Gene targeting provides the means for creating strains of mice with mutations in virtually any gene. This technology permits the evaluation of the function of genes in an intact mammal and the systematic dissection of the most complex of biological processes, such as development and learning. The investigator not only chooses which gene to modify but also has virtually complete control over how that gene's DNA sequence is modified. This not only permits disruption of the gene's activity in every cell of the mouse but also can restrict the gene's activity in selected cells of the mouse and/or to distinct temporal periods during the mouse's developmental or postdevelopmental life span. Because nearly all biological phenomena are mediated or influenced by genes, this technology is impacting the analysis of all such phenomena in mammals, including the study of cancer, neurobiology, development, immunology, and human genetic disease.

One major project being pursued in our laboratory is the use of gene targeting to genetically dissect the functions of the murine Hox complex. These genes function as master transcription control switches, specifying positional information along the major mammalian embryonic axes. As a consequence, the network of Hox genes orchestrates the regionalization of the embryo and thereby mediates the formation of the vertebrate body plan. In humans and mice, the Hox complex contains 39 genes present in four linkage groups on four separate chromosomes. The four linkage groups are believed to have arisen early in vertebrate evolution as a result of successive duplications of a single ancestral linkage group common to both vertebrates and invertebrates. This expansion may well have played a critical role in the evolutionary progression of vertebrates by supplying to this network of genes the necessary regulatory complexity to accommodate the development of more advanced body plans. In addition, Hox genes appear to have been usurped during evolution of vertebrates to mediate a wide range of functions during the postdevelopmental period (i.e., in adults). For example, we have demonstrated that mice with targeted disruptions in Hoxc13, in addition to having defects in the formation of the posterior aspect of the skeleton, do not produce any body hair as adults. Similarly, mice mutant for Hox9 paralogous genes, in addition to exhibiting embryonic defects, fail to form mammary tissue in response to pregnancy.

We have generated mice with targeted disruptions in all 39 Hox genes. Analysis of mice homozygous for individual Hox gene mutations defines their separate roles in patterning the mammalian embryo. However, this complex of genes functions as a highly interactive network. To reveal the interactions among Hox genes, we are analyzing the phenotypic consequences of combining two or more Hox gene mutations in a single embryo. Hox genes at the same relative position in each linkage group are most closely related with respect to DNA and protein sequence, and they exhibit similar expression patterns. Such genes are designated as belonging to the same paralogous family. It is anticipated that paralogous Hox genes will perform closely related functions.

An excellent example of redundancy within paralogous Hox gene family members is the role of Hox10 and Hox11 paralogous genes in patterning the mammalian skeletal system. The spectrum of perturbations of the mammalian skeleton from either gain- or loss-of-function mutations in individual Hox genes has been difficult to interpret in terms of a coherent model of how these genes participate in the patterning of the axial skeleton. For example, loss-of-function mutations have yielded changes in vertebral morphology that have been interpreted as anterior homeotic transformations, as posterior transformations, and as simple malformations. Typically, these morphological changes involve perturbations in one or a small number of vertebrae. By generating mice in which all members of the Hox10 or Hox11 paralogous group are disrupted, we provide evidence that the mammalian Hox genes are involved in global, as opposed to localized, patterning of the axial skeleton. In the absence of Hox10 paralogous function, no lumbar vertebrae are formed. Instead, ribs project from all vertebrae, extending caudally from the last thoracic vertebrae to beyond the sacral region, and these elements assume a thoracic morphology. In the absence of Hox11 paralogous group function, sacral vertebrae are not formed and instead, these vertebrae assume lumbar identity. Remarkably, the redundancy among these paralogous family members is so great that this global aspect of Hox gene patterning is not apparent even in mice mutant for five of the six paralogous alleles.

From the above results we can begin to postulate mechanisms explaining how changes in Hox gene expression could account for variation of the axial formula in different vertebrate taxa. For instance, one would predict that shifts of the boundaries of Hox10 paralogous gene expression, rostrally or caudally, would contract or expand the number of thoracic vertebrae present in a vertebrate animal. Similarly, shifts in the expression of the Hox11 paralogous genes would predict an alteration in the position and number of sacral vertebrae. Many fish and primitive tetrapods have ribs projecting from all vertebrae, extending from the head to the tail. This has led paleontologists and vertebrate anatomists to suggest that the ground state for vertebrae includes rib projections. Our data from the mouse support this hypothesis and provide a mechanism whereby Hox genes have been used during evolution to suppress and modify rib formation in the lumbosacral region.

Recently we have shown that mice with targeted disruption of Hoxb8 exhibit a dramatic behavioral defect. Surprisingly, mice homozygous for this mutation show no apparent external or internal morphological malformations. Instead, at approximately 3.5 weeks of age and older, these mutant mice show, with 100 percent penetrance and in two different genetic backgrounds, excessive grooming that leads to hair removal and self-inflicted wounds at those sites. Behavioral tests suggested that these mutant mice have normal reactions to touch, pressure, and pain-inducing stimuli. Further analysis has not revealed any evidence of peripheral nerve or skin abnormalities that could account for this behavior. The most convincing evidence that the excessive grooming is not a response to local irritation is, however, that when these mutants are placed together with wild-type littermates, the mutants also groomed the littermates to the extent of inducing hair loss.

Videotape analysis has shown that the behavioral defect is specific to grooming, body licking, and biting. Other behaviors—such as locomotion, eating, drinking, nest building, and exploration—appear normal. Furthermore, the pattern of grooming (i.e., the grooming syntax) is normal and occurs at normal times (i.e., before rest). It is the control of grooming behavior that is abnormal. The mutant mice spend twice as much time grooming and initiate grooming bouts more frequently than control, sex-matched littermates.

In humans, exaggerated grooming behavior occurs in obsessive-compulsive (OC) disorder and the related OC-spectrum disorders. Of particular interest is the human OC-spectrum disorder, trichotillomania, which involves uncontrollable hair pulling that can be so severe as to lead to baldness and loss of eyelashes and eyebrow hair. Functional imaging analysis of patients with OCD has shown abnormal activity in the striatum, the orbital cortex, and the anterior cingulate cortex, which has become known as the OCD circuit. Interestingly, Hoxb8 is strongly expressed in these regions of the central nervous system as well as in regions previously implicated in the execution of grooming behavior. It will be of great interest to determine whether subsets of human patients with trichotillomania have defects in the HOXB8 gene or in paralogous family members, HOXC8 and HOXD8. To address whether Hoxb8 mutations affect the formation of the neural circuitry involved in repetitive behavior, the modulation of this circuitry after its formation, or both, we have generated conditional alleles of this gene. These conditional alleles will also allow us to study the contribution of subcomponents of this circuitry to the behavioral output.

Using conditional mutagenesis, our laboratory has recently modeled alveolar rhabdomyosarcoma and synovial sarcoma, both very aggressive childhood and young adult cancers. New therapies will require better understanding of the molecular etiologies of these cancers. Both cancers appear to be initiated by separate, distinct translocations. Expression of the respective fusion genes early in embryonic development leads to lethality, explaining the failure of earlier attempts to model this cancer. We have generated conditional knock-in models of the above cancers that allow production of the respective fusion proteins in chosen tissues of the mouse, before or after birth. With respect to modeling alveolar rhabdomyosarcoma it was clear that it is a muscle cancer, but it was unknown which cell of the muscle lineage gives rise to this cancer. On the other hand, neither the tissue nor cell of origin for synovial sarcoma was known.

Surprisingly, activation of the Pax3:Fkhr fusion gene in fully differentiated muscle but not muscle satellite cells led to the formation of alveolar rhabdomyosarcoma. As expected, the frequency of tumor formation was low (i.e., 1 tumor/200 mice/year) but the tumors were exclusively alveolar rhabdomyosarcoma. However, if we added second-site mutations, conditionally in the same tissue, such as p53 mutations that are commonly found in human alveolar rhabdomyosarcoma, the frequency of these tumors increased enormously, with mice showing multiple tumors at 4 months of age. These tumors are histologically and immunohistochemically extremely similar to the human tumors. The Pax3:Fkhr fusion gene appears to lead to dedifferentiation of the muscle tissue, formation of rhabdomyoblasts, and subsequent progression toward a malignant cell phenotype. The pattern of metastases in these mice also appears to be very similar to that of the human tumor.

Expression of the SYT:SSX fusion gene, the signature genetic event leading to human synovial sarcoma, in Myf5-expressing muscle myoblasts alone is sufficient to initiate formation of synovial sarcomas with 100 percent penetrance. The tumors are exclusively synovial sarcoma-like and resemble human synovial sarcomas in terms of time of origin (young adults), place of origin (near joints), histopathology (monophasic, biphasic, and poorly differentiated), immunohistochemistry, and transcriptional profile (microarray analysis). The mouse model defines a potential tissue and cell of origin. Expression of the SYT:SSX fusion gene ubiquitously early in embryogenesis or early in the muscle lineage (Pax3- or Pax7-expressing cells) leads to embryonic lethality. Expression of the SYT:SSX fusion gene in differentiated muscle leads to a myopathy without tumor formation. Thus, only expression of the signature fusion gene in muscle myoblast (i.e., Myf5-expressing cells) results in the formation of synovial sarcoma-like tumors. Both of the above mouse models should prove useful for elucidating the molecular etiologies of these aggressive tumors and function as platforms for the generation of tumor-specific therapies.