Multimodal Retinal Structural and Functional Tests Identify Local and Layer-Related Alterations in Early Type 2 Diabetes

Abstract

Purpose

To use a multimodal approach to provide deeper insights into subclinical, layer-related, and localized retinal structural and functional deficits associated with prediabetes (PreDM) and early type 2 diabetes (T2DM).

Methods

Twenty controls, 13 PreDM, and 15 T2DM age-matched (45.58 ± 9.40 years) groups participated. Color perimetry was performed at four locations 3° from fixation, using 0.5° circular red, green, blue, and yellow stimuli. Multifocal electroretinograms (mfERGs) were recorded (103 hexagons), and implicit times (ITs) and amplitudes were averaged and analyzed in ring 2, approximately corresponding to the perimetric testing. Then, 20° optical coherence tomography (OCT) and 10° OCT angiography (OCTA) images were acquired. Perfusion density (PD; all plexuses), retinal layers, and choroidal thicknesses were analyzed (central 2 mm).

Results

mfERG P1 ITs were delayed in T2DM (P = 0.03 vs. controls). Thresholds obtained for red increased with delayed P1 ITs (P = 0.02) and with increasing thickness of the inner nuclear layer (P = 0.03). Choroidal thickness showed a negative relationship with blue and yellow thresholds (P < 0.05). Thresholds obtained for red and yellow were higher in the T2DM (all P < 0.05 vs. controls) but green and blue were not (P > 0.05). PD of the intermediate capillary plexus was lower in the T2DM group (P = 0.001 vs. all groups) and was associated with HbA1c (P = 0.02).

Conclusions

This combination of tests showed alterations in middle retinal health, which could occur in retinal bipolar cells early in T2DM. A longitudinal investigation would clarify the temporal onset of these neural and vascular changes in the disease process.

The global burden of prediabetes (PreDM), a precursor of type 2 diabetes (T2DM), and T2DM is rapidly rising.^1,2 Diabetic retinopathy (DR) is the leading cause of preventable vision loss among the working-age population.³ This presents a major public health challenge and highlights the importance of a thorough understanding of the mechanisms involved in T2DM and DR for improved interventions.

Increasing evidence suggests that early T2DM is associated with subclinical, progressive microvascular and neural compromise in the retina.^4–6 It is not clear whether neuronal abnormalities precede detectable vascular changes or arise as a consequence of them. Despite advancements in detecting subclinical retinal changes, a cohesive mechanistic framework that links layer-level, localized retinal structural/microvascular changes and functional changes across the glucose dysfunction spectrum is lacking. Moreover, comprehensive information about the specific retinal neurons and visual processing pathways affected in the earliest stages of T2DM and DR remains limited. Although some investigations have highlighted inner retinal structure and function impairment as early events in T2DM and DR,^7–9 other findings have pointed to cone photoreceptor defects.^10,11 Thus, whether retinal dysfunction in early T2DM is primarily driven by inner- or outer-retinal metabolic stress, or by a combination of both, remains unresolved. Studies that have relied on a single modality often report inconsistent findings.^12,13 Here, we evaluated tests of microvascular and neuronal health in the retina across glucose dysfunction to better understand local and layer-related changes.

Adequate retinal perfusion is essential for normal retinal function. Studies have reported reduced vessel density, perfusion deficits, and enlarged foveal avascular zone (FAZ) area and perimeter in people with early T2DM.^13–18 Nesper and colleagues¹² reported that FAZ area and perfusion deficits in the full retina and superficial capillary plexus (SCP) showed a positive linear relationship with increasing DR severity. Although the retina is supplied by three capillary plexuses, work by Ong and Fawzi¹⁶ indicated that the intermediate capillary plexus (ICP) has not been independently assessed in several studies due to software challenges with available optical coherence tomography angiography (OCTA) systems. Segmented analysis of each of the three retinal capillary plexuses improves the detection of non-perfusion and other vascular defects in DR.^19,20 The current study provides a more detailed assessment of each of the retinal capillary plexuses together with retinal structural and functional assessment at discrete retinal locations in the context of PreDM and T2DM.

It is well known that multifocal electroretinography (mfERG) in people with T2DM without visible fundus changes shows decreased amplitudes and delayed implicit times (ITs).^10,11,21 We employed this technique to assess the health of photoreceptors and bipolar cells at predefined retinal locations in PreDM and early T2DM. In early T2DM, color vision test results are dependent on the test administered.²² This could lead to an underestimation of the type and prevalence of color vision defects. Studies that used the Farnsworth–Munsell 100 hue test²³ and the Cambridge color test⁷ have reported diffuse color vision defects in T2DM without DR. Conversely, those employing the Farnsworth D-15²⁴ and Adams desaturated D-15⁶ tests reported a predominant tritan (koniocellular) defect. Insight into the red–green (parvocellular) color vision loss in early T2DM remains limited. Altogether, this shows that accurate detection of subclinical color vision defects requires more sensitive tools, such as color perimetry.²⁵

Our group developed a novel color perimeter that can measure chromatic thresholds at discrete retinal locations. We established its repeatability in people with and without PreDM or T2DM.²⁶ The device can detect changes in chromatic sensitivity in people with T2DM across both koniocellular and parvocellular pathways. These changes are highly correlated with HbA1c.²⁶ Thresholds of the achromatic stimulus also correlated with Mars contrast sensitivity.²⁷ Thus, the perimeter serves as a reliable and sensitive tool for studying early color vision loss in T2DM.

Here, we used the novel color perimeter with mfERG, OCT/OCTA, the L'Anthony D-15 color test, and the Mars contrast sensitivity test to allow direct spatial, structural, and functional correlation among color vision test, electrophysiology, microvascular changes and retinal structural integrity. The macula mediates color vision and high-resolution acuity,²⁸ and vascular dropout in the extrafoveal regions (outside the central 1 mm) has been associated with DR progression.²⁹ Therefore, this extrafoveal region within the macula was of interest in our study. Leveraging this multimodal approach allowed us to develop a comprehensive report of localized retinal neurovascular alterations and structure–function relationships associated with diabetic retinal disease progression.

Methods

Study Design, Participants, and Ethical Considerations

This study enrolled 50 participants, including 21 controls (no diabetes, HbA1c ≤ 5.6%), 13 with PreDM (HbA1c = 5.7%–6.4%), and 16 with T2DM (HbA1c ≥ 6.5% or physician-diagnosed, with no or mild retinopathy) based on their HbA1c³⁰ without restriction on ethnicity or sex. All participants had a visual acuity of at least 20/32 in the right eye on the Early Treatment Diabetic Retinopathy Study (ETDRS) chart, which was the eye used in all tests and analyses. The groups were age matched at enrollment. Individuals with systemic or ocular conditions or treatments that can affect color vision were excluded. Ethical approval was granted by the University of Houston Institutional Review Board. Each participant provided written informed consent. The study conformed to the tenets of the Declaration of Helsinki.

Demographic Data and Clinical Assessments

The self-reported demographic and health status data included age, sex, race, presence of PreDM/T2DM, duration of PreDM/T2DM, and medications. We evaluated logMAR visual acuity, Mars contrast sensitivity, L'Anthony D-15 color test, slit-lamp biomicroscopy/angle assessment (SL-D301 with camera; Topcon Corporation, Tokyo, Japan), auto kerato-refractometry (KR800; Topcon Corporation), blood pressure, and HbA1c (from finger-tip blood; DCA Vantage Analyzer; Siemens Medical Solutions, Malvern, PA, USA). The D-15 color test was used as a clinical tool to compare its performance to the novel color perimeter in detecting subtle color vision deficits. It was used in our previous studies to identify color vision changes in diabetes.³¹

Color Perimetry

Chromatic thresholds were measured in the undilated right eye (with the left eye occluded) using the novel color perimetry described in our previous work.^26,27 Briefly, chromatic thresholds were measured at four diagonal locations (3° eccentricity), and participants indicated the perceived location of stimuli with a button press. Stimuli were 0.5°-diameter circular red, green, blue, and yellow stimuli defined along the Derrington–Krauskopf–Lennie (DKL) color space axes against a dim gray (10 cd/m²) background centered at the white point.³² Stimuli were presented in a randomized order amid dynamic luminance noise to prevent non-chromatic cues for detection. Chromatic thresholds were determined using a randomly interleaved adaptive one-up/one-down staircase procedure. Thresholds from specific color stimuli were also averaged to evaluate potential selective involvement of chromatic pathways in the pathophysiology of PreDM and T2DM.

Multifocal Electroretinogram

After the color perimetry, the right eye was dilated with 1% tropicamide and 2.5% phenylephrine for mfERG and other tests. The mfERG responses were recorded as in our other studies.^21,31 Briefly, we used a VERIS Science 6.4.8 system (Electro-Diagnostic Imaging, Milpitas, CA, USA) with a Burian–Allen electrode and a 103-hexagon stimulus with a 4-minute m-sequence near 100% contrast. Data was processed with one iteration of artifact removal and 10% spatial averaging from neighbors. The first-order N1 and P1 ITs were analyzed for ring 2 (∼2.5°–7.0°, which captures the locations tested in the color perimetry and OCT/OCTA), and the entire 103 hexagons to confirm adequate statistical power.

OCT and OCT Angiography

A 25-line volumetric OCT scan (20° × 20° area) and an OCTA (10° × 10° area), all with enhanced depth imaging and centered on the fovea, were acquired using a spectral-domain OCT (SPECTRALIS HRA2; Heidelberg Engineering, Heidelberg, Germany). OCT/OCTA images were manually inspected, and poor-quality images such as those with segmentation, motion, and shadowing artifacts were excluded from analyses.³³ The eight OCT/OCTA scans excluded from analysis included three controls, three participants with PreDM, and two participants with T2DM. The examiner was masked to the study groups. All images were corrected for retinal magnification during the analysis. The choroid, full retinal, and individual retinal layer thicknesses, as well as the perfusion density (PD), were analyzed within the central 2-mm region (∼6.6° diameter or ∼3.3° from fixation). This matches the location tested in the color perimetry and it is within the ring 2 hexagons of the mfERG.

The OCT scans were exported as RAW (.vol) files and then imported into MATLAB (MathWorks, Natick, MA, USA). Choroidal thickness was measured using a custom MATLAB deep learning segmentation algorithm.³⁴ The use of Heidelberg Eye Explorer (Heyex) 1.10.4.0 software allowed automatic segmentation and measurement of the full retinal and individual retinal layer thicknesses from the OCTA scans. It also automatically segmented and distinguished the angiograms of the individual retinal capillary plexuses (SCP, ICP, and deep vascular/capillary plexus [DCP]) and the choriocapillaris (CC) from the OCTA scans. The CC was delineated as the vascular layer located between 10 µm and 30 µm beneath Bruch's membrane (20 µm thick).

Each capillary plexus and CC angiogram was extracted from the OCTA scans as .tif files. Images were imported into MATLAB, and PD (in %) was measured with a custom MATLAB program using the Otsu (SCP, ICP, and DCP) and Phansalkar (CC) binarization methods.^35–37 The FAZ area, perimeter, and circularity index were analyzed for the ICP using ImageJ (National Institutes of Health, Bethesda, MD, USA) with manual tracing of the FAZ.³⁸ The FAZ metrics were assessed in the ICP because it has been less extensively studied and could provide additional information on early microvascular changes in T2DM.

Fundus and Lens Photography

A 45° color fundus photograph (TRC-NW400 Retinal Camera; Topcon Corporation) was taken and used together with the OCT/OCTA scans to rule out any posterior segment abnormalities (including moderate to severe diabetic retinopathy) from the analyses. Lens photographs (camera mounted to the slit lamp with external monitor) were captured and independently graded by two examiners (optometrists) using the Lens Opacity Classification System III (LOCS III).³⁹ Participants with a LOCS III grade > 2 (averaged between the two examiners) were excluded from the study to ensure that lens changes did not impact our findings.

Data Analysis and Statistical Tests

Prior to statistical analysis, all variables were evaluated for outliers and distributional form using formal Shapiro–Wilk tests for normality and visual inspection via distributional plots. Quantitative measurements were expressed as mean ± SD or median (interquartile range [IQR]), where relevant. Power calculations to detect differences between clinical diabetic groups across all measured variables were computed using mfERG testing (based on P1 IT across all 103 hexagons) and assuming a significance level of α = 0.05 and β = 0.2. We calculated the effect size of the mfERG to be 1.3 (Cohen's d) for the mfERG P1 IT. Thus, the sample required for 80% power was 13 per group, or 39 total. One-way analysis of variance (ANOVA) followed by Tukey's multiple comparison test for normally distributed data or Kruskal–Wallis test followed by Dunn's multiple comparison test for non-normally distributed data were used to statistically test for group-level differences. We evaluated exploratory associations between retinal layers and choroidal thicknesses and color perimetry thresholds using linear regression models for the red, green, blue, and yellow stimuli. Two models were developed to assess relationships between parameters. For all colors, model 1 estimates the univariable β coefficient of structural thickness only, and model 2 adjusts for age and body mass index (BMI).

Results

Demographic and Initial Clinical Characteristics of Participants

We enrolled 50 participants in this study. Two participants (one control and one with T2DM) were excluded from data analysis due to a staircase failure during the color perimetry and an unconventional mfERG waveform (criteria predetermined before data analysis). The mean age of the included participants (24 males and 24 females) was 45.6 ± 9.4 years. There were no statistically significant differences in age, visual acuity, contrast sensitivity, or color confusion index between the groups (see the Table). Fundus photographs revealed mild DR in three participants with T2DM. The mean duration of disease was 3.4 ± 7.3 years in the T2DM group and 8 months in the PreDM group. Many of the T2DM participants (40%) were newly diagnosed within the last 6 months, and the majority (69.2%) of participants in the PreDM group self-reported as controls but were classified as having PreDM based on their measured HbA1c.

Table.

Demographic and Initial Clinical Assessments

Characteristic	Control (n = 20)	PreDM (n = 13)	T2DM (n = 15)	Adjusted P
Age (y), mean ± SD	43.1 ± 9.1	44.2 ± 10.9	50.1 ± 7.0	0.07
Sex (M:F), n	12:8	7:6	5:10	0.29
Visual acuity (logMAR), mean ± SD	0.02 ± 0.15	0.02 ± 0.11	0.05 ± 0.13	0.73
Mars contrast sensitivity (logCS), median (IQR)	1.80 (1.72–1.80)	1.76 (1.66–1.80)	1.76 (1.72–1.80)	0.56
L'Anthony D-15 (color confusion index), median (IQR)	1.00 (1.00–1.19)	1.17 (1.00–1.26)	1.23 (1.00–1.31)	0.16
Systolic blood pressure (mmHg), mean ± SD	122.2 ± 12.2	123.4 ± 21.8	123.2 ± 15.1	0.97
Diastolic blood pressure (mmHg), mean ± SD	86.3 ± 8.1	83.8 ± 13.9	85.3 ± 9.2	0.79
HbA1c (%), median (IQR)	5.35 (5.13–5.45)	5.80 (5.80–6.00)	6.85 (6.50–7.80)	<0.0001

Groups were compared using either one-way ANOVA or Kruskal–Wallis test (depending on the normality of the data), followed by appropriate multiple comparison tests.

All but one participant in the T2DM group were taking T2DM medications, including metformin (10 participants), glucagon-like peptide 1 (GLP-1) receptor agonists (eight participants), sodium–glucose cotransporter-2 (SGLT2) inhibitors (four participants), and insulin (one participant). Eight were also treated for co-morbid hypertension. Only one participant in the PreDM group was on treatment (metformin), and two were taking medication for hypertension.

Chromatic Thresholds Across Diabetes Status

We compared chromatic thresholds (log cone contrast [logCC]) across groups for each color (Fig. 1; Supplementary Table S1). The thresholds obtained using the red stimulus were significantly higher in the T2DM group compared to the controls (−1.38 ± 0.20 vs. −1.53 ± 0.16 logCC; P = 0.02). Similarly, thresholds obtained using the yellow stimulus were significantly higher in the T2DM group than in controls (−0.61 ± 0.11 vs. −0.76 ± 0.15 logCC; P = 0.004). The yellow color stimulus had 86.7% sensitivity and 65.0% specificity for identifying T2DM compared to controls (Supplementary Fig. S1). The area under the curve was 0.77 for the yellow stimulus. Averaged thresholds for colors representing the parvocellular pathway (red and green stimuli) and koniocellular pathway (blue and yellow stimuli) were significantly higher in the T2DM group compared to controls (all P < 0.05). The correlations between red and green thresholds (r = 0.76) and blue and yellow thresholds (r = 0.81) were very high. Thresholds obtained using the green and blue stimuli did not significantly differ among the three groups (all P > 0.05). No differences were observed in any of the stimulus thresholds between the controls and PreDM group (all P > 0.05).

Figure 1.

People with T2DM have higher chromatic thresholds. Chromatic thresholds were measured in the undilated right eyes of 20 healthy controls (HbA1c ≤ 5.6%), 13 participants with PreDM (HbA1c = 5.7%–6.4%), and 15 participants with T2DM (HbA1c ≥ 6.5%). (A, B, D, E) Within-group average thresholds (logCC) across the four positions tested for the red, green, blue, and yellow stimuli, respectively, during different segments of the test. (C, F) Thresholds obtained from the red and green stimuli were averaged (C), as were those from the blue and yellow stimuli (F), to assess potential selective involvement of the parvocellular and koniocellular pathways, respectively, in the pathophysiology of PreDM and T2DM. Numbers on the plots represent the adjusted P values. Adjusted P < 0.05 was considered statistically significant.

Diabetes and mfERG Implicit Times and Amplitudes

The mfERG P1 ITs were significantly delayed in the T2DM group compared to controls for the average of the six hexagons surrounding the foveal hexagon (all ring 2 hexagons) and the average of all 103 hexagons (all P < 0.05). We found no significant differences in N1 ITs or any of the amplitudes among the groups for all of the regions analyzed (all P > 0.05). A summary of the mfERG findings is presented in Figure 2 and Supplementary Table S2.

Figure 2.

Multifocal electroretinogram P1 implicit times are delayed in type 2 diabetes. The mfERG responses were recorded in dilated right eyes using a Burian–Allen electrode, a 103-hexagon stimulus with a 4-minute m-sequence near 100% contrast on a VERIS System. (A, B) The P1 implicit time (A) and amplitude (B) were averaged and analyzed for all of the ring 2 hexagons. (C) The 103 hexagons with the ring 2 hexagons highlighted in red. Numbers on the plots represent the adjusted P values. Adjusted P < 0.05 was considered statistically significant. Healthy controls: HbA1c ≤ 5.6% (n = 20); PreDM: HbA1c = 5.7%–6.4% (n = 13); T2DM: HbA1c ≥ 6.5% (n = 15).

Perfusion Density, FAZ Metrics, Retinal, and Choroidal Thicknesses

After excluding the eight poor-quality scans, the final OCT/OCTA dataset included 17 controls, 10 participants with PreDM, and 13 participants with T2DM. We compared PD within the central 2-mm region (∼6.6°) across groups for each retinal capillary plexus and the CC (Fig. 3; Supplementary Table S3). In the ICP, PD was significantly lower in the T2DM group compared to the control (P = 0.001) and PreDM (P = 0.001) groups (Fig. 4). PD in the SCP trended lower in the T2DM group compared to the other groups (adjusted P = 0.06). PD in the DCP and CC were not significantly different among the groups (all P > 0.05). To further investigate the impact of PreDM and T2DM on the ICP (which is less studied in literature compared to the other plexuses), we analyzed FAZ area, perimeter, and circularity index for this plexus. All of the FAZ metrics in the ICP were similar among the groups (all P > 0.05).

Figure 3.

Perfusion density of the intermediate capillary plexus is lower in T2DM. Fovea-centered OCTA (10° × 10°) with enhanced depth imaging was acquired for all of the participants using the SPECTRALIS HRA2. The en face angiograms of each capillary plexus were extracted as .tif files. Images were then binarized and the PD of each plexus within the central 2-mm region was measured using a custom MATLAB program after correction for retinal magnification. The examiner was masked to the participant groups during analysis. Poor-quality images were excluded from analysis. Numbers on the plots represent the adjusted P values. Adjusted P < 0.05 was considered statistically significant. Healthy controls: HbA1c ≤ 5.6% (n = 17); PreDM: HbA1c = 5.7%–6.4% (n = 10); T2DM: HbA1c ≥ 6.5% (n = 13).

Figure 4.

Representative measurement of PD in the intermediate capillary plexus. (A) Angiograms were extracted as .tif files from the OCTA scans and imported into a custom MATLAB program. (B) Images were binarized using the Otsu method, and PD was measured within the central 2-mm region (green ring). Non-perfused areas are black, and perfused areas are white. PD was calculated as the percentage of white pixels within the green ring to the total number of pixels within the green ring.

We observed an overall group difference in the outer plexiform layer (OPL) thickness (P = 0.03); however, we did not detect statistically different OPL thickness after multiple-comparison tests. The full retinal layer, ganglion cell layer (GCL), inner nuclear layer (INL), outer nuclear layer (ONL), inner retinal layer, outer retinal layer, and choroidal thicknesses were not statistically different among the groups (all P > 0.05) (Supplementary Table S3).

Associations Between Functional and Structural Measures

Color perimetry, mfERG, and OCT/OCTA testing were all evaluated at corresponding retinal locations within the same eye, providing complementary assessments of retinal function and structure in corresponding regions. Thresholds obtained using the red stimulus increased with delayed mfERG P1 ITs (positive correlation) for region-aggregated ring 2 hexagons (β = 0.05; 95% confidence interval [CI], 0.01–0.09; P = 0.02) for the entire cohort of participants. This was most strongly influenced by the control group (β = 0.07; 95% CI, 0.01–0.13; P = 0.03). There was no significant relationship between the chromatic thresholds obtained using the other perimetry stimuli and mfERG metrics for all of the regions analyzed (all P > 0.05). Perfusion density of the ICP within the central 2-mm region was associated with HbA1c (adjusted for age and BMI; β = −1.51; 95% CI, −2.75 to −0.27; P = 0.02) but not with chromatic thresholds obtained using any of the perimetry stimuli (all P > 0.05).

Models evaluating the relationship between chromatic thresholds and individual retinal layer and choroidal thicknesses (model 1, unadjusted; model 2, adjusted for age and BMI) were not significant for most color stimuli and structural thickness. However, thresholds obtained using the red stimulus were positively associated with increasing INL thickness (β × 100 = 1.21; 95% CI, 0.14–2.27; P = 0.03) but negatively associated with increasing choroidal thickness (β × 100 = −0.16; 95% CI, −0.28 to −0.04; P = 0.01) in model 1. This observation in the INL trended toward significance after adjusting for age and BMI (model 2: adjusted β × 100 = 0.93; 95% CI, −0.02 to 1.89; P = 0.06). Similar trends were observed with the green stimulus, reaching significance for only the choroidal thickness in the unadjusted model (model 1: β × 100 = −0.14; 95% CI, −0.25 to −0.03; P = 0.02) (Fig. 5; Supplementary Tables S4A–S4D).

Figure 5.

Forest plot of structural coefficients of retinal layers and choroidal thicknesses. Points represent β × 100 estimates and horizontal lines represent 95% CIs for the associations between structural thicknesses and the red, green, blue, and yellow stimuli thresholds, respectively. Two models are shown. Model 1 estimated the univariable β coefficient of structural thickness only; model 2 adjusted for age and BMI.

Discussion

We investigated localized retinal functional and structural alterations across age-matched healthy controls, PreDM, and T2DM groups using a multimodal approach. We report decreased PD in the retinal ICP, loss of chromatic sensitivity to red and yellow stimuli, and delayed mfERG P1 ITs in early T2DM but not in PreDM. These findings are unlikely to be influenced by hypertension, as both systolic and diastolic blood pressures were similar across groups. A reduction in retinal vessel density may indicate worsening diabetic disease, whereas an increase may reflect disease onset.^12,40 Notably, in the present study, the PD showed a consistent trend toward an increase in the PreDM group compared to controls across all of the capillary plexuses. Previous studies have described changes in metabolic autoregulation during the early stages of diabetes and DR, characterized by dilation of blood vessels and increased perfusion.^12,16,41,42 Thus, our findings extend these observations to hypothesize that this compensatory mechanism in response to T2DM may begin during the PreDM stage. Prospective longitudinal studies are required to strongly establish this observation.

The inner retinal layers, supplied by non-fenestrated retinal capillaries, are relatively hypoxic compared to the outer retinal layers, which are mainly supplied by fenestrated choroidal blood vessels.^43,44 The high metabolic demand of the retina, coupled with this oxygen gradient, increases the susceptibility of the inner retina to diabetic injury.⁴⁵ In this study, perfusion within the ICP was significantly reduced in individuals with T2DM compared to the other groups. The ICP, unlike the other capillary plexuses, is a narrow layer extending 6 to 7 mm from the fovea, a region often prone to ischemia in DR.⁴⁶ It supplies metabolically demanding neurons in the middle retina (part of the inner retinal layers),^28,46 including bipolar cells, horizontal cells, and amacrine cells located in the IPL, INL, and OPL. This makes the ICP highly sensitive to hypoxia. It has been identified as the site of most microaneurysms,^17,46,47 a key early clinical sign of diabetic retinal disease.^17,46 It is also closely associated with Müller cells, making it particularly susceptible to hyperglycemia-induced neurovascular dysfunction in early diabetes.¹⁵ This potentially explains the decreased PD of the ICP in the T2DM group in our study.

The decrease in PD suggests early involvement of the ICP and compromised blood supply to the middle retina during the early stages of T2DM. This interpretation is consistent with the observations of Huang et al.⁵ and Nesper et al.,¹² who reported that a marked decrease in the SCP and DCP vessel density is linked to more advanced stages of DR. Our findings further support independent autoregulation of the ICP and variations in perfusion among the three retinal capillary layers,^48,49 particularly in the context of diabetes. Accordingly, separate analysis of the three distinct plexuses in both clinical and research settings is necessary to better characterize disease onset, progression, and treatment response. The decrease in ICP PD in the T2DM group despite similar FAZ metrics among all the groups suggests that a decrease in PD during diabetes is not only accounted for by the FAZ but also by capillary dropout in the extrafoveal region. This aligns with the increased extrafoveal avascular area in people with diabetes that was reported in the work of You et al.²⁹

In contrast, Frizziero and colleagues¹⁵ found reduced SCP, but not ICP, vessel area density in patients with diabetes without DR. Discrepancies with our study results may arise from cohort and methodological differences. Our study included newly diagnosed T2DM (average duration of 3.4 years) participants with contemporaneous HbA1c measurements for all participants to minimize group misclassification. Their cohort included both type 1 and 2 diabetes patients with longer duration (14.5 years) and no mention of glycemic assessment at the time of examination. Thus, their results could reflect a more advanced stage of diabetes. Nonetheless, all of the above studies point to a compromise in blood supply to the inner retinal layers during diabetes.

The red and yellow stimuli of our novel color perimeter revealed reduced sensitivity (higher thresholds) in the T2DM participants compared to healthy controls. The results from these two color stimuli suggest a diffuse color vision impairment in the early stages of T2DM as opposed to the predominantly blue–yellow defects reported in other studies.^6,24 Our findings are consistent with other studies reporting both red–green and blue–yellow color vision defects in T2DM.^7,23,50 In contrast, performance on the L'Anthony D-15 test, a widely used tool for clinical assessment of color vision, did not differ among groups. This finding is consistent with prior work by Feitosa-Santana et al.,⁵⁰ who reported that the conventional D-15 procedure lacks sensitivity for detecting very mild color vision deficits in early stages of T2DM without DR. Most participants in the PreDM and T2DM groups were newly diagnosed or undiagnosed before their study visits, and only three T2DM participants had signs of DR on fundus photographs. Therefore, a more sensitive tool, such as color perimetry, was required to detect subclinical functional changes.²⁵ The decrease in PD of the ICP and the trending thickening of the OPL in the T2DM group, together with the significant association between increasing thresholds obtained from the red stimulus and increasing INL thickness in this study, suggest and point to vulnerability in the middle retina (including bipolar cells) during the early stages of T2DM.

Although the novel color perimetry detected deficits at discrete retinal locations, it did not delineate the specific retinal layers or neurons affected in early T2DM. Therefore, mfERG was performed at corresponding locations to validate the findings and to identify the affected retinal neurons. In mfERG, cones largely generate the initial negative trough (N1), and the first positive peak (P1) is primarily driven by ON-bipolar cells.^51–53 Hence, the delayed mfERG P1 IT of the averaged ring 2 hexagons in the T2DM group in this study is consistent with early dysfunction of the middle retina, including impaired ON-bipolar cell activity. This supports our findings from the OCT/OCTA measurements in this study. An earlier study in our lab similarly found mfERG IT abnormalities without amplitude changes in T2DM without retinopathy, whereas prediabetes did not differ from controls.²¹ Other researchers have also reported prolonged mfERG IT in people with T2DM with or without DR.^15,54,55

The findings indicate that, under normal retinal physiology (controls/normoglycemia), chromatic threshold testing using the red stimulus is associated with local retinal implicit time (mfERG P1 IT). The lack of a consistent within-group relationship in PreDM and T2DM suggests that functional deficits detected by one functional measure may occur independently or with different sensitivity than abnormalities detected by the other. A comprehensive functional assessment of diabetic retinal disease would ideally include both tests rather than either alone. This suggests that the two modalities may be complementary measures of retinal function rather than interchangeable in the study of diabetic retinal disease. The pathophysiological process that disrupts this relationship requires further investigation. The lack of association between the mfERG and retinal capillary metrics in the current study may reflect differences in the timing or sensitivity of neural and vascular measures during the early stages of T2DM, as reported elsewhere.^4,41 The parvocellular stimuli (red and green) appear to have promising association with differences in retinal layer and choroidal thicknesses as observed in this study. These preliminary findings require further investigation to determine the underlying nature of these associations.

We carefully considered lens-related effects on our data. Participants with lens changes greater than LOCS III grade 2 were excluded. This limited the influence of lens yellowing and cataract on the observed differences. Our participants were heterogeneous, and all of our measurements were taken under the same environmental conditions. This limits the influence of any potential environmental or technical factors on our data. The high rates of undiagnosed PreDM in this study cohort, close to the 80% reported by the American Diabetes Association and the Centers for Disease Control and Prevention,^56,57 suggest an urgent need for improved research, diagnostic tools, and health education targeting this population.

The study suggests neural and vascular alterations involving the middle retina during PreDM and early T2DM. These alterations could occur at distinct time points. However, the cross-sectional design and modest sample size of this study, especially in the PreDM group, limit causal inference regarding the temporal sequence of structural and functional retinal changes in the disease process. This limits our conclusions regarding changes in PreDM. Moreover, the incomplete spatial matching between the imaging and functional testing modalities may have attenuated, rather than exaggerated, the strength of observed associations. This suggests that the true associations may be even stronger than reported. Some participants were receiving treatment at the time of assessment. Although this could act as a potential confounding factor, it reflects real-world clinical heterogeneity rather than a source of systematic bias. This likely enhances the generalizability of our findings to typical clinical populations. Moreover, recent literature has reported that GLP-1 receptor agonists and SGLT2 inhibitors preserve retinal microvascular network and neuroretinal function.^58–60 Therefore, the use of these types of medication by some of the participants in the T2DM group in the current study could have potentially attenuated rather than exaggerated the observed outcomes. Likewise, three participants had mild DR which is an expected clinical consequence of T2DM and could influence the vascular and functional outcomes. Future prospective longitudinal studies incorporating spatially co-registered multimodal assessments in larger cohorts and accounting for treatments are warranted to confirm and extend our current findings.