Targeting of Slc25a21 Is Associated with Orofacial Defects and Otitis Media Due to Disrupted Expression of a Neighbouring Gene

Homozygosity for Slc25a21tm1a(KOMP)Wtsi results in mice exhibiting orofacial abnormalities, alterations in carpal and rugae structures, hearing impairment and inflammation in the middle ear. In humans it has been hypothesised that the 2-oxoadipate mitochondrial carrier coded by SLC25A21 may be involved in the disease 2-oxoadipate acidaemia. Unexpectedly, no 2-oxoadipate acidaemia-like symptoms were observed in animals homozygous for Slc25a21tm1a(KOMP)Wtsi despite confirmation that this allele reduces Slc25a21 expression by 71.3%. To study the complete knockout, an allelic series was generated using the loxP and FRT sites typical of a Knockout Mouse Project allele. After removal of the critical exon and neomycin selection cassette, Slc25a21 knockout mice homozygous for the Slc25a21tm1b(KOMP)Wtsi and Slc25a21tm1d(KOMP)Wtsi alleles were phenotypically indistinguishable from wild-type. This led us to explore the genomic environment of Slc25a21 and to discover that expression of Pax9, located 3′ of the target gene, was reduced in homozygous Slc25a21tm1a(KOMP)Wtsi mice. We hypothesize that the presence of the selection cassette is the cause of the down regulation of Pax9 observed. The phenotypes we observed in homozygous Slc25a21tm1a(KOMP)Wtsi mice were broadly consistent with a hypomorphic Pax9 allele with the exception of otitis media and hearing impairment which may be a novel consequence of Pax9 down regulation. We explore the ramifications associated with this particular targeted mutation and emphasise the need to interpret phenotypes taking into consideration all potential underlying genetic mechanisms.


Introduction
Knockout mice are invaluable tools for studying the function of genes both during embryonic development and in the adult. Classically, gene disruption in mice is achieved by replacing a part of the target gene with a selectable marker, e.g. a neo cassette, leading to constitutive ablation of the targeted gene. More contemporary approaches include the creation of conditional alleles and targeted gene traps, as well as small hairpin RNA (shRNA) and the use of lentiviral transgenesis, reviewed in [1]. The Sanger Institute Mouse Genetics Project [2] is contributing to an international effort to conduct systematic, large-scale gene function analysis in a mammalian system, through the generation of mice using targeted ES cells available from the EUCOMM and KOMP resources [3]. The typical EUCOMM and KOMP knockout first conditional ready targeted trap [tm1a(EUCOMM) and tm1a(KOMP)] is a powerful allele configured with the potential to convert to a conditional allele or a lacZ tagged null allele [3]. This tm1a allele was selected for pragmatic reasons, including there being no requirement for additional breeding to modify the allele, as targeted traps typically disrupt expression of the targeted gene [3,4], and the major advantage that the allele, with full potential for conversion, would be cryopreserved for subsequent community use. However, it has previously been reported that the presence of selectable markers used for gene targeting, such as the neo cassette in the EUCOMM and KOMP tm1a allele, can interfere with the expression of neighbouring genes [5,6,7,8,9]. It is, therefore, important that phenotypes discovered in mice carrying the knockout first conditional ready allele, are verified using the derived deletion allele.
SLC25A21 has previously been reported to be ubiquitously expressed in all tissues tested [10], and codes for the 2-oxoadipate mitochondrial carrier (ODC) that catalyses the movement of several substrates, primarily 2-oxoadipate and 2-oxoglutarate, across inner mitochondrial membranes [10,11,12]. It is suspected that ODC functions by a counter exchange mechanism, catalysing the uptake of 2-oxoadipate into the mitochondrial matrix, in exchange for internal 2-oxoglutarate [10]. The physiological role of the ODC suggests that it may be involved in the human disease 2-oxoadipate acidaemia (OMIM 204750), characterised by an accumulation of 2-oxoadipate and associated metabolites in urine [11]. This inborn error of metabolism shows clinical symptoms of intellectual disability varying from mild to severe, hypotonia, seizures, motor and developmental delay, and cerebellar ataxia [13,14,15,16].
Here we report characterization of the allelic series derived from the KOMP targeted allele of Slc25a21. From this, three distinct observations were made. Firstly, ablation of Slc25a21 was not found to cause symptoms of 2-oxoadipate acidaemia in mice; animals homozygous for Slc25a21 tm1b(KOMP)Wtsi and Slc25a21 tm1d(-KOMP)Wtsi null alleles were normal for all parameters tested. Secondly, we report that Slc25a21 tm1a(KOMP)Wtsi homozygous mice present with profound dental, orofacial and hearing/middle ear phenotypes caused by off-target effects on the expression of the neighbouring gene Pax9. Finally, the resulting novel, hypomorphic allele of Pax9 opens the door to further studies into how Pax9 is involved in the patterning of the palatal rugae and may represent a novel model of otitis media in the mouse. With the growing resource of knockout first conditional ready targeted ES cells created by the EUCOMM and KOMP initiatives, these results represent a timely reminder that the phenotype observed in the knockout first conditional ready mice should be verified by creating and analysing the deletion allele.

Ethics Statement
The care and use of all mice in this study were in accordance with UK Home Office regulations, UK Animals (Scientific Procedures) Act of 1986, and were approved by the Wellcome Trust Sanger Institute Ethical Review Committee. All efforts were made to minimize suffering including housing mice in a specific pathogen-free unit at a density of 3-5 animals per cage in polysulfone individually ventilated cages [Tecniplast Seal Safe 1284L (overall dimensions of caging: (L6W6H): 36562076140 mm, floor area = 530 cm 2 )]. Sterilised Aspen bedding substrate and standard environmental enrichment of nestlet, cardboard tunnel, and three wooden chew blocks were provided. The light cycle was maintained at 12 h light/12 h dark with lights off at 19:30 hours and no twilight period. Room temperature was 2162uC and humidity was regulated at 55610%. Mice were given food and water ad libitum unless otherwise stated [2]. Animal welfare is our primary consideration and was assessed routinely for all mice involved in this study. Adult mice were killed by terminal anaesthesia followed by exsanguination and either cervical dislocation or removal of the heart. Embryos were killed by decapitation or cooling followed by immersion in cold tissue fixative. Experimental work was performed in accordance with the Arrive guidelines [17].
Mice were fed Mouse Breeders Diet (Lab Diets 5021-3, Richmond, Indiana, USA) unless otherwise specified, including Slc25a21 tm1d(KOMP)Wtsi mice processed through the Mouse Genetics Project screening protocol [2]. Slc25a21 tm1a(KOMP)Wtsi , Slc25a21 tm1b(-KOMP)Wtsi and Slc25a21 tm1c(KOMP)Wtsi mice which went through the Mouse Genetics Project screening protocol were transferred to a high-fat (21.4% fat by crude content) dietary challenge (Special Diet Services Western RD-829100, Witham, UK) at 4 weeks of age. Cages were processed randomly and different genotypes could be housed together, hence there was no pattern to the order in which animals were processed. The experimental unit throughout the study was a single mouse.

Validation of targeting event
To confirm the integrity and validity of each allele studied, detailed molecular characterisation was performed using shortrange, long-range and quantitative PCR strategies, as described previously [20] and detailed in File S1 and Table S1.
Work by Santagati [21] indicated the presence of a regulatory element, named conserved non-coding sequence +6 (CNS+6), for Pax9 within the allele design for Slc25a21 tm1a(KOMP)Wtsi . A PCR assay was designed (Table S1) to amplify Slc25a21 tm1a(KOMP)Wtsi samples and the resulting products sequenced to determine if the CNS+6 element was disrupted.
RNA sequencing was performed using 5 mg of total RNA from a subset of the above E13.5 embryo head samples [3 homozygotes and 2 wild-types for each allele (12 mutant and 8 control samples in total)] as detailed in File S1.

Physical assessment, body composition, radiography and clinical chemistry
Homozygous mice (7M and 7F) from each of the alleles (Slc25a21 tm1a(KOMP)Wtsi , Slc25a21 tm1b(KOMP)Wtsi , Slc25a21 tm1c(K-OMP)Wtsi and Slc25a21 tm1d(KOMP)Wtsi ), as well as age, sex and genetic background matched controls, were analysed using the standard Sanger Institute Mouse Genetics Project phenotyping screen [2]. Assays of particular relevance are described in File S1, whilst those not part of the standard phenotyping screen are described below.
At necropsy, skulls were collected and the teeth (both upper and lower jaw) and palate were reviewed and imaged (MZ16A dissecting microscope, Leica, Wetzlar, Germany; DFC490 digital camera, Canon Powershot G5, Japan). A range of 42 additional tissues and organs from homozygous Slc25a21 tm1a(KOMP)Wtsi mice (2M and 2F), and age, sex and genetic background matched controls, were fixed in 10% neutral buffered formalin (Leica Biosystems, Peterborough, UK) for 15-20 hrs and processed to wax. 5 mm sections were stained with haematoxylin and eosin then reviewed by an experienced pathologist.
Anaesthetised mice were placed on a heating blanket inside a sound attenuating booth and recording electrodes inserted in the skin to record responses of the left ear. For ABR threshold determination, click (0.01 ms duration) and tone pip (6,12,18,24, and 30 kHz of 5 ms duration, 1 ms rise/fall time) stimuli over a range of intensity levels from 10-95 dB sound pressure level (SPL) in 5 dB steps were presented in free-field. Averaged responses to 256 stimuli, presented at 42.2/s, were analysed and thresholds established as the lowest sound intensity giving a visuallydetectable ABR response.

Middle ear analysis
Following ABR testing, Slc25a21 tm1a(KOMP)Wtsi mice aged 9 weeks (wt, n = 8; het, n = 11; hom, n = 8) were killed by cervical dislocation and the right external and middle ears were examined in detail using a dissecting microscope for any signs of malformation or inflammation, including: appearance of excess cerumen in the external ear canal; thickening, whitening, sponginess or excess vascularisation of the bulla wall; clarity and vascularisation of the tympanic membrane; presence of fluid or white inflammatory material in the middle ear cavity; and solid masses or bony outgrowths of the middle ear wall or ossicles. In addition, the left side of the head of these mice was fixed in 10% formalin for 48 hours, and decalcified in 10% EDTA diluted in phosphate buffered saline (PBS) for 10 days, dehydrated, embedded in paraffin wax, sectioned at 8 mm and stained with haematoxylin and eosin.

Data analysis and statistics
A reference range approach was used to assess continuous data, including time course. For categorical data, a Fisher's exact test was used to identify phenotypic variants. More details of both approaches are presented in File S1. Since both of the above approaches to automatically identify significant calls are known to be conservative and assess each parameter in isolation, they were complemented by a manual assessment made by a biological expert who used knowledge of events on the day, or across sexes, or between related variables, to highlight additional potentially abnormal phenotypes (Table S2).
To assess the ABR data, a one-way Kruskall-Wallis ANOVA on Ranks was used where three genotypes were assessed and Mann-Whitney rank sum test where only homozygote and wild-type animals were assessed. The qPCR gene expression results for mutant mice were compared to littermate wild-type mice using a Student's t-Test assuming two-sample unequal variance. P-values for the RNA sequencing were adjusted for multiple testing with the Benjamini-Hochberg procedure. Pre-weaning lethality was assessed using the Test for One Proportion.
Those homozygous animals surviving to weaning, presented with decreased body weight ( Fig. 1A-B), fat mass ( Fig. 1C-D) and fat percentage estimate (data not shown), in both sexes, compared to wild-type controls. Furthermore, a range of dental abnormalities were observed in Slc25a21 tm1a(KOMP)Wtsi homozygous mice. At 10 weeks of age 13 out of 14 homozygous Slc25a21 tm1a(KOMP)Wtsi mice were found to have an abnormally shortened snout (data not shown). At 16 weeks of age all fourteen homozygous Slc25a21 t-m1a(KOMP)Wtsi mice were found to have macroscopically visible dental abnormalities ranging from white or translucent to severely hypoplastic lower incisors ( Fig. 2A-B) compared to wild-type (Fig. 2C). The molars of homozygous Slc25a21 tm1a(KOMP)Wtsi adult mice (Fig. 2D), in particular of the lower jaw, were found to be strongly stained with X-gal (requiring b-galactosidase activity) compared to heterozygous and wild-type ( Fig. 2E-F). Microscopic examination of sections from representative animals confirmed that in unaffected areas, the molars were fully developed with normal odontoblast morphology. In affected areas of the molars, there was focal fracturing of the surface dentine and necrosis of the pulp with associated inflammation and bacterial colonies (Fig. 2G).
Micro-CT scanning revealed that the lower incisor root area in homozygous Slc25a21 tm1a(KOMP)Wtsi mice was severely underdeveloped and that the lower incisors appeared to be lacking the enamel coating ( Fig. 2H-I) compared to wild-type (Fig. 2J). Furthermore, two dimensional X-ray images revealed several changes of the manidibular morphology including pronounced pogonion, pronounced anterior tuberosity, under-developed incisor process and alveolus (data not shown), potentially indicating a change in the developmental dynamics of the mandible in homozygous Slc25a21 tm1a(KOMP)Wtsi mice. Taken together, the 2D X-ray images, micro-CT and X-gal staining in tooth cavities suggests feeding might be uncomfortable for homozygous Slc25a21 tm1a(KOMP)Wtsi mice which may explain their reduced body weight and fat mass.
Finally, 14 week old Slc25a21 tm1a(KOMP)Wtsi homozygous mice were found to have a hearing impairment as they presented with elevated auditory brainstem response thresholds across all frequencies, with severity of impairment increasing with age ( Fig. 4A-D). Slc25a21 tm1a(KOMP)Wtsi homozygous mice displayed an approximately parallel pattern of threshold elevation across the range of frequency stimuli used, a profile which is typical of conductive hearing loss. Fig. 4E shows that click threshold in wildtypes was stable with increasing age, at least up to 26 weeks of age, whereas a progressive deterioration was seen in Slc25a21 tm1a(-KOMP)Wtsi homozygous mice over the same age range. At 26 weeks of age, some wild-type mice showed age-related high-frequency hearing loss (indicated by larger standard deviations at 24-30 kHz). The thresholds recorded in the Slc25a21 tm1a(KOMP)Wtsi homozygous mice at all ages were significantly elevated above agematched wild-types (p,0.001) (Fig. 4F).
To investigate the morphology of the middle ear, the right temporal bone, containing the entire ear, was dissected from Slc25a21 tm1a(KOMP)Wtsi homozygous mice at 9 weeks of age (n = 8) along with wild-type (n = 8) and heterozygous (n = 11) littermates. Compared to wild-types (Fig. 5A), most of the homozygotes (7/8) showed opaque fluid or white debris in the middle ear cavity indicating otitis media (Fig. 5B). Other features of otitis media were more variable, including excessive vascularisation (n = 3) or whitening of the bulla (n = 4) or tympanic membrane (n = 1). The one homozygous Slc25a21 tm1a(KOMP)Wtsi mouse in which exudate was not observed in the middle ear also had ABR thresholds only slightly above those seen in wild-type mice. Heterozygotes had normal middle ears.
The left side temporal bones from the same eight Slc25a21 tm1a(-KOMP)Wtsi homozygous mice and wild-type controls were sectioned through the middle ear, stained with haemotoxylin and eosin, and examined microscopically. Inflammatory changes and accumulation of exudate consistent with otitis media were observed in six of the eight homozygotes (Fig. 5D, F).
The middle ear cavity of the six affected Slc25a21 tm1a(KOMP)Wtsi homozygous mice contained variable amounts of an amorphous eosinophilic exudate containing cholesterol crystals, foamy macrophages and neutrophils, indicating an active inflammatory response. In these mice, the lining epithelium of the middle ear cavity was variably hyperplastic and the underlying lamina propria was oedematous with congestion of blood vessels (Fig. 5D, F). None of the littermate controls showed evidence of inflammation ( Fig. 5A, C, E).
Of the two Slc25a21 tm1a(KOMP)Wtsi homozygous mice showing little or no effusion one had comparatively normal ABR thresholds, whilst the other had raised ABR thresholds.
The tympanic ring size was normal in Slc25a21 tm1a(KOMP)Wtsi homozygous mice; assessment of micro-CT images revealed no change in adults whilst alizarin red/alcian blue stained embryos revealed no changes at E18.5 (data not shown).
Systematic histopathology assessment of a range of tissues from homozygous Slc25a21 tm1a(KOMP)Wtsi mice did not reveal any further abnormalities, and in particular, no indication of any inflammation other than in the middle ear was found.

Ablation of Slc25a21 does not cause signs of 2oxoadipate acidaemia in mice
It has been proposed that loss of function of ODC, the protein encoded by SLC25A21, causes the human disease 2-oxoadipate acidaemia which is clinically characterised by hypotonia, seizures, motor and developmental delay, cerebellar ataxia and varying severities of intellectual disability [13,14,15,16]. Unexpectedly, homozygous Slc25a21 tm1a(KOMP)Wtsi mice did not phenocopy this human disease. In particular, lean mass, grip strength and gait, correlates of the human disease manifestation, were all normal for these mice (data not shown), and no seizures were detected. To ascertain the molecular consequence of the targeting event, expression analysis of Slc25a21 tm1a(KOMP)Wtsi was performed on RNA extracted from embryo heads (E13.5). When using the Slc25a21 TaqMan assay which spanned exons 8-9, 39 of the cassette insertion, expression was found to be reduced in homozygous Slc25a21 tm1a(KOMP)Wtsi samples (28.7% of wild-type expression, p = 2.0e-05) but not completely ablated (Fig. 6A). The hypomorphic nature of this allele may have explained the absence of the predicted phenotypes, therefore an allelic series was created (Fig. 6B-E) where tm1b was a lacZ reporter tagged deletion with the critical exon (exon 4) removed, tm1c a conditional allele with wild-type function restored, and tm1d the deletion allele where both exon 4 and the lacZ reporter were removed. Expression analysis of each of these Slc25a21 alleles was performed on RNA extracted from embryo heads (E13.5). Slc25a21 RNA expression was found to be ablated in homozygous Slc25a21 tm1b(KOMP)Wtsi (0% of wild-type expression, expression was below the limit of detection) and greatly reduced in Slc25a21 tm1d(KOMP)Wtsi homozygotes (12.9% of wild-type expression, p = 0.043) (Fig. 6A). As expected, expression levels of Slc25a21 in embryos homozygous for Slc25a21 tm1c(KOMP)Wtsi were not significantly different from wildtype expression (Fig. 6A).
These results led us to conclude that the phenotypes observed in mice homozygous for the Slc25a21 tm1a(KOMP)Wtsi targeted allele were not due to ablation of Slc25a21 function, and speculate that off-target effects due to the selection cassette present in the tm1a allele affected expression of neighbouring genes. The 1 Mb genomic interval containing Slc25a21 (500 kb flanking the target gene) is predicted to contain seven known protein coding genes, one non coding RNA, one novel antisense and one putative processed transcript (Fig. S1). Of particular interest is Pax9, located at the 39 end of Slc25a21 and orientated on the opposite strand. The 39UTR regions of Slc25a21 and Pax9 overlap by 188 bp. Consistent with the phenotypes observed in Slc25a21 tm1a(KOMP)Wtsi homozygous animals, mice carrying a hypomorphic Pax9 allele had previously been reported as causing reduced viability at weaning [25], and presented with a range of dental and craniofacial abnormalities [25,26] similar to those described herein.
RNA expression analysis of Pax9 was performed on embryo heads (E13.5) for all four alleles in the Slc25a21 series. Pax9 expression was found to be down regulated [approximately 34% of wild-type Pax9 levels (p = 1.1E-06)] in homozygous Slc25a21 tm1a(-KOMP)Wtsi samples (Fig. 8) confirming our hypothesis that we had inadvertently created a Pax9 hypomorphic allele. Pax9 levels were found to be unaffected in homozygous Slc25a21 tm1b(KOMP)Wtsi , Slc25a21 tm1c(KOMP)Wtsi and Slc25a21 tm1d(KOMP)Wtsi samples (Fig. 8). A known Pax9 regulatory element (CNS+6) was located within the targeting construct used to create the Slc25a21 tm1a(KOMP)Wtsi allele [21]. PCR and sequence analysis was used to demonstrate that this sequence was not disrupted during the vector construction or at the homologous recombination stage (Table S1; data not shown). Consequently, we hypothesise that the presence of the promoter driven neomycin cassette in the homozygous Slc25a21 tm1a(KOMP)Wtsi mice caused the reduction of Pax9 expression.
RNA sequencing was performed to further explore the impact of the targeting event on the genomic environment around Slc25a21. Comparing RNA from wild-type and homozygous Slc25a21 tm1a(KOMP)Wtsi E13.5 embryo heads, the only gene in the 1 Mb interval surrounding Slc25a21 that had significantly altered expression levels was Pax9 [q (adjusted p value) = 7.10E-06 (data not shown)]. Pax9 was not found to be differentially expressed in homozygous Slc25a21 tm1b(KOMP)Wtsi , Slc25a21 tm1c(KOMP)Wtsi and Slc25a21 tm1d(KOMP)Wtsi embryo heads. The low endogenous expression level of Slc25a21 in wild-type mice was below the threshold where significant changes could be detected.
The phenotypes we observed in homozygous Slc25a21 tm1a(-KOMP)Wtsi mice were broadly consistent with those reported previously in mice carrying hypomorphic Pax9 alleles [26] with the exception that the otitis media and hearing impairment reported in this study have not been assessed previously in Pax9 mutant mice and may represent a novel consequence of Pax9 suppression.

Discussion
Characteristics of 2-oxoadipate acidaemia are absent in mice ablated for Slc25a21 expression We report the generation and phenotypic characterisation of the Slc25a21 allelic series derived from a typical KOMP-CSD allele [3]. Subject to the sensitivity and scope of the phenotyping performed, ablation of Slc25a21 expression was not found to phenocopy the human disease 2-oxoadipate acidaemia (OMIM 204750) for which it is a candidate gene [11]. There are several possible explanations for this. Whilst no hypotonia, ataxia or seizures were detected in Slc25a21 null mice [homozygous for the tm1b (KOMP)Wtsi or tm1d (KOMP)Wtsi alleles], a subset of other clinical symptoms associated with 2-oxoadipate acidaemia, including intellectual disability and urine metabolites, were not assessed by the standard Mouse Genetics Project primary phenotyping pipeline with which these animals were screened [2]. Therefore it cannot be ruled out that additional tests, or analysis performed with increased sensitivity, may reveal subtle disease similarities. Alternatively, mitochondrial transport of metabolites is a complex process involving many carrier proteins. Unknown compensatory mechanisms may be at play in Slc25a21 null mice that prevent the cytoplasmic build-up of oxoadipate and the associated disruption in the catabolism of key amino acids such as lysine [11]. For example, it is known that the substrate specificity of Slc25a21 itself varies between species [11]. Finally,   definitive proof of the candidacy of Slc25a21 for 2-oxoadipate acidaemia is not published and it has been speculated that the disease may be due to defective 2-oxoadipate dehydrogenase, an alternative gene in the same metabolic pathway [11].

Slc25a21 tm1a(KOMP)Wtsi allele affects expression of a neighbouring gene
The Slc25a21 tm1a(KOMP)Wtsi homozygous mice investigated in this study were found to be sub-viable. Those homozygous animals surviving to weaning presented with growth retardation, orofacial abnormalities, brittle, white incisors and hearing impairment. Given the absence of any detectable phenotype in mice homozygous for the Slc25a21 tm1b (KOMP)Wtsi, tm1c (KOMP)Wtsi or tm1d (KOMP)Wtsi derived alleles, we conclude that the phenotypes observed in mice homozygous for Slc25a21 t-m1a(KOMP)Wtsi were not due to ablation of Slc25a21 function, and propose that off-target effects of the tm1a allele affected expression of a neighbouring gene. The spatial, temporal and quantitatively correct activity of a gene requires the presence of not only intact coding sequence but also properly functioning regulatory control. Interference of the expression of neighbouring genes by the presence of selectable marker cassettes or short DNA sequences [5,6,7,8,9] has been reported for genes in close vicinity to the target gene and as far as .100 kb from the targeted gene [6]. Transcriptional silencing of neighbouring genes due to CpG rich sequences within the standard, non-codon optimised, lacZ reporter gene used in this KOMP-CSD allele has also been reported [27]. Furthermore, disruption of cis-regulatory elements has been linked with abnormal phenotypes in both mice [5] and humans [28,29]. In addition, a wide variety of mechanisms which cause disruption of gene expression by a change in its chromosomal environment, while leaving the transcription unit intact, have been identified [29].
At E13.5, RNA sequencing revealed that the only gene within the 1 Mb genomic interval surrounding Slc25a21 that was differentially expressed, compared to wild-type, in homozygous Slc25a21 tm1a(KOMP)Wtsi embryos was Pax9, an observation confirmed by quantitative PCR. Pax9 is located in a 21.06 kb genomic interval directly 39 of Slc25a21 and orientated on the opposite strand. Evolutionarily conserved non-coding sequences are present both up-and down-stream of Pax9, some of which have been demonstrated to induce Pax9 expression [21]. One such conserved region, CNS+6, [21] was located within the homology arm of the Slc25a21 tm1a(KOMP)Wtsi targeting construct, and therefore a candidate for causing altered Pax9 expression. However, sequencing confirmed the integrity of CNS+6 in the Slc25a21 tm1a(KOMP)Wtsi allele. Since homozygous Slc25a21 tm1b(KOMP)Wtsi mice, which still have part of the targeting cassette present (Fig. 6C), were found to be phenotypically normal, interference by the polyA-b-actin::neo cassette is the most likely cause of the reduction of Pax9 expression seen in homozygous Slc25a21 tm1a(KOMP)Wtsi mice. The findings in this study highlight the need to confirm phenotypes found in animals carrying the knockout first conditional ready [tm1a (KOMP)Wtsi or tm1a (EUCOMM)Wtsi] allele by creating and analysing the derived deletion allele. This is a timely reminder, with direct implications on the 3Rs, given the growing resource of , and Slc25a21 tm1d(KOMP)Wtsi mice (C,F) (body weight n = 7, fat mass n = 5) compared to local wild-type controls run on the same week (green symbols) (n$9). Male homozygotes for each of these three alleles and two parameters were also phenotypically normal (data not shown). Mean (6 SD) body weight is plotted against age (A-C). Fat mass is presented as a boxplot, with median, 25 th and 75 th percentile (box), and the lowest and highest data point within 1.56 inter quartile range (whiskers) shown (D-F). The median and 95% reference range (2.5-97.5 th percentiles, dotted lines) for all wildtype mice of the same genetic background and sex (n.80) are displayed on the pale green background. ABR thresholds were normal in homozygous Slc25a21 tm1b(KOMP)Wtsi (G), Slc25a21 tm1c(KOMP)Wtsi (H) and Slc25a21 tm1d(KOMP)Wtsi mice (I) (red symbols) compared to wild-type controls (green symbols). doi:10.1371/journal.pone.0091807.g007 knockout first conditional ready targeted ES cells created by the EUCOMM and KOMP initiatives.

Hypomorphic expression of Pax9 may account for orofacial abnormalities
The Pax family of transcription factors plays a pivotal role in embryonic patterning and disease. The expression of each Pax gene is highly regulated in a temporal and spatial manner [30]. During mouse embryogenesis, Pax9 is expressed mainly in the sclerotome of somites, the pharyngeal pouch endoderm and its derivatives, developing limb buds, and in facial mesenchyme of neural-crest cell origin, including nasal and jaw processes and tooth buds [26,31]. There is strong expression of Pax9 in the medial edge epithelium during the critical time points of palate fusion [32]. In adult mice, Pax9 is expressed in oesophagus and tongue [33]. Looking at published models of Pax9 deficiency, homozygous Pax9 knockout mice die shortly after birth and exhibit a wide range of developmental defects [26] including the absence of teeth. Death has been attributed to cleft palate, affecting the maxillary and palatine shelves, resulting in an inability to suckle and respiratory issues [26]. Two hypomorphic alleles of Pax9 have been published previously [25]. The first, which expressed wildtype Pax9 mRNA at 7% of control levels, presented with hypoplastic upper incisors lacking enamel, missing lower incisors, unilateral or bilateral absence of the third molar in the upper jaw and the second and third molar in the lower jaw, and the lower molars that were present showed severe attrition. In comparison, the hypomorphic allele with 20% of wild-type Pax9 mRNA expression relative to controls presented with less severely affected dentition including relatively normal upper incisors, hypoplastic lower incisors lacking enamel and unilateral or bilateral absence of the upper and lower third molar. We now report homozygous Slc25a21 tm1a(KOMP)Wtsi mice which express wild-type Pax9 mRNA at 34% of control levels and present with an even milder dental phenotype including upper incisors that were commonly maloccluded, which may be due to abnormal snout morphology or the fact that the incisors do not wear against each other. We found fragile, damaged or missing lower incisors which were white or translucent; normally incisors in mice are brownish/yellow due to incorporation of iron-containing pigment in the enamel [34]. The lower incisor root area was severely under-developed. Lower jaw molars showed focal fracturing of the surface dentine and necrosis of the pulp with associated inflammation and bacterial accumulation. The presence of lacZ activity within the cavities of the lower molars of homozygous Slc25a21 tm1a(KOMP)Wtsi mice but not heterozygous mice, confirms that bacterial b-galactosidase is the cause of the staining, not expression of the lacZ reporter gene within the targeted allele. Molars in the upper jaw were less affected. The more severe findings in the lower jaw are consistent with findings in the other Pax9 hypomorphic alleles [25].
Only limited assessment of the palate has been published for the previous Pax9 hypomorphic lines and no abnormalities were reported [25]. In this study, more extensive analysis was performed and whilst the phenotype was significantly less severe than that reported in Pax9 null mice [26], abnormalities were identified, characterised largely by discordance of rugae 3 and an incomplete fusion of the vomer bone to the maxilla.
This gradated allelic series of Pax9 expression levels, extended by the data reported herein, clearly demonstrates that dosage of this gene affects orofacial development and that lower jaw dentition in the mouse is more susceptible to reduced Pax9 expression [25]. This is consistent with humans where dentition is not uniformly affected by a reduced Pax9 expression [35,36,37] and the observation that the minimal Pax9 gene dosage required for the formation of individual teeth varies, the posterior tooth in each tooth family being the most sensitive to a reduction in Pax9 levels.
Otitis media and hearing impairment in mice may be due to disrupted Pax9 expression A reduction in the size of the tympanic ring has previously been described in Pax9 mutant mice [26], however the effect on hearing has not previously been assessed. The homozygous Slc25a21 tm1a(-KOMP)Wtsi mice have a normal sized tympanic ring but display a clear hearing impairment with increased thresholds at all frequencies tested. The moderate hearing impairment seen in homozygous Slc25a21 tm1a(KOMP)Wtsi mice is robust and reproducible across different cohorts of mice, tested at ages from 4 to 26 weeks of age. The degree of impairment in mutant mice is progressive from 4-14 weeks old, with slower progression to 26 weeks of age. The parallel shifts in audiometric profiles across frequencies at the different ages (Fig. 4) are consistent with an underlying conductive hearing loss. Middle ears of the majority of homozygous Slc25a21 tm1a(KOMP)Wtsi mice were filled with fluid of varying viscosity at all ages examined and sections revealed inflammation of the lining of the middle ear. The presence of eosinophilic exudate and the thickening of the mucosa are evident in the homozygous Slc25a21 tm1a(KOMP)Wtsi mice, indicating an on-going inflammatory process. Whilst the aetiology of otitis media is complex [38] and was beyond the scope of this study, the correlation between signs of otitis media and elevated ABR thresholds supports our suggestion that the otitis media is a major contributor to the hearing impairment observed in the young adult homozygous Slc25a21 tm1a(KOMP)Wtsi mice. However, cochlear dysfunction cannot be ruled out especially in the 26 week-old group of both mutants and controls. ABR data from Slc25a21 tm1b(KOMP)Wtsi , Slc25a21 tm1c(KOMP)Wtsi and Slc25a21 tm1d(KOMP)Wtsi homozygotes indicate that hearing thresholds are normal.

Conclusion
In conclusion, ablation of Slc25a21 expression was not found to phenocopy the human disease 2-oxoadipate acidaemia (OMIM 204750) for which it is a candidate gene. The phenotypes observed in mice homozygous for Slc25a21 tm1a(KOMP)Wtsi were not due to ablation of Slc25a21 function, but instead the presence of the selection cassette affected expression of the neighbouring gene, Pax9. The resulting mutant line confirmed and extended the existing knowledge of Pax9 gene dosage effect on orofacial development and represents a novel model of otitis media that may be due to reduced Pax9 expression. Table S1 The sequence of primers used for molecular characterisation. Molecular characterisation of the targeting event was performed using a combination of PCR assays. The 59 to 39 primer sequence along with the expected product size (bp) is presented.

(DOC)
Table S2 Comparison of the outcomes from automatic data evaluation and manual assessment. Comparison of the outcomes from automatic data evaluation and manual assessment for each parameter included in the dataset is presented. Discrepancies between these two methods of assessment are highlighted, and the rationale behind the manual assessment is provided in each instance there was a discrepancy. (DOC)