Neuroretinal Dysfunction With Intact Blood-Retinal Barrier and Absent Vasculopathy in Type 1 Diabetes

Diabetic retinopathy (DR) is a major cause of vision loss (1). The prevalence of DR in type 1 diabetic patients is very high (∼90% after 15 years of disease duration), as shown by long-term epidemiological studies (2). Among the several known risk factors for DR development, disease duration is indeed one of the most relevant (3–6), together with metabolic control (5,7). Nevertheless, a fraction of patients presenting good metabolic control (∼10%) do develop DR, and still another proportion of patients with poor metabolic control, nevertheless, do not develop this complication (8). The role of vasculopathic mechanisms in DR is well established. Accordingly, classification of ocular fundus changes is relevant to the management and treatment of complications of this disease (9–11). The concept of a preretinopathy stage is characterized by the absence of lesions in ocular fundus examination and of the presence of subclinical changes in the blood-retinal barrier (BRB) (3,12,13). Leakage of this barrier has been proposed as the earliest manifestation in DR (3).

It is possible that independent neural damage may also occur in diabetes. This is difficult to study in the presence of concomitant vascular damage. However, if early neural damage occurs, it can be studied by searching for evidence of neuroretinal changes in the absence of retinal microvasculopathy. This is possible in the pre-DR stage (14–17). Some functional studies have suggested visual impairment in diabetic patients without apparent DR, but these studies did not exclude changes in integrity of the barrier function using a quantitative method (2,18–25). The occurrence of structural changes indicating vascular damage is also quite disputed in type 1 diabetic patients at a preclinical stage (23,26), possibly because the way patients are phenotyped may vary across studies. The criterion for determining the absence of DR lesions in type 1 diabetic patients is indeed a common problem across studies, which often are solely based on retinography or standard angiography. Definition of early vascular (endothelial) changes, as defined above, concerning measures of inner barrier function, may provide a more clear pathophysiological cutoff. Thus we aimed at establishing objective criteria for the absence or presence of initial vasculopathy by including the quantitative measurements of the permeability of the BRB.

Type 1 diabetic patients without clinical evidence for DR may show impaired objective responses to stimuli of distinct spatial frequencies (19), sometimes concomitantly associated with different contrast levels (18). A link of such deficits with parvo- and magnocellular impairment is, however, missing. Besides the changes in achromatic contrast sensitivity (20,22), there are also studies showing changed psychophysical and physiologic changes in these patients without DR, related to color contrast sensitivity (24,25) and multifocal electroretinogram (mfERG) responses (21). It remains unclear whether these changes can be present independently of vascular damage (e.g., if they could reflect direct neural damage).

The current study aimed to circumvent the above-mentioned methodological limitations by providing, as stated above, a clear-cut pathophysiological cutoff between presence and absence of BRB damage (quantitative analysis of BRB permeability with vitreous fluorometry [VF]). We combined this measure with objective structural and functional (electrophysiological) measures of visual function, as well as psychophysical thresholds. In summary, we investigated retinal neural dysfunction in a population of type 1 diabetic patients, from the psychophysical and neurophysiological point of view, with characterization of BRB leakage and metabolic control level. The goal was to identify whether neural damage can occur independently of alterations of the BRB and normal fundus images.

The study followed the tenets of the Declaration of Helsinki and was approved by our ethics committee. Prior to the inclusion in the study, informed consent was obtained from all subjects after a full explanation.

We tested 42 eyes of 42 patients (mean age ± SD = 26.6 ± 5.3 years) with preserved visual acuity (VA) and divided them into two groups: one group without any clinical signs of DR and normal BRB permeability as assessed by VF (n = 14 eyes; VA = 1.11 ± 0.15) and another group with BRB breakdown/vasculopathy with no clinical signs of DR or mild nonproliferative DR (n = 28 eyes; VA = 1.06 ± 0.14). These data were compared with those obtained in 25 age-matched controls (27.4 ± 5.8 years). Sample sizes were sufficient to probe evidence for isolated neural damage as assessed by electrophysiological measures as the main outcome (amplitude of central retinal responses, R1, sample size calculation for a α = 0.05 and power = 0.8; predicted effect size of 30 nV/deg² between controls and patients, as expected from our previous data and the literature cited in the article's introduction and discussion).

The criterion for choosing the study eye was the one with larger number of lesions, when applicable. If no DR signs were visible and BRB permeability values were normal, the right eye was chosen if the year of the patient’s birthday was an even number and the left eye if it was an odd number. This last principle (to prevent bias) was also applied to healthy controls.

Participants were submitted to a complete ophthalmological examination, including the best-corrected VA, slit lamp examination of anterior chamber, intraocular pressure measurement (Goldmann applanation tonometer), angle and fundus examination (noncontact lens), and subjective visual complaints. Exclusion criteria included media opacities, neuroophthalmological/retinal diseases (besides DR), and high ammetropies (sphere > ±4D; cylinder >± 2D).

Data were collected at the Institute for Biomedical Imaging in Life Sciences, the Association for Innovation and Biomedical Research on Light and Image, and the University Hospital of Coimbra until 2011.

In order to evaluate the patients’ metabolic control level, blood samples were collected for analysis of glycated hemoglobin (HbA_1c). This was assessed by high-performance liquid chromatography (Variant II, Bio-Rad).

Color fundus photographs (35°) of both eyes were taken in seven standard fields of the retina according to Early Treatment Diabetic Retinopathy Study (ETDRS) procedures (9), with a Topcon TRC-50IA Retinal Ophthalmic Camera with a Sony DXC-950P digital camera with a resolution of 0.5 MPixels/768 × 576. These photographs were always obtained after psychophysical and electrophysiological assessment to prevent artifactual changes in retinal adaptation.

Besides color fundus photography, fluorescein angiography (FA) was sequentially performed (TRC-50 IA Topcon; digital camera: B/W Megaplus 1.4i, resolution of 1 MPixel/1,340 × 1,037) to increase sensitivity in detecting topographic changes on BRB permeability.

A red-free photograph was taken before 10% fluorescein sodium solution administration, which was injected in the antecubital vein in a volume adjusted to the patients’ weight prior to VF.

A series of 20 early-phase pictures during the arterial and capillary filling were taken in the eye with larger number of lesions, when applicable. If no DR signs were visible, the right eye was chosen if the year of the patient’s birthday was an even number and the left eye if it was an odd number. After this series, a macula-centered photograph was taken from the contralateral eye, followed by acquisition of quadrantic field images of the selected and the contralateral eyes, respectively. Bilateral late-phase pictures (5 and 10 min) were finally taken in order to first classify in a qualitative manner the presence or absence of leakage to help substantiate the classification of the DR level (10).

VF is a valuable tool for studying the intraocular fluid dynamics and the permeability of the blood-ocular barriers, namely, the BRB. VF brought an important clinical contribution for quantifying BRB permeability in DR and is of particular interest in the preclinical stage since diabetic eyes show increased permeability values even in apparently normal ocular fundi (27,28).

For this purpose, we used a commercial fluorophotometer, the FM-2 Fluorotron Master (Coherent, Palo Alto, CA) and followed the standard protocol for BRB permeability measurement (29). A dose of 14 mg/kg was injected, and measurements of fluorescein concentration in the posterior vitreous were obtained 1 h after injection, after discounting for the natural fluorescence of the ocular tissues (preinjection scan). Blood samples were collected at 10, 15, and 50 min after injection for posterior measurements of the plasmatic fluorescein concentration (a cuvette that was adapted to the optical head of the Fluorotron was used for measuring the concentration of fluorescein in plasmatic solutions). With all these parameters, it was possible to obtain the penetration ratio (PR) of the fluorescein across the BRB (30).

We have applied several psychophysical and electrophysiological methods to characterize visual function and estimate neurophysiological parameters that could be used to study structure-function correlations in our groups.

Chromatic contrast sensitivity was tested along three parallel, randomly interleaved staircases (31) (Cambridge Colour Test, CRS, Rochester, U.K.). We simultaneously assessed the three cone confusion axes, modulated in the CIE 1976 u'v' color space. The chromatic stimulus (a Landolt-like C-shaped ring) appears over the noisy pattern of gray circles of different sizes/luminance levels, and participants chose the gap position in a forced four-choice task. Stimuli were presented monocularly on a 21-inch γ-corrected monitor (GDM-F520; Sony, Tokyo, Japan; refresh rate 100 Hz) at a viewing distance of 1.8 m in a darkened room. Far vision refractive correction was worn by participants, and tinted contact/spectacle lenses were replaced by lenses in a trial frame when necessary. The test ended after 11 reversals of each adaptive staircase, and the mean of the last 7 reversals was obtained as the threshold expressed in confusion vector length for each protan, deutan, and tritan axes in CIE 1976 u'v' color space units (32,33).

Achromatic contrast sensitivity within the magnocellular pathway was probed using a perimetric test based on frequency-doubling technology (FDT) (33–35). Vertically oriented sinusoidal grating stimuli (0.25 cpd, 25 Hz) were presented monocularly using a Humphrey Matrix Perimeter (Carl Zeiss Meditec Inc.).

An N-30-F (nasal 30°) threshold testing strategy was chosen. This procedure consisted of a modified binary search using a four-reversal rule to determine the threshold level at each of the 19 tested locations. The range of possible raw data are between 0 dB (maximum contrast/lowest patient sensitivity) and 56 dB. The formula to calculate sensitivity in dB units was log₁₀*(2,048/c)*10*H, where c ranges from 1 to 2,048 (scaled minimum and maximum contrast, respectively) and H (Humphrey scaling factor) is ∼2.

Statistical analysis was performed considering both global parameters (mean deviation [MD] and pattern SD [PSD]) and contrast sensitivity pooled from five regions: the 5° central area (C) and the four visual field quadrants (superior temporal [ST], superior nasal [SN], inferior nasal [IN], and inferior temporal [IT]).

The electrophysiological procedures mentioned below were performed with pupil dilation.

mfERG was recorded using RETIscan (Roland Consult, Germany) with DTL-Plus electrodes to measure local photopic activity within photoreceptor/bipolar cell circuitry (36). The stimulus consisted in a 30° central visual field hexagonal pattern with 61 elements adjusted for the magnification factor, pseudo-randomly presented binocularly according to the standard m-sequence on a 20-inch cathode ray tube monitor (rate 60 Hz; viewing distance 33 cm). Participants’ far vision refractive correction was compensated, when applicable, according to the adjustment scale for this purpose, and near refractive correction was added if necessary.

Active voltage range of the signal was ±200 μV using a band-pass filter of 5–100 Hz and amplification at a gain of 100,000. Eight 47-s cycles were obtained for averaging at an artifact rejection level of 10%.

Analyses were performed using the first-order kernels. For each hexagon and concentric ring (eccentricities, in diameter of visual angle: ring 1, 4.4°; ring 2, 4.4–13.6°; ring 3, 13.6–25.8°; ring 4, 25.8–40.8°; ring 5, 40.8–58.7°), the amplitude (nV/deg²) and implicit times (ms) of the N1 trough and the P1 peak were calculated.

Given the observed relative violation of ANOVA assumptions, nonparametric analyses (Mann-Whitney U testing/Spearman rank correlation) at a significance level of P < 0.05 were performed using StatView (SAS, Cary, NC), with correction for the number of group comparisons.

Given the conservative approach of only including one eye per participant (see 2Research Design and Methods), we also replicated this analysis with an approach that uses all eyes, with correction for dependence using generalized estimating equations (see Supplementary Data with this replication analysis).

Receiver operating characteristic (ROC) analyses (MedCalc, version 12.2.1.0, Mariakerke, Belgium) were performed to determine sensitivities at defined levels of specificity (∼90%) and to compare differences between controls and patients without signs of DR and preserved BRB. Diagnostic accuracies of psychophysical and electrophysiological tests were assessed by comparing areas under the ROC curves (AUCs). Statistic significance was set at P < 0.05.

Table 1 provides a global summary of the demographic and clinical data of the included diabetic patients (age, diabetes duration, HbA_1c, and presence/degree of vascular ocular fundus lesions). Group analyses are presented below.

The study group included 42 eyes of 42 type 1 diabetic patients (14–37 years; mean age ± SD = 26.6 ± 5.3 years) with preserved VA (best-corrected values ranging from 0.9 to 1.3) divided into two subgroups according to the values of BRB permeability assessed by VF and the ETDRS classification system (based on color fundus photography and FA analysis): one group with normal BRB permeability (≤3.1 × 10⁻⁶ min⁻¹) and no fundus signs of DR (n = 14 eyes; VA = 1.11 ± 0.15) and another group that included patients with mild nonproliferative DR, or no DR but increased BRB permeability values (n = 28 eyes; VA = 1.06 ± 0.14).

The mean duration of disease ranged from 7 to 29 years (group with no DR, mean ± SD = 14.43 ± 4.86 years; group with initial DR, mean ± SD = 18.93 ± 5.04 years).

All participants completed the protocol, so no cases of missing data were present.

These patients had thorough and systematic local surveillance of metabolic control levels since childhood, at the time of their diagnosis of diabetes: one visit to the endocrinology department, every 6 months in the initial follow-up, and one visit yearly thereafter. In the group with normal BRB, HbA_1c varied from 6.2% (44 mmol/mol) to 9.9% (85 mmol/mol; mean ± SD values = 7.74 ± 1.11% [61 ± 12.1 mmol/mol]). In the group with disruption of BRB, HbA_1c varied from 6.2% (44 mmol/mol) to 13.0% (119 mmol/mol; mean ± SD values = 8.70% ± 1.64% [72 ± 17.9 mmol/mol]). This difference did not reach statistical significance (P = 0.07).

These two methods were jointly used to classify the DR level: 10 (17 eyes, 40.5%), 20 (15 eyes, 36%), 35C (4 eyes, 9.5%), and 35D (6 eyes, 14%). See Table 1 for more details (right column).

In the group with disrupted BRB, PR values (10⁻⁶ min⁻¹) ranged from 1.78 to 12.39 (mean ± SD = 4.78 ± 2.79), while in the other group, these values ranged from 1.08 to 2.81 (mean ± SD = 2.12 ± 0.49). These are below the limit for normal values (≤3.1 × 10⁻⁶ min⁻¹) identified in the literature, thereby enabling us to define the participants in this group as having normal BRB permeability (37) according to the inclusion criteria.

Table 2 provides a detailed characterization of the diabetic group characterized by no changes in BRB and, simultaneously, by the absence of signs of DR.

Figure 1 shows different examples of psychophysical and neurophysiological data in three patients with different patterns of damage (patients 5 and 29 present no DR, patient 5 with minimal functional changes and patient 29 with clear impairment on visual function testing; patient 4 had initial DR).

As illustrated in Fig. 2, chromatic contrast sensitivity measures were significantly different between controls and patients. Considering the group with disrupted BRB, the comparison with controls showed higher thresholds for all chromatic axes (protan, P = 0.007; deutan, P < 0.0001; tritan, P < 0.0001). Importantly, significantly higher thresholds were also found, but only for the tritan axis in the group with normal BRB permeability and no DR (P = 0.009). When comparing both patient groups, significant differences are only found for the deutan axis (P = 0.01).

MD, a global parameter of magnocellular performance (FDT testing; see 2Research Design and Methods) yielded significant differences between controls and each group of patients (with normal BRB permeability and no DR, P = 0.002; with leakage and/or DR, P = 0.0002), with no significant effect detected between patient groups. Concerning PSD (field heterogeneity measure), a significant difference was found between controls and patients with diabetes without leakage and no DR (P = 0.02) and no difference between patients groups. For details, see Fig. 3A.

Please note that MD and PSD provide essentially different information, and MD is more useful as a lesion indicator since it defines absolute/global damage. This value is higher in the group of initial DR, as expected.

To probe whether regional differences were underlying these effects, we then analyzed magnocellular achromatic contrast sensitivity in distinct regions. Significant differences were found in all five locations, between controls and both patient groups, except in the C region and IT quadrant of the visual field in patients with no lesions (group with preserved BRB and no DR: ST, P = 0.04; SN, P = 0.01; IN, P = 0.058—group with disrupted BRB: C, P = 0.02; ST, P = 0.01; SN, P = 0.003; IN, P < 0.0001). Comparisons between both patient groups showed no significant differences (see Fig. 3B for details).

The mfERG showed a significant decrease in amplitude in all concentric rings in both patient groups when compared with control subjects (P1 wave, all eccentricity rings, P < 0.0001 in the group of disrupted BRB—group with preserved BRB, R1, P < 0.0001; R2, P = 0.0003; R3–R5, P < 0.0001) (Fig. 4A), suggesting a neural basis for damage at the outer retinal level (photoreceptor and bipolar cell circuit level) since dysfunction occurs irrespective of evidence for early vascular damage.

Analysis of implicit times showed a significant decrease in patients with preserved BRB and no DR, as compared with controls, but only in ring 2 (P = 0.006). This difference is also evident in ring 2 (P = 0.02) in the group with disrupted BRB (see Fig. 4B for details).

A similar pattern was found concerning the N1 component: decrease in amplitude values in both patient groups (with disrupted BRB, R2, P = 0.001; R3, P = 0.0003; R4, P = 0.0006; R5, P < 0.0001—with preserved BRB, rings 2–5, P < 0.0001), with no changes in implicit times (Fig. 5A and B).

It is important to point out that no significant differences were obtained for amplitude and implicit time values of P1 and N1 waves between both clinical groups.

Significant correlations could be observed between PR and functional measures in the group with leakage, namely, in chromatic thresholds (deutan, r = 0.39, P = 0.04; tritan, r = 0.37, P = 0.05) and in mfERG P1-wave implicit time (most peripheral rings, R5, r = 0.46, P = 0.002).

Duration of disease correlated with implicit time of mfERG, namely, P1 wave (most peripheral rings, R4, r = 0.43, P = 0.03; R5, r = 0.44, P = 0.02; and marginally lost in ring 1, r = 0.25, P = 0.06) and N1 wave (ring 1, r = 0.44, P = 0.02), in the group with BRB changes. In the other clinical group, duration of diabetes only correlates with a chromatic protan thresholds axis: r = 0.72, P = 0.002. These findings suggest distinct pathophysiological mechanisms in the presence or absence of BRB breakdown.

ROC curves were generated for all included tests, allowing us to compare sensitivities (probability of positive testing results when the clinical condition is present) for detecting impairment at fixed specificity levels. In our analysis, the chosen cutoff for this parameter was set near 90%. The highest sensitivity values were found in mfERG amplitude values (P1 wave, sensitivities of 78% across the five concentric rings). N1 wave amplitude values also showed high sensitivities (up to 82%) in paracentral rings, except in the central ring. Chromatic contrast tests showed the lowest sensitivity, in particular, concerning protan and deutan axes (Table 3 and Fig. 6).

The AUC summarizes the diagnostic accuracy of each parameter (AUC = 1 would represent a perfect discrimination between controls and diabetic patients with no signs of DR/preserved BRB; AUC = 0.5 means discrimination at the chance level). Our results showed AUCs between 0.946 and 0.532 (Table 3). The higher AUC values (P < 0.0001) correspond to the mfERG test (P1 and N1 wave amplitudes, except ring 1 of the N1 wave).

In this study, we have found evidence for an independent neural phenotype in pre- or absent vascular DR in type 1 diabetes. This was possible to establish by using a clear pathophysiological cutoff defining the onset of endothelial lesions. This cutoff is based on in vivo objective measures of BRB permeability.

Retinal neuronal changes occurring in type 1 diabetic patients with no breakdown of the BRB or onset of vasculopathy were probed by psychophysical (achromatic/chromatic contrast sensitivity) tests of ganglion cell magno-, parvo- and koniocellular pathways and electrophysiological recordings (mfERG). The latter suggest a neural correlate of dysfunction at the outer retina.

Interestingly, we found that neurophysiological changes correlate with increased BRB permeability in the group of patients with leakage, which confirms the existence of an independent mechanism in the absence of BRB disruption. This corroborates the idea that isolated neural damage can occur even when the (usually dominant) vascular damage is not present. The fact that these measures were more tightly linked with disease time duration in the group with vascular lesions suggests that vascular damage superimposed on primary and secondary neuronal changes can dominate during the subsequent period in the natural history of disease. These issues can best be clarified in a longitudinal study.

Concerning the functional pathways that were affected even in the prevascular stage, we identified damage of both parvocellular and koniocellular pathways (as demonstrated by chromatic contrast sensitivity testing). Some previous studies also report significant color vision defects in type 1 diabetes (24,25). However, these studies did not provide a clear cutoff between prevascular and vascular stages.

Impairment of magnocellular function was also observed (as revealed by decreased thresholds in FDT in C and peripheral regions), even in the group with normal BRB measures and without clinical signs of DR. Previous studies reported a nonselective decrease in contrast sensitivity (18,19,22,38). Our study used a specific perimetric test biased to the activation of the magnocellular performance, probing central and peripheral vision, and our results are consistent with a previous study documenting FDT changes in diabetic patients with no clinically detectable DR (39,40). mfERG findings showed generalized decrease amplitudes that are consistent with changes at photoreceptor/bipolar cell circuits. Recent studies (41–43) reported local neurophysiological changes in type 1 diabetes, as assessed by mfERG. Although these local neurophysiological changes show the expected interindividual variability (in terms of range of amplitude values), ROC and AUC analyses showed that relatively high sensitivity is observed for electrophysiological measures at high specificities, confirming the presence of sensitive early measures of specific neural damage that may potentially be used as biomarkers for detecting neural changes independently of vascular lesions.

Accordingly, studies by the Adams group have shown that local mfERGs are highly predictive of the subsequent development of DR in adult patients, suggesting the location of future lesions, based on previous local changes in implicit time (44–46).

In our study, we also found early mfERG changes. However, these are not directly comparable to the above-mentioned results because of group definition criteria. Changes in wave morphology and amplitude might lead to different forms of change and even decreases in the local implicit time values, especially in cases of amplitude reduction. However, in patients with minimal DR, implicit time tended to increase. As mentioned above, small differences across studies might be due to the criterion for defining the absence of retinal lesions. In previous studies, it was based on fundus photography and not on an increase in BRB permeability, which is the first sign of vascular damage (3,12,13). In other words, BRB permeability measurements may be critical to precisely define the presence or absence of a vascular phenotype.

In summary, our study uniquely combined objective measures of the BRB permeability and DR staging to define diabetic patient groups with and without vascular damage. A neural phenotype could then be investigated with functional information gathered from psychophysical and neurophysiological measures. We found that these could discriminate between groups with high sensitivity and specificity, as confirmed by ROC analyses, which support the generalizability of our results. These findings show that neural changes may probably occur in the retina in addition to and independently of vascular lesions, with potential implications for early neuroprotective treatment (47).

The main limitation of this study is that it is cross-sectional and cannot infer on the relative time courses of neural and vascular processes. Moreover, the BRB permeability assay may be differently sensitive to vascular pathology in different regions of the retina when compared with retinal function testing across broad regions. Future longitudinal studies should elucidate the time course of isolated diabetes-related neural impairment in relation to the likely dominant neural phenotype secondary to vascular damage that is also present in the natural history of DR.

We conclude that retinal neuronal changes can occur in type 1 diabetic patients independently of the breakdown of the BRB or vasculopathy.

Funding. This research was supported by Fundação para a Tecnologia: FCT/PTDC/SAU/NEU/68483/2006, COMPETE/PEst-C/SAU/UI3282/2013; the National Brain Imaging Network of Portugal, Comissão de Coordenação Região Centro, CENTRO-07-ST24-FEDER-00205; and Agência de Inovação, QREN-COMPETE/DoIT/Diamarker.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Authors Contributions. A.R. designed the experiments, acquired and analyzed data, and wrote and edited the manuscript. C.M. acquired and analyzed data. P.M. and J.F. acquired data. J.C.-V. contributed to the discussion. M.C.-B. designed the experiments, analyzed data, contributed to the discussion, and wrote the manuscript. A.R. and M.C.-B. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.