Cerebral white matter lesions – associations with Aβ isoforms and amyloid PET

Small vessel disease (SVD) and amyloid deposition may promote each other, with a potential association between SVD and altered production or clearance of β-amyloid (Aβ) affecting its cleavage products. We investigated the relationship between SVD, multiple isoforms of Aβ in cerebrospinal fluid (CSF) and cortical Aβ in 831 subjects with cognitive performance ranging from normal to Alzheimer’s disease (AD) (the Swedish BioFINDER study). SVD was estimated as white matter lesions (WML) and lacunes. 18F-flutemetamol PET was performed in 321 subjects. Lower CSF levels of Aβ38 and Aβ40 were consistently associated with increased WML in all subgroups, while lower levels of CSF Aβ42 were associated with WML mainly in AD. CSF Aβ38 and Aβ40 were associated with regional WML in all regions, while CSF Aβ42 was associated with temporal WML only. A composite measure of 18F-flutemetamol uptake was not associated with WML, and regional 18F-flutemetamol uptake only with temporal WML. Lacunes were not associated with Aβ isoforms nor 18F-flutemetamol uptake. Our results suggest that WML may be associated with alterations in the production or clearance of Aβ species, particularly of Aβ38 and Aβ40. However, in AD cases, Aβ42 pathology might be associated with WML, especially in the temporal lobe.


standardized protocol. 23
The patients with MCS (180 males and 179 females; mean ± SD age: 70.7 ± 5.7) were 24 enrolled consecutively at three memory outpatient clinics in Sweden and were thoroughly 25 assessed by physicians with special interest in dementia disorders. The inclusion criteria were: 26 3 The third sample, comprising 89 cases (34 female, 65 male; mean ± SD age: 65.4 years ± 11.2 1 years) with Parkinson´s disease (PD), henceforth referred to as the "PD cohort", is also part of 2 the prospective and longitudinal Swedish BioFINDER (Biomarkers For Identifying 3 Neurodegenerative Disorders Early and Reliably) study (www.biofinder.se). Participants were 4 recruited at the Neurology and Memory Clinics. At the initial visit, all patients were assessed 5 by physicians experienced in movement disorders and underwent thorough physical, 6 psychiatric and neurological examination. PD diagnosis was set according to the NINDS 7 Diagnostic Criteria. 10 Demographic characteristics of the cohorts are presented in Table 1. In the CHE, SCD and MCI subjects, MR imaging was performed at a 3 T Siemens Trio 11 system and in the PD subjects at a 3 T Siemens Skyra system, equipped with standard head 12 coils with 12 and 20 channels, respectively. Axial T2 FLAIR (27 and 23 slices, slice thickness 13 5.2 mm and 5 mm, respectively), coronal MPRAGE (180 slices, slice thickness 1.2 mm and 1 14 mm, image resolution 1 × 1 × 1.2 mm 3 and 1 × 1 × 1 mm 3 , respectively) were acquired. In the 15 AD subjects, imaging data included axial CT images (slice thickness 3-5 mm) in 96 cases and 16 axial FLAIR images (slice thickness 5-6 mm at 1.5 T systems) in 14 cases. implemented in SPM8 (http://www.applied-statistics.de/lst.html). The LST segments WML in 21 native space from a combination of high resolution T1-weighted and FLAIR images using a 22 lesion growth algorithm. 11 The T1 image is segmented into three main tissue classes 23 (CSF, GM and WM). This information is then combined with the FLAIR intensities in 24 order to calculate lesion belief maps. By thresholding these maps by a user determined 25 threshold (kappa), an initial binary lesion map is obtained which is subsequently 1 grown along voxels that appear hyperintense in the FLAIR image. The result is a 2 lesion probability map. The disadvantage of this algorithm, the choice of the initial 3 threshold kappa, is overcome by a routine using manual reference segmentations for a 4 few images. Here, manual delineation of WML on FLAIR images, coregistered to the native 5 high resolution T1 was performed in twelve individuals from this study, four controls, four 6 MCI patients and four PD patients comprising. The manually segmented volume from these 7 eight individuals ranged from 0.9 to 106.3 mL; the resulting optimal κ of 0.4 was used in the 8 subsequent automated segmentation for all participants. The final result of the LST 9 segmentation is a total lesion volume [mL], henceforth named 'WML volume', for each 10

individual. 11
For MR data, visual rating of WML on FLAIR images was performed firstly according to the 12 3 point scale from Fazekas and colleagues. 12  infratentorial and the basal ganglia, is rated separately and in addition a total score is 20 calculated. WML are rated as absent=0, focal=1, beginning confluence=1, diffuse 21 involvement =3. Lesions were included if the diameter was larger than 2 mm, except for the 22 striatum, globus pallidus, and thalamus where the cut off diameter for lesion detection was 5 23 mm in order to separate WML from diffuse changes surrounding perivascular spaces that are 24 frequently occurring in these areas. CT images in the AD subjects were assessed for WML 25 according to the Fazekas scale. The Fazekas and ARWMC scores based on MR images in the 1 CHE, SCD and MCI subjects and on CT images in the AD subjects, correlated similarly 2 (R 2 =0.883 and R 2 =0.871 respectively, Pearson correlation). For statistical analysis, scores 3 from the left and right hemispheres were summarized. 4 The presence of lacunes was assessed according to Wardlaw et al. 14 using the FLAIR and 5 MPRAGE sequences. Thus a lacune was defined as a round or ovoid, subcortical, fluid-filled 6 cavity (signal similar to CSF) of 3 -15 mm in diameter, on CT and often surrounded by a 7 hyperintense rim on FLAIR, consistent with a previous acute small subcortical infarct or 8 haemorrhage in the territory of one perforating arteriole. The total number of lacunes was 9 recorded and this variable was dichotomized as lacunes present or absent. 10 Hippocampal volume and a scaling factor to correct for head size differences were determined 11 in the CHE, SCD and MCI groups using Adaboost 15 ; the mean of the left and right 12 hippocampus multiplied by the scaling factor, was used in the statistical analysis. In the PD 13 cohort, hippocampal and intracranial volume (ICV) were determined using VolBrain 14 (http://volbrain.upv.es); here, the mean hippocampal volume was multiplied by the ICV to 15 account for head size differences. In the AD group, the medial temporal lobe atrophy (MTA) 16 score was determined and the mean score of the left and right MTA was used. 16 17 18 18 F-Flutemetamol PET imaging and analysis 19 In 122 CHE, 101 SCD and 98 MCI subjects, the cerebral Aβ burden was measured using 20 images were spatially normalized to Montreal Neurologic Institute (MNI) standard space 26 using a PET-only adaptive template registration method. 13 A volume of interest (VOI) 1 template defined in MNI space was applied bilaterally in frontal, partietal, occipital and 2 temporal VOIs, as well as in a composite VOI. The standardized uptake value ratio (SUVR) 3 was defined as the tracer uptake in a VOI, normalized for the mean uptake in the cerebellar 4 cortex, since this is free of fibrillar plaques. 5 6

Statistics 7
Logtransformation was performed of the WML volume, CSF levels of Aβ38, Aβ40 and Aβ42 8 as well as for the composite and regional SUVR for 18 F-flutemetamol. The distribution before 9 and after logtransformation was checked by (i) visual inspection of the data using box-plots and 10 histograms, (ii) comparing the mean and median for the individual variables, and (iii) comparing 11 the distance between median and upper/lower quartile (Supplemental Table 1). After 12 logtransformation, the composite and regional SUVR were less normally distributed.

15
Correlations between the independent and dependent variables, respectively 16 CSF levels of Aβ38 and Aβ40 did correlated slightly with CSF Aβ42 (r = 0.273 and 0.305, 17 respectively, partial correlation, corrected for age, gender and hippocampal volume, data from 18 the CHE, SCD and MCI groups). No significant correlation with the composite SUVR was 19 found (r = 0.16 and 0.050, respectively). Only CSF Aβ42 showed a strong, inverse, 20 correlation with the composite SUVR (r = -0.659). Thus, no variance attributable to AB42 of 21 significant magnitude (r> 0.7) is shared with the other amyloid variables simultaneously 22 included in the model. 23 The WML volume and total Fazekas score correlated strongly, as expected since these 24 variables are different measurements of the abundance of white matter lesions (r = 0.752).

7
The correlation of these two with lacunes was only weak (r = 0.262 and 0.280, respectively) 1 and thus no variance attributable to lacunes was shared by measures of white matter lesions. Additional analysis was performed for the composite SUVR, now using binary regression with 5 the 1.42 as cut-off for normalisation as previously described by Palmqvist et al [14]; results were 6 similar for the logtransformed and dichotomized composite SUVR (Supplemental Table 1 -0.081 -0.20 Values represent standardized beta (linear regression). Values are corrected for age, gender and hippocampal volume; the latter was determined using Adaboost, the mean value of the left and right hippocampus was normalized using the scaling factor provided by this software to account for head size differences. Data on WML volume were not available for the AD group, where imaging mainly was performed using CT. * = p<0.05, ** = p<0.01, *** = p>0.005