Different associations of white matter lesions with depression and cognition

Subjects were comprised of subjects recruited from 3 academic dementia centers (University of California Davis, University of California San Francisco, and University of Southern California) as part of a multicenter collaborative study examining contributions of subcortical ischemic vascular dementia and Alzheimer’s disease to cognitive impairment in older adults. Exclusion criteria included age younger than 55 years, non-English speaking, severe dementia (Clinical Dementia Rating (CDR) > 2), evidence of alcohol or substance abuse, head trauma with loss of consciousness lasting longer than 15 minutes, severe medical illness, neurologic or psychiatric disorders other than dementia, or currently taking medications likely to affect cognitive function. Severely depressed patients who had recurrent suicidal ideas or severe impairments in social function were also excluded. In addition, subjects were excluded if the MRI showed evidence of cortical infarction, hemorrhage, or structural brain disease other than atrophy, lacunes, or WML.

Participants were grouped by the levels of cognitive impairment as normal cognition control (CDR = 0), cognitively impaired but not demented (CIND) (CDR = 0.5), and demented (CDR ≥ 1). Alzheimer’s disease were diagnosed using National Institute of Neurologic and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) [12] diagnostic criteria and ischemic vascular dementia were diagnosed by California Alzheimer's Disease Diagnostic and Treatment Centers (ADDTC) [13] criteria.

The study had been reviewed and approved by the institutional review boards of all study sites and written informed consent had been obtained from the study participants or their legal representatives.

All participants received a standardized battery of neuropsychological tests. The composite scores were calculated using item response theory methods and transformed to a standard scale with a mean of 100 and a SD of 15 [14]. Verbal memory score was calculated based on the Word List Learning tasks of the Memory Assessment Scales, visual memory score on the tasks of the Biber Figure Learning Test and executive score on the Initiation-Perseveration task from the Mattis Dementia Rating Scale, the letter fluency task, the digit span backward and the visual span backward tasks from the Wechsler Memory Scale-Revised. Every subject received clinical interview for depression by clinical psychologists using Minimum Uniform Data Set (MUDS) structured interview [15]. Depressive score was defined as the number of depressive symptoms meeting the nine criterion symptoms by Diagnostic and Statistical Manual of Mental Disorders, third edition, revised (DSM-IIIR) [16]. The number of depressive symptoms based on DSM criteria reflects severity of disease well [17]. The CDR and the Mini Mental State Examination were also used.

The entire brain was imaged using a 1.5 T Magnetom VISION system (Siemens, Erlangen, Germany) equipped with a standard quadrature head coil. We derived WML volume from proton density (PD)-weighted (PD) and T2-weighted spin-echo axial images from a double spin-echo sequence (repetition time (TR)/echo time (TE)1/TE2 = 2500/20/80 msec; 1.0 × 1.4 mm² in-plane resolution; slice thickness 3 mm oriented along the AC-PC line) and regional frontal gray matter volumes from T1-weighted coronal three dimensional images from the magnetization-prepared rapid gradient Echo (MPRAGE) (TR/TE/TO = 13.5/7/300 msec; 1.0 × 1.0 mm² in plane resolution; slice thickness 1.4 mm oriented along the long axis of the hippocampus).

Total WML volume and total intracranial volume (TIV) were based on the multi-channel segmentation of Expectation Maximization Segmentation [18] and lobar WML segmentation were based on the atlas-based deformable registration method [19]. The technical details of these procedures are described in prior publication [11]. Lacunes were defined as small (> 2 mm) areas of subcortical gray and white matter with increased signal relative to Cerebrospinal fluid (CSF) on proton density MRI.

Frontal regional cortical volumes were based on the Freesurfer image analysis software 4.0 version (http://​surfer.​nmr.​mgh.​harvard.​edu/​). The technical details of these procedures are described in prior publication [20]. Cortical volume combined both thickness and area information and parcellation of the cerebral cortex into units based on gyral and sulcal structure [21] were done to get regional frontal volumes.

To account for variations in head sizes, all gray matter volumes and WML volume were normalized to each TIV according to: VOLn = VOLr × TIVm/TIVr. Here VOLn and VOLr are the normalized volume and the raw volume of a subject, respectively; and TIVm and TIVr are the mean TIV from all subjects and the TIV of the subject, respectively. Volume unit is cm³. WML volumes were log-transformed to achieve normal distribution. We did same analysis in subjects who had not lacunar infarcts to test these relationships without the effect of lacunes.

We used multiple linear regression models. Independent variables were WML volume, age, gender, years of education, and groups and dependent variable was each lobar cortical volume or each regional frontal cortical volume. The interaction between WML and group status on predicting frontal cortical volume was also tested in the same linear regression model by examining the difference of regression coefficients between WML volume and frontal gray matter volume among control, CIND subjects, and dementia patients. Plots of residuals were used to evaluate model assumptions. Residuals appeared normal and linear regression assumptions were satisfied. Regression coefficients were standardized by all variables centered on the mean and scaled by the standard deviation to compare them easily.

We employed the structural equation modeling (SEM) to test for the mediation effect of frontal cortex on WML association with depressive symptoms or cognitive function. Mediation means that independent variable (A: WML) causes an intervening variable (B: frontal cortex), which in turn causes the dependent variable (C: depression or cognition) [22] (Figure1). Four conditions must be met for B to be a mediator: 1) A is significantly associated with C; 2) A is significantly associated with B; 3) B is significantly associated with C after controlling for A; 4) the impact of A on C is significantly less after controlling for B [23]. We used SEM to test these conditions because we had multiple measures for frontal cortex and cognition [23]. We reported model fitness by the χ², the comparative fit index (CFI), and the root mean-square error of approximation (RMSEA). An excellent fitting model is usually indicated by a non-significant χ², CFI >0.90, and RMSEA <0.1 [24]. Path coefficients were standardized by subtracting the sample mean and then dividing by the sample SD. The Amos Version 18.0 software (SPSS, Chicago, Illinois) was used for all SEM analyses utilizing maximum likelihood estimation. The Kruskal-Wallis test was used to test for mean differences among groups. A p value less than 0.05 was considered significant.

Table1 shows demographic data, results of cognitive tests, and MRI volumes for all 161 subjects. The mean age of subjects was 74.4 (SD = 7.8) and the mean years of education was 14.9 (SD = 3.2). 56 percent of participants were male (n = 90). Sixty-four subjects had normal cognition, 44 subjects were CIND, and 53 subjects were diagnosed with dementia. Thirty-four of the demented subjects were diagnosed with Alzheimer’s disease, 9 with mixed Alzheimer's disease and subcortical ischemic vascular dementia, and 10 with subcortical ischemic vascular dementia. Seventy-three subjects had no lacune (41 in normal controls, 12 in cognitively impaired but not demented subjects, 20 in dementia subjects). Twenty-five subjects were diagnosed with current major depressive disorder or had been diagnosed with previous major depressive disorder (4 in normal controls, 11 in cognitively impaired but not demented subjects, 10 in dementia subjects).

We conducted linear regression to determine the relationship between total WML volume and each lobar cortical volume after controlling the effects of age, gender, years of education and groups. Total WML predicted only frontal cortical volume (standardized β = −0.24, SE = 0.07, p = 0.002) (Figure2) and did not predict other lobar cortical volumes. There was no interaction between group status and WML volume in predicting frontal cortical volume (F (2, 154) = 0.64, p = 0.53). It indicated that group status did not affect the association between WML and frontal cortical volume. In subjects who had no lacune, total WML predicted only frontal cortical volume, too (standardized β = −0.24, SE = 0.08, p = 0.002).

All lobar WML volumes [frontal WML (β = −0.24, SE = 0.08, p = 0.003); temporal WML (β = −0.14, SE = 0.07, p = 0.05); parietal WML (β = −0.27, SE = 0.07, p < 0.001); and occipital WML (β = −0.15, SE = 0.07, p = 0.04)] predicted frontal cortical volume.

We conducted linear regression to determine the relationship between total WML volume and each regional frontal cortical volume after controlling the effects of age, gender, years of education and group status (Table2). Among regional frontal cortices, only medial orbitofrontal cortical volume (β = −0.16, SE = 0.07, p = 0.04) was significantly associated with total WML (Figure2). Pars opercularis (β = −0.15, SE = 0.08, p = 0.08) and precentral gyrus (β = −0.14, SE = 0.08, p = 0.08) showed a trend to be associated with total WML. In subjects who had no lacune, only medial orbitofrontal cortical volume was significantly associated with total WML, too (β = −0.16, SE = 0.08, p = 0.04).

Because total WML predicted frontal cortical volume atrophy, we used SEM to test relationship among WML volume, frontal cortex, and depressive symptoms (Figure3). Frontal cortex was defined as a latent variable and constructed by three indicators which had a trend to be associated with WML: medial orbitofrontal cortical volume, pars opercularis cortical volume, and precentral gyral cortical volume. All three indicators predicted frontal cortex significantly (p < 0.001).

In the direct effect model, we included total WML volume as a predictor and depressive score as a dependent variable. WML predicted depressive score (β = 0.30, SE = 0.08, p < 0.001) significantly.

In the mediation model, we added frontal cortex as a mediator between total WML and depressive score. This model showed an excellent fit to the data (χ² = 4.4, df = 4, p = 0.36, CFI = 0.99, RMSEA = 0.03) and total WML predicted frontal cortex (β = −0.47, SE = 0.12, p < 0.001) and depressive score directly (β = 0.29, SE = 0.10, p = 0.003) but frontal cortex did not predict depressive score (β = −0.02, SE = 0.10, p = 0.85). It indicated WML - depression path was direct. When WML - depression path were constrained to zero, χ² change was significant (χ² change = 8.0, df change = 1, p < 0.01). It indicated WML - depression path was not mediated by frontal cortex. Therefore, we adopted the direct WML - depression effect model as a final depression predicting model (Figure3). In subjects who had no lacune, the direct WML - depression effect model was also well-fitted.

In multiple linear regression controlling for age, gender, years of education, and group status, among all regional frontal cortex and all lobar WML, depressive score was also only predicted by lobar WML: frontal WML (β = 0.26, SE = 0.09, p = 0.003) and temporal WML (β = 0.22, SE = 0.08, p = 0.005).

We used SEM to test relationships among total WML volume, frontal cortex and cognitive function (Figure3). Cognition was defined as a latent variable and constructed by three indicators: verbal memory score, visual memory score, and executive score. All indicators predicted cognition significantly (p < 0.001).

In the direct effect model, we included WML volume as a predictor, cognition as a dependent variable. Total WML predicted cognition (β = −0.25, SE = 0.09, p = 0.01) significantly.

In the mediation model, we added frontal cortex as a mediator between total WML and cognition. This model showed a good fit to the data (χ² = 20.7, df = 12, p = 0.06, CFI = 0.97, RMSEA = 0.08). Total WML significantly predicted frontal cortex (β = −0.50, SE = 0.10, p < 0.001) but did not predict cognition directly (β = 0.003, SE = 0.09, p = 0.97). Frontal cortex predicted cognition (β = 0.51, SE = 0.13, p < 0.001) significantly. It indicated the effect of total WML on cognition was indirect. When WML - cognition path were constrained to zero, χ² change was not significant (χ² change = 0.1, df change = 1, p > 0.05). It indicated WML - cognition path was mediated by frontal cortex. Because direct WML - cognition path was not significant, we adopted the mediation model without direct WML - cognition path as a final cognition predicting model. In subjects who had no lacune, the mediation model without direct WML - cognition path was also well-fitted.

In multiple linear regression, cognition was associated with regional frontal cortex and lobar WML: verbal memory score was predicted by medial orbitofrontal cortex (β = 0.11, SE = 0.05, p = 0.05), superior frontal gyrus (β = 0.11, SE = 0.05, p = 0.05), temporal WML (β = 0.11, SE = 0.05, p = 0.05), and parietal WML (β = 0.13, SE = 0.06, p = 0.03); visual memory score was predicted by medial orbitofrontal cortex (β = 0.15, SE = 0.06, p = 0.01) and lateral orbitofrontal cortex (β = 0.14, SE = 0.06, p = 0.02); and executive score was predicted by precentral gyrus (β = 0.16, SE = 0.08, p = 0.04).