We distinguished affected mice in the dearisch colony by auditory brainstem response (ABR) threshold measurements. Mice display a range of ABR thresholds to click stimuli, from normal (approximately 15 to 30 dB sound pressure level (SPL)) to moderate hearing impairment (between 50 and 80 dB SPL), with a bimodal distribution (n = 250; Figure 1a). Affected mice were defined as having a click threshold of 50 dB SPL or over, and mice with click thresholds of 30 dB SPL or below were defined as unaffected mice. Measurements of thresholds at a range of frequencies at 12 weeks old showed approximately 40 dB hearing loss across the majority of frequencies in affected mice (Figure 1b). This consistent loss across frequencies, mirroring the shape of the audiogram in unaffected, hearing mice, associated with a hearing loss of rarely more than 40 dB and normal growth of waveform amplitudes and reduction in latencies with increasing stimulus intensity above threshold (Figure 1c, d), are all consistent with conductive pathology as the most likely cause for the hearing impairment.

Repeated ABR testing on a cohort of aging mice demonstrated that affected dearisch mice have hearing impairment from the earliest age tested (3 weeks), and this surprisingly does not generally progress with age (Figure 1e).

Gross anatomy of the inner ear appears normal (Figure 2a-d) and the round and oval window areas are not significantly different between unaffected and affected mice (Student's t-test; P-value 0.24 and 0.86, respectively; data not shown). Ultrastructural anatomy of the cochlea assessed using scanning electron microscopy shows normal sensory hair cell morphology and layout (Figure 2e-j).

However, middle ear examination revealed chronic otitis media with an intact tympanic membrane (Figure 3). Affected mice displayed a variety of pathological features associated with otitis media, including: white bony bulla instead of translucent bone (12 of 14); an abnormally vascularized bulla (5 of 14); a vascularized tympanic membrane (5 of 14); fluid in the middle ear-mostly thick, white, opaque, but not sticky fluid (11 of 14); mucosal oedema (6 of 14); crystalline deposits around the malleus (6 of 14); bony outgrowths that sometimes included fusion of ossicles (9 of 14); and excessive cerumen in the external ear canal (12 of 14). The severity of otitis media was variable and this may account for the variability of the ABR findings. The ABR thresholds did not fluctuate substantially in most individual mice over time (Figure 1c), implying the hearing impairment is due to chronic middle ear disease rather than recurrent acute otitis media. Middle ears of unaffected mice with normal click thresholds were not entirely normal, and showed some abnormal signs, including: a white bony bulla (2 of 14); a vascularized bulla (1 of 14); a vascularized tympanic membrane with engorged capillaries (1 of 14); fluid in the middle ear, either clear or turbid (4 of 14); edema of the middle ear lining (1 of 14); crystalline deposits (4 of 14); bony overgrowths (2 of 14); and cerumen in the external auditory canal (5 of 14). Mild and less frequent pathology in mice with normal thresholds is not entirely unexpected, as the apparent reduced penetrance of the phenotype means some hearing mice will carry the mutated gene and may exhibit some features of otitis media without this being severe enough to compromise ABR thresholds.

Histology of normally hearing mice revealed a single cell thick mucosa lining the middle ear, while in affected mice there was evidence of thickened mucosa with fibrocytes, granulocytes and granulation tissue (Figure 4). This is typical of chronic otitis media. The middle ear cavity of affected mice contained cellular effusion including foamy macrophages and neutrophils, suggesting an acute, possibly infective, otitis media superimposed upon the chronic otitis media. While no unaffected mice grew any bacteria on culture of external and middle ear swabs, two out of four affected mouse middle ears and one out of four of their external ear canals grew Proteus sp. (DJ Pickard, personal communication)

The current dearisch colony is derived from a single male on a C3HeB/FeJ background. This original founder male had mild hearing loss (click threshold 34 dB SPL) on ABR, suggesting variable expressivity of the mutation. When crossed with known wild-type females from the original C3HeB/FeJ background, the male produced some mildly and some moderately affected offspring in the F1 generation, suggesting dominant inheritance. The male was able to produce both affected male and female progeny, suggesting that X-linked inheritance is unlikely. The colony has been outcrossed at least five times to wild-type mice from a C3HeB/FeJ colony that had not been exposed to ENU, diluting out ENU-induced mutations that are unrelated to the dearisch phenotype. There were smaller numbers of affected mice in the colony than could be explained by a simple Mendelian model with full penetrance.

We attempted to map the mutation by outcrossing an affected male to C57BL/6J females, then backcrossing affected outcross offspring to known wild-type C57BL/6J mice. Five affected outcross mice were found out of 168 tested, but when these were backcrossed there were no affected backcross offspring out of 77 tested so we were unable to map the mutation by the usual linkage analysis approach.

We used the Agilent SureSelect ^XT mouse all exon kit for sequence capture followed by Illumina Genome Analyzer II next-generation sequencing to search for the causative mutation using one DNA sample from an affected dearisch mouse and one sample from the C3HeB/FeJ colony (Table 1). Agilent reports 49.6 Mb capture of 221,784 exons from 24,306 genes using this kit 54 . Sequencing reads were mapped to NCBI build 37 of the mouse genome (C57BL/6J) using bwa 0.5.7 55 and duplicate fragments were marked using picard 1.15 56 . SAMtools 0.1.8 57 was used to obtain a list of single nucleotide variants (SNVs) and short insertions and deletions. These were filtered to remove variants found in both wild-type (C3HeB/FeJ) and dearisch mutant sequences, and then to remove variants known to be present in other strains, from dbSNP (build 128 58 ) 59 and from the resequencing of 17 inbred strains 60 (Table 2). Variants were finally filtered on the basis of SNP quality (with a lower limit of 20), mapping quality (with a lower limit of 45) and read depth (with a lower limit of 10). This resulted in approximately 8,000 variants. These were then prioritized on the basis of type and consequence. Those SNVs that were predicted to cause either the gain or loss of a stop codon, that resulted in an amino acid change in the protein or that were within an essential splice site (defined as being in the first or last two base pairs of an intron) were chosen for further analysis. There were 23 SNVs that fitted these criteria (Tables 2 and 3).

Of the 23 variants of interest, all were autosomal and 14 were present as heterozygotes, consistent with the expected autosomal dominant pattern of inheritance. All 23 variants were analyzed further by capillary sequencing using the original two DNA samples, which resulted in exclusion of most of the variants as false positive variant calls on the basis that the DNA sample from the mutant DNA was identical to that of the wild-type C3HeB/FeJ DNA at that position (Table 3). The high number of false positives is due partly to the presence of small inserts or deletions causing the SAMtools SNP caller to misread SNVs either side of the indel. Most of the other false positives can be seen to have low consensus and/or SNP quality scores for either or both dearisch and C3HeB/FeJ sequences; SNVs were not filtered on consensus score at all, and only lightly on SNP quality score, because we preferred false positives to false negatives. Only one SNV has high consensus quality, SNP quality, mapping quality and read depth scores, and this has been found by capillary sequencing to be a correct call. This SNV is a point mutation in Isl1 leading to a T to C base pair transition at position MMU13:117098488 causing a substitution of tyrosine by cysteine (Y71C; Figure 5a, b). This missense mutation affects an amino acid within the first LIM domain of Isl1.

Capillary sequencing of this position in 21 wild-type strains and in 5 individual C3HeB/FeJ wild-type mice reveals that all are homozygous (T/T) for the reference allele. Indeed, this T to C transition in dearisch mutants alters a tyrosine residue that is highly conserved in orthologous proteins in other species (Figure 5c, d). Having detected this promising candidate mutation, we sequenced DNA samples from throughout the dearisch colony. All 28 affected dearisch mice (born between 2009 and 2011) were heterozygotes (T/C). All of the mice with thresholds above 50 dB SPL were found to have one copy of the Isl1 mutation (Table 4). Of the offspring of known heterozygote by heterozygote matings, no pups out of 111 were detected as homozygous for the Isl1 mutation, suggesting severely reduced homozygote viability. The penetrance of raised ABR thresholds (> 50 dB SPL) in known heterozygotes is 23.1%. Interestingly, most of the mice with ABR click thresholds of 30 to 50 dB SPL were also heterozygous for the dearisch Isl1 mutation (Table 4; Figure 6), giving a penetrance of 51.2% if the more mildly affected mice are included. Furthermore, most of the 'unaffected' mice with thresholds of 30 dB SPL or less but with signs of subclinical middle ear inflammation mentioned earlier were found to be carriers of the Isl1^Drsh mutation (data not shown).

The close linkage of the Isl1 variant with the otitis media phenotype is strong support for this being the causative mutation. However, it remains a possibility that the Isl1 variant is simply a linked marker. In order to exclude linkage between the Isl1 mutation and any other potentially causative mutation, it is important to exclude other mutations on chromosome 13 (Table 5). Of the 23 SNVs (non-synonymous, stop gained and splice site mutations) identified by exome sequencing, the Isl1 mutation is the only one on chromosome 13 (Table 3). Four other chromosome 13 SNVs were excluded at the final filtering step, one in a noncoding transcript of Tpmt, one in the 5' UTR of Smad5 and two in the 3' UTRs of the genes Histh1a and Sdha, the closest of which is 70 Mb from the Isl1 mutation. We also examined indels from chromosome 13. The SAMtools variant caller identifies short indels as well as SNVs, and these indels were not included in the final analysis of 23 variants. Thirteen deletions and twelve insertions were identified on chromosome 13, although only one and five, respectively, were within coding regions. Of the insertions and deletions within 10 Mb of Isl1, none were within coding regions.

We next asked if Isl1 protein is expressed in the middle ear. Immunohistochemistry of the adult wild-type middle ear revealed clear, widespread expression of Isl1 within the single cell mucosal lining of the middle ear cavity, including the single cell layer covering the ossicles, but less pronounced on the inner surface of the tympanic membrane (Figure 5e, f). Expression is also seen in the epithelial layer of the external ear canal and outer layer of the tympanic membrane. At postnatal day 4, the expression is more diffuse but is present in the immature mucosa where the middle ear has cavitated and in the outer cellular layer surrounding the ossicles (Figure 5g).

According to Pfam 61 , the Isl1 protein consists of four Pfam domains: two LIM domains, a homeodomain and a Gln-rich domain. Each LIM domain contains two zinc fingers, which each bind a zinc atom. The LIM-homeodomain (LIM-HD) combination is thought to represent a 'LIM code' that governs transcriptional regulation in the control of cell type specification in different tissues and organs 62 . Isl1 is a member of the LIM-HD family of proteins. The two LIM domains are responsible for interaction with other proteins while the homeodomain uses its helix-turn-helix motif to bind DNA sequences containing the sequence 5'-ATTA-3' and so initiate transcription of the appropriate genes.

Proteins binding to LIM-HD proteins do so via a LIM-interaction domain (LID), which consists of around 30 residues. The Y71C mutation is located within the first LIM domain and so may affect the strength of this binding. To predict how it might do so requires knowledge of the protein's three-dimensional structure.

To date, there have been no experimental determinations of the three-dimensional structure of Isl1 protein (other than fragments of the carboxy-terminal domain). However, there are many structural models of related proteins in the Protein Data Bank (PDB) 63 . One of these, PDB entry 2xjy, is of particular interest. This is a structural model, solved by X-ray crystallography to 2.4 Å resolution, of human rhombotin-2 (aka LMO2). The protein is a LIM-only (LMO) protein; that is, it consists of two LIM domains only. However, the structure is a complex between this protein and a 35-residue fragment of a LID from human LIM domain-binding protein 1. As such, it provides a general idea of how LIM domains recognize their interaction partner. The three-dimensional structure reveals that the LID fragment binds in an extended conformation along a groove running along the length of the two LIM domains.

Thus, to help understand the structural effects of the Y71C mutation, we built a homology model for Isl1, using the rhombotin-2 protein from PDB entry 2xjy as a template. The sequence identity of the two LIM domains in the two proteins is 34% over 126 residues, giving an E-value of 9 × 10^-32, so the model is expected to be a good approximation of the structure of Isl1. Figure 7 shows the model, with the LID from PDB entry 2xjy retained to show the interactions that one might expect between the LIM domains of Isl1 and the LIDs of the protein(s) they bind to. Of particular interest is Tyr71. The equivalent of this residue in the PDB 2xjy structure is Tyr84. This makes a side chain-side chain hydrogen bond with Asp354 in the LID of the partner protein. It turns out to be the only side chain-side chain hydrogen bonded interaction across the interface between the two proteins. In all, 12 pairs of residues interact via hydrogen bonds across this interface and all but the Tyr84-Asp354 interaction are hydrogen bonds between main chain atoms. So mutations to any of these other residues are far less likely to disrupt the binding of the two proteins. Indeed, it seems to be a feature of the LID-LIM interface that it is particularly tolerant to mutation 64 . The exception would appear to be the Tyr84-Asp354 interaction.

We propose that the Isl1 Y71C mutation leads to the predisposition of heterozygotes to develop otitis media, for several reasons. Following exome resequencing, the Isl1 variant was the only candidate that was confirmed by capillary sequencing. The tyrosine residue at this location is highly conserved among many species and in other mouse strains. The Isl1 mutation segregates with the phenotype, with all affected mice carrying the mutation in heterozygote form. No other likely pathogenic DNA changes linked to Isl1 on chromosome 13 were identified. Isl1 is expressed in the middle ear mucosa of wild-type mice. Finally, three-dimensional modeling of LIM domain interactions pinpoints the amino acid altered by this mutation as being particularly important in protein-protein interactions. As it was not possible to map the locus of the causative gene in dearisch using traditional backcross matings due to the low penetrance of the phenotype, exome resequencing has proved to be invaluable in identifying the likely causative mutation.

Isl1 is a transcription factor that acts as an insulin enhancing gene 65 . It contains two LIM domains and one carboxy-terminal homeodomain involved in protein-protein and protein-DNA interactions. Our modeling suggests this protein-protein interaction is likely to have been interrupted by the mutation we discovered in dearisch mutants. Isl1 has one isoform in mice and seven isoforms in humans and is located on chromosomes 13 and 5, respectively. Several mouse mutations affecting Isl1 exist, and the most widely studied is the Isl1^tm1Tmj allele 66 , which consists of a neo cassette insertion into the DNA sequence encoding the second LIM domain. Mice with this mutation are homozygote lethal at embryonic day (E)11.5. Dearisch also appears to be homozygote lethal, although the age and cause for this has yet to be determined. Of four embryos so far harvested from dearisch heterozygote by heterozygote matings at E9.5, one has been genotyped as a homozygote. This pup looked immature and abnormal on external inspection (data not shown). Homozygotes of Isl1^tm1Tmj exhibit malformed vasculature, including the dorsal aorta, foregut and pancreatic malformations, and exhibit no motor neuron development. Heterozygote carriers of Isl1^tm1Tmj have not been reported to have any middle ear or inflammatory defects. However, Isl1 is expressed in both immature cochlear hair cells and in auditory neurons 67 . Over-expression of Isl1 results in protection of neurons from age-related and noise-induced hearing loss 68 . No electrophysiological studies of inner ear function in Isl1 mutants have been previously reported. Surprisingly, despite evidence of widespread neuronal irregularities in Isl1 knockout mice and the known expression of Isl1 within the inner ear, no evidence of sensorineural abnormalities were detected in the affected dearisch mice. This suggests that one copy of the wild-type Isl1 allele is sufficient for normal development of auditory neurons and hair cells.

Prior to this study, Isl1 expression in the middle ear had not previously been reported. However, Isl1 expression has been documented within other mucosal epithelial linings. Expression of Isl1 is strong in the ultimobranchial epithelium of the pharynx at embryonic stages 69 , and was increased in specification of lung bud at E8.5 to E9.5 70 . Isl1 expression has been found in somatostatin-expressing cells of the gastric mucosa in juvenile rats, suggesting that Isl1 may have a role in regulating somatostatin gene expression 71 . In the lungs, somatostatin is known to decrease substance-P-related mucous secretion from submucosal glands 72 . This suggests that Isl1 may affect mucous secretion from the mucosa through effects on somatostatin. We found that Isl1 is expressed in the wild-type adult middle ear mucosa. This might be expected, as middle ear mucosa is often described as being a respiratory-type mucosa. Through secretion of protective factors such as lactoferrin, lysozyme and mucus, the middle ear mucosa can reduce risk of infection 73 . Isl1 may contribute to predisposition to otitis media by affecting the constituents, amount or protective nature of middle ear mucosal secretions.

The innate immune system offers non-specific immediate defense against infection. The cytokines form part of this system, recruiting immune cells and initiating or reducing inflammation by acting as chemical mediators to specific genetic pathways. Interleukin 6 is one such cytokine. It binds the gp130 component of the type 1 cytokine receptor complex, resulting in activation of the receptor, which initiates intracellular signaling. JAK1 and STAT3 are known to be activated by this process 74 . The JAK-STAT pathway is involved in acute phase response and chronic inflammation in a variety of tissues, including the lungs and gut 75 . Isl1 has been shown to physically interact with both JAK1 and STAT3, forming a complex in both human and monkey immortal cell lines 76 . This results in the activation of STAT3, which acts as an important signal transducer and activator of transcription. JAK1 is also activated and is able to dock and recruit further signaling proteins. STAT3 has been shown to be necessary for lung and bladder epithelium to respond effectively to Gram-negative bacteria 77 78 . Without Isl1 the function of both of these genes in the prevention of infection or inflammation via innate immunity is potentially disrupted. Like Isl1 mutants and dearisch, Stat3 knockout mice suffer from embryonic lethality, while Jak1 knockout in mice results in perinatal mortality 79 .

The importance of innate immunity in reducing otitis media is already well documented. For example, toll like receptors (Tlrs) recognize bacterial endotoxin, stimulating TNFα production, which in turn affects production of immunoglobulins, cytokines and mucin 46 . Mice that are genetically deficient for Tlr4, such as the C3H/HeJ inbred strain, develop chronic otitis media due to an inability to clear Gram-negative bacteria 32 . Between 35% and 60% of these mice were affected by otitis media at some point during their life span. Unlike affected dearisch mice, there was also evidence of bony remodeling of the round window and elements of inner ear inflammation in some C3H/HeJ mice. Knockout of gp130 suggests that the Tlr4 31 32 pathway response to bacterial endotoxin may be modulated by the Stat3 pathway 80 . The role of Isl1 in innate immunity is yet to be fully elucidated, but wild-type Isl1 levels in heterozygote dearisch mice may be sufficiently low to reduce their ability to clear bacteria from the middle ear.

In humans, several rare point mutations in ISL1 have been shown to lead to maturity onset diabetes of the young 81 . An increased incidence of otitis media has not been reported in people with mutations of this gene but a general increased propensity to infection is well-recognized in diabetics. Otitis media is very common and therefore an increased prevalence of otitis media in these patients may have gone unnoticed.

The founder mouse was detected in a large-scale ENU mutagenesis program aimed at detecting new dominantly inherited phenotypes 9 . Impaired hearing was detected by screening for lack of an earflick (Preyer reflex) in response to a 20 kHz calibrated sound burst using a custom-made clickbox. Mice that did not respond were studied in more detail. Affected dearisch mice (also known as DEA2) appeared to lose their Preyer reflex from several months of age. The colony was managed by mating affected mice with unaffected littermates, and the line was maintained on the original genetic background of the mutagenized males, C3HeB/FeJ.

ABRs were measured with recovery anesthesia using three scalp electrodes 82 . Responses were recorded to broadband clicks and tone bursts at 3, 6, 12, 18, 24, 30, 36 and 42 kHz and at a wide range of intensities from 10 to 97 dB SPL in 3 dB steps. Thresholds were determined using a stack of response waveforms and identifying the lowest stimulus at which an identifiable waveform occurs. This ABR protocol was performed on 9 mice at single time points and 16 mice at 4-weekly intervals from 8 to 24 weeks. Input/output functions were then calculated using amplitude and latency of P1/N1 and P4/N4 waveform components plotted with respect to stimulus intensity (n = 13 affected and 13 unaffected mice at 3 to 15 weeks of age).

A short ABR protocol taking approximately 6 minutes per mouse, consisting of a 70 dB SPL test click, broadband clicks from 10 to 97 dB in 3 dB steps followed by a further test 70 dB SPL click and anesthetic recovery, was used to screen the entire colony (n = 85). Following this, all mice born underwent this short ABR protocol at 6 to 8 weeks of age (n = 348 in total) and the results used to plan matings. However, only mice born between 2009 and 2011 (n = 250) were analyzed for constructing frequency histograms to avoid bias due to selective retention of affected mice born before 2009. Mice with thresholds above 50 dB were defined as affected based on the population distribution of click thresholds shown in Figure 1a. This short ABR protocol was used to assess younger mice from the dearisch colony at 3, 6 and 8 weeks of age (n = 66, with 35 mice undergoing single recordings and 31 undergoing repeated ABR measurements).

An outcross was performed with an affected dearisch male and a female from the C57BL/6J inbred colony. F1 offspring (n = 168) were screened using the short ABR protocol. Affected F1 mice were mated with another C57BL/6J mouse to create a backcross. The backcross offspring from these matings (n = 77) were screened using the short ABR protocol.

A pedigree was drawn up using information from ABR tests over several generations of the dearisch colony. This has been combined with data from Isl1 genotyping.

Inner ear clearing was performed using glycerol as described previously 83 (n = 5 affected and 5 unaffected littermate controls, aged 15 months). Round and oval window measurements were taken from images of cleared inner ears using Adobe Photoshop. Each measurement was performed four times and averaged. Scanning electron microscopy (n = 3 affected and 3 unaffected littermate controls, aged 2 months) was performed following fixation in 2.5% glutaraldehyde, a standard osmium-thiocarbohydrazide-osmium OTOTO protocol, dehydration, critical point drying and examination in a Hitachi S-4800 scanning electron microscope.

Middle ear dissections were performed on fresh tissue (n = 14 affected and 14 unaffected littermate controls, aged 9.3 to 24.0 months-mean 16.8 months, standard deviation 4.2 months) and observations were recorded on a standard tick sheet. First the tympanic membrane was inspected, the tissue covering the bulla was dissected away and the bulla inspected. The bulla was carefully removed and the tympanic membrane inspected a second time. The tympanic membrane was removed and the presence of fluid, inflamed mucosa or debris recorded. The malleus, incus and stapes were removed and photographed before removing the inner ear for clearing.

For histology, half heads of mice were fixed in 10% formalin and decalcified using EDTA for 10 days. Following alcohol dehydration the half heads were embedded in paraffin wax, sectioned to 8 μm and stained according to a standard hematoxylin/eosin protocol (n = 4 affected and 4 unaffected littermate controls, aged 6 months). Isl1 expression was inspected on sections from the same mice (n = 3 littermate controls, aged 6 months) using Isl1 antibody (AbCam: 20670, Cambridge, Cambridgeshire, UK) according to the immunohistochemistry protocol described previously 84 . Postnatal day 4 pups were also used for immunohistochemistry (n = 4 wild types), but no decalcification step was required.

Swabs from the outer and middle ear of affected and litter mate controls (n = 4 affected and 2 unaffected littermate controls, aged > 15 months) were firstly grown on nutrient broth and on L-agar plates (Oxoid Ltd, Basingstoke, UK). The bacteria were identified by plating on selective media that included CLED, MaConkey's and UTI brilliance agar (Oxoid Ltd). Oxidase testing was used as a final confirmatory step.

One deaf dearisch mouse and one mouse of the original background C3HeB/FeJ were used for exome sequencing using a pre-market product from Agilent (Agilent SureSelect^XT mouse all exon kit for exome sequence capture). This uses 55,000 biotinylated cRNAs to identify the exome and surrounding intronic and intergenic information, including microRNAs. Magnetic beads are then used to pull-down the relevant DNA. Remaining DNA is washed away and the cRNA digested.

DNA (2 to 3 μg in TE) was sheared to 100 to 400 bp using a Covaris E210 (Covaris, Woburn, MA, USA). Sheared DNA was subjected to Illumina paired-end DNA library preparation according to manufacturer's recommendations (NEBNext DNA Sample Prep Set 1; New England BioLabs, Ipswich, MA, USA) and the adapter-ligated libraries were amplified for five to six cycles using Herculase II (Agilent Technologies) with PE1.0 and PE2.0 oligonucleotides (Illumina, San Diego, CA, USA). Amplified library (500 ng) was hybridized to the mouse bait library (SureSelect^XT Mouse All Exon Kit; Agilent Technologies, catalogue number G7500A) according to the manufacturer's recommendations. Hybridized material was captured using streptavidin-coated beads (Invitrogen, Paisley, UK) and amplified for 10 to 11 cycles using Herculase II with PE1.0 and PE2.0 oligonucleotides (Illumina). Captured libraries were sequenced on the Illumina Genome Analyzer II platform as paired-end 76-bp reads according to the manufacturer's protocol. Two lanes of sequence were generated for each mouse.

Sequence data have been deposited in the European Nucleotide Archive (accession number ERP000744).

Primers to amplify the regions containing the potential DNA changes detected by Illumina sequencing were designed using Primer 3 85 and ordered from Sigma, (Haverhill, Suffolk, UK) (Table 6). DNA from the original sequenced mice underwent capillary sequencing to exclude false positives. The same Isl1 primer and capillary sequencing were used to assess mice from the rest of the dearisch colony and other C3HeB/FeJ mice. Indels and SNVs originally excluded by the final filtering step before capillary sequencing were examined along the entire length of chromosome 13 to exclude any potentially pathogenic mutation that may be linked to Isl1. The following wild-type mouse strains were also sequenced to establish the Isl1 sequence: NOR/Lt, BUB/BnJ, I/LnJ, C3HeB/FeJ, FVB/N, 129P2/OlaHsd, CBA, PL/J, 101/H, C57BL/6J, SWR/J, P/J, BALB/c, LG/J, CHMU/LeJ, MA/MyJ, SB/Le, PN/nBSwUmabJ, DBA/1J, DA/HuSn, and SM/J.

We used the PDBsum database 86 to find all structural models containing one or more LIM domains (Pfam identifier PF00412), and then examined those having two tandem LIM domains to find any that might be in complex with a binding partner. One such was PDB entry 2xjy, solved by X-ray crystallography to 2.4 Å resolution. This is a complex of human rhombotin-2 (aka LMO2) and a 35-residue fragment of a LIM-interaction domain (LID) from human LIM domain-binding protein 1.

We used the SWISS-MODEL server 87 to build automatically a three-dimensional homology model of ISL1 using the three-dimensional structure of rhombotin-2 from PDB entry 2xjy as our template. The sequence identity of the two LIM domains in the two proteins is 34% over 126 residues, giving an E-value of 9 × 10^-32, so the model is expected to be a good approximation of the structure of Isl1. To our model we added the LID fragment from PDB entry 2xjy (by cut-and-paste between PDB files), and noted that the Tyr84-Asp354 side chain interaction from 2xjy was retained as Tyr81-Asp354 in our model.