Retinal Oxygen Metabolism in Patients With Type 2 Diabetes and Different Stages of Diabetic Retinopathy

Diabetic retinopathy (DR) is a frequent ocular complication of diabetes and among the most prevalent causes for severe and irreversible vision loss in adults (1). Data from recent epidemiological studies have shown a dramatic increase in the prevalence of type 2 diabetes, which is characterized mainly by peripheral insulin resistance and frequently associated with lifestyle factors such as lack of physical activity or a high-calorie diet (2–5). More than 50% of patients with type 2 diabetes develop DR within 20 years from diagnosis (6), which underscores the need of a better understanding of the pathophysiology of DR in order to identify high-risk patients early and allow for tailored therapy (7). Furthermore, as the human eye allows for the noninvasive in vivo visualization of the microvasculature, retinal changes may be an early indication of microvascular complications in patients with diabetes (8,9).

There is increasing evidence that hypoxia and impaired adaptive responses to hypoxia are important pathophysiological factors involved in the development of DR and diabetic macular edema (10,11). It has been hypothesized that hypoxia in the retina triggers the release of vascular growth factors, which in turn lead to pathological vessel growth and leakage with well-described clinical consequences (12). Thus, a close monitoring of retinal oxygen metabolism may be an important biomarker for risk stratification, treatment decision making, and follow-up (13). However, the in vivo investigation of oxygen metabolism in humans is challenging. The introduction of retinal function imaging devices, such as retinal spectroscopy, has allowed for the noninvasive measurement of oxygen saturation in vivo (14–16). Using this technology, several studies have shown that retinal oxygen saturation is altered in patients with diabetes (17–19) but does not directly reflect oxygen consumption or oxygen extraction, which are also determined by the level of retinal perfusion.

In this study, we investigated retinal oxygen extraction using a previously published and validated model (20). Measuring retinal blood flow with Doppler optical coherence tomography (OCT) and oxygen saturation with spectroscopic fundus imaging allows us to calculate tissue oxygen extraction. We have recently shown by using this approach that oxygen extraction is reduced in patients with type 1 diabetes compared with healthy, age-matched subjects (21). We extend this approach to patients with type 2 diabetes to determine whether oxygen extraction depends on severity of disease.

The study protocol was approved by the ethics committee of the Medical University of Vienna and national competent authorities and followed the Declaration of Helsinki and Good Clinical Practice guidelines of the European Union. All subjects were selected by the Department of Clinical Pharmacology, Medical University of Vienna, and provided written informed consent before any study-related procedures. All included subjects passed a prestudy screening in the 4 weeks before the first study day, including the following examinations: medical history; pregnancy test in women of childbearing potential; blood pressure and heart rate measurements; laboratory testing for HbA_1c, blood glucose, and hemoglobin levels; assessment of visual acuity; slit lamp biomicroscopy; indirect fundoscopy; standard seven-field color fundus photography for grading of DR; and measurement of intraocular pressure (IOP) using Goldmann applanation tonometry. If any clinically significant medical condition other than diabetes was found as part of the screening examination, the subject was not included. Further exclusion criteria were ametropia of more than six diopters, smoking, and personal history or family history of epilepsy. No patients with diabetic macular edema or history of diabetic macular edema were included.

The current study was a parallel group design using a blinded observer analysis. On the screening and study days, all subjects arrived in the morning after 7–8 h of sleep. All subjects abstained from alcohol and stimulating beverages containing xanthine derivates (e.g., tea, coffee, or cola drinks) 12 h before the study day. At the beginning of the study day, a pregnancy test was performed in females of childbearing potential. After instillation of one drop of Mydriaticum Agepha (Agepha, Vienna, Austria) into the study eye, a resting period of at least 20 min was scheduled before measurements were started to ensure constant hemodynamic conditions. Capillary blood glucose levels were assessed by a glucose meter. Retinal nerve fiber layer thickness was measured using a commercially available OCT system. The same system was used to measure retinal vessel density (VD) using an OCT angiography (OCT-A) device. Then, oxygen saturation was measured with a retinal vessel analyzer, and total retinal blood flow (TRBF) using Doppler OCT was assessed. Before and after blood flow measurements, blood pressure, heart rate, and IOP were measured. Ocular perfusion pressure was calculated as 2/3 MAP − IOP (22).

An automated oscillometrc device (Infinity Delta; Dräger, Vienna, Austria) was used to measure systolic and diastolic blood pressure, mean arterial pressure, and pulse rate using a fingertip pulse oximeter.

IOP was measured with a slit lamp–mounted Goldmann applanation tonometer. Before each measurement, one drop of oxybuprocaine hydrochloride combined with sodium fluorescein was instilled to obtain corneal anesthesia.

Capillary blood glucose levels were measured with a commercially available glucose meter (Accu-Chek Go; Roche Diagnostics GmbH, Vienna, Austria). Capillary blood was sampled from one fingertip using a single-use lancet. To induce capillary vasodilation for sample collection, Finalgon paste (Boehringer Ingelheim Regional Center Vienna, Vienna, Austria) was applied on the earlobe. A lancet incision was made, and a thin tube was used to draw capillary blood. Using an automatic blood gas analysis system (ABL800 FLEX; Drott Medizintechnik GmbH, Wiener Neudorf, Austria), pH, Pco₂, and Po₂ were determined.

Measurement of retinal oxygen saturation is based on the image analysis using the software of the retinal vessel analyzer camera, which was coupled to the Doppler OCT system. In two monochromatic fundus images obtained by the camera and filter assembly at different wavelengths of 610 nm and 540 nm, the operator marked the vessel of interest by a mouse click. Photometric edges in the vicinity of the mouse cursor were located as vessel walls. If the software determined edges, the search was continued in their vicinity. To determine the direction in which a vessel is traced, three or more edge segments needed to be found. For analysis of each vessel segment, 3–10 pixels in length, 6 pixels outside the vessel (3 on either side) and all pixels inside the vessel, excluding those representing the wall, were considered. To calculate the oxygen extraction, the brightness values of the pixels inside and outside the vessel in the green as well as red camera channel were averaged. Obtained O₂ saturation values were averaged over the vessel (14). Oxygen measurements were taken at the same location on the vessel where perfusion measurements were performed.

Peripapillary retinal nerve fiber layer thickness was assessed using the glaucoma module of a commercially available spectral domain OCT (SPECTRALIS; Heidelberg Engineering, Heidelberg, Germany).

To gain VD information, a commercially available spectral domain OCT device with an OCT-A module (SPECTRALIS) was used to perform OCT-A measurements. Macula-centered, high-resolution (512 B-scans, 512 A-scans/B-scan) 10° × 10° OCT-A scans were performed. The superficial vascular plexus (SVP), intermediate vascular plexus (ICP), and deep capillary plexus (DCP) were the layers of interest in this investigation. Quality checks were performed to filter images with poor quality. The raw enface scans of the SVP, ICP, and DCP were exported using MATLAB version 2018b software (MathWorks, Natick, MA). A fovea-centered annulus with an inner diameter of 1 mm and outer diameter of 2.5 mm was generated. Images were then binarized using the mean values inside the annulus to obtain the corresponding VD values.

Statistical analysis was performed using SPSS version 26 software (IBM Corporation, Armonk, NY). All values are presented as mean ± SD. Normal distribution for all variables was confirmed using the Shapiro-Wilk test. Descriptive statistics are reported for all values obtained. One-way ANOVA was used to assess overall differences among the four groups (healthy subjects, patients with no DR, patients with mild DR, and patients with moderate to severe DR). In addition, comparisons among groups were performed using least significant difference testing within the ANOVA model. P < 0.05 was considered as the level of significance.

The data sets generated during the current study are available from the corresponding author upon reasonable request.

A total of 87 subjects were included, of whom 67 had diabetes with various stages of nonproliferative DR. Thirty-four had no DR, 15 had mild DR, and 18 had moderate to severe DR. In addition, 20 healthy control subjects were included. The demographic and baseline characteristics of the four study groups are shown in Table 1. Significant differences between groups were found for BMI (P = 0.010), HbA_1c (P = 0.006), and plasma glucose level (P = 0.034). No differences between groups were found for diabetes duration, systemic hemodynamics, IOP, ocular perfusion pressure, or retinal nerve fiber layer thickness.

TRBF was 40.1 ± 9.2 μL/min in patients with no DR (P = 0.323 vs. healthy subjects) and 41.8 ± 15.0 μL/min in patients with mild DR (P = 0.215 vs. healthy subjects), while it was 37.2 ± 5.7 μL/min in healthy subjects and 34.6 ± 10.4 μL/min in patients with moderate to severe DR (P = 0.445 vs. healthy subjects) (Fig. 1A).

Retinal oxygen saturation in arteries was significantly different among the four groups (95.6 ± 3.3% in healthy subjects, 93.1 ± 4.0% in patients with no DR, 96.0 ± 3.6% in patients with mild DR, 95.6 ± 3.9% in patients with moderate to severe DR). When comparing single groups, significant differences between patients with no DR and healthy subjects (P = 0.019), patients with mild DR (P = 0.020), and patients with moderate to severe DR (P = 0.039) were found.

Retinal venous oxygen saturation was comparable between healthy subjects (60.6 ± 10.0%) and patients with no DR (60.5 ± 7.2%, P = 0.967), while it was significantly higher in subjects with mild DR (68.7 ± 9.5%, P = 0.025 vs. healthy subjects and P = 0.009 vs. patients with no DR). There was no significant difference between patients with moderate to severe DR (65.3 ± 9.3%) and healthy subjects (P = 0.162).

The arteriovenous difference in oxygen saturation was highest in healthy subjects (34.9 ± 7.5%), followed by patients with no DR (32.5 ± 6.3%, P = 0.243 vs. healthy subjects) and moderate to severe DR (30.3 ± 6.5%, P = 0.057 vs. healthy subjects). It was also found to be the lowest in patients with mild DR (27.3 ± 8.0%, P = 0.010 vs. healthy subjects and P = 0.044 vs. patients with no DR) (Fig. 1B).

Calculated retinal oxygen extraction was significantly different between healthy subjects and patients with moderate to severe DR (P = 0.040) (Fig. 1C). In particular, retinal oxygen extraction was numerically the highest in healthy subjects (2.24 ± 0.57 μL O₂/min), followed by patients with no DR (2.14 ± 0.65 μL O₂/min, P = 0.619 vs. healthy subjects), mild DR (1.90 ± 0.77 μL O₂/min, P = 0.164 vs. healthy subjects), and moderate to severe DR (1.78 ± 0.57 μL O₂/min, P = 0.040 vs. healthy subjects) (Fig. 1C).

When looking at the retinal microcirculation, differences in central macular VD among groups were found in the DCP and ICP. In the DCP, VD was highest in healthy subjects (25.5 ± 5.7%), followed by patients with no DR (21.3 ± 3.4%), mild DR (17.8 ± 5.2%), and moderate to severe DR (16.3 ± 4.0%) (Fig. 1D). These differences were significant between healthy subjects and all DR groups (P = 0.010 vs. no DR, P < 0.001 vs. mild DR and vs. moderate to severe DR). Significant differences in DCP VD were also found among patients with no DR and mild DR (P = 0.028) or moderate to severe DR (P = 0.001). The same pattern was found for VD in the ICP. Again, the highest values were found in healthy subjects (28.1 ± 4.5%) compared with patients with no DR (24.5 ± 4.2%, P = 0.029 vs. healthy subjects), mild DR (22.5 ± 4.9%, P = 0.004 vs. healthy subjects), or moderate to severe DR (19.4 ± 4.3%, P < 0.001 vs. healthy subjects). Differences were also significant between patients with no DR and patients with moderate to severe DR (P = 0.001). In the SVP, differences among groups were less pronounced. While VD was 11.9 ± 4.0% in healthy subjects, it was 9.9 ± 4.1% in patients with no DR, 8.4 ± 2.5% in patients with mild DR, and 8.7 ± 1.7% in patients with moderate to severe DR. These differences were significant between healthy subjects and patients with mild DR (P = 0.021), as well as between healthy subjects and patients with moderate to severe DR (P = 0.029).

This study is the first to demonstrate that retinal oxygen extraction is reduced in patients with type 2 diabetes and moderate to severe DR. Furthermore, our data indicate that the impairment of oxygen metabolism depends on the severity of disease, supporting the concept that functional imaging may in the future serve as biomarker to identify high-risk patients.

Tissue hypoxia has been observed in several human tissues of patients with diabetes, including the skin, kidney, and retina (28–31). As for retinal tissue, the introduction of noninvasive, in vivo imaging of oxygen saturation has stimulated research on oxygen metabolism in patients with diabetes (16). Using multiwavelength spectroscopy, a variety of studies have reported altered oxygen saturation in patients with diabetes (17–19). As such, it has consistently been shown that arteriovenous differences in oxygen saturation of retinal vessels are decreased in patients with diabetes (17–19,32).

The interpretation of the findings from these studies remains difficult since the amount of oxygen delivered to the retina depends on perfusion, which is highly autoregulated (20,33–35). Thus, measuring oxygen saturation in retinal arteries and veins alone may not be sufficient to gain a full picture with regard to metabolic changes in the retina. It has been known for a long time that oxygen-induced vasoconstriction as well as hypoxia-induced vasodilation are altered in patients with diabetes, indicating an impaired metabolic autoregulation in this group (11,36–38). The mechanism behind these altered responses to hyperoxia and hypoxia is still not fully clarified but may be related to the impairment of hypoxia-inducible factor (HIF) regulation, as described in the literature (39). HIFs are transcriptional factors that induce the expression of angiogenesis growth factors, such as vascular endothelial growth factor (40). This leads to the formation of pathological vessels as seen in DR (41). HIFs have been found to also have an impact on vascular tone, which may change the vasomotor response to hyperoxia and hypoxia (42). Thus, it is difficult to decide whether changes in oxygen saturation in retinal vessels may be attributed to a reduced oxygen consumption or by a vascular impairment in patients with diabetes.

Our results may help to further elucidate the complex metabolic changes in the retinal tissue of patients with type 2 diabetes. They indicate that retinal oxygen extraction in patients with moderate to severe DR is significantly lower compared with healthy, age-matched subjects. This finding supports the concept that tissue hypoxia is caused by a reduced oxygen delivery to the retinal tissue, which in turn may contribute to the development of diabetic complications (17,43). Consistent with the main hypothesis of our study, our data show that the arteriovenous difference in oxygen saturation is highest in healthy subjects, followed by patients with diabetes and no DR, and is lowest in patients with moderate and severe DR. The latter results are in keeping with previous studies, also indicating a decline of arterio-venous oxygen difference with increasing severity of DR (19,44).

The hypothesis that the altered oxygen extraction is caused by reduced oxygen delivery to the tissue is also in line with the OCT-A data of our study. OCT-A imaging indicates a microvascular rarefication in the DCP and ICP of the retina, confirming data from previous studies using OCT-A to assess microvascular changes in patients with diabetes (45–48). Thus, it is reasonable to assume that such a capillary dropout and capillary nonperfusion may lead to a decrease of oxygen delivery to the tissue, which is reflected in the decrease of oxygen extraction as measured in our study. However, to finally prove this hypothesis, methods for the in vivo assessment of oxygen saturation on the level of retinal capillaries would be necessary and currently do not exist.

Interestingly, our results indicate an increased retinal arterial oxygen saturation. This is in keeping with previously published data consistently showing increased arterial oxygen content in patients with diabetes (44,49–51). It has been hypothesized that the difference in retinal arteriolar oxygen saturation may be due to countercurrent oxygen exchange in the optic nerve, where the central artery and vein are closely adjacent (44). More specifically, two effects may cause this observation. On the one hand, there is evidence for increased retinal blood flow in patients with diabetes as indicated from the current study and several previous reports (52–54). This may result in a reduced oxygen loss to the adjacent vein. On the other hand, lower oxygen extraction and, in turn, higher venous oxygen saturation in the countercurrent venous flow may also result in a higher arterial oxygen saturation due to the reduced oxygen gradient between the artery and vein (50). Direct evidence for these mechanisms is, however, lacking.

Methodological issues may also contribute to the observed effect. One could argue that differences in optical densities in diabetic tissues or vessel walls may contribute to the observation of an increased retinal oxygen saturation (50). Although this effect cannot be fully excluded, there have been experiments validating our approach. During graded hypoxia, we have shown that despite a decrease in arteriovenous oxygen saturation difference during hypoxic conditions, the concomitant increase in blood flow leads to a constant retinal oxygen extraction (55). This further proves the close interaction between blood flow and oxygen extraction and the validity of the technique.

In considering the hypothesis that tissue hypoxia is present in patients with type 2 diabetes, one might expect a counterregulatory increase in ocular blood flow to satisfy the tissues’ oxygen need, depending on the severity of disease. Interestingly, there is still controversy in the literature on blood flow changes in patients with diabetes. Although some studies found a decrease of retinal blood flow in patients with diabetes (56,57), others have reported an increase in retinal perfusion (21,52,58). In the current study, we found a tendency toward increased retinal blood flow in patients with type 2 diabetes and no or mild DR, but this effect was not significant. This is in line with a previous study from our group where we observed a slight increase in retinal blood flow in patients with early type 1 diabetes and no signs of DR compared with healthy subjects (21). Further studies are needed to investigate blood flow in different stages of DR.

The current study has some strengths and limitations. One of the major strengths is that the combination of Doppler OCT with the fundus camera–based spectroscopic measurement of oxygen saturation allowed us to noninvasively assess oxygen extraction of the retinal tissue in vivo. Thus, our conclusions are not based on absolute oxygen saturation levels but on the integrated effect of the arteriovenous difference in oxygen saturation together with perfusion data. This is of particular importance because oxygen metabolism and blood flow are strongly coupled, and hypoxia causes counterregulatory vasodilation (55). Furthermore, we have validated our approach in previous studies and successfully used this method to assess oxygen extraction in patients with type 1 diabetes (21) as well as in patients with systemic neurodegenerative and neuroinflammatory diseases (59,60).

As a limitation, the current study is cross-sectional in nature. Thus, we cannot determine whether the reduction in oxygen extraction observed in patients with type 2 diabetes is a cause or a consequence of the disease. To further elucidate this question, a longitudinal approach would be necessary. Furthermore, we cannot fully exclude that factors such as diabetes duration, glucose levels, or blood pressure may partially influence our results. In the current study, no correlation between retinal hemodynamic factors and these clinical variables was observed (data not shown), but the sample size may be too small to elucidate such associations.

In summary, our data indicate that oxygen extraction is reduced in patents with type 2 diabetes and moderate to severe DR. Further longitudinal studies are needed to assess whether oxygen saturation may in serve as a biomarker in the future to identify high-risk patients.

Funding. This study was supported by the Austrian Science Fund (grants KLI529 and KLI721), National Medical Research Council (grants CG/C010A/2017_SERI, OFLCG/004c/2018-00, MOH-000249-00, MOH-000647-00, MOH-001001-00, MOH-001015-00, MOH-000500-00, and MOH-000707-00), National Research Foundation Singapore (grants NRF2019-THE002-0006 and NRF-CRP24-2020-0001), A*STAR (A20H4b0141), Singapore Eye Research Institute & Nanyang Technological University (SERI-NTU Advanced Ocular Engineering [STANCE] Program), SERI-Lee Foundation (grant LF1019-1), and Duke-NUS Medical School (grant Duke-NUS-KP[Coll]/2018/0009A).

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

Author Contributions. N.H., M.K., A.S., K.H., D.S., and G.G. recruited patients and performed the study. N.H., R.M.W., D.S., L.S., and G.G. analyzed the data and performed the statistical analysis. N.H., D.S, L.S., and G.G. planned the study. N.H. and G.G. wrote the manuscript with input from all authors. G.G. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.