Selective ablation of αv integrins in the central nervous system leads to cerebral hemorrhage, seizures, axonal degeneration and premature death

Among the billions of neurons and glia in the vertebrate central nervous system (CNS) is an elaborate network of blood vessels. Besides carrying essential oxygen and nutrients to cells of the CNS, blood vessels also provide growth, guidance and survival cues that regulate various aspects of CNS development and maintenance (Louissaint et al., 2002; Mi et al.,2001; Palmer et al.,2000; Song et al.,2002; Zerlin and Goldman,1997). In turn, cells of the CNS microenvironment provide factors that influence blood vessel growth and survival(Haigh et al., 2003). Thus,deciphering how blood vessels communicate with neurons and glia is important for understanding the proper development of both the vascular and central nervous systems.

αv integrins are necessary for proper interactions between embryonic cerebral blood vessels and brain parenchymal cells(McCarty et al., 2002). Mouse embryos containing a null mutation in the αv gene, and thus lacking all five αv integrin family members; αvβ1, αvβ3,αvβ5, αvβ6 and αvβ8, exhibit normal vascular morphogenesis except in the brain. These mutants develop intracerebral hemorrhage and die shortly after birth. By contrast, deletion of both β3 and β5 integrin subunits has no effect on cerebral vascular development(McCarty et al., 2002);however, deletion of the β8 integrin gene leads to a cerebral hemorrhage phenotype similar to that seen in the αv-null mice(Zhu et al., 2002). Thus, loss of one specific integrin, αvβ8, can account for the selective cerebral blood vessel defects observed in the αv-nulls. αvβ8 is a receptor for multiple extracellular matrix (ECM) proteins(Milner et al., 1999; Mu et al., 2002; Nishimura et al., 1994; Venstrom and Reichardt, 1995),and is reportedly expressed on embryonic and postnatal glial cells and neurons in the CNS (Nishimura et al.,1998). The molecular mechanisms underlying how αvβ8 integrin functions to regulate cerebral blood vessel integrity remain to be determined, although normal endothelial cell branching and sprouting, as well as normal pericyte recruitment and apposition to the vasculature, all occur in the brains of αv-null embryos (Bader et al., 1998; McCarty et al.,2002).

To circumvent the early neonatal lethality in the complete αv-nulls and to explore further the cellular basis for the cerebral vascular defects,we have selectively ablated αv expression in cells of the embryonic and postnatal CNS. Ablation of αv expression on CNS radial glial cells and astrocytes leads to embryonic and neonatal cerebral hemorrhage. Deletion ofαv expression on both CNS neurons and glia also leads to cerebral hemorrhage, as well as additional neurological abnormalities, including seizures, ataxia and loss of hindlimb coordination. This study proves that theαv integrin subunit expressed on neural cells, particularly glia, is necessary for proper regulation of embryonic cerebral blood vessel development. Additionally, we reveal a novel function for αv integrins on postnatal CNS axons.

All progeny were confirmed by standard PCR-based genotyping of DNA isolated from tail snips or yolk sacs. Embryo staging and analysis was performed by timed matings, with noon on the plug date defined as E0.5. The following primer sequences were used for PCR genotyping: CRE,5′-ACCAGCCAGCTATCAACTC-3′ and 5′-TATACGCGTGCTAGCGAAGATCTCCATCTTCCAGCAG-3′ (the CRE primers yield a single PCR product of ∼200 bp); αv-flox,5′-TTCAGGACGGCACAAAGACCGTTG-3′ and 5′-CACAAATCAAGGATGACCAAACTGAG-3′ (PCR-mediated amplification of the αvflox allele generates a product of 350 base pairs; theαv^+/+ or αv^-/- alleles yield a PCR product of 250 bp); and NEO, 5′-AAGATGGATTGCACGCAGGTTCTC-3′ and 5′-CCTGATGCTCTTCGTCCAGATCAT-3′ (the NEO PCR product is approximately 1000 bp).

Nestin-Cre mice have been described elsewhere(Tronche et al., 1999) and were purchased from Jackson Laboratories. The Tie2-Cre mice have been previously described (Kisanuki et al.,2001). The hGFAPCre transgene contains a 2.2 kb fragment of the human GFAP gene driving expression of Cre cDNA. Results presented here were generated using one transgenic line containing 75-100 transgene copies.

The following antibodies were used for immunohistochemistry: anti-Nestin and anti-RC2 (Developmental Studies Hybridoma Bank); anti-BLBP (Nathaniel Heintz, Rockefeller University); anti-GFAP (DAKO); mouse anti-β-tubulin III (Sigma); mouse anti-Cre mAb (BabCo); mouse anti-PECAM (Pharmingen); and mouse anti-Smooth Muscle α-Actin (Sigma). The anti-myelin/oligodendrocyte specific protein, anti-neurofilament, anti-NeuN and anti-GFAP mAbs antibodies were purchased from Chemicon. Secondary antibodies used for immunofluorescence were Alexa-conjugated goat anti-rabbit,goat anti-mouse and goat anti-rat (Molecular Probes).

The anti-αv antiserum (Bossy and Reichardt, 1990) was used at a 1:200 dilution. Prior to application to sections, the diluted antiserum was pre-absorbed usingαv-null acetone-extracted brain protein, prepared by standard methods(Harlow and Lane, 1999). The antiserum was then applied to unfixed or Carnoy's-fixed histological sections. The anti-β8 antiserum was generated using a synthetic peptide corresponding to the C-terminal tail of the humanβ8 cytoplasmic region, CTRAVTYRREKPEEIKMDISKLNAHETFRCNF. The specificity of the anti-β8 antibody was tested using COS cell lysates transiently transfected with the pcDNA3.1 plasmid (Invitrogen) containing a human β8 cDNA. The expressed β8 protein contains a C-terminal V5 epitope tag. Alternatively, cortical astrocytes cultured from αv^+/- orαv^-/- neonates were cell surface labeled with EZ-Link Sulfo-NHS-LC-Biotin (Pierce). Lysates were immunoprecipitated with anti-β8, resolved by SDS-PAGE and immunoblotted with Streptavidin-HRP(Vector Laboratories).

Adult mice were anesthetized and fixed by perfusion with 4%paraformaldehyde (PFA). For frozen embedding, adult brains were removed and post-fixed for 12-16 hours. Embryonic and neonatal brains were dissected and immersed in Carnoy's fixative or 4% PFA for 4 or 16 hours, respectively. PFA-fixed tissue was cryopreserved in sucrose at 4°C and embedded in Tissue Tek OCT (Miles). Alternatively, brains were dehydrated and processed for standard paraffin embedding and basic histological analyses. Semi-thin (1μm) plastic sections of embryonic brains were prepared according to manufacturer's instructions (Polysciences).

Neonatal brains (P1-3) were removed and placed in sterile, ice-cold Hank's Balanced Salt Solution (HBSS). Whole neocortices were dissected and the hippocampus and internal structures were removed to leave only the cortical sheets. The meninges were stripped away, and the cortical sheets were minced with a razorblade and digested for 30 minutes at 37°C in Dulbecco's modified essential media containing 150 units/ml collagenase (Worthington) and 40 μg/ml deoxyribonuclease (Sigma, St Louis, MO). The cortical tissue was then triturated in DMEM containing 10% calf serum (Sigma, St Louis, MO) and filtered through a 50 μm sterile mesh. The resulting single-cell suspension was plated onto T-75 tissue culture flasks that had been pre-coated with 10μg/ml mouse laminin (Sigma, St Louis, MO). Generally, cells from two or three neonatal brains were plated per T-75 flask. After 7-10 days the astroglial cells formed a confluent monolayer, with neurons, oligodendrocytes and fibroblasts growing on top. These contaminating cells were removed by rotary shaking the flasks overnight at 250 revolutions per minute. The resulting cultures were composed of more than 95% astrocytes, as assessed by anti-GFAP immunoreactivity. Immunostaining of cells was performed using standard procedures. Briefly, astrocytes were trypsinized and replated onto laminin-coated coverslips for 24 hours. Cells were fixed in 4% PFA,permeabilized in 0.1% NP40/PBS, and immunostained with primary and secondary antibodies.

Adult brains were dissected and digested with collagenase and deoxyribonuclease. Brains were then minced with a razorblade and triturated to generate a cellular suspension. Brain endothelial cells were isolated using magnetic beads (Dynal) that were coated with anti-PECAM and anti-Flk rat mAbs(Pharmingen, Inc.) according to manufacturer's instructions.

A Rotarod device (Ugo Basile) was used to measure motor coordination and strength. The times at which mice fell from the rod were recorded. Mice were allowed to stay on the rod for a maximum of 360 seconds (6 minutes). Six separate trials were performed over a 2-day period (three trials per day).

We concluded in a previous study that αv-null mice develop brain-specific hemorrhage owing to defective associations between blood vessels and surrounding neural cells (neurons and/or glia)(McCarty et al., 2002);however, the identity of the αv-expressing cell type initiating the hemorrhage remained unknown. Analysis of αv protein expression patterns in the embryonic brain parenchyma revealed no readily detectable αv protein on endothelial cells or pericytes(Fig. 1A-F), suggestingαv integrin on cerebral vascular cells did not contribute to the cerebral hemorrhage phenotype. However, it remained possible that a low level of αv protein (undetectable by standard immunofluorescence) was present,and functioned in cerebral blood vessel development. To address this possibility, we genetically ablated αv expression selectively in the vascular endothelium using the Tie2-Cre transgene(Kisanuki et al., 2001). The Tie2 promoter drives Cre expression in some hematopoietic cells, as well as the majority of vascular endothelial cells, including those comprising blood vessels of the CNS (Allende et al.,2003; Gerety and Anderson,2002; Kisanuki et al.,2001; Liao et al.,2001).

Tie2-Cre+; αv^+/- mice were crossed with mice containing a conditional (floxed) αv allele (A.L.-H., J.S. and J.R., unpublished). The resulting mutant progeny are hemizygous for the Tie2-Cre transgene, and carry one αv-flox and one αv-null allele. Littermate controls are hemizygous for the Tie2-Cre transgene, and carry one αv-flox and oneαv wild-type allele. Recombination of the αv-flox allele by the Tie2-Cre transgene was verified using freshly isolated brain endothelial cells. Analysis of genomic DNA revealed significant reduction in the 350 bp PCR band representing the αv-flox allele(Fig. 1G,H).

Tie2-Cre⁺; αv^flox/- mutants were born in expected Mendelian ratios and at birth displayed no phenotypic defects. Gross analyses of neonatal (Fig. 1K),as well as embryonic and adult mutant brains (data not shown) revealed no obvious cerebral vascular defects. Furthermore, microscopic analysis of mutant brains showed intact cerebral blood vessels with no indication of dilation or hemorrhage (Fig. 1L). Thus,loss of αv expression in vascular endothelium does not account for the cerebral hemorrhage phenotype observed in the complete αv knockouts.

Previous papers using different anti-αv integrin antibodies have reported αv protein expression on embryonic neuroepithelial and radial glial cells (Anton et al.,1999; Hirsch et al.,1994). To confirm these results, we used these same anti-αv antibodies to immunostain E13.5 αv^+/- andαv^-/- brain sections. Surprisingly, we observed a neuroepithelial expression pattern in both αv^+/- andαv^-/- brain sections (data not shown). To eliminate this non-specific αv immunoreactivity, we pre-absorbed anti-αv antiserum (Bossy and Reichardt,1990) using acetone-extracts prepared from αv-null embryos. Immunostaining αv^+/- and αv^-/- brain sections revealed αv-specific neuroepithelial immunoreactivity, as determined by co-localization with nestin(Fig. 2A-C). The anti-αv immunoreactivity did not colocalize with markers for newly differentiated neurons, such as β-tubulin III or MAP2 (see Fig. S1 in the supplementary material and data not shown).

Mouse embryos null for the β8 integrin subunit also develop embryonic intracerebral hemorrhage that appears identical to that in the completeαv-nulls (Zhu et al.,2002). We generated an affinity-purified polyclonal antibody directed against a peptide corresponding to the β8 cytoplasmic tail and showed that it recognized a predominant protein band of ∼90 kDa from COS cell lysates transfected with a cDNA encoding the β8 protein(Fig. 2M). Analysis of primary astrocytes from wild-type neonatal mice showed that anti-β8 immunoprecipitated two cell surface proteins of ∼130 kDa and ∼90 kDa,both consistent with the molecular weights of αv and β8 integrins(Fig. 2N). These bands were not detected when anti-β8 was used to immunoprecipitate cell surface proteins from αv-null astrocytes (Fig. 2N). Immunostaining E11.5 αv^+/- embryonic brain sections (Fig. 2G,I) revealedβ8 protein expression on neuroepithelial cells. However, using double-label immunofluorescence analyses, no β8 protein expression was detected on endothelial cells or pericytes that comprise cerebral blood vessels (data not shown). We detected β8 protein expression inαv-null embryos; however, the spatial pattern was abnormal: β8 protein was not detectable on neuroepithelial cell processes(Fig. 2K), but was primarily localized to neuroepithelial cell bodies(Fig. 2J,L). Thus, we concluded that, in the embryonic brain, αvβ8 integrin is expressed on glial cell processes, and that it is likely that the loss of αvβ8 expression on these cell types leads to intracerebral hemorrhage.

We attempted to ablate αv expression selectively on embryonic and postnatal glial cells using Cre under the control of the human GFAP promoter. In most mammalian species, GFAP is expressed primarily by postnatal astrocytes; however, in primates, GFAP is also expressed by embryonic radial glia and independently generated hGFAP-Cre transgenes show Cre expression in both embryonic and post-natal glial cells(Bajenaru et al., 2002; Malatesta et al., 2003; Marino et al., 2000; Zhuo et al., 2001).

We used an anti-Cre antibody to monitor Cre protein expression in our hGFAP-Cre transgenics. Nuclear Cre protein was most prominent in cells adjacent to, and occasionally within, the ventricular zone of the neocortex(Fig. 3A,B) and ganglionic eminence (data not shown) as early as E15. Based on this pattern, we analyzed whether Cre protein was present in radial glial cells. Using antibodies recognizing the brain lipid-binding protein (BLBP) and the astrocyte-specific glutamate transporter (GLAST) (Anthony et al., 2004; Malatesta et al.,2003), we showed Cre protein in many cortical radial glial cells in E15 embryos (Fig. 3A,B). By contrast, at E15, no co-localization of Cre and β-tubulin III (a marker for neurons) was detected in the neocortex(Fig. 3C) or ganglionic eminence (data not shown). Cre protein was also expressed by many GFAP-positive Bergmann glia in the cerebellum as well as astrocytes in the neonatal cerebral cortex (Fig. 3D,E). hGFAP-Cre transgene expression in astrocytes and neurons was also observed in vivo using the Rosa26-LoxSTOPLox-lacZ reporter strain (Soriano, 1999) to monitor hGFAP-driven Cre activity (data not shown). Owing to their multi-potential nature, CNS radial glial cells give rise to neurons throughout the brain (Anthony et al.,2004; Malatesta et al.,2003). Therefore, it is likely that in our hGFAP-Cre strain, some embryonic and postnatal neurons may lack αv expression. Any contributions that these neurons normally make to cerebral blood vessel function would also be defective in the hGFAPCre conditional αv mutants.

Consistent with the mosaic Cre protein expression pattern in vivo, most astrocytes cultured from hGFAP-Cre mice expressed Cre, but some had no detectable level of Cre expression (Fig. 3F). PCR analysis of DNA from cultured astrocytes revealed a significant reduction in the 350 bp band representing the αvflox allele(Fig. 3G,H). Likewise, using extracts prepared from control and mutant astrocytes, we showed a significant reduction in αv protein expressed in cells isolated from hGFAP-Cre conditional αv mutant brains (Fig. 3I). αv expressed by contaminating cells may account for the low residual level of αv protein, although our astrocyte cultures are generally more than 95% pure (GFAP positive). The low level of αv protein present in the mutant astrocyte cell cultures was most likely due to the mosaic expression pattern of the hGFAPCre transgene.

hGFAP-Cre⁺; αv^+/- progeny were mated with mice containing a loxP-flanked (floxed) αv allele (Lacy-Hulbert et al.,submitted). Thus, the resulting mutant progeny are hemizygous for the hGFAP-Cre transgene, and carry one αv-flox and one αv-null allele. Littermate controls were hemizygous for the hGFAP-Cre transgene, and carry oneαvflox and one αv-wild type allele.

hGFAP-Cre⁺; αv^flox/- mutants were born in expected Mendelian ratios (18 mutants from 68 total F1 progeny, or 26%). Analysis of hGFAP-Cre conditional αv mutant brains revealed that intracerebral hemorrhage occurred with 100% penetrance. P5 mutant neonates(Fig. 3K) or E18 embryos (data not shown) developed obvious punctate regions of microhemorrhage in the cerebral cortex, mid-brain and occasionally in the cerebellum(Fig. 3M; data not shown). Brains from older neonates between P7 and P14 revealed progressive reduction in cerebral microhemorrhage (data not shown), and gross and microscopic analyses of adult mutant brains revealed no obvious signs of hemorrhage or edema (Fig. 3O,Q). Immunolabeling with anti-laminin antibody revealed intact intracerebral blood vessels with no indications of distension or rupture (data not shown). Furthermore, perfusion of a small molecular weight reactive biotin tracer,revealed no obvious blood-brain barrier abnormalities in adult hGFAP-Cre conditional αv mutants (see Fig. S2 in the supplementary material).

We hypothesized that the significantly less severe hemorrhage observed in the hGFAP-Cre conditional αv mutants (as compared with completeαv-nulls) was connected to the late onset (∼E15) of Cre expression. To ablate αv expression earlier during CNS development we used a Nestin-Cre transgenic mouse strain which expresses Cre under control of a Nestin promoter (Graus-Porta et al.,2001; Haigh et al.,2003; Tronche et al.,1999). The Nestin-Cre transgene drives Cre expression primarily in CNS neural precursor cells that give rise to both neurons and glia(Haigh et al., 2003).

Nestin-Cre+; αv^+/- mice were mated with mice containing a conditional (floxed) αv allele. The resulting mutant progeny were hemizygous for the Nestin-Cre transgene, and carried one αv-flox and oneαv-null allele. Littermate controls were hemizygous for the Nestin-Cre transgene, and carried one αvflox allele and one αv wild-type allele. Cre-mediated recombination of the αv-flox allele was confirmed by PCR analysis of DNA from adult mutant brains. DNA isolated from the tail yielded PCR products of 350 and 250 bp, indicating no detectable Cre-mediated recombination (Fig. 4A,B). DNA isolated from adult brain tissue, however, showed a dramatic reduction in the 350 bp product (αv-flox), owing to Cre-mediated recombination of theαv-flox allele. The low level of 350 bp band(Fig. 4B) is probably due to the absence of Cre-mediated recombination in non-neural cells, e.g. circulating blood cells, fibroblasts, etc. Additionally, immunoblotting lysates prepared from adult cerebral cortex and cerebellum with anti-αv antibody revealed marked reduction in αv protein in Nestin-Cre conditional αv mutants (Fig. 4C). The protein band at ∼80 kDa(Fig. 4C) is nonspecific. It is not a fragment of full-length αv, as it is observed in αv-null cell and tissue lysates (data not shown).

Nestin-Cre conditional αv mutants were born in the expected Mendelian ratios (27 mutants of 111 total progeny, or 24%). All Nestin-Cre conditionalαv mutants developed small bilateral cerebral hemorrhages beginning at E14.5, which became progressively more severe throughout embryogenesis(Fig. 4E). Gross inspection of mutant neonatal (P7) brains also revealed obvious cerebral hemorrhages(Fig. 4G,I). The cerebral hemorrhage in the Nestin-Cre conditional αv mutants was more widespread than that in the hGFAP-Cre conditional αv mutants; however, the hemorrhage was significantly less severe than that seen in the completeαv-nulls (Fig. 4H). Similar to the complete αv nulls(Bader et al., 1998), the vascular abnormalities in the Nestin-Cre conditional αv mutants initiated in the developing ganglionic eminences of the forebrain(Fig. 4K). Unlike the close juxtaposition between cerebral vessels and the surrounding parenchyma in brains of Nestin-Cre control littermates(Fig. 4J), cerebral vessels in Nestin-Cre mutants were separated from the surrounding neural tissue(Fig. 4K). Mutant cerebral blood vessels were lined by endothelial cells and invested by a pericyte/vascular smooth muscle cell layer, but appeared tortuous and distended (Fig. 4M). Mutant neonatal brains contained abnormally high numbers of GFAP-positive astrocytes,indicative of reactive gliosis (Fig. 4Q). The increased numbers of astrocytes may represent a mechanism to repair the cerebral hemorrhage in the Nestin-Cre conditional αv mutants.

Nestin-Cre conditional αv mutants survived beyond the neonatal period and were behaviorally normal until 2-3 weeks after birth. At this point,∼30% (13 of 39 mutants identified so far) began to display obvious signs of motor dysfunction (Fig. 5B). These conditional αv mutants also developed spontaneous convulsions, and ultimately died by 4 weeks of age (data not shown). Histological analyses of CNS tissue isolated from mutants revealed focal regions of microhemorrhage throughout the brain and spinal cord (Fig. 5D,F). As hemorrhage begins in utero and is present prior to the onset of apparent seizure (Fig. 4E,G,I), it is very likely that the cerebral hemorrhage causes the seizures and premature death. The incomplete penetrance associated with the seizures and early death may be related to genetic heterogeneity, as the mice used in this study were a mixture of C57BL/6, 129S4 and FVB genetic backgrounds.

The remaining 70% of the Nestin-Cre mutants survived for several months. At∼2-3 months of age, 100% of these mutants began to display defects in hind limb coordination. The hind limbs of mutants remained reflexive; however,mutant muscle tone was spastic, leading to a rigid posture and gait(Fig. 5H). Indeed, gait analyses revealed that mutants partially dragged their hind limbs(Fig. 5I), and were physically weaker and less coordinated, as assessed by Rotarod testing(Fig. 5J). Mutants also displayed toe-walking, involuntary tail wriggling and ataxia (see Movie 1 in the supplementary material; data not shown). By 6 months of age, the mutants developed more frequent seizures and more obvious symptoms of hind limb dysfunction, indicative of spastic paraparesis. Nestin-Cre mutants also developed urinary dysfunction, leading to severe bacterial infection of the bladder and, for ethical reasons, were euthanized by 8 months of age.

We initially expected the hindlimb dysfunction in the adult mutants to be linked to neurological damage because of persistent cerebral hemorrhage and stroke. However, gross and microscopic examination of adult mutant brains revealed no pathological indication of hemorrhage (see Fig. S3 in the supplementary material). CNS deletion of the β1 integrin subunit also leads to ataxia and premature death, largely owing to abnormal anchorage of glial end-feet to marginal zone basement membranes in the cortex and cerebellum (Graus-Porta et al.,2001). Thus, it was possible that selective loss ofαvβ1 could account for the neurological defects observed in the Nestin-Cre conditional αv mutants. However, immunofluorescence analysis of Nestin-Cre conditional αv mutants revealed normal glial end-feet attachments in the cortex and cerebellum (see Fig. S3 in the supplementary material). Histological analysis of brains from neonatal and adult Nestin-Cre mutants, both by Hematoxylin and Eosin staining, as well as immunohistochemistry using an antibody directed against the neuronal nuclear marker NeuN, revealed relatively normal neuronal patterning in the cerebral cortex and cerebellum (see Fig. S3 in the supplementary material; data not shown). Additionally, silver staining to label axonal neurofilaments, or immunohistochemistry using an anti-MAP2 antibody to label dendrites, revealed normal neuronal arborizations (see Fig. S3 in the supplementary material; data not shown).

We also suspected that the onset of spastic paraparesis in the Nestin-Cre conditional αv mutants might be due to neuromuscular defects. However,examination of mutant quadriceps showed no indications of muscular atrophy or wasting (see Fig. S4 in the supplementary material). Furthermore, inspection of hindlimb peripheral nerves as well as peripheral nerves connecting to thoracic and lumbar spinal cord dorsal root ganglia revealed no obvious pathological indication of peripheral axonal degeneration or demyelination(see Fig. S4 in the supplementary material; data not shown). This is unlike the paralysis and demyelination defects observed in mutant mice where β1 integrin is selectively ablated in peripheral nervous system Schwann cells(Feltri et al., 2002).

Analyses of grey matter spinal cord regions from controls and Nestin-Cre conditional αv mutants did not reveal motor neuron dystrophy or degeneration (Fig. 6C,D). However, analysis of spinal cord white matter revealed demyelination(Fig. 6F) and axonal dystrophy(data not shown) within the fasciculus gracilis, an axonal tract that regulates hind limb proprioception and discriminative touch. Additionally, an obvious macrophage infiltrate was present, with many macrophages containing myelin fragments (Fig. 6F) as well as axonal fragments (data not shown), suggesting both glial and axonal degeneration.

Axons of the cerebellar white matter also regulate basic hind limb proprioception and coordination (Bosco and Poppele, 2003). Luxol Fast Blue staining to visualize myelin revealed significant demyelination in the caudal cerebellum of Nestin-Cre conditional αv mutants (Fig. 6J). Additionally, analysis of silver-stained white matter axons within the cerebellum revealed obvious axonopathy(Fig. 6L). A macrophage infiltrate was also present, and many macrophages contained internalized myelin and neurofilament fragments (Fig. 6J,L). However, a lymphocyte infiltrate, which would be indicative of autoimmunity, was not present in the spinal cord or cerebellar lesions.

Ablation of the β8 gene leads to a cerebral hemorrhage phenotype(Zhu et al., 2002), which is essentially identical to that observed in the complete αv-nulls(Bader et al., 1998). With this knowledge, we hypothesized that the neurological defects observed in the Nestin-Cre conditional αv mutants were probably due to the selective loss of αvβ8 function. To determine the expression pattern forαvβ8 protein in the adult cerebellum, we used an antibody directed against the β8 integrin subunit (Fig. 7A,D,G). αvβ8 was expressed on axons in the caudal cerebellum, as determined by co-localization with an anti-neurofilament antibody (Fig. 7C). However,αvβ8 was not detected on cerebellar white matter astrocytes(Fig. 7F) or oligodendrocytes(Fig. 7I). αvβ8 protein was also expressed on spinal cord axons, including the fasciculus gracilis (data not shown). Interestingly, αvβ8 was also expressed on axons in white matter regions of the cerebral cortex (data not shown),although no pathological lesions were observed in these regions at the time of death.

We have used Cre/Lox technology to analyze the functions of αv integrins in the embryonic and postnatal CNS. Selective ablation of αv expression in endothelial cells does not produce any obvious embryonic or neonatal developmental defects. By contrast, ablation of αv expression on neural cells, particularly glia, has significant effects. A central result from this work is that αvβ8 integrin expressed on embryonic CNS neural cells is necessary for the proper development of cerebral blood vessels. This finding extends our earlier studies of complete αv-null mice (McCarty et al., 2002),and firmly implicates αvβ8 as a neural regulator of embryonic cerebral blood vessel function. Additionally, we present the novel finding that αvβ8 integrin expressed on CNS axons plays an important function in mediating long-term axonal survival. Collectively, these results suggest that αvβ8 may function in mediating proper cellular juxtaposition both between neural cells and blood vessels in the embryonic CNS, and between axons and glia in the postnatal CNS.

Cerebral blood vessels are composed of endothelial cells and pericytes surrounded by a vascular basement membrane. The close juxtaposition between cerebral blood vessels and surrounding neural cells, particularly glia, evokes an attractive model, whereby αvβ8 integrin on the glial processes mediates interactions with blood vessels via adhesion to ECM ligands within the vascular basement membrane. Neural cell adhesion via αvβ8 may provide physical support that maintains proper blood vessel morphology. Alternatively, αvβ8 may play a more dynamic role in mediating the adhesion and communication between neural cells and blood vessels. We propose a model where αvβ8 integrin is necessary for the initiation and maintenance of neural cell-blood vessel adhesion and communication, perhaps including proper assembly of the basement membrane. A wave of secondary events would then serve to affect proper neural cell-blood vessel juxtaposition. We are currently identifying and characterizing the signal transduction pathways activated by αvβ8 adhesion in an attempt to understand better neural cell-mediated regulation of cerebral blood vessel development.

Why do 100% of complete αv-nulls die by the first day of birth,whereas the majority of the conditional αv mutants survive for several months? First, conditional αv mutants do not develop other non-CNS abnormalities that occur with variable penetrance in complete αv-nulls(Bader et al., 1998). Second,the hemorrhage in the conditional αv knockout mice is significantly less severe than that in the complete αv-nulls, most probably because of differences in Cre spatial expression. Additionally, there may be a temporal delay from the onset of Cre expression (and αv-flox deletion) to the point where αv integrin mRNA and protein are completely absent from the cells. Last, epigenetic events may preclude all glial cells from efficiently recombining the αv-flox allele, leading to a mosaic pattern of Cre expression. Such mosaic Cre expression would decrease the overall proportion of αv-null glial cells, thus leading to a reduction in the severity of hemorrhage. Alternatively, it is possible that loss of αv expression on multiple CNS cell types is necessary to recapitulate fully the severe cerebral hemorrhage observed in the complete αv-nulls.

We initially hypothesized that the neurological phenotypes observed in the adult Nestin-Cre conditional αv mutants were the result of CNS trauma caused by the early embryonic and neonatal hemorrhage. Indeed, this is the case for humans with various forms of cerebral palsy, where localized CNS trauma during prenatal or neonatal periods leads to neuronal cell death and the unfortunate symptoms associated with this disease(Ozduman et al., 2004). However, the hemorrhage in the Nestin-Cre conditional αv mutants occurs throughout the embryonic and neonatal brain, whereas the neurological lesions first develop in specific regions of the adult spinal cord and cerebellum,suggesting a more localized defect responsible for the neurological phenotype.

Alternatively, the neurological lesions may result from progressive hydrocephaly resulting from intraventricular hemorrhage. Indeed, completeαv-null mutants develop intraventricular hemorrhage as early as E12.5,and display severe hydrocephaly at birth(Bader et al., 1998). Interestingly, severe hydrocephaly in humans can cause pathological lesions in the long white matter tracts of the cerebral cortex and cerebellum, leading to symptoms such as seizures, ataxia, lower limb spasticity and urinary dysfunction (Silverberg,2004). However, we have yet to observe intraventricular hemorrhage or obvious hydrocephalus in the Nestin-Cre conditional αv mutants. Nonetheless, it remains possible that subtle changes in ventricle volume may account for some of the neurological abnormalities.

A more likely explanation for the neurological phenotypes, and one supported by our experimental data, is that the adult-onset defects are unrelated to the embryonic and neonatal hemorrhage. Indeed, to date, no neurological defects or premature death are observed in the hGFAP-Cre conditional αv mutants, which also develop embryonic and neonatal cerebral hemorrhage, albeit less severe than we see in the Nestin-Cre conditional αv mutants. Additionally, the progressive neurological phenotypes observed in the Nestin-Cre conditional αv mutants develop over a period of several months, well beyond the time of embryonic and neonatal hemorrhage. Last, αvβ8 integrin protein is expressed on axons in the spinal cord and cerebellum where pathologic lesions develop in the Nestin-Cre conditional αv mutants. These data collectively suggest that the CNS abnormalities observed in the Nestin-Cre conditional αv mutants are likely to be related to the specific loss of αvβ8, as they do not occur in β3/β5 integrin double knockout mice(McCarty et al., 2002), orβ6 integrin knockouts (Huang et al.,1996), and are distinct from the neurological defects in mice lacking β1 integrins in the CNS or PNS(Graus-Porta et al., 2001; Feltri et al., 2002). The confirmation of this hypothesis awaits generation and analysis of conditionalβ8 knockout mice.

Interestingly, in the Nestin-Cre conditional αv mutants, we observe not only spinocerebellar axonal degeneration, but also significant demyelination. Thus, analogous to its role in regulating glia-blood vessel juxtaposition, axonally expressed αvβ8 may also mediate proper axonal-glial interactions. CNS axons intimately associate with myelinating oligodendrocytes via direct cell-cell contacts, as well as cell-ECM adhesion. Recent evidence also shows cooperative signaling between integrins and a variety of axonal guidance and survival factors, including the Ephs/ephrins(Zou et al., 1999),semaphorins (Pasterkamp et al.,2003), netrins (Yebra et al.,2003) and slits (Stevens and Jacobs, 2002). It will be interesting to determine whetherαvβ8 integrin signaling events `crosstalk' with these other pathways, and how these events may regulate axonal survival.

The cerebral hemorrhage in the conditional αv mutants is very similar to pathologies associated with the human genetic disorder cerebral cavernous malformation (CCM). Individuals with CCM develop CNS hemorrhage characterized by tortuous blood vessels devoid of surrounding brain parenchyma, leading to clinical symptoms ranging from mild headaches to sporadic seizures and loss of motor function (Marchuk et al.,2003). Mutations in the KRIT1 gene have been linked to∼50% of all familial CCM cases (Gunel et al., 2002; Laberge-le Couteulx et al., 1999). We are currently investigating whetherαvβ8 integrin plays a role in regulating Krit1 function and how these events may be connected to development of CCM.

Finally, the neurological abnormalities observed in the adult Nestin-Cre conditional αv mutants are similar to a human genetic disorder known as hereditary spastic paraparesis (HSP)(Kobayashi et al., 1996; McDermott et al., 2000). Individuals with HSP develop progressive lower limb spasticity partly because of deterioration of spinocerebellar axons. It will be interesting to determine if the Nestin-Cre conditional αv mutants might serve as a useful mouse model for studying the pathogenesis and treatment of symptoms related to HSP.

We thank Dr Ann Graybiel for comments on the manuscript; Drs Louis Reichardt and Guido Tarone for providing anti-αv antibodies; Dr Nathaniel Heintz for the anti-BLBP antiserum; Dr Masashi Yanigasawa for providing the Tie2-Cre mice; Dr Stephen Nishimura (UCSF) for the human β8 cDNA; and Aaron Cook for help with mouse genotyping. This work was supported by PO1 HL66105-03 to R.O.H. A.L.-H. is funded by Wellcome Trust Grant 064487 to J.S. R.O.H. is an investigator and J.H.M. was an associate of the Howard Hughes Medical Institute.