utilized whole‐exome sequencing to identify the genes responsible for numerous spontaneous mutations in 124 inbred mouse strains [57, 58].
Such programs, especially ENU, have proved their long‐term success but depend very much on having targeted phenotype screens that can be carried out at high throughput and strategies for recovery of the mutated gene. This has become much easier with the development of next‐generation sequencing technologies and successful sperm freezing techniques. Two of the first projects were set up in 1997 at the Helmholtz Institute in Munich and the Medical Research Council (MRC) Mammalian genome unit in the United Kingdom. These have been followed by similar projects in Australia, the United States, and Japan and to date more than 20 projects have been run worldwide [59, 60]. These projects have yielded a steady stream of novel mouse models, often generating very useful allelic series across a wide range of domains [55]. Particularly notable is the innate immunity screen set up in Bruce Beutler's laboratory with an innovative methodology for rapidly identifying the mutated allele associated with an immunological phenotype [53].
The work involved in identifying mutant alleles from ENU screens gave rise to the second strategy for randomized mutation – that of using insertional mutagenesis making use of the insertion of transposons, such as Sleeping Beauty or Piggy‐Bac, which has high mutagenesis efficiency and makes recovery of mutant alleles relatively straightforward [61].
The most comprehensive strategy, however, has been a reverse genetic strategy to individually knockout every protein‐coding gene in the mouse genome one by one, and to phenotype the subsequent strains. The first of these very ambitious projects began in 2007 with the International Knockout Mouse Consortium using a set of tailored constructs for embryonic stem cell (ES cell) mediated homologous recombination, which has been followed up by the use of the much faster CRISPR/Cas9 technology which has sped up this process and completely replaced the various molecular technologies that generated traditional null mutations. CRISPR/Cas9 is capable of generating very precise mutations that can and do mimic polymorphisms found in human populations [42, 62–65].
Unfortunately, results often varied between laboratories, most likely due to differences in environment, genetic background, and experimental approaches. Moreover, many laboratories lacked pathology support. The transition to large scale, partially blind approaches using more standardized methodologies, has greatly expanded and speeded up this process. Early, large‐scale, industrialized programs were designed to identify novel drug targets but also proved useful in predicting human disease mutations [66]. Currently the international public sector program, the US NIH Knockout Mouse Program (KOMP) and the International Mouse Phenotyping Consortium (IMPC), has systematically inactivated one gene at a time with a standardized phenotyping program [67]. Development of the comprehensive high throughput phenotyping has been critical in using these mutants to discover information about gene function and to establish new mouse models for human diseases [67, 68] To date more than 18 000 genes have been conditionally knocked out and more than 6000 of these mutant ES cells turned into mouse strains and phenotyped [69]. Combined, these programs have identified many new diseases in mice and humans [70–73].
The importance of phenotype in the approach to model discovery is critical and this has consequently lead to computational approaches to identify candidate genes for human diseases from mouse mutants by comparing phenotypes of patients with unknown lesions, to those of mice with known knockouts. Advances in biosemantics have led to the development of several platforms supporting this approach, based on semantic similarity between phenotypes of two or more species [74–76]. While the approach is able to recapitulate known gene–phenotype relationships most of the time, there remains some concern about the ability of such algorithms to identify new genes due to perceived overfitting of training sets. In addition, the IMPC report that comparing phenotypes from International Knockout Mouse Consortium (IKMC) mutants, made on a C57BL/6N background, to those already in the mouse genome informatics (MGI) database created through hypothesis‐driven research showed that only 40% had any phenotype in common at all [68]. While this figure could be due to incomplete or biased phenotyping by investigators, or incorrect phenotyping calls by the IMPC, the likelihood is that it represents strain background effects, an issue that reinforces the need for careful and detailed phenotyping on multiple backgrounds in order to find useful models.
From Human Polymorphism to Mouse Models
Some from the clinical genetics community, seeing successes from genome‐wide association studies (GWAS), claim that with modern genetics and next generation sequencing “the model for the human is now the human” and the use of other organisms is redundant. However, when combined with human genetic studies, such as GWAS for complex genetic diseases, it is now possible to reverse engineer mouse models from development toward validation of the human discoveries. This has been ongoing for many years using traditional recombineering approaches. For example, mutations in the Abcc6, ATP‐binding cassette, sub‐family C (CFTR/MRP), member and member 6 (ABCC6) gene were identified by several groups to cause pseudoxanthoma elasticum (PXE) in humans, a systemic disease characterized by ectopic mineralization. Two groups made null mutations of this gene in mice that recapitulated many of the phenotypes seen in humans [77, 78]. Many other genes produce genocopies of ectopic mineralization, and the discovery of four inbred strains with the same spontaneous Abcc6 hypomorphic allele but very different severity of PXE lesions [79] revealed many modifier genes in the mouse and by extension in humans [42].
Complex human genetic diseases, such as psoriasis, have been difficult to model in mice [80]. However, GWAS studies identified several single nucleotide polymorphisms in the caspase recruitment domain family, member 14 (CARD14) gene [81, 82]. One of these single nucleotide polymorphism (SNPs) was reproduced by three different laboratories in mice, all of which recapitulated the skin disease [83–85]. One lab made the second allelic mutation, which did not have any effect [85]. These are a few of many examples where genetic engineering in mice can either validate or fail to match molecular discoveries in human patient cohorts.
New Mouse Genetic Tools
The power of mouse genetics is being increasingly exploited, for example by using the resources of inbred strains which are now being used to construct genetic reference populations, such as the Collaborative Cross [86], which mimics the genetic complexity of human populations and the Diversity Outcross mice, a stock of truly outbred mice derived from the incipient inbred collaborative cross strains [87], that can provide defined heterozygosity. Together with the ability to manipulate the genome through reverse genetics, these novel mouse genetic tools provide complementarity to human clinical and genetic studies never before available. Medicine will continue to face major challenges for which mouse models will prove to be invaluable; for example, the ability to predict whether a particular individual will develop one or more of hundreds of different diseases for which they carry predisposing alleles, the accurate prediction of risk or prognosis or a particular disease in an individual and the prioritization of the many candidate genes arising from GWAS studies [88–90]. Success in translational research depends on the combined efforts of investigators in the laboratory and clinic to create the ideal slippers (mice) for Cinderella to go to the Ball.
Patient Derived Xenograft (PDX) Models
Mouse cancer models, be they spontaneous, induced by a variety of means, or genetically engineered, have been invaluable as tools to dissect underlying mechanisms of disease. However, they have had a smaller impact on translational studies to directly impact cancer treatment in humans. Development of refined immunodeficient mouse models [64, 91] has enabled engraftment of human cancer cells into these mice providing a tool to directly assess the human cancers in this xenograft model. These patient derived xenografts (PDX models) are used for target‐validation and drug‐assessment studies [92].
Acknowledgments
The authors thank Zoe Reifsneider for her help in preparing the graphics. This work was supported in part by grants from the National Institutes of Health (R13 OD010920, to JPS and JMW; R01‐CA089713 and