Cognitive performance and MR markers of cerebral injury in cognitively impaired MS patients

Participants.

This study was approved by the institutional review board for human subject research at the State University of New York (SUNY) at Stony Brook. Participants were 37 individuals with definite MS19 who had undergone baseline cognitive testing as part of a clinical trial to enhance cognitive function with an acetylcholinesterase inhibitor (donepezil). Subjects were largely recruited from the MS Comprehensive Care Center at Stony Brook as well as outside physician referrals.

Eligibility criteria included the presence of at least mild cognitive impairment, operationally defined as a score at least 0.5 SD below age- and gender-based normative data on the Rey Auditory Verbal Learning Test (RAVLT), a test of verbal learning and memory.20 Patients with severe cognitive impairment on the Mini-Mental State Examination (MMSE < 26) were excluded (mean = 28.5, SD = 1.3).21 One person initially scoring 24 on the MMSE was admitted after scoring 26 upon later readministration of the task. Patients with severe depression on the Montgomery–Asberg Depression Scale (MADRS > 14) were also excluded.22 All participants were ambulatory and had Expanded Disability Status Scale (EDSS)23 scores of ≤6.5 (mean = 3.8, range = 0 to 6.5, SD = 1.9). Concurrent antidepressants, antispasticity agents, and disease-modifying therapies (interferon-β or glatiramer acetate) were permitted so long as dosage had been constant for at least 1 month prior to the evaluation. Persons currently taking benzodiazepines were excluded. Other exclusion criteria included current alcohol or substance abuse, history of head injury, or other medical condition known to affect cognition.

The MS subjects ranged in age from 20 to 55 years (mean = 44.2 years, SD = 8.1 years), with 10 to 20 years of education (mean = 15.0 years, SD = 2.4 years). Participants were predominantly women (81.1%). The disease course was relapsing remitting in the majority (59.5%) and secondary progressive in the remainder.

Neuropsychological tests.

A modified version of the Brief Repeatable Battery (BRB)24 formed the core of the testing protocol. The BRB tasks are among those most sensitive to cognitive impairment in MS.1,25⇓ The battery has served as an outcome measure in several clinical trials and longitudinal studies.10,13,26,27⇓⇓⇓ The BRB was supplemented by the Tower of Hanoi (TOH) to further assess executive functions beyond verbal fluency assessed by Controlled Oral Word Association (COWA) in the BRB.28,29⇓

The specific BRB based tasks were the following: 1) Selective Reminding Test (SRT; six-trial version), a verbal learning/memory task. Measures used were total recall, long-term storage, consistent long-term retrieval, and delayed recall. 2) 10/36 Spatial Recall Test (10/36), a visuospatial learning/memory task. Measures used were total correct during the three learning trials and delayed recall. 3) Symbol–Digit Modalities Test (SDMT; oral version), which assessed sustained concentration and speeded visual information processing. The measure used was total correct. 4) Paced Auditory Serial Addition Test (PASAT), which requires working memory, sustained concentration, and information-processing speed. The measures used were total correct for the 2- and 3-second forms of the task as well as the mean total number of dyads (consecutive correct responses) across both forms. 5) COWA, Category Fluency, which was used to measure semantic fluency. The measure used was the mean number of correct words in the categories of animals and fruits/vegetables. 6) TOH, considered a test of executive functions, used to measure conceptual planning as well as procedural learning.30 Two alternative forms were developed by dividing the original problems into two sets matched according to difficulty.28 Each subject performed both halves. The score used was the mean number of items solved in the two TOH forms. Overall neuropsychological performance was also assessed, by calculating the unweighted z score mean of six measures, one from each task: SRT total recall, 10/36 total, SDMT total, PASAT total for 2- and 3-second forms combined, COWA mean across both forms, and TOH mean across both forms.

MR procedures.

MRI/MRS scanning sessions were performed using a 1.5 T Marconi Edge whole-body scanner (Marconi Medical Systems, Cleveland, OH) with the body coil as the transmitter and a birdcage head coil as the receiver. A three-dimensional spoiled gradient-recalled (SPGR) sequence was employed to acquire volumetric T1-weighted axial images covering the whole brain with 30° flip angle, echo time (TE) of 5 milliseconds, repetition time (TR) of 30 milliseconds, 24-cm field of view (FOV), and 256 × 256 matrix size. Slice thickness was 3.0 mm for the first 16 subjects and was decreased to 1.5 mm for the last 17 in an effort to improve the precision of cerebral atrophy measurement. The results of statistical analyses were quite similar for the two groups assessed separately, and so the analyses presented are based on the combined groups. A three-dimensional sequence with fat suppression was used to collect volumetric T2-weighted axial images with the same acquisition location and parameters except for TE of 95 milliseconds, TR of 4,000 milliseconds, and echo train length of 136. Fluid-attenuated inversion recovery (FLAIR) images were also acquired from each patient with the same slice location, slice number, slice thickness, and FOV. Single-slice multivoxel ¹H MRS was performed with a PRESS sequence with TE of 135 milliseconds, TR of 1,500 milliseconds, 16-cm FOV, two-dimensional phase encoding (16 × 16), and two scan averages. Only the center 8 × 8 array of voxels of the MRS FOV was refocused to give MRS signals. The outer frame was not refocused to prevent signal from scalp lipid from being aliased. The slice of interest with 2-cm thickness was taken through the posterior and anterior aspects of the corpus callosum.18 The resulting nominal size of each MRS voxel is 2 cm³. The total MRI/MRS scanning time was <1 hour.

The three-dimensional T2-weighted images were used for stripping away the skull/scale and extracting a brain mask volume. A multispectral segmentation scheme based on a hidden Markov random field model and an expectation maximization algorithm31 were then applied to both T1- and T2-weighted images for classifying voxels within the mask, resulting in volumes of gray matter (GM), white matter (WM), and total CSF. The central CSF volume was delineated by both morphology and region-growing technologies, beginning with the selection of a central CSF seed. In datasets where the central CSF and peripheral CSF were connected, these connections could induce an overestimation of the central CSF. Therefore, a user-friendly semiautomated tool was developed to assist the radiologists to appropriately cut the connections between the central and peripheral CSF regions. The addition of FLAIR images helped distinguish lesions. Small lesions or low-probability lesions were eliminated first. FLAIR image artifacts, which are easily distinguished after segmentation, were eliminated through the semiautomatic tool. Then, the detected lesions were subtracted from the segmented CSF result, avoiding overestimation of the CSF volume. The primary measure of interest was central cerebral atrophy, assessed by the percent of central CSF to total intracranial volume (CCSF/[GM + WM + CSF + lesion] × 100). Central atrophy has been found to have particularly high rates of increase per year in longitudinal MS studies.32

Given the high prevalence of periventricular lesions in MS and our prior experience correlating NAA levels in this region with cognitive deficits in MS,18 we examined the left and right posterior periventricular (PPV) regions in the MRS analyses. The PPV voxels were selected through elimination of voxels with >30% CSF and by anatomic placement.18 Proton spectra from these voxels were processed with 3-Hz line-broadening, phase, and baseline correction and fitted using a nonlinear least-squares fitting procedure with a Levenberg–Marquardt algorithm. The areas of resonance peaks of NAA, Cr, and Cho were measured, and their ratios were calculated. There were two ratio measures of interest: NAA/Cr and NAA/Cho.

Procedure.

Neuropsychological tests were administered as part of a larger battery of tasks requiring approximately 2.5 to 3 hours of testing in a single session. MR measures were collected within 2 weeks of neuropsychological testing. The reported neuropsychological and MR assessments were performed at baseline prior to the initiation of treatment for cognitive impairment.

Statistical analyses.

All neuropsychological and MR variables met conditions of normality,33 with the exception of central cerebral atrophy, which was positively skewed because of three or four patients with particularly large central atrophy ratios. To reduce these outliers, a log-10 transformation was performed on the central cerebral atrophy measure. Because transformed variables can sometimes be difficult to interpret, results for both raw and transformed versions of this variable are presented for comparison. The results were quite similar in either case. It should also be noted that Spearman correlations were also analyzed and led to results statistically similar to those presented here. The primary analyses in this study consisted of partial correlations (partial r ), controlling for age and education, to assess the relation between neuroimaging measures and cognition. These partial correlations represent the correlation between an MR and cognitive variable that remains after removing any subject differences on either variable that arise owing to either age or education. All analyses involving cognition controlled for the effects of age and education. For the purpose of this exploratory study, a significance value of p ≤ 0.05 (two tailed) was used throughout, despite the large number of correlations, though many of those reached significance at far more stringent levels. All analyses were performed with SPSS 11.0 statistical software (Chicago, IL).

Neuropsychological performance.

This group of MS patients showed a range of cognitive deficits. As shown in table 1, the MS participants performed most poorly on the verbal learning/memory measures of the SRT, performing 1.0 to 1.7 SD below norms, generally consistent with the classification of mild cognitive impairment as compared with the largest available published sample of comparable healthy persons.26,34⇓ The relatively greater deficit on verbal learning and memory measures was consistent with the selection criterion of persons having difficulty on the RAVLT, another verbal learning/memory task. Whereas the screening criterion for the RAVLT was 0.5 SD below age- and gender-based normative data,20 the participants actually performed at a lower level, consistent with their performance on the SRT measures (mean = −1.5, SD = 0.7). The current sample also performed approximately 1 SD below the performance on the SRT measures of a large sample of patients with relapsing–remitting MS and approximately 0.4 SD below on the other BRB measures.13,26,27,35⇓⇓⇓

Correlations involving neuropsychological and MR measures.

There was a clear and consistent association between cognitive performance and MR measures. The partial correlations between the neuropsychological and MR measures, controlling for age and education, resulted in numerous significant correlations (table 2). Large correlations were identified between the global measure of overall neuropsychological performance and MR measures. The correlation between overall neuropsychological performance and central cerebral atrophy accounted for almost half the shared variance (partial r ² = 0.467; figure 1), representing a large effect size.36 Each of the cognitive tasks significantly correlated with at least one MR measure, though correlations for the SRT were less likely to reach significance. Among the tasks, the SDMT measure of sustained concentration and information-processing speed/efficiency tended to display some of the strongest correlations with the MR measures. The correlation between SDMT and central atrophy accounted for approximately half the shared variance (partial r ² = 0.494). At least one of the four MRS measures was found to correlate with each of the cognitive tasks, with the exception of the SRT. Correlations for MRS measures with cognitive functioning tended to be stronger in the right hemisphere locations. Lesion volume correlated with all cognitive tasks with the exception of the SRT.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Relation between overall neuropsychological performance z score and central cerebral atrophy (log-10 transformation of central CSF percent of total intracranial volume) (r = −0.699, p < 0.001).

The different MR parameters also correlated with one another. Again, correlations involving the MRS measures were strongest for right hemisphere locations. Pearson correlations between central atrophy and MRS measures ranged from r = −0.218 (p = 0.189) with left PPV NAA/Cho to r = −0.611 (p < 0.001) with right PPV NAA/Cr. The correlation between central atrophy and lesion volume was r = 0.537 (p = 0.001). Correlations between MRS measures and lesion volume ranged from r = −0.275 (p = 0.095) with left PPV NAA/Cho to r = −0.607 (p < 0.001) with right PPV NAA/Cr.

A multiple regression analysis assessing the combination of all MR measures (central atrophy, all four NAA ratios, lesion volume) was more highly associated with overall neuropsychological performance than were any of the neuroimaging measures in isolation (R = 0.796, p < 0.001), accounting for well over half the variance in cognitive performance (R ² = 0.634, adjusted R ² = 0.561). With the addition of age and education to the equation, the multiple regression model reached R = 0.809, p < 0.001. The combination of all MR measures accounted for a great deal more of the subject variability in cognition beyond that explained by age and education alone (R ² change = 0.586, p < 0.001).

The strongest combination of central atrophy and lesion volume with a single MRS measure was with right PPV NAA/Cho (R = 0.785, p < 0.001; figure 2), a combination that displayed an association with overall neuropsychological performance that rivaled that of the combination of all neuroimaging measures (R ² = 0.616, adjusted R ² = 0.581). With the addition of age and education, the regression model reached R = 0.794, p < 0.001. The combination of the three MR measures (central atrophy, lesion volume, and right PPV NAA/Cho) accounted for far more variance in cognition than age and education alone (R ² change = 0.564). To assess the unique contribution of each MR variable in the latter analysis, partial correlations of each MR variable with overall cognition were conducted, controlling for age, education, and the other two MR variables. The resulting correlations were as follows: central atrophy (partial r = −0.531, p = 0.001), right PPV NAA/Cho (partial r = 0.318, p = 0.072), and lesion volume (partial r = −0.313, p = 0.076).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Relation between overall neuropsychological performance z score and standardized predicted score based on a multiple regression model containing the following predictors: central cerebral atrophy, lesion volume, and RPPV NAA/Cho (R = − 0.785 p <0.001).

Correlations involving EDSS were also examined. EDSS was found to correlate with overall neuropsychological performance (r = −0.340, p = 0.040), central atrophy (r = 0.381, p = 0.020), and lesion volume (r = −0.415, p = 0.011). Correlations of EDSS with MRS measures ranged from (r = −0.119, p = 0.485) for left PPV NAA/Cr to (r = −0.277, p = 0.097) for right PPV NAA/Cho.