven by the two nsADIs with highest MD values. Same baseline patterns were observed among participants who were rescanned at follow-up.

Figure 3

Patterns of longitudinal frontal white matter changes in (A) MD, (B) FA, (C) volume and (D) NAA concentration observed in participants who were scanned at baseline and follow-up: the P-values pertain to longitudinal changes. Please, note the significant decrease in MD of nsADIs (A) is not accompanied by any volume change (C), whereas the volume increase in sADIs (C) is not accompanied by a MD change (A). The range of corresponding measures in 22 nsLDs is depicted in grey (mean ± standard error). NAA concentrations reported after correction for age and BMI.

Fractional anisotropy

The omnibus dMANOVA comparing lobar FA in nsADIs and sADIs between baseline and follow-up demonstrated a significant effect for time [F(4,16) = 4.04, P = 0.019] and a trend for an interaction between smoking status and time [F(4,16) = 2.43, P = 0.09]. The follow-up ANOVAs demonstrated a significant effect for time [F(1,19) = 12.9, P = 0.002] and an interaction between smoking status and time [F(1,19) = 10.57, P = 0.004] in temporal white matter. These findings appeared to be driven by significant increases in temporal white matter FA of nsADIs (3.5 ± 2.6%, P = 0.003). nsADIs also showed a trend for increasing FA in the frontal white matter (1.5 ± 3.0%; P = 0.06), whereas sADIs demonstrated no FA changes over the same interval (P > 0.16, per cent changes <0.57%). Covariation for group differences in total lifetime consumption of ethanol did not appreciably alter these findings.

Mean diffusivity

dMANOVA yielded no main effects or interactions for smoking status or time. Although the follow-up repeated measures ANOVAs pointed to some effects for time and/or time by smoking status interactions, they were not significant after correction for multiple comparisons. Importantly and consistent with FA changes, only nsADIs demonstrated a pattern of MD decrease in white matter of frontal (−1.5 ± 2.3%, P = 0.04, uncorrected), temporal (−1.8 ± 2.0%, P = 0.01, uncorrected), parietal (−1.8 ± 2.3%, P = 0.02, uncorrected) and occipital lobes (−2.6 ± 4.0%, P = 0.03, uncorrected). No corresponding changes over the same interval were observed among sADIs (P > 0.16, uncorrected, per cent changes <0.34%). Covariation for group differences in total lifetime consumption of ethanol did not appreciably alter these patterns.


The omnibus dMANOVA yielded a significant interaction between smoking status and time [F(4,15) = 3.05, P = 0.05]. This result reflected pattern of volume increases in frontal (1.2 ± 1.5%, P = 0.01, uncorrected) and temporal white matter (1.5 ± 2.1%, P = 0.03, uncorrected) of sADIs, with no patterns of volume changes in nsADIs (P > 0.27, uncorrected, percent change <0.5%). Finally, we have obtained similar patterns using monthly rates of changes. This assures that the observed differences in recovery between nsADIs and sADIs are not a consequence of the numerically larger inter-scan interval in nsADIs (37.5 days) compared with sADIs (31.2 days).


dMANOVAs for NAA, Cho and m-inositol concentrations did not yield any significant effects of time (P > 0.27) or time by smoking status interactions (P > 0.55).

Finally, additional comparisons indicated that none of the above results were appreciably influenced by participants with co-morbid conditions.

Cross-sectional group differences at 5-week follow-up

In the smaller longitudinal sample of 21 sADIs and nsADIs, the group differences between nsADIs and sADIs at 5-week follow-up and nsLDs re-scanned after 1 year did not reach statistical significance for lobar white matter MD and FA, lobar white matter volumes or mean lobar metabolite concentrations [all F(8,46) < 1.57, all P > 0.23], possibly due to normalization of magnetic resonance measures and/or low statistical power.

Relationships between outcome measures at baseline

We tested the hypotheses that in alcohol-dependent individuals (nsADIs and sADIs combined) at 1 week of abstinence (baseline) higher lobar MD is associated with lower FA, smaller white matter volumes and lower concentrations of NAA within the same white matter regions.

At baseline, larger white matter volumes were moderately related to lower MD and higher FA, but these associations did not survive corrections for multiple comparisons. Greater age in alcohol-dependent individuals was associated with higher MD in all regions (0.40 < r < 0.57, P < 0.03) and lower FA in frontal and temporal white matter (r < −0.46, P < 0.01). After correction for age, higher FA was associated with lower MD in total white matter and all lobar regions (r < −0.70, P < 0.004); smoking status did not affect these relationships. These patterns remained significant when participants with co-morbid mood disorders, hypertension or white matter lesions were excluded from analyses. In addition, when participants with co-morbid mood disorders only were excluded, we observed a positive statistical association between frontal white matter volume and frontal NAA concentration (r = 0.66, P = 0.008). Finally, a higher amount of alcohol consumed over lifetime (in kilograms) was associated with higher MD (r = 0.29, P = 0.04) and lower FA (r = −0.34, P = 0.02) in total white matter, and surprisingly, later onset of heavy drinking was associated with higher total white matter MD (r = 0.35, P = 0.02). Otherwise, the FA and MD measures in alcohol-dependent individuals were not related to the average numbers of drinks per month over 1, 3 and 8 years, prior to enrolment or with any measures of smoking severity in sADIs.

Relationships between outcome measures changes over 1 month of abstinence from alcohol

We hypothesized that over 1 month of abstinence from alcohol, decreasing lobar white matter MD is associated with corresponding increases in FA, white matter volumes and NAA concentrations, whereas increasing Cho is associated with increasing white matter volumes in the same regions.

Over the month of abstinence from alcohol and within alcohol-dependent individuals, the MD decreases were associated with FA increase in total white matter (r = −0.44, P = 0.02) and among lobes in frontal (r = −0.56, P = 0.016), temporal (r = −0.49, P = 0.048) and occipital white matter (r = −0.53, P = 0.024), supporting our hypothesis. There were no patterns of associations between longitudinal changes of markers obtained with different magnetic resonance modalities within the same regions and no such correlation survived corrections for multiple comparisons. Additionally, when participants with white matter lesions were excluded from the statistical analyses, the volume increase in frontal white matter correlated with Cho increase (r = 0.65, P = 0.044). Later onset of heavy drinking was associated with faster FA recovery in total white matter (r = 0.52, P = 0.008) and temporal white matter (r = 0.53, P = 0.024). Finally, none of the correlations was appreciably affected by exclusion of participants with depressive disorders, cardiovascular disease or history of drug abuse/dependence. The associations of volumetric and spectroscopic changes with cognitive measures and their changes with abstinence will be reported elsewhere in a larger cohort.


This is the first report to assess white matter injury in smoking and non-smoking alcoholics during abstinence using DTI, structural MRI and 1H MRS imaging together in the same cohort and within comparable white matter regions.

At 1 week of abstinence from alcohol, nsADIs had higher MD in frontal, temporal and parietal white matter than nsLDs, whereas sADIs had higher MD than nsLDs only in frontal white matter. Longitudinally, over 1 month of abstinence from alcohol, there was an overall pattern of FA increases and FA by smoking status interactions, reflecting faster microstructural recovery in nsADIs than in sADIs. These FA changes were accompanied by a tendency for MD decrease exclusively in nsADIs. Conversely, the interactions between time and smoking status for volumetric changes reflected white matter volume increases in frontal and temporal white matter of sADIs, but not of nsADIs. Lobar metabolite concentrations did not change significantly over time in either group. Only FA and MD measures in the same lobes and their changes were interrelated. At baseline, higher white matter MD was associated with lower white matter FA. Moreover, later onset of heavy drinking was associated with faster FA recovery in total white matter and in temporal white matter, suggesting that cumulative effects of heavy drinking impairs the ability of white matter to recover from injury. Taken together, the outcome measures obtained with different imaging modalities were not significantly related to each other and their changes during abstinence from alcohol demonstrated different patterns of brain injury and recovery thereof in nsADIs and sADIs. This suggests that each modality, being differentially sensitive to various brain changes in alcohol-dependent individuals, provides unique and complementary information on white matter status cross-sectionally and during abstinence.

Cross-sectional findings

Our DTI results in lobar white matter are consistent with previous reports of elevated MD in the corpus callosum of alcohol-dependent individuals (Pfefferbaum and Sullivan, 2002; Pfefferbaum et al., 2005a). The corresponding group differences of FA were not significant, possibly because FA measures within the same voxels are more affected by crossing fibres (Hirsch et al., 1999; Alexander et al., 2001) and more susceptible to errors arising from inclusion of non-white matter voxels than MD (Pfefferbaum and Sullivan, 2003). Higher lobar MD was significantly associated with lower lobar FA, but contrary to Pfefferbaum et al. (2005a) who reported associations between callosal FA and callosal volume, our DTI indices did not relate to the corresponding white matter volumes. However, our results are consistent with Fjell and colleagues’ (2008), who did not find any consistent pattern in the relationships between white matter volumes and FA measures among healthy individuals. Taken together, all the findings suggest that correlations between measures of microstructural and macrostructural integrity are region-specific and perhaps restricted to regions of high-density fibre bundles. The lack of significant volumetric abnormalities in our cohort in the presence of diffusion abnormalities suggest that DTI is more sensitive to brain injury than conventional structural MRI, as previously noted (Pfefferbaum and Sullivan, 2002).

Surprisingly, the nsADIs group demonstrated more widespread MD abnormalities than the sADIs group. As sADIs consumed more alcohol over lifetime than nsADIs and started drinking at heavy levels a full decade earlier than nsADIs, it seems unlikely that less microstructural injury (lower MD) in sADIs than in nsADIs is due to the differences in drinking history. Although unexpected, this result is qualitatively consistent with Paul and colleagues (2008), who found higher FA in corpus callosum of non-clinical (otherwise healthy) cigarette smokers compared with their non-smoking counterparts. They attributed this effect to stimulation of nicotinic receptors in oligodendrocytes by nicotine that could result in better microstructural white matter integrity in cigarette smokers. Nevertheless, less widespread MD abnormalities in sADIs versus nsADIs may reflect neurodegenerative processes.

Our sADIs had a qualitatively higher degree of white matter signal hyperintensity than nsADIs, consistent with population studies (e.g. Jeerakathil et al., 2004), which suggests that smoking adds an additional burden to white matter integrity. We and others have demonstrated brain perfusion abnormalities in smoking alcoholics and controls (Rourke et al., 1997; Gazdzinski et al., 2006). Chronic hypoperfusion results in axonal damage and demyelination accompanied by formation of redundant myelin and mild astrogliosis (e.g. Farkas et al., 2004). These processes could lead to decreased water content within white matter and the observed lower MD in sADIs versus nsADIs. Alternatively, chronic cigarette smoking has been equated to a type of repeated acute (mild) carbon monoxide poisoning (Alonso et al., 2004); carbon monoxide poisoning is associated with white matter cytotoxic oedema and restricted diffusivity in white matter lesions (see Terajima et al., 2008 and references therein), consistent with decreased MD.

nsADIs and sADIs demonstrated lower NAA in parietal and frontal white matter, respectively, consistent with most earlier MRS reports in abstinent alcoholics (Bendszus et al., 2001; Schweinsburg et al., 2001; Ende et al., 2005; Bartsch et al., 2007), but not in Parks et al. (2002). We did not observe any significant Cho abnormalities in white matter, consistent with Parks et al. (2002), but not Bendszus et al. (2001), Schweinsburg et al. (2001), Ende et al. (2005) and Bartsch et al. (2007). Also, similarly to our earlier study of a largely different cohort (Durazzo et al., 2004), sADIs demonstrated patterns of more NAA abnormalities than nsADIs. The DTI indices were not related to metabolite concentrations within alcohol-dependent individuals, similar to the findings by Cader et al. (2007) among patients with multiple sclerosis. Taken together, these results suggest that DTI and MRS imaging yield complimentary information on white matter injury. We are aware of published associations between spectroscopic and DTI measures in cohorts including both patients and healthy controls, such as in a study that compared a small cohort of patients with multiple sclerosis with healthy controls (Irwan et al., 2005). In fact, we also observed relationships between NAA and MD within frontal and parietal white matter of all alcohol dependent individuals combined and nsLD groups (r < −0.38, P < 0.048, corrected), which were absent within alcohol-dependent individuals and within nsLDs, suggesting that the correlations were driven by (on average) lower NAA and higher MD in alcohol-dependent individuals versus nsLDs (cf. Fig. 2).

Changes over 1 month of abstinence from alcohol

In the absence of significant white matter volume increases in nsADIs, FA increased significantly only in temporal white matter and this 1 month change was accompanied by a pattern of MD decreases in multiple white matter regions, suggesting remyelination and/or recovery of axonal membranes during abstinence from alcohol in nsADIs. However, much to our surprise, we observed a consistent pattern of volumetric increases in sADIs that were not accompanied by changes in DTI indices. In nsADIs, the result could be interpreted as microstructural recovery, likely due to remyelination, preceding measurable white matter volume increases. The sADIs, however, demonstrated qualitatively less microstructural injury than nsADIs at 1 week of abstinence and the white matter recovery processes were reflected as significant volumetric increases. In fact, in a larger sample of abstinent nsADIs, we have observed no white matter volume changes over the first month of abstinence, followed by white matter volume increases over the subsequent 6 months of abstinence. Similarly, the white matter volumetric increases in sADIs were also reproduced in a much larger sample of sADIs and this volumetric enlargement continued over the ensuing 6 months of abstinence (to be reported elsewhere). These different patterns of white matter volumetric changes with abstinence in sADIs and nsADIs suggest that the discrepant results of previous volumetric studies (Pfefferbaum et al., 1995; Agartz et al., 2003) might be explained by different proportions of smokers and non-smokers in these cohorts and by the different group patterns of volume changes during abstinence.

The observed DTI and volumetric changes were not paralleled by significant NAA increases in either group, suggesting little metabolic recovery in axons. The lack of accompanying significant changes in Cho concentrations that are associated with membrane turnover may reflect a shift in equilibrium between membrane anabolic and catabolic processes, which does not change the total amount of choline-containing metabolites. This interpretation is consistent with a computer tomography study that demonstrated increasing tissue density during abstinence from alcohol (Trabert et al., 1995) and with no change in spectroscopic signal from tissue water (Bartsch et al., 2007), both over about 1 month of abstinence from alcohol.


The limitations of this study included use of median diffusion indices calculated over large regions that could be affected adversely by crossing fibres. The three slices of the MRS image acquisition spatially limited the brain regions used for analyses. Although we had a relatively small longitudinal cohort, the patterns of volumetric changes with abstinence are consistent with the results that we obtained in a significantly larger cohort. The cohort used in our study included mostly male participants recruited at a Veterans Administration Medical Centre, so that sex effects of concurrent alcohol dependence and cigarette smoking could not be assessed. Potential unrecorded differences in nutrition, exercise, general health and genetic predispositions between study groups may influence the results described in this study. Finally, future studies assessing injury and recovery of specific white matter tracts with higher spatial resolution and/or higher magnetic field strengths may yield more reliable diffusion indices and allow evaluating specific white matter neurocircuitry known to be disrupted by substance dependence.


Different imaging modalities seem to provide complimentary information and a more detailed understanding of brain injury in alcohol dependence and neurordegenerative processes during abstinence from alcohol. Short-term abstinence from alcohol is associated with white matter recovery and appears to be prominently driven by its glial component both in smoking and non-smoking alcoholics. This process may be reflected either as white matter microstructural improvement in nsADIs or as white matter volume increases in sADIs. Thus, co-morbid cigarette smoking may modulate results in studies of neural repair. We believe our results to be important, as they demonstrate that use of a single magnetic resonance imaging modality provides an incomplete picture of neurobiological processes associated with alcohol induced brain injury and recovery thereof or even lead to improper interpretation of results.


This material is the result of work supported with resources and the use of facilities at the Radiology Research Service of the Veterans Administration Medical Centre in San Francisco. National Institutes of Health (AA10788 to D.J.M.); Radiology Research Service of the Veterans Administration Medical Centre in San Francisco.


We thank Mary Rebecca Young and Bill Clift of the San Francisco VA Substance Abuse Day Hospital and Dr David Pating, Karen Moise and their colleagues at the San Francisco Kaiser Permanente Chemical Dependency Recovery Program for their valuable assistance in recruiting research participants. We thank Dr Susanne Mueller for help with setting up the Expectation–Maximization Segmentation procedures.


  • Abbreviations:
    body mass index
    choline-containing compounds
    dependent measures multivariate analysis of variance
    diffusion tensor imaging
    effect size
    fractional anisotropy
    1H MRS
    proton magnetic resonance spectroscopy
    multivariate analyses of co-variance
    mean diffusivity
    non-smoking alcohol-dependent individual
    non-smoking light drinkers
    smoking alcohol-dependent individual


View Abstract