Loss of pericytes from the capillary wall is a hallmark of diabetic retinopathy, however, the pathogenic significance of this phenomenon is unclear. In previous mouse gene knockout models leading to pericyte deficiency, prenatal lethality has so far precluded analysis of postnatal consequences in the retina. We now report that endothelium‐restricted ablation of platelet‐derived growth factor‐B generates viable mice with extensive inter‐ and intra‐individual variation in the density of pericytes throughout the CNS. We found a strong inverse correlation between pericyte density and the formation of a range of retinal microvascular abnormalities strongly reminiscent of those seen in diabetic humans. Proliferative retinopathy invariably developed when pericyte density was <50% of normal. Our data suggest that a reduction of the pericyte density is sufficient to cause retinopathy in mice, implying that pericyte loss may also be a causal pathogenic event in human diabetic retinopathy.
All capillaries are partially covered by mural cells of the vascular smooth muscle cell (vSMC) lineage, referred to as pericytes. A number of murine gene knockout mutations interrupting signaling by platelet‐derived growth factor (PDGF)‐B/PDGF receptor‐β (PDGFR‐β), angiopoietin‐1/Tie‐2, TGF‐β and EDG‐1 lead to pericyte‐ and/or vSMC deficiency. These mutants are all lethal before birth due to cardiovascular dysfunction (Levéen et al., 1994; Soriano, 1994; Suri et al., 1996; Lindahl et al., 1997; Patan, 1998; Li et al., 1999; Yang et al., 1999; Liu et al., 2000; Oh et al., 2000). TGF‐β signaling appears to be critical for early vSMC/pericyte formation, whereas PDGF‐B/R‐β and EDG‐1 signaling seems to be involved in later proliferation and migration of these cells. Angiopoietin‐1 targets primarily Tie‐2 receptors on endothelial cells and its role in vSMC/pericyte formation might therefore be indirect.
The only disease strongly linked to pericyte deficiency is diabetic retinopathy. The pericyte density is higher in retinal than in other capillaries (Sims, 1986), probably reflecting a particularly important function at this location. Pericyte loss is the earliest morphological sign of retinal vascular abnormalities in diabetes (Cogan et al., 1961); however, whether it constitutes a causal event in, or a mere consequence of, the pathogenesis of diabetic retinopathy is unknown. Pericytes may produce survival signals for endothelial cells (Benjamin et al., 1998, 1999), and loss of pericytes may therefore provoke endothelial death and the formation of acellular (regressing) capillaries, which are typically increased in diabetic retinopathy. It has also been suggested that pericyte loss may cause local weakenings leading to outpouchings (microaneurysms) in the capillary wall (Cogan et al., 1961; Buzney et al., 1977). Lack of pericytes in PDGF‐B‐ or PDGFR‐β‐deficient embryos leads to the formation of capillaries with highly variable diameter (Hellström et al., 2001). Since vessel wall tension is proportional to its radius at constant blood pressure (law of Laplace), focal dilations may rapidly expand to microaneurysms. Indeed, pericyte‐deficient PDGF‐B and PDGFR‐β null mice develop numerous rupturing microaneurysms at late gestation, when blood pressure is increasing (Levéen et al., 1994; Soriano, 1994; Lindahl et al., 1997; Hellström et al., 1999). These vascular lesions are reminiscent of those seen in human diabetic microangiopathy, providing support for a causal role of pericyte loss in the pathogenesis of this disease. Moreover, the importance of the PDGF‐B/PDGFR‐β signaling pathway for pericyte recruitment in the retina was demonstrated recently in a mouse model in which the intracellular domain of PDGFR‐β was exchanged for that of PDGFR‐α (Klinghoffer et al., 2001). These mice develop severe vascular changes and retinopathy, most likely because of failure of proper pericyte recruitment. Data from diabetic rodents, however, conflict somewhat with the view that pericyte loss is a causal event in diabetic retinopathy. Such animals lose up to 50% of the retinal pericytes (Buscher et al., 1989; Engerman, 1989) and develop increased numbers of acellular capillaries, but microaneurysms are rare, and more importantly, progression into the severe proliferative type of retinopathy is never seen. Thus, only the earliest and mildest signs of human diabetic retinopathy are seen in diabetic rodents, in spite of significant pericyte loss.
Here, we have used the Cre‐loxP system to target an inactivating mutation to the mouse PDGF‐B gene selectively in endothelial cells. Such mutant mice survived postnatally and showed a high degree of inter‐ and intra‐individual variation in the recruitment of pericytes to blood vessels in the CNS, including the retina. Hence, they were suitable for the analysis of the consequences of different extents of pericyte deficiency in the retina. We took advantage of the variation in pericyte recruitment to establish a strong correlation between the degree of pericyte reduction and the development of a spectrum of retinal vascular changes reminiscent of those observed in non‐proliferative, as well as in proliferative, diabetic retinopathy.
Generation of mice with endothelium‐restricted deletion of PDGF‐B
We placed loxP sites on each side of PDGF‐B exon 4, thereby producing a PDGF‐Bflox allele, and mice carrying this allele were generated (Figure 1 and Materials and methods). PDGF‐Bflox/flox and PDGF‐Bflox/− mice were normal, demonstrating functional activity of the PDGF‐Bflox allele. We crossed mice expressing the Cre recombinase under the control of the endothelial Tie‐1 promoter (Tie1Cre) (Gustafsson et al., 2001) with PDGF‐B+/− mice, and subsequently with PDGF‐Bflox/flox mice. In the resulting litters, endothelium‐restricted mutants (PDGF‐Blox/−) appeared together with different types of control analogous to PDGF‐B+/+ and PDGF‐B+/− mice (see Materials and methods for details). In contrast to PDGF‐B−/− mice, which are embryonic lethal (Levéen et al., 1994), PDGF‐Blox/− mice were born at Mendelian ratios, reached adulthood and were fertile. To assess the efficiency of Cre‐mediated recombination at the PDGF‐B locus in capillary endothelial cells in vivo, we isolated capillary fragments from brains of embryonic day (E) 16.5 PDGF‐Blox/+ embryos, and determined the relative abundance of the flox and lox alleles by PCR (Figure 1D). There was a noticeable individual variation in flox:lox PCR fragment ratio in different individuals. Assuming equal efficiency of amplification of the wild type, flox and lox alleles, the Cre‐mediated recombination of the flox allele in endothelial cells from different individuals varied between ∼20 and 90% (Figure 1D).
Impaired pericyte recruitment in the CNS correlates with development of retinopathy
We bred the Tie1Cre transgene and the different PDGF‐B alleles onto the XlacZ4 background in order to visualize and quantify pericyte recruitment to CNS microvessels. The XlacZ4 transgene is expressed in vSMCs and pericytes from late gestation onwards in correlation with other markers for these cells, such as α‐smooth muscle actin, desmin and NG2 (Klinghoffer et al., 2001; Ozerdem et al., 2001; Tidhar et al., 2001; Abramsson et al., 2002; Stalmans et al., 2002). Whole‐mount staining of E15.5 brains visualized the pericyte abundance and distribution in superficial CNS vessels and revealed that the pericyte density in PDGF‐Blox/− embryos was intermediate between that of PDGF‐B+/+ and PDGF‐B−/− embryos at this site. However, there was a noticeable inter‐individual variation in the pericyte density between PDGF‐Blox/− embryos, ranging from near normal to near complete absence (Figure 2A–D and data not shown).
The pericyte deficiency in PDGF‐Blox/− mice did not normalize postnatally. The density of XlacZ4‐positive pericytes and vSMCs was reduced in PDGF‐Blox/− mutants at several sites in the CNS and in pial vessels at 3–4 weeks of age (Figures 2E and F and 3). The reduction affected arteries, veins and capillaries, which all showed intermittent stretches with very sparse coverage of XlacZ4‐positive cells. Quantification of pericytes in four regions of the brain (pial plexus, forebrain cortex, thalamus and cerebellum) showed a similar overall degree of reduction in different regions of the same mouse, but extensive variation between individual PDGF‐Blox/− mice (Figure 3). The retinas in individuals with the lowest overall CNS pericyte density, however, deviated from this correlation by displaying focal regions of increased pericyte density (Figure 3D–F red asterisks and bars). Such retinas also showed gross abnormalities; they were contracted (Figure 3D1–F1) and often attached to the retinal pigment epithelial (RPE) cells and the lens.
At high magnification we could see distinctive vascular aberrations in all PDGF‐Blox/− individuals, including variable capillary and venous diameter, irregular capillary density, abnormal capillary ring structures, increased number of regressing capillary branches and the presence of microaneurysms (Figure 4D and E and data not shown). The abundance of these types of abnormality correlated with the degree of pericyte deficiency, but in the most severely affected retinas (e.g. Figure 3D–F), they were found only in regions with low pericyte density (Figure 4B and E). In the regions with highly increased pericyte density there was a complete loss of regular vascular pattern and formation of a dense, chaotic vascular network with large numbers of endothelial cells, pericytes and possibly also other cell types (Figure 4C). We confirmed these observations by studying vascular preparations from protease‐digested retinas (Figure 4F–H). These preparations also revealed in the same retinas areas of rather normal overall vascular pattern showing an increased number of acellular capillaries and pyknotic nuclei, as well as regions of chaotic vascular organization with markedly increased vascular density (Figure 4F–H).
In the regions of increased vascular density, we also saw vessel branches penetrating into the vitreous as well as from the choroid through the photoreceptor and RPE layers (Figure 5A–H). Thus, the regions with high pericyte density show the typical hallmarks of proliferative retinopathy. These regions also showed loss of organization of the neural layers and folding of the photoreceptor layer producing typical photoreceptor rosette profiles (Figure 5A–H). Analysis of 52 mice of different genotypes showed a strong correlation between the overall CNS pericyte density and the presence of proliferative retinopathy. We chose the cerebellum as reference site for pericyte quantification in the CNS because it was easy to select analogous anatomical locations for analysis, and because cerebellum pericyte density provided a good reflection of the degree of overall CNS pericyte density (Figure 3). All individuals with <52% of the normal pericyte density in the cerebellum showed patches of proliferative retinopathy affecting at least one eye, whereas none of the individuals with >52% of the normal pericyte density showed any signs of proliferative retinopathy (Figure 5I).
Capillary regression in the outer retinal plexus correlates with proliferative retinopathy
Two interconnected vascular plexuses develop in the retina, an inner (superficial) plexus with arteries, veins and capillaries, and an outer (deep) plexus consisting mostly of capillaries. In PDGF‐Blox/− mice, both plexuses displayed characteristic abnormalities including vascular occlusion and regression. To address whether this correlated directly with pericyte deficiency, we simultaneously scored the frequency and pattern of capillary occlusion and pericyte density in the outer plexuses in a number of PDGF‐Blox/− mutants (Figure 6A). Ten random fields were analyzed in each retina, representing together about half of the retinal area. In controls, branch occlusions were observed in regular patterns. This is known to be part of the normal vessel remodeling and it affects mostly singular branches interconnecting neighboring plexus units, as defined by supplying arterioles (not shown). In contrast, PDGF‐Blox/− mutants displayed both an increased density of regression profiles and a distinctive change in their pattern; several sequential branches were often affected, leading to the formation of characteristic Y‐shaped, or more complex, regression profiles (Figure 6A). In addition to an inter‐individual correlation between pericyte deficiency and regression (Figure 6B), there was a similar intra‐individual correlation in different retinal regions, seen as an inverse correlation between number of pericytes and number of regression profiles in individual microscopic fields (Figure 6C and D). A complete lack of pericytes correlated with the complete regression of the outer plexuses (Figure 6C3). This, in turn, correlated with the occurrence of proliferative changes at the retinal surface and the formation of abnormal vascular penetration of the bottom layers of photoreceptor and RPE cells, as shown by confocal z‐scans (Figure 5A–D).
Previous studies have shown that pericyte progenitors express PDGFR‐β and require PDGF‐B for their recruitment to new vessels in the course of angiogenesis (Lindahl et al., 1997; Hellström et al., 1999; Klinghoffer et al., 2001). Loss of pericytes in these mutants leads to microvascular changes similar to those typical of diabetic microangiopathy, such as the formation of microaneurysms and increased microvascular leakage. Whereas the complete knockouts of PDGF‐B or PDGFR‐β are lethal during late embryonic development or at birth (Levéen et al., 1994; Soriano, 1994), the exchange of the intracellular domain of PDGFR‐β for that of PDGFR‐α produced viable mice; however with a severe deficit in retinal pericytes and the development of retinopathy (Klinghoffer et al., 2001). These and other studies on the cardiovascular defects in PDGF‐B and PDGFR‐β mutant mice (Crosby et al., 1998; Lindahl et al., 1998; Ohlsson et al., 1999; Tallquist et al., 2000; Hellström et al., 2001), have demonstrated the critical importance of PDGF‐B/PDGFR‐β signaling for proper pericyte recruitment in developmental angiogenesis in a number of different organs, including the retina.
In contrast to the previous models of genetic PDGF‐B or PDGFR‐β deficiency, the endothelium‐restricted PDGF‐B knockout reported here produces a wide spectrum of pericyte‐deficient states, which vary at both the inter‐ and intra‐individual levels. This variation is most likely dependent on different degrees of chimerism with regard to PDGF‐B‐negative and PDGF‐B‐positive endothelial cells occurring as a result of variation in Cre‐mediated recombination efficiency. The reason for this variation is unclear; however, genetic background is likely to play a role as we have noticed a significant degree of familial clustering of individuals with severe or mild pericyte loss and retinal changes, respectively (data not shown). Due to the crossing of mice carrying PDGF‐B null, and flox alleles with Tie1Cre transgenics, the genetic background is a mixture of C57Bl6, 129 and CBA. Further inbreeding of PDGF‐Blox/− mutants should allow us to determine how strong the genetic component is, and to what extent epigenetic mechanisms may also play a role, and if the variation depends on Cre‐protein level, recombination efficiency at the PDGF‐B locus, or both.
Irrespective of mechanism underlying the variation in Cre‐mediated recombination and PDGF‐B expression, the resulting inter‐ and intra‐individual variation in pericyte density allowed us to analyze the effects of a wide range of pericyte‐deficient states, ranging from near‐normal to near‐complete lack of such cells. In mutants with >50% of normal overall CNS pericyte density, the retinal vasculature displayed irregular microvessel diameter, microaneurysms and increased vascular regression. In individuals with <50% of normal pericyte density, the retinas developed regions with a massive increase of abnormal vessels extending into the vitreous and choroid. The somewhat paradoxical pericyte proliferation associated with these proliferative changes is apparently independent of endothelium‐derived PDGF‐B, suggesting that different mechanisms govern pericyte proliferation in association with normal or pathological angiogenesis. Importantly, regions of vascular proliferation in the inner plexus, complete outer plexus regression, and vitreous‐ and choroid neoangiogenesis, correlated without exception, and bordered sharply to non‐proliferative regions. The inverse correlation between the numbers of pericytes and regressing capillaries at both the inter‐ and intra‐individual levels strongly suggests that pericyte deficiency triggers capillary occlusion. This may be tolerable up to a threshold level, above which neoangiogenic responses are initiated, leading to proliferative retinopathy.
The endothelium‐specific PDGF‐B mutant represents a model of pericyte deficiency independent of diabetes. Since these mice develop a broad spectrum of the retinal vascular changes highly similar to the different stages of diabetic retinopathy, our data provide strong support for the view that pericyte loss constitutes an early and important causal event in the pathogenesis of diabetic retinopathy. Our data also suggest that pericyte deficiency in the retina is tolerable down to a threshold level below which proliferative retinopathy will develop. The reason why this level is not reached in response to diabetes in rodents, and consequently why diabetic retinopathy does not progress into proliferative states, remains to be established. It is possible that human and rodent pericytes are differently sensitive to hyperglycemic challenges, or alternatively, that human and rodent retinal vessels are differentially sensitive to pericyte deficiency. It is also likely that duration of the diabetic state (months in rodents compared with decades in humans) contributes to the species differences in disease progression. In humans, duration of diabetes is a major risk factor for retinopathy, and correlates with its severity (Klein et al., 1984a,b).
Materials and methods
Generation of PDGF‐Bflox mice
We generated a targeting vector in which the fourth exon, which codes for the major part of the PDGF‐B protein, was flanked by a loxP‐flanked PGK‐neo cassette in intron 3 and a single loxP fragment in intron 4 (Figure 1). We linearized the targeting vector with NotI and transfected and selected G418‐selected E14.1 embryonic stem (ES) cells as described (Levéen et al., 1994). The vector had integrated by homologous recombination in approximately one‐fifth of the selected clones, as judged by Southern blotting. Of ES‐cell clones in which a single copy of the targeting construct had integrated correctly at the PDGF‐B locus, about one‐third contained the loxP site in intron 4 and two‐thirds were lacking it, showing that recombination could take place on either side of this loxP site. We chose an ES‐cell clone containing a single correctly integrated targeting construct (flox‐neo allele) for deletion of the loxP‐flanked PGK‐neo cassette by transient expression of hCMV‐Cre (pBS185; kindly provided by Philippe Soriano, Seattle, WA). We distinguished clones in which the PGK‐neo cassette was deleted but the PDGF‐B exon 4 remained (flox allele) from clones in which recombination had deleted both PGK‐neo and exon 4 (lox allele) by Southern blot analysis. We created germ line chimeras and subsequently PDGF‐Bflox/+ PDGF‐Bflox/flox and PDGF‐Bflox/− mice (Levéen et al., 1994). All these were born at expected Mendelian ratios, reached adulthood and were phenotypically indistinguishable from PDGF‐B+/+ and PDGF‐B+/− mice, respectively. This and other types of analysis indicated that the PDGF‐Bflox allele was functionally equivalent to the PDGF‐B wild‐type (PDGF‐B+) allele.
Endothelium‐restricted ablation of PDGF‐B
We crossed Tie‐1Cre transgenic mice (Gustafsson et al., 2001) with PDGF‐B+/− mice to generate Tie1Cre+, PDGF‐B+/− offspring, which were subsequently crossed with PDGF‐Bflox/flox mice to create Tie1Cre+ PDGF‐Blox/− (PDGF‐Blox/−) animals together with various controls (Tie1Cre+PDGF‐Blox/+, Tie1Cre0PDGF‐Bflox/+, Tie1Cre0PDGF‐Bflox/−). We bred the mice onto the background of the XlacZ4 reporter mice that express β‐galactosidase in vSMCs/pericytes (Tidhar et al., 2001). Subsequently, we intercrossed PDGF‐Blox/− mice, which generated in addition to the above‐mentioned genotypes also Tie1Cre+, PDGF‐Blox/lox mice (PDGF‐Blox/lox), in which Cre must inactivate two alleles in each cell in order to generate a null situation, and PDGF‐B−/− mutants (embryonic lethal). The following PCR primers were used for genotyping: BF: 5′‐GGGTGGGACTTTGGTGTAGAGAAG‐3′; BB1: 5′‐TTTGAAGCGTGCAGAATGCC‐3′; BB2: 5′‐GGAACGGATTTTG GAGGTAGTGTC‐3′; BBlox: 5′‐TCTGGGTCACTGCTTCAGAATA GC‐3′.
For genotyping mice during breeding, we mixed the BF, BB1 and BB2 primers with tail or toe DNA in Gittchier buffer. Forty cycles of PCR were run as follows: 96°C 30 s, 57.9°C 30 s, 65°C 2 min. This generated diagnostic fragments of 265, 400 and 624 bp for the PDGF‐B+, PDGF‐Bflox and PDGF‐B− alleles, respectively (Figure 1C). The extent of recombination in vivo was controlled by PCR analysis of microvascular fragments isolated from brains of E16.5 PDGF‐Blox/+ embryos. The microvascular fragments were prepared essentially as described (Gargett et al., 2000) and contained ∼80% endothelial cells, 10% pericytes and 10% non‐vascular cells as judged by marker analysis. DNA isolated from the microvascular fragments was subject to a three‐primer PCR using the BF, BB2 and BBlox primers (PCR and cycle program as above). This PCR protocol provides diagnostic PCR fragments of 540, 400 and 265 bp representing the PDGF‐Blox, PDGF‐Bflox and PDGF‐B+ alleles (Figure 1C and D).
We carried out whole‐mount β‐galactosidase staining of whole tissue and sections as described (Hogan et al., 1994). Pericyte densities were quantified by counting XlacZ4‐positive nuclei on images captured from whole‐mount preparations of various parts of the CNS (including the retina) or 300 μm thick vibratome sections studied in a dissection microscope at fixed magnification. An image area containing >300 lacZ‐positive nuclei in a wild‐type individual was first determined. Subsequently the same area was counted in different lox/– individuals. The number of pericytes in lox/–individuals is described as a percentage of wild type. X‐gal‐stained retinas were post‐fixed for 10 min in 4% paraformaldehyde, followed by isolectin staining (Bandeiraea simplicifolia, Sigma L‐2140). Retinas were incubated in 1% bovine serum albumin, 0.5% Tween in phosphate‐buffered saline (PBS) overnight, washed twice in PBS pH 6.8 containing 1% Tween, 0.1 mM CaCl2, 0.1 mM MgCl2, 0.1 mM MnCl2 (PBlec) and incubated in biotinylated isolectin (20 μg/ml in PBlec) at 4°C overnight. Following washes in PBS, isolectin was detected using 10 μg/ml of a fluorescent streptavidin conjugate (Alexa Fluor 488, Molecular Probes, S‐11223). TO‐PRO3 (1:1000; Molecular Probes) was used for nuclear counterstaining. Whole‐mount isolectin staining on post‐fixed X‐gal‐stained P21 brains was achieved using peroxidase‐conjugated isolectin B4 from B.simplicifolia (Sigma L‐5391). Endogenous peroxidase was blocked prior to lectin staining, by 0.6% H2O2 in PBS. Peroxidase activity was detected by standard DAB staining. Vibratome brain sections (200 μm) were stained with isolectin B4 as described for retinas. GFAP labeling was achieved using a polyclonal rabbit antibody (1:75, Dako Z0334) followed by Alexa‐568‐conjugated secondary antibody (Molecular Probes). Confocal images were taken on a Leica TCS NT microscope system and processed in Adobe Photoshop. Retinal vascular preparations were obtained using a pepsin–trypsin digestion technique as described previously (Hammes et al., 1991). Briefly, a combined pepsin (5% pepsin in 0.2% hydrochloric acid for 1.5 h)–trypsin (2.5% in 0.2 M Tris for 15–30 min) digestion was used to isolate the retinal vasculature. Subsequently, the samples were stained with periodic acid Schiff's (PAS) stain.
We thank Helen Hjelm and Monica Elmestam for excellent technical assistance and Per Lindahl and Mats Hellström for valuable comments on the manuscript. The study was supported by the Novo Nordisk Foundation, the Swedish Cancer Foundation, IngaBritt and Arne Lundberg Foundation, the European Community and Göteborg University. H.G. is supported by an EMBO postdoctoral fellowship.
- Copyright © 2002 European Molecular Biology Organization