Vascularization typically occurs via two processes: vasculogenesis and angiogenesis. Vasculogenesis is the de novo formation of blood vessels that arises from the coalescence of angioblasts to form the primary capillary plexus.³ Vasculogenesis occurs largely during embryonic development; however, the presence of a population of circulating endothelial progenitor cells (EPCs) originating from the bone marrow, suggests that it may also occur in the postnatal stages of life.^4,5 Conversely, angiogenesis is the sprouting of vessels from existing vasculature and occurs during development and throughout life. This process involves endothelial cells (ECs) degrading the basement membrane of the parent vessel, collectively migrating through the provisional extracellular matrix (ECM) and subsequently establishing mature neovessels stabilized by pericytes.⁶ This requires the coordinated expression of cytokines, proteases, adhesion molecules and other cell cycle regulatory factors. During embryonic development, angiogenesis is the means by which vessels expand from the primary capillary plexus into an extensive, closed circulatory network. Excluding corpus luteum formation and endometrium redevelopment, angiogenesis generally does not occur in adults, but is initiated during wound healing or in response to tumor growth.

Activation of angiogenesis under pathological conditions is controlled by what Judah Folkman termed the “angiogenic switch,”^7,8 which was originally coined to describe how tumors generate an angiogenic stimulus to support their growth and viability. This concept states that at homeostasis, there exists a balance between pro-angiogenic and anti-angiogenic factors, but under pathological conditions, expression of pro-angiogenic factors tilts the balance toward vessel formation, and conversely, when the process is complete the anti-angiogenic factors prevail, resulting in vessel pruning and regression. Tumors were shown to mimic features of healing wounds by producing copious amounts of angiogenic factors (e.g., VEGFA and FGF2) and macrophage chemoattractants (e.g., CCL2 and CSF1).^8-10 Macrophage infiltration at the tumor site augments the pro-angiogenic environment by increasing the local availability of pro-angiogenic cytokines.^11-17 However, exogenous addition of anti-angiogenic factors, such as thrombospondin 1, causes vessel regression by reversing the tilt of the angiogenic switch.¹⁸ Restoration of tumor suppressor genes that limit tumor growth and subsequent metabolic demand can also shift the balance toward an anti-angiogenic environment.^19,20

Homeobox genes are divided into families and subfamilies based on the similarity of their DNA-binding homeodomains. Most families contain between two to nine genes; these include the Msx, En, Nkx, Pax and Dlx families. However, the largest, and most extensively studied, family of homeobox genes is the class I Hox family, which includes 39 members in mammals. The Hox genes are arranged in clusters—with eight genes clustered on a single chromosome in Drosophila, while mammals have 39 genes grouped in clusters of 9–13 genes on each of four chromosomes. During vertebrate development, there exists a strict temporospatial control of Hox gene expression in a manner such that 3′ Hox genes are expressed prior to 5′ genes within a given cluster.²⁴ This expression pattern is responsible for controlling segmental identity and morphology on the anteroposterior axis; 3′ Hox genes are associated with the development of anterior or rostral tissues and 5′ genes are associated with posterior or caudal tissues. Mutations or deletions that lead to misexpression of Hox genes have shown to result in homeotic transformations in Drosophila and mammals, in which anterior features develop in posterior regions and vice versa.^22,25,26 The finding that Hox genes are arranged into highly conserved chromosomal clusters, with the lowest density of interspersed repeats in the entire human genome,²⁷ is most likely due to the need to preserve cis-regulatory elements to maintain this temporospatial control.

In humans, the 39 mammalian Hox genes are arranged into four chromosomal clusters; designated HOXA, HOXB, HOXC and HOXD, located on chromosomes 7p14, 17q21, 12q13 and 2q31, respectively. Each of these 39 genes is assigned to one of 13 paralog groups, with no single cluster containing a gene from each group (Fig. 1). Hox paralogs, defined by their homeodomain sequence, occupy the same relative positions on a given chromosomal cluster (e.g., HOXA3, HOXB3 and HOXD3).²⁸ Paralogous group members have been shown to exhibit some level of functional redundancy, as they are able to perform compensatory functions during development upon mutation or deletion of a fellow group member.^29-34 In these experiments, mice that are double or triple mutants for paralogous Hox group members display more complete and severe phenotypes than single mutants of any of these paralogs. Functional redundancy between non-paralogous, neighboring Hox genes,^35-37 as well as non-paralogous genes in separate clusters has also been documented.³⁸

The initial phases of embryonic vascular development consist of the formation of blood islands through the proliferation and differentiation of mesodermal hemangioblasts into angioblasts and hematopoietic progenitor cells.³ Fusion of these blood islands results in the formation of the primary capillary plexus and the primitive heart tube.

Early studies by Chisaka and Capecchi linked Hox function to embryonic vascular development in vertebrates.³⁹ They observed that Hoxa3 (originally referred to as Hox1.5)-knockout mice died at or shortly after birth, with abnormalities including an absent right carotid artery and irregularly sized heart compartments. These effects could be ameliorated by inserting Hoxd3 within the Hoxa3 locus, suggesting that the paralogs are somewhat interchangeable if expressed “properly” in space and time.³⁴

In addition, the non-Hox homeobox genes Prx1 and Prx2 are normally expressed in the vessel wall during development. Prx1^−/− mice displayed an irregularly curved aortic arch along with a misdirected and elongated ductus arteriosus.⁴⁰ In contrast, Prx2^−/− mice did not display any vascular deformities. However, Prx1^−/− Prx2^−/− double mutant mice presented similar, but more pronounced, abnormalities as the Prx1^−/− mice. Interestingly, although the great arteries were malformed in the double mutants, the heart developed normally compared with wild-type mice. Additionally, vascular smooth muscle cells (VSMCs), which represent the major cellular component of the vessel wall, were found to express HOXA5, HOXA11, HOXB1, HOXB7 and HOXC9 in fetal tissue, while Hox gene expression was absent in their adult counterparts, suggesting their importance during early vascular development.⁴¹

Subsequent studies revealed that homeobox genes, at least partially, direct the differentiation of EPCs into ECs. Transactivation of Vegfr2 occurs via cis-acting Hoxb5 in embryoid bodies, resulting in their differentiation into angioblasts. These angioblasts expand and further differentiate into an endothelial phenotype.⁴² We have also profiled the dynamic expression of several Hox genes during the differentiation of embryonic stem cells into ECs in culture.⁴³ Early expression of HOXA3 and HOXD3 was concomitant with the emergence of an EC-phenotype and upregulation of pro-angiogenic factors such as CCL2, MMP14 and the β3 integrin subunit. Furthermore, Hoxa3 is critical in maintaining the endothelial phenotype in embryonic vasculature and preventing its hematopoietic differentiation.⁴⁴ At later time points, HOXA3 and HOXD3 are downregulated while the anti-angiogenic HOXA5 and HOXD10 are upregulated. This corresponds with the dowregulation of the aforementioned pro-angiogenic factors and the expression of the potent angiogenesis inhibitor thrombospondin 2. A separate study identified HOXA9 as a master regulator of EC differentiation from progenitor cells by directly enhancing transcription of VEGFR2 and eNOS.⁴⁵

The non-Hox homeobox gene Hex plays a complex role in vascularization. During embryonic vasculogenesis, Hex is expressed in the blood islands, and is required for endothelial and hematopoietic differentiation.⁴⁶ Hex-knockout mice displayed embryonic lethality typified with several defects, including impaired vasculogenesis.⁴⁷ However, Hex inhibits EC tube formation in culture through the downregulation of angiogenic factors such as VEGFR1, VEGFR2, TIE1, TIE2 and the α_v integrin subunit,^48,49 suggesting that Hex promotes vasculogenesis rather than angiogenesis, or that other factors are integral in determining the outcome of Hex signaling (the anti-angiogenic effects of homeobox genes will be subsequently detailed in this review). The precise hierarchy and interactions between each of these Hox factors during EC differentiation remains to be established, but suggests a complex network of factors required for EC development.

Hox genes also serve important roles in vascularization post-development in response to injury or other pathological conditions. HOXD3, through binding of cofactor PBX1, directly enhances EC expression of the α₅ and β₃ integrin subunits in response to FGF2.^50-52 Heterodimerization of these subunits with separate subunits form the fibronectin receptors α₅β₁ and α_vβ₃, which allow adhesion and migration through fibronectin-rich matrices found in wound healing and tumor microenvironments. HOXD1 was recently shown to provide a synergistic effect to this by directly enhancing the EC expression of the β₁ integrin subunit.⁵³ Disruption of HOXD3 expression inhibited the expression of the α₅ and β₃ integrin subunits, and was sufficient to inhibit angiogenesis, consistent with previous findings that these integrins are essential for angiogenesis.^54,55 Additionally, HOXD3 upregulated the expression of the serine protease urokinase plasminogen activator (uPA), which locally activates plasmin during the degradation of the provisional matrix,⁵⁰ suggesting that EC HOXD3 is associated with the invasive phases of angiogenesis. Subsequent studies revealed that the paralogous Hox members Hoxa3 and Hoxb3 also display pro-angiogenic capabilities, albeit through distinct pathways, consistent with the notion of functional redundancy among Hox paralogs.³⁴ EC expression of Hoxa3 correlated with increased EC migration and expression of the proteases uPA receptor (uPAR) and Mmp14,⁵⁶ while HOXB3 increased the expression of the angiogenic ligand ephrin A1 and induced capillary morphogenesis in endothelial sprouts.⁵⁷ Because adult angiogenesis is generally limited to pathological conditions, it is not surprising that these pro-angiogenic HOX3 paralogs are not normally expressed in quiescent vasculature, but are induced in response to cytokines linked to tissue injury or tumors.^50,51,56,58

The 5′ Hox gene HOXD10 was shown to be expressed in normal, quiescent vascular endothelium, but not in highly angiogenic, tumor-associated vasculature.²⁰ Overexpression of HOXD10 in ECs mitigated their angiogenic response to FGF2 and VEGFA in two and three-dimensional cultures. Furthermore, HOXD10-expressing ECs showed a marked decrease in the angiogenic factors RHOC, β₄ and α₃ integrins, MMP14, uPAR and cyclin D1. The expressions of the pro-angiogenic HOXA3, HOXB3 and HOXD3 were not altered, suggesting that HOXD10 functions to mute the downstream effects of those genes, rather than attenuate their expression.

In addition to its direct anti-angiogenic effect on ECs, HOXD10 also attenuates tumor angiogenesis by acting within tumorigenic epithelial cells to reduce expression of angiogenic factors including VEGFA.⁶¹ Restoring expression of HOXD10 in mammary tumor cells, from which it is frequently lost, reestablished polarity and the formation of acinar structures in three-dimensional cultures and reduced EC recruitment.⁶¹ It is also worth noting that the ability of HOXD10 to reduce VEGFA expression was not due to a direct inhibition of VEGFA transcription, or interference with binding of the transcriptional activator HIF1A, but arose indirectly through the re-establishment of tissue polarity. When HOXD10 was expressed, but tissues were prevented from adopting basolateral polarity, expression of VEGFA and other angiogenic factors remained high. Thus, the ability of Hox genes to modulate the angiogenic capacity of tissues can arise directly through transcriptional modulation of angiogenic effector genes, or indirectly via the Hox mediated changes in tissue architecture.

The findings that HOXD10, whether expressed in epithelial or endothelial cells has a similar inhibitory effect on angiogenesis, also highlights the fact that Hox genes have the capacity to coordinate the responses of different cell populations within a tissue. These coordinated actions represent one means by which these morhpregulatory genes can work to collectively maintain differentiated tissue function.

HOXA5 is another gene expressed in quiescent, but not activated, endothelium.⁶² Similar to HOXD10, HOXA5-expressing ECs did not respond to the angiogenic stimuli provided by either VEGFA or FGF2. Among the angiogenic genes that were downregulated in ECs were VEGFR2, ephrin A1, HIF1A and COX2. Correspondingly, these cells also upregulated the expression of the anti-angiogenic thrombospondin 2. Further evidence demonstrated that EC HOXA5 stabilized adherens junctions through β-catenin retention.⁶³ Presumably, this prevents endothelium dissociation in the parent vessel—the first step in angiogenesis. Interestingly, paralogous Hoxb5 in ECs was shown to promote both vasculogenesis⁴² and angiogenesis,⁶⁴ counterintuitive to the paralog redundancy theory.

The homeobox gene GAX (MEOX2) is also largely confined to quiescent cardiovascular tissue in adults. Originally found to induce G₀/G₁ cell cycle arrest in VSMCs,^65,66 GAX, and its related gene MEOX1, were also found to be expressed in the endothelium and induced similar growth arrest in ECs.^67,68 Additionally, GAX-expressing ECs, through NFκB inhibition, exhibited an impaired angiogenic response to stimuli by VEGFA, FGF2 and TNFα.⁶⁹ Translation of both GAX and HOXA5 is directly regulated by MIR130A (located outside of the Hox clusters).⁷⁰ EC expression of MIR130A downregulates these homeobox genes and effectively restores EC activation by angiogenic stimuli.

Thus, the ability of tissues to activate the angiogenic switch relies upon upregulation of pro-angiogenc genes and suppression of anti-angiogenic or vascular-stabilizing genes (Fig. 2).

Given the essential role of angiogenesis in wound healing, it is not surprising that several homeobox genes contribute to effective wound repair. Gene delivery of the pro-angiogenic Hoxd3 accelerated wound closure in poorly healing diabetic mice and was accompanied by a robust induction of angiogenesis.⁵⁸ As transgene uptake is evident in ECs, as well as keratinocytes and fibroblasts, wound closure was accompanied by increased collagen (Col1a1) production mediated by fibroblasts. Similar to Hoxd3, the paralogous Hoxa3 was found to be upregulated in ECs in a wound healing mouse model, whereas diabetic mice were unable to induce its expression;⁵⁶ Hoxa3 transfer into the wounds of these diabetic mice significantly improved angiogenesis and wound closure. Mechanistically, in addition to the angiogenic response by ECs detailed previously, HOXA3 promoted uPAR-dependent keratinocyte migration and subsequent epithelialization—a feature that was not directly enhanced by HOXD3. Furthermore, HOXA3 expression within wound tissue recruited bone marrow-derived progenitor cells,⁷³ and promoted their differentiation into pro-angiogenic, CD11b⁺ Gr-1⁺ myeloid cells.⁷⁴ The synergistic functions of HOXA3 and HOXD3 in wound healing mimic their complementary roles in angiogenesis, and further support the functional redundancy concept conceived from double Hoxa3/Hoxd3-knockout mice.²⁹

While wound contraction is a normal part of wound healing, excessive or prolonged contraction leads to diminished tissue function and appearance. Enhanced angiogenesis within the wound bed reduces the contractile activity of myofibroblasts and scarring.⁷⁵ Building from this concept, we have engineered composite skin grafts consisting of a dermal fibroblast layer, which produces a secretable form of HOXA3, overlaid with an epidermal keratinocyte layer. Following wounding and engraftment onto mice, the secretable HOXA3 is taken up by keratinocytes and ECs within the wound tissue, and these grafts displayed increased angiogenesis and reduced contraction compared with controls (unpublished observations, manuscript submitted). Additionally, cells within the graft produced less inflammatory cytokines, consistent with the reduced inflammatory response in wounds treated with the HOXA3 transgene.

One of the cornerstones of physiological angiogenesis is the transition from vessel invasion to vessel maturation. ECs in the maturation phase show decreased proliferation relative to their invasive counterparts, and also intimately associate with supporting pericytes, which function to regulate vessel tone and leakage. As mentioned, EC HOXD3 expression correlates with the invasive angiogenic phenotype, and correspondingly, Hoxd3 overexpressing vessels formed in the chick chorioallantoic membrane (CAM) assay did not mature, but formed large endothelima-like masses.⁵⁰ This result is in agreement with the finding that HOXD3 is expressed in proliferating hemangiomas, but not involuting ones.⁷⁸ Similarly, tumor-associated vasculature, which is typically comprised of immature, leaky vessels, also highly expressed HOXD3.⁵¹ It is also worth noting that HOXA5, which can stabilize vessels and reduce permeability, is absent in both endothelioma tissues and tumor vessels,^62,63 suggesting that the unopposed actions of HOXD3 contribute to aberrant vascular morphology. Indeed, ectopic expression of HOXA5 in endotheliomas leads to a reduction in cavernous cyst formation and restores the phenotype of normal, differentiated vessels.⁷⁹

Tumor expression of HOXB7 or HOXB9 has correlated with an increase in VEGFA, and the neutrophil attractants CXCL1 and IL8.^80-82,84 HOXB13-mediated inflammation within wound tissue is characterized by neutrophil infiltration and the development of irregular vasculature.⁷⁷ Additionally, HOXA9 was found to bind the CCL2 promoter and enhance its transcription in individuals displaying a single nucleotide polymorphism within the promoter region.⁹⁰ Conversely, we have found the anti-angiogenic HOXD10 reduces tumor cell production of CCL2 and CXCL12, and abates leukocyte recruitment in culture (unpublished data). Interestingly, while HOXA3 also attenuates recruitment of CD45⁺ cells in wound tissues, it selectively promotes recruitment/differentiation of endothelial precursors and pro-angiogenic myeloid cells.^73,74

Expression of HOXA9 during EPC differentiation and angiogenesis is reliant on histone deacetylase (HDAC) activity,⁴⁵ while upregulation of both HOXA9 and HOXD3 has been shown to occur via histone trimethylation by the TrxG protein MLL.¹⁰⁰ The expression of the PcG gene MEL18 (PCGF2) is linked to an upregulation in VEGFR2 and ANGPTL2, with a corresponding increase in EC migration and tube formation;¹⁰¹ however, its downstream effectors and global function is still debated, as another group was able to show that MEL18 exhibits anti-angiogenic activity.¹⁰²

An exhaustive study of Hox gene expression in several esophageal carcinomas revealed abnormally high expression of several 5′ Hox genes in a tissue that normally displays low levels of 5′ Hox genes and high levels of 3′ Hox genes.¹¹² Similarly, expression of the 5′ HOXB13 is lost in colorectal tumor cells, leading to increased cell proliferation,^116,117 The 3′ HOXB7 was shown to directly upregulate EGFR and confer tamoxifen resistance to estrogen receptor-positive breast cancer cells,¹¹⁸ while loss of the 5′ HOXA9 and HOXD10 are frequently associated with breast cancer progression.^109,119 Contrastingly, overexpression of HOXA9 in several leukemias correlates with chemotherapy resistance and poor patient prognosis. Transgenic studies with Hoxa9^−/− mice demonstrated an inhibition of leukemogenesis, but also impaired normal hematopoiesis in the process;¹²⁰ Hoxa9^+/− mice developed normally with no leukemogenesis, demonstrating a dose-dependent response of Hox gene expression mediating the shift between organogenesis and oncogenesis.¹²¹ Thus, the potential of HoxA9 to act as either a tumor suppressor in breast tissue or an oncogene in leukocytes underscores the tissue-specific and context-dependent function of Hox proteins. Whether tissue-specific repertoires of Hox cofactors and/or differences in tissue architecture (adherent vs. circulating cells) contribute to the differential responses has not been directly addressed. It is worth noting again, that the ability of HOXD10 to reduce VEGFA expression and revert the tumorigenic phenotype of breast epithelial cells is highly context-dependent, as it requires cell adhesion to the basement membranes and acquisition of apical-basal polarity.⁶¹

As mentioned in the previous section, the expression of several Hox genes is regulated by histone trimethylation by TrxG or PcG proteins. Changes in chromatin conformation allow CpG motifs in Hox promoter regions to be further methylated or demethylated. Epigenetic deregulation occurs when members of this regulatory process become mutated or misexpressed. The TrxG gene MLL is mutated in some leukemias, causing irregular histone methylation and increased expression of HOXA4, HOXA5, HOXA7 and HOXA9.^114,115 The inactivation or loss of the tumor suppressor gene P16 was shown to upregulate the PRC2 proteins and consequently repress HOXA9 in mammary epithelial cells, thereby perpetuating tumorigenesis.¹⁰⁶

Additionally, miRNAs that regulate Hox expression may also have altered expression patterns in tumor cells through methylation or other means.^122,123 For instance, upregulation of MIR10B in mammary tumor cells inhibits translation of HOXD10 mRNA and produces robust invasion and metastasis.¹²⁴

Hox gene dysregulation arises through a variety of mechanisms and produces a wide array of aberrant responses in several tumor types; it is still unclear whether Hox dysregulation is a cause or consequence of tumor progression. However, in many instances, acquisition of the tumorigenic phenotype by Hox misexpression results in the upregulation of angiogenic factors that act on adjacent ECs and alter expression of endogenous Hox genes, leading to the formation of new vessels (for a comprehensive review of Hox genes and cancer, see ref. ¹²⁵).

This phenomenon, termed “phenotypic suppression” in invertebrates^126,128 and “posterior prevalence” in vertebrates,¹²⁹ is believed to be part of an ancient, poorly understood, cross-regulatory mechanism, in which posterior Hox transcription factors can inactivate anterior Hox proteins possibly through mechanisms that do not involve transcriptional repression (i.e., even if both genes are expressed, the posterior one is functionally dominant).¹³⁰ Alternatively, two groups of miRNA sequences, MIR10 and MIR196, which downregulate, several 3′ Hox genes have been identified within the Hox chromosomal clusters, and possibly contribute toward this phenomenon (Fig. 1).^124,131,132

However, regardless of the mechanistic basis, the posterior prevalence theory has its share of shortcomings, as several groups have reported instances where the 3′ Hox gene exerts dominance,¹³³ 3′ and 5′ Hox genes act cooperatively,¹³⁴ or the functional response is dose-dependent.¹³⁵ In all of these models, however, the prevailing theme is that repression of an effector gene product takes precedence over its activation, as is the case for other transcriptional control pathways.¹³⁶

One plausible model for repression involves the requirement of cofactors for Hox proteins to efficiently bind DNA. Most 3′ Hox proteins contain a hexapeptide motif that allows them to bind members of the PBX family of transcription factors, while most 5′ Hox proteins (members of paralog groups 10–13) lack this motif.^137,138 3′ Hox-PBX-DNA complexes exhibited dramatically increased stability compared with 3′ Hox-DNA complexes alone. Many of the Hox genes discussed in this review have been characterized to complex with PBX proteins, including HOXA9,¹³⁹ HOXB3, HOXB5, HOXB7, HOXB9¹³⁸ and HOXD3.⁵² The common theme among these factors is that they are all activators of angiogenesis, wound healing or both (Table 1). In contrast, anti-angiogenic HOXD10, and related paralogs, can bind target DNA with much higher affinity in the absence of PBX proteins. Thus, by requiring additional factors for transcriptional activation by the pro-angiogenic Hox genes, it is evident how repressor Hox proteins could exert functional dominance over activators. Interestingly, ectopic expression of Hoxd10 in Drosophila antennal imaginal discs inhibited nuclear translocation of EXD (Drosophila homolog of PBX1 and PBX2) and rendered it unavailable for cooperative DNA binding with 3′ Hox genes.¹⁴⁰ It remains to be seen if this mechanism is recapitulated in mammalian cells, but it does suggest a possible means by which the anti-angiogenic HOXD10 can dominantly counteract Hox3-mediated pathological vascularization.

For instance, overexpression of HOXD3 in healthy endothelium may trigger an unwanted inflammatory response or generate large cavernous vessels reminiscent of endotheliomas. To address this issue, our group commonly uses Hox DNA expression plasmids embedded within methylcellulose pellets to limit gene uptake to cells within wound tissue and their duration of expression.^56,58

While topical application presents a straightforward approach for cutaneous tissues, targeted delivery strategies must be developed for tumors in internal organs. The mammary tumor suppressing properties of HOXA5 and HOXD10 have been documented,^109,143 but a clinically viable treatment for breast cancer would need to target delivery of these Hox genes/proteins to tumors and their metastases in an effective manner. Like most systemic chemotherapies, systemic Hox delivery would likely produce unwanted side effects; therefore, engineering targeted-gene delivery vehicles can reduce off-target effects, but will require identifying appropriate tumor biomarkers for targeting. Moreover, while localized delivery to large tumors may be effective in reducing primary tumor angiogenesis, additional approaches must be considered to limit further progression linked to metastases.

Homeobox gene therapy can also be leveraged in the field of regenerative medicine to induce vascularization within engineered tissue; however, in these cases it will be important to temporally regulate expression levels to effectively transition through the different phases of angiogenesis. In situations such as these, designing inducible homeobox gene expression systems, placing homeobox expression cassettes downstream of phase-specific promoters, or delivering Hox proteins rather than genes may prove beneficial.

No potential conflicts of interest were disclosed.