Aging and central vision loss: Relationship between the cortical macro-structure and micro-structure

Aging and central vision loss are associated with cortical atrophies, but little is known about the relationship between cortical thinning and the underlying cellular structure. We compared the macro- and micro-structure of the cortical gray and superficial white matter of 38 patients with juvenile (JMD) or age-related (AMD) macular degeneration and 38 healthy humans (19-85 years) by multimodal MRI including diffusion-tensor imaging (DTI). A factor analysis showed that cortical thickness, tissue-dependent measures, and DTI-based measures were sensitive to distinct components of brain structure. Age-related cortical thinning and increased diffusion were observed across most of the cortex, but increased T1-weighted intensities (frontal), reduced T2-weighted intensities (occipital), and reduced anisotropy (medial) were limited to confined cortical regions. Vision loss was associated with cortical thinning and enhanced diffusion in the gray matter (less in the white matter) of the occipital central visual field representation. Moreover, AMD (but not JMD) patients showed enhanced diffusion in lateral occipito-temporal cortex and cortical thinning in the posterior cingulum. These findings demonstrate that changes in brain structure are best quantified by multimodal imaging. They further suggest that age-related brain atrophies (cortical thinning) reflect diverse micro-structural etiologies. Moreover, juvenile and age-related macular degeneration are associated with distinct patterns of micro-structural alterations.


Structural brain measures 217
The brain macro-and micro-structure was quantified by several tissue-dependent and 218 diffusion-based measures. Diffusion-based measures (Fig. 1b) consisted of the three 219 eigenvalues (λ 1 , λ 2 , λ 3 ) of the tensor model (Basser et al., 1994) and several derived measures. 220 This included measures of diffusion-strength: mean (MD), axial (longitudinal) (AD), and 221 radial (perpendicular) diffusion (RD). Note that AD corresponds to the major eigenvalue (λ 1 ). 222 In addition, measures of diffusion-shape such as fractional anisotropy (FA), the mode of 223 diffusion (MO), and the inter-voxel coherence (IVC) were calculated. FA is a normalized 224 measure of anisotropy ranging from 0 (isotropic) to 1 (anisotropic diffusion) (Basser and 225 Pierpaoli, 1996). MO describes the mode of anisotropic diffusion along a normalized 226 continuum from -1 (planar) to +1 (tubular) (Ennis and Kindlmann, 2006). IVC reflects the 227 average difference angle of the primary eigenvector between a voxel and its adjacent voxels 228 (weighted by FA and distance) (Pfefferbaum et al., 2000). The mean angle was rescaled to a 229 maximum of 1 and subtracted from 1 resulting in values ranging from 0 (incoherence: 230 different orientation across voxels) to 1 (coherence: same orientation across voxels). 231 aging and central vision loss 9 Tissue-dependent measures included T1-weighted (T1w) and T2-weighted (T2w)  232 intensities. T1w were based on the MPRAGE images. T2w were based on the mean b-zero 233 images of the DTI runs, which is essentially a T2-weighted signal. Both T1w and T2w values  234 were normalized (rescaled) by the mean intensity of the white matter. 235 Finally, several measures of the brain macro-structure based on the Freesurfer 236 reconstruction were examined. This included thickness estimates across the cortical surface 237 but also global measures (brain volume, surface area, and gyrification indices). Two brains 238 showed extreme values in gyrification indices (> 3 standard deviations from the mean) and 239 were excluded from this analysis. 240 241 2.6 Whole-brain analysis 242 All measures (except for global measures) were probed along nine projection levels 243 relative to the white-gray matter boundary (WGB) of each individual reconstructed brain. 244 This was done in order to evaluate differences (or similarities) between the gray and the 245 superficial white matter and layer-specific properties within each structure. Projection levels 246 ranged from the deep white matter (DWM) 2 mm below the white matter surface to the white-247 gray matter boundary in steps of 0.5 mm and from the white-gray matter boundary to the pial-248 gray matter boundary (PGB) in steps of 25% of the cortical thickness. In order to assess the 249 consequences of partial volume effects (e.g., values at gray matter voxels being contaminated 250 by adjacent cerebro-spinal fluid), two projection levels (25% and 50%) beyond the pial-gray 251 matter boundary were also analyzed. 252 All individual surface maps were spherically registered and mapped to the standard 253 brain of Freesurfer. No volumetric smoothing was applied. Surface smoothing was performed 254 by iterative nearest-neighbor averaging resulting in the equivalent of a 2D Gaussian kernel 255 with a full-width half-maximum of 15 mm (Hagler et al., 2006). Note that this smoothing 256 width is substantially less than in previous studies (22 mm diameter) examining age-related 257 aging and central vision loss 10 cortical thinning (Salat et al., 2004) as we wanted to avoid substantial blurring of the spatially 258 limited effects expected by sensory deprivation. 259 For each measure and vertex, age-related and CVL-related variations across brains were 260 assessed by a general linear model. The design matrix included regressors for age and CVL 261 group, nuisance regressors for sex and MR sequence, and a constant. The age regressor coded 262 the age for the whole sample relative to a reference age of 20 years. CVL-related variations 263 were modeled by two regressors: JMD (1 for JMD patients, 0 otherwise) and AMD (1 for 264 AMD patients, 0 otherwise). Note that these regressors model deviations from NV cases. Sex 265 was coded by 1 for men (0 otherwise) and MRI sequence was coded by 1 for 30 orientations 266 (0 otherwise). Subsequently, contrasts were defined: These included a Base effect (based on 267 the constant regressor and corresponding to a NV brain at 20 years), an Age effect (slope of 268 linear variations accounted by the age regressor), and a CVL group effect (combined 269 contribution of the JMD and AMD regressor). Moreover, for a more detailed analysis of 270 differences between JMD and AMD patients, separate contrasts were defined for the JMD and 271 the AMD group effect. 272 The whole-brain analysis included three main steps: The first step of the analysis aimed 273 to quantify global aspects of brain structure. Therefore, the mean values of the whole cortical 274 surface (sparing the medial wall adjacent to subcortical structures) were compared for each of 275 the three main contrasts. Second, a factor analysis based on principal components (PCA) 276 (Jolliffe, 2002) was conducted to identify relevant and independent measures of brain 277 structure. Third, cortical surface distributions of relevant measures were compared. 278 The PCA aimed to investigate redundancy and independence, respectively, across 279 measures. Many previous studies restricted their analysis to one or a few primary measures 280 (e.g., FA) or brain regions (e.g., WM). However, this approach may disregard unknown but 281 potentially relevant markers of brain integrity. Hence, we adopted instead the data-driven 282 approach of a factor analysis that clarifies which measures and projection levels are sensitive 283 to the same or different cellular structures. Three analyses with the three main contrasts (Base 284 level, Age effect, CVL effect) were conducted. The analysis was performed with custom-285 made Matlab (Mathworks) scripts. The data matrix containing the MRI measures and 286 projection levels as variables and the vertices across the surface as observations was 287 normalized (z-scaled). The number of principal components was identified by the Kaiser-288 Guttman criterion (eigenvalues greater than one). Subsequently, a varimax rotation of the 289 reduced factor space was performed. Individual variables were assigned to one of the factors 290 based on a minimum factor loading (correlation between variable and component) of .4 291 (Stevens, 2002) and the Fürntratt criterion (ratio of squared factor loading to communality ≥ 292 .5) (Fürntratt, 1969). 293 Finally, a whole surface analysis was conducted for combinations of the variables (e.g., 294 mean across projections) that were grouped together by the PCA. Surface-based significance 295 maps were thresholded to a p-value of .01. Cluster-wise correction for multiple comparisons 296 was performed by comparing voxel-wise significance maps (voxel-wise threshold of p ≤ .01) 297 with cluster-wise significance maps (cluster-wise threshold of p ≤ .01) that were based on the 298 Monte Carlo simulated distribution of cluster size (CSD) as implemented in Freesurfer. Only 299 clusters exceeding the CSD-simulated cluster size were preserved. As structural changes 300 related to sensory deprivation were primarily expected in the occipital cortex, a less rigid 301 cluster-wise correction (minimum cluster size of 50 mm 2 ) was adopted for the group effect 302 (CVL vs. NV). 303 304

ROI analysis 305
For a more detailed quantification of the brain structure in CVL patients, a region-of-306 interest analysis of the visual cortex was performed (Fig. 1d). This was motivated by previous 307 studies showing changes in brain structure (Boucard et  Results 318

Whole-brain global measures 319
First, we analyzed the global (across the whole cortical surface) macro-and micro-320 structure. MRI measures were probed along nine projection levels relative to the white-gray 321 matter boundary ranging from the deep white matter to the pial-gray matter boundary ( Tissue-dependent measures of brain structure (T1w, T2w) varied across projection 337 levels from the deep white matter (2 mm below the white-gray matter boundary) to the pial-338 gray matter boundary (see Fig. 2). As expected, Base levels of T1w values were relatively 339 high in the white matter and low in the gray matter. T1w values significantly (p ≤ .01) 340 decreased along the projection axis (from white to gray matter). However, at the deepest white 341 matter projection a significant increase relative to superficial white matter was observed. An 342 almost opposite pattern across projection levels was observed for T2w values, which were 343 relatively low in white matter and which increased significantly along the projection axis 344 (from white to gray matter) except for the deepest white matter projection (

Whole-brain factor analysis 377
The global whole-brain analysis showed age-related changes for all measures of brain 378 structure. In order to test whether these measures are correlated (e.g., reflecting related 379 structural properties) or whether they are independent (e.g., reflecting different structural 380 properties), a principal component factor analysis was performed. This analysis tested all 381 measures for correlations across vertices of the brain surface and followed the rationale that it 382 should only reveal one or very few components, if diffusion and tissue-dependent measures 383 are highly correlated (and hence redundant), but a relatively large number of components, if 384 the various measures are sensitive to different aspects of the brain structure. Separate PCAs 385 were performed for the Base level, the Age effect, and the Group effect (CVL vs. NV).
According to the eigenvalue criterion (see methods) a minimum of 9 components was needed 387 to explain 95% (Base level), 95% (Age effect), and 92% (Group effect) of the variance. 388

Whole-brain surface maps 406
The results of the factor analysis suggest that diffusion-based and tissue-dependent 407 measures are sensitive to distinct components of age-related changes in brain structure. 408 However, it does not reveal the scope of age-related brain changes. This is shown in Figure 4

ROIs of retinotopic areas 464
For a more detailed quantification of the brain structure in CVL patients, a ROI analysis 465 of the visual cortex was performed (Fig. 1d). This was motivated by previous findings that 466 reported changes in brain structure (Boucard et  all ROIs (Fig. 6a). However, the cortex was thicker in V2 than V1. T2w values were lower in 477 central representations of the visual field (e1) than in more peripheral representations (e2, e3, 478 e4). Values of diffusion-strength (MD, AD, RD) tended to be higher for central (e1, e2) than 479 for peripheral (e3, e4) representations of the visual field and more pronounced diffusion was 480 found in V1 than in V2 and for gray as compared to white matter. Measures of diffusion-shape 481 (MO, IVC) showed fairly uniform values across visual cortex ROIs with the exception of FA, 482 which was relatively low in both white and gray matter, but tended to be higher in central 483 compared to peripheral representations of the V1 and V2 gray matter. 484 Almost all measures showed an Age effect (Fig. 6b) Significant differences between CVL patients and the NV control group (Fig. 6c)

JMD versus AMD 502
Our CVL group contained both JMD and AMD patients. The results reported above 503 were based on both sub-groups combined based on the rationale that JMD and AMD patients 504 suffer from similar sensory deprivations (leading to similar secondary degeneration). 505 Nevertheless, it is possible that they differ in the type of structural changes. Therefore, we 506 performed additional analyses separate for the JMD and AMD sub-groups, respectively (Fig.  507 7, see also Supp. Fig. S5 and Supp. Tab. S1, S2, and S3). Although the main structural 508 changes observed in both sub-groups were cortical thinning and increased diffusion of the 509 occipital pole gray matter, the two sub-groups differed in magnitude and scope. In the JMD 510 group the major structural change was cortical thinning and only a moderate increase in gray 511 matter diffusion limited to the occipital pole and central visual field representations (e1). By contrast, the AMD sub-group showed more wide-spread structural changes: Cortical thinning 513 was not dominant at the occipital pole or central visual ROIs, but instead was primarily 514 observed in the posterior cingulate cortex. However, gray matter diffusion was pronounced in 515 the occipital pole and central visual ROIs (e1, e2). Moreover, increased gray matter diffusion 516 was observed beyond early retinotopic cortex extending to lateral occipital and temporal 517 cortex and the posterior cingulate cortex. The latter effect was also pronounced in direct 518 comparisons of the AMD group with the JMD group (see Supp. Discussion 522 We examined the brain macro-and micro-structure of patients with central vision loss 523 and healthy humans in the age range from 19 to 85 years by multimodal MRI including DTI. 524 We found that both, aging and central vision loss, were associated with structural changes in 525 the cortex. 526

Degeneration by age 527
Our results showed that even 'normal' aging is associated with alterations in the cortical 528 brain macro-and micro-structure. We observed age-related reductions in cortical volume and Our results further showed pronounced age-related differences in diffusion-based 541 measures. Aging was primarily associated with increased diffusion in both the gray and the 542 superficial white matter. Reduced (less anisotropic) FA, reduced (more planar) MO, and 543 increased (more coherent) IVC was also observed. These findings seem to be inconsistent 544 with previous research that did not find aging-related changes of diffusion properties in the 545 gray matter (Helenius et al., 2002). Note, however, that a coarser analysis approach (ROI-546 based, less diffusion directions) was adopted in this previous work. Our results suggest that 547 the micro-structure in the gray and superficial white matter as revealed by DTI is not only 548 affected by brain maturation (< 20 years) (Wu et al., 2014), but also by aging (> 20 years). The observed differences in MRI measures suggest age-related degeneration. Although 553 cortical thinning considered in isolation might be interpreted in alternative ways (e.g., 554 enhanced neural density rather than degeneration), the full pattern suggests the involvement of 555 degenerative mechanisms. For instance, cortical thinning was accompanied by increased 556 diffusion (e.g., MD levels). Increased diffusion likely reflects cellular lysis (e.g., neural loss) 557 and/or associated gliosis (Govindan et al., 2013;Sotak, 2002;Sykova et al., 1998;Welch et 558 al., 1995). For instance, age-related astrogliosis in rats has been shown to reduce tortuosity 559 (corresponding to increased MD) (Sykova et al., 1998). However, glia proliferation may only 560 account for the Age effects in frontal brain regions that also showed elevated T1w values, 561 because a high macromolecular mass fraction (as to be expected in gliosis) would be 562 associated with fast longitudinal relaxation (corresponding to high T1w values) (Rooney et al., 2007). In other brain regions no differences in T1w were observed. Here, increased 564 diffusion most likely reflects a decline in neural density and/or increase in extracellular space. 565 For instance, increased MD was observed during long-term cell damage (e.g. necrosis) 566 following stroke (Sotak, 2002;Welch et al., 1995) and in brain areas affected by neuronal loss 567 and gliosis such as the seizure-onset regions of epileptic children (Govindan et al., 2013). The 568 occipital pole (including central V1 and V2) was hardly affected by age-related cortical 569 thinning or alterations in diffusion, but instead by decreased T2w values. These T2w 570 reductions likely reflect a regional-specific increase in nonheme iron deposits (Korogi et al., 571 1997  are unlikely accounts in those regions affected by FA declines, because no reduction in IVC 581 was observed (Pfefferbaum et al., 2000). Hence, the increased MD and the reduced FA likely 582 reflects de-myelination. We also observed changes in inter-voxel coherence. Originally, this 583 measure was proposed to detect axonal decline in the callosum, which would result in reduced 584 coherence (Pfefferbaum et al., 2000). However, this original work did not observe age-related 585 reductions in inter-voxel coherence, but instead a tendency for increased coherence. Similarly, 586 our results showed that inter-voxel coherence increased by age in some parts of the temporal 587 cortex. Possibly, this increase in coherence reflects reduced axonal complexity (e.g., decline 588 in local association fibers) of the superficial white matter. 589 aging and central vision loss 23 Previous studies proposed that age-related degeneration is observed uniformly across 590 the brain (Penke et al., 2010) or in a gradient-like (e.g., anterior to posterior or superior to 591 inferior) manner (Pfefferbaum et al., 2000;Sullivan et al., 2010;Zahr et al., 2009). Our 592 multimodal analysis approach showed that age-related structural changes across the cortex 593 varied by measure and were grouped around several cortical regions. Hence, our findings 594 support a regional-specific view (Burzynska et al., 2010) rather than a global view (at least at 595 the level of the gray and superficial white matter). Our factor analysis (Fig. 3) showed that 596 tissue-dependent and diffusion-based measures were quite independent measures of brain 597 structure suggesting that they are sensitive to distinct age-related mechanisms of 598 degeneration. The surface distribution of cortical thinning and alterations in tissue-dependent 599 measures (T1w or T2w) were hardly related (loading on distinct factors) with age-related 600 changes in DTI metrics. Although measures of diffusion-strength (MD, AD; RD) were highly 601 inter-correlated within gray and white matter, respectively, they were not or only poorly 602 related with measures of diffusion-shape (see factor loadings in Fig. 3). Hence, gradient 603 notions of age-related degeneration are likely constrained to certain brain parts (e.g., deep 604 white matter) and/or are based on unimodal MRI analyses approaches. 605 showed less variability and more reliable effects than other measures of brain structure. 626 Hence, CVL-related micro-structural changes in the superficial white matter seem to be either 627 absent or subtler than those observed within the gray matter or those related to aging. It seems 628 that axonal connections (e.g., thalamo-cortical fibers) in the superficial white matter of V1 629 and V2 were not or only mildly affected by long-lasting sensory deprivation. This is 630 consistent with functional MRI reports that found recovery of visual cortex activity following 631 However, there were also marked differences. Structural changes in JMD patients primarily 643 involved cortical thinning with only a modest increase in GM diffusion suggesting brain 644 atrophy without substantial decline in cellular structures. Moreover, the structural changes 645 were limited to the lesion projection zone of early retinotopic visual areas. By contrast, 646 structural changes in AMD patients were primarily observed in the micro-structure (increased 647 gray matter diffusion). Moreover, structural degeneration was observed beyond early 648 retinotopic visual areas: Enhanced gray matter diffusion was also found in the lateral occipital 649 and temporal cortex. In addition, cortical thinning and increased diffusion was observed in the 650 posterior cingulate cortex. These findings suggest that cortical degeneration in AMD patients 651 is more generalized than in JMD patients. 652 The structural alterations of AMD patients in lateral brain regions might mediate the 653 reduced functional connectivity of the lateral occipital cortex observed in a previous study 654 (Zhuang et al., 2018). To our knowledge, we are the first to show micro-structural alterations 655 in the lateral cortex of AMD patients. However, altered cytochrome-oxidase staining patterns 656 were observed in area V5 (posterior medial temporal cortex) in a post-mortem examination of 657 an 82-year-old woman (Clarke, 1994). Moreover, reduced gray matter volume of the lateral 658 occipital cortex was observed in patients with monocular blindness (Prins et al., 2017). Note 659 that these wide-spread structural alterations of AMD patients (compared to JMD) are likely 660 not attributable to a more severe visual deprivation. Disease duration and scotoma size tended 661 to be less and visual acuity tended to be higher in AMD than in JMD patients suggesting that 662 low-level visual functions tended to be even less impaired in the AMD than in the JMD sub-663 group. Hence, the extensive micro-structural alterations in AMD patients are more likely 664 related to higher-level visual functions (e.g., reading) or strategies to compensate for the CVL. 665 Consistent with this notion, reading speed was more strongly compromised in AMD than in 666 JMD patients. It remains elusive though whether these behavioral limitations are the cause or the consequence of the altered brain structure. 668 It is possible that the AMD-specific alterations in brain structure beyond low-level 669 visual pathways reflect a link between AMD and other neurodegenerative mechanisms such 670 as mild cognitive impairment or Alzheimer's disease (Ikram et al., 2012). According to this 671 view, the pathological mechanisms responsible for the cellular degeneration in the retina of 672 AMD (but not JMD) patients are also operating in the brain. Epidemiological studies showed 673 that age-related macular pathologies are associated with enhanced risks for Alzheimer's 674 disease (Klaver et al., 1999)  to the white-gray matter boundary. Low diffusion (e.g., MD) and low anisotropy (FA) in gray 690 as compared to white matter was expected based on previous work (Helenius et al., 2002). 691 However, the variation across projection levels was not monotonic (see Fig. 2a). In particular, 692 diffusion of the first eigenvalue (AD) showed an inverted U-shaped distribution suggesting that deep white matter was characterized by other cellular properties than superficial white 694 matter. 695 The surface-based analysis also mitigated methodological problems of volumetric 696 whole-brain approaches such as inter-subject normalization (Bach et al., 2014). However, it 697 may be argued that partial-volume effects and spatial smoothing across projection levels (.5 698 mm in WM) may have compromised our findings as the voxel size (2.5 mm isotropic for 699 diffusion-based measures) was larger than the distance between projection levels. 700 Nevertheless, volume averaging likely had only a negligible effect on our main results: As 701 demonstrated by the factor analysis ( Fig. 3), white matter projections were sharply separated 702 from gray matter projections. In order to test whether volume averaging with the cerebro-703 spinal fluid contaminated our findings, we also examined MRI measures outside the gray 704 matter. This analysis showed that MRI properties differed for projections beyond the pial-gray 705 matter boundary from those observed within the gray matter. 706 Age effects in our study were only examined by a linear regressor. Previous studies 707 suggested a non-linear (e.g., an inverse U-shaped) relationship (Chang et al., 2015;Yeatman 708 et al., 2014). For instance, MRI correlates increase during childhood and adolescence, peak at 709 around 20 to 40 years, and decline in elderly. However, these previous studies examined a 710 broader age range (e.g., 7 -87 years) and a linear model is likely an acceptable approximation 711 for the age range of the current study (19 to 84 years). We examined both Age effects and 712 CVL effects in a combined analysis. This allowed us to control for confounding effects of Age 713 on CVL. However, it might be argued that our Age effects were contaminated by CVL effects. 714 Therefore, we performed control analyses including only the normally sighted (NV) group. 715 These analyses (Sup. Fig. S2 and S3) showed very similar result patterns as our main analysis. Variability within the same brain tissue (e.g., same projection level of white or gray matter) was quite low. Variability across brains primarily reflected Age and Group effects. Diffusion-720 based measures were quite specific. Little correlation was found with other measures of brain 721 structure. Cortical thickness showed less specificity than any of the diffusion-based measures. 722 However, a single measure of diffusion (e.g., mean diffusion) is not sufficient to fully explain 723 micro-structural changes due to aging or central vision loss. Although changes in MD were 724 most prominent, some mechanisms were best detected by measures of diffusion-shape. 725 726 5 Conclusions 727 Our surface-based DTI analysis revealed significant age-related changes of the brain 728 micro-structure in the gray and superficial white matter. Aging resulted in cortical thinning, 729 enhanced T1w, reduced T2w, enhanced diffusion, reduced anisotropy, and less tubular 730 diffusion. Structural changes were observed across most of the cortex. However, DTI-based 731 measures, tissue-dependent measures, and cortical thickness were sensitive to relatively 732 distinct degenerative mechanisms: Whereas frontal regions showed atrophy that is likely 733 accompanied by gliosis, structural changes in occipital brain regions likely reflect reduced 734 neural density, demyelination, and enhanced iron concentrations. Central vision loss resulted 735 in enhanced diffusion in the occipital gray matter (central visual field representation) but 736 preserved integrity of the superficial white matter suggesting that visual deprivation is 737 associated with reduced neural density and gliosis but preserved axonal structures in the 738 lesion projection zone. The micro-structural differences of JMD patients are consistent with 739 secondary degeneration due to sensory deprivation, whereas the micro-structural differences 740 of AMD patients likely result from additional degenerative mechanisms. 741 742 aging and central vision loss 29