Mixed-location cerebral microbleeds as a biomarker of neurodegeneration in a memory clinic population

Characteristics of participants

Table 1 shows the characteristics of the patients with and without CMB. Of 477 patients, 41.9% (n=200) had at least 1 CMB. Among 200 patients, 53.5% (n=107) were strictly lobar, 17.5% (n=35) were strictly deep and 29.0% (n=58) were mixed-location CMBs. Among the patients with mixed-location CMBs (n=58), the median number of lobar CMBs was 5 and the median number of deep CMBs was 2. Patients with CMBs had higher white matter hyperintensities (WMH) volume (p<0.001) and had a higher prevalence of lacunes (p=0.032) compared to those without CMB. There were no differences in demographics, vascular risk factors (hypertension, hyperlipidemia and diabetes), total enlarged perivascular spaces (ePVS) and neurodegenerative markers (global cortical thickness, subcortical, total white matter and total intracranial volume) in patients with and without CMBs (p>0.05).

Association between CMBs and cortical thickness

Table 2 shows the association between CMBs and cortical thickness. Mixed-location CMBs were associated with smaller global cortical thickness (p<0.005). Region specific analysis showed that mixed-location CMBs were associated with smaller cortical thickness in frontal, temporal and limbic lobes (p<0.05). On applying Bonferroni correction, the association between mixed-location CMBs and cortical thickness of limbic lobe remained significant. On categorizing mixed-location CMBs into presence of 2-4 CMBs and presence of >4CMBs; the presence of >4 CMBs were associated with smaller global cortical thickness as well as region specific cortical thickness in frontal, temporal, limbic and insular lobes when compared to patients with no CMBs. Most of these associations survived multiple testing. Similar associations were observed when CMBs locations (strictly lobar, strictly deep and mixed-location) were treated as categorical data (presence vs absence) [Supplementary Table 1]. No association was observed between strictly lobar and strictly deep CMBs with global and region specific cortical thickness.

Association between CMBs and subcortical structures volume

Table 3 shows the associations between CMBs and subcortical structural volumes. Mixed-location CMBs were associated with smaller accumbens volume (p<0.05) and a borderline significance with total hippocampus volume (p=0.075) independent of demographic and vascular risk factors. On applying Bonferroni correction, association with acumbens volume remained significant. Further categorization of mixed-location CMBs into 2-4 and >4 CMBs revealed an association between presence of >4 CMBs with smaller accumbens and brain stem volumes. By contrast, mixed-location CMBs were associated with a larger thalamic volume and presence of >4 mixed-location CMBs were associated with larger lentiform nucleus volume (p<0.05). These associations survived multiple testing. In case of strictly lobar CMBs, 1 CMB and 2-4 CMBs were associated with smaller lentiform nucleus volume and >4 CMBs were associated with smaller hippocampus and brain stem volumes. Most of these associations were still observed in the analysis involving presence and absence of CMBs [Supplementary Table 2]. No significant association was found between strictly deep CMBs with subcortical structures volumes.

Figure 1 shows the association between mixed-location CMBs and hippocampus subfield volumes. Mixed-location CMBs were associated with smaller volumes of hippocampus tail, subiculum, CA1, presubiculum, molecular layer, GCMLDG, CA3 and CA4 (p<0.05). No significant association was found between strictly lobar and strictly deep CMBs with hippocampus subfield volumes [data not shown]. When treating location of CMBs as categorical data (presence vs absence) no significant association was again found between presence of strictly lobar, strictly deep and mixed-location CMBs with hippocampus subfield volumes [data not shown].

Figure 1. Forest plot for the association between mixed-location CMBs with hippocampus subfields volume. Effect estimates adjusted for age, gender, intracranial volume, hypertension, hyperlipidemia and diabetes.Bold values represents statistically significant associations at p =0.05. * Statistically significant after Bonferroni correction (0.05/12 ~ 0.0041). # Statistically significant after further adjusting for white matter hyperintensities volume, presence of lacunes and total enlarged perivascular spaces (n=328) (p<0.05). CMBs, cerebral microbleeds; CA, Cornu Amonis; GCMLDG, Molecular and Granule Cell Layers of the Dentate Gyrus; HAAT, Hippocampal Amygdala Transition Area.

Association between CMBs and white matter volume

Table 4 shows the association between CMBs and white matter volumes. Strictly lobar CMBs were associated with smaller total white matter volume as well as smaller volumes of frontal, parietal, temporal and occipital white matter (p<0.05). These associations remained significant after applying Bonferroni correction. On categorizing strictly lobar CMBs into presence of 1, 2-4 and >4 strictly lobar CMBs; presence of >4 strictly lobar CMBs were associated with smaller total white matter volume as well as in the frontal, parietal and temporal lobes white matter volumes, which did not survive multiple testing. Deep CMBs were associated with larger parietal, temporal and occipital white matter volume which were stronger in the presence of >4 deep CMBs. On applying Bonferroni correction, only association with occipital lobe remained significant. Similar associations were observed when CMBs locations (strictly lobar, strictly deep and mixed-location) were treated as categorical data (presence vs absence) [Supplementary Table 1].

Subgroup analysis with other MRI markers

In the subgroup analysis consisting of 328 patients, mixed-location CMBs were associated with smaller global cortical thickness [β= -0.03; 95% CI= -0.05, -0.01, p=0.013] as well as cortical thickness in frontal [β= -0.02; 95% CI= -0.04, -0.00, p=0.043], parietal [β= -0.02; 95% CI= -0.04, -0.00, p=0.004], occipital [β= -0.02; 95% CI= -0.04, -0.00, p=0.020] and limbic [β= -0.03; 95% CI= -0.05, -0.01, p=0.010] lobes, smaller brainstem volume [β= -0.03; 95% CI= -0.04, -0.01, p=0.001] and smaller volumes of all the hippocampus subfields (p<0.05) when further adjusted for total WMH volumes, presence of lacunes and total ePVS. Similarly, strictly lobar CMBs were associated with smaller total white matter volume and region specific white matter volumes in frontal, parietal, temporal and occipital lobes (p<0.05) [data not shown].

Stratified analysis by diagnostic groups

On performing secondary analysis among patients with NCI, CIND and dementia, we found that in patients with CIND, strictly deep CMBs were associated with smaller cortical thickness in temporal lobe [β= -0.14; 95% CI= -0.27, -0.01, p=0.040] and with larger parietal [β= 0.16; 95% CI= 0.02, 0.31, p=0.027], temporal [β= 0.14; 95% CI= 0.00, 0.27, p=0.044] and occipital [β= 0.33; 95% CI= 0.20, 0.47, p<0.001] white matter volumes whereas, mixed-location CMBs were associated with smaller accumbens volume [β= -0.03; 95% CI=- 0.04, -0.01, p=0.001]. However, in patients with dementia, mixed-location CMBs were associated with larger thalamic [β= 0.02; 95% CI= 0.02, 0.03, p<0.001] and lentiform nucleus volumes [β= 0.01; 95% CI= 0.00, 0.02, p=0.010] and smaller total and region specific white matter volumes (p<0.05). No significant associations were found between strictly lobar, strictly deep and mixed-location CMBs with hippocampus subfield volumes among patients with NCI, CIND and dementia [data not shown].

Study population

For this study, data was obtained from the ongoing prospective memory clinic based case-control study. The controls were individuals who may have subjective cognitive complains, but no objective cognitive impairment on comprehensive neuropsychological tests or functional impairment and were diagnosed as No Cognitive Impairment (NCI). Cases were participants with subjective memory complains and impairment on neuropsychological assessment and were diagnosed with cognitive impairment no dementia (CIND) and dementia. CIND was defined as impairment in at least one cognitive domain on comprehensive neuropsychological test, but did not meet the criteria for dementia according to the Diagnostic and Statistical Manual for Mental Disorder-Fourth Edition (DSM-IV). Dementia was diagnosed according to the DSM- IV criteria. The etiological diagnoses of dementia were based on internationally accepted criteria: Alzheimer’s Disease (AD) was diagnosed using the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA), AD with cerebrovascular disease or mixed dementia was defined as subjects fulfilling criteria of diagnosis of AD together with concomitant cerebrovascular disease on MRI (presence of multiple infarcts or extensive white matter hyperintensities) and vascular dementia (VaD) was defined using the National Institute of Neurological Disorders and Stroke and Association Internationale pour la Recherché et l' Enseignement en Neurosciences (NINDS-AIREN) criteria [44–46].

A total of 578 patients were recruited for this study from August 2010 to January 2016. All subjects underwent comprehensive evaluation including physical, medical and neuropsychological assessment along with 3T magnetic resonance imaging (MRI), all done on the same day. Of 578 patients, 12 did not perform MRI scans (3 were claustrophobic, 1 refused MRI, 2 were uncooperative and could not follow instructions properly and 6 could not perform MRI due to other medical conditions), 18 patients with incomplete and poor quality MRI scan and 71 patients with cortical stroke were excluded, leaving 477 patients for the final analysis [NCI=104 (21.8%), CIND=211 (44.2%) and Dementia=162 (34.0%) (AD=65; 40.1%, VaD=20; 12.3%, mixed dementia=77; 47.6%)].

This study was approved by the National Healthcare Group Domain-Specific Review Board and was conducted in accordance with the Declaration of Helsinki. A written informed consent was obtained from all subjects or their caregivers prior to the recruitment for this study.

Covariates

A detailed demographic questionnaire was administered for each subject to collect information on age, gender, years of formal education and smoking. History of hypertension, hyperlipidemia, and diabetes mellitus was noted and verified with medical records. Hypertension was defined as systolic blood pressure ≥140mmHg and /or diastolic blood pressure ≥90mmHg or a history of hypertension, or use of antihypertensive medication. Hyperlipidemia was defined as total cholesterol level ≥4.14 mmol/l or a history of hyperlipidemia, or use of lipid-lowering medication. Diabetes mellitus was defined as glycated hemoglobin ≥6.5% or a history of diabetes mellitus, or the use of any glucose-lowering medication.

Neuroimaging

MRI scans of all the patients were performed at the Clinical Imaging Research Center of the National University of Singapore, using 3T Siemens Magnetom Trio Tim Scanner system, with a 32-channel head coil. The standardized neuroimaging protocol in this study included a three dimensional T1-weighted, T2-weighted, fluid-attenuated inversion recovery (FLAIR) and susceptibility weighted image (SWI) sequences. SWI sequence was used to detect CMBs with echo time = 20 ms; repetition time= 27 ms; flip angle= 15 degrees; field of view= 256 mm; field of view= 75%; image matrix= 192x256; slice thickness= 1.50 mm.

Cerebrovascular disease markers

MRI markers of cerebrovascular disease (lacunes, white matter hyperintensities, enlarged perivascular spaces and cerebral microbleeds) were graded using the Standards for Reporting Vascular changes on Neuroimaging (STRIVE) criteria [47].

1. CMBs were graded according to the Microbleed Anatomical Rating Scale [48]. CMBs were classified manually into two different locations: lobar (frontal, parietal, temporal, occipital and insula) and deep (basal ganglia, thalamus, internal and external capsule, brainstem and cerebellum). Based on the location of CMBs, patients were classified into 3 groups: strictly lobar CMBs (having CMBs exclusively in lobar region), strictly deep CMBs (having CMBs exclusively in deep region) and mixed-location CMBs (having CMBs in both lobar and deep locations). The total number of CMBs in each location was recorded. Both strictly lobar and strictly deep CMBs were further categorized in three groups according to CMBs burden: presence of 1 CMB, presence of 2-4 CMBs and presence of >4 CMBs, whereas mixed-location CMBs were categorized into two groups: presence of 2-4 CMBs and presence of >4 CMBs [49, 50].

2. Lacunes were defined as round or ovoid lesions, 3 to 15 mm in diameter in the subcortical regions, with a high signal on T2-weighted images and a low signal on T1-weighted and FLAIR images, and a hyperintense rim with a center following the cerebrospinal fluid intensity [47].

3. WMH volume was quantified using T1 and T2 weighted images. Image preprocessing and the tissue classification algorithm have been described elsewhere [51]. Region-specific WMH volume (ml) was calculated for frontal, parietal, occipital and temporal lobes. Total WMH volume was calculated as the sum of WMH volumes in above mentioned 5 regions [52].

4. ePVS were defined as round or linear lesion, which were hypointense on T1 weighted and hyperintense on T2 weighted images. It was considered dilated when lesion was ≥1mm in diameter. ePVs were visually graded in 4 different regions of brain: centrum semiovale, basal ganglia, mesencephalon and hippocampus. Total numbers of ePVS in each region were recorded. Total ePVS was calculated as sum of centrum semivale, basal ganglia, hippocampus and mesencephalon ePVS [52].

All the MRI scans were graded by two independent raters (B.G & M.A.S) blinded to the clinical history. All potential MRI markers of SVD were discussed in the weekly consensus meeting. Any disagreement was further discussed with an experienced neuroimaging fellow (SH) to make a final decision. Inter-rater and intra-rater agreement was excellent, which has been published previously [53, 54].

Neurodegenerative markers

Quantitative MRI analyses were performed using automated segmentation procedure at the Department of Medical Informatics, Erasmus University Medical Center, Netherlands using a model based automated procedure (FreeSurfer, v.5.1.0) on T1 weighted images (TR=7.2 ms, TE=3.3 ms, matrix=256×256×180 mm³). For each patient, the following MRI markers were computed.

1. Cortical thickness was measured as the shortest distance between gray/white matter boundary and pial surface at each vertex. Global and regional thickness averages were converted to millimeters (mm). Regional average was calculated from left and right lobes thickness using parcellation guide on sulcus and gyrus of cerebral cortex. Region specific cortical thicknesses were calculated for the frontal, parietal, occipital, temporal, insular, and limbic regions [55].

2. Subcortical structures volumes were calculated for accumbens, amygdala, lentiform nucleus, thalamus, hippocampus, and brainstem. Segmentation was performed using rigid body registration and nonlinear normalization of images to a probabilistic brain atlas. During segmentation process, each voxel was labeled automatically as corresponding brain region on parcellation guide. Volumes of accumbens, amygdala, lentiform nucleus, hippocampus and thalamus and were calculated for left and right hemispheres separately. Finally, average volumes of each structure were calculated and then converted into milliliters (ml) [39, 56].

3. Hippocampus subfield volumes were calculated in 12 subfields based on an ultra-high resolution ex-vivo atlas and includes: Cornu Amonis regions 1, 2 and 3 combined, and 4 (CA1, CA3, and CA4), parasubiculum, presubiculum, subiculum, molecular and granule cell layer of dentate gyrus (GCMLDG), hippocampus amygdala transition area (HATA), fimbria, molecular layer, hippocampal fissure, and hippocampal tail. For each subject both left and right hemisphere hippocampus subfield volumes were calculated separately. Finally, average volumes of each subfield were calculated and then converted into ml.

4. White matter volume was quantified using T1 and T2 weighted images. The used image preprocessing steps and the tissue classification algorithm have been described elsewhere [51]. Briefly, k-nearest-neighbor brain tissue classifier technique was used to classify voxels into cerebrospinal fluid (CSF), gray matter, white matter, white matter hyperintensities (WMH). Volume (ml) was calculated for all biomarkers from these segmentations. Region-specific white matter volume was calculated for frontal, parietal, occipital and temporal lobes. Total white matter volume was calculated as the sum of white matter volumes in the above mentioned five regions. Intracranial volume was the sum of the CSF, gray matter, white matter and WMH.

Statistical analysis

In order to examine the differences in the baseline characteristics of patients with and without CMBs, t-test was performed for normally distributed continuous variables (age, total WMH volume and neurodegenerative markers), Mann-Whitney U Test for skewedly distributed continuous variable (total ePVs), and chi-square test for binary variables (gender, hypertension, hyperlipidemia, diabetes and presence of lacunes). For regression analysis logarithmical transformation of WMH volumes was performed to ensure normal distribution. Other neurodegenerative markers: global cortical thickness, region specific cortical thickness, total white matter volume, region specific white matter volumes, subcortical structures volumes and hippocampus subfields volumes were standardized by dividing each variable by its standard deviation [39, 55]. CMBs were treated as both count and categorical data. For categorical data, we classify CMBs by location: presence of strictly lobar vs absence, presence of strictly deep vs absence and presence of mixed-location vs absence and by numbers: strictly lobar and strictly deep CMBs as 1 vs absence, 2-4 vs absence and >4 vs absence and mixed-location CMBs as 2-4 vs absence, and >4 vs absence. Total numbers of CMBs in each location: strictly lobar, strictly deep and mixed-location were treated as count data. Linear regression models were performed to determine the association between CMBs with cortical thickness (global and region specific thickness), white matter volumes (total and region specific volumes), subcortical structure volumes (accumbens, amygdala, lentiform nucleus, thalamus and brainstem) as well as hippocampus subfield volumes (CA1, CA3, CA4, parasubiculum, presubiculum, subiculum, GCMLDG, HATA, fimbria, molecular layer, hippocampal fissure, and hippocampal tail). In our stratified analysis, we divided our patients into three groups i.e. patients with NCI, CIND and dementia (AD, VaD and mixed dementia). In the analysis we treated CMBs as determinants and cortical thickness, subcortical structures volume and white matter volume as outcomes. Regression models were first adjusted for age, gender, intracranial volume, hypertension, hyperlipidemia, and diabetes. Furthermore, in a subgroup of 328 patients regression models were further adjusted for presence of lacune, total ePVs and total WMH volume (ePVS grading were only available for 328 patients). Mean difference with 95% confidence intervals (CIs) from the regression models were reported.

Results were considered significant at p<0.05. In view of multiple testing performed between CMBs and neurodegenerative markers, we used Bonferroni correction to obtain revised statistical significance level of 0.05/6 ~0.0083 for cortical thickness and 0.05/5~0.010 for subcortical structural volumes. When analyzing the association between CMBs and region specific white matter volume, we used the revised statistical significant level set at 0.05/4 ~ 0.013. Similarly, when analyzing association between CMBs and hippocampus subfield volumes statistical significance level was set at 0.05/12~0.0041. All data were analyzed using SPSS software package (version 25).