HOXA3 induces cell migration in endothelial and epithelial cells promoting angiogenesis and wound repair

Homeobox (HOX) genes encode transcriptional regulators with well-described and evolutionarily conserved roles in directing morphogenesis during embryonic development. In humans, the class I homeobox genes are found in four clusters, each located on a different chromosome, with up to 13 members in each cluster. The conserved organization and sequence homology allows paralogous relationships to be established between clusters. Paralogous HOX genes are believed to carry out complementary and sometimes overlapping functions. In addition to their early developmental roles, HOX genes also direct tissue remodeling in adults, including postnatal development of the mammary gland during pregnancy (Chen and Capecchi, 1999). Furthermore, altered expression of HOX genes has been linked to pathologically induced tissue remodeling associated with tumorigenesis, as well as the process of wound repair (Cantile et al., 2003; Cillo et al., 1999; Raman et al., 2000; Stelnicki et al., 1998). These findings are not surprising as embryonic development, tumorigenesis and wound repair share many common features at the cellular level (for reviews see Coussens and Werb, 2002; Martin and Parkhurst, 2004). Although the full suite of direct target genes of HOX transcription factors are unknown, accumulating evidence indicates that genes involved in cell-cell and cell-extracellular matrix (ECM) interactions are frequently found as targets (Boudreau and Varner, 2004; Chen and Ruley, 1998; Cillo et al., 1996; Jones et al., 1992; Lorentz et al., 1997; Myers et al., 2000; Penkov et al., 2000).

We have investigated the role of HOX genes in angiogenesis and in the remodeling of the ECM, as neovascularization (or angiogenesis) is a critical component of tumor progression and wound repair characterized by dynamic alterations of cell-cell and cell-ECM interactions. We and others have shown that adult endothelial cells (EC) express a number of HOX genes, including HOXB3, HOXD3, HOXA9 and HOXD10, whose expression can be modulated by exposure to various factors that induce angiogenesis. These include bFGF, TNF-α and VEGF (Belotti et al., 1998; Boudreau et al., 1997; Bruhl et al., 2004; Patel et al., 1999). We have also shown that although HOXD3 is not highly expressed in quiescent endothelium in vivo, expression in dermal microvessels is markedly elevated within 24 hours following wounding (Uyeno et al., 2001). Furthermore, we have previously demonstrated that HOXD3 upregulates expression of the matrix-degrading serine proteinase, urokinase plasminogen activator (uPA), as well as both the β3 subunit of the αvβ3 integrin and the α5 subunit of α5β1 integrin (Boudreau et al., 1997; Boudreau and Varner, 2004). Together, activation of these HOXD3 targets contributes to the invasive and migratory stages of angiogenesis. HOXD3 was shown to improve wound repair in diabetic mice in vivo by increasing angiogenesis as well as the synthesis of collagen (Hansen et al., 2003b). The HOXD3 paralog, HOXB3, has also been shown to promote angiogenesis in vivo (Myers et al., 2000). Rather than enhancing EC migration or invasion, HOXB3 induces new vascular sprouts to undergo capillary morphogenesis by upregulating expression of ephrin A1 protein, which in turn facilitates the cell-cell interactions necessary for capillary morphogenesis (Myers et al., 2000). The discovery of these distinct, yet clearly complementary roles for HOXD3 and HOXB3 support a model in which paralogous HOX genes function in a similar, yet complementary manner within a particular tissue (Condie and Capecchi, 1994; Manley and Capecchi, 1997).

Although the role of HOXA3 (a paralog of HOXD3 and HOXB3) has not been previously examined in angiogenesis, studies have demonstrated that during embryogenesis HOXD3 and HOXA3 genes can be functionally interchangeable when expressed at the proper place and time (Greer et al., 2000). Specifically, the embryonic lethal phenotype of HOXA3 null mice was rescued by placing HOXD3 into the HOXA3 locus. Complementing this result, HOXA3 was found to rescue the skeletal defects of HOXD3 null mice when expressed from the HOXD3 locus. What is not clear is whether these interchanged HOX genes regulated the transcription of identical target genes when expressed in a different context, or whether they modulated expression of distinct, yet functionally related genes resulting in phenotypic rescue. Moreover, studies of compound mutants of HOX9 group genes revealed markedly different effects on embryonic and adult mammary gland development, suggesting that some paralogous HOX genes perform roles that may change with time (Chen and Capecchi, 1999).

In the present study we have examined the function of HOXA3 in three contexts: in cultured adult microvascular EC and keratinocytes, during developmental angiogenesis and during wound healing in vivo. We show that HOXA3 expression is upregulated during neovascularization in response to injury and that HOXA3 promotes angiogenesis, alters expression of genes associated with cell-cell and cell-ECM interactions and induces changes in EC behavior similar to the effects of HOXD3 or HOXB3, but through distinct target genes. Furthermore, we report that gene transfer of HOXA3 greatly accelerates wound healing in diabetic mice, and unlike HOXD3, also induces keratinocyte migration in response to wounding. We present evidence that this is mediated in part by the transcriptional activation of uPAR. These results provide new insight into the adult function of HOXA3 and suggest that HOXA3 may be a potent therapeutic agent for accelerating wound closure in diabetic patients.

An immortalized human dermal microvascular endothelial cell (EC) line HMEC-1 (Ades et al., 1992) was a gift from T. Lawley (Emory University, Atlanta, GA). These cells have previously been shown to maintain many properties of primary dermal microvascular cells in culture including the ability to undergo capillary morphogenesis when cultured on basement membrane (Matrigel, BD Biosciences, San Jose, CA) and maintain expression of a number of EC surface markers (Xu et al., 1994). Cells were maintained in media MCDB 131 supplemented with 10% FBS, Gentamicin and 1% hydrocortisone, and passaged using calcium and magnesium-free PBS supplemented with 0.053 mM EDTA. Primary cultures of human dermal microvascular endothelial cells were purchased from Clonetics (Palo Alto, CA). Recombinant human VEGF was purchased from R&D Systems (Minneapolis, MN). EC culture on basement membrane (Matrigel) was performed as previously described (Boudreau et al., 1997). To release cells from Matrigel for protein or mRNA isolation, cultures were suspended in PBS without Ca²⁺ or Mg²⁺, containing 0.5 mM EDTA and incubated on ice for 1 hour to allow the Matrigel to disperse.

Total RNA was isolated from HMEC-1 using the Qiagen RNeasy kit (Qiagen, Valencia, CA). For northern blot analysis, 10 or 20 μg of total RNA were separated on 1% agarose formaldehyde gels as previously described using standard methods (Boudreau et al., 1997). Ribosomal RNA was visualized by staining with 5 μg/ml ethidium bromide. Labeled [³²P]dCTP probes were prepared using the Decaprime Kit (Ambion, Austin, TX) and purified using Sephadex G-25 columns (Roche, Indianapolis, IN). Blots were probed with 1×10⁶ cpm/ml Hybridsol I hybridization buffer (Oncor, Gaithersburg, MD) and exposed to Kodak M5 X-Omat film at –70°C. The cDNA probe for human uPAR was purchased from American Type Culture Collection (Manassas, VA), and the cDNA probe for human MMP-14 corresponds to GenBank accession number NM004995.

1 μg total RNA was reverse transcribed using MMuLV RT (Invitrogen, Carlsbad, CA) for 1 hour at 42°C in a total volume of 25 μl; 0.1, 1, 2 and 10 μl of this RT reaction was then amplified for 20, 30 or 35 cycles of 95°C, 58°C and 72°C for 30, 30 and 90 seconds, respectively. Primers used for PCR were as follows: forward primer for EC and wounds, 5′-TGC-GAT-CAA-GAT-CGT-GAA-ACA-ACG-C-3′ corresponding to nucleotides 93080-93056 of the genomic sequence contained in human PAC clone (RP-1167F23), accession number AC004079; EC reverse primer, 5′-AGA-CTC-TCC-TGG-CGC-GTA-GCC-CCA-A-3′ corresponding to 90277-90312 of the genomic sequence (AC004079); wounds reverse primer, 5′-GGG-CTC-ATA-TGG-GAC-ACT-GTT-3′. The expected 1.3 kb PCR product was visualized by electrophoresis on 1% agarose gels containing ethidium bromide. From this analysis it was determined that amplification of 1 μl of the total 25 μl RT reaction for 30 cycles gave optimal, reproducible results within the linear range for amplification. To normalize for total RNA, 18S RNA was amplified in the same reaction with commercially available primers sets for human or murine 18S RNA (Ambion, Austin, TX). The full-length human cDNA was cloned by isolating the 1.37 kb PCR product described above and ligating into the pCRII-TOPO TA cloning vector (Invitrogen, Carlsbad, CA). Sequences were verified by sequencing reactions performed at the Biomolecular Resource Center, UCSF.

To construct the HOXA3 gene expression plasmid, an EcoRI fragment was generated from the pCRII-TOPO HOXA3 clone, removing the last eight amino acids and stop codon, and ligated in-frame into the pcDNA3.1 myc/His vector (Invitrogen, Carlsbad, CA) to generate a HOXA3 myc/His fusion protein under control of the CMV promoter. Transfections of the HMEC-1 cell line were performed using Effectene (Qiagen, Valencia, CA) according to the manufacturer's instructions and pools of stably transfected cells were selected using 35 μg/ml of G418 (Invitrogen, Carlsbad, CA).

Western blots of control and HOXA3-transfected EC were probed for human uPAR and MMP-14. Total protein was isolated from cell lysates of HOXA3 or control-transfected HMEC-1 and 20 μg were loaded onto a 10% SDS-PAGE gel, transferred onto Immobilon-P membrane and blocked with 1% BSA in T-TBS. Blots were incubated with 1:500 rabbit anti-human uPAR (American Diagnostica) or 1 μg/ml mouse monoclonal anti-MMP-14 (clone 113-5B7, Oncogene), followed by donkey anti-rabbit-HRP or sheep anti-mouse-HRP at 1:2000, and detected with the ECL system (Vector Labs). To detect the HOXA3/myc/His fusion protein, transfected cells or homogenized whole wounds were lysed with 6 M urea lysis buffer and diluted in Laemmli buffer. 40 μg total cell lysate were separated on a 10% SDS-PAGE gel, transferred to Immobilon-P, and blocked with 1% BSA. Blots from transfected cells were probed with a 1:100 dilution of a monoclonal antibody against a His₆ epitope tag (Invitrogen, Carlsbad, CA), followed by sheep anti-mouse IgG conjugated to HRP at 1:2000, and visualized with enhanced chemiluminescence. Blots from wounds were probed with a 1:200 dilution of rabbit anti-His-tag (Novus Biologicals, Littleton, CO), followed by a 1:2000 dilution of donkey anti-rabbit IgG conjugated to HRP, and detected with enhanced chemiluminescence (SuperSignal, Pierce, Rockford, IL).

Western blot analysis of uPAR was carried out using goat anti-uPAR (R&D Systems, Minneapolis, MN) at 1:300, followed by rabbit anti-goat IgG conjugated to biotin at 1:2000, followed by ABC-HRP (Vector Labs, Burlingame, CA) and detected with enhanced chemiluminescence (SuperSignal, Pierce, Rockford, IL).

Modified Boyden chambers were coated with 20 μg/ml type I collagen (Invitrogen, Carlsbad, CA) for 2 hours at 37°C and cells were plated and allowed to migrate to the lower chambers for 4 hours. Cells that had migrated were stained with Diff Quick and counted as previously described (Myers et al., 2000).

Approximately 5×10⁵ HMEC-1 cells were seeded with 0.5 ml of a solution containing 1 mg/ml Cytodex-3 gelatin-coated microcarrier beads (Pharmacia/Pfizer, New York, NY) in PBS and maintained in suspension in bacterial plates in MCDB 131 media containing 10% FBS until they reached confluency. Confluent beads were subsequently embedded into three-dimensional fibrin gels prepared as described (Nehls and Drenckhahn, 1995). Briefly, 0.5 ml of beads containing HMEC-1, were suspended in a mixture of 800 μl fibrinogen (5.45 mg/ml in PBS, pH 7.2) and 300 μl thrombin (2 U/ml, Sigma, St Louis, MO) and incubated at 37°C for 30 minutes to allow fibrin clotting. When indicated 25 μg/ml of control IgG or function blocking antibodies against human uPAR (No. 399R, American Diagnostica, Stamford, CT) were also included in the gel mixture. Following clotting of the gels, 1 ml of MCDB 131 media containing 5% FBS was added. Fresh media (with or without soluble factors) was added to the fibrin matrices every 48 hours. Cell migration was observed by phase-contrast microscopy using a Nikon TE300 inverted microscope and photographed using a Hamamatsu Orca digital camera and Open Lab Improvision software. Migration was quantified by measuring the distance that EC had migrated away from the cytodex bead after 48 hours. At least four samples were counted in each group. Statistical significance was assessed using a paired t-test.

A human cell-cell interaction array was purchased from Clontech (Palo Alto, CA). 60 μg total RNA isolated from control or HOXA3-transfected HMEC-1 were labeled using the Atlas Pure Total RNA Labeling System (Clontech). Poly-A⁺ RNA enrichment was performed using a streptavidin magnetic bead preparation and cDNA probes were synthesized following reverse transcription and amplification with the supplied primer mix and [α-³²P]dATP. Membranes were hybridized according to manufacturer's instructions. Following hybridization, membranes were exposed to BiomaxMS film (Kodak, Rochester, NY) for 1-4 days at –70°C. Signal quantification was performed by scanning densitometry and subsequent analysis performed using NIH Image 1.61 software. Microarray analysis was performed on at least two independent samples of RNA harvested from either HOXA3 or control-transfected cells.

We used 10-day-old SPAFAS pathogen-free chick embryos and performed angiogenesis assays as described (Myers et al., 2000). Briefly, either 25 μg control (β-gal) or HOXA3 expression plasmids were added to the CAM by incorporating the DNA into a 1% methylcellulose carrier. For positive controls, 50 ng recombinant VEGF¹⁶⁵ incorporated into methylcellulose carrier was added. After 72 hours, CAMs were harvested and vascular density, morphology and immunohistochemical analyses were performed. Angiogenesis was quantified by counting the number of branch points arising from the tertiary vessels in a 6-mm-square area where recombinant angiogenic factors or expression plasmids were added. Measurements were made in 12 samples treated with VEGF, control or HOXA3-expressing cDNA plasmids from three separate experiments. Statistical significance was assessed using a paired t-test.

CAMs were fixed in situ in 4% paraformaldehyde, embedded in O.C.T. medium, frozen in a dry ice/ethanol bath and stored at –70°C. 7 μm cryosections of the CAMs were used for immunohistochemistry. Following brief acetone fixation, sections were air-dried and subsequently blocked in PBS containing 2% BSA for 1 hour followed by staining with appropriate antibodies. A 1:200 dilution of a polyclonal rabbit anti-human antibody against von Willebrand Factor was used followed by Texas Red-conjugated goat anti-rabbit secondary antibody (Calbiochem, San Diego, CA). To detect HOXA3 from the expression plasmid, a 1:40 dilution of rabbit anti-myc-tag (Santa Cruz Biotechnology, Santa Cruz, CA) was used followed by FITC-conjugated anti-rabbit IgG.

C57BL/6J mice (Jackson Laboratories, Bar Harbor, ME) were anesthetized with ketamine/xylazine and a 1 cm linear full thickness wound was made through the skin. The entire region, including the panniculus carnosus layer, surrounding the incision, was harvested after 1 or 4 days, fixed in formalin and embedded in paraffin. Control, unwounded skin was harvested in parallel. 7 μm sections were prepared and deparaffinized by heating at 80°C for 30 minutes followed by two washes in xylene for 5 minutes, as previously described (Uyeno et al., 2001). Sections were rehydrated through an ethanol series, post-fixed for 5 minutes with 4% paraformaldehyde, digested with 1 μg/ml Proteinase K (Sigma, St Louis, MO) for 10 minutes and hybridized using 800 ng/ml of digoxigenin (DIG)-labeled riboprobes as described previously (Boudreau et al., 1997). Riboprobes to detect murine Hoxa3 were generated using a DIG RNA labeling kit (Roche, Indianapolis, IN) with either T7 or Sp6 RNA polymerase from a 395 bp KpnI/EcoRI fragment from the 3′ end of human HOXA3 subcloned into the pCRII-TOPO TA cloning vector (Invitrogen, Carlsbad, CA).

Total RNA was isolated from harvested wounds using TRIZOL reagent (Life Technologies, Rockville, MD) as previously described (Hansen et al., 2003a). Semi-quantitative and real-time RT-PCR of murine MMP-14 and uPAR, respectively, were carried out using the following primers: MMP-14 forward primer, 5′-CAA-CAC-TGC-CTA-CGA-GAG-GA-3′; MMP-14 reverse primer, 5′-GTT-CTA-CCT-TCA-GCT-TCT-GG-3′. uPAR forward and reverse primers were purchased as `Assays-on-Demand' (Applied Biosystems, Foster City, CA) and sequences were not disclosed, but flank the sequence of the Taqman probe, 5′-ACT-ACC-TGT-GTC-CCA-GCC-TCC-CAG-G-3′. MMP-14 amplification products were analyzed using Alpha Innotech software and normalized using 18S RNA cDNA amplified in the same reaction. Two independent experiments were averaged for MMP-14. Statistical significance was determined using a paired t-test. uPAR and control (β-glucuronidase, GUS) amplification plots from four independent samples of HOXA3- or control-treated wounds were generated using an ABI Prism 7000 Sequence Detection System and analyzed using ABI 7000 software. uPAR expression was normalized to GUS expression for each case and the results were averaged. Statistical significance was determined using a two-tailed t-test.

All animals used in this study were housed at the University of California, San Francisco animal care facility. The Committee on Animal Research approved all procedures. The mice were between 8 and 12 weeks of age at the time of wounding. Diabetic and non-diabetic littermate controls of the B6.Cg-m +/+ Lepr^db/J (Stock #000697, Jackson Labs, Bar Harbor, ME) were anesthetized with 2-3% isoflurane in oxygen at 2 litres/minute. The dorsum of the mouse was shaved and sterilized with betadine and a 2.5 cm diameter open wound was excised including the panniculus carnosus layer. Control or HOXA3 expression plasmids (4×25 μg incorporated in a 1% methylcellulose carrier) were placed directly onto open wounds. Animals received buprenorphine as needed for pain. Wound areas were measured every 7 days by planimetry, and subsequent analysis was performed using Adobe Photoshop 5.5 software. At least six diabetic animals were used in each group and at least two non-diabetic controls. For molecular and immunohistological analyses, wounds were harvested at the described time points by sacrificing the animal and removing the entire wound area, including a 2 mm region outside the wound edge, as well as the granulation tissue within the wound. Tissue samples were snap-frozen in liquid nitrogen until processing for RNA or protein isolation, or, for immunohistological analyses, tissues were fixed in formalin and embedded in paraffin, or frozen in O.C.T. compound until sectioning.

Wounds (0.8cm) were harvested at 4 and 7 days after application of either control (β-gal) or HOXA3 expression plasmids. Blind scoring was used to evaluate histological properties of wounds, according to the method of Greenhalgh and colleagues (Greenhalgh et al., 1990). Briefly, each specimen was given a score of 1 to 12: 1 to 3, none to minimal cell accumulation and granulation tissue or epithelial migration; 4 to 6, thin immature granulation tissue dominated by inflammatory cells; 7 to 9, moderately thick granulation tissue, ranging from being dominated by inflammatory cells to more fibroblasts and collagen deposition; and 10 to 12, thick, vascular granulation tissue dominated by fibroblasts and extensive collagen deposition.

Keratin 10 staining was performed as previously described (Coussens et al., 2000). Keratin 5 staining was performed using rabbit anti-murine keratin 5 (PRB-160, Covance, Berkeley, CA) at 1:1000 on deparaffinized sections in 0.5× blocking buffer (1×: 5% normal goat serum, 2.5% BSA, 1× PBS), followed by biotinylated donkey anti-rabbit at 1:200 (Pierce, Rockford, IL) and ABC-HRP (Vector Labs, Burlingame, CA) and incubation with NovaRed substrate (Vector Labs, Burlingame, CA). CD31 staining was performed using a 1:100 dilution of a rat anti-murine CD31 antibody (Pharmingen, San Diego, CA) followed by a 1:250 dilution of anti-rat FITC (MP Biomedicals, Irvine, CA).

Scratch wounds (0.1 mm) were created in confluent cultures of either control or HOXA3-expressing BALB/C MK immortalized keratinocytes and analyzed as previously described (Hansen et al., 2003a).

To test the potential of HOXA3 to promote angiogenesis, we stably transfected immortalized human microvascular endothelial cells (HMEC) with a myc/His-tagged HOXA3 expression plasmid. Expression of both the mRNA and the resulting 46 kDa HOXA3/myc/His fusion protein was confirmed by semi-quantitative RT-PCR and western blotting using an anti-His-tag antibody (Fig. 1a). To determine whether elevated expression of HOXA3 influences EC behavior, we assayed migration in HOXA3-transfected cells and control-transfected cells (Fig. 1b). Constitutive expression of HOXA3 strongly enhanced migration compared with controls, and was similar to application of recombinant VEGF, a potent inducer of EC migration (Fig. 1b). In order to directly test whether HOXA3 increases angiogenesis in vivo, we applied the HOXA3 expression plasmid to the chick chorioallantoic membrane (CAM) from 10-day-old embryos. Forty-eight hours following application of HOXA3, we observed an increase in the number of new vessels when compared with control plasmid treated CAM (Fig. 1c). Antibody staining for the myc-epitope tag confirmed expression of the HOXA3 transgene product (not shown). Similar to our findings for EC migration, HOXA3 is as potent in inducing angiogenesis in the CAM as administration of recombinant human VEGF¹⁶⁵ (Fig. 1c). Quantification of tertiary branch points confirmed that gene transfer of HOXA3 resulted in a statistically significant increase in angiogenesis in vivo when compared with CAM treated with control cDNA (Fig. 1d).

Our previous studies demonstrated that HOX gene expression profoundly affects cell-ECM interactions (Boudreau et al., 1997; Myers et al., 2000). Therefore, we performed microarray analysis to detect changes in the expression of 238 genes involved in cell-cell or cell-ECM interactions (Clontech Cell Interaction Array). When RNA harvested from control and HOXA3-expressing HMEC was hybridized, we observed upregulation of several genes that are known to contribute to the increased angiogenic potential of endothelial cells. These included uPAR and MMP-14 (Table 1) (Blasi and Carmeliet, 2002; Fibbi et al., 1998; Lakka et al., 2003). To confirm the array data, we performed both northern and western blot analysis for uPAR and MMP-14. Both exhibited upregulation of mRNA and protein levels in HOXA3-expressing EC when compared with controls (Fig. 2a,b). In order to test if these genes represented specific targets of HOXA3, we performed RT-PCR following transfection with HOXB3 or HOXD3, however, neither were upregulated by either of the paralogs (data not shown). Furthermore, overexpression of HOXA3 did not lead to changes in the expression of either HOXD3 or HOXB3, or their respective targets genes, such as β3 integrin and ephrin A1 (data not shown). Together these results show that HOXA3 promotes an angiogenic phenotype, similar to HOXD3 and HOXB3. However, HOXA3 activates expression of pro-angiogenic genes that are distinct from those activated by the products of the paralogous HOXD3 or HOXB3 genes.

As uPAR is a component of the plasminogen activator system, involved in the degradation of fibrin and other ECM constituents, we wished to test whether HOXA3-mediated activation of uPAR might cause enhanced migration of EC and/or angiogenesis. We performed a migration assay in three-dimensional fibrin matrices where fibrinolytic activity is necessary for EC migration (Kroon et al., 1999). Control-transfected EC showed little or no migration in the absence of added cytokines (Fig. 3a). Interestingly, HOXA3-expressing EC showed a marked ability to invade the fibrin gels without the addition of angiogenic cytokines (Fig. 3b). This HOXA3-induced invasion can be completely inhibited by the addition of function-blocking antibodies against uPAR (Fig. 3c,d), suggesting that the effects of HOXA3 on migration of EC may be mediated primarily by uPAR.

Having established that HOXA3 promotes EC migration and angiogenesis we next determined if murine Hoxa3 was differentially expressed in quiescent or angiogenic microvascular endothelial cells in vivo. We performed in situ hybridization for Hoxa3 in both unwounded (quiescent) and wounded (angiogenic) skin. Despite strong positive hybridization in hair follicles from unwounded tissue, Hoxa3 expression was low to undetectable in the adjacent microvessels (Fig. 4a). However, by 24 hours following administration of a full thickness linear wound, expression of Hoxa3 increased in microvessels adjacent to the wound area (Fig. 4b). After 4 days following wounding, angiogenic vessels forming in the granulation tissue expressed high levels of Hoxa3 (Fig. 4c). Sense control probes did not detect any expression (Fig. 4d). These findings establish that Hoxa3 expression is increased in endothelial cells in an angiogenic environment, and in conjunction with our previous results, further suggest that HOXA3 plays an important role in mediating angiogenesis in adults during wound repair.

We have previously shown that expression of the Hoxa3 paralog, Hoxd3, is induced during wound-mediated angiogenesis. However, wound-induced expression of Hoxd3 is markedly reduced in genetically diabetic (Lepr^db/Lepr^db) mice that exhibit both impaired angiogenesis and wound repair (Uyeno et al., 2001). Therefore we investigated whether Hoxa3 expression was lacking in diabetic mice following wounding. RT-PCR of wound tissue harvested from Lepr^db/Lepr^db (diabetic) and Lepr^db/+ (non-diabetic) mice 4 days after administration of a full-thickness excisional wound revealed that Hoxa3 expression was undetectable in diabetic mice compared with wound tissue from non-diabetic mice (Fig. 4e).

In order to test the effects of restoring HOXA3 expression in the diabetic mouse model, we applied HOXA3 or β-gal (control) expression plasmids to wounds in Lepr^db/Lepr^db (diabetic) and Lepr^db/+ (non-diabetic) mice using the methylcellulose gene delivery system previously shown to restore HOXD3 expression in vivo (Hansen et al., 2003a). Western blot analysis demonstrated that the HOXA3/myc/His fusion protein expressed by the transgene was detectable in transfected wounds after 4 days (Fig. 5a), with expression declining by the seventh day (not shown). Expression analysis of HOXA3 target genes previously identified in cell culture experiments showed activation of these genes by HOXA3 in vivo. MMP-14 mRNA expression showed a 1.5- to 2-fold upregulation by 7 days following wounding (Fig. 5b). This is a significant increase compared with control-treated wounds, and indistinguishable from MMP-14 expression levels in wounds tissue harvested from Lepr^db/+ (non-diabetic) animals (Fig. 5b). Expression analysis of uPAR mRNA expression at four days following wounding, revealed a 2- to 2.5-fold increase in expression in wounds treated with HOXA3 compared with control treated wounds (Fig. 5c). This increase in uPAR mRNA expression correlated with a dramatic increase in uPAR protein expression in HOXA3-treated wound tissue (Fig. 5d). These data support the hypothesis that HOXA3 function can be restored, at least in part, in diabetic wounds in vivo.

In addition to upregulation of downstream target genes, we found that restoring HOXA3 expression in wounds led to an improvement in overall appearance of wounds (Fig. 6, compare a and b) and wound closure rates (Fig. 6c) as early as 7 days, and persisting to 42 days after wounding when compared with control-treated wounds. Application of HOXA3 expression plasmids did not accelerate wound healing in non-diabetic littermate controls (not shown). This is consistent with the finding that wild-type animals already express abundant Hoxa3 in response to injury.

To determine whether restoring HOXA3 expression could improve angiogenesis in wound tissue we assayed endothelial cell density in the wound 7 days following application of HOXA3. Immunohistochemistry on tissue sections through the wounded area revealed increased endothelial cell density in the wounds of HOXA3-treated mice, when compared with control plasmid-treated Lepr^db/Lepr^db mice. This is apparent following staining with CD31 (Fig. 6d,e). Quantification of angiogenesis, epithelial migration, collagen deposition and other processes involved in wound repair was performed by blind assignment of a histological score to wound sections 7 days following treatment with HOXA3 or control expression plasmids (see Materials and Methods). As expected, the improvement in overall wound appearance and increased expression of the angiogenic marker CD31 correlated with a significant increase in the histological score of HOXA3-treated wounds as compared with control-treated wounds (Fig. 6f). Despite rapid wound closure following HOXA3 treatment, we did not observe any evidence of dysplastic or abnormal skin architecture in HOXA3-treated wounds, compared with untreated, healed Lepr^db/Lepr^db (diabetic) or Lepr^db/+ (non-diabetic) wounds. Each group also displayed similar ratios of keratin 5/keratin10 expression (not shown), further indicating that despite accelerated closure, HOXA3-treated keratinocytes were undergoing normal differentiation.

We have previously observed that administration of HOX genes via methylcellulose carriers can introduce trangenes into a variety of cell types in the wound, including keratinocytes (Hansen et al., 2003a). This raised the possibility that HOXA3 function may directly influence keratinocyte behavior. To determine whether improved keratinocyte migration/re-epithelialization was occurring through the direct influence of HOXA3 or indirectly via enhanced angiogenesis and granulation tissue formation, we stably transfected the BALB/C MK mouse keratinocyte line with control, HOXA3 or HOXD3 expression plasmids. Scratch wound analyses indicated that HOXA3-expressing keratinocytes showed a more rapid rate of closure than the control (Fig. 7a,b), whereas HOXD3 had a slightly negative effect on keratinocyte migration, as previously described (Hansen et al., 2003a). Western blot and semi-quantitative RT-PCR analyses revealed that, similar to EC, HOXA3-transfected keratinocytes showed increased levels of uPAR mRNA and protein (not shown). To test whether the HOXA3-mediated increase in uPAR expression also contributed to the enhanced keratinocyte migration, we added function-blocking antibodies against murine uPAR. This treatment significantly impaired HOXA3-induced keratinocyte migration (Fig. 7c). BrdU labeling indicated that HOXA3-expressing keratinocytes did not show any significant increases in the rate of DNA synthesis as compared with control-transfected keratinocytes (not shown). Together these results indicate that HOXA3 directly stimulates migration of keratinocytes as well as EC. Thus, we propose that HOXA3 is a potent inducer of wound closure by promoting cell migration and subsequently re-epithelialization and angiogenesis, via upregulation of genes such as MMP-14 and uPAR that have immediate effects on these changes in cell behavior.

Here we have shown that HOXA3 acts on epithelial and endothelial cells to induce migration and angiogenesis in response to injury. Moreover, although HOXA3 is functionally similar to the paralogous HOXD3 and HOXB3 genes (Boudreau et al., 1997; Myers et al., 2000), its ability to promote angiogenesis arises via activation of a unique set of functionally related target genes. Whereas HOXD3 functions by activating transcription of the genes encoding Collagen 1A1 and the β3 subunit of the αvβ3 integrin (Boudreau and Varner, 2004; Hansen et al., 2003b), and HOXB3 activates ephrin A1 (Myers et al., 2000), we have found that HOXA3 increases the expression of MMP-14 and uPAR, two genes known to play an important role in cell migration. These target genes are activated both in cell culture and in vivo. The connection between HOXA3 and uPAR is functionally supported by the demonstration that HOXA3-induced migration can be blocked by inhibition of uPAR function.

Gene transfer of HOXA3 suggests that HOXA3 also functions as a potent inducer of wound repair in genetically diabetic animals. A single application of HOXA3 results in complete closure of wounds in an average of 42 days, whereas control plasmid treated wounds require on average 77 days for complete closure. In our previous studies we noted that HOXD3 also accelerates closure of wounds in genetically diabetic animals that exhibit impaired angiogenesis. However, when HOXD3 expression was restored, the average time to closure was still 63 days. Although both HOXD3 and HOXA3 are potent inducers of angiogenesis in vivo, only HOXA3 is capable of directly promoting migration of keratinocytes, whereas HOXD3 is not (Hansen et al., 2003a). Therefore, the ability of HOXA3 to enhance angiogenesis and directly promote re-epithelialization, two key processes involved in efficient wound repair, probably explains the superior ability of HOXA3 to promote wound healing.

HOXA3-induced migration of endothelial cells and keratinocytes is accompanied by an increase in expression of uPAR, the receptor for urokinase plasminogen activator. Our findings that addition of function-blocking antibodies against uPAR can abolish HOXA3-dependent migration in both EC and keratinocytes strongly suggests that uPAR function is a critical mediator of HOXA3-dependent effects in improving wound repair.

In addition to promoting uPA-dependent proteolysis, uPAR has been shown to exhibit a variety of other functions, including colocalization with and binding to integrins, facilitating downstream signaling (Blasi and Carmeliet, 2002). However, as addition of aprotinin, a serine protease inhibitor, impaired HOXA3-induced migration to the same extent as addition of antibodies against uPAR (data not shown), it is likely that the HOXA3-mediated enhancement of angiogenesis and keratinocyte migration during wound repair primarily involves the proteolytic functions of uPAR.

Together, these findings are in agreement with previous studies showing that an increase in uPAR expression or activity contributes to enhanced migration of both EC and keratinocytes in culture (Daniel and Groves, 2002; Kroon et al., 1999). Loss of plasminogen activator inhibitor-1 (PAI-1), which inhibits uPA/uPAR activity, leads to improved wound healing in mice, and in isolated keratinocytes, loss of PAI-1 leads to an enhanced rate of wound closure (Li et al., 2000). Moreover, direct application of uPA, the ligand for uPAR to human dermis grafted onto severe combined immunodeficient (SCID) mice was found to be as potent an inducer of angiogenesis in dermal wounds as VEGF. Interestingly, the same study noted that application of antisense RNA against uPAR resulted in thrombosis and eventual thickening of the epidermis and papillomatosis indicating that uPAR is essential for normal dermal remodeling following wounding (Gruss et al., 2003). However, when expression of uPAR is sustained in keratinocytes of mice expressing a keratin5-uPAR transgene, skin develops normally and displays levels of keratin 10, a keratinocyte differentiation marker, similar to that in wild-type animals (Zhou et al., 2000). Not surprisingly, we observed that following wounding in Lepr^db/Lepr^db diabetic mice, application of HOXA3 and subsequent upregulation of uPAR did not alter normal skin architecture or keratin 10 expression compared with that in non-diabetic littermate (heterozygous) control mice. As diabetic mice show delayed expression of uPAR when compared with wild-type control animals following tissue grafting, we propose that addition of HOXA3 immediately following wounding may act to restore normal uPAR levels early in wound healing (Vasir et al., 2000).

Nonetheless, it is unlikely that HOXA3 and uPAR expression are sustained in our model. Similar to our previous results using HOXD3 methylcellulose-mediated gene transfer, the HOXA3 transgene could not be detected 21 days after application (Hansen et al., 2003a). PCR analysis of genomic DNA harvested from HOXA3-treated mice indicated that a functional HOXA3/myc/His cDNA could not be detected in the recipient genome (K.A.M. and N.B., unpublished). This is consistent with previous studies showing that plasmid-mediated gene transfer rarely leads to permanent integration and expression (Wang et al., 2004).

Because addition of antibodies against uPAR completely blocked HOXA3-mediated migration, the role of MMP-14 induction by HOXA3 is less clear. Previous studies have indicated that MMP-14 can also promote fibrinolysis and invasion into fibrin; however, the MMP-14-dependent invasion was assayed in tissues that lacked a functional plasminogen activator system (Hiraoka et al., 1998). Recently MMP-14 was found to be indispensable for endothelial cell invasion and neovessel formation in a collagen-rich environment, whereas the plasminogen system was not required (Chun et al., 2004). Several other groups have also shown that MMP-14 plays an essential role in mediating capillary tubule formation of invasive EC (Collen et al., 2003; Lafleur et al., 2002). It is possible that MMP-14-dependent migration into fibrin-rich environments may not be readily apparent in tissues with high levels of uPAR or other components of the plasmin/plasminogen system, whereas the contribution of uPAR may be less relevant in tissues expressing high levels of MMP-14. Additionally, other cell types certainly express other matrix metalloproteinases that function in a complementary manner. For example, although MMP-14 may be important in EC migration, MMP-10 appears to play an important role in keratinocyte migration during wound healing (Krampert et al., 2004). However, the plasminogen system may be required in keratinocytes during wound healing, as studies have shown that hypoxic induction of keratinocyte migration in cell culture can be blocked by uPA inhibitors (Daniel and Groves, 2002). In short, because of the dynamic environment of the wound, it is likely that the combined and complementary activities of HOXA3-induced uPAR and MMP-14 function to contribute to the overall increase in keratinocyte migration and functional microvessels observed in vivo.

Though the enhancer/promoter regions for both MMP-14 and uPAR contain several putative HOX consensus binding sites, at present it has not been established if these sequences are directly bound and activated by HOXA3 or whether HOXA3 upregulation of these genes is indirect. It is possible that the activation of other transcription factors, such as EGR-1, a known activator of MMP-14 (Haas et al., 1999), is required for the activation of these downstream genes.

Previous studies have indicated that although HOXD3 and HOXA3 play distinct roles during development, they appear functionally interchangeable when expressed in the correct context (Greer et al., 2000). Indeed, many of the functions of the HOX3-group genes have been found to be identical during developmental patterning (Gaufo et al., 2003). However, very little is known about their function during adult processes such as tissue remodeling during wound healing or cancer. Our results in adult endothelial cells show that although these genes are each capable of inducing angiogenesis and promoting wound repair, they do so by activating non-overlapping, but functionally related genes. The studies by Greer and colleagues, in which HOXD3 mutant phenotypes were rescued by introducing HOXA3 into the HOXD3 locus, emphasize again that both temporal and spatial regulation of these genes are critical aspects of proper HOX function (Greer et al., 2000). It is of interest to note that HOXD3, which does not promote migration of keratinocytes, is not normally expressed in this cell type, whereas HOXA3, which is expressed in primary keratinocyte cultures, can stimulate migration of these cells (Hansen et al., 2003a). This may be due to differences in target gene selection. It is interesting to note that HOXA3 activates transcription of uPAR, and not uPA, whereas HOXD3 activates transcription of uPA and not uPAR. This is an excellent example of how paralogous HOX genes may function in concert yet have subtly different effects. Thus, the distinct effects of the paralogous HOXA3 and HOXD3 genes on cells within the wound may reveal a growing divergence in their expression patterns and function in vivo. Additionally, differential target gene selection and regulation may arise from tissue-specific availability of HOX co-factors or chromatin modifications. Although there is likely to be overlap in function between the HOX3-group paralogs, the ability of HOXA3 to promote both angiogenesis and keratinocyte migration suggests that this HOX gene plays a critical role in coordinating the repair programs of different cell populations within the wound environment, whose integrated functions are essential for efficient healing. Future studies will address the mechanisms underlying HOXA3-directed changes in cell behavior, and determine how these changes affect specific tissues during wound healing. Moreover, the potential therapeutic benefits to patients with impaired wound healing using transient gene transfer techniques, such as the one described in this study, warrant further investigation of the mechanisms of HOXA3 function during adult tissue remodeling processes. We speculate that further analysis of specific cellular targets and functions of HOX proteins in adult tissue remodeling, especially during wound repair and tumorigenesis, will contribute to a deeper understanding of the enigmatic activities of Hox proteins during embryogenesis.

We are grateful to David Morris for providing the BALB/C MK cells, Michelle Rohde for excellent technical assistance and Matthew Ronshaugen for helpful comments on the manuscript. We would also like to thank Lisa Coussens for help with the keratin 5/keratin 10 staining and analysis of tissue histology. This work was funded in part by the National Institutes of Health (NRSA F32GM2084901 to S.L.H., KO8GM00674 to D.M.Y. and CA85249 to N.B.).