Progression Biomarkers of Microvascular and Photoreceptor Changes Upon Long-Term Evaluation in Type 1 Diabetes

Abstract

Purpose: To assess demographic, metabolic, and imaging predictors influencing microvasculature and photoreceptors changes over a 4-year follow-up in type 1 diabetes mellitus (DM1).

Methods: This prospective cohort study enrolled patients with DM1 with mild non-proliferative diabetic retinopathy. Complete medical records, glycosylated hemoglobin (HbA1c), optical coherence tomography angiography, and adaptive optics were collected for the 4 years of follow-up. The main outcome measures included perfusion density at the superficial capillary plexus (SCP) and deep capillary plexus (DCP), choriocapillaris (CC) flow deficits (FDs, %), cone density, linear dispersion index (LDi), and heterogeneity packing index (HPi).

Results: The SCP presented a dichotomic perfusion trend, with increasing PD at 1 and 2 years and a subsequent decline (P < 0.001). DCP presented a similar trend in the first 2 years (P < 0.01) but not at the following time points, whereas CC FDs constantly increased over time (P < 0.01). The best-fitted model for the microvascular parameters demonstrated that the main factors affecting SCP included time (P < 0.001), duration of diabetes (P = 0.007), and HbA1c (P = 0.03), whereas the DCP was influenced by LDi modifications (P = 0.006). The LDi and HPi were mainly influenced by SCP and CC perfusion in the parafovea (P = 0.02).

Conclusions: This study demonstrated an initial vasodilatory phenomenon resulting from a compensatory mechanism from the superficial vasculature, followed by capillary dropout. Initially, it would seem that there was an adaptive response by the DCP to the needs of the photoreceptors. Although the SCP may initially support the DCP, when the microvascular damage becomes diffuse and involves the SCP and CC it directly affects photoreceptor integrity.

Diabetic retinopathy (DR) is one of the most relevant complications in type 1 diabetes mellitus (DM1) affecting age-working individuals.¹ Patients with DM1 represent a population with fewer cardiovascular, neuropathic, and nephropathic complications compared to young-onset type 2 diabetes mellitus, rendering this group a better model for studying the microvascular changes.² 

Most of the existing literature on DR has focused on vascular abnormalities, which are predominant and involved in vision loss and severe retinal complications. Although the importance of DR as a microvascular disease is undeniable, neurodegeneration is a critical aspect that requires reconsidering, especially in the early stages of the disease.^3–6 

Previous studies have postulated an association between microvascular damage and photoreceptor loss, using different methodologies to assess the photoreceptor alterations.^7–12 Our group recently investigated the direct interpolation of cone photoreceptors in response to microvascular changes taking advantage of adaptive optics (AO), which provides high-resolution in vivo cone assessment. The microvascular networks were analyzed at different anatomical levels, including the superficial capillary plexus (SCP), deep capillary plexus (DCP), and choriocapillaris (CC), which were studied through swept-source optical coherence tomography angiography (OCTA).¹³ Although this pilot study provided exciting insights into the structural–microvascular interconnections in mild non-proliferative diabetic retinopathy (NPDR), our preliminary findings investigated the population only cross-sectionally. Therefore, the present study represents a refinement and an evolution of our preliminary findings by investigating the longitudinal modifications of retinal microvasculature and cone metrics in eyes with mild NPDR from DM1. This study aimed to identify potential demographic, metabolic, and imaging predictors influencing microvascular impairment and structural alterations of photoreceptors over a 4-year follow-up. 

Methods

Study Participants

This prospective observational study enrolled patients with DM1 with mild NPDR evaluated at the IRCSS-Bietti Foundation between September 2017 and May 2022. The study adhered to the tenets of the Declaration of Helsinki and was approved by the institutional review board of the IRCCS-Fondazione Bietti (RET03-22/12/2016). All patients gave their written consent before being included in the study. 

Patients with DM1 with mild NPDR, a minimum age of 18, and a best-corrected visual acuity of 20/20 were consecutively enrolled. Complete medical records, glycosylated hemoglobin (HbA1c), and multimodal imaging, including color fundus photographs, optical coherence tomography (OCT), OCTA, and AO, were collected over a 4-year follow-up. Patients with DM1 were enrolled if they had adequate glycometabolic control and no concomitant systemic comorbidities as determined by our referral diabetologists (FP, SF). All of the patients included in the study utilized insulin pumps. The diagnosis of mild NPDR was made by two experienced examiners (MP, EC) according to the modified Early Treatment of Diabetic Retinopathy Study (ETDRS) severity scale, considering both clinical and instrumental examinations at baseline.¹⁴ 

Patients were excluded if they had other ophthalmic comorbidities, intraocular or vitreoretinal surgery, axial length greater than 26 mm and/or a spherical error > 6 diopters, any cystic changes or macular edema, any dioptric media opacities, or poor-quality images. The signal strength index (SSI) was set at 8/10 as a cut-off value for optimal OCTA quality and the absence of motion artifacts. 

Multimodal Imaging Modalities

The acquisition of color fundus photographs was performed at baseline using a Topcon TRC-50DX (Topcon Healthcare, Tokyo, Japan). Spectral-domain OCT was acquired using the SPECTRALIS HRA+OCT (software version 1.10.2.0; Heidelberg Engineering, Heidelberg, Germany). The minimum scanning protocol consisted of a 20° × 20° square centered on the fovea with 25 B-scans. Swept-source OCTA using the PLEX Elite 9000 (software version 1.7.027959; Carl Zeiss Meditec, Jena, Germany) was acquired with a 3-mm-volume cube using active eye tracking with projection artifact removal enabled during image analysis. The PLEX Elite 9000 built-in software automatically segmented the SCP, DCP, and CC slabs. The segmentation boundaries were checked by a single operator (SF); in case of misalignments or segmentation errors, the boundaries were manually readjusted and the images reprocessed. 

An AO retinal camera (rtx1; Imagine Eyes, Orsay, France) was used to acquire cone mosaic images. For optimal image acquisition, the pupils were dilated using 1% tropicamide. AO images were acquired at different locations, covering an area of 5° × 4° centered on the preferred locus of fixation (coordinates x = 0° and y = 0° as the foveal reference point). A sequence of 40 frames with a rate of 9.5 frames per second was obtained at each retinal location. The images were corrected for distortions and artifacts but also registered and frame-averaged to produce a final image with an enhanced signal-to-noise ratio using a proprietary program. The images were then registered, aligned, and montaged using the montage tool provided by i2k Retinal Pro (DualAlign, Clifton Park, NY, USA). Cone labeling and identification were performed using an enhanced algorithm for imaging processing toolbox in MATLAB (MathWorks, Natick, MA, USA), as previously described.^13,15 

Image Processing

Outcome measures were collected at baseline and every year until the 4-year follow-up, when available. OCTA images obtained at each follow-up, including SCP, DCP, and CC slabs, were exported using a .jpg file format with a dimension of 1024 × 1024 pixels for further analysis. Perfusion density (PD) and CC flow deficits (FDs, %) were calculated using ImageJ (version 2.3.0/1.53f; National Institutes of Health, Bethesda, MD, USA) and the previously described procedure.¹³ The OCTA slabs were aligned and registered using the Landmark Correspondences method. A customized ETDRS grid was overlaid on the registered images to perform precise cropping for the quantitative analysis. The customized grid consisted of six different regions of interest (ROIs) created and saved in Fiji software. Two circular areas were delineated, the larger delimiting the central 2.25-mm diameter and a concentric smaller circle of 1 mm diameter, centered on the fovea. Starting from these two concentric rings, the remaining four ROIs depicted the parafoveal subfields (superior, nasal, inferior, temporal), based on the original ETDRS ring centered on the fovea. Each parafoveal sector corresponded to an area of approximately 0.78 mm² and was processed separately, as detailed in Figure 1. The entire parafovea was calculated by averaging the four parafoveal sectors for all the variables considered. 

Figure 1.

 

Image processing. (A) SCP slab. (B) The vascular slab was overlaid with the ETDRS ring and cropped. (C) After removing the 1-mm central annulus, the remaining parafoveal sectors were cropped for processing. (D) The background was removed to avoid false overestimation of perfusion given by the white background. (E) The resulting image was binarized using the mean threshold before calculating the perfusion density.

The PD was calculated as a percent for the different parafoveal sectors of the SCP and DCP slabs after binarizing, following the previously validated procedures.^16–18 The CC FD% was calculated based on the four different ROIs representing the parafoveal sectors after being imported into Fiji software for binarization using auto local thresholding with the Phansalkar method. The binarized ROIs were processed using the Analyze Particles tool that automatically counts the flow deficits, as previously described.^9,19 

The same customized ETDRS grid was applied to the AO montage (Fig. 2). The four parafoveal regions were cropped using Fiji software, exported as single .tiff files on a black background, and then processed using the MATLAB algorithm, as previously detailed in our pilot study.¹³ The main AO outcome measures included (1) cone density (CD), calculated as number of cones per square millimeter (cones/mm²); (2) linear dispersion index (LDi), which measures the distance between neighboring cones, or cone spacing; and (3) heterogeneity packing index (HPi, %), which describes alterations of spatial distribution patterns of the cones, or cone arrangement. This index is the fractional increase in four-sided and eight-sided non-hexagonal Voronoi tiles for six-sided cells: 6n – (8n + 4n).¹⁵ 

Figure 2.

 

Image processing for AO. (A) AO images were registered, aligned, and montaged using the montage tool i2k Retina Pro (DualAlign, Clifton Park, NY, USA). The area cropped for the processing corresponded to a 2.25-mm square (yellow). (B) The customized ETDRS ring, consisting of two concentric circles measuring 1 mm and 2.25 mm, was then applied to the cropped image. The four parafoveal sectors (superior, inferior, nasal, and temporal) were traced, and each measured 0.78 mm². (C) Each parafoveal sector was then cropped using Fiji software and processed separately using a MATLAB algorithm to measure CD, LDi, and HPi.

Statistical Analysis

Based on the results from the literature,⁶ sample size calculations were performed considering SCP vessel density parameter changes in the placebo group over the 3 years of follow-up (baseline value, 31.20 ± 4.56; 3-year value, 26.87 ± 7.55), at a significance level of 5% and a power of 80% (paired t-test, two tails). A sample size of at least 21 cases was required. The normality of distribution was assessed with the Shapiro–Wilk test. Data were reported as mean ± standard deviation (SD) or median and interquartile range after verifying the normality of distribution. Repeated-measures analysis of variance was used to assess paired differences in the study group. Post hoc comparisons were carried out using Tukey’s honestly significant difference test for multiple comparisons. A linear mixed model analysis was performed to deal with missing data and random effects from subject-specific effects following a previously reported model.²⁰ The data are structured considering each individual and time point (year); participants were dropped from the model only for any specific time points that were missing and not for all of the time points. Different mixed models were created considering the time and the different variables individually for every parafoveal subfield. Age, HbA1c, diabetes duration, CD, LDi, CD, SCP, DCP, and CC alone or combined were considered fixed effects. The different mixed models for all of the parameters were compared using the Bayesian information criterion (BIC), and the model that minimized it was chosen as the best fitting. The P value threshold was set at 0.05 (two-sided) for all of the analyses. Statistical analysis was performed using SPSS Statistics 25 (IBM, Chicago, IL, USA) and RStudio-2022.07.1 for R. 

Results

Of the 54 patients initially considered (54 eyes), 32 patients (32 eyes) were excluded for a lack of OCTA examinations, low-resolution imaging (SSI < 8), and/or motion artifacts (24 eyes). Eight eyes were excluded for the absence of an AO examination matching the OCTA follow-up time points, the presence of motion artifacts, or poor image quality. A total of 22 patients (22 eyes) with mild NPDR were considered eligible for the study. The demographic features of the participants are summarized in Table 1. 

Table 1.

 

Baseline Characteristics of the Cohort With Mild NPDR

 

Baseline Topographical Distribution of Microvascular and AO Metrics Per Sector

The topographical distribution of OCTA and AO parameters at baseline is summarized in Figure 3. SCP and CC topographical analyses revealed no significant differences among the sectors analyzed (P > 0.05, both), whereas the DCP parafoveal sectors differed significantly, F(3, 84) = 4.29, P = 0.007. Post hoc evaluation revealed a relevant difference in PDs between the superior and temporal sectors (−0.06; 95% confidence interval [CI], −0.10 to −0.02; P = 0.003). AO analysis demonstrated significant differences between the CD, F(3, 84) = 5.85, P = 0.001, and LDi, F(3, 84) = 3.02, P = 0.03, parafoveal sectors. Post hoc analysis showed that the inferior parafoveal sector presented the greatest CD compared to others (superior, P = 0.002; nasal, P = 0.03; temporal, P = 0.003), whereas the LDi differed significantly between superior and temporal sectors (P = 0.02). 

Longitudinal Changes in Microvascular Parameters

The means and 95% CIs of the longitudinal variations with respect to baseline in the SCP and DCP at each parafoveal location are summarized in Table 2. A dichotomic trend can be observed in the SCP, with an increasing PD (%) at 1 and 2 years compared to baseline and a subsequent decline of PD at 3 and 4 years. The same trend is evident in the average parafovea and among the single parafoveal sectors analyzed. When observing the DCP longitudinal variations, a similar PD (%) increase was observable in the first 2 years (Table 2), but with no evidence of a univocal trend at the following time points for the entire parafovea and all of the sectors analyzed. Figure 4 displays the average PDs for both the SCP and DCP from baseline to the 4-year follow-up at each parafoveal sector. CC FDs (%) tended to continuously increase at the different time points with the only exception being year 2, where a non-significant deflection was observable (P > 0.05 for the average parafovea and all of the parafoveal sectors). Moreover, between years 3 and 4, a plateau is evident in the linear graph shown in Figure 4, confirmed by the non-significant variations in all of the sectors analyzed (superior, P = 0.79; inferior, P = 0.85; nasal, P = 0.99; temporal, P = 0.95). For further details, see Table 3 for the means and 95% CIs of the CC FD (%) variations, as well as Figure 4. 

Table 2.

 

Mean and 95% CIs for Longitudinal Microvascular Variations Among the Parafoveal Locations As Estimated From the Linear Mixed Model

 

Figure 3.

 

Box plots displaying the distribution of microvascular and photoreceptor parameters at baseline in the different parafoveal sectors.

Figure 4.

 

Linear graphs showing the 4-year longitudinal changes in the SCP, DCP, and CC FDs.

Table 3.

 

Mean and 95% CIs of Longitudinal Variations in the CC Among the Parafoveal Locations As Estimated From the Linear Mixed Model

 

Cone Mosaic Variations on Longitudinal Analysis

Based on the same analysis method, the longitudinal variations for CD were not significant for all of the parafoveal sectors analyzed (superior, P = 0.31; nasal, P = 0.15; inferior, P = 0.74; temporal, P = 0.86). The longitudinal variations in the HPi affected only the parafoveal superior sector at 4 years, with an average difference of −3.5% (95% CI, −6.6 to −0.3; P = 0.03) compared to baseline, and the inferior parafovea at 4 years with a difference of −1.9 (95% CI, −3.3 to −0.7; P = 0.006). Regarding LDi, the most relevant variations involved the nasal parafovea at 1 year with an average difference of 0.003 (95% CI, 0.000–0.005; P = 0.02) compared to baseline, and the inferior parafovea at 4 years with an average of 0.003 (95% CI, 0.001–0.005; P = 0.01). For further details, see Supplementary File S1. 

Factors Influencing the Microvascular and Morphological Modifications Over Time

The best-fitted model for the microvascular parameters included a combination of age, duration of diabetes, HbA1c, and LDi, all considered at baseline and at each time point until the 4-year follow-up. Table 4 summarizes the main effects and P values for each retinal location. The main factors influencing the SCP included time and diabetes duration for all of the sectors evaluated and HbA1c, except for the nasal parafoveal sector (P = 0.45). In the DCP, LDi and time significantly affected all of the locations analyzed except the temporal parafovea, whereas age influenced the superior parafovea. Regarding CC FDs (%), the main parameter is represented by time in superior parafovea diabetes duration, and HbA1c had a significant effect (P = 0.02, both), which was not replicable in the remaining sectors. 

Table 4.

 

P Values for Fixed Effects and Time for the Best Model According to the BIC At Each Retinal Location

 

The model for cone metrics included age, time, diabetes duration, HbA1c, SCP PD, DCP PD, and CC FDs. Table 5 summarizes the main effects and P values for each retinal location. CD appeared to be poorly influenced by the variables included in the model, except for the temporal sector, which presented significant influences with age, diabetes duration, and HbA1c. Moreover, the duration of diabetes influenced the parafoveal and inferior sectors (P = 0.01, both). Many parameters affected both HPi and LDi with similar topographic distributions, among which the microvascular features add the most interesting insights. As shown in Table 5, the HPi was influenced by the modifications of SCP PD for all of the sectors, excluding the nasal and temporal parafovea, which were excluded from the model for those locations based on the BIC values. Moreover, HPi was also influenced by CC FDs (%) in all of the sectors except for the nasal parafovea. Similar findings were found for the LDi, where SCP PD influenced all of the locations, except the nasal and temporal parafovea, and the CC FDs (%) in all of the sectors, except the superior and nasal parafovea. 

Table 5.

 

P Values for Fixed Effects and Time for the Best Model According to the BIC At Each Retinal Location

 

Discussion

The present study investigated the longitudinal metabolic, microvascular, and photoreceptor modifications in DM1 eyes with mild NPDR. The baseline topographical analysis demonstrated a reduced PD in the DCP, particularly affecting the temporal parafoveal sector. Perfusion alterations in the DCP corroborated the predominant capillary nonperfusion at this anatomical level that is evident at the early stages of DR, as previously postulated.^21–24 With regard to the longitudinal changes, our study revealed an interesting perfusion trend in mild NPDR eyes, particularly evident in the SCP. At this level, PD persistently increased in the first 2 years, at which point it reached its highest point and then tended to decrease markedly in years 3 and 4. Likewise the DCP showed an increasing PD during the first 2 years, but at the following time points the changes were not significant. 

Our findings support initial compensatory vasodilation in response to hypoxia, which was sustained for a relatively long interval of 2 years. This effect was notable in all of the parafoveal sectors considered in the SCPs and DCPs, further corroborating its relevance. In experimental models of DR, increased arterial capillary perfusion represents the initial perfusion change.^25,26 These findings were explained by hypothesizing that modest reductions in perfusion can lead to compensatory vasodilation as a part of an autoregulatory capacity able to reduce the flow resistance and increase the blood transit into the tissues.^26,27 The autoregulatory capacity of the SCP can also lead to apparent preservation of the blood flow limiting the progression of the microvascular damage over time.²⁸ 

Similar to our findings, Onishi et al.²⁹ found increasing flow in the SCP, measured through an approximating index (adjusted flow index), in eyes with DM without DR. The authors hypothesized that the blood flow modification in the SCP could result from ischemic changes occurring in the middle and DCP. This discrepancy can also be explained by the arrangement of microvascular plexuses. As postulated, DCP is the most susceptible to hypoxic conditions due to a possible predominant in-series vascular arrangement, where the DCP is the central venous outflow system. Under an increased arteriovenous difference in oxygen saturation, the oxygen extraction fraction becomes maximal at the SCP, leading to exaggerated hypoxia in the distal capillaries, a phenomenon known as “misery perfusion.”^27,30 Therefore, the reduced DCP perfusion already seen at baseline implies that this layer can suffer from basal hypoperfusion, attenuating the differences seen at 3 and 4 years compared to baseline. Accordingly, a misery perfusion may be precociously observable in the DCP, and thus a compensatory increase in the superficial blood flow can temporarily affect the DCP. When the microvascular response moves toward the ischemic condition, with loss of capillaries in the SCP, the DCP is no longer dilated, and the capillary dropout becomes evident again. It is essential to investigate this hypothesis further with functional correlates that may be altered precociously in response to hypoperfusion.^6,31,32 Of note, the SCP was also influenced by other baseline factors in our model, such as diabetes duration and HbA1c, further confirming the microvascular damage resulting from the underlying diabetic disease. 

Our in vivo model added relevant insights by interpolating microvascular and cellular components. Among the factors influencing DCP perfusion, our results point toward the LDi, representing an index measuring the distance between neighboring cones. Considering our findings in light of the existing evidence, the DCP appears to modify its perfusion according to photoreceptor structural changes. In experimental models, the DCP supplies the photoreceptors with around 10% to 15% of the oxygen requirements.³³ But, more importantly, during hypoxic conditions, the inner retina microvasculature can compensate for the deficient oxygen supply from the choroidal circulation.^8,34 The association between capillary non-perfusion at the DCP level and photoreceptor disruption observed on OCT⁸ was further reinforced by the colocalization with areas of reduced retinal sensitivity, serving as a surrogate for photoreceptor viability.³⁵ Considering the influence of the photoreceptors arrangement over the DCP, it may also be hypothesized that the perfusion trend seen for the DCP at 3 and 4 years can be attenuated by photoreceptor remodeling, increasing the need for oxygen. 

More interestingly, the cone metrics representing the cone spacing (LDi) and arrangement (HPi) were influenced by both the SCP and CC perfusion that progressively deteriorated over time (Table 5). The involvement of the SCP can be partially explained by the evidence previously discussed, starting with the autoregulatory capacity of this vascular layer in response to ischemic stimuli from the deeper vascular layers.^22,23,28 Therefore, the SCP may initially sustain the hypoxic changes from the DCP and choriocapillaris with a compensatory vasodilation supporting the photoreceptors. Still, when the microvascular impairment of the SCP becomes evident, likely reducing the support to the deeper microvasculature, the photoreceptors may suffer, leading to a structural rearrangement. 

Capillary dropout of the CC was previously demonstrated in eyes with no signs of DR or with mild DR on histopathological specimens, where the evidence of a beaded morphology with focal dilations and narrowing was observed.³⁶ Hidayat and Fine³⁷ analyzed the choriocapillaris in diabetic eyes affected exclusively by DM1 with the aim of avoiding interference with arteriosclerotic or senile changes. The authors demonstrated a basement membrane thickening with luminal narrowing affecting the CC and small choroidal blood vessels. Such basement membrane thickening mainly affects the outer (scleral) side of the CC, where the pericytes are typically present. Furthermore, when oxygen pressure falls, the choroid cannot compensate and vasodilate,²⁶ explaining our results showing a progressive decline in CC perfusion with time (Fig. 3). A correlation between cone parameters and CC FDs was found in our previous cross-sectional evaluation and by others who have considered ellipsoid zone reflectivity to be a surrogate for photoreceptor integrity.^9,10,13,38 In contrast, a recent cross-sectional study demonstrated no direct association between HPi and CC FDs, but the analysis was not comparable to our methodology, and the subjects enrolled had type 2 DM.³⁹ The topographical distribution of cone metric variations over time is interesting. As noted, the temporal parafovea presented the most significant associations, including diabetes duration, HbA1c, and CC FDs. These findings may confirm the temporal predilection for diabetic-driven damage in NPDR.^13,40 

Strengths of the present study are its long follow-up time, the customized topographical assessment based on our previous pilot study,¹³ the interpolation between microvascular and structural modifications at a photoreceptor level, and the inclusion of patients with DM1 without systemic comorbidities. Nevertheless, the study has significant limitations that should be considered. First is the sample size and the loss of follow-up or missing data, especially in light of the fact that the patients were followed during the COVID pandemic. Also, the acquisition of both OCTA and AO images is strongly dependent on patients’ cooperation and intrinsic factors that may affect image quality and resolution. 

In conclusion, the present study demonstrated characteristic patterns of microvascular rearrangement in eyes with mild NPDR secondary to type 1 DM. A vasodilatory phenomenon was initially predominant, suggesting a compensatory mechanism in the superficial and deep vasculature. In the second phase (years 3 and 4), reduced perfusion with capillary loss was evident in the SCP but not the DCP. At the same time, the DCP appeared to have an early involvement at baseline, providing additional vascular support to the photoreceptors, as demonstrated by the influence of the LDi on perfusion. These findings may justify the irrelevant microvascular variations from year 3. One possibility is that pre-existing DCP damage at baseline may have attenuated changes when the vasodilatory phenomenon ceased, or the DCP may even have adapted to the needs of the photoreceptors. Nevertheless, it is reasonable to suggest that photoreceptors are threatened by microvascular damage progression that involves both the SCP and CC. Further studies with larger sample sizes and functional assessments of the cellular response would be desirable. 

Acknowledgments

Supported by the Italian Ministry of Health and Fondazione Roma, Italy. 

Disclosure: S. Fragiotta, None; E. Costanzo, None; F. Picconi, None; P. Giorno, None; D. De Geronimo, None; D. Giannini, None; S. Frontoni, None; M. Varano, Allergan (F), Bayer (F), Novartis (F), SIFI (F); M. Parravano, Alfaintes (F), Allergan (F), Bayer (F), Novartis (F), Omikron (F), Roche (F), Zeiss (F)