Early Complement Activation and Decreased Levels of Glycosylphosphatidylinositol-Anchored Complement Inhibitors in Human and Experimental Diabetic Retinopathy

Human postmortem eyes were obtained from certified eye banks through the National Disease Research Interchange. Criteria for inclusion in the study were age <75 years, diabetes duration <15 years to address mostly nonproliferative retinopathy, and the fewest possible chronic pathologies other than diabetes. Criteria for exclusion were retinal, hematological, and inflammatory diseases; uremia; chemotherapy; and use of life support measures. The causes of death were almost exclusively cardiovascular in both groups, and hypertension was reported in one-third of donors in each group. The eyes used for immunohistochemical studies were from 13 diabetic and 14 nondiabetic donors (Table 1). The eyes were fixed in 10% buffered formalin by the eye banks; one retina from each donor was cryopreserved and embedded in OCT compound for sectioning, and the other was in most cases digested with trypsin to isolate the intact microvascular network (trypsin digests) (3). The retinal trypsin digests of seven donors in each group had previously been studied for microvascular cell apoptosis and histological lesions of nonproliferative diabetic retinopathy (3). The eyes used for protein isolation were from 17 diabetic and 15 nondiabetic donors (Table 1).

Six-week-old male Sprague-Dawley rats (Taconic Farms, Germantown, NY) were assigned randomly to a diabetic and a control group. Diabetes was induced by intravenous administration of streptozotocin (55 mg/kg body wt, dissolved in citrate buffer pH 4.5). The care of the diabetic rats and insulin treatment to prevent weight loss were as described previously (14). The rats were killed after 10 weeks of diabetes. A total of nine diabetic and nine control rats were studied; from each rat, one retina was embedded in OCT to prepare frozen sections, and the other was homogenized in lysis buffer for isolation of protein.

Sections (10 μm) and trypsin digests from the formalin-fixed human retinas were rehydrated and blocked as described (15) and incubated overnight at 4°C with the primary antibodies diluted in PBS containing 2% BSA and 0.3% Triton X-100. The reactions were visualized by peroxidase immunohistochemistry or immunofluorescence (15,16). The primary antibodies were mouse monoclonal antibodies (mAb) aE11 anti-human C5b-9 (0.5 μg/ml; Dako, Carpinteria, CA), which react with a neoepitope on poly C9 exposed upon formation of MAC; 1A4 anti-human smooth muscle actin (1.8 μ/ml; Dako); E4.3 anti-human CD46 (1 μg/ml; Santa Cruz Biotechnology, Santa Cruz, CA); MEM-43 anti-human CD59 (1 μg/ml; Serotec, Oxford, U.K.); and rabbit polyclonal antibodies to human C3d (1:1,000), C1q (1:1,000), C4c (1:1,000), and von Willebrand factor (vWf; 1:7,000), all from Dako. The rat retinal sections were fixed in acetone and studied with mAb 2A1 anti-rat C5b-9 (1:150; provided by W.G. Couser). Negative controls were obtained by substituting the primary antibodies with an equivalent concentration of nonimmune isotypic mouse IgG or rabbit IgG as appropriate.

For isolation of protein, the fresh retinas were homogenized in ice-cold lysis buffer containing phosphatase and protease inhibitors as described (16). The homogenate was sonicated three times for 2 s and centrifuged at 16,000g for 15 min at 4°C, and the supernatant was collected and stored in aliquots at −80°C. Protein concentration was determined with the Bradford method using BSA as standard (protein assay kit; Bio-Rad, Hercules, CA). Retinal proteins were resolved by SDS-PAGE and immunoblotted as described (14). For the study of CD46, CD55, and CD59, SDS-PAGE was under nonreducing conditions. In the human retinas, CD46 was detected with rabbit polyclonal antibody H-294 (1:1,000; Santa Cruz), CD55 with mAb BRIC 216 (1 μg/ml; Serotec), and CD59 with both mAb MEM-43 (1 μg/ml) and goat polyclonal Ab N-20 (1:1,000; Santa Cruz). Neuron-specific enolase was detected with mAb BBS/NC/VI-H14 (1 μg/ml; Dako). In the rat retinas, CD46, CD55, and CD59 were detected with rabbit polyclonal antibody H-294 (1:1,000), H319 (1:200), and R-79 (1:1,000), respectively (all from Santa Cruz), and β-actin with mouse mAb AC-15 (1:200,000; Sigma, St. Louis, MO).

To determine whether the antibodies chosen could detect CD59 quantitatively in diabetic samples irrespective of glycation, we tested the antibodies against red blood cells (RBCs) CD59 glycated in vitro. Normal human RBCs were incubated in standard storage medium containing 55 mmol/l glucose or in storage medium containing 150 mmol/l glucose for 40 h at 30°C, following a protocol previously described (17). Glycation of hemoglobin was measured with the Glyc-Affin GHb Assay (Perkin Elmer Wallac, Norton, OH). Equal amounts (200 μg) of solubilized protein from the high- and control-glucose RBC were incubated with mouse mAb BRA-10G anti-human CD59 (20 μg/ml; Ancell, Bayport, MN) at 4°C overnight. The immunocomplexes were precipitated by the addition of protein-G Sepharose (Sigma) for 2 h, and the resulting immunoprecipitates were analyzed by Western blotting (14). CD59 glycation, known to lead mostly to formation of Amadori product (13), was measured with a polyclonal antibody (provided by N. Taniguchi) that reacts with ε-(1-deoxyhexitolyl)-lysine, the reduced form of the Amadori product (18). Hence, the blots were treated with NaHBH4 (50 μmol/l in PBS) for 4 h at room temperature, followed by a 15-min wash in PBS acidified to pH 5.0 with acetic acid to stop the reaction. After additional PBS washes, the membrane was blocked and reacted with the anti-hexitol-lysine antibody (1:1,000). The amount of immunoprecipitated CD59 was measured in companion blots probed with the anti-CD59 mAb MEM-43 or polyclonal antibody N-20. Similar immunoprecipitation experiments were performed with lysates of individual retinas of diabetic and nondiabetic donors to assess the level of glycation of retinal CD59 in human diabetes.

Data are summarized as the mean ± SD. Statistical analysis was performed with the unpaired t test.

Radial retinal sections from 12 of the 13 diabetic donors studied showed C5b-9 immunostaining. The staining localized at small and mid-size vessels (Fig. 1), which were mostly negative for smooth muscle actin. In contrast, no staining or only occasional punctiform staining was observed in vessels of nondiabetic donors. Also, the retinal capillaries, not readily visualized in sections and therefore studied in trypsin digest preparations, showed MAC immunostaining in the diabetic but not nondiabetic donors (Fig. 1). In diabetic retinal vessels observed in confocal microscopy, MAC was seen on the endothelial surface, often colocalized with vWf (Fig. 2). Occasionally, MAC was also detected in the middle and outer layers of the wall of large vessels, concentric to the lumen and possibly reflecting cumulation over time in basement membranes (Fig. 2).

In the retinal sections of diabetic donors, MAC showed perfect colocalization with C3, the complement component on which the classical and alternative pathways of activation converge. In confocal microscopy, colocalization was captured on the surface of vascular endothelium (data not shown). In the same retinas, C1q and C4, the two complement components unique to the classical pathway, were either not detected at all or detected occasionally in vascular lumens reflecting their presence in blood (Fig. 3). This pattern is consistent with complement activation occurring in diabetic retinal vessels via the alternative pathway.

Because activation of the alternative pathway of complement is critically modulated by inhibitors, we compared the levels of inhibitors in the retina of diabetic and nondiabetic donors and investigated their topography in the human retina. In SDS-PAGE and Western blot (Fig. 4A), CD55 was detected as a unique band of 70- to 80-kDa apparent molecular mass; its levels were reduced by 48% in the diabetic retinas when compared with the control retinas (P = 0.001). CD59 migrated as a broad band at 20–25 kDa, reflecting a mixture of glycoforms (19), and its levels measured using either Ab MEM-43 or N-20 were decreased as well in the retinas of diabetic donors (P = 0.01). CD46 showed in the retina the two isoforms (65 and 55 kDa) reported in other organs and accounted for by alternative splicing, leading to different degrees of glycosylation (20). An additional band migrating at 48 kDa possibly represents intracellular precursors (20). In contrast to CD55 and CD59, which are anchored to the plasma membrane by GPI linkage, CD46 is a transmembrane protein, and its levels were similar in diabetic and nondiabetic retinas (the sum of the three bands was used to measure CD46 levels in each sample).

The measurement of CD59 levels took into account that this protein is a target of nonenzymatic glycation in diabetes. Exposure of human RBCs to 150 mmol/l glucose in vitro resulted in nonenzymatic glycation of hemoglobin (7.4 vs. 5.1% in control RBCs incubated in 55 mmol/l glucose) and of CD59 (Fig. 4B). However, the intensity of the CD59 band as detected by mAb MEM-43 or polyclonal antibody N-20 was similar in the immunoprecipitate from RBCs exposed to 55 or 150 mmol/l glucose, indicating that documented glycation of CD59 did not affect its reactivity with the two antibodies (Fig. 4B). Confirmation was obtained in immunoblots comparing the levels of CD59 in glycated and nonglycated RBC lysates not subjected to immunoprecipitation (data not shown). There was a suggestion that CD59 was more glycated in diabetic than in nondiabetic retinas (the ratio of the hexitol-lysine to the CD59 signal was 1.3 ± 0.8 in the diabetic and 0.7 ± 0.4 in the nondiabetic retinas), but the difference was not statistically significant.

We did not find antibodies suitable for immunohistochemical detection of CD55 in the fixed human retina, but both CD59 and CD46 could be localized exclusively to vessels (Fig. 5). Retinal vessels thus seem to be the site where the selective decrease in GPI-anchored complement inhibitors occurs in diabetes.

Radial retinal sections from rats with 10 weeks of diabetes (GHb levels 16 ± 4% vs. 4 ± 1% in controls) showed vascular deposition of MAC, similar to that observed in the retina of human diabetic donors (Fig. 6A). No MAC immunostaining was found in the retina of nondiabetic rats. Also in agreement with the findings in human diabetic donors were the decreased levels of CD59 and CD55 but not CD46 in the retina of diabetic rats when compared with levels in nondiabetic rats (Fig. 6B).

We found that retinal vessels, one of the main targets of the long-term effects of diabetes, show evidence of complement activation early in diabetes, associated with a prominent and selective decrease in the levels of GPI-anchored complement inhibitors. This indicates that complement regulation is altered in diabetes and suggests mechanisms for the pathogenesis of diabetic microangiopathy.

The presence of MAC in diabetic retinal vessels is likely to reflect local assembly, possibly on the endothelium. Confocal images captured colocalization of MAC and vWf, which, coupled with the knowledge that the C5b-9 complex becomes partially embedded within the plasma membrane of target cells (21), indicates that vascular endothelium is one of the sites of MAC assembly in diabetic retinal vessels. However, in larger retinal vessels of diabetic donors we also observed MAC deposits removed from the luminal endothelium. Increased permeability of retinal vessels in diabetes cannot be the only explanation for MAC deposited in the vessel wall because there was no evidence of extravasation of complement components (we tested C1q and C4), which have molecular sizes smaller than MAC. Moreover, when complement is activated by the subendothelial extracellular matrix, the reaction seems not to proceed to MAC formation (22). In the renal glomeruli of patients with diabetes, immunoelectron microscopy had shown that MAC is deposited on membranous structures, most likely cellular debris trapped between layers of existing and newly formed lamellae of basement membranes (10). Reduplication of basement membranes is a characteristic feature of diabetic vessels (23). It is thus possible that MAC forms initially on the luminal endothelium of diabetic vessels and that, over time, fragments of disrupted plasma membrane still bearing MAC become embedded within adjacent basement membranes. Such a paradigm would be consistent with the absence of MAC deposits in the wall of the larger retinal vessels of diabetic rats, which were studied after only 10 weeks of diabetes.

The observations in the rat model identify complement activation and decreased levels of inhibitors as early events in the course of diabetes, detectable before other manifestations of retinal vascular pathology. The human retinal vessels were from donors with a duration of diabetes that had set the stage also for the occurrence of vascular cell apoptosis (3) and microthrombosis (1), both of which could contribute to complement activation (4,5). However, these two processes are not detected in the diabetic rat model until after 6–8 months of diabetes (3,14,24), and this points to complement activation as an upstream event. Decreased levels of complement inhibitors seem likewise to be upstream of other known vascular pathology. Complement regulators are lost from necrotic cells and tissues (25), but the prominent decrease in CD55 and CD59 was disproportionate to the small number of apoptotic cells detected at any given time in the retinal vessels of diabetic donors (3) and preceded by several months the development of acellular capillaries in the diabetic rats (3). The decreased levels of inhibitors also cannot be readily attributed to complement activation because the latter results in upregulation of inhibitors (7). The reduced levels of CD55 and CD59 thus may be viewed as a primary effect of diabetes and one of the mechanisms for complement activation in diabetic vessels.

The selective decrease in GPI-anchored complement inhibitors suggests effects of diabetes on common regulatory steps in the synthesis or the processing of these molecules. Studies in cultured endothelial cells have shown that, similar to what we observed in diabetes, high glucose concentrations decreased cellular content of both CD55 and CD59 but not CD46 and that the levels of GPI-anchored inhibitors decreased on the cell surface but increased in the culture medium (12). Transfer to a glucose medium induced in yeast spheroplasts the activation of an endogenous GPI-specific phospholipase C (26). Collectively, these observations propose that a mechanism for the selectively reduced levels of CD55 and CD59 in diabetic retinal vessels may be sought in hyperglycemia-induced activity of phospholipases capable of cleaving GPI anchors. Hyperglycemia can additionally compromise the protective effects of CD59 (and other complement inhibitors) by causing nonenzymatic glycation in the vicinity of the active site (13). In this context, one may ask why patients who have diabetes with poor metabolic control do not manifest symptoms of paroxysmal nocturnal hemoglobinuria. Complete deficiency of CD59 is probably required for the syndrome to become apparent (9). In addition, the mechanism that leads to decreased levels of CD59 (and CD55) in diabetes may be cell- or tissue-specific and not involve RBCs. It was noteworthy in our experiments that the RBCs exposed to high glucose in vitro manifested extensive Amadori adduct formation on CD59 but did not show decreased CD59 levels.

This work proposes several unsuspected pathways whereby diabetes/hyperglycemia may alter the phenotype of vascular cells to induce the characteristic abnormalities of retinal microangiopathy (27). C3a and C5a anaphylatoxins generated during complement activation could contribute to increased permeability and neutrophil adhesion. Even if not lytic, repeated attacks by MAC will, at a minimum, burden the endothelium with the energetic and metabolic cost involved in recovering from the attacks (28) and may lead to a state of chronic activation (6), heightening the risk of acute events such as apoptosis and microthrombosis over time. The finding that diabetes alters the synthesis or processing of GPI-anchored proteins further widens the spectrum of implications for the endothelial phenotype to include possible effects on other GPI-linked molecules and on signaling systems compartmentalized in the plasmalemmal microdomains in which GPI-anchored molecules cluster (26,29). New experiments will address these issues to clarify further the biology of diabetic vascular disease and to test whether the events surrounding complement activation and/or complement activation itself should become targets in preventative strategies.

This work was supported by National Institutes of Health Grant EY 09122, the Juvenile Diabetes Research Foundation Center for Diabetic Retinopathy at the Schepens Eye Research Institute, and the George and Frances Levin Endowment.

We thank Dr. William Couser for mAb 2A1 and Dr. Naoyuki Taniguchi for the antibody to hexitol-lysine.