Introduction

Clinical diagnosis and staging of diabetic retinopathy (DR) is based on the characteristic microvascular changes in the retina observed by biomicroscopy, fundus photography and fluorescein angiography. However, more advanced techniques in ocular imaging also enable to study subtle changes in retinal microanatomy as well as perfusion alterations in vivo, that otherwise remain clinically unnoticed. Degenerative changes in both the neural and the vascular components of the retina have been demonstrated as early features in experimental diabetes and it is often debated which one is affected first in DR [1].

An increasing number of clinical studies using optical coherence tomography (OCT) indicates that changes in DR include the degeneration of neuroretinal tissue, which has been described as diabetic neuroretinopathy [2,3,4,5,6]. This degeneration is seen mostly in the inner retinal layers of the perifoveolar region [6,7,8,9,10,11], where these layers are thickest, and it worsens with advancing stages of DR [12, 13]. In line with these more recent findings, multiple studies have demonstrated subclinical impairment of neuroretinal function in patients with early DR, and on a smaller scale also in diabetes patients without DR [14,15,16,17,18,19,20,21,22]. In addition, reduced capillary density in the fovea can be observed early on using fluorescein angiography or OCT angiography [23, 24]. It worsens in correlation with inner retinal layer thinning during progression of DR [13, 25,26,27], and leads to reduced capillary blood flow in neuroretinal tissue [28, 29].

Blood flow in the retina is locally controlled by autoregulation, which enables a constant blood supply during changes in perfusion pressure [30]. In addition, when neuronal cells in the retina are activated, their metabolic and blood demand increase. This is compensated by a regulatory increase in blood flow, a response termed neurovascular coupling or functional hyperemia [31, 32]. This response is mediated by signaling through glial cells that are in direct contact to neurons and vessel walls. Due to the tight interaction of neurons, glia and pericytes in the regulation of neurovascular coupling, these three cell types can collectively be termed as the neurovascular complex. Accordingly, both vascular and neuronal alterations in the diabetic retina could play a part in the disturbed blood flow response to retinal stimulation that has been observed in many studies [33]. Stimuli suitable for assessing neurovascular coupling in the retina include transient exposure to flickering light or the transition from dark-adapted state to light. Both stimuli are known to increase blood flow and vessel diameters in the healthy retina, which can be measured non-invasively using various imaging devices [33].

Hyperglycemia significantly affects vascular reactivity to retinal stimulation already in healthy eyes [34, 35], and impaired neurovascular coupling is also among the earliest detectable changes in the retina of diabetes patients [36,37,38]. Interestingly, this alteration has been found associated with but also independent from retinal ganglion cell dysfunction [39, 40]. It has also been shown that neurovascular coupling deteriorates further with increasing stages of DR [37, 41, 42]. However, the influence of neurodegenerative changes in this context is not known.

To our knowledge, direct interactions between retinal layer structure and the disturbed neurovascular coupling in DR have not been investigated so far. Therefore, we aimed to examine associations of the functional hyperemic response in the retina during flicker light stimulation with the thickness of the individual neuroretinal layers in patients with early non-proliferative DR, in whom retinal layer thinning might be expected.

Materials and methods

The study was performed at the Department of Ophthalmology and the Department of Clinical Pharmacology of the Medical University of Vienna in adherence to the tenets of the Declaration of Helsinki and Good Clinical Practice guidelines. The study protocol and its design was approved by the institutional Ethics Committee. All participants signed written informed consent before the study.

Study participants

Thirty adult patients with type 1 diabetes featuring mild (n = 15) or moderate (n = 15) non-proliferative DR and 14 healthy subjects registered as probands at the Department of Clinical Pharmacology and matched for sex and age were included. The sample size of this study was specified to detect differences in vessel diameters and vessel reactivity between groups, as observed in previous experiments [36, 39, 40]. A sample size of 30 patients was estimated sufficient for exploratory correlation analysis. All participants had a screening examination including detailed medical history, a complete ophthalmologic examination, and assessment of best-corrected visual acuity (BCVA) using standardized logarithmic visual acuity charts (“ETDRS” Charts, Precision Vision, La Salle, IL). Resting blood pressure was measured and venous blood samples were taken to confirm normal blood pressure, blood count and glucose levels in healthy individuals. Hemoglobin A1c (HbA1c) was quantified in diabetes patients but not in healthy controls. DR was classified according to the criteria set out in the Early Treatment Diabetic Retinopathy Study (ETDRS) [43]. Patients without DR or with more advanced stages than moderate DR, patients after a previous treatment with intravitreal injections or laser and patients with macular oedema were not included. Further exclusion criteria for all participants were or consumption of illicit drugs or nicotine products including tobacco smoking, a body mass index above 30 kg/m², other eye diseases, anti-inflammatory medication in the past 3 weeks, and refractive errors of more than 6 diopters. If both eyes had DR, the eye showing more signs of DR was chosen as the study eye. In diabetes patients capillary blood glucose was measured during the study day. All participants had to refrain from consuming alcohol or caffeine during 12 h before the study.

Dynamic vessel analyzer

Study eyes were instilled with tropicamide for pupil dilation. After a resting period of 15 min, we measured the diameters of one major temporal retinal artery and vein supplying the macular area, at a location up to two disc diameters from the optic disc margin, using the Dynamic Vessel Analyzer (DVA, IMEDOS GmbH, Jena, Germany). The DVA is described in detail elsewhere [44]. In short, it enables continuous real-time analysis of retinal vessel diameters from digitized images. Baseline measurements were performed for 60 s. This was followed by 60 s of stimulation with full field flickering light at a frequency of 12.5 Hz, during which the response of vessel diameters could be assessed. Flicker light-induced vasodilation was defined as the ratio of the average diameter values during the last 20 s of the stimulation period over the average values during the whole baseline recording, and was expressed as % change over baseline. Measurement accuracy of the DVA is higher in retinal veins [44]. Therefore, the venous response was chosen as the primary outcome variable of vessel reactivity.

Optical coherence tomography

Spectral domain OCT (Spectralis OCT, Heidelberg Engineering, Heidelberg, Germany) was performed using a 20 × 20 degree scan pattern of 49 horizontal B-scans centered on the macula. Technical properties of this OCT system are detailed elsewhere [45]. Activated eye-tracking with averaging of 20 frames per B-scan was used to minimize motion artefacts and background noise, and to enhance scan contrast [46]. The built-in segmentation software (HRA viewing module version 6.7.17; Heidelberg Engineering) was used for automated segmentation of individual retinal layers. All B-scans were checked for segmentation errors by the same investigator (BP), and obvious errors were manually corrected. Thus, three-dimensional maps of retinal nerve fiber layer, ganglion cell layer (GCL), inner plexiform layer (IPL), inner nuclear layer, outer plexiform layer, and outer nuclear layer were obtained. A circular ETDRS grid was centered on the foveola, and the average layer thicknesses in the ring areas measuring 1–3 mm, and 3–6 mm around the center were used for quantitative analysis. The inner ring area was chosen as primary outcome variable of retinal layer thickness, because it contains the maximum thickness of GCL and IPL [9].

Statistical analysis

Continuous variables were expressed as means ± standard deviation (SD). Normal distribution was assessed by the Shapiro-Wilk test and accepted at p > 0.05. Unpaired t-tests were used to compare normally distributed data between groups. Diameter changes in retinal veins were assessed by analysis of covariance (ANCOVA) using baseline vessel diameters as a co-variable. Categorical variables and non-normally distributed data were compared using the Mann-Whitney U test. Univariate analyses with Pearson correlations were calculated to assess correlations of flicker light response or retinal layer thickness with other variables. In addition, multiple regression models with flicker light response or retinal layer thickness as dependent variables were performed. A p-value of 0.05 was considered as the level of significance for all calculations. The Statistica software (Release 6.1; StatSoft Inc., Tulsa, OK) was used for all statistical analyses.

Results

Subject characteristics and study parameters

Table 1 displays the characteristics and measurement results in the groups of investigated patients and healthy controls. The primary outcome variables were normally distributed in all groups.

Table 1 Summarized data of healthy control subjects and diabetes patients, and of patient subgroups
Full size table

Age, sex, height, weight, visual acuity and vessel diameters were not significantly different between patients and healthy controls. In addition, all parameters in the two patient subgroups with mild or moderate DR did not differ statistically. Besides daily insulin, 20 patients had additional regular blood pressure and/or lipid-lowering medication including angiotensin-converting-enzyme inhibitors, angiotensin II receptor blockers, beta blockers and statins.

Vessel diameter changes

Vessel dilation in response to stimulation with flickering light was significantly reduced in the diabetes patients compared to healthy controls in retinal veins (3.0%±2.3% vs. 6.1%±3.1%, p < 0.001; Fig. 1A) as well as in retinal arteries (1.3%±2.3% vs. 5.1%±5.2%, p = 0.005). The flicker light response was neither significantly different in patients with mild or moderate DR (Table 1) nor in patients with or without co-medication (veins: 2.7%±2.1% vs. 3.6%±2.6%, p = 0.35; arteries: 1.4%±2.5% vs. 1.0%±1.8%, p = 0.69).

Fig. 1
figure 1

Box-and-whisker plots presenting retinal vessel dilation in response to stimulation with flickering light (A) and macular ganglion cell layer thickness (B) in patients with diabetic retinopathy (DR) and in healthy controls. Data are presented as median (bar), interquartile range (box), non-outlier minimum and maximum (whiskers) and outliers (dots)

Full size image

DVA and OCT measurements in two exemplary diabetes patients are shown in Fig. 2.

Fig. 2
figure 2

Vessel diameters during the flicker light experiment using the DVA, and macular ganglion cell layer thickness heatmaps and data from OCT in two diabetes patients without (A) and with (B) diabetic neuroretinopathy. Blue and red curves represent venous and arterial diameters, respectively; yellow marks indicate beginning and end of the stimulation period. The thickness heatmap is overlaid with a 6 mm ETDRS grid [43]. Average thickness in the individual sectors is shown in the adjacent diagram

Full size image

Correlation analyses

Table 2 summarizes the results of univariate correlation analyses between the venous flicker light response and individual retinal layer thickness in diabetes patients.

Table 2 Summary of univariate correlation analyses between venous flicker light response and individual retinal layer thickness as measured by OCT in diabetes patients
Full size table

In diabetes patients only, univariate analysis showed a significant correlation of the diminished flicker light response in retinal veins with GCL thickness (r = 0.46, p = 0.010; Fig. 3A) and to a lesser degree with IPL thickness (r = 0.40, p = 0.027), whereas all other retinal layer thickness values did not correlate. In addition, the venous flicker light response correlated negatively with HbA1c levels (r=-0.41, p = 0.023; Fig. 3B) and with age (r=-0.38, p = 0.037; Fig. 3C), but not with baseline vessel diameter (r = 0.08, p = 0.69), glucose level (r = 0.00, p = 0.98), or diabetes duration (r=-0.27, p = 0.15). In a multiple regression model including GCL thickness, age, disease duration, baseline venous diameter, blood glucose, HbA1c and arterial hypertension as predictor variables, only GCL thickness (p = 0.017, β = 0.42) and HbA1c (p = 0.045, β=-0.35) remained significantly associated with the vascular flicker light response.

Fig. 3
figure 3

Scatterplots with correlation analyses of the flicker light response in retinal veins with ganglion cell layer (GCL) thickness (A), hemoglobin A1c (HbA1c) levels (B) and age (C) in diabetes patients. Straight lines show linear fits, and dotted lines show 0.95 confidence intervals

Full size image

Mean GCL thickness was slightly but not significantly lower in diabetes patients, more obvious in patients with moderate DR (Table 1). Figure 1B illustrates a more negative skew of GCL thickness values in eyes with moderate DR. GCL thickness did neither show significant correlations with age, diabetes duration or HbA1c in univariate analysis nor in a multivariate model. Including DR stage as an additional categorical factor also showed no correlations.

In healthy control subjects no significant association between flicker light-induced vasodilation with any other variable, including GCL thickness (r = 0.28, p = 0.34) and age (r=-0.20, p = 0.50), was detected. Interaction analysis of vessel reactivity in the combined groups showed a statistically significant interaction term of group and GCL thickness (p = 0.036) and an increase in r² (0.36 vs. 0.13), which indicates an effect of GCL thickness on vasodilation in diabetes patients.

Discussion

Our results demonstrate an association of the impaired flicker light-induced hyperemic response in eyes with early DR with the gradual reduction of inner retinal layer thickness, most prominently in the macular GCL. Not all patients were equally affected by GCL thinning. Hence, GCL thickness was not significantly different between the study groups. However, its significant correlation with retinal vessel reactivity in the diabetes patients indicates a neurodegenerative component in the worsening of neurovascular coupling in eyes with DR.

The results are in good agreement with previous OCT studies in diabetes that found subtle atrophic changes primarily in the inner retinal layers [3,4,5,6,7,8,9,10,11,12, 47]. Recent longitudinal observations have shown a progressive GCL thinning already before clinical emergence of DR [13, 48, 49]. Long-term follow-up revealed a mean parafoveal GCL + IPL loss of 0.29 to 0.46 μm per year, independent of HbA1c, age, and sex [6, 48]. Apoptosis of neuroretinal cells and GCL reduction are early and persistent features in diabetes [1, 6, 50,51,52,53]. However, since the GCL contains retinal ganglion cells, Müller cells, astrocytes, microglia, displaced amacrine cells, and capillaries of the superficial retinal vascular plexus [11], layer thinning may not only involve ganglion cell degeneration but also glial and microvascular loss.

Obviously, not all patients are affected by diabetic neuroretinopathy to a similar extent. The reason for this is not clear. Despite a relatively long disease duration in our cohort we detected no association of GCL thinning with disease duration or medium-term glycemic control. In addition, there was no significant difference in retinal layer thickness between the two subgroups of non-proliferative DR. Patients with moderate DR tended to have longer disease duration and lower GCL thickness. However, a considerable variability of these parameters is noted in both patient subgroups. The sample size may have been too small to detect retinal layer thinning on a group level as previously shown by others.

It has been speculated that retinal neurodegeneration in diabetes may play a role in the formation of early microvascular changes in DR, including breakdown of the blood-retina barrier, vasoregression and impairment of neurovascular coupling [4]. However, the causal relationship of neuroretinal degeneration and diabetic microangiopathy has not been clarified. Subclinical dysregulation and changes in the microcirculation could also induce or enhance neurodegenerative processes via an increased vulnerability to damage from metabolic injury.

In line with our results, multiple studies in diabetes patients have shown a reduction of vessel reactivity to flickering light, with comparable values in mild and moderate DR [36, 37, 39, 41], indicating abnormal neurovascular coupling. This impairment can often be observed already before the first visible signs of DR [33, 40, 54], and to a lesser extent in pre-diabetic patients [55] or in healthy subjects during hyperglycemia [34]. A large prospective cohort study in patients with type 2 diabetes found that reduced vasodilation during flicker light stimulation is an independent risk factor for DR incidence and progression [56], and several cross-sectional studies showed that the flicker light-induced vasodilation of retinal vessels further decreases with increasing stages of DR [37, 41, 42]. Hence, neurovascular coupling may deteriorate with advancement of the disease. Our results indicate that neurodegeneration may be involved in this worsening.

Similar to findings in human patients, flicker light-induced vasodilation is early reduced in animal models of diabetes [57, 58]. In diabetic rats upregulation of inducible nitric oxide (NO) synthase (iNOS) increased retinal levels of NO [57], which. have also shown neurotoxic effects on retinal ganglion cells [59]. Inhibition of iNOS restored impaired neurovascular coupling in an animal model [57], and also improved the flicker light response in retinal veins in diabetes patients without retinopathy [60]. Several medical treatments have shown potential in improvement of neurovascular coupling [61,62,63,64]. However, following from our results the chance to restore this impairment may be higher if neurodegeneration has not yet occurred.

It is long known, that ocular stimulation with flickering light results in a reactive increase of retinal vessel diameters and retinal blood flow [65, 66]. Although the mechanisms behind this functional hyperemia are not yet known in complete detail, there is a general agreement that it is induced by increased neural activity in retinal ganglion cells, and ensures adequate supply of nutrients and metabolites to the activated neuroretinal tissue with increased metabolic demand [33]. Whereas it was traditionally assumed that blood supply in neural tissue is strictly controlled by the local energy demand, recent findings show that this rapid regulatory process is the result of a tight interaction of neurons, glia cells and pericytes, usually referred to as the neurovascular unit or complex [67].

Glia cells directly transmit light-evoked vasomotor responses from retinal neurons to capillaries without neuronal intermediates [25]. Hence, functional hyperemia is absent when the connection of neurons to glia is interrupted [25]. Diabetes disturbs the normal interaction between glia cells, retinal ganglion cells and capillaries very early on a functional level [68], and glial degeneration is followed by hypoxia and retinal ganglion cell loss [69, 70], which may then also contribute to the worsening of vascular autoregulation in the diabetic retina.

Pericytes also play a role in the regulation of retinal capillary perfusion. They can constrict and dilate capillaries independently from arterioles [71]. This makes them crucial in the regulation of retinal blood flow, because retinal capillaries form the largest volumetric portion of this vascular system. Pericyte loss is among the earliest histologic signs of diabetic retinopathy [72]. But already before pericytes are lost, damage to gap junctions, and most likely also to the recently detected interpericyte tunneling nanotubes [73, 74] lead to a loss of coordinated light-evoked vasomotor response [75,76,77]. Pericyte damage-induced neurovascular impairment in diabetes is therefore very likely and may additionally disturb the blood supply to retinal ganglion cells, thus contributing to their functional impairment and sensitization to degenerate on the long term.

Besides the neurodegenerative component, we also detected an association of worse glycemic control in diabetes estimated by HbA1c levels with the reduced vascular response to flicker-light stimulation, whereas age showed less influence, since it lost its significant association in the multiple regression model. The observed negative correlation with HbA1c goes in line with the results from the large population-based Maastricht Study [78]. This study also reported a negative correlation of glucose levels and flicker light response, which we did not observe. Chronic hyperglycemia in diabetes may initiate alterations in the retinal vasculature including capillary basal lamina thickening, pericyte loss and capillary non-perfusion [67, 79], as well as dilation and endothelial dysfunction in larger vessels [80, 81]. These changes could be more prominent in patients with continuously higher glucose levels reflected by higher HbA1c, and also impair vascular reactivity. However, previous studies showed that the diminished retinal vasodilation response in diabetes is not solely due to endothelial dysfunction or pre-dilation of retinal veins [81, 82]. Consistent with this, the baseline vessel calibers in the current study were not different between patients and controls, and showed no influence on vessel reactivity. In addition, chronic hyperglycemia also induces oxidative stress and low-grade inflammation which are considered to play an important role in the development of DR [50, 83], and could also contribute to neurodegenerative changes.

There was considerable variability of the vasodilation response also among healthy individuals but no significant association with GCL thickness, age or other assessed parameters. Still, higher age variability in combination with a smaller sample size may have induced this observation [84].

Our study has some limitations that need to be considered: first, this is a cross-sectional study in a relatively small cohort. Future studies in larger cohorts are needed to validate our initial results. These studies should also include a full range of diabetic retinopathy stages and perhaps diabetic maculopathy to investigate if neurovascular coupling is progressively impaired together with further increasing changes in retinal layer structure. Although not statistically significant, diabetes patients were slightly older than healthy controls, and patients with moderate DR had a longer disease duration on average, which both could have introduced some degree of measurement bias. The inclusion of age, disease duration and arterial hypertension as predictors in the multiple regression model aimed to control for possible bias. In addition, because we did not include patients without clinical DR, the detected correlations may not be directly applicable to eyes without DR, where neurodegeneration may be less or absent. In diabetes patients without neuroretinopathy, neurovascular coupling impairment may be primarily induced by other mechanisms as explained above. However, our data correspond well to previous observations that independently found either more reduction in GCL thickness [12, 49] or a further decline in functional hyperemia [37, 41, 42] in advancing stages of DR. Our study results are limited to morphological changes in retinal layers, which also include other cell types than neuronal cells. As discussed above, disturbance and degeneration of the neurovascular complex in diabetes seems to affect all different components already early on. Although degeneration of retinal ganglion cells should logically result in reduced inner neuronal activity, direct consequences of retinal ganglion cell loss on neurovascular coupling in patients with DR remain to be shown. The current evidence from experimental studies indicates that ganglion cell impairment and loss rather happens consequently to vascular dysfunction. Longitudinal studies investigating the vascular, neuronal and glial components of the neurovascular complex in detail, could provide more information about the underlying pathways of changes in perfusion regulation during progression of DR. Future studies should also aim at identifying additional factors that contribute to degeneration of the neurovascular complex in diabetes in order to distinguish possible treatment targets. Improvement of glycemic control, reflected by lower HbA1c, may be an appropriate strategy to restore neurovascular coupling in diabetes patients. However, our results imply that this effect is limited by the degree of retinal neurodegeneration. This should be also considered in future trials assessing the effectiveness of therapeutic agents specifically addressing neurovascular dysfunction.

In conclusion, our current results corroborate the hypothesis that the reduced reactivity of retinal vessels to light stimulation in diabetes involves an independent regulatory impairment of the retinal neurovascular complex. Furthermore, GCL thinning indicating structural degeneration in the neurovascular complex of the inner retina might contribute to the advancing disturbance of neurovascular coupling in clinically established DR, in addition to the previously observed role of chronic hyperglycemia. Because the correlation between GCL thickness with neurovascular coupling was not observed in healthy individuals, it may be specific to diabetes.