Introduction

Neurodegenerative diseases, such as Alzheimer’s disease (AD) and other forms of dementia, pose significant challenges to the aging global population. These conditions are becoming increasingly prevalent as life expectancy rises, placing a substantial burden on patients, caregivers, and healthcare systems worldwide. Early diagnosis and intervention are crucial for managing these diseases and mitigating their impact1.

Among the various early diagnostic biomarkers being investigated for dementia, the retina has garnered significant interest. Studies have shown that retinal thickness is reduced in patients with AD, suggesting that the retina may reflect pathological changes occurring in the brain2,3,4,5. Retinal abnormalities in AD patients have been linked to the accumulation of amyloid-beta, a hallmark of AD pathology6,7,8. Notably, the thinning of the retinal nerve fiber layer (RNFL), as measured by optical coherence tomography (OCT)3, correlates with both the severity of cognitive impairment and the progression of AD9,10, indicating that retinal imaging could serve as a non-invasive biomarker for early detection of neurodegenerative changes2,5. More recently, studies in non-demented older adults have also reported associations between reduced RNFL thickness and cognitive impairment or structural brain changes9,11. Specifically, reduced RNFL thickness has been associated with smaller volumes in the cingulate cortex, hippocampus, and entorhinal cortex—brain regions involved in early neurodegeneration—and with progressive decline in episodic memory12. These findings support the relevance of RNFL thickness as a potential early marker of preclinical neurodegeneration even in non-demented older adults.

While AD pathology provides an important clinical framework for understanding retinal changes, the present study focuses on older adults without dementia. In our previous research10, we identified a significant association between macular RNFL thickness and cognitive decline in a community-based elderly cohort. This finding suggests the possibility that reductions in brain volume or cortical thickness, which are recognized as markers of neurodegeneration13, may be the neural correlates or underlying mechanisms linking macular RNFL thinning to cognitive decline. However, previous studies investigating the association between RNFL and brain volume have reported inconsistent results11,14,15,16,17,18,19,20,21,22,23,24,25,26,27, with most research focusing on peripapillary RNFL11,12,14,15,17,18,19,20,21,22,23,24,25. In contrast, the relationship between macular RNFL thickness and structural brain measures remains less thoroughly explored.

Therefore, in this study, we aimed to investigate the association between macular RNFL thickness and brain volume/cortical thickness in older adults without dementia, using baseline MRI data from our community-based elderly cohort in which a link between macular RNFL thickness and cognitive decline had previously been established9,10.

Results

The 166 participants included in our study were younger on average (75.2 ± 5.3 years old), had a higher proportion of females (61.4%), and attained higher levels of education (10.4 ± 5.2 years) compared to the 263 individuals excluded due to the absence of MRI (Table 1). While there was no statistically significant difference in total macular RNFL, the inner RNFL thickness was greater (94.8 ± 13.0 vs. 91.2 ± 13.5, P = 0.006). Cognitive functions, assessed by CERAD-TS and MMSE, were better among the included group, which had no dementia cases and a lower rate of mild cognitive impairment compared to those excluded (Table 1).

Table 1 Demographics and clinical characteristics of the participants.
Full size table

In the uncorrected analyses, increased gray matter volumes correlating with outer RNFL thickness were present in the right inferior parietal (t = 3.81, cluster size = 135 voxels), the left superior frontal (t = 3.71, cluster size = 129 voxels), and inferior temporal (t = 3.95, cluster size = 183 voxels) cortices. The total RNFL thickness was associated with the volume of the right inferior parietal cortex (t = 3.62, cluster size = 101 voxels). However, these findings were statistically insignificant after correction for multiple comparisons using Family-Wise Error rate (FWE) < 0.05. The detailed results of the VBM analysis are presented in Table 2; Fig. 1. Additionally, VBM analysis of peripapillary RNFL revealed a significant positive association with gray matter volume in the right precentral cortex (t = 3.79, cluster size = 280 voxels; Supplementary Table S1).

Table 2 Location and size of brain regions showing significant associations with retinal nerve fiber layer (RNFL) thickness.
Full size table
Fig. 1
figure 1

Voxel-based morphometry analysis showing associations between RNFL thickness and regional brain volumes. Results are displayed at an uncorrected threshold of p < 0.001, with a cluster size threshold of 100 voxels applied to reduce the risk of false positives. Panel (a) shows results for outer RNFL thickness, with significant clusters located in the left superior frontal, right inferior parietal, and left inferior temporal cortices (from top to bottom). Panel (b) shows a significant cluster for total RNFL thickness in the right inferior parietal cortex. The color bar represents t-values. For each significant cluster, a scatter plot is presented to the right, showing the association between RNFL thickness (μm) and cluster mean gray matter volume, adjusted for age, sex, and total intracranial volume. Each plot displays unstandardized β coefficients, P-values, and R² values from multiple linear regression models.

Full size image

Prior to correction for multiple comparisons, cortical thickness ROIs that showed a positive correlation with total RNFL thickness included the banks of the left superior temporal sulcus (r = 0.154, P = 0.049), left fusiform (r = 0.161, P = 0.040), left lingual (r = 0.158, P = 0.043), left pars opercularis (r = 0.162, P = 0.038), left posterior cingulate (r = 0.228, P = 0.003), left superior temporal (r = 0.169, P = 0.031), both supramarginal gyri (right: r = 0.270, P < 0.001; left: r = 0.217, P = 0.005), right lateral occipital (r = 0.191, P = 0.014), right lateral orbitofrontal (r = 0.161, P = 0.039), and right superior parietal cortex (r = 0.176, P = 0.024). The left cerebral mean thickness also showed a significant positive association (r = 0.163, P = 0.037; Table 3; Fig. 2). The left caudal middle frontal cortical thickness showed a significant association only with the outer RNFL thickness (r = 0.157, P = 0.044), while the bilateral cuneus (right: r = 0.153, P = 0.049; left: r = 0.158, P = 0.044), right caudal anterior cingulate cortex (r = 0.158, P = 0.043), and left transverse temporal regions (r = 0.160, P = 0.041) showed significant associations only with inner RNFL thickness. Notably, the left posterior cingulate cortex (outer: r = 0.197, P = 0.011; inner: r = 0.215, P = 0.006; center: r = 0.248, P = 0.001; total: r = 0.228, P = 0.003) and right supramarginal cortex (outer: r = 0.244, P = 0.002; inner: r = 0.243, P = 0.002; center: r = 0.258, P = 0.001; total: r = 0.270, P < 0.001) were significantly associated with all RNFL measures, including outer, inner, center, and total (Table 3; Fig. 2). However, after applying False Discovery Rate (FDR) correction for multiple comparisons, none of these associations remained significant at the p < 0.05 level. No significant correlations were found between peripapillary RNFL and cortical thickness across any of the ROIs analyzed (Supplementary Table S2).

Table 3 Estimates of partial correlations (r) between retinal nerve fiber layer (RNFL) thickness and cortical thickness.
Full size table
Fig. 2
figure 2

Regions of significant partial correlations (p < 0.05) between RNFL thickness and cortical thickness in 3D brain image. (a) Outer RNFL, (b) Inner RNFL, and (c) Total RNFL thickness. r represents the correlation coefficient (Pearson’s partial correlation adjusted for age and sex).

Full size image

Discussion

In this exploratory study, we investigated potential associations between macular RNFL thickness and structural brain measures in non-demented elderly individuals. However, it is crucial to interpret our findings with considerable caution. None of the observed associations remained statistically significant after correction for multiple comparisons, and the partial correlation coefficients were low (r = 0.153 to 0.270), indicating weak relationships. Therefore, our results should be considered preliminary and hypothesis-generating rather than conclusive evidence of meaningful brain-retina associations.

Although not statistically significant after multiple comparisons, we found that the thicker the RNFL, especially the outer part, the greater the volume of the left superior frontal, right inferior parietal, and left inferior temporal regions. Additionally, the thickness of the macular RNFL was associated with increased cortical thickness in several brain areas, including the left posterior cingulate and both supramarginal regions. This positive directionality is biologically plausible, as the retina and brain share embryonic origins and are susceptible to similar pathological mechanisms or aging processes28. A systemic aging or neurodegenerative process, therefore, would be expected to lead to simultaneous tissue loss in both structures11. Conversely, the few negative associations between cortical thickness and RNFL thickness (Table 3) were not statistically significant both before and after correction and lack a clear biological rationale. Therefore, they are most likely attributable to chance, a known risk in exploratory analyses involving a large number of comparisons.

The brain regions showing potential associations in our uncorrected analyses - left superior frontal, right inferior parietal, and left inferior temporal cortex areas - are associated with higher cognitive functions and sensory processing. The left (dominant) superior frontal gyrus is a key component in the neural network of working memory as well as spatial processing29,30. This region is often associated with the Executive Control Network (ECN), which is involved in high-level cognitive functions such as decision-making, problem-solving, and executive functions including planning and impulse control31. The right inferior parietal cortex, beyond its fundamental roles in attention, language, and social cognition, also forms part of the Default Mode Network (DMN), becoming active during the brain’s resting state32,33,34. The left inferior temporal cortex plays a significant role in various cognitive functions, including language processing, semantic memory, and visual object recognition (“What Pathway” in visual processing)35,36. Given the lack of statistical significance after correction, the potential relevance of these specific anatomical regions remains speculative and would require replication in larger studies to be considered reliable evidence.

Cortical thickness showed weak correlations with RNFL thickness in regions involved in visual processing and perception (both cunei, left fusiform, left lingual, right lateral occipital, both supramarginal)37,38,39, language and semantic processing (left pars opercularis, left superior temporal, left transverse temporal, both supramarginal)40,41,42, spatial awareness (both supramarginal)43,44, executive function and cognitive control (left caudal middle frontal, right lateral orbitofrontal)45,46, and social cognition and emotional processing (left posterior cingulate, both supramarginal)47,48. Notably, the posterior cingulate, a central node of the DMN, is a critical area for early AD detection, as it accumulates beta-amyloid from the early stages of the disease49. The supramarginal gyrus may show changes in functional connectivity and structural atrophy as Alzheimer’s disease progresses50,51. While these uncorrected findings may suggest that early pathological changes could involve both retinal and brain structures, particularly in regions known to be affected in AD, such interpretation remains highly speculative in the absence of robust statistical support.

Both the retina and the brain originate from the same embryonic tissue, rendering them susceptible to similar pathological mechanisms28. Key pathologies of AD, such as amyloid-beta plaques and tau protein, are observed in both the brain and the retina6,7,8,52. Additionally, the RNFL consists of axons of retinal ganglion cells, and neurodegeneration in AD leads to the loss of these axons, resulting in RNFL thinning15,53. Neurodegeneration in AD also causes brain volume reduction and cortical thinning54. Microvascular pathology can also induce neurodegenerative changes in both the retina and the brain6. It remains unclear whether RNFL thinning precedes brain atrophy or if the overall neurodegenerative process occurs simultaneously in the RNFL and the brain. Thus, further research through longitudinal studies is necessary to elucidate the temporal relationship55.

Previous studies predominantly examined the association between peripapillary RNFL and brain volume11,14,15,17,18,19,20,21,22,23,24,25, while fewer studies investigated the link between macular RNFL and brain volume, often yielding inconsistent results17,21,25,26,27,56. Some discussions in the literature suggest that macular RNFL may show fewer significant correlations with brain volume due to measurement issues compared to peripapillary RNFL17,57. Alternatively, macular RNFL thinning might precede peripapillary RNFL thinning in the early phase of cortical degeneration10, which could explain the fewer significant associations found between macular RNFL and cortical volume/thickness in comparison to peripapillary RNFL. The present study adds evidence to the small body of research exploring potential associations between macular RNFL and brain volume/cortical thickness25,56 by selecting macular RNFL, which has been significantly associated with longitudinal cognitive decline in previous longitudinal studies9,10.

Among the brain areas that showed significant associations with macular RNFL thickness in prior studies25,56—such as the cingulate, lingual gyrus, cuneus, and occipital lobe—our uncorrected analyses also found correlations with cortical thickness in these regions. Our study suggests potential connections between macular RNFL and the posterior cingulate, a known early pathologic point for AD pathology49, and identifies possible associations with several cortical brain volumes related to higher cognitive functions. A previous study observed a decrease in the macular RNFL that is moderately correlated with positron emission tomography imaging evidence of amyloid aggregation58.

Several important limitations must be acknowledged in interpreting our findings. Firstly, none of the observed associations remained statistically significant after correction for multiple comparisons. While the theoretical rationale for brain-retina associations remains compelling given their shared embryological origin and pathological susceptibilities, the weak associations before correction and their loss of significance afterward suggest that any such relationships may be more complex or subtle than initially hypothesized. Secondly, given this lack of significance after multiple comparisons, we cannot completely exclude the possibility of Type I errors. In addition, the extensive multiple testing burden (whole-brain VBM analysis and 280 partial correlations) and the modest sample size of 166 participants likely reduced statistical power, increasing the risk of Type II errors after multiple comparison corrections and limiting the ability to detect true associations. This modest sample size may also limit the generalizability of our findings. Thirdly, the cross-sectional design precludes any inference about causality or temporal relationships. Fourthly, despite controlling for age and sex, residual confounding cannot be ruled out. Finally, the exclusion of individuals without MRI data may introduce selection bias.

In conclusion, this exploratory study identified potential associations between OCT-measured macular RNFL thickness and brain volumes/thickness in regions linked to cognitive functions and early structural brain alterations related to cognitive decline. However, the weak effect sizes and lack of statistical significance after multiple comparison correction require cautious interpretation. These findings may serve as preliminary data for hypothesis generation in future research. Larger, adequately powered population-based longitudinal investigations are necessary to confirm these associations and to clarify the temporal dynamics between retinal and brain changes. Ultimately, such studies are needed to evaluate the potential utility of macular RNFL as a non-invasive biomarker for neurodegeneration.

Methods

Participants

We enrolled Korean older adults aged over 60 years from two population-based longitudinal cohort studies: the Korean Longitudinal Study on Health and Aging (KLoSHA)59 and the Korean Longitudinal Study on Cognitive Aging and Dementia (KLOSCAD)60, collectively termed the KLOSHA-Eye61 and KLOSCAD-Eye studies. Specific details of the two cohorts are described in our previous work10. Five hundred participants were enrolled from September 2010 to September 2011 and, 70 were excluded due to high myopia (axial length > 26 mm), high intraocular pressure (IOP > 21 mmHg), self-reported glaucoma history, and combined ocular pathologies that might affect retinal layer thickness, such as age-related macular degeneration, epiretinal membrane, diabetic macular edema, and diabetic retinopathy found in the OCT infrared image, leaving 430 participants10. Among these 430 participants, 166 had undergone brain MRI within 3 years before or after the OCT examination and were included in the current analysis. The Institutional Review Board of Seoul National University Bundang Hospital (SNUBH) approved this population-based longitudinal cohort study, which adhered to the tenets of the Declaration of Helsinki (IRB No. B-0912-089-010). Written informed consent was obtained from all the study participants.

Ophthalmic and cognitive function assessments10

The baseline assessment comprised comprehensive ophthalmic examinations, including spectral-domain optical coherence tomography (SD-OCT, Spectralis, Heidelberg Engineering, Germany), best-corrected visual acuity (BCVA), intraocular pressure, auto-kerato-refractometry, and optical biometry with axial length measurement (IOL Master; Carl-Zeiss Meditec, Dublin, CA, USA). The macula protocol, consisting of a raster-scan composed of 31 horizontal lines centered on the fovea, was performed with 25 frames averaged for each OCT B-scan, along with 6 mm length and automatic real time (ART) processing61. We obtained the retinal layer thicknesses of both the macula/fovea and peripapillary areas. The macula/fovea area adopted nine macular fields based on the Early Treatment Diabetic Retinopathy Study group, consisting of three concentric rings centered on the fovea measuring 1 mm (center), 3 mm (inner), and 6 mm (outer)(supplementary Fig. S1 online). We measured six individual retinal layers (the RNFLs, ganglion cell layer, inner plexiform layer, inner nuclear layer, outer plexiform layer, and outer nuclear layer) using a built-in automated program (HEYEX software) (supplementary Fig. S1). In our study, the outer nuclear layer (ONL) includes the thick hyporeflective band, the anatomical ONL and Henle fiber layer (HFL), before the 2014 IN-OCT consensus62. Two independent retina specialists (H.M.K. and Y.J.P.) manually measured the subfoveal choroidal thickness using the enhanced depth imaging mode, and the average of repeated measures were analyzed. We collected and analyzed retinal thickness data for the outer ring, inner ring, and total macular areas. The individual participant’s right eye was selected as the analyses of the retinal thickness data.

Research neuropsychologists or trained research nurses performed comprehensive neuropsychological assessments, including the Korean version of the Consortium to Establish a Registry for Alzheimer’s Disease Assessment Packet (CERAD-K)63. A panel of research neuropsychiatrists diagnosed dementia according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition diagnostic criteria64. MCI was diagnosed according to the consensus criteria from the International Working Group on MCI65. The presence of objective cognitive impairment was ascertained when the performance of the participants was − 1.5 standard deviation (SD) or below the age-, sex-, and education-adjusted norms in any of the neuropsychological tests.

MRI acquisition and preprocessing

We obtained three-dimensional (3D) structural T1-weighted spoiled gradient echo MR images of the participants using a 3.0 T Achieva Scanner (Philips Medical Systems; Eindhoven, Netherlands) at Seoul National University Bundang Hospital. The images were acquired using the following parameters: voxel size of 1.0 × 0.5 × 0.5 mm31.0 mm sagittal slice thickness with no inter-slice gap, echo time of 4.6 ms, repetition time of 8.1 ms, flip angle of 8° and a matrix size of 175 × 480 × 480 in the x, y, and z dimensions. We used the original Digital Imaging and Communications in Medicine (DICOM) format images and converted them to the Neuroimaging Informatics Technology Initiative (NIfTI) format for analysis using MRIcron software. Subsequently, We resliced the T1 images to isovoxels sized 1.0 \(\:\times\:\) 1.0 \(\:\times\:\) 1.0 mm³.

Voxel-Based morphometry (VBM) analysis

VBM analysis was performed using Statistical Parametric Mapping (SPM) software (SPM12, Wellcome Centre for Human Neuroimaging, London, UK) implemented in MATLAB R2020b (MathWorks, Natick, MA, USA). Preprocessing included spatial normalization to MNI space, segmentation into gray matter, white matter, and cerebrospinal fluid, and smoothing with an 8-mm full-width at half-maximum (FWHM) Gaussian kernel.

Cortical thickness analysis

For surface-based analysis, we employed FreeSurfer version 6.0 (http://surfer.nmr.mgh.harvard.edu) to automatically segment and parcellate cortical structures based on the Desikan-Killiany-Tourville (DKT) atlas66. The processing pipeline (recon-all) includes motion correction, intensity normalization, skull stripping, Talairach transformation, topology correction, and surface reconstruction. After parcellation, we extracted mean cortical thickness values from all regions of interest (ROIs)67.

Statistical analyses

We compared the demographic and clinical characteristics of the 166 participants included in this analysis with the 263 individuals excluded due to the absence of MRI using Student’s t-test (Continuous variables) and Pearson’s chi-square tests (Continuous variables).

To identify brain regions associated with RNFL thickness, we employed voxel-based morphometry (VBM). In the VBM analysis, inner, outer, and total RNFL thickness were each included as variables of interest in a multiple regression model to examine their associations with regional brain volumes. We controlled for potential confounding factors, including age and sex, as well as total intracranial volume. For VBM, significance was determined at a p-value of 0.001 (peak level, uncorrected), with a cluster size threshold set at 100 voxels. Subsequently, we analyzed the association between cortical thickness in all 70 regions of interest (ROIs) and outer/inner/center/total RNFL thickness using Pearson’s partial correlation, controlling for age and sex. A p-value of < 0.05 was considered statistically significant for these initial analyses.

To account for multiple comparisons, we applied correction methods suited to each analysis type. For the VBM analysis, we used Family-wise Error (FWE) correction at p < 0.05 (peak level), a stringent approach appropriate for controlling for false positives across the thousands of voxels tested in whole-brain analysis68. For ROI-based cortical thickness correlations, we applied the False Discovery Rate (FDR) correction (Benjamini–Hochberg method)69 to account for the multiple ROIs being tested (p < 0.05). This method controls the expected proportion of false positives and is well-suited for analyses involving numerous discrete comparisons.

VBM was used to identify voxel-wise associations with gray matter volume across the whole brain, while region-based surface analysis using FreeSurfer was used for cortical thickness analysis. These complementary methods differ in spatial resolution and sensitivity to anatomical features, enabling a more comprehensive assessment of structural brain changes. VBM and partial correlation analyses were also conducted for the ganglion cell layer, inner plexiform layer, inner nuclear layer, outer plexiform layer, and outer nuclear layer, as well as for the peripapillary RNFL (Supplementary tables S1, S2 online). All statistical analyses were performed using R (version 4.3.1, R Foundation for Statistical Computing, Vienna, Austria).