High-throughput screening of mouse gene knockouts identifies established and novel skeletal phenotypes

Abstract

Screening gene function in vivo is a powerful approach to discover novel drug targets. We present high-throughput screening (HTS) data for 3 762 distinct global gene knockout (KO) mouse lines with viable adult homozygous mice generated using either gene-trap or homologous recombination technologies. Bone mass was determined from DEXA scans of male and female mice at 14 weeks of age and by microCT analyses of bones from male mice at 16 weeks of age. Wild-type (WT) cagemates/littermates were examined for each gene KO. Lethality was observed in an additional 850 KO lines. Since primary HTS are susceptible to false positive findings, additional cohorts of mice from KO lines with intriguing HTS bone data were examined. Aging, ovariectomy, histomorphometry and bone strength studies were performed and possible non-skeletal phenotypes were explored. Together, these screens identified multiple genes affecting bone mass: 23 previously reported genes (Calcr, Cebpb, Crtap, Dcstamp, Dkk1, Duoxa2, Enpp1, Fgf23, Kiss1/Kiss1r, Kl (Klotho), Lrp5, Mstn, Neo1, Npr2, Ostm1, Postn, Sfrp4, Slc30a5, Slc39a13, Sost, Sumf1, Src, Wnt10b), five novel genes extensively characterized (Cldn18, Fam20c, Lrrk1, Sgpl1, Wnt16), five novel genes with preliminary characterization (Agpat2, Rassf5, Slc10a7, Slc26a7, Slc30a10) and three novel undisclosed genes coding for potential osteoporosis drug targets.

Introduction

Following successful sequencing of the human and mouse genomes, a major goal of the genomics community has been to determine the functions of the ∼20 000 mammalian protein-coding genes by generating and examining phenotypes of knockout (KO) mice for each gene. Initial efforts towards this goal were pursued by Lexicon Pharmaceuticals (summarized in this report) and Deltagen.¹ More recently, academic centers joined forces to form the International Knockout Mouse Consortium (IKMC) encompassing efforts by the Knockout Mouse Project, European Conditional Mouse Mutagenesis Program, North American Conditional Mouse Mutagenesis Program and the Texas A&MInstitute of Genomic Medicine^2–6.

From 2000 through 2008, Lexicon Pharmaceuticals performed high-throughput mouse knockout and comprehensive phenotypic analyses (Genome5000™) for >4 650 genes. KO strategies (described below) involved both gene trapping,⁷ using the OmniBank® I embryonic stem (ES) cell library, and homologous recombination technologies. Lexicon generated a second ES cell library (OmniBank^® II) for the TIGM gene trap repository⁸ and since 2009, over 100 studies have used this resource to examine gene disruptions in mice (http://www.tigm.org/publications/). Phenotyping involved a battery of tests in the areas of behavior, cardiology, immunology, metabolism, oncology and ophthalmology, and included serum chemistry, histopathology and a high fat diet obesity challenge. Mice and ES cells from many KOs generated through this Genome 5000™ program are available through the USA NIH Mutant Mouse Regional Resource Center (https://www.mmrrc.org/catalog/overview_Major_Collection.php), Wellcome Trust (http://www.tigm.org/wellcome-trust/) and Taconic Farms (http://www.taconic.com/KO). The USA NIH Mutant Mouse Regional Resource Center collection includes 472 genes analyzed as part of Genentech’s Secreted Protein Discovery Initiative⁹ and having published phenotypes.¹⁰ Lexicon published body composition data for gene KOs (through 2005) that resulted in lean and obese phenotypes among the first 2 322 KO lines evaluated.¹¹ The present report describes the high-throughput screen employed to detect skeletal phenotypes and provides data on genes this screen identified that influence bone mass and architecture.

The International Mouse Phenotyping Consortium (IMPC, http://mousephenotype.org) was launched in September 2001 to coordinate phenotyping efforts of 15 worldwide groups involved in the IKMC.^12,13 Considerable effort and discussions have focused on optimizing these comprehensive high-throughput screens (HTS) to phenotype the mouse gene KOs.^6,14–23 Our goal is not to review current screening protocols, but to describe the successful HTS strategy employed by Lexicon for identifying skeletal phenotypes.

Lexicon’s protocol to identify potential skeletal phenotypes involved three complementary analyses: (i) whole body, femur and spine BMD by DEXA scans of anesthetized mice; (ii) microCT scans of dissected bones to determine bone architecture of the LV5 spine vertebral body and midshaft femur; and (iii) histological examinations of decalcified bones (long bones, sternums and nasal turbinates). Full experimental details are provided below.

The goal of HTS is to identify the few genes that influence phenotypes among the many knockouts examined. The conflict between thoroughly studying individual lines of KO mice and examining many different KO lines involves difficult compromises. The number of mice examined, the number of bones examined from each mouse, the skeletal measurements made and the criteria employed to identify potential skeletal phenotypes all must be simultaneously optimized. Undoubtedly, both false positives (genes falsely identified as having KO phenotypes) and false negatives (genes with true KO phenotypes that are missed) occur.

Without confirmation in additional cohorts of mice, phenotypes identified in HTS can represent false positive findings due to the necessity of examining multiple bone parameters in a small number of mice. Bone HTS results reported by three groups all limited their analyses to primary radiographic and DEXA screens. Deltagen¹ examined 750 gene KOs and found 9 (1%) with skeletal phenotypes. Genentech¹⁰ reported 150 of 476 (32%) KO lines showed bone phenotypes. The Wellcome Trust Sanger Institute reported bone phenotypes in 10 of 100 genes²⁴ and subsequently 9 body BMD and 33 radiographic phenotypes in 146 KOs (of 250 total) with viable mice.²⁵

Although the possibility of missing true phenotypes can never be completely eliminated, this risk of false negatives can be estimated by the ability of the HTS to identify previously established KO gene phenotypes. As described below, Lexicon’s skeletal HTS successfully confirmed published results of established genes including Fgf23, Klotho, Crtap, Ostm1 and Src and for these genes no further bone measurements beyond HTS were made. For Lrp5,²⁶ Sost,²⁷ Wnt10b,²⁸ Dkk1²⁹ and Sfrp4³⁰ bone phenotypes observed in the HTS were confirmed and extended in subsequent cohorts of mice. Many well-studied genes known to affect bone, such as Dmp1, Ctsk, Rank and Rankl, were not examined.

Lexicon’s motivation for undertaking this KO and phenotyping program involved identifying novel therapeutic targets.^31–33 This motivation dictated specifically targeting the roughly 5 000 genes in the druggable genome^34,35 and the selection of HTS assays measuring phenotypes providing guidance for treatable diseases. Thus, the genes analyzed were highly enriched in enzymes, receptors and secreted proteins, but transcription factors such as Runx2, Osterix and Msx2, and genes coding for structural proteins were omitted. Abnormalities in developmental processes, such as craniofacial³⁶ and digit anatomy, and longitudinal bone growth, result in serious human diseases, but are not easily amenable to treatment with small molecule drugs or neutralizing antibodies. For bone, our primary interest involved genes controlling bone formation by osteoblasts and bone resorption by osteoclasts. Identifying novel genes in these pathways can potentially lead to new osteoporosis therapies. Genes involved with diseases directly resulting from gene mutation such as osteogenesis imperfecta and chondrodysplasias were examined but not studied extensively, as they are less likely to yield targets for drug development.

Most importantly, given the goal to discover novel genes not previously known to regulate bone mass, Lexicon identified several novel genes, including Cldn18, Fam20c, Lrrk1, Sgpl1 and Wnt16. KO of Wnt16 reduces cortical bone mass and multiple human GWAS studies subsequently identified SNPs in the WNT16 gene region that affect cortical bone mass and strength.^37,38 KO of Lrrk1³⁹ produces severe osteopetrosis from osteoclast dysfunction, whereas KO of Cldn18 results in reduced bone mass from hyperactive osteoclasts.⁴⁰ The FAM20C protein was recently established as the kinase phosphorylating secreted proteins⁴¹ and Fam20c KO mice have hypophosphatemic rickets.⁴² Among other actions,⁴³ sphingosine-1-phosphate is a key mediator of osteoclast/osteoblast communication⁴⁴ and KO of Sgpl1⁴⁵ results in compromised immune function and osteopetrosis.

Materials and methods

Mouse production

Gene trapping offers a high-throughput approach for producing large numbers of insertional mutations in the mouse genome, while gene targeting by homologous recombination allows precise manipulation of genetic sequences in the mouse. Lexicon utilized both methods to generate KO mice for the Genome 5000™ project. KO mice were generated by homologous recombination using both a λ phage KOS shuttle system⁴⁶ and PCR-based targeting vector strategies as described.⁴⁷ Gene trapped lines were derived from the OmniBank® I library.⁴⁸ Details of each mutation are provided in Supplementary Table S1. To achieve effective gene disruption when using gene-trap mutations, intragenic insertions intersecting all known transcript units were selected after identifying the precise location of vector insertion using inverse genomic PCR. Oligonucleotide primers complementary to the gene-trap vector were used to amplify the vector insertion site for each clone, which was then compared to mouse genome sequence assemblies to localize the insertion with respect to the exons and introns of the gene. Gene disruption in vivo for gene-trap mutations was confirmed by a direct analysis of gene expression using RT-PCR. RNA was extracted from at least two tissues of wild-type (WT) and homozygous mutant mice using a bead homogenizer and RNAzol (Ambion, Austin, TX, USA) according to manufacturer’s instructions. Reverse transcription was performed with SuperScript II (Invitrogen, Carlsbad, CA, USA) and random hexamer primers, according to the manufacturer’s instructions. PCR amplification was performed with oligonucleotide primers complementary to exons flanking the insertion site.

Targeted or gene-trap mutations were generated in strain 129SvEv^Brd-derived ES cells. The chimeric mice were bred to C57BL/6J albino mice to generate F1 heterozygous animals. These progeny were intercrossed to generate F2 WT, heterozygous and homozygous mutant progeny. This generation was used for HTS phenotyping. On rare occasions, for example when very few F1 mice were obtained from the chimera, F1 heterozygous mice were crossed to 129SvEv^Brd/C57 hybrid mice to yield additional heterozygous animals for the intercross to generate the F2 mice. KO mice could not be generated for three genes, as heterozygous Pstk mice were infertile and, confirming published observations, KO of Cask⁴⁹ and Dll4⁵⁰ resulted in heterozygous lethality.

Lexicon’s KO strategies for Agpat2, Clcn7, Cldn18, Fam20c, Gnptab, Lrp5, Lrrk1, Sgpl1, Stk36, Tph1, Tph2 and Wnt16 are provided in the publications of these phenotypes. KO strategies used to generate 4 077 of Lexicon’s KO mouse lines are provided at the Taconic Farms website (http://www.taconic.com/KO). Supplementary Table S1 summarizes KO strategies for all 93 genes discussed in this review.

A total of 139 X-linked genes were KO'd, with bone data for 133 KOs reported. Male mice having an X chromosome gene KO'd are designated hemizygous. KO of two lines resulted in reduced viability and KO of four X-linked genes (Ebp, Mmgt1, Porcn, Prps1) resulted in hemizygous lethality. Data for 105 male hemizygous KO lines were analyzed. For 28 X-linked KOs DEXA data were analyzed for hemizygous and WT males plus homozygous female KO mice compared to age-matched WT female mice from the breeding colony. For 90 of these 105 KOs, data for only two male WT mice were available for DEXA calculations.

High-throughput screening assays

DEXA BMD was determined in 14-week-old mice anesthetized with 250 mg·kg⁻¹ tribromoethanol given intraperitoneally using a GE/Lunar PIXImus scanner. For each KO line, mean body spine and femur BMD ratios for KO/WT littermates were calculated separately for both male and female mice, and then these male and female data were averaged to yield a normalized BMD value for each site. For most KO lines, four male KO (actual mean=4.2), two male WT (mean=2.2), four female KO (mean=4.2) and two female WT (mean=2.1) mice were analyzed. Lines with fewer than four KO mice or fewer than three WT mice were excluded. When there was an uneven distribution of male and female mice, normalized BMD values were weighted to account for the actual number of mice analyzed. Body vBMD was calculated by dividing body BMD by the square root of body bone area. Body BMD and vBMD were correlated but vBMD had a lower variation. Spine BMD, femur BMD and body BMC/lean body mass (LBM) ratio were also determined. BMD values for left and right femurs were averaged for the HTS, but left femur BMD alone was employed for secondary screens.

These DEXA scans also provided body fat and LBM data for our obesity phenotyping program. The body composition measurements were validated against carcass composition determined by chemical analysis.⁵¹ A summary of the first 2 322 KO genes evaluated has been published¹¹, with identification of Ksr2 as a novel hyperphagic obesity gene. For most KO lines, architectural parameters of LV5s and femurs from four male KO (actual mean=4.1) and two male WT (mean=2.1) mice were determined at 16 weeks of age using a Scanco Medical µCT40 (Brüttisellen, Switzerland). Trabecular bone within the vertebral body was evaluated.

MicroCT X-ray voltage and current were 55 keV and 145 μA, respectively. Isotropic voxel dimensions for the HTS scans were 16 µm for LV5s and 20 µm for femurs, but higher resolution scans (8 µm voxel dimensions) were often employed for many of the secondary screen analyses. Vertebral body trabecular BV/TV, thickness and number were analyzed in LV5 scans and midshaft cortical thickness and total area (a surrogate for bone diameter) were analyzed in femur scans. Scans of 10 random LV5s with a range of trabecular bone mass showed the expected linear decrease in BV/TV values as the microCT threshold value was increased. After consultation with Scanco Medical, a threshold value of 240 was employed for all scans.

To generate high-throughput microCT scans, we engaged a machinist to construct Plexiglas inserts to hold multiple bones and fit snuggly inside the µCT40 sample holders. These inserts held 48 LV5s (12 rows of 4 bones per row) for overnight scanning, allowing four LV5s to be scanned simultaneously. Separate inserts held 18 femurs (three rows of six bones per row) allowing six femurs to be scanned simultaneously in 10 min. Photographs of these inserts are provided in Figure 1.

Figure 1

Photograph showing Plexiglas plastic inserts employed to align LV5 vertebral bodies (left) and femurs (right) inside sample holders for the Scanco µCT40 microCT scans. The red dots indicate placement of LV5s, with 12 rows of 4 columns. The lack femurs indicate placement of femurs, with three rows of six columns. Bones are fixed in place by adhering soft tissue and removable tape around the outside of the insert.

Histological examinations of bone were included as part of a comprehensive analysis of multiple tissues.²² The HTS analyses examined femur, tibia, sternum and nasal turbinates. Additional bones (vertebrae, calvarium and forelimbs) were examined in follow-up studies for selected KO mice having skeletal phenotypes in the HTS.

Advanced bone phenotyping assays

Deciding which KOs showing potential skeletal phenotypes in the primary HTS merited advancement to secondary screening involved both statistical and judgmental considerations. Given that multiple parameters were measured from a small numbers of mice, strict statistical tests were not employed. Knowing the means and standard deviations of each parameter, we sought consistency among the various DEXA and microCT values. Values for KO mice were compared to both littermate WT controls and historical WT data.

Additional methods, such as ovariectomy, daily subcutaneous teriparatide treatment, measurement of serum levels of PINP as an index of bone formation⁵² and bone breaking strength, all involved standard protocols. Biomechanical parameters were measured at Numira Biosciences (previously SkeleTech, MDS Pharma Services and Ricerca Biosciences, Salt Lake City, Utah, USA) using standard procedures for LV5 compression and femur shaft four-point bending. Body CT scans were performed using an ImTek scanner (Siemens, Munich, Germany). Three-dimensional images were reconstructed using the Feldkamp algorithm with ImTek 3D RECON software.

Lexicon developed two neutralizing mouse antibodies to Dickkopf 1 (DKK1). Mice were immunized with purified mouse DKK1 protein produced in HEK293 cells. Total RNA was obtained from the spleens of immunized mice and a phage library displaying FAb fragments was constructed. Phage displaying DKK1-specific FAbs were selected on immobilized DKK1 protein and monoclonal FAbs were generated in Escherichia coli. The specificity of FAbs for DKK1 was confirmed by ELISA. Chimeric proteins composed of combinations of the N-terminal leader/CYS1 domain and the C-terminal CYS2 domain/tail of DKK1, DKK2 and DKK4 were produced by transient transfection in HEK293 cells and used to map FAb epitopes. The ability of FAbs to inhibit the activity of DKK1 was determined using the CellSensor^® LEF/TCF-bla FreeStyle™ 293F reporter cell line (Invitrogen, Grand Island, New York, USA) in the presence of exogenous Wnt3a (R&D Systems, Minneapolis, Minnesota, USA) and exogenous DKK1 protein. The ability of FAbs to inhibit binding between DKK1 and LRP6 was determined in an ELISA-based binding assay utilizing purified DKK1 protein and purified LRP6 ectodomain-Fc fusion protein (R&D Systems). Based on their ability to inhibit DKK1 function in vitro and their binding mapping to distinct domains of DKK1, two FAbs were selected for conversion to full length mouse IgG1 antibody, production from CHO cells, and testing in vivo. Full-length antibody affinities for mouse and human DKK1 were measured using a Biacore 3000 (GE Healthcare, Pittsburgh, Pennsylvania, USA) with DKK1 in the solution phase.

Mouse husbandry

Mice were housed in micro-isolator cages within a barrier facility at 24 °C on a fixed 12-h light and 12-h dark cycle and were provided ad libitum acidified water and Purina rodent chow # 5001 (Purina, St Louis, MO, USA). Procedures involving animals were conducted in conformance with Lexicon Pharmaceuticals’ Institutional Animal Care and Use Committee guidelines, that are in compliance with state and federal laws and the standards outlined in the Guide for the Care and Use of Laboratory Animals (National Research Council, 2011). Quarterly sentinel surveillance showed no evidence of pathogenic rodent viruses, Mycoplasma or Helicobacter species in the Lexicon Pharmaceuticals source colonies.

Results

High-throughput screen data

Viable mice from 3 762 distinct KO genes were evaluated by various combinations of DEXA (N=3 651), microCT LV5 trabecular bone (N=3 399) and/or microCT midshaft femur cortical bone (N=3 366) scans. All three analyses were performed on 87% (N=3 255) of the KO lines. A Venn diagram showing the overlap of these assays is presented in Figure 2. Sixty-four percent of gene KOs were generated by homologous recombination and 36% by gene-trap technologies. These numbers include only KO lines for which viable mice survived through 14 weeks of age. Bone phenotypes were also identified by DEXA and microCT scans in heterozygous mutant Dkk1 mice and from two KO lines with reduced viability (Klotho and Sgpl1). Histologic examinations identified detrimental bone phenotypes in seven KO lines that did not survive until 12 weeks of age (Fgf23, Nppc, Npr2, Ostm1, Slc25a1, Slc30a10, Sumf1).

Figure 2

Venn diagram showing the numbers of KO lines undergoing HTS analyses for DEXA, LV5 microCT and femur shaft microCT.

Beyond these KO lines examined, additional KO lines were generated but are not included in this count. Of these, embryonic or neonatal lethality was observed in over 840 KO lines while the number of WT and/or KO mice generated for 41 KO lines was considered too low to provide adequate data for HTS bone measurements (see above) and were thus excluded from analyses. Three KO lines were generated prior to purchase of our DEXA scanners. Thus, mouse KOs were generated for more than 4 650 genes.

Figure 3 presents histograms of the gene KOs showing means, s.d.s and normal distribution values for DEXA vBMD, LV5 trabecular bone BV/TV and midshaft femur cortical thickness. For illustration, selected HTS values of mouse KOs having bone phenotypes are indicated. HTS data for all genes of interest are provided in the tables and text below.

Figure 3

Histograms showing distributions of vBMD (N=46 228 mice), LV5 Tb BV/TV (N=13 979 mice) and midshaft femur cortical thickness (N=13 845 mice) HTS values for all gene KOs examined, with s.d. indicated. Lines show normal distributions for the calculated means and s.d.s. HTS data for selected genes having clear bone phenotypes are indicted by arrows. s.d., standard deviation.

HTS data for Lexicon’s published KOs

The HTS (Table 1) clearly identified skeletal phenotypes for Cldn18, Fam20c, Lrp5, Lrrk1, Sgpl1, Sost and Wnt16 and comprehensive follow-up studies for each of these KO lines have been published.

Table 1 Primary screen data for seven gene KOs published by Lexicon

The 24 mammalian claudin proteins are components of membrane tight junctions regulating permeability between cells.⁵³ Although Cldn18 is highly expressed in osteoblasts,^40,54 bone cells are believed to contain gap junctions but not tight junctions.⁵⁵ Consistent with high expression of Cldn18 in gastric cells and in agreement with a previous report,⁵⁶ Lexicon’s Cldn18 KO mice had severe gastritis.⁴⁰ Osteoclast activity was increased with KO of Cldn18, resulting in low bone mass.⁴⁰ Cldn18 KO mice fed a high calcium diet remained osteopenic, demonstrating an intrinsic osteoclast defect⁵⁷ and not a secondary osteoclastic response to a possible low intestinal calcium absorption from reduced gastric acid secretion.

Fam20c KO mice grew poorly after weaning (LBM is 53% of normal), having dental lesions, hypophosphatemic rickets and greatly elevated serum levels of FGF23.⁴² Identical phenotypes were observed in an independent mouse KO.⁵⁸ Human mutations result in Raine syndrome with neonatal lethality the usual outcome.⁵⁹ During the time these mouse KOs were being characterized two independent groups demonstrated that FAM20C is a Golgi kinase that phosphorylates proteins undergoing secretion.^60,61 FAM20C has recently been shown to phosphorylate FGF23 at serine 180, inhibiting O-glycosylation of threonine-178 and thereby promoting proteolysis to inactive N- and C-terminal fragments.⁶² FAM20C also phosphorylates casein, BMP15 and SIBLING bone matrix proteins.

As established by other groups, KO of Lrp5^63–65 and Sost^66,67 result in low and high bone mass (HBM), respectively. In agreement with an independent Lrp5 KO mouse study⁶⁸ and a recent human trial examining a patient with a LRP5 mutation,⁶⁹ Lexicon showed that Lrp5 KO mice respond to the anabolic actions of teriparatide.²⁶ Lexicon demonstrated that male Sost KO mice continue to gain cortical bone through 2 years of age, with male and female KO mice losing bone following castration.²⁷ These Sost KO mouse data will be presented separately. Lexicon also contributed to a multi-institution study showing that Lrp5 acts locally in osteoblasts to influence bone mass,⁷⁰ arguing against systemic actions of gut-derived serotonin on bone mass.⁷¹ Lrp5 KO mice generated independently at Lexicon and the Max Delbrück Center for Molecular Medicine (Germany) did not show any disturbances in serotonin metabolism.⁷⁰

Patients having loss-of-function mutations in LRP5 and Lrp5 KO mice have eye vascularization defects.⁶³ KO of any of four distinct genes in mice leads to failure of retinal hyloid vessels to regress during development: the secreted proteins Ndp^72,73 and Wnt7b,⁷⁴ and the coreceptors Frd4⁷² and Lrp5.^64,75 Not described in our Lrp5 publication²⁶ is the fact that Lexicon’s KO mice have these ocular defects. Histologic examination (Supplementary Figure S1a) showed that photoreceptors, inner retinal cells and ganglion cells were distributed in three cellular layers, the outer nuclear layer, inner nuclear layer and ganglion cell layer, respectively, in WT retinas. Synaptic connections among these cells occur in the outer plexiform layer and inner plexiform layers. In Lrp5 KO mice, the vitreous, which is clear in healthy retinas, contained protein exudates (asterisks) and the inner nuclear layer was disrupted by large infiltrating blood vessels (arrowheads) that apparently failed to disperse, but formed a deep capillary plexus. Lexicon has also confirmed and extended findings of similar ocular vascularization defects in Fzd4 KO mice.⁷⁶

LRRK1 is a serine/threonine kinase containing three ankyrin repeat domains, seven leucine-rich repeat domains, a Roc GTPase domain, a COR domain and a kinase domain.⁷⁷ Mutations in the related LRRK2 gene are the most common genetic cause of Parkinson’s disease and since these mutations often result in enhanced kinase activity, drug development efforts are underway to discover small molecule inhibitors of this target. Lexicon’s Lrrk1 KO mice (which lack the GTPase domain as a result of targeted deletion of exons 16–19) have severe osteopetrosis resulting from dysfunctional osteoclasts.³⁹ A second Lrrk1 KO line (C57BL/6-Lrrk1^tm[1.]^1Mjff/J, lacking exons 24–29 which encode the kinase domain) and Lrrk1 KO rats (LEH-Lrrk1^em1sage[−/− ]) are available from the Michael J. Fox Foundation⁷⁸ through Jackson Laboratories (Bar Harbor, Maine, USA) and SAGE Labs (Boyertown, Pennsylvania, USA), respectively. C57BL/6-Lrrk1^tm1.1Mjff/J KO mice suffer from neonatal lethality and reasons for phenotype differences resulting from the two different KO strategies are unclear.

Sphingosine-1-phosphate (S1P), a signaling phospholipid produced from ceramides by the actions of ceramidase and sphingosine kinase, is a ligand for five GPCR receptors (S1P1 through S1P5) and is inactivated by S1P lyase (SGPL1). Among other actions, S1P inhibits migration of T lymphocytes from thymus and lymph nodes into the circulation and therefore pharmacological analogues of S1P (fingolimod) or treatments that increase endogenous S1P levels are immunosuppressive. Plasma S1P levels are elevated in postmenopausal women with vertebral fractures.⁷⁹ Several laboratories, including Lexicon,⁴⁵ have shown that global KO of Sgpl1 results in sickly mice having lymphopenia and markedly reduced lifespan. Lexicon’s comprehensive histological examination showed lung, heart and urinary tract lesions, along with osteopetrosis. Osteoclasts are plentiful but have increased cytoplasmic volume and show increased incidence of degeneration and apoptosis. Thickened trabecular bones suggest that osteoclasts are not fully functional. Complementing the histological bone sections demonstrating osteopetrosis,⁴⁵Figure 4 shows microCT images with greatly increased femoral trabecular bone content in KO mice. In addition to influencing osteoclast function, S1P is also produced by osteoclasts and stimulates bone formation by osteoblasts.^44,80 Complete genetic rescue of the non-lymphoid lesions, including osteopetrosis, was achieved in transgenic mice having mouse Sgpl1 replaced by human SGPL1. These humanized mice have spleen SGPL1 enzyme activity 17% of normal and are healthy, but continue to exhibit lymphoid defects.⁴⁵ Pharmacological inhibition SGPL1 in mice⁸¹ and humans⁸² increased circulating S1P levels and decreased circulating T-lymphocytes.

Figure 4

MicroCT scans of femurs from WT (left) and Sgpl1 KO (right) mice.

Lexicon’s Wnt16 KO mice are healthy with normal bone length and trabecular bone mass, but reduced cortical bone diameter and thickness^37,38 and fail to add new periosteal bone after mechanical loading.⁸³ Additional data for Lexicon’s Wnt16 KO mice are provided in Supplemental Table S2, as part of the Wnt16/Sfrp4 double KO (DKO) study described below. These Lexicon Wnt16 KO mice were on a mixed (hybrid) C57BL/6J–129SvEv genetic background but another group found that Wnt16 KO mice on a C57BL/6J genetic background developed spontaneous fractures associated with reduced thickness but normal diameter cortical bone.⁸⁴ Thus, a cortical bone deficit occurs in both mouse lines with mechanistic differences in the development of skeletal phenotypes. Periosteal bone formation is reduced in hybrid mice but normal in C57BL/6J mice, with endocortical bone resorption near normal in hybrid mice but increased in C57BL/6J mice.

KOs with HTS data only—confirming published KOs

Lexicon’s HTS confirmed previously published skeletal phenotypes of mouse KOs for numerous genes. These successful confirmations demonstrate the power of the HTS measurements and provide confidence that there were minimal false negatives when screening for novel genes affecting bone mass. Table 2 presents HTS data for Cebpb, Crtap, Enpp1, Kremen1, Kremen2, Klotho Slc39a13 and Src. Since these established skeletal phenotypes are unambiguous, no bone analyses were performed on additional cohorts of mice.

Table 2 Primary screen data Cebpb, Crtap, Enpp1, Klotho, Kremen1, Kremen2, Neo1, Pstn, Slc39a13 and Src

Four groups have shown that the transcription factor Cebpb (CCAAT/enhancer binding protein beta) KO mice are small with reduced bone mass^85–88 and these observations were confirmed in Lexicon’s KO mice as LBM was 82% of normal and BMD values were reduced.

Cartilage-associated protein is an enzyme that hydroxylates proline at carbon-3 in collagen and mutations cause osteogenesis imperfecta with connective tissue defects in humans and mice.^89,90 Crtap KO mice had reduced viability with low BMD and microCT parameters. Nasal turbinates (Supplementary Figure S1b) were fragile during handling and showed increased basophilia, diffuse thinning and numerous discontinuities in the bone plates.

Dual oxidation maturation factor 2 (DUOXA2) is a transmembrane protein required for plasma membrane localization of the hydrogen peroxide generating enzyme DUOX2. Hypothyroidism results from mutations in DUOX2 and DUOXA2 in humans⁹¹ and KO of Duox2⁹² or DKO of Duoxa1/Duoxa2⁹³ in mice. As expected, Duox2 KO mice are dwarfs with low femur BMD.⁹⁴ Lexicon’s Duoxa2 KO mice showed clear signs of hypothyroidism, including hypertrophic TSH-secreting cells in the pituitary, goitrous thyroid, growth retardation, bone epiphyseal flattening, reduced medullary area, altered articular surfaces and increased numbers of brown fat adipocytes containing large fat droplets. HTS DEXA scans showed reduced LBM (57% of normal) and femur BMD (85% of normal), but due to small body size, body vBMD and spine BMD were normal. As expected, body BMC (69% of normal) and BMD (90% of normal) were greatly reduced. In agreement with published data,⁹⁵ KO of Duox1 did not result in hypothyroidism (data not shown).

Mice with the ectonucleotide pyrophosphatase/phosphodiesterase family member 1 (Enpp1) gene KOed⁹⁶ or disrupted by spontaneous mutation⁹⁷ or ENU mutagenesis^98,99 have ectopic calcifications. Lexicon’s Enpp1 KO mice had low bone mass and modest hypophosphatemia (5.2±0.2 versus 7.2±0.3 mg·dL⁻¹, P=0.004) with ankylosing intervertebral and peripheral joint hyperostosis, multifocal calcification of arteries, ligaments, tendons and articular cartilage and the splenic capsule.

KLOTHO is a FGFR coreceptor that binds FGF23.^100,101 Confirming published data, Lexicon’s Klotho KO mice had poor growth and few survived past 10 weeks of age. The bone phenotype was unusual as high trabecular bone mass coexisted with low cortical bone mass¹⁰² and these defects were rescued by simultaneous deletion of PTH.¹⁰³ These skeletal phenotypes resulted in elevated values for body vBMD, spine BMD and spine trabecular BV/TV, but reduced values for femur BMD and cortical thickness. A histological section of a femur from a KO mouse is shown in Supplementary Figure S1c.

KREMEN1 and KREMEN2 are single transmembrane proteins that bind to DKK1 and LRP6, but the exact mechanism of KREMEN action is unclear. We did not observe any obvious skeletal phenotypes in our HTS screens for Kremen1 and Kremen2 KO mice and this finding was independently confirmed in mice examined at 12 weeks of age.¹⁰⁴ Interestingly, double KO mice with disruptions of both Kremen1 and Kremen2 genes are viable and have high trabecular bone mass resulting from elevated bone formation.¹⁰⁵ However, this same group subsequently showed that Kremen2 has high bone expression, osteoblast-specific overexpression of Kremen2 produces low bone mass and that Kremen2 KO mice have elevated trabecular bone mass at 24 weeks of age.¹⁰⁵

Neogenin is a cell surface protein modifying BMP receptor association with membrane lipid rafts and Neo1 KO mice have abnormal chondrocyte maturation and endochondral bone growth.¹⁰⁶ Lexicon’s KO mice had low bone mass parameters in the HTS screen and growth plate hypoplasia histologically (Supplementary Figure S1d).

Periostin is a secreted matricellular protein highly expressed in the periosteum.^107,108 Postn KO mice have low bone mass,^109,110 a minimal bone formation response to exercise and axial compression and increased bone fatigue damage in response to loading.¹¹¹ Reduced bone mass was observed in Lexicon’s HTS screen.

SLC39A13 is a zinc transporter. Gene disruption is associated with multiple abnormalities, including low bone mass, in KO mice,¹¹² and the spondylocheiro dysplastic form of Ehlers–Danlos syndrome in humans.¹¹³ Lexicon’s HTS confirmed growth retardation (LBM is 73% of normal), a generalized low bone mass phenotype and dysplasia of the epiphyseal growth plate cartilage. The cartilage was thickened (2–3 times normal) and disorganized with increased interchondrocyte matrix material of a loose fibrillar appearance. The cytoplasm of mature chondrocytes was distended by large eosinophilic granules. KO mice also exhibited modest thickening of the articular cartilage (Supplementary Figure S1e).

Src was first identified as an osteopetrosis gene in 1991.¹²⁶ Lexicon’s KO mice were small (LBM was 81% of normal) with high values for BMD and trabecular bone by microCT. Cortical bone thickness was low. Histologic examination showed osteopetrosis and dysplasia of incisor teeth. Bones were characterized by medullary cavities (Supplementary Figure S1h) filled with trabecular bone and thinning of the cortical bone. Nasal turbinate bones were distorted and up to five times thicker than normal.

The DEXA, microCT and histological assays used in HTS at Lexicon accurately confirmed these nine established skeletal KO phenotypes (and normal bone mass in Kremen1 and Kremen2 KO mice). In addition, three KO lines having severe bone phenotypes (Fgf23, Ostm1 and Nppc) but not surviving to undergo DEXA and microCT scans were confirmed histologically.

FGF23 is secreted by osteocytes to stimulate renal phosphate excretion and calcitriol synthesis. Fgf23 KO mice have a maximum lifespan of 13 weeks, are hyperphosphatemic and have multiple skeletal defects.¹¹⁴ Lexicon’s KO mice were small and sickly and euthanized at 6 weeks of age without serum P determination. Histologically, defective mineralization of bone was observed, characterized by retention of cartilage cores in cortical and trabecular bone in long bones and thickening of turbinate and calvarial bones, as well as apposition of woven bone on the endosteal surfaces of the diaphysis. Although osteoblasts in regions with normally high bone turnover had a normal appearance, the overwhelming majority of osteoblasts in other areas were abnormal. The abnormal osteoblasts were up to fivefold normal size and frequently filled the intertrabecular marrow spaces. The cytoplasm of affected osteoblasts and osteocytes was distended by abundant basophilic material, presumably matrix proteins. Mineralization was consistently present in the kidney and duodenum and regions in the aorta, trachea and stomach. Ectopic mineralization in the kidney was relatively mild and limited in distribution to proximal tubules located near arcuate vessels.

Mutations in OSTM1 are responsible for osteopetrosis in humans and the spontaneous grey lethal osteopetrotic mouse.^115,116 This gene codes for a transmembrane protein that forms the beta subunit of the CLNC7 chloride transporter, mutations in which also lead to osteopetrosis.¹¹⁷ Lexicon’s Ostm1 KO mice were small and failed to thrive with a grey coat color and no teeth. Microscopic analysis revealed retinal degeneration, reactive astrogliosis (hypertrophy with inclusion bodies), neuronal necrosis and osteopetrosis. The medullary cavities of all long bones, vertebrae and sternebra were filled with trabecular bone (Supplementary Figure S1f). Osteoclast numbers were increased and in some areas there were degenerating and necrotic osteoclasts. Osteoblasts were also numerous, although they tended to be elongated and fibroblastic. Bones in the skull, nasal trabecula and epiphyses of long bones contained abundant loosely woven bone and trabecula.

C-type natriuretic peptide (CNP, gene symbol NPPC) stimulates longitudinal bone growth though natriuretic peptide receptor B (NPR2). KO of mouse Nppc¹¹⁸ and three spontaneous mouse (cn, slw and lbab) mutations^119–121 all result in dwarfism. The involvement of NPR2 in growth plate development is shown by mutations in humans and mice,¹²² with a gain-of-function mutation in humans leading to tall stature.¹²³ Most Npr2 KO mice at Lexicon died young and the few survivors were small and sickly. Achondroplastic dwarfism, resulting from laminar hypoplasia of the epiphyseal cartilage, was confirmed by histological examination. Limb bones had nearly normal diameters but were markedly shortened, especially in comparison to the craniofacial bones formed by intramembranous ossification. Zones of resting and proliferating cartilage had nearly normal thickness but there was marked thinning of the zones of hypertrophic and degenerating chondrocytes. The zones of calcification that are normally composed of cartilage cores lined by hypertrophic osteoblasts were almost entirely absent (Supplementary Figure S1g). There was severe dental malocclusion due to unequal growth rates of the calvarium and mandible.

Osteocrin is a secreted protein with homology to A- and B-type natriuretic peptides. Osteocrin binds to natriuretic peptide receptor C (NPR3), a clearance receptor for natriuretic peptides, and thereby increases local concentrations of these natriuretic peptides. An ENU-induced point mutation in the mouse Npr3 extracellular domain, presumably inhibiting CNP clearance, results in elevated longitudinal bone growth.¹²⁴ Mice with transgenic overexpression of osteocrin in osteoblasts, expected to inhibit CNP clearance, have normal bone mass but elongated tails and femurs¹²⁵. Lexicon’s Ostn KO mice had normal bone mass in the HTS (data not shown). Femoral and tibial lengths in both male and female mice determined from DEXA scans were unaltered (P>0.6, data not shown). Thus, CNP signaling in the bone growth plate appears to be unaffected by disruption of osteocrin, but stimulated by both elevated osteocrin levels and disruption of the NPR3 clearance receptor.

Genes involved in WNT signaling

Great interest in the roles of Wnt signaling on bone mass followed the 2001/2002 discoveries that various LRP5 and SOST gene mutations in humans are responsible for dramatic alterations in patients with osteoporosis pseudoglioma, the HBM phenotype, sclerosteosis and van Buchem’s disease. Along with Frizzled receptors, LRP5 and LRP6 are coreceptors for the 19 WNT ligands with sclerostin (coded by the SOST gene) blocking WNT binding to LRP5/6. Since inactivating LRP5 (osteoporosis pseudoglioma syndrome) and LRP6^127,128 mutations decrease bone mass, whereas activating LRP5 mutations (HBM phenotype) and mutations disrupting SOST expression (sclerosteosis and van Buchem’s disease) increase bone mass, WNT signaling must activate bone formation. As described above, KO of Lrp5 and Sost in mice produce the identical skeletal phenotypes as the human inactivating mutations. Furthermore, transgenic mice with Lrp5 activating mutations have extremely HBM.^70,129

WNT signaling can also be inhibited by DKKs that bind LRP5/6 and Secreted frizzed-related proteins (SFRPs) that bind WNTs. As is the case with sclerostin, reducing levels of DKKs and SFRPs in bone should remove inhibitors of WNT signaling, thereby presumably increasing bone mass. Support for this idea was available in the original HBM publication describing the LRP5 activating mutation, which in cells eliminated the inhibitory actions of DKK1.¹³⁰ The key role of WNT signaling in bone prompted us to examine KO mice for all four Dkks, five Sfrps and Wif1 (Table 3). KO of Dkk1 leads to embryonic lethality, but mice with disruptions of the other eight genes were viable.

Table 3 Primary and secondary screen data for WNT antagonists

HBM is observed in doubleridege mice having a spontaneous mutation leading to reduced DKK1 expression,¹³¹ Dkk1 heterozygous mutant mice¹³² and Dkk1 homozygous KO mice generated on a Wnt3 heterozygous mutant background.^133,134 Osteoblast-specific overexpression of Dkk1 in mice leads to severe osteopenia.¹³⁵ Lexicon confirmed observations of HBM in heterozygous Dkk1 KO mice. These and other observations prompted several groups, including Novartis,¹³⁶ Merck,^137,138 Amgen,¹³⁹ Eli Lilly¹⁴⁰ and Lexicon, to develop anti-DKK1 neutralizing antibodies for potential therapeutic treatment of osteoporosis and other bone diseases, particularly multiple myeloma.¹⁴¹

Lexicon’s DKK1 Antibody #1 recognizes an epitope in the C-terminal half of DKK1 that prevents DKK1 binding to LRP6. Antibody #2 recognizes an epitope in the N-terminal half of Dkk-1 that does not affect Dkk1 binding to LRP6. Identical anti-DKK1 epitope specificities for LRP6 binding have been reported by Amgen.¹⁴² Both Lexicon antibodies have potencies below 500 pM as determined by binding affinities to human DKK1 (416 pM for Ab #1 and 347 pM for Ab #2) and mouse Dkk1 (128 pM for Ab #1 and 233 pM for Ab #2) and IC50 values (270 pM for Ab #1 and 370 pM for A b#2) in a mouse cell-based assay employing Wnt3a as a stimulator of Wnt signaling.

Three preliminary studies examining these two antibodies in adult male mice (>12 weeks of age) at weekly doses of 3–30 mg·kg⁻¹ for 6 or 12 weeks all provided evidence of bone efficacy (data not shown). Since DKK1 expression in rat bone declines with age,¹⁴³ we performed a fourth study of 4 weeks duration in male mice starting at 4 weeks of age. Antibody #1 and antibody #2 were each given subcutaneously weekly at doses of 30 mg·kg⁻¹. A negative control group of mice was treated similarly with a control antibody. A positive control group of mice was treated at the start of the study with a single subcutaneous dose of 50 µg·kg⁻¹ zoledronate to inhibit bone resorption.

Treatment with antibody #2 and zoledronate increased spine and femur BMD, and cancellous BV/TV in the LV5 vertebral body and distal femur metaphysis (Figure 5). Focusing on the LV5 vertebral body, antibody #2 treatment increased trabecular thickness but not trabecular number, whereas zoledronate treatment increased trabecular number but not trabecular thickness. This differential mechanism of cancellous bone gain is consistent with proposals that DKK1 inhibition acts to increase bone formation but zoledronate treatment acts to block bone resorption. Treatment with antibody #1 consistently showed trends to increase bone mass but this antibody was clearly inferior to antibody #2.

Figure 5

DKK1 antibody treatment in young mice. Data are means±s.e.m. for 10 mice per group treated between 4 and 8 weeks of age with antibody doses of 30 mg·kg⁻¹ weekly or a single 10 µg·kg⁻¹ zoledronate dose. Statistical evaluation involved ANOVA followed by Dunnett’s test for multiple comparisons.

In agreement with previous reports,^144,145 Lexicon’s Dkk2 KO mice had numerous ocular abnormalities, including a lack of Harderian glands and malformed eyelids. The cornea and sclera contained sebaceous glands and hair follicles with multifocal keratitis and ulceration and fibrosis of the collageneous stroma. Although Dkk2 KO mice have a normal skeletal architecture, their bone contains an excess of unmineralized osteoid.¹⁴⁶ Lexicon’s KO mice had normal, or even slightly elevated, BMD. Undecalcified bone sections were not examined and therefore the excess osteoid phenotype was not evaluated. We rescanned eight WT and eight KO midshaft femurs at the highest microCT resolution possible (6 μm voxel size) and did not observe any hypomineralization of cortical bone, as material BMD was unaffected (0.7% higher with P=0.27) in bones from Dkk2 KO mice. No gross skeletal phenotypes were previously detected in Dkk3 KO mice¹⁴⁷ and Lexicon did not observe any bone abnormalities in the HTS.

Several groups generated viable Sfrp2, Sfrp5^148,149 KO or ENU-mutant Sfrp5¹⁵⁰ mice without describing any bone abnormalities. Interestingly, DKO of Sfrp1/Sfrp2 results in embryonic lethality,¹⁵¹ whereas Sfrp2/Sfrp5 DKO mice are viable.¹⁵⁰ Lexicon did not detect any skeletal phenotypes in Sfrp2 or Sfrp5 KO mice. Sfrp2 KO mice have subtle limb deformities including brachydactyly and syndactyly¹⁵² and these digit abnormalities were likely missed in Lexicon’s HTS.

Published reports describe skeletal phenotypes in Sfrp1 and Sfrp3 KO mice. Global overexpression of Sfrp1 in transgenic mice results in reduced bone mass.¹⁵³ Sfrp1 KO mice have elevated trabecular bone mass after 13 weeks of age,^154,155 but Lexicon’s primary and secondary screens did not confirm this observation. Interestingly, the published Sfrp1 KO (targeting exon 1) mice examined were generated at Lexicon prior to 2000. These mice are distinct from the KO mouse subsequently studied at Lexicon. The reason for this discrepancy is unclear and additional studies are required. DEXA HTS analyses performed at the Harwell MRC site as part of the Europhenome mouse KO project observed reduced BMD in Sfrp1 KO mice. At least three additional Sfrp1 KO mice lines have been generated^151,156,157 without skeletal analyses performed.

Frzb (Sfrp3) KO mice have normal trabecular bone mass but a 7% increase in cortical bone thickness and increased articular cartilage loss during arthritis triggered by instability, enzymatic injury or inflammation.¹⁵⁸ KO mice also showed an exaggerated cortical bone gain in response to repetitive loading of the ulna, with the added bone being predominantly periosteal.¹⁵⁷ Loss of bone following ovariectomy was not affected by Sfrp3 KO.¹⁵⁷ GWAS analysis identified SFRP3 SNPs that influence hip osteoarthritis and geometry.¹⁵⁹ Lexicon’s HTS and secondary screens did not show an obvious bone phenotype. Further analysis of the secondary screen microCT data (excluding one KO outlier) found a 6% increase in midshaft femur cortical thickness (P=0.07, with N=8 for both WT and KO mice). No challenge studies, such as mechanical loading, were performed.

Including Lexicon’s data, three independent laboratories have presented data in abstracts showing bones from Sfrp4 KO mice have increased trabecular bone mass, elevated cortical bone diameter and reduced cortical bone thickness.^30,160,161 High trabecular bone mass results in increased compressive breaking strength of the LV5 vertebral body, whereas reduced cortical bone thickness leads to reduced four-point breaking strength of the femur shaft.³⁰ Differential effects of disrupting Sfrp4 function in cortical and trabecular bone likely reflect distinct signaling pathways as canonical WNT signaling is involved in trabecular bone formation but non-canonical Wnt signaling is involved in cortical bone formation.¹⁶² The Yale/Harvard and Lexicon data are being prepared for a joint publication. Here we show our HTS and secondary screen data characterizing these skeletal phenotypes and data from a Sfrp4/Wnt16 DKO study.

As described above, Wnt16 KO mice have normal trabecular bone mass but reduced cortical bone diameter and thickness. We examined the effects of disruption of WNT16 signaling on the skeletal phenotypes of male and female Sfrp4 KO mice by microCT at 16 weeks of age. In the absence of an effect of Wnt16 KO on trabecular bone, the elevated trabecular bone mass in Sfrp4 KO mice should not be influenced by disruption of Wnt16. However, the enlarged cortical bone diameter in Sfrp4 KO mice might involve WNT16 signaling. Since cortical bone thickness is reduced in both Sfrp4 and Wnt16 KO mice, simultaneous disruption of both genes might result in greatly reduced thickness and even spontaneous fractures.

As shown in Supplementary Table S2, skeletal phenotypes in both male and female Sfrp4 and Wnt16 single KO mice are exactly as determined previously. LV5 trabecular bone is elevated in Sfrp4 KO mice but normal in Wnt16 KO mice. Cortical bone total area (diameter) is elevated in Sfrp4 KO mice and reduced in Wnt16 KO mice. Cortical thickness is reduced in both KOs, but to a greater extent with Sfrp4 than Wnt16 disruption. KO of Wnt16 did not influence the elevated LV5 trabecular bone mass in Sfrp4 KO mice. Bones from DKO mice had intermediate cortical total area between elevated Sfrp4 KO and reduced Wnt16 KO values (interaction P values >0.50), indicating these two genes have independent actions on bone diameter. Bone areas were reduced similarly in both KOs and not further influenced by DKO. Marrow areas were greatly elevated with KO of Sfrp4, slightly reduced with KO of Wnt16 and elevated in DKOs. Therefore, inactivation of Wnt16 does not prevent the stimulation of endocortical bone resorption responsible for the enlarged marrow cavity in Sfrp4 KO mice. Bones from DKO mice had lower cortical thickness than the already reduced values in single KOs but no spontaneous fractures were observed.

Like SFRPs, WNT inhibitory factor 1 antagonizes Wnt signaling by binding to WNTs. Wif1 KO mice have a normal skeleton but are sensitive to radiation-induced osteosarcomas¹⁶³ and mice with osteoblast-specific Wif1 overexpression display no overt bone phenotype.¹⁶⁴ Lexicon’s Wif1 KO mice had no clear bone phenotypes in HTS or secondary screens.

WNT10A suppresses adipogenesis and stimulates osteoblastogenesis in ST2 and 3T3-L1 cells.¹⁶⁵ Lexicon’s HTS did not identify a bone phenotype in Wnt10a KO mice, but follow-up studies showed a 5% decrease in body length in both male and female KO mice at 33 weeks of age (data not shown). WNT10B is expressed in primary cultures of mature mouse calvarial osteoblasts¹⁶⁶ and activates canonical WNT signaling in the immune system, mammary gland, adipose tissue, bone and skin.¹⁶⁷ Wnt10b KO mice develop low bone mass with age.¹⁶⁸ Transgenic mice with high Wnt10b expression driven by the FABP4 promoter in adipocytes and the osteocalcin promoter in osteoblasts have HBM.^169,170 The anabolic effect of teriparatide treatment involves WNT10B stimulation of T lymphocytes.¹⁷¹ Lexicon’s Wnt10b KO mice showed modestly low bone mass in the HTS and follow-up studies. At 32 weeks of age (Supplementary Table S3), vBMD was reduced 8% in both male and female mice. MicroCT analyses indicated reduced LV5 trabecular number but normal cortical bone total area (diameter) with minimal decreases in cortical thickness.²⁸

KOs with robust bone phenotypes

Table 4 provides HTS and secondary screen data for seven KOs having clear bone phenotypes.

Table 4 Primary and secondary screen data for KOs with robust bone phenotypes

The seven-pass transmembrane protein DCSTAMP is required for fusion of mouse osteoclast precursors,^172–174 and this observation was confirmed in Lexicon’s Dcstamp KO mice. All osteoclasts are mononuclear, indicating a complete block in osteoclast cell fusion. There appeared to be increased numbers of mononuclear osteoclasts and bone architecture appeared normal by histology (Supplementary Figure S1i). Surprisingly, bone mass was only minimally elevated in the original work and normal in Lexicon’s HTS and secondary screens.

GNPTAB codes for the alpha and beta subunits of the GlcNAc-1-phosphodiester α-N-acetylglucosaminase, which is synthesized as a type III membrane precursor protein subsequently activated by site-1 protease. Patients with mutations in GNPTAB suffer from mucolipidosis type II which involves missorting of lysosomal proteins. Among other phenotypes, these patients have disabling skeletal abnormalities. Transgenic mice with a single base-pair insertion into Gnptab mimicking a human mutation are osteopenic from reduced bone formation and enhanced osteoclast activity.¹⁷⁵ Lexicon’s HTS and secondary screen showed KO mice have reduced bone mass in the spine. Lexicon has published a full analysis of the retinal defects in these KO mice, with data on body weight and length and cartilage histologic abnormalities.¹⁷⁶ Gnptg KO mice with disruption of the gamma subunit of this enzyme have only minor defects, presumably due to residual beta-glucuronidase activity.¹⁷⁷

KISS1R (kisspeptin1 receptor, GPR54) is a hypothalamic G-coupled receptor activated by the peptide Kisspeptin to initiate gonadotropin releasing hormone secretion at puberty. Mutations in either this ligand or its receptor result in hypothalamic hypogonadism in both sexes and KO mice continue to provide valuable insights into reproductive physiology.^178,179 Lifelong KO of Kiss1 or Kiss1r in mice should produce similar phenotypes to those observed with orchidectomy or ovariectomy prior to puberty. As expected, histological examinations showed Lexicon’s Kiss1 and Kiss1r KO mice both had typical lesions of hypogonadotropic hypogonadism including immature gonads and absence of sexual dimorphism (kidney and salivary glands in the male, mammary gland in the female). All bone parameters were low in the HTS and secondary screens. For Kiss1r KOs, DEXA and microCT data from the secondary screen are provided in Supplementary Figure S2. Both male and female KO mice were obese, but through different mechanisms. Male mice had normal body fat content but low LBM, whereas female mice had normal LBM but excess body fat.¹⁸⁰

Myostatin is a secreted protein that inhibits myogenesis with spontaneous mutations in humans, dogs and cattle resulting in dramatic muscle hypertrophy. In addition to muscle hyperplasia, Mstn KO mice have elevated bone mass.^181–183 In Lexicon’s HTS, Mstn KO mice had high muscle mass as determined by DEXA LBM (129% of normal), QMR LBM (126% of normal), gastrocnemius muscle weight (198% of normal) and increased muscle fiber number but not size by histology. An additional seven KO and five WT male mice were examined at 116 weeks of age (Supplementary Table S4). As expected, QMR LBM (120%) and gastrocnemius weight (188%) were increased. Spine trabecular bone, femur and tibia lengths (data not shown) and cortical bone diameters at the femoral midshaft or tibia–fibula junction were all unaltered in KO mice. Cortical bone thickness was elevated at the femoral midshaft and tibia–fibula junction. Cortical bone in the distal tibia (halfway between the tibia–fibula junction and the distal bone end) had an enlarged diameter but normal thickness. These site-specific bone changes demonstrate that distinct skeletal locations are influenced by lifelong muscle hypertrophy.

SLC30A5 is a zinc transporter with Slc30a5 KO mice reported to have extremely low bone mass¹⁸⁴ and this observation was confirmed for Lexicon’s KO mice in both primary and secondary screens. Mechanisms responsible for this severe osteopenia are unclear.

KOs with minor bone phenotypes

Table 5 provides HTS and secondary screen data for ten KOs having minor bone phenotypes.

Table 5 Primary and secondary screen data for KOs with minor bone phenotypes

KO of the calcitonin receptor (Calcr) in mice led to embryonic lethality when exons 6 and 7 were targeted¹⁸⁵ but normal viability and trabecular bone mass in spine and femur when exons 13 and 14 were targeted.¹⁸⁶ Lexicon’s Calcr KO mice were generated by targeting exon 1 and no clear bone phenotype was observed in the HTS.

Gremlins are BMP inhibitors and osteoblast-specific KO of Grem1 results in transiently elevated trabecular bone mass.¹⁸⁷ Global Grem1 KO mice, examined in a C57BL/6-FVB hybrid background to minimize neonatal lethality occurring in C57BL/6 mice, had reduced growth rates, missing fibula and generally low trabecular bone mass.¹⁸⁸ Human GWAS studies identified GREM2 SNPs that influence hip bone mass¹⁸⁹ and trabecular but not cortical bone.¹⁹⁰ Craniofacial development in zebrafish is influenced by GREM2 expression, as part of a regulatory system involving BMP4, endothelin-1 and jagged 1b.¹⁹¹ Lexicon’s Grem2 KO mice showed elevated spine and femur BMD in the HTS and dental defects.¹⁹² Additional studies are required to characterize possible skeletal phenotypes in this mouse KO.

IL6RA codes for the IL-6 receptor-alpha and an antibody to this receptor (tocilizumab) is an effective treatment for rheumatoid arthritis. In studies of Il6ra KO mice, no skeletal phenotypes were observed but one of two studies^193,194 indicated that KO mice are protected from ovariectomy-induced bone loss. Treating normal mice with an IL-6 neutralizing antibody following ovariectomy has given conflicting results, both inhibiting bone loss¹⁹⁵ and being without effect.¹⁹⁶ Secondary hyperparathyroidism from dietary calcium deficiency produced similar bone loss in WT and IL-6 KO mice.¹⁹⁷ Transgenic mice overexpressing IL-6 are growth retarded and have low bone mass.¹⁹⁸ Continuous low-dose PTH infusion in normal mice increases levels of serum IL-6 and bone resorption markers. This PTH-stimulated bone resorption, but not bone formation, was suppressed with cotreatment with an IL-6 neutralizing antibody.¹⁹⁹

FACS analysis confirmed that isolated spleen cells from Lexicon’s Il6ra KO mice (N=10 for both KO and WT) did not bind a commercial IL-6 receptor antibody. KO mice had normal bone mass in Lexicon’s HTS and normal body, spine and femur BMD in two additional cohorts (21 and 25 weeks of age (Supplementary Table S5A). Mice in both cohorts underwent sham or ovariectomy surgery with bone analyses on the combined cohorts after 6 weeks. There was no effect of Il6ra KO on bone loss following ovariectomy, determined by both DEXA BMD scans and microCT analyses of spine BV/TV and midshaft femur diameter and cortical thickness (Supplementary Table S5B). Treating a third mouse cohort (97 weeks of age) with teriparatide produced similar elevations in serum PINP levels in control and KO mice (Supplementary Figure S3a), suggesting IL-6 is not involved in the anabolic actions of teriparatide.

SLC26A7 is an anion transporter with Slc26a7 KO mice having reduced gastric acid secretion and distal renal tubular acidosis.²⁰⁰ Lexicon’s Slc26a7 KO mice were hypothyroid with males more strongly affected than females. Serum thyroxine levels (N=6 or 7) were reduced by 87% in males (P<0.001) and 47% in females (P=0.003). Histologic observations showed hyperplastic thyrotrophs in male but not female mice. Bone mass was slightly elevated in the HTS screen. Follow-up studies examined male and female mice separately between 18 and 21 weeks of age (Supplementary Table S6). Consistent with modest hypothyroidism, there were minimal effects of Slc26a7 KO on bones from female mice. Male mice, being severely hypothyroid, had modest reductions in LBM (20%) and femur length (6%), but elevated vertebral trabecular bone as indicated by high spine BMD (20%), LV5 trabecular BV/TV (42%) and LV5 trabecular thickness (28%).

The sodium/iodide symporter SLC5A5 plays a major role in thyroid iodide uptake and mutations in this gene result in hypothyroidism. Analyses of 12 different SLC5A5 mutations shows marked clinical heterogeneity in patients with the same and different mutations without a clear genotype-phenotype correlation. This observation ‘… suggests that additional unknown factors are involved in the clinical manifestation’ of this defective iodide transport.²⁰¹ There do not appear to be any data for Slc5a5 KO mice. SLC26A4 (PENDRIN) is an anion exchanger with broad specificity, including iodide, and Slc26a4 KO mice are deaf without a thyroid phenotype.²⁰² Since Slc26a7 KO mice are hypothyroid, this anion transporter might also contribute to iodide uptake by thyrotrophs as mouse SLC26A7 is capable of both chloride and iodide transport.²⁰³

SLC39A1 is a zinc transporter expressed in osteoclasts.²⁰⁴ Triple KO of mouse Slc39a1, Slc39a2 and Slc39a3 did not produce an obvious phenotype unless the mice were fed a zinc-deficient diet.²⁰⁵ Lexicon’s Slc39a1 KO mice had normal BMD and LV5 trabecular bone in the HTS but elevated cortical thickness. MicroCT analyses of bones from a second cohort of female mice examined at 42 weeks of age (N=18 for WTs and N=21 for KOs) confirmed the lack of a LV5 trabecular bone phenotype and femur length was normal. Midshaft cortical bone scans at 6 micrometer voxel size showed elevated total area (23%, P<0.001) and cortical thickness (6%, P=0.09) but reduced material BMD (3.2%, P<0.001) consistent with cortical porosity visually observed in the scans.

SPARC (secreted protein, acid, cysteine-rich, also known as osteonectin) is a widely expressed matricellular glycoprotein that has been extensively studied. Two of three Lexicon Sparc KO mice examined had cataracts, as previously reported.^206,207 KO mice have increased subcutaneous fat with normal body weight.²⁰⁸ Lexicon’s KO mice did not have an obesity phenotype when fed mouse chow or high fat diet (data not shown) but subcutaneous fat was not examined. KO mice develop intervertebral disk degeneration with aging,²⁰⁹ but this defect was not noticed in Lexicon’s mature KO mice. Several studies have reported modest osteopenia in KO mice^210–212 and Lexicon’s Sparc KO mice showed trends for low bone mass.

Secreted phosphoprotein 1 (osteopontin) is a widely-expressed, highly acidic SIBLING phosphoprotein degraded into small fragments by PHEX²¹³ and having many non-skeletal actions. Numerous abnormalities, including fibrosis, abnormal wound healing, defective cardiac remodeling, tumor metastases, aberrant angiogenesis and ectopic calcification, have been observed in Spp1 KO mice. Trabecular bone mass has been reported to be both normal^214–216 and high.^217–220 Lexicon’s Spp1 KO mice had normal bone mass in the HTS.

Actin plays a critical role in osteoclast activity and is a component of the podosome and ruffled borders. Actin dynamics are regulated by many factors²²¹ including cofilin, one of the most highly expressed genes in osteoclasts.²²² There are three cofilin genes: cofilin1, cofilin2 and destrin (actin depolymerizing factor). Lexicon’s KOs of Cfl1 and Cfl2 were both embryonic lethal, with no bone phenotypes in HTS or secondary screens for heterozygous mutant mice. KO of destrin increases bone loss following unloading.²²³ Cofilin activity is inhibited by LIM kinase phosphorylation and reactivated after dephosphorylation by slingshot phosphatases.^224–226 Osteoclasts isolated from Limk1 KO mice are hyperactive and published Limk1 KO mice have spine abnormalities²²⁷ and reduced bone mass.²²⁸ However, Lexicon’s Limk1 and Limk2 KO mice had no HTS skeletal phenotypes. There are three members in the Slingshot (SSH) family and disruption of Slingshot activity might suppress osteoclast activity. KOs of Ssh2 and Ssh3 did not have HTS skeletal phenotypes (data not shown), but Ssh1 KO mice had HBM in the HTS and three additional cohorts of KO mice were examined (data not shown). Elevated bone mass was confirmed in Cohort #1 female, but not male mice; was not confirmed in Cohort #2 female mice (males were not examined); and was confirmed in Cohort #3 male and female mice. Ssh1 KO did not protect Cohort #3 mice from ovariectomy-induced bone loss between 16 and 24 weeks of age. Given its small magnitude, the inconsistency of this skeletal phenotype is not surprising. Normal bone loss following ovariectomy made this gene uninteresting as an osteoporosis drug target. Ssh2 and/or Ssh3 might also be active in osteoclasts and a double or triple KO study is required to examine this topic.

Sclerostin domain containing 1 (SOSTDC1) codes for Wise (Ectodin) protein, which has a close homology to sclerostin and also inhibits canonical Wnt signaling by binding to Lrp5 and Lrp6. Lexicon’s Sostdc1 KO mice did not show any obvious skeletal phenotypes in the HTS, consistent with observations of no phenotype²²⁹ and mildly elevated BMD through 3 months of age that disappears at 4 months of age.²³⁰

Robinow syndrome, with major skeletal abnormities, results from mutations in WNT5A.²³¹ KO of Wnt5a results in perinatal lethality²³² with mouse fetuses having numerous dental²³³ and skeletal²³⁴ defects. Wnt5a heterozygous mutant mice exhibit impaired osteoclastogenesis.²³⁵ Skeletal actions of Wnt5b are less clear, as KO mice have been examined only for neuronal development.²³⁶ Chondrocyte hypertrophy and bone mineralization are delayed in transgenic mice overexpressing Wnt5b in cartilage.²³⁴ Lexicon’s Wnt5b KO mice had elevated bone mass in the HTS and secondary screens. Studies on additional cohorts of male and female mice showed Wnt5a KO mice often had slightly elevated bone mass, but this phenotype was not consistently observed (data not shown). The low magnitude of this potential elevated bone mass phenotype does not permit definitive conclusions about Wnt5b.

Novel genes with detrimental bone phenotypes

SLC10A7 is a Na cotransporter with unknown substrate specificity. Lexicon’s Slc10a7 KO mice exhibited moderate skeletal dysplasia (osteochondrodyplasia), characterized by markedly shortened and mildly bowed limbs. In contrast to the shorted diaphyseal regions of the bone, the epiphyses and joints were of normal size and the proximal metaphyses were mildly flared, suggesting that limb shortening was due to reduced endochondral bone growth (data not shown). At necropsy, the knees and other joints were loose due to ligamentous laxity, which would contribute to bowing of the legs during development. KO mice were small, as LBM was 77% of normal. The HTS indicated reduced bone mass (88% for spine BMD and 90% for femur BMD). MicroCT scans gave normal values for LV5 BV/TV (16%) and midshaft femur (269 µm for CT and 1.8 mm² for total area).

SLC25A1 is a mitochondrial citrate transporter. Lexicon’s Slc25a1 KO mice were small and sickly and surviving mice were examined at 2 weeks of age. Microscopic analysis revealed generalized hypoplasia. At the growth plate, the zones of hypertrophy were narrowed and there were notable reductions in the number of osteoblasts and the amount of osteoid. These lesions are suggestive of a metabolic deficit, resulting in a generalized hypoplasia that was most severe in liver, bone and bone marrow (data not shown).

SLC30A10 is a manganese transporter and patients with gene mutations have hypermanganesemia and Parkinsonian-like gate disturbances.²³⁷ Lexicon’s Slc30a10 KO mice were sickly and euthanized at 8 weeks of age. There were multiple abnormalities, including osteopenia, characterized by the loss of most trabecular bone at the metaphyses (data not shown).

Miscellaneous genes/topics

For most genes, the KO strategies employed completely inactivate gene function and genetically null mice are examined. Occasionally translation of some exons results in expressed proteins having residual activities, providing information on functions of various protein domains and their tissue specificity. This topic was mentioned above for Lrrk1, as Lexicon’s KO mouse is osteopetrotic whereas a different KO targeting strategy results in neonatal lethality. Two additional examples (Clnc7, Hdac4) from Lexicon’s database should be mentioned.

Mutations in the CLCN7 chloride channel gene produce osteopetrosis in human patients and several groups have observed osteopetrosis in Clcn7 KO mice. Transgenic mice having their Clcn7 gene engineered to contain the human CLCN7 G213R missense mutation develop lethal osteopetrosis.²³⁸ Although Clcn7 KO mice generated at Lexicon showed retinal and neuronal degeneration, their skeletons and coat color were normal. All Clcn7 KOs show growth retardation with poor viability and Lexicon’s mice were examined at 6 weeks of age. Lexicon’s KO mice were generated by disruption of exon 1, allowing expression of an active bone-specific Clcn7 isoform that did not involve translation of the first exon.²³⁹ Previous studies involved disruption of exons 3–7^240,241 or 8–10.²⁴¹

The histone deacetylase 4 gene contains 31 exons, with the N-terminal portion of the protein interacting with MEF2C and RUNX2 and the C-terminal portion containing the deacetylase domain. Hdac4 KO mice generated by targeting exon 6 do not survive to weaning and have chondrocyte hypertrophy similar to that observed with KO of Runx2.²⁴² This lethality and chondrocyte defect can be rescued in the presence of Mef2c heterozygosity.²⁴³ Lexicon examined Hdac4 KO mice generated by gene trapping with an insertion between exons 15 and 16, allowing (reduced) expression of the MEF2C and RUNX2 interacting domains but disrupting the C-terminal deacetylase domain.²⁴⁴ These Hdac4 KO mice survived for at least 12 months without any obvious skeletal abnormalities and minimal neurological defects. HTS DEXA scans gave 101% for vBMD, 101% for spine BMD and 96% for femur BMD. MicroCT scans gave normal values for LV5 BV/TV (17%) and midshaft femur (222 µm for CT and 2.68 mm² for total area). LBM was slightly low, at 89% in the HTS and 86% in high fat diet challenged mice.

Mutations in PLEKHM1 in both human and the spontaneous incisors absent rat cause severe osteopetrosis.^245,246 In contrast, Lexicon’s Plekhm1 mouse mutation had normal bone mass in the HTS with no evidence of osteopetrosis. This allele was generated through gene-trapping and carries an insertion within the first intron of the gene. RT-PCR of RNA collected from heart and kidney tissue clearly demonstrated the lack of endogenous transcription in the homozygous mutant animals (Supplementary Figure S3b). This lack of bone phenotype was independently confirmed in 8-week-old male mice, with lack of PLEKHM1 protein expression verified by western blotting of osteoclasts generated from KO mice (Fraser Coxon, University of Aberdeen; personal communication). The reason for this phenotype difference between mice and humans/rats with gene disruption is unclear.

Since Lexicon is primarily interested in identifying genes affecting bone mass for evaluation as osteoporosis drug targets, KO mice were not specifically examined for craniofacial abnormalities. However, histological examinations detected several KOs with abnormal craniofacial structures. We previously described hydrocephalus with craniofacial abnormalities in Stk36 (serine threonine kinase 36) KO mice as part of an analysis of 12 KO lines having hydrocephalus.²⁴⁷ Lexicon’s KO of Usp34 (ubiquitin-specific peptidase 34) resulted in half of the mice dying prior to 4 weeks of age. Surviving mice grew poorly and exhibited numerous neurologic abnormalities with lymphoid and hepatocyte hypoplasia. Flattened skulls were noted during behavior testing, but no additional observations were made. There does not appear to be any published mouse KO for this gene.

Disruption of the Pgap1 (post-GPI attachment to proteins 1) gene by KO²⁴⁸ and ENU-generated mutation²⁴⁹ results in severe developmental facial abnormalities with incomplete penetrance in KOs and strain dependence with ENU mutations. Two Lexicon Pgap1 KO mice were born with no facial features (Supplementary Figure S1j), lacking nostrils, mouth, tongue, mandible and ear canals. Eyes and other structures of the face were hypoplastic and deformed. However, most KOs survived into adulthood with reduced BW and length. For survivors, DEXA scans gave 85% for LBM, 97% for vBMD, 82% for spine BMD and 87% for femur BMD.

RASSF5 is a tumor suppressor gene and the published mouse KO study focused on oncologic parameters.²⁵⁰ Lexicon’s Rassf5 KO mice had craniofacial malformations noted by gross observations and CAT scans (skulls had reduced length, a short nasal bone, a short zygomatic bone and a short mandible with a stunted coronoid process). There was histological evidence of sternal malformation consistent with chondrodysplasia (Supplementary Figure S1k). Post-cranial skeletal architecture appeared generally normal at 102% for vBMD and 99% for femur BMD. Trabecular bone mass might be elevated as spine BMD was 110% of normal and LV5 trabecular BV/TV (23%) was slightly elevated.

The three steps in ascorbic acid synthesis starting from D-glucuronate are catalyzed by aldehyde reductase/aldose reductase, gulonolactase, and gulonolactone oxidase. The genes coding for these four enzymes are AKR1A1, AKR1B3, RGN and GULO, respectively. Rgn²⁵¹ and Gulo²⁵² KO mice develop lethal scurvy and spontaneous fractures occur in Rgn KO mice. Haplorhini primates, guinea pigs, Shionogi (ODS) rats²⁵³ and spontaneous fracture (Sfx) mice^254,255 all have spontaneous GULO gene mutations. KO of Akr1a1 was performed at Lexicon, but these KO mice were not examined in Lexicon’s HTS. Akr1a1 KO mice develop normally but have spontaneous bone fractures during pregnancy and following ovariectomy. Further analyses showed ascorbate levels in these KO mice were 15% of normal, which are sufficient for normal development but result in scurvy during states involving oxidative stress. DKO of Akr1a1 and Akr1b3 results in complete ascorbic acid deficiency and scurvy.²⁵⁶ Surprisingly, the contributions of aldehyde and aldose reductases to ascorbic acid synthesis were not known prior to this study.

Many genes affecting bone are also involved in tooth development.^257,258 As summarized in Supplementary Table S7, KO of Cebpb, Fam20a,⁴² Fam20c,⁴² Grem2,¹⁹² Ostm1,²⁵⁹ Postn,²⁶⁰ Sostdc1^261,262 and Src²⁶³ resulted in dental abnormalities.

KOs not confirming published bone phenotypes

Tryptophan hydroxylase (TPH) is the rate-limiting enzyme in serotonin synthesis and different genes code for peripheral (TPH1) and neuronal (TPH2, with two splice variants) enzymes. Greater than 90% of peripheral serotonin is produced by gut enterochromaffin cells and acts to modulate gut motility. 5-hydroxyindoleacetic acid (5-HIAA) is a metabolic breakdown product of serotonin that cross the blood-brain barrier and urinary 5-HIAA levels can be used to estimate body serotonin turnover. 5-HIAA is also produced in the gut from the metabolism of foods, such as bananas, that contain high levels of serotonin and therefore subjects must consume restrictive diets if urinary 5-HIAA is to provide useful information.

Mouse KOs of both Tph1 and Tph2 are viable without gross phenotypes.^70,264–266 Lexicon examined one Tph2 KO mouse line²⁶⁴ and two distinct KOs of Tph1^264,266. Brain and gut serotonin contents of Tph2 and Tph1 KO mice, respectively, are extremely low.

From studies on Tph1 and Tph2 KO mice, considerable attention has focused on potential skeletal actions of both neuronal and gut-derived systemic serotonin. Briefly, there are reports that Tph1 KO mice with negligible circulating serotonin levels have HBM resulting from elevated bone formation,⁷¹ whereas Tph2 KO mice with negligible brain serotonin content have low bone mass.²⁶⁷ DKOs with disruptions of both Tph1 and Tph2 have low bone mass.²⁶⁷ These results suggest that an orally active inhibitor of gut TPH1 that does not cross the blood–brain barrier might be a novel anabolic agent for treating osteoporosis. Lexicon explored this proposed gut-derived serotonin bone anabolic response as its orally active TPH1 inhibitors are undergoing clinical trials for irritable bowel and carcinoid syndromes.^268–271

Lexicon’s Tph1 and Tph2 KO mice had no skeletal phenotypes in HTS analyses (Table 6). Given the therapeutic importance of TPH1 as potential drug target, additional cohorts of Tph1 and Tph2 KO mice were examined. These studies do not support proposals that gut-derived serotonin inhibits bone formation and brain serotonin is required for proper bone development. Bone data for Lexicon’s Tph1 KO mice (males at 4 and 7 months plus females at 7 months), along with independent data obtained in multiple cohorts of male and female KO mice from 3 through 6 months of age, have been presented.⁷⁰ Lexicon has subsequently analyzed additional Tph1 KO mice, Tph2 KO mice²⁷² and Tph1/Tph2 DKO mice and detailed analyses of these studies will be reported separately.

Table 6 Primary screen data for Tph1 and Tph2 KO mice

Genes of interest with embryonic/neonatal lethality

KO of several genes having important roles in bone metabolism results in lethality of homozygous embryos. Data for heterozygous mutant Dkk1 mice are described above. Our HTS screens evaluated heterozygous mutant mice for Fam20b, osteogenesis imperfecta gene Tmem38b, WNT coreceptors Lrp4 and Lrp6, Lrp5/Lrp6 chaperone Mesdc2, WNT secretion chaperone Wntless (Gpr177), WNT signaling regulator R-spondin2 (Rspo2), R-spondin receptors Lgr4 (Gpr48) and Lgr5 (Gpr49), glucocorticoid receptor Nr3c1, fibroblast growth factor receptors Fgfr1 and Fgfrl1, exostosis gene Ext1, Wiskott–Aldrich syndrome protein Wasf1 and its homolog Wasf2 involved in actin dynamics, and the bisphosphonate drug target Fdps. In each case, there were no dramatic bone phenotypes in heterozygous mice.

SLC4A22 is a chloride-bicarbonate anion exchanger and two groups have shown Slc4a2 KO mice survive to adulthood with reduced growth and severe osteopetrosis.²⁷³ Lexicon’s KO mice were small and sickly with death occurring about 3 weeks of age. The explanation for this difference in survivability may involve differences in mouse strains, as the other groups examined FVB/N and 129S6/SvEv Tac/Black Swiss hybrid mouse strains.

Mutations in the human gene and KO of the mouse gene coding for the lipid biosynthetic enzyme 1-acyl-sn-glycerol 3-phosphate O-acyltransferase 2 (AGPAT2) result in severe lipodystrophy and reduced viability.^274,275 Two of four mice Agpat2 KO mice examined histologically at ten days of age had spontaneous proximal femur (hip) fractures.²⁷⁶ No HTS bone phenotypes were observed in heterozygous mutant mice.

Sulfatase modifying factor 1 (SUMF1, formylglycine-generating enzyme) enzymatically activates all 17 sulfatases by converting a critical cysteine in the active site to formylglycine and patients with SUMF1 mutations suffer from multiple sulfatase deficiency characterized by the combined effects of simultaneous deficiency of all sulfatases. Sumf1 KO mice have multiple abnormalities, including reduced viability, growth retardation, seizures, kyphosis, joint deformaties and lysosomal storage disease involving GAG accumulation.^277,278 These findings were confirmed in Lexicion’s KO mice, having juvenile lethality and disrupted epiphyseal cartilage maturation with absence of normal columnar arrays of proliferating and hypertrophic chondrocytes.

Example of a HTS false-positive bone phenotype

All HTS screens detect false-positive phenotypes and miss true phenotypes (false-negatives). False-positive hits are detected by failure to confirm the HTS phenotype in subsequent studies. One false-positive example from Lexicon’s HTS is Cytl1 (cytokine-like 1), a cytokine expressed in chondrocytes²⁷⁹ that promotes chondrocyte differentiation in mouse limb bud mesenchymal cells.²⁸⁰ Lexicon’s Cytl1 KO mice had a clear skeletal phenotype in the DEXA HTS, with values of 107% for body vBMD, 106% for spine BMD and 111% for femur BMD. MicroCT parameters were slightly elevated, with LV5 BV/TV being 22% and a midshaft femur cortical thickness value of 256 µm. This HTS high BMD phenotype could not be confirmed in follow-up DEXA or microCT scans of bones (N=10 for both male and female mice) from an additional cohort of KO mice examined at 18 weeks of age (data not shown).

Discussion

This report summarizes the most comprehensive analysis to date of skeletal phenotypes observed in a HTS of gene KO mice. The number of genes examined in viable mice is many-fold higher than previous studies^1,10,24,25 and replication of phenotypes for genes previously established to affect bone mass and architecture demonstrated the effectiveness of the HTS and a low likelihood of false negatives. Identification of eight genes not previously known to have dramatic KO skeletal phenotypes suggests that many additional genes important in bone biology remain to be discovered.

Our successful HTS for identifying skeletal phenotypes in KO mice involved several key decisions.

1. Mouse phenotypes can be strongly affected by strain,^281–283 with three dramatic metabolic examples being (i) cold sensitivity in Ucp1 KO mice;²⁸⁴ (ii) obesity, diabetes and endocrine parameters in leptin-deficient ob mice;²⁸⁵ and (iii) impaired insulin secretion in C57BL/6J mice.^286,287 Numerous studies have shown that bone architecture and skeletal responses to ovariectomy and exercise are influenced by the mouse strains examined. Compared to most other strains, cortical bones in C57BL/6J mice have large diameters but low thickness, and therefore low areal BMD. As shown in Supplementary Table S8, we have confirmed this common finding. As expected, midshaft femur cortical bone thickness (245 µm) in F2 hybrids is intermediate between values of the two parental strains. Values for femur length and trabecular bone in the LV5 vertebral body and distal femur metaphysis were similar for these strains. With the exception of 28 C57BL/6J KO lines (three of which were embryonic lethal), phenotyping was performed in F2 hybrids of C57BL/6J and 129 SvEv KO mice.

2. Use of DEXA allowed us to simultaneously screen both BMD and body composition (body fat and LBM). Body composition data for Lexicon’s first 2322 KO lines have been published.¹¹

3. Bone cell morphology was examined in decalcified histologic sections, which allowed distinction between osteopetrosis (Src, Ostm1 and Lrrk1) and osteosclerosis (Sost) in KOs with HBM. The absence of multinucleated osteoclasts was detected for Dcstamp KO mice and bones from Fam20c KO mice had osteomalacia. As argued by others,^22,288,289 Lexicon’s experience supports inclusion of histopathology in HTS screens of KO mice.

4. We did not include bone strength analyses in our primary HTS because biomechanical testing has high variability and we thought the possibility that bone strength would be affected independently of bone mass and architecture was low. Bone strength measurements were performed in KO lines of interest after confirmation of the skeletal phenotype. Compressive strength of LV5 is elevated with disruptions of Sost and Sfrp4,³⁰ whereas the strength of the femur shaft determined by three-point or four-point binding tests is greatly increased in bones from Sost KO mice. but decreased with Sfrp4³⁰ and Wnt16³⁸ disruption. Supplemental Figure 3c shows the excellent correlation of femur shaft strength and cortical thickness (three-point bending) in WT and Wnt16 KO mice.³⁸

5. Although we did not include bone length measurements in our primary screen, we believe this information is important in secondary screens. Bone length was unaffected in KOs of Wnt16 and Sfrp4 but reduced slightly with KO of Lrrk1.³⁹

6. Analyzing global gene KOs to identify drug targets provides information on the effects of gene disruption throughout the body. Examining tissue-specific conditional KOs can address mechanistic questions related to cell types involved in biochemical pathways, but drugs act throughout the body. As a hypothetical example, consider a gene that when disrupted increases bone mass and also results in lung pathology. A drug targeting the protein coded by this gene would be expected to produce on-target pulmonary toxicity in addition to any beneficial skeletal effects. Examining global KOs allows drug target validation prior to starting medicinal chemistry programs.

With the experience gained performing this HTS and follow-up studies on skeletal phenotypes, several additional topics merit discussion.

1. With the exception of Slc26a7 KO mice (with severe hypothyroidism in males but subclinical hypothyroidism in females), we did not observe clear sexual dimorphism in any of our bone phenotypes. In our opinion, gender differences in bone parameters during HTS are often spurious statistical findings or suggestive evidence of a weak phenotype. Our experience with Ssh1 KO mice, described above, provides a good example. High bone mass was inconsistently observed in various cohorts of KO mice, with a detectable phenotype switching between male and female mice in different cohorts. Known mutations causing established human skeletal and mineral phenotypes occur in both sexes. Both approved (bisphosphonates, teriparatide, donosumab) and experimental (cathepsin K inhibitors, sclerostin antibodies) osteoporosis therapies are effective in men and women. Mouse KO and transgenic mice show minimal skeletal sexual dimorphism with disruptions of Lrp5, Sost, Rankl, Csk and Lrrk1 genes.

2. Failures by independent laboratories to confirm published preclinical findings have drawn considerable attention.^290–292 For bone research, guidelines²⁹³ and perspectives²⁹⁴ for describing skeletal phenotypes have been published. A key component of Lexicon’s mouse gene KO campaign has involved comparisons to published data, with confirmations of skeletal phenotypes for 23 previously reported genes (Calcr, Cebpb, Crtap, Dcstamp, Dkk1, Duoxa2, Enpp1, Fgf23, Kiss1/Kiss1r, Klotho, Lrp5, Mstn, Neo1, Npr2, Ostm1, Postn, Sfrp4, Slc30a5, Slc39a13, Sost, Src, Sumf1 and Wnt10b). As described above, such confirmation was not obtained for Sfrp1, Tph1 and Tph2. Lexicon’s data for Sfrp1 are limited and further studies are clearly indicated. Lexicon negative findings for Tph1 KO mice have been independently replicated and are consistent with additional studies performed at multiple laboratories.⁷⁰ The failure to confirm published Clcn7 and Hdac4 KO phenotypes is believed to result from different gene KO targeting strategies. The failure to observe osteopetrosis in Plekhm1 KO mice, which occurs with gene disruption in rats and humans, requires further study. We strongly believe that KO comparisons to WT littermate/cagemate mice are essential to control for environmental variables and possible genetic drift of the colony.

3. Analyzing LV5 vertebral bodies and midshaft by microCT provided data on both trabecular (LV5) and cortical (femur) bone. As shown by the skeletal phenotypes observed in Klotho, Src, Wnt16 and Sfrp4 KO mice, these two bone compartments can have distinct responses to gene deletion. For trabecular bone examinations, we believe that LV5 is superior to the distal femur and proximal tibia, as the amount of trabecular bone present is far greater and independent of both sex and age. Performing microCT scans of the distal femur metaphyseal trabecular bone remained an option since the entire femur was obtained at necropsy. Using both DEXA and microCT scans provided both global (DEXA) and focused (MicroCT) assessments of bone mass. Having both DEXA spine BMD and microCT LV5 trabecular bone volume data provided two vertebral measurements as a check for internal consistency.

4. Strength is compromised if bone is either under- or over-mineralized. Bone mineralization is readily assessed by material BMD determined during the microCT scans. Consistent with hypophosphatemia and histologic osteomalacia, bones from Fam20c KO mice had reduced material BMD.⁴²

5. We developed microCT methods to determine architecture of the mouse femoral neck and observed bone phenotypes at this site are consistent with the phenotypes observed at LV5 and the midshaft femur.²⁹⁵ Additional experience indicates the thickness of the LV5 cortical shell responds similarly to genetic variations in midshaft femur cortical thickness. Both cortical sites lose bone following ovariectomy and gain bone with teriparatide treatment (data not shown).

6. KO mouse phenotypes can vary depending on the genetic alteration employed. If a specific protein domain is selected for disruption and the corresponding exons are deleted, other exons of the gene can still be transcribed and translated, and partial gene function may remain. Two examples described above were Lexicon’s KO alleles of Clcn7 and Hdac4. Another example comes from studies of the vitamin D receptor (VDR), which contains an N-terminal DNA-binding domain and a C-terminal ligand-binding domain. Nutritional vitamin D deficiency and KO of the vitamin D 1-hydroxylase (Cyp27b1) result in hypocalcemic rickets in mice, although mice have normal fur. Initial KO studies of Vdr described rachitic and hairless (alopecia) mice, whereas humans with VDR mutations are always rachitic but alopecia is variable. Mutations in subjects with hair occur in the ligand-binding domain, leaving the DNA-binding domain intact.²⁹⁶ Thus, the VDR DNA-binding domain acts (through heterodimerization with RXR) in keratinocytes to promote hair growth without requiring binding of vitamin D. This phenotypic domain specificity also occurs in mice, as Vdr KO mice had rickets but normal fur if the Vdr ligand-binding domain was disrupted with the Vdr DNA-binding domain remaining intact.²⁹⁷

7. Although attention is usually focused on genes yielding phenotypes when disrupted, mutated genes that do not have bone and/or any obvious phenotypes also provide important and oftentimes surprising information.²⁹⁸ For example, Lexicon’s albumin (Abl) KO mice, having negligible levels of serum albumin, showed no obvious phenotype (data not shown). This observation is consistent with findings in analbuminemic humans and spontaneous mutant analbuminemic rats.²⁹⁹ Mouse KOs of S100g (Calbindin9K), a Ca-binding protein,^300–302 and Trpv6, a putative membrane Ca transporter,³⁰³ both failed to show expected intestinal calcium malabsorption phenotypes. Three examples of human genes acquiring loss-of-function mutations with only benign phenotypes are MC1R (red hair), ABO (O blood group) and CCR5 (HIV resistance). Recent explorations of the human genome have identified healthy people having genes with predicted loss-of-function homozygous mutations.^304–306 The occurrence of genes in healthy individuals having loss-of-function compound heterozygous alleles has been less thoroughly explored.

Genes can have critical functions only under environmental conditions not occurring during standard mouse phenotyping. Examples include (i) the GCN2 kinase regulating feeding behavior in response to meals with imbalances of essential amino acids;³⁰⁷ (ii) KO of the mitochondrial uncoupling protein UCP1 influences body fat depending on the age of the mice and the ambient temperature at which they are housed;³⁰⁸ (iii) normal iron status with KO of the iron-binding protein lactoferrin,³⁰⁹ but elevated colorectal dysplasia following inflammation;³¹⁰ and (iv) Gc KO mice without the serum vitamin D-binding protein have normal serum bioactive vitamin D levels but respond more quickly to dietary deficiency and show hypercalcemic toxicity at lower vitamin D doses.^311,312 Compared to sedentary KO mice, physical exercise (treadmills and running wheels) can uncover pathological phenotypes in otherwise healthy KO mice (Mybpc3, Lipe) or prevent obesity in Mc4r KO mice.³¹³ Thus, KO mouse lines without skeletal phenotypes in HTS might show abnormal disturbances in bone metabolism with increasing age or following challenges such as ovariectomy, mechanical loading or unloading, fracture healing, arthritis induction, dietary deficiencies and reproduction.

8. Comparisons of skeletal phenotypes showed excellent agreement for genes having mutations or SNPs in humans and corresponding mouse KOs. Phenotype concordance was observed for CRTAP, FAM20A, FAM20C, GNPTAB, LRP5, NPR2, OSTM1, SLC39A13, SOST and WNT16, with PLEKHM1 being the single exception.

9. During 2012, the IMPC estimated that ∼30% of mouse KO lines are embryonic or perinatal lethal³¹⁴ and increasing efforts are being focused on examining embryos to determine mechanisms involved in the developmental defects.^314,315 With 556 KO lines examined, the February 2014 IMPC update lists lethal (21%) and subviable (9%) lines. Analyses of the MGI database found lethality in 23% of 6812 mouse KO genes curated³¹⁶ and 2 472 lethal gene KOs (39%) among 6283 genes annotated to have direct human orthologs.³¹⁷ A comprehensive review described birth trauma, respiratory failure, homeostasis deficiencies and inability to suckle as common causes of lethality during the first 24 h after birth.³¹⁸ KO lethality is influenced by mouse strain and can sometimes be rescued by simultaneous disruptions in different genes, as is the case of Dkk1 KO survivability in the presence of Wnt3 heterozygosity described above.^133,134 The lower 18% lethality value observed by Lexicon might be related to examining few genes coding for transcription factors and structural proteins and hybrid vigor from studying C57BL6/J–129SvEv hybrid mice. As described above, Dkk1 heterozygous mutant mice and KO mice examined prior to 12 weeks of age due to reduced viability can show skeletal phenotypes (Fgf23, Klotho, Nppc, Npr2, Ostm1, Sgpl1, Slc25a1, SLC30a10, Sumf1).

Mouse resources

There are many resources providing both KO mice and phenotype data. During 2006, Francis Collins, Director of the US NIH, stated ‘A graduate student shouldn’t spend a year making a knockout that’s already been made. It’s not a good use of resources’.³¹⁹ There are many reviews describing the evolution of the IKMC since the 2003 Banbury Organizing Conference.³²⁰ KO mouse genes with available ES cells and/or cryopreserved sperm are compiled in the International Mouse Strain Resource, with a link from the Mouse Genomic Informatics (MGI) website and published links to centralized mouse repositories.³²¹ The Gene Expression Database³²² is included within the MGI resource. To facilitate tissue-specific gene KOs, most IKMC lines include the cre driver system during the gene KO targeting design and the MGI website maintains a CrePortal providing up-to-date information. Underappreciated complications of conditional cre-loxP excision include unexpected activity in off-target tissues, mosaicism and parent-of-origin effects.^323,324 Problems with the osterix-cre mouse line have been described.^325,326

The first compendium of 263 published mouse KO phenotypes appeared in 1995.³²⁷ The IMPC¹³ coodinates phenotype data from the IKMC, with 441 KO mouse lines fully analyzed and 643 KO mouse lines under examination (September 2014). Recent results include KOs of Lrrk1 and Wnt16, both having skeletal phenotypes similar to those described by Lexicon. The MGI includes phenotype data for 251 NIH-sponsored Deltagen and Lexicon KO mouse lines.

Commentaries on genome-wide KO efforts involving zebrafish,³²⁸ nematodes³²⁹ and yeast³³⁰ provide interesting perspectives for mouse KO efforts. Information on rat gene KOs is compiled in the Rat Genome Database.³³¹

Summary

Lexicon Pharmaceuticals, during the decade preceding the initiation of the IKMC program, successfully employed industrialized biology to generate and phenotype over 4 650 mouse gene knockout lines. The genes examined were highly represented in the druggable genome, foreshadowing the Illuminating the Druggable Genome program within the NIH Common Fund (http://commonfund.nih.gov/idg/overview). Global KO strategies involved both gene trapping and homologous recombination technologies. Phenotyping screens included a battery of tests in the areas of behavior, bone, cardiology, immunology, metabolism, oncology and ophthalmology, and included serum chemistry, histopathology and a high fat diet obesity challenge. Software was written for an internal database that tracked mice undergoing breeding and moving through the phenotyping screens and allowed investigators to view all phenotyping data for all mice. Lexicon’s KO mouse campaign successfully identified previously published and novel genes influencing bone mass, architecture, strength and mineralization (the Skeletome). Three of these novel genes (undisclosed) code for enzymes or secreted proteins that are potential osteoporosis drug targets.

Further review of the Lexicon dataset showed elevated spine bone mass phenotypes in both primary and secondary screens for Slc37a2 (glucose-6-phosphate/phosphate antiporter) and Slc41a2 (Mg transporter). Further studies are required to fully characterize skeletal phenotypes in these two gene KOs.