Ganglion Cell – Inner Plexiform Layer Damage in Diabetic Patients: 3-Year Prospective, Longitudinal, Observational Study

Diabetes is expected to accelerate age-related ganglion cell–inner plexiform layer (GC-IPL) loss, but there is limited information on the rate of reduction in GC-IPL thicknesses. We aimed to evaluate the reduction rate of GC-IPL thickness in diabetic patients, and to compare the rates between patients without and with diabetic retinopathy (DR). We included 112 eyes of 112 patients with diabetes [49 eyes without DR (no-DR group) and 63 eyes with mild to moderate non-proliferative DR (NPDR group)] and 63 eyes of 63 normal controls (control group) in this study. Macular GC-IPL thickness in all participants was measured for 3 years at 1-year intervals. The reduction rates of GC-IPL thickness were determined by linear mixed models and compared among the three groups. The estimated reduction rates of the average GC-IPL thickness in the no-DR (−0.627 μm/year) and NPDR (−0.987 μm/year) groups were 2.26-fold (p = 0.010) and 3.56-fold (p = 0.001) faster, respectively, than the control group (−0.277 μm/year). Age, duration of diabetes, and baseline average GC-IPL thickness were associated with longitudinal changes in average GC-IPL thickness. The GC-IPL reduction rate was significantly faster in diabetic patients, with and without DR. Physicians should therefore be aware that GC-IPL damage continues even if there is no DR.

The GC-IPL thickness at each visit. The average thickness, and that of the six segments, of the GC-IPL showed a significant reduction during the 3-year period in all three groups (all, p < 0.05), except the superotemporal (p = 0.107), inferotemporal (p = 0.057), and superonasal sector (p = 0.146) thicknesses in the control group. Moreover, significant differences among the three groups were observed at all visits. Using post hoc analysis, the no-DR and NPDR groups had significantly reduced thickness measurements compared with the control group (all, p < 0.05), except at the initial visit with respect to the superior, superotemporal, and inferior sectors, which showed differences only between the control and NPDR groups. No significant difference was found between the no-DR and NPDR groups, at any visit or for any thickness parameter. Details of all these analyses are provided in Table 2 Table 3), respectively, and the rate in two diabetic groups were 2.26-fold (p = 0.010) and 3.56-fold (p = 0.001) faster than the control group. The estimated mean reduction rate was significant for all parameters, in all three groups (all, p < 0.05), except for the superotemporal (p = 0.834), inferotemporal (p = 0.070), and superonasal (p = 0.097) sector thicknesses in the control group. When comparing the rate of reduction of the thickness parameters among the three groups, the no-DR group showed a higher reduction rate in the average (p = 0.010), superotemporal (p = 0.001), inferonasal (p = 0.032), and superonasal (p = 0.042) values than the control group. The NPDR group showed a Factors associated with longitudinal changes of GC-IPL thickness. In univariate linear mixed models, age (p = 0.001), duration of diabetes (p = 0.005), BCVA (p < 0.001), and baseline average GC-IPL thickness (p < 0.001) were significantly associated with longitudinal changes in the average GC-IPL thickness (Table 4). In multivariate linear mixed models including the above significant variables in univariate analyses, age (p = 0.019), duration of diabetes (p = 0.016), and baseline average GC-IPL thickness (p < 0.001) were again significantly associated with longitudinal changes in the average GC-IPL thickness.

Discussion
In this prospective study, we followed-up 112 diabetic patients and 63 normal controls for 3 years and analyzed longitudinal changes in macular GC-IPL thickness. The results showed a significant reduction in GC-IPL thickness over time, and the rate of reduction of the average GC-IPL thickness in the no-DR and NPDR groups was 2.26-and 3.56-fold faster than that in the normal control group, respectively; this confirms that diabetes is a critical factor accelerating age-related loss of the GC-IPL thickness, where this was observed regardless of whether DR was present.  Table 2. Changes in ganglion cell-inner plexiform layer thickness by visit. All values are the mean ± standard deviation. DR = diabetic retinopathy; NPDR = nonproliferative diabetic retinopathy; FU, follow-up. *p-value for one-way analysis of variance (ANOVA) with Bonferroni correction. † p-value for repeated-measures ANOVA. ‡ Indicates control group was significantly greater than the no-DR and NPDR groups. § Indicates control group was significantly greater than the NPDR group. Significant differences are in bold font.
Retinal blood flow is affected by neural activity and retinal metabolism, which is known as neurovascular coupling and involves complex mechanisms 19 . In diabetic patients, hyperglycemia triggers metabolic pathways, such as the polyol and hexosamine pathways, resulting in the production of free radicals and advanced glycation end products; along with inflammation and ischemia, these are essential processes for the development of DR 20,21 . The activation of these pathways causes abnormalities in the neural retina, resulting in retinal neurodegeneration and retinal microangiopathy in the capillary bed. The most important features of DRN are neural apoptosis and reactive gliosis 3 . A previous study reported that retinal ganglion and amacrine cells were the retinal neurons associated with apoptosis, the process of which is accompanied by reactive gliosis occurring in astrocytes and Müller cells 22,23 . These processes cause a reduction in the thickness of the inner retinal layer, including the GC-IPL and RNFL, which can be detected by OCT.
Recent studies have suggested that DRN occurs prior to vascular abnormalities in diabetic patients, and is involved in the development of early microvascular changes. Neurodegeneration has been shown to cause breakdown of the blood-retinal barrier (BRB) 24,25 , as well as vasoregression 26 and impairment of neurovascular coupling 27,28 . In addition, glutamate accumulation induced by DRN increases the secretion of vascular endothelial  Table 3. Estimated mean rates of reduction in the ganglion cell-inner plexiform layer thickness: results from linear mixed models. DR = diabetic retinopathy; NPDR = nonproliferative DR; CI = confidence interval. *p-value for between-group differences and the follow-up duration based on linear mixed models. Significant differences are in bold font. (2020) 10:1470 | https://doi.org/10.1038/s41598-020-58465-x www.nature.com/scientificreports www.nature.com/scientificreports/ growth factor, which leads to damage to the BRB 29 . Considering these mechanisms, DRN is a crucial factor in the development of DR, and it could explain our finding of significant GC-IPL loss in patients without DR.
Although ganglion cell loss in diabetic patients has been identified by histological and experimental studies 4 , cross-sectional studies using OCT have reported different results. Studies by van Dijk et al. 5,8,30 reported a decrease in GC-IPL and macular RNFL thickness in patients with, but not without, DR. However, Ng et al. 31 reported GC-IPL loss in patients both with and without DR; the loss was progressive in advanced DR, with a decrease in inner retinal layer thickness seen in both the no-DR and NPDR groups. The retinal ganglion cell layer has been associated with DRN and may show changes that can be detected, because the ganglion cell layer in the macular area is 10-20 times thicker than their axons in the RNFL. In addition, several abnormalities in retinal function have been observed in diabetic patients and rats without DR, based on electroretinograms 32-34 . Considering the above results and those of the present study, which showed a significantly higher rate of reduction of GC-IPL thickness in the no-DR group than in the control group, inner retinal thinning in patients without DR supports the view that DRN is an early event in the pathogenesis of DR that precedes retinal microvascular changes.
Although the rate of reduction of GC-IPL thickness in this study was higher in the NPDR group than in the no-DR group, there was no significant difference between the groups, nor in any other thickness parameter. This could be explained by insufficient statistical power to reveal differences between the groups, due to the relatively small group sizes. In addition, GC-IPL loss may have been masked by subclinical macular edema not detected by OCT.
The results of this study were similar to our previous study, which studied pRNFL thickness using a similar study design 35 . The previous study determined that diabetic eyes had a significantly greater decrease in pRNFL over 3 years than normal eyes regardless of mild DR progression. Several previous studies have reported progressive age-related loss of GC-IPL thickness, based on spectral domain OCT, in normal subjects and patients with certain ocular conditions (Fig. 2). The average rate of GC-IPL thickness reduction in normal subjects was reported as −0.249 ~ −0.53 μm/year 36-39 , whereas Hanmmel et al. 40 reported no significant age-related reduction. These values were then compared with those for abnormal conditions, particularly glaucoma, and the rate of reduction in glaucoma patients was found to be significantly higher than that in normal subjects. There was also a difference in the rate according to glaucoma status, which was reported to be in the range of approximately −0.5 ~ −1.0 μm/year [38][39][40][41] , and Lee et al. 42 reported a rate of −1.46 μm/year in pseudoexfoliation glaucoma. The rate of GC-IPL loss in the control group in the present study (−0.277 µm/year) was consistent with the above studies, and the reduction rates in the no-DR and NPDR groups (−0.627 µm/year and −0.987 µm/year, respectively) were similar to those of glaucoma patients. Although it is difficult to directly compare these values because of differences in study design, our results suggested that diabetes is a significant factor in GC-IPL loss, with an importance comparable to that of glaucoma.
This study had several limitations. First, the longitudinal data on blood glucose levels were not obtained. Although the initial HbA1c levels were not associated with changes in GC-IPL thickness, and no relationship between serial HbA1c levels and progression of DRN was reported in the previous study 43 , longitudinal changes in HbA1c levels might have affected the rate of GC-IPL reduction. Second, inner and outer nuclear layers also expected to change considering the mechanism of DR. However, we did not analyze because nuclear layer thicknesses cannot be measured automatically on our platform. Additional research will be needed for to investigate the retinal nuclear layer. Third, although we carefully checked for glaucomatous findings, such as RNFL defects and glaucomatous optic disc based on OCT findings, we did not perform a visual field test and the patient's data was not reviewed by a glaucoma specialist; thus, it is possible that we enrolled patients with pre-perimetric glaucoma. Finally, we could not analyze the association between DRN and functional changes. Additional well-designed prospective studies are needed to overcome these limitations.
In conclusion, we confirmed that the GC-IPL reduction rate was significantly faster in patients with diabetes, both without and with DR, compared with normal controls. This is the first study to demonstrate a temporal association of GC-IPL reduction with progression of diabetes. Our results suggest that DRN antecedes the microvasculopathy in people with DR, in turn suggesting that DRN may be an important component in the pathogenesis and fluorescein angiography using an HRA Spectralis system (Heidelberg Engineering, Heidelberg, Germany). We divided the diabetic patients into two groups: a no-DR group and a nonproliferative DR (NPDR) group. NPDR was graded according to the International Clinical Diabetic Retinopathy Disease Severity scale 45 . The NPDR group consisted only of mild to moderate NPDR cases, and all patients with a more advanced grade of DR, such as severe NPDR or proliferative DR were excluded. Among the subjects who visited our Retina and Vitreous clinic for various reasons (routine checkups for ocular diseases such as peripheral vitreous floaters, cataract, health screening checkups), those who met eligibility criteria and had glucose level data [fasting plasma glucose < 100 mg/dL or hemoglobin A1C (HbA1c) < 5.7%] were enrolled in the normal control group. Normal subjects (control group) had no history of diabetes and no ocular disease, with a BCVA ≥ 20/25, a normal IOP range, and a SE within ± 6.0 D.
All participants underwent examinations including BCVA, IOP, slit-lamp examination, dilated fundus examination, photography, and OCT every 12 months for 36 months (total of four examinations). Patients who were diagnosed with progressed DR grade (except mild to moderate NPDR) or macular edema, were lost to follow up, or underwent intraocular surgery were excluded.
OCT protocol. Spectral domain OCT scan was performed using a Cirrus HD-OCT ® instrument (Carl Zeiss).
In the macular scan, GC-IPL thickness was measured using the ganglion cell analysis algorithm in the macular cube 200 × 200 protocol 46 . This algorithm calculates the average thickness of six sectors (superotemporal, superior, superonasal, inferonasal, inferior, and inferotemporal) of the elliptical annulus (vertical inner and outer radii of 0.5 and 2.0 mm, respectively; horizontal inner and outer radii of 0.6 and 2.4 mm, respectively), centered on the macular area (Fig. 3). The pRNFL thickness and optic nerve head (ONH) parameters, such as the rim area and average cup/disc ratio were measured using the optic disc 200 × 200 protocol. All participants underwent two OCT scans by an experienced examiner in all participants, and we selected the best scan with signal strength ≥ 7. We excluded results from patients with OCT scan signal strength < 7 or segmentation errors. R Foundation for Statistical Computing, Vienna, Austria) and SPSS statistical software (version 21.0; SPSS Inc., Chicago, IL, USA). Snellen BCVA results were converted into the logarithm of the minimum angle of resolution (logMAR). Continuous variables are presented as the mean ± standard deviation. Differences were considered significant at p < 0.05.
Baseline demographics and OCT measurements, including CMT, average GC-IPL, average pRNFL thickness, and ONH parameters were compared among groups using one-way analysis of variance (ANOVA) with post-hoc Bonferroni correction and the chi-square test; GC-IPL thickness at each visit was also compared. Repeated measures ANOVA was used to analyze longitudinal changes in GC-IPL thickness in each group. Linear mixed effects models were used to calculate the rate of reduction, with 95% confidence interval (CI), of GC-IPL thickness over time; the rate was then compared among the three groups. Age, sex, SE, AL, follow-up duration, and interaction among the groups were included as fixed effects, and the subject was included as a random effect. In addition, univariate and multivariate linear mixed models were also fitted to determine factors correlated with longitudinal changes in GC-IPL thickness.

Data availability
Data supporting the findings of the current study are available from the corresponding author on reasonable request.