Ganglion Cell-Inner Plexiform Layer Thickness is Associated with Persistently Cognitive Decline -The Rugao Longevity and Aging Study

Background: The simple, convenient and well-validated biomarkers are imperative for detection of cognitive decline (CD). The powerful evidence is lacked for verifying the reliability and clinical utility of retinal biomarkers for detection of CD with repeated assessments. To investigate the association of retinal thickness with CD using repeated assessments. Methods: This study included 446 older adults with three-time repeated assessments of cognitive function during 5-years follow-up. Retinal thickness measured on spectral-domain optical coherence tomography. Logistic regression models were conducted to analyze the association of retinal thickness with cognitive function. Results: According to cognitive status in three assessments, individuals were categorized into consistently normal cognition groups (N = 159), persistently CD groups (N = 134), progressed to CD groups (N = 70), and reverting or uctuating CD groups (N = 83). Thinner ganglion cell-inner plexiform layer (GC-IPL) was associated with persistently CD (odds ratio [OR] per 1-μm decrease: 1.09, 95% condence interval [CI], 1.02-1.18; per standard deviation [SD] decrease: 1.78, 95%CI, 1.04-3.19) rather than progressed to CD, reverting or uctuating CD. No signicant relationship was found between retinal nerve ber layer and any CD subgroups (p > 0.05). Conclusions: Thinner GC-IPL was associated with persistently CD, suggesting retinal neurodegeneration may be a promising biomarker for persistently CD. Further studies, including both longitudinal and repeated measurements of retinal layer thickness and cognitive function, are needed to assess the possibility of retinal thickness as a biomarker for persistent CD.

prospective longitudinal studies found that thinner retina was a risk factor for incidence of AD and future cognitive impairments, suggesting the possibly predictive role of retinal thickness [13,14]. Hence, overwhelming studies demonstrated the possible role of retinal thickness in diagnosis and detection of MCI and AD. However, it was still lacking powerful evidences to draw conclusions with respect to the reliability or clinical utility of any potential retinal biomarkers with CD in repeated assessments [15,16]. Therefore, we aimed to investigate the association of retinal thickness with CD using repeated assessments among 5-years follow-up.

Study population
The Rugao Longevity and Ageing Study (RuLAS) is an observational, prospective and community-based cohort study [17]. The baseline survey was conducted from November to December 2014 (Wave 1). A total of 1788 adults aged 70-84 years were recruited. Then, three follow-up surveys were conducted in In December 2019, SD-OCT was added to perform at wave 4 of the cohort. A total of 2200 older adults were recruited in wave 4. SD-OCT was available for 1721 subjects (78.22%). Subjects who did not undergo SD-OCT (n = 479) or had upgradable OCT due to poor quality scans (n = 538) were excluded. A total of 1183 subjects who had complete retinal thickness data were recruited in wave 4. Among the 1183 subjects, 480 subjects completed both three visits (December 2014, November 2017 and December 2019) and had cognitive information. In addition, subjects with retinal related disease were excluded, including glaucoma (n = 4), age-related macular degeneration (n = 4), diabetic retinopathy (n = 3), pathological myopia (n = 16) and other retinal disease (n = 7). At last, 446 participants with three repeated measurements of cognitive function were included and analyzed in our study (Fig. 1).

Outcomes
In our study, cognitive function was evaluated by the revised Hasegawa's dementia scale (HDS-R) [17], which comprising of orientation, memory, attention/calculation and verbal uency [18]. HDS-R has been widely accepted in Asian populations in clinical and epidemiological surveys for the assessment of cognitive impairment [19]. Previous studies observed that HDS-R was similar to Mini-Mental State Examination and had better diagnostic accuracy for screening AD [20]. In addition, the HDS-R was more robust to demographic in uence (such as level of education, age and gender) than MMSE [20,21]. In conclusion, it was a brief and reliable tool to assess global cognitive function in Asian [22,23]. In our study, individuals who scored higher than 21.5 were de ned as normal cognition, while who scored 21.5 or below were de ned as CD [17,24,25]. Cognitive function was assessed in November 2014, November 2017 and December 2019.
Individuals were categorized into four subgroups according to cognitive function in three repeated assessments among 5-years follow-up [26]. Individuals with consistently normal cognition (Normal-Normal-Normal) in both three visits were categorized into consistently normal cognition group (N = 159); Individuals with consistently CD in both three visits (CD-CD-CD) were categorized into persistently CD group (N = 134); Individuals who progressed to CD from normal cognition within three visits (Normal-Normal-CD or Normal-CD-CD) were categorized into progressed to CD group (N = 70). Individuals who returned to CD from normal cognitive (Normal-CD-Normal, or CD-Normal-CD), or returned to normal cognitive from CD in any two visits (CD-Normal-Normal, or CD-CD-Normal) were categorized into reverting or uctuating CD group (N = 83).

SD-OCT scan
SD-OCT scanning was performed without pupil dilation using the spectral domain OCT-HS100 (Canon Inc, Tokyo, Japan) with a macular 3D scans over a 10 × 10 mm area (1024 A-Scan × 128 B-Scan). The Early Treatment Diabetic Retinopathy Study (ETDRS) grid focused on the macular was used for OCT measurements (Fig. 2). The ETDRS grid consisted of three concentric circles of 1, 3, and 6 mm diameters. The 3-and 6-mm circles were each divided into superior, inferior, nasal and temporal quadrants. Macular retinal nerve ber layer (mRNFL) and ganglion cell layer-inner plexiform layer (GC-IPL) were segmented.
The right eye was used for measurements if the Signal Strength Index was ≥ 5. Otherwise, the left eye was used. The exclusion criteria included: (1) with signal strength index of both eyes < 5; (2) any ocular disorders, including pathological myopia, age-related macular degeneration, diabetic retinopathy, glaucoma, vitreoretinal interface abnormalities (e.g. epiretinal membrane, macular hole), and other eye pathology; (3) a history of clinical stoke, uncontrolled diabetes and hypertension. The thickness were automatically segmented, and manually corrected was conducted by two masked examiners (Shen H and Gong W) if necessary. The mean thickness was calculated.

Covariates
Demographic, physiologic and clinical data were collected from the RuLAS. Demographic data included age, gender, married status, educational status, smoking and alcohol consumption. Physiologic variables, including body mass index (BMI), high-density lipoprotein (HDL), low-density lipoprotein (LDL), triglyceride, serum creatinine, fasting blood glucose (FBG) and High-sensitivity C-reactive protein (Hs-CRP) were also measured. Clinical variables, including, major cardiovascular disease (CVD), diabetes mellitus, depressive status (assessed by the 15-item Geriatric Depression Scale [27]), hypertension were collected. Major CVD included cerebral infarction, stroke, cerebral hemorrhage, coronary heart disease, myocardial infarction and heart failure. In addition, mobility (measured by time up and go test [28]) and grip strength was also assessed in our study.

Statistical Analyses
Firstly, we described the characteristics of participants in our study. Continuous and categorical variables were presented as mean with standard deviation (SD) and frequency (%), respectively. Group differences between the four groups were analyzed by chi-square or t test. The differences between consistently normal cognition and CD subgroups were also analyzed. Secondly, we tted logistic regression models, treating memberships in all four group as outcome variables, with the consistently normal cognition group as reference, to assess the effect of mRNFL and GC-IPL in four models. Model 1: unadjusted; Model 2: adjusted for age, body mass index, sex, smoking, alcohol accumulation, married status and education; Model 3: adjusted for model 2 + serum fasting glucose, triglyceride, HDL, LDL, creatinine and Hs-CRP; Model 4: adjusted for model 3 + CVD, hypertension, diabetes mellitus, mobility, grip strength and depressive status. Lastly, sensitivity analyses using logistic regression models were conducted to validate the relationship in participants without CVD and/or diabetes. A p-value (two-tailed) less than 0.05 determined as statistical signi cant. All analyses were conducted by SPSS 22.0 or R (Version 3.6.1: www.r-project.org/).

Characteristics of included population
In our study, a total of 446 participants were analyzed. Table 1 summarized demographic, physiologic and clinical data of the participants according to different CD subgroups. One hundred and fty-nine (35.65%), 70 (15.70%), 83 (18.61%) and 134 (30.04%) individuals were categorized into consistently normal cognition group, progressed to CD group, reverting or uctuating CD group and persistent CD group, respectively. Difference was observed in the four subgroups. The mean age were 78.96 (3.05), 80.16 (3.81), 80.00 (3.24) and 80.37 (3.52) years, respectively. The detail information was showed in Table 1. The association of retinal thickness with cognitive subgroups in primary analyses  The association of retinal thickness with cognitive subgroups in sensitive analyses In order to validate our results, we conducted sensitive analyses in older adults without major CVD and/or diabetes.

Discussion
To our knowledge, it is still lacking evidences to validate the association of retinal thickness with CD using repeated assessments in older adults. This is the rst study, involving 3 times repeated assessments of cognitive function among 5-years follow-up, to investigate the association. In this community-based study, we found that thinner GC-IPL was associated higher risk of persistent CD but not progressed to CD, or reverting or uctuating CD, independent of demographic, physiologic and clinical confounder factors, with consistently normal cognition groups as reference. In contrast, no signi cant association of mRNFL with any CD subgroup was found.
The association of GC-IPL with cognitive function was widely studied. Our study illustrated signi cant association of GC-IPL with persistent CD. Similarly, previous studies found that patients with AD and MCI had thinner GC-IPL than controls [7,11,29]. Thinner GC-IPL was also associated with lower brain white matter microstructure integrity and grey matter volume [30,31]. Additionally, animal models also suggested the close association of GC-IPL with cognitive function. GC-IPL is composed of retinal ganglion cells (RGCs), while amyloid-β deposits in the RGCs layer may cause neurodegeneration in the retina of a transgenic mouse model of AD [32]. More importantly, previous animal study illustrated that dendritic changes in RGCs preceded cell loss in a mouse model of AD [33]. Therefore, the GC-IPL may be a useful biomarker for the diagnosis of CD.
In our study, there was no signi cant association of mRNFL thickness with any CD groups after adjusted for potential confounder factors. The results were consistent with a recent study conducted in biomarkercon rmed preclinical AD population, in which no difference were found in macular or peripapillary RNFL between amyloid positive (Aβ+) and negative (Aβ-) individuals [34]. Similar results were also observed in the Rotterdam Study [13]. In that study, Mutlu et.al illustrated that thinner peripapillary RNFL was not associated with prevalent dementia in the Rotterdam Study. While Ko et.al found that thinner mRNFL was associated with worse cognitive function as well as future cognitive decline in UK biobank [14]. The heterogeneity of included population may be the reason of the inconsistency. The age of subjects in the Rotterdam Study (mean age: approximately 68.9 years) and RuAS (mean age: approximately 80.0 years) were much older than that in UK biobank (mean age: approximately 56.0 years). Due to the retinal thickness was thinner with increased age [35]; the real and accurate association of RNFL with cognitive function might be concealed by aging process of retina. Interestingly, a recent study with wide adult age range (mean age: 54.3 years; age range: 30-94 years) found no signi cant association of peripapillary RNFL with performance in any cognitive domain in the Rhineland Study [36], which proved our results in some degree.
As we all known, the RNFL is composed of RGCs axons, while GC-IPL is composed of RGCs bodies and dendrites [37,38]. The damage of optic nerve may gradually swell from bodies and dendrites to axons [39,40]. RNFL thickness may decrease gradually over very long time. The similar condition was also observed in the Rotterdam study that thinner GC-IPL but not RNFL thickness was associated with prevalent dementia [13]. Due to the cross-sectional setting of community-based cohort and the older age in analyses, it was limited to explore reasons of different effects of GC-IPL and mRNFL on cognitive function. Despite this, our study was rst to explore the association of retinal thickness with cognitive function using repeated measurements and found exciting results. In the future, larger prospective and community-based cohort study with wider age range and repeated measurements would be conducted to validate our results.
There were several limitations in our study. Firstly, a cross-sectional study was conducted to evaluate the association of retinal thickness with CD. Our ndings only illustrated a correlative rather than causative relationship. Hence, longitudinal studies should be conducted to validate our ndings in the future. Secondly, the sample in our study was relative small, which limited the statistic power in some degree.
Despite this, we found robust and independent association of GC-IPL with persistent CD. More subjects would be recruited to validate the results in further studies. Lastly, we just measured the right eye using SD-OCT in our study, which may partly affect the measurement of retinal thickness. However, previous study did not nd signi cant difference in retinal layer thicknesses between left and right eye in the Rotterdam Study [13]. Moreover, the thinning of retina was generalized in older adults and it should not be limited to one eye. Therefore, the signal-eyes measurements did not in uence our ndings.
Strengths were also included in our study. Firstly, we assessed the cognitive function for three times in each subjects among 5-years follow-up. Previous studies concentrated on the relationship between retinal thickness and cognitive function assessed by single one-point and found close relationship between them [7,[41][42][43]. However, not all individuals with CD progressed to dementia or consistently stabilized CD, while 18%-24% of mild cognitive impairments individuals reverted to normal cognition [26,[44][45][46], which potentially leading to partial association of neurodegeneration with retinal thickness. Using repeated assessments would increase the consistency and accuracy of assessing cognitive function. Therefore, we could accurately explore the relationship between retinal thickness and cognitive function.
Secondly, only one experienced physician (Dr. Shen) using the same machine performed the SD-OCT, which will signi cantly reduce the measurement errors. Finally yet importantly, structural retinal changes are not unique to preclinical or clinical AD. Similar changes may occur in other disease, such as hydroxychloroquine retinal toxicity [47] and glaucoma [48]. Exactly, we conducted the study according to rigorously exclusion criteria, which excluded individuals with retina-related disease, such as glaucoma, age-related macular degeneration, diabetic retinopathy and pathological myopia.

Conclusion
In this community-based cohort, we found thinner GC-IPL was associated with persistent CD rather than progressed to CD, or reverting or uctuating CD, independent of demographic, physiologic and clinical confounder factors. Thinner GC-IPL thickness may be a promising biomarker for persistent CD. Flow diagram of study population Figure 2