Under stress, the absence of intermediate filaments from Müller cells in the retina has structural and functional consequences

Intermediate filaments (IFs) are the least understood part of the cytoskeleton. Their composition depends on the cell type and on its developmental and functional state. In some cell types, such as epithelial and muscle cells, IFs provide resistance against physical stress (Fuchs and Cleveland, 1998; McLean and Lane, 1995). It is not known whether IFs provide structural integrity in other cell types.

In astroglial cells – the most abundant cells in the mammalian central nervous system (CNS), with key functions in health and disease – IFs are formed by glial fibrillary acidic protein (GFAP) and vimentin. In many pathological situations, astroglial cells become reactive and increase their expression of GFAP and vimentin; they also begin to re-express nestin, an IF protein normally expressed by immature astrocytes (Eddleston and Mucke, 1993; Eng and Ghirnikar, 1994; Eng et al., 2000; Frisen et al., 1995; Ridet et al., 1997). However, nestin cannot form IFs on its own or with GFAP; it requires vimentin as a polymerization partner (Eliasson et al., 1999). Studies of mice deficient in GFAP (GFAP^–/–) (Gomi et al., 1995; Liedtke et al., 1996; McCall et al., 1996; Pekny et al., 1995), vimentin (Vim^–/–) (Colucci-Guyon et al., 1994) or both (GFAP^–/–Vim^–/–) (Eliasson et al., 1999) have provided important insights into the function of astrocytic IFs in response to stress. In brain or spinal-cord trauma, the absence of astrocytic IFs results in altered glial scar formation, slows wound healing (Pekny et al., 1999) and leads to the increased synaptic loss immediately after brain injury, whereas later on it is associated with increased axonal sprouting and synaptic regeneration (Menet et al., 2003; Wilhelmsson et al., 2004). The absence of IFs in astroglial cells in the retina also leads to improved integration of retinal transplants (Kinouchi et al., 2003; Emsley et al., 2004; Quinlan and Nilsson, 2004; Pekny et al., 2004). However, the function of IFs in astroglial cells in the absence of any challenge remains an enigma.

The retina, an extension of the CNS into the eye, does not enjoy the same degree of mechanical protection as the brain or the spinal cord, which are protected by bone and cerebrospinal fluid. The retina is a thin, laminated structure lining the innermost wall of the eye, with the inner side facing the liquid vitreous body and the outer side adjacent to the choroidea (Dowling, 1987). It contains two types of astroglial cells: astrocytes, which are confined to the innermost layers; and Müller cells, which span the entire thickness. The unique position of Müller cells suggests that they provide structural support for the retina (Müller, 1851; Bringmann and Reichenbach, 2001) and they were reported to express variable amounts of IFs (Bignami and Dahl, 1979; Ekstrom et al., 1988).

Here, we wanted to elucidate whether, as in epithelial and muscle cells (McLean and Lane, 1995; Fuchs and Cleveland, 1998), IFs are also important for providing resistance to mechanical stress in the Müller cells of the retina. We assessed the mechanical integrity of the retina in wild-type, GFAP^–/–, Vim^–/– and GFAP^–/–Vim^–/– mice under normal conditions, and found that IFs of Müller cells provide the retina with structural support against severe mechanical stress. Furthermore, we analysed the pathophysiological relevance of this in oxygen-induced retinopathy (OIR) (Smith et al., 1994), the murine model of retinopathy of prematurity with vascular changes reminiscent of the proliferative stage of diabetic retinopathy.

Mice with null mutations in the GFAP and/or vimentin loci have been described (Eliasson et al., 1999; Pekny et al., 1999). All mice were from a mixed C57BL/129 genetic background and maintained in a barrier animal facility. Mice used in this study were handled in accordance with Swedish animal welfare laws.

For light-microscopy analyses, mice were treated in one of two ways. (1) Eyes of adult mice were excised immediately after the mice had been killed by cervical dislocation or perfusion through the left ventricle with 4% paraformaldehyde (PFA), then postfixed in 4% PFA (`stress-inducing' treatment causing a mechanical stress on the retina). (2) The mice were killed by cervical dislocation and the eyes excised after postfixation of the whole head in 4% PFA overnight (`gentle' treatment that does not cause mechanical stress). For electron microscopy, adult mice were killed by cervical dislocation. The eyes were excised immediately and opened at their corneal margin; the vitreous body was removed and the eyecups were fixed in a buffered mixture of 0.5% glutaraldehyde and 4% PFA overnight and postfixed in 1% osmium tetroxide for 2 hours.

The fixed eyes were embedded in paraffin, cut with a microtome into 6 μm sections through the cornea and parallel to the optic nerve, and stained with hematoxylin and erythrosine (H&E) for morphological examination. For whole-mount analysis, retinas were dissected free and cleared from retinal pigment epithelium, choroidea, sclera and hyaloid vasculature.

The following antibodies were used: rabbit-derived anti-S100 (Z0311, Dako; 1:200); sheep-derived anti-carbonic-anhydrase-II (CA-II) (Biogenesis; 1:50); mouse monoclonal antibody against retinoschisin (Rs1) (3R10, gift from R. S. Molday; 1:10); rabbit-derived anti-fibronectin (A0245, Dako; 1:20); rabbit-derived anti-laminin (9393, Sigma; 1:50); and rabbit-derived anti-neurofilament-M (AB1987, Chemicon; 1:200). Blood vessels were detected with biotinylated isolectin B4 (L2140, Sigma; 20 μg ml^–1), cell nuclei were visualized with TO-PRO 3 (Molecular Probes; 1:1000). Corresponding Alexa-488- or -568-conjugated secondary antibodies, Alexa-568-conjugated streptavidin (Molecular Probes; 10 μg ml^–1) and FITC-conjugated sheep immunoglobulin (FI-6000, Vector Laboratories; 1:100) were used.

For immunodetection on sections, antibodies were diluted in PBS with 0.1% bovine serum albumin (BSA) and 0.05% Triton X-100, incubated for 1 hour at room temperature, and washed with PBS. For retinal whole mounts, the dilution buffer was PBS with 0.5% BSA and 0.25% Triton X-100. Isolectin was diluted in PBS, pH 6.8, containing 1% Tween-20, 0.1 mM CaCl₂, 0.1 mM MgCl₂ and 0.1 mM MnCl₂. Whole mounts were permeabilized in PBS containing 1% BSA and 0.5% Triton X-100, incubated at 4°C overnight, washed three times with the above dilution buffer for 30 minutes each and mounted with Mowiol 4-88 (Hoechst) supplemented with 2.5% antibleaching agent (DABCO, Sigma).

Retinas were analysed by fluorescence microscopy with a Nikon E1000 microscope equipped with a digital camera (Nikon Coolpix 990) and by laser-scanning confocal microscopy with a Leica LCS NT microscope. Digital images were processed with Adobe Photoshop 7.0.

The fixed eyecups were rinsed, dehydrated in ethanol and incubated overnight in 70% ethanol saturated with uranyl acetate. After further dehydration in absolute ethanol and propylene oxide, the samples were embedded in Araldite 502 (Sigma, Steinheim, Germany) and sectioned on a Reichert FCR Ultracut ultramicrotome (Reichert, Bensheim, Germany). Ultrathin sections were stained with uranyl acetate and lead citrate, and examined with a Zeiss EM 10 electron microscope.

Neovascular retinopathy was induced as described (Smith et al., 1994). Briefly, postnatal day 7 (P7) mice were placed with their mothers in an airtight incubator and exposed to 75% oxygen for 5 days at 20°C. The oxygen concentration was checked twice daily with an oxygen analyser. At P12, the mice were removed from the incubator and placed in room air; some mice were killed (by cervical dislocation), the eyes were excised immediately and fixed in 4% PFA, and blood vessel regression was assessed on whole mounts. The remaining mice were killed at P17 by either cervical dislocation or perfusion fixation, or at P19 by cervical dislocation, and the eyes were prepared by the `standard' procedure or used for RNA preparation.

Neovascularization was quantified as the mean number of vascular cell nuclei on the vitreal side of the inner limiting membrane (ILM) in 20 H&E-stained sections (Smith et al., 1994). P17 littermates not exposed to hyperoxia served as controls. In whole-mount retinas, vessels were visualized using isolectin, and the extent of revascularization was calculated as the ratio of the vessel-free areas to the whole retinal area (Sennlaub et al., 2001).

Total RNA was extracted from individual retinas with the RNeasy kit (Qiagen). Purified RNA was treated at 37°C for 15 minutes with 10 units DNase I (Amersham Pharmacia Biotech) in 40 mM Tris-HCl, pH 7.5, containing 6 mM MgCl₂. RNA (1 μg) was reverse transcribed with AMV reverse transcriptase (1483188, Roche) and random primer (dN)₆.

PCR primers against the gene encoding vascular endothelial growth factor A (VEGF-A) and the housekeeping gene HPRT (hypoxanthine-guanine phosphoribosyltransferase) were designed with Primer Express software (Perkin-Elmer) and synthesized by Oligos Etc. (Wilsonville, OR). The sequences were as follows: VEGF-A, forward 5′-AGTCCCATGAAGTGATCAAGTTCA-3′ and reverse 5′-TCCGCATGATCTGCATGG-3′; HPRT, forward 5′-TATCAGACTGAAGAGCTACTGTAATGATC-3′ and reverse 5′-TTACCAGTGTCAATTATATCTTCAACAATC-3′. The VEGF-A probe sequence was 5′-FAM-TGCCCACGTCAGAGAGCAACATCAC-TAMRA-3′ the HPRT probe sequence was 5′-JOE-TGAGAGATCATCTCCACCAATAACTTTTATGTCCC-TAMRA-3′^†. Quantitative real-time polymerase chain reactions (PCRs) were performed with the TaqMan universal PCR master mix kit (PE Applied Biosystems) and an ABI Prism 7700 sequence detector using the manufacturer's protocol (http://www.appliedbiosystems.com/support/apptech/). Samples were run in triplicate. Gene expression was normalized to HPRT expression as internal control. Data are expressed as the percentage of expression in P17 wild-type control mice not exposed to OIR.

Whole-mount retinas of wild-type and mutant mice were examined at P0, P3, P5, P7 and P9. Vessels were visualized with isolectin as described above and astrocytes with antibodies against S100 and fibronectin [an extracellular matrix protein produced both by astrocytes and endothelial cells (Jiang et al., 1994)]. The distance from the center of the optic nerve to the leading front of developing vessels was measured and compared with the total length of the retina in each retinal quadrant (Stalmans et al., 2002) in three to four retinas at each time point. Adobe PhotoShop 7.0 was used for the quantification.

Values are mean ± s.e.m. A two-tailed Student's t-test was used for statistical evaluation. Differences were considered to be significant at P<0.05.

The retinal morphology of wild-type (n=6) and GFAP^–/–Vim^–/– mice (n=5) was first assessed on H&E-stained sections prepared from mice killed by cervical dislocation and eyes excised after overnight fixation in situ (`gentle' tissue preparation). The overall layered structure and abundance of individual cell types was comparable between the two genotypes (Fig. 1). Next, the retinal morphology of wild-type, GFAP<sup>–/–</sup>, Vim<sup>–/–</sup> and GFAP<sup>–/–</sup>Vim<sup>–/–</sup> mice killed by either cervical dislocation or perfusion fixation was studied after `stress-inducing' preparation (i.e. excision of the eye before postfixation) (Fig. 2; wild-type, n=50; GFAP^–/–, Vim^–/– and GFAP^–/–Vim^–/– mice, n=25 each). Also, the single mutants showed the normal layered structure of the retina, with comparable abundance of individual cell types (Fig. 2B,C). However, upon the `stress-inducing' preparation of the eyes, the ILM (Fig. 2F) and a variable amount of adjacent tissue including parts of Müller cells in the ganglion cell layer (Fig. 2H) were completely separated from the other retinal layers in all GFAP^–/–Vim^–/– mice and partially separated in all Vim^–/– mice (Fig. 2C,D,F,H). Such a separation was never seen in wild-type or GFAP^–/– mice (Fig. 2A,B,E,G) and was independent of whether the mice were killed by cervical dislocation (GFAP^–/–Vim^–/–, n=15; Vim^–/–, n=15) or perfusion fixation (GFAP^–/–Vim^–/–, n=10; Vim^–/–, n=10). In GFAP^–/–Vim^–/– retinas, neurons were normally distributed within the retinal layers (Fig. 2D) and the distribution and morphology of astrocytes and Müller cells (visualized by antibodies against S100) appeared normal (Fig. 2H). Moreover, immunostaining of wild-type and GFAP^–/–Vim^–/– retinas showed comparable and normal distribution of retinoschisin (Fig. 2J), a cell-adhesion protein in photoreceptor and retinal bipolar cells that was found to be missing in patients with retinal fragility and proposed to play a role in retinal-cell-layer architecture (Molday et al., 2001; Weber et al., 2002).

We then asked whether the above-mentioned separation of the ILM and inner retinal tissue occurred during the excision of a live eye or during the fixation procedure. For this purpose, eyes from wild-type and GFAP^–/–Vim^–/– mice killed by cervical dislocation were prepared by the `stress-inducing' procedure before immersion fixation and transmission electron microscopy of retinas was performed. In wild-type mice, the ILM was consistently nearly straight and consisted of a basal lamina overlying a regular arcade-like pattern of thick, funnel-shaped Müller-cell endfeet<sup>‡</sup>, which contained abundant rough endoplasmic reticulum. As previously reported (Bignami and Dahl, 1979; Kinouchi et al., 2003), IFs were clearly present (Fig. 3A) and the endfeet formed a boundary between the vitreous body (and basal lamina) and bundles of optic axons in the nerve fiber layer. In the GFAP<sup>–/–</sup>Vim<sup>–/–</sup> mice, however, the ILM was widely detached from the rest of the retina and the Müller cells had small, irregularly shaped `endfeet' – often with only very thin lamellae (Fig. 3B) and ragged borders (Fig. 3C) – and their cell processes were devoid of IFs. In contrast to wild-type retinas, GFAP^–/–Vim^–/– retinas contained many blood vessels not covered by Müller-cell processes. Instead, the basal lamina of the pericytes formed the innermost surface of the retinal edge but was not continuous with the ILM. In some instances, detached blood vessels and adhering axon bundles accompanied by small Müller-cell compartments were found between the detached ILM and the other retinal layers in GFAP^–/–Vim^–/– retinas (Fig. 3D).

Most of the Müller cell endfeet in GFAP^–/–Vim^–/– retinas were torn off and stayed connected with ILM after it separated from the rest of the retina. However, the membranes of the processes of these cells and their cytoplasm appeared to be intact (Fig. 3B-D). This finding suggests that the ILM detachment occurred before fixation, because living cells close membrane defects within seconds (Newman, 1984) and only a few Müller-cell processes were found with an incomplete membrane and missing cytoplasm (Fig. 3E). The detachment was probably caused by mechanical stress on the eyecup due to the release of intraocular pressure after removal of the vitreous body before fixation for electron microscopy. When the vitreous body was left in situ during fixation, the ILM remained attached to the rest of the retina (Kinouchi et al., 2003). Thus, the absence of GFAP and vimentin in Müller cells in the retina appears to weaken it in a specific region, resulting in separation during mechanical stress.

To assess the pathophysiological effects of the decreased mechanical stability of the retina, we subjected the mice to OIR. In this model (Fig. 4A), the neovascular response is maximal at P17 (Smith et al., 1994) and occurs in the same region where the retinal fragility was observed. Both retinal cross sections and whole-mount preparations from mice killed either by cervical dislocation or retrieved immediately after perfusion fixation were examined.

On sections from P17 mice exposed to OIR, independent of the killing paradigm, the ILM and adjacent tissue were completely separated from the rest of the retina in all GFAP^–/–Vim^–/– mice and partially separated in all Vim^–/– mice; no separation was seen in wild-type or GFAP^–/– mice (Fig. 4B-E). The Müller-cell endfeet in wild-type and GFAP^–/–Vim^–/– retinas were strongly immunopositive for CA-II, a marker of Müller cells (Linser et al., 1984), and the retinal separation in GFAP^–/–Vim^–/– mice invariably occurred within this layer (Fig. 4G). On both sides of the separation, the endfeet had a ragged appearance (Fig. 4E, G) similar to that in adult GFAP^–/–Vim^–/– retinas exposed to `stress-inducing' preparation (Fig. 2D,H, Fig. 3C). In sections obtained from eyes that were postfixed in situ overnight and handled with extreme care, the separation of the ILM was almost undetectable. This suggests that the ILM in GFAP^–/–Vim^–/– mice was not detached in the course of the neovascular response but became separated as a consequence of a physical stress applied during the eye removal.

The vascular response in the OIR model was then analysed. Whole mounts of P12 retinas showed that the vasculature had regressed to the same extent in all genotypes (i.e. the response to hyperoxia was identical) (data not shown). Next, we assessed the degree of neovascularization into the vitreous body that occurs in response to the relative hypoxia, at P17 (Fig. 5A-D). GFAP^–/–Vim^–/– mice had 66-79% fewer cell nuclei in the vitreous body than wild-type mice [6.1±1.2 (n=6) vs 28.5±0.05 (n=2), P<0.01; 10.5±1.5 (n=6) vs 31.1±5.8 (n=6), P<0.001; two independent experiments] and 74% fewer capillary tufts [2.7±0.25 (n=6) vs 10.8±0.6 (n=6), P<0.001]. In another experiment, the number of cell nuclei was 69% lower in GFAP^–/–Vim^–/– than in wild-type mice [6.6±1.0 (n=6) vs 21.2±3.4 (n=5), P<0.002] and 55% lower in Vim^–/– mice [9.5±2.5 (n=4), P<0.05], but not significantly different in GFAP^–/– mice (19.0±1.6, n=5). In yet another experiment, Vim^–/– mice had 31% fewer cell nuclei than wild-type mice [22.4±2.0 (n=6) vs 32.5±1.8 (n=7), P<0.005], demonstrating a dose response in Vim^–/– but not GFAP^–/– mice.

The analysis of retinal cross sections at P19 (i.e. 2 days later) showed that GFAP^–/–Vim^–/– retinas still had 61% fewer cell nuclei in the vitreous body than wild-type retinas [15.8±3.4 (n=4) vs 40.4±3.3 (n=5), P<0.001] (Fig. 5E). This suggests that the intravitreal part of the neovascular response to OIR in GFAP^–/–Vim^–/– mice was not merely delayed.

We then asked whether it was only the intravitreal component of the neovascular response to OIR that was impaired in GFAP^–/–Vim^–/– mice. For this purpose, we analysed the neovascular response to OIR by assessing the extent of vascular-free zones in retinal whole mounts at P17 (Fig. 5F,G). No significant differences were found between wild-type and GFAP^–/–Vim^–/– mice [15±2% (n=7) vs 18±2% (n=5)] (Fig. 5H). These data suggest that the formation of new blood vessels in response to relative hypoxia is not per se altered in GFAP^–/–Vim^–/– mice.

To substantiate this finding, we analysed the expression of VEGF-A-encoding RNA in wild-type and GFAP^–/–Vim^–/– P17 retinas by real-time quantitative PCR (TaqMan). VEGF-A is produced by astrocytes and is the most powerful neovascularization-driving growth factor (Robinson and Stringer, 2001). In retinas not subjected to OIR, VEGF-A expression was similar in wild-type and GFAP^–/–Vim^–/– mice [100±24% (n=5) vs 86±26% (n=7)] (Fig. 5I). In the ischemic retinas, VEGF-A expression was about two times higher in both wild-type and GFAP^–/–Vim^–/– mice than in controls [198±17% (n=5) vs 248±38% (n=9)] (Fig. 5I). Thus, astrocytes devoid of IFs produce and secrete normal amounts of VEGF-A both under normal and ischemic conditions.

During development, astrocytes spread from the optic nerve head and form a network that precedes and drives the formation of the retinal vascular system (Fruttiger et al., 1996; Gerhardt et al., 2003; Stone and Dreher, 1987). To determine whether this process is altered in GFAP^–/–Vim^–/– mice, we assessed the formation of the astrocytic network and quantified the extent of vascularization in the developing retina (Fig. 6). We did not detect any difference in the appearance of the astrocytic and vascular networks between wild-type and GFAP^–/–Vim^–/– retinas, as revealed by S100, isolectin and fibronectin immunostaining at P7 (Fig. 6A-D). In addition, the progression of the vascular network, determined as a proportion of the retinal radius, was similar in wild-type and GFAP^–/–Vim^–/– retinas at each of the five developmental stages examined (Fig. 6E,F). These data show that IF-deficient astrocytes support normal vascularization in developing GFAP^–/–Vim^–/– retinas.

Despite an increased understanding of the function of IFs in astroglial cells (for a review, see Pekny, 2001), the possible role of astroglial IFs in maintaining the mechanical integrity of these cells and of the CNS tissue remains to be elucidated. We show here that the absence of IFs from retinal Müller cells results in decreased resistance of the retina to mechanical stress. In GFAP^–/–Vim^–/– mice, a stress that did not affect the structure of the retina in wild-type or GFAP^–/– mice invariably led to a complete separation of the ILM and adjacent tissue from the rest of the retina. This separation occurred in the retinas of all GFAP^–/–Vim^–/– mice in which the eyes were not fixed in situ, irrespective of whether the mice were killed by cervical dislocation or by formalin perfusion. In Vim^–/– mice, whose astroglial cells contain fewer abnormally tightly bundled IFs composed solely of GFAP (Eliasson et al., 1999), the same stress consistently led to a partial separation. Ultrastructural examination revealed that the separations occurred within the endfeet of Müller cells, which are normally rich in IFs but lack IFs in GFAP^–/–Vim^–/– mice.

This had a clear pathophysiological consequence and resulted in a reduced frequency of crosses of newly formed blood vessels into the vitreous body in retinal ischemia. The intravitreal neovascularization was reduced by more than two-thirds in GFAP^–/–Vim^–/– mice and by one-third to one-half in Vim^–/– mice (Fig. 5A-D).

The dose effect seen in Vim^–/– mice in both the degree of the ILM detachment and the reaction to retinal ischemia might reflect the higher ratio of vimentin to GFAP in Müller cells than in mature astrocytes (Bignami and Dahl, 1979), which would result in a more prominent reduction of IFs in Vim^–/– than in GFAP^–/– Müller cells. There are at least two possible ways to connect the absence of IFs in Müller cells and the weakness of the region just beneath the ILM with the impaired growth of newly formed blood vessels into the vitreous body. The first scenario is that the penetration of the ILM by the vessels requires structural support in the region just beneath the ILM, which is available in wild-type but not in IF-deficient mice. The other scenario is that, in wild-type mice, penetration of the ILM by newly formed vessels is preferred because the inner retinal layers are tightly packed with IF-reinforced Müller cell processes and endfeet, which provide resistance to intraretinal growth of blood vessels; in the absence of IFs in Müller cells, this resistance is reduced, and blood vessel growth might easily occur within the retina. Both scenarios would accommodate the dose effect seen in Vim^–/– retinas, in which GFAP can be expected to compensate partially for the loss of vimentin (Galou et al., 1996; Eliasson et al., 1999; Pekny, 2001). These scenarios are also compatible with only a limited hypertrophy of cellular processes observed in GFAP^–/–Vim^–/– reactive astrocytes (Wilhelmson et al., 2004).

The neovascular response to ischemia was normal in GFAP^–/–, Vim^–/– and GFAP^–/–Vim^–/– mice, as demonstrated by the normal density of newly formed blood vessels in retinal whole mounts. In all mutant mice, the neovessels were structurally normal and equally abundant. These findings show that the absence of vimentin, and thus of IFs, in vascular endothelial cells in both GFAP^–/–Vim^–/– and Vim^–/– mice (Pekny et al., 1999) does not adversely affect the formation, growth or stability of new blood vessels. Thus, the observed phenotype results from the absence of IFs (GFAP^–/–Vim^–/–) and partial depletion of IFs (Vim^–/–) specifically in Müller cells.

Astroglial cells are highly involved in retinal vascularization because they are the main producers of VEGF-A during both normal (Stone et al., 1995) and pathological (Ozaki et al., 2000) vessel formation. The vascular system developed normally in GFAP^–/–Vim^–/– mice, and VEGF-A RNA was expressed at normal levels in both ischemia-exposed and control GFAP^–/–Vim^–/– retinas. This suggests that the absence of IFs in astroglial cells does not affect their function in retinal vascularization in either health or disease.

Our findings show that partial or complete deficiency of IFs in Müller cells in the retina does not have any structural consequences under normal conditions. However, IF deficiency causes weakness in the endfeet of Müller cells that becomes apparent during mechanical stress and in response to pathological situations that place increased structural and functional demands on the retina. We conclude that IFs are important for the ability of Müller-cell endfeet, and thus the retina, to withstand mechanical stress, and also for newly formed retinal vessels to traverse the ILM.

We thank R. S. Molday (Department of Ophthalmology, University of British Columbia, Canada) for antibodies against retinoschisin and H. Gerhardt and M. Pekna (The Sahlgrenska Academy at Göteborg University, Sweden) for valuable discussions. This study was supported by grants from the Swedish Cancer Foundation (project 3622), the Swedish Medical Research Council (project 11548), the Swedish Society for Medicine, the Swedish Society for Medical Research, the King Gustaf V Foundation, Volvo Assar Gabrielsson Fond, the Swedish Stroke Foundation and the Novo Nordisk Foundation Program on Diabetic Microangiopathy. A.R. was supported by the Bundesministerium für Bildung, Forschung und Technologie (BMB+F), Interdisciplinary Center for Clinical Research at the University of Leipzig (01KS9504, Project C5), and Deutsche Forschungsgemeinschaft (RE 849/8-2). H.W. was supported by the Tübingen Fortüne Program (1038-0-0).