Systematic screening for skin, hair, and nail abnormalities in a large-scale knockout mouse program

The International Knockout Mouse Consortium was formed in 2007 to inactivate (“knockout”) all protein-coding genes in the mouse genome in embryonic stem cells. Production and characterization of these mice, now underway, has generated and phenotyped 3,100 strains with knockout alleles. Skin and adnexa diseases are best defined at the gross clinical level and by histopathology. Representative retired breeders had skin collected from the back, abdomen, eyelids, muzzle, ears, tail, and lower limbs including the nails. To date, 169 novel mutant lines were reviewed and of these, only one was found to have a relatively minor sebaceous gland abnormality associated with follicular dystrophy. The B6N(Cg)-Far2tm2b(KOMP)Wtsi/2J strain, had lesions affecting sebaceous glands with what appeared to be a secondary follicular dystrophy. A second line, B6N(Cg)-Ppp1r9btm1.1(KOMP)Vlcg/J, had follicular dystrophy limited to many but not all mystacial vibrissae in heterozygous but not homozygous mutant mice, suggesting that this was a nonspecific background lesion. We discuss potential reasons for the low frequency of skin and adnexal phenotypes in mice from this project in comparison to those seen in human Mendelian diseases, and suggest alternative approaches to identification of human disease-relevant models.


Introduction
How many genes are involved in the normal ontogeny and physiology of the skin and its adnexal structures? How do they interact and how are they controlled? These questions, simple to articulate, have complex answers at best, but are predicated on our knowledge of gene function and more importantly the pleiotropic functions of genes, which can only be elucidated though experimentation and particularly through genetic manipulation. Faced with this challenge, we might initially search major databases such as Mouse Genome Informatics (MGI; http://www.informatics.jax.org/), Online Mendelian Inheritance in Man (https://www.omim. org/), or Orphanet (http://www.orpha.net/) and ask whether naturally-occurring or genetically a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 number of parameters in a large number of mice and consequently is designed to pick up relatively clear-cut examples of phenodeviance.
Prior to the initiation of the current international efforts, both Lexicon and Deltagen undertook large-scale systematic mutagenesis and phenotyping projects, of which Lexicon's used systematic histopathology as a routine phenotyping tool (see [10] for discussion) and generated histopathological analyses on nearly 5,000 mutant strains. Detailed descriptions of these strains are now in the process of being published in the literature [11,12].
More recently, the Sanger Mouse Genetics Program reported a preliminary general histopathology workup of 50 randomly selected knockout lines yielding 2 with skin phenotypes, using single samples of dorsal skin from two males and two female mice [13]. This was complemented by a "special skin workup" as part of an extended primary phenotyping pipeline [14]. The histopathology protocol involved collecting one piece of dorsal skin, fixing, and processing it routinely to create hematoxylin and eosin stained slides. Nearly 800 mutant mouse skin samples were collected from 562 mutant lines in this high-throughput protocol. One and sometimes two slides with two or three serial sections of dorsal skin from one mouse in the study were provided. Mice were approximately 15 weeks of age. Only 23 lines (4.1%) had some sort of skin phenotype while all the others were normal, implying that many of the genes inactivated had no function in the skin; at least not evident from this screening protocol.
The International Mouse Phenotyping Consortium database currently (accessed 17 February 2017) shows that overall 2.22% of 3,199 lines show abnormalities of skin morphology and 5.35% of 2,971, abnormalities of the adnexa. This group reported no abnormalities of skin physiology in its database to date. A search of the Europhenome database, which overlaps with the International Mouse Phenotyping Consortium, shows 12 genes with annotations on skin morphology and physiology (excluding nociception; 27) of 563 genotypes derived from thesemutagenesis efforts (2.1%). The low number of skin phenotypes in these samples contrasts the frequent reporting of skin and adnexal phenotypes both in human Mendelian diseases and in the gathered data of MGI. Detailed examination of the MGI database shows that 1,069 of 9,045 targeted genes with phenotypes are annotated to either skin morphology or phenotypes; approximately 12%. In OMIM, 1,314/3,710 gene entries with phenotypic annotations are annotated to "abnormality of skin, HP:000095" or its children in the Human Phenotype Ontology database (35%) [15]. Assuming that we might expect skin and adnexal phenotypes in both humans and mice to result from loss of function of the same gene, the key question is where are the missing phenotypes? Even considering the exceptionally detailed phenotyping carried out in clinical practice, this is a large discrepancy. One possibility is that in the mouse, loss of function genes involved in the development or function of the skin and adnexa might be embryonic or neonatal lethal. However, this may also be an issue of detection; either through skewed phenotyping (i.e. not focused on the skin or not reported) or failure to detect phenotypes. There are projects associated with the US KnockOut Mouse Program that examine embryonic and perinatal lethality, yet none of them look beyond gross skin phenotype as reasons for early mortality [4]. From the reasoning above, we believe this problem can be addressed through careful targeted skin phenotyping.
The Jackson Laboratory is one of three sites in the United States generating knockout mutant mice. Histopathology is an optional parameter in the phenotyping pipeline, and thus is not performed and/or reported universally from individual centers. To better address finding skin lesions, representative retired breeder mice were received, histology was done on skin and nails from multiple sites, and slides were screened by a veterinary pathologist specializing in phenotyping skin mutants. Candidate models were subsequently reviewed by a physician pathologist specializing in dermatopathology. In so doing 169 novel mutant lines were reviewed and of those only one was found to have a relatively minor sebaceous gland abnormality.

Results
Skin, hair follicles, and nail units of both sexes were screened for lesions in 169 novel mutant lines of knockout mice generated at JAX (Table 1). Of these 116 (68.6%) were homozygous and 61 (36.1%) were heterozygous mutant mice. Four lines had both heterozygous and homozygous mice screened (2.4%). Of the heterozygous mice, 62.3% were pre-wean lethal and 11.5% were sub-viable, which is defined as yielding fewer than half of the expected number of homozygotes from an intercross (<12.5%). For these strains no homozygous mutant mice were evaluated. A total of 743 (2.2 average per line and sex) mice were necropsied and the skin and adnexa were evaluated histologically. Nine lines (two females and three males for heterozygotes only and three females and one male for homozygotes only) had only one sex evaluated.
Nonspecific, sporadic lesions were found in many of the mice in this study. Most commonly these were ruptured hair follicles with hair shaft fragments free in the dermis or hypodermal fat layer surrounded by a variety of inflammatory cell types, usually neutrophils around the hair shaft and a layer of macrophages (trichogranuloma, Fig 1). These were often solitary and not consistently found in all the mice within a mutant line, which was a primary criterion to differentiate these lesions from gene mutation associated phenotypes. Other mice had follicular dystrophy consistent with C57BL/6 alopecia and dermatitis, a low frequency, strain-specific skin disease resembling human central centrifugal cicatricial alopecia [16].
Mice with null mutations in their fatty acyl CoA reductase 2 (Far2 tm2b(KOMP)Wtsi /2J) gene were found to consistently develop focal alopecia as adults on the dorsal skin extending distally from the base of the ears (Fig 2). Histologically, in areas where the skin appeared to be relatively normal there were minor lesions affecting the pilosebaceous and Meibomian (eyelids) glands that could be easily over-looked. Other modified sebaceous glands (Zymbal, clitoral, or preputial glands) were not examined in this study. The sebaceous glands superficially appeared to be normal but within the sebaceous duct the sebocytes became brightly eosinophilic and, when these holocrine cells ruptured, their secretion remained present and clumped. By contrast, normal sebocytes were slightly basophilic with fine, uniformly sized vacuoles in their cytoplasm. When the normal cells ruptured there was no evidence of their cytoplasmic content within the follicular infundibulum or on the skin surface (Fig 3). In areas of alopecia, the epidermis was mildly acanthotic with orthokeratotic hyperkeratosis associated with areas of follicular dystrophy and rupture and an associated foreign body inflammatory response that would lead to follicular scarring, a form of cicatricial alopecia (Fig 4). Skin surface lipidomic and time course studies are in progress to better define these changes. Further, non-integument phenotypes affecting lipid metabolism, behavior, immunity, and the cardiovascular system have been reported for this strain, and can be found on the IMPC database website (http:// www.mousephenotype.org/data/genes/MGI:2687035#section-associations).
Heterozygous mutant mice for the protein phosphatase 1, regulatory subunit 9B (Ppp1r9b tm1.1(KOMP)Vlcg /J) gene had between one and five vibrissae hair shafts undergoing degenerative changes (follicular dystrophy) in each of three females and two males examined. Each section contained 10-15 vibrissae follicles (Fig 5). Only mystacial vibrissae were affected, not the periorbital or lower leg vibrissae. No other hair types were affected. This change was not found in two female and two male homozygous mutants suggesting that this was a nonspecific background lesion, as vibrissae follicular dystrophy was also seen occasionally in many mouse lines but only in one or two follicles and not consistently in all vibrissae in any section. The skin and nail sections from almost all of the mice in these studies were clinically and histologically normal.

Discussion
We have investigated the apparent deficiency of skin and adnexal phenotypes in a random set of strains from a systematic, genome-wide, knockout study using histopathological phenotyping. We report here that numbers detected confirm an earlier study [14] which also showed low numbers of detectable skin phenotypes, suggesting that ascertainment is not the cause of the "missing phenotypes".
Of over 100 homozygous mutants screened, only one line, Far2 tm2b(KOMP)Wtsi /2J, had lesions affecting sebaceous glands with what appeared to be a secondary follicular dystrophy, which probably progresses to a form of cicatricial alopecia as seen with a number of other mouse mutations with abnormalities of the sebaceous glands [17]. A rare autosomal recessive neurological disease without hair abnormalities was reported for humans with mutations in FAR1 [18]. While the biochemistry of these two proteins, FAR1 and FAR2, is known no disease is currently associated with mutations in FAR2 in humans (http://omim.org/entry/616156? search=Far2&highlight=far2). However, Far2 transcripts were reported in the eyelid and skin, presumably localized in the Meibomian and sebaceous glands, respectively [19]. Information obtained from phenotyping these mice will be useful to help assess sebaceous gland abnormalities in humans as not all genes that are expressed in sebaceous glands result in sebaceous gland abnormalities or clinically obvious hair problems, such as mouse null mutations in Soat1 [20,21]. Detailed studies are in progress to define that pathogenesis of the follicular dystrophy and to match it with a specific human disease in these Far2 mutant mice.
A second line, Ppp1r9b tm1.1(KOMP)Vlcg /J, had follicular dystrophy limited to many but not all vibrissae in heterozygous but not homozygous mutant mice suggesting this was a nonspecific background problem. A targeted mutant mouse line made for this gene (Ppp1r9b tm1Jfe ) was reported to only have neurological abnormalities [22,23]. As the vibrissae are an important somatosensory organ in the mouse [24] this structure should be carefully evaluated in such phenotyping studies.
The International Mouse Phenotyping Consortium has provided a unique, internally consistent dataset with genome-wide coverage and has yielded a great deal of valuable information on gene function. At the outset anticipation was high that systematic creation of null mice for all known protein coding genes would yield large amounts of information on the pathobiology of human disease. The high frequency of embryonic and neonatal lethal lines [4], consistent with previous estimates [6], was not a surprise, but potentially limits the value of these mice for Skin phenotypes in KOMP mutant mice Skin phenotypes in KOMP mutant mice detailed investigation of adult phenotypes and diseases which will require carefully designed conditional strategies. The relatively low frequency of skin phenotypes in the International Mouse Phenotyping Consortium mice found in this and an earlier study [11] was surprising given the frequency expected from global surveys of MGI and OMIM discussed above. The histopathological survey presented here was very detailed focusing on the skin, and we believe this supports the assertion that the "missing phenotypes" are not due to lack of ascertainment, at least in the random sample selected. The remaining possibilities are that genes involved in skin morphogenesis and physiology are overrepresented in those essential for embryogenesis, the C57BL6N background is failing to show phenotypes, or that the complete removal of gene function in the mouse knockouts have more severe functional consequences than the point mutations, regulatory lesions, and insertions/deletions (indels) that constitute most disease-causing mutations in humans. Protein truncating variants which provide the greatest likelihood of complete functional loss ("knockouts") are very rare in the human population [25].
In the current study, 74% of the heterozygous lines were pre-wean lethal or sub-viable as homozygotes (that were not examined) presenting the possibility that mutants having skin phenotypes might also be embryonic essential genes. While it is possible to argue that essential genes should never show up as disease genes in a population due to negative selection, there is increasing evidence that essential genes are often 'disease genes"-i.e. carry mutations that are associated with abnormal or disease phenotypes. This is by no means a settled conclusion however. The relationship between gene essentiality and disease association is discussed in detail in Cooper et al. [25].
Studies conducted around a decade ago, based on a limited gene set, suggested that mouse genes that are essential-i.e., genes related to more severe phenotypes in mouse, are more likely to be human disease genes, but that embryonic lethals were less likely [26]. However, a . However, the sebaceous glands were mildly hypoplastic with brightly eosinophilic mature sebocytes that did not immediately rupture as they entered the infundibulum (C). As in other mouse mutations that cause hypoplasia of sebaceous glands, the hair shafts in these mice did not exit the follicular ostium resulting in perforation of the root sheaths, release of the hair shaft into the dermis, and an inflammatory reaction (B). This healed by follicular scarring, or cicatricial alopecia. By contrast, aged and sex matched B6N(Cg) controls rarely exhibited ruptured follicles (D). Sebaceous glands varied in size with the hair cycle, which was normal, and sebocytes had light pink cytoplasm that becomes very pale as the cells matured and abruptly ruptured into the duct. (E). Hematoxylin and eosin stain, bar size indicated in each figure. https://doi.org/10.1371/journal.pone.0180682.g003 Skin phenotypes in KOMP mutant mice more recent study strongly suggested that about 30% of human disease genes (from OMIM) have a mouse ortholog with a lethal knockout phenotype (673/1965) [27]. The discrepancies here reflect to an extent the methodology used to calculate the association and the inclusion or exclusion of genes with no phenotypic data in the study by Park et al. [26]. This association will certainly stand further scrutiny now that more data are available, but does not explain why the IMPC/KOMP strains show fewer skin phenotypes than the bulk of the loss-of-function allele carrying strains in MGI. Core analysis carried out on Ingenuity™ using the lists of lethal essential genes from supplementary table 3 of Dickinson et al. [4] shows no significant enrichment of annotations to dermatological disease or skin development, mitigating against the possibility that genes associated with skin disease are overrepresented in the essential segment of the genome. A working conclusion, therefore, is that either there is a problem with the strain background in this resource showing full expressivity of mutations affecting the skin, or that we have so far insufficient coverage of the genome using high-resolution phenotyping to account for the "missing phenotypes" from the International Mouse Phenotyping Consortium program. Of concern is the observation that most human disease associated mutations are hypomorphic alleles which, in the event of lethal homozygosity of complete KOs, are much more likely to present a phenotype compatible with survival. The potential for missing these with a complete loss of function strategy is significant, especially when examining completely novel genes where there are no existing experimental or inferred annotations. The power of looking for skin phenotypes using a strategy such as that developed by Beutler et al. [28] using ENU mutagenesis is very persuasive. Currently Mutagenetix (https://mutagenetix. utsouthwestern.edu/home.cfm), accessed 1 Mar 2017) shows 49 genes with skin associated phenotypes recovered as phenotypic mutations from a total of 262 phenotyped genes (119 of 498 mapped mutations) and should be considered in the future along with full characterization of spontaneous and ENU generated mutations in large collections.

Generation of mutant mice
Embryonic stem cell lines. Initially mice included in the program were produced from embryonic stem cell lines acquired from repositories (https://www.komp.org/ or http://www. eummcr.org/). These resources comprised two major allele types, the Velocigene "definitive null" allele produced by the Regeneron Corporation (Vlcg) [1] and the "knockout first, conditional ready" design produced by the Wellcome Trust Sanger Centre and partners (Sanger) [2]. The Vlcg alleles delete the entire coding sequence of the target gene, including all introns, while inserting a LacZ reporter cassette. The neo resistance cassette is flanked by loxP sites for removal by cre-recombinase. The Sanger allele is a multifunctional design. In its native form, the strong splice acceptor typically generates a null allele. When bred to Flp-expressing mice, the entire 5' LacZ and neo cassette are removed leaving a loxP-flanked critical exon for conditional mutagenesis. When bred to cre-recombinase transgenic mice, this allele is converted into a "definitive null" which retains the 5' LacZ cassette and removes the critical exon and neo marker.
All animal work was approved by The Jackson Laboratory Animal Care and Use Committee (approval number 07005 "Mouse Models for Dermatology Research").
Generating live mice. Following importation at The Jackson Laboratory, clones are injected into blastocyst hosts and resulting chimeras are bred to C57BL/6NJ (STOCK#5304) or C57BL/6NJ-Tyr em3J /GrsrJ (STOCK#21999; to select for coat color transmission) inbred mice (see Table 1) to assure isogenicity of the resource. Germline transmitting (GLT) clones are quality confirmed (both Vlcg and Sanger) for targeting. Following GLT, mice are bred to congenic cre-recombinase transgenic mice (B6N.Cg-Tg(Sox2-cre)1Amc/J; STOCK#14094) to remove the neo cassette and to generate a definitive null allele, generating the tm1.1 allele (Vlcg and non-conditional Sanger clones) and tm1b allele (from conditional-ready Sanger clones). Four lines were generated by CRISPR/Cas9 mutagenesis, using standard approaches [29] to generate small deletions in the early part of the transcript. These mice are expanded, quality control-confirmed, and intercrossed to generate knockout mice. Approximately 30% of lines display lethality or sub-lethality (defined as yielding less than 12.5% homozygous mice) [4]. For these strains, heterozygous mice were phenotyped. For other strains, homozygous breeder pairs established for colony maintenance used in this study when they became retired breeders to maximize chances of finding adult onset diseases.

Systematic analysis of mutant mice
Animal logistics. Two female and two male retired breeder mice were provided from The Jackson Laboratory knockout mouse production program (Table 1). Homozygous lethal embryos and perinatal lethal mouse lines were specifically excluded from these studies. In these cases, heterozygous animals were aged and evaluated. All lines, whether Vlcg definitive null or Sanger conditional-ready mice were evaluated post-cre excision, which removed the neo cassette and heterologous promoter and in the case of Sanger alleles also removing the critical exon to create a definitive null allele (Vlcg tm1.1 and Sanger tm1b). All mice had skin, hair, and nails collected (see below).
Necropsy procedures, slide generation, and histopathologic analyses. As the skin is anatomically and molecularly distinct throughout the body in both mice and humans [30,31] and diseases can be restricted to specific anatomic sites, skin was collected from several areas. As dorsal and ventral trunk skin are histologically similar in the mouse, they were collected with other anatomically identifiable tissues and put into respective cassettes: 1) dorsal skin and ear skin; 2) ventral skin, eyelid skin, muzzle skin; 3) digits to include the foot pads and nail unit; and 4) cross and longitudinal sections of the tail. This allows for evaluation of all major hair follicle types, mucocutaneous junctions, the nail unit, as well as the footpad (including eccrine glands) [32]. Detailed protocols on skin removal, orientation, and processing are described elsewhere [33]. Tissues were fixed in Fekete's acid-alcohol-formalin solution. Sections were stained with hematoxylin and eosin (H&E) for routine screening.