MBTPS2 mutations cause defective regulated intramembrane proteolysis in X-linked osteogenesis imperfecta

Osteogenesis imperfecta (OI) is a collagen-related bone dysplasia. We identified an X-linked recessive form of OI caused by defects in MBTPS2, which encodes site-2 metalloprotease (S2P). MBTPS2 missense mutations in two independent kindreds with moderate/severe OI cause substitutions at highly conserved S2P residues. Mutant S2P has normal stability, but impaired functioning in regulated intramembrane proteolysis (RIP) of OASIS, ATF6 and SREBP transcription factors, consistent with decreased proband secretion of type I collagen. Further, hydroxylation of the collagen lysine residue (K87) critical for crosslinking is reduced in proband bone tissue, consistent with decreased lysyl hydroxylase 1 in proband osteoblasts. Reduced collagen crosslinks presumptively undermine bone strength. Also, proband osteoblasts have broadly defective differentiation. These mutations provide evidence that RIP plays a fundamental role in normal bone development.

O steogenesis imperfecta (OI), or 'brittle bone disease', is a well-known heritable skeletal dysplasia. Affected individuals have low bone mass and increased fracture susceptibility. OI is genetically heterogeneous. Most cases have autosomal-dominant mutations in the type I collagen genes, COL1A1 or COL1A2 (ref. 1). More recently, multiple genes causing mostly autosomal recessive forms of OI were delineated, generating a new paradigm for OI as a collagen-related disorder. The products of these genes (CRTAP, LEPRE1, PPIB, FKBP10, PLOD2, SERPINH1, BMP1, SERPINF1, IFITM5 and SPARC) interact with collagen post-translationally for prolyl 3-hydroxylation, folding, processing, fibrillogenesis, cross-linking and/or mineralization 1,2 . No cases of X-linked OI have been reported.
Regulated intramembrane proteolysis (RIP) involves cleavage of membrane-spanning regulatory proteins by proteases within the plane of the membrane 3 . The three intramembrane cleaving protease (iCLiP) families include serine proteases, site-2 metalloprotease (S2P) and aspartyl proteases. The best-described RIP-mediated signalling in humans involves sequential cleavage of diverse substrates by site-1 protease (S1P), a serine protease encoded by membrane-bound transcription factor peptidase, site 1 (MBTPS1), and S2P, encoded by membrane-bound transcription factor peptidase, site 2 (MBTPS2). S1P and S2P are located in the Golgi membrane. They cleave regulatory proteins transported from the ER membrane in times of ER stress or decreased sterol metabolites, releasing mature N-terminal fragments that shuttle to the nucleus and activate gene transcription. These substrates include old astrocyte specifically induced substance (OASIS)/cAMP responsive element binding protein 3-like 1 (CREB3L1), a transcription factor expressed in astrocytes and osteoblasts, activating transcription factor 6 (ATF6), a component of the unfolded protein response pathway and sterol regulatory element binding protein (SREBP), involved in cholesterol synthesis [4][5][6] . MBTPS2 missense mutations were previously identified in the related dermatological conditions IFAP/BRESHECK (ichthyosis follicularis, atrichia, and photophobia, OMIM #308205) 7,8 and keratosis follicularis spinulosa decalvans (KFSD, X-linked, OMIM #308800) 9 .
In this report, we present two independent OI pedigrees with an X-linked inheritance pattern, and without symptoms of IFAP/ KFSD. Linkage analysis and next-generation sequencing (NGS) identified a novel MBTPS2 missense mutation in each pedigree. The resulting substitutions are in or near the S2P NPDG motif crucial for metal ion coordination 10 . Mutant S2P protein is stable but cleavage or activation of S2P substrates OASIS and ATF6, respectively, is impaired, consistent with reduced proband collagen secretion. Osteoblasts with mutant S2P have defective differentiation, and bone tissue collagen from one proband has decreased hydroxylation of the critical crosslinking lysine residue. These X-linked OI-causing mutations extend RIP functioning to normal bone formation.

Results
Clinical features. Family I. From an extended Thai OI pedigree, 6 of 12 affected males were examined (Fig. 1a,b and Supplementary Table 1). The proband (P1/I; V7) had prenatal fractures of ribs and long bones (Fig. 2a). At 2 years, his physical exam was notable for moderate short stature, blue sclerae, pectus carinatum and bowing of lower extremity long bones. His L2-L4 bone mineral density z-score ¼ À 4.7 (0.315 g cm À 2 ). Affected adult males had fractures beginning during gestation, short stature, white sclerae, variable scoliosis and pectal deformity, striking tibial anterior angulation and generalized osteopenia (see Case Reports).
Family II. A non-consanguineous German pedigree has two affected males, the 26-year-old proband (P1/II) and his 68-year-old maternal uncle (P2/II) ( Fig. 2c; Supplementary  Fig. 2a and Supplementary Table 1). As a newborn, the proband was noted to have bowing of humeri, radii and tibiae, and fractures of femora, ribs and clavicles, as well as generalized osteopenia ( Fig. 2b and Supplementary Fig. 2b). Both patients had numerous fractures of upper and lower extremity long bones during childhood, but fractures declined post puberty. Both have white sclerae, kyphoscoliosis with anterior vertebral wedging and short stature, with final height of 5-6-year-old males. The uncle has severe pectus excavatum.
None of the affected individuals had ichthyosis follicularis, alopecia, photophobia, intellectual disability or seizures.
Mutation identification. Sequencing of a set of genes involved in OI and bone fragility (ALPL, BMP1, COL1A1, COL1A2, CREB3L1, CRTAP, FKBP10, IFITM5, LEPRE1, LRP5, PLOD2, PLS3, PPIB, SERPINF1, SERPINH1, SP7, TMEM38B and WNT1) did not reveal functional changes in the probands of either pedigree. Each proband had a neutral COL1 single-nucleotide polymorphism (SNP) that was inherited from his healthy father ( Supplementary Fig. 3). Interestingly, the urinary lysyl pyridinoline (LP) to hydroxylysyl pyridinoline (HP) ratio, which has been suggested as a diagnostic marker for OI 11 , was markedly elevated in an adult from Family I and two adults from Family II (0.305-0.450 patients versus controls 0.211±0.008) and in the index case from Family I (0.319 versus paediatric controls 0.208±0.030) (Supplementary Table 2). The LP/HP ratio was not elevated in a carrier from each OI Family, or in a case of IFAP. The elevated LP/HP ratio is similar to findings in a mouse model of type IX OI 12 , although X-OI probands have normal PPIB sequences.
X-chromosome inactivation studies of five obligate carriers from Family I showed a skewed pattern of inactivation ( Supplementary Fig. 4). Linkage analysis of the X-chromosome in Family I defined a 21.6-Mb critical region on Xp22 from markers DXS7108 to DXS1067 (LOD score 3.31) (Fig. 1b and  Supplementary Table 3). Comparative genomic hybridization (CGH) analysis did not reveal imbalances in the critical region, which was then sequenced by NGS in affected male IV17 and unaffected male IV10. After excluding known SNPs, and variants present in the unaffected relative, two candidates remained. A mutation in MBTPS2 exon 11, c.1376A4G (nucleotide: NM_015884.3, GRCh38) predicting a p.N459S substitution (protein: NP_056968.1), was confirmed by Sanger sequencing of proband DNA (Fig. 2c), segregated with OI in the pedigree (Fig. 1b) and was absent in 644 X chromosomes of unrelated Thai controls (181 females, 282 males). A FAM48B1 variant, c.532G4A (p.V178M), occurred in 4 of 22 alleles from Thai controls (Supplementary Table 4).
In Family II, SNP-genome-wide linkage analysis supported X-linked recessive inheritance, despite normal X-inactivation (0.70) in the mother of P1/II, with a positive LOD score in a 30-Mb region of chromosome X between rs11094708 and rs5906168 (Supplementary Fig. 1a and Supplementary Table 5). In view of the modest positive LOD score, the entire X-exome was sequenced by NGS, resulting in two variants with population frequencies below 0.5% within the 30-Mb critical region. The first variant, a hemizygous c.1778G4A (p.R593H) missense SNP in CCDC120, was detected in proband P1/II but not in his affected uncle (P2/II). The second variant, a hemizygous c.1515G4C mutation in MBTPS2 (p.L505F; Supplementary Fig. 1b), co-segregated with the phenotype and was also found in a heterozygous state in the proband's mother (Fig. 2d).
Multiple programs (Polyphen2, SIFT, Provean, Mutation-Taster) predict both MBTPS2 (S2P) variants p.N459S and p.L505F as pathogenic since they are likely to interfere with function (Supplementary Table 6). Residues N459 and L505 are highly conserved in S2P, as identified by ClustalX (Fig. 2e) Table 6). N459 is the first residue of the NPDG motif, which is important for metal ion coordination 10 .
Effect of MBTPS2 mutations on transcripts and protein.
Effect of mutant S2P on RIP functions. We examined the impact of S2P substitutions on activation of RIP substrates OASIS, ATF6 and SREBP (Fig. 4a-c). On western blot, mature intracellular OASIS cleavage fragment (50 kDa, S1P/S2P cleaved) was decreased, while the 55-kDa cleavage product (S1P cleaved) was increased, in fibroblasts with each mutation and in S2P p.L505F osteoblasts, indicating S2P cleavage impairment in OI probands comparable to IFAP and KFSD cells (Fig. 4a). Treatment with ALLN (N-acetyl-leucyl-leucyl-norleucinal) to prevent degradation of OASIS cleavage products increased the total OASIS detected in all fibroblast and osteoblast cell lines, as well as the relative proportion of fully cleaved OASIS in normal control, IFAP and KFSD fibroblasts and normal control osteoblasts, more than in OI cells (Fig. 4a). Treatment with tunicamycin to induce ER stress had no further effect on OASIS cleavage.    Effect of defective S2P on type I collagen biochemistry. Since OI is a collagen-related bone dysplasia and fully processed OASIS is reported to activate Col1a1 transcription 13 , we examined type I collagen biochemistry in fibroblasts and osteoblasts with mutant S2P. Collagen transcripts were variable in mutant cells, with decreased COL1A1 transcripts detected in fibroblasts of OI proband 1/I (p.N459S), IFAP (p.R429H) and KFSD (p.N508S) fibroblasts, but not in OI fibroblasts or osteoblasts with the S2P p.L505F mutation (Fig. 5a).
Fibroblasts from OI probands secreted a significantly reduced amount of type I collagen (20-73% of control), as did IFAP and KFSD fibroblasts (E40%) (Fig. 5b). Proband osteoblasts were insufficient for this assay. Electrophoretic migration of proband collagen ( Fig. 5c) was normal, consistent with normal thermal stability (T m ) and helical resistance to proteolytic digestion at physiologic temperature ( Supplementary Fig. 5a,b). Collagen deposited into matrix in culture by S2P p.N459S fibroblasts from P1/I had a decreased proportion of collagen containing mature cross-links (10% versus 22% of collagen deposited by normal control fibroblasts) ( Supplementary Fig. 5c), suggesting that hydroxylation of the type I collagen K87 residue crucial to crosslinking may be impaired.
Effect of mutant S2P on bone tissue and cells. Tissue and osteoblast characteristics of OI were demonstrated in a bone sample from P1/II (p.L505F). Osteoblasts differentiated in culture revealed impaired expression of transcripts related to osteoblast maturation and RIP pathways (Fig. 6). Alkaline phosphatase (ALPL) expression is about half of control in mid-to late differentiation. OASIS (CREB3L1) expression was significantly reduced throughout osteoblast differentiation. Mature OASIS normally complexes with SMAD4 to upregulate expression of matrix genes 14 . However, SMAD4 transcripts were reduced throughout p.L505F osteoblast differentiation, underlying the reduced expression of the proteoglycan matrilin-1 (ref. 15). Interestingly, COL1A1 expression in P1/II osteoblasts was comparable to control (Fig. 5a).

Discussion
We identified an X-linked form of osteogenesis imperfecta in two independent pedigrees. Phenotypic inheritance pattern, linkage analysis and NGS were used to localize the causative gene in each family to MBTPS2 at Xp22. The gene defects co-segregate with the phenotype in each pedigree and all affected individuals are the sons of obligate carriers. This gene identification extends the inheritance pattern of OI to include X-linked recessive. Notably, the OI phenotype of these MBTPS2 defects is distinct from that of PLS3 mutations, which cause an X-linked form of osteoporosis that is mostly apparent in middle-aged adults, but does cause fractures in some children 17   MBTPS2 encodes S2P, an integral membrane protein critical to RIP and essential for cholesterol metabolism 18 . Mutations previously reported to cause substitutions in S2P ( Supplementary  Fig. 8) underlie a spectrum of dermatological conditions: IFAP syndrome, characterized by a triad of ichthyosis follicularis, atrichia and photophobia 7 , KFSD 9 , BRESHECK syndrome 8 and Olmsted syndrome 19 . The affected individuals in our OI pedigrees do not have any symptoms associated with these conditions, even into adulthood. Both OI-causing substitutions in S2P are located in highly conserved intramembrane residues near the carboxyl end of the protein, and likely contribute to a specialized function. In Family I, the MBTPS2 missense mutation results in a p.N459S substitution at the first residue of the S2P NPDG motif, which coordinates the zinc metal ion critical to protease catalytic function 10 . In addition, the serine substitution at this site introduces a potential O-GlcNAc glycosylation site (http://www.cbs.dtu.dk/services/YinOYang/), and addition of a bulky group might interfere with S2P interaction with its substrates. The Family II MBTPS2 mutation results in S2P p.L505F. This substitution, in which leucine is replaced with the bulkier phenylalanine, is predicted to be pathogenic by multiple mutation analysis programs and may also interfere with S2P substrate interactions. Interestingly, the nearby p.N508S and p.N508T have a mild IFAP and KFSD phenotype 9 , and most IFAP mutations are located adjacent to this critical region (p.R429-N508) 20 .
S2P and S1P span the Golgi membrane and function sequentially in RIP to process crucial regulatory substrates transported from the ER membrane in times of ER-stress or sterol restriction, including OASIS, ATF6 and SREBP (Fig. 7c) 3,5,13,21 . The mature processed proteins then enter the nucleus where they activate pathways involved in cholesterol homoeostasis, unfolded protein response and bone formation. Our demonstration of S2P functioning in human bone development is supported by prior studies on defects of RIP components in animal models. Both S1P and S2P knockdown zebrafish showed abnormal cartilage development, independent of lipid defects 22 . Cartilage-specific S1P knockout mice (S1P cko ) have a lethal chondrodystrophy, with loss of hypertrophic chondrocytes and absent endochondral bone formation 23 . S1P cko chondrocytes also display differential downregulation of transcripts related to cholesterol and lipid biosynthesis 24 . In addition, ablation of the RIP substrate OASIS, a bZIP transcription factor, causes spontaneous fractures and severe osteopenia in mice 13 and homozygous deletion of CREB3L1 (OASIS) causes recessive OI in one family 25 .
The MBTPS2 mutations causing X-linked OI do not impair MBTPS2 expression or destabilize S2P protein, nor do MBTPS2 mutations causing IFAP/KFSD. Instead, they disrupt S2P processing of RIP substrates, interfering with their downstream functions. First, both p.N459S and p.L505F fibroblasts and p.L505F osteoblasts display impaired OASIS cleavage. OASIS/CREB3L1 expression is decreased in p.L505F osteoblasts, as is the expression of its nuclear binding target SMAD4 (ref. 14), which, along with the reduced expression of osteoblast maturation marker alkaline phosphatase, indicates a broad defect in osteoblast differentiation. SMAD4 in turn co-activates transcription of matrix proteins such as matrilin-1 (ref. 15), consistent with decreased expression of MATN1 in OI proband P1/II's osteoblasts. Second, type I collagen secretion is decreased in fibroblasts of all OI probands. The combination of normal collagen expression and reduced collagen secretion found in mutant MBTPS2 fibroblasts differs from the expected decreased SMAD4 transcriptional activation of Col1a1 (ref. 14), as well as from the murine Oasis knockout, in which reduced Col1a1 expression was detected 13 . However, reduced collagen secretion was not detected in fibroblasts of patients with mutant CREB3L1 (OASIS) 25 . Furthermore, S1P cko chondrocytes retain type IIB procollagen due to a trafficking defect but do not exhibit decreased Col2a1 expression 24 . Finally, the activation of RIP substrates ATF6 and SREBP during ER stress and sterol-free conditions, respectively, were significantly reduced in CHO-M19 cells transfected with reporter constructs containing mutant versus normal MBTPS2.
Notably, bone tissue from one X-OI proband revealed an underlying collagen-related defect. Type I collagen extracted directly from P1/II bone had significantly reduced hydroxylation of helical lysine residue 87 (K87) in both alpha chains, consistent with the proband's increased urinary LP/HP ratio. Although bone was not available from P1/I, his elevated urinary LP/HP ratio suggests a similar effect on collagen crosslinking occurs in his bone, while the LP/HP ratio was normal in two X-OI carriers and IFAP/KFSD patients (Supplementary Table 2). K87 is critical for collagen crosslinking in bone, which is a major contributor to bone strength 26 . Similarly reduced hydroxylation of type I collagen K87 to about half of wild-type control values occurs in a Ppib/CyPB knockout mouse model of type IX OI 12 . In the Ppib knockout mouse, the level of LH1/PLOD1, the enzyme which modifies K87, is normal, but its function is impaired by absence of its PPIase, CyPB. In contrast, LH1 protein is reduced in osteoblasts from X-linked OI proband P1/II by an unknown mechanism, although the level of CyPB is normal. Thus, the bone collagen defect in X-linked OI combines reduced collagen secretion into matrix and impaired collagen crosslinking. Disturbance of additional OASIS functions may contribute to the severe skeletal phenotype of X-linked OI, including apoptosis, downregulation of osteocalcin, osteopontin and bone sialoprotein, and ER stress, which overlaps ATF6 activation 27 .
It is apparent that mutations at different positions in MBTPS2 cause distinct syndromes, with no evidence at this time to suggest a spectrum disorder. A variety of skeletal malformations have been reported in IFAP-family syndromes, specifically vertebral malformation and cleft hands 28 . For example, the p.R429H IFAP mutation studied in our reporter assays causes vertebral abnormalities 7,8 . Short stature, inguinal hernia and microcephaly have also been noted in some IFAP patients. However, IFAP patients do not have the fracture susceptibility or generalized bone dysplasia that are the hallmarks of OI. However, the mechanism by which different mutations in MBTPS2 cause totally different developmental disorders remains unclear. The biochemical studies comparing X-linked OI and IFAP fibroblasts show no clear distinction between phenotypes in cleavage of RIP substrates OASIS, ATF6 and SREBP. The major distinction between syndromes resides in the urinary crosslinks, with elevated LP/HP ratios in X-OI probands and not in X-OI carriers or an IFAP patient. This presents a reasonable hypothesis that impaired hydroxylation and crosslinking of bone collagen is the critical distinction for X-OI versus IFAP. Future studies elucidating the RIP pathways involved in bone development may reveal tissue-specific differences in collagen post-translational modification and crosslinking, or critical S2P bone-related substrates other than OASIS, ATF6 and SREBP. These studies of RIP pathways in bone development are of fundamental importance and will be facilitated by animal models and bone samples from additional patients with both X-OI and IFAP.

Methods
Detailed case reports. Family I. The proband (P1/I; V7) of Family I, the Thai pedigree, was delivered at 39 weeks gestation by caesarean section due to multiple long bone fractures detected at 4 months gestation. His birth weight was 2,860 gm (10th centile), length 45 cm (o3rd centile), head circumference 34 cm (40th centile) and anterior fontanelle 5 Â 5 cm. He had fractures of ribs, humeri, femora and tibiae (Fig. 2a). The proband received pamidronate starting at age 3 weeks.
He stood with support at 18 months. No additional fractures occurred through 24 months of age, when his weight was 10 kg (3rd centile), length 76.5 cm (50th centile for a 13-month-old boy) and head circumference 47.8 cm (40th centile). He had closed anterior fontanelle, positional plagiocephaly, blue sclerae, normal teeth, pectus carinatum, straight spine and upper extremity long bones but bowing of femora and tibiae. Karyotype showed normal 46, XY.
The pedigree contains 12 affected members, with seven living (Fig. 1a). Clinical assessments and blood samples were obtained from 18 individuals, including six affected (Supplementary Table 1). The proband's 32-year-old mother was healthy with normal weight, height and bone mineral density. The proband and his mother had normal serum calcium, phosphorus, alkaline phosphatase and lipid profiles. Affected adult males had fractures beginning during gestation, white sclerae, normal teeth, head circumference 52-58 cm (within mean ± 2.5 s.d. for Thai adults), variable scoliosis and pectal deformity, bowing of extremities with striking anterior angulation of tibiae, generalized osteopenia and final adult stature from 110 to 140 cm (50th centile for age 5-11 years).
Family II. Patient P1/II is a 26-year-old man ( Fig. 2b and Supplementary  Fig. 2a). The pregnancy was uneventful and he was delivered spontaneously at term. Birth weight was 2,970 g (10th-25th centile), with a length of 46 cm (3rd-10th centile). Direct postpartal bowing of the distal humeri and the proximal radii were indicative of osteogenesis imperfecta. Radiography showed bilateral fractures of the clavicles, several rib fractures and a consolidated prenatal fracture of the left femur ( Supplementary Fig. 2b). Between age 14 months and 16 years, numerous fractures occured after only minor and moderate traumatic events, involving the femur 11 times (this stopped after the insertion of Bailey-Dubow elongation rods at age 8 years), the humerus 7 times, the lower arm 2 times and the tibia 3 times. Two fractures of the mandibulae occured after adequate traumatic events. There were no new fractures until age 24 years when one traumatic fracture of the lower leg occured. Treatment with 9 mg kg À 1 per day intravenous bisphosphonate was started at age 13 for 3 years. Current height at the age of 26 years is 110 cm (50th centile for 6-year-old male) and weight is 37 kg. Moderate kyphoscoliosis of the thoracic and lumbar spine is present. There are no signs of dentinogenesis imperfecta and the sclerae are white. An active wheelchair is mainly used. Unsupported standing is possible for a longer time; unsupported walking only a few steps, with support 5-10 m, and on a treadmill more than 1 km.
The maternal uncle P2/II is a 68-year-old man who was diagnosed with osteogenesis imperfecta quite soon after birth. Fracture rate and bending of the long bones were similar to P1/II. Since puberty, fractures are rarely observed unless through adequate trauma. He has a severe connatal pectus excavatum. His current height is 105 cm (50th centile for 5-year-old male) and his weight is 45 kg. Kyphoscoliosis of the thoracic and lumbar spine is more pronounced than in his nephew, and is rather progressive. He has no signs of dentinogenesis imperfecta and his sclerae are white. He can walk slowly with two crutches.
Molecular genetic investigations of Family I. DNA from blood and dermal fibroblasts. Proband and family whole blood and skin samples were collected after written informed consent was obtained under a protocol approved by the Institutional Review Board of the Faculty of Medicine, Chulalongkorn University. High-molecular-weight DNA was extracted from leukocytes of all the available family members, and from cultured skin fibroblasts of the proband, using the ArchivePure DNA Blood Kit (5 PRIME Inc., Gaithersburg, MD), according to the manufacturer's instructions.
X-inactivation assay. Eight female members (III1, III5, III7, IV5, IV13, IV18,  IV21 and IV23) and one unaffected male (IV19) were subjected to X-inactivation analysis as previously described 29 . A total of 500 ng blood-derived DNA was digested with 10 U of the methylation-sensitive enzyme HpaII at 37°C for 16 h and enzyme-inactivated at 95°C for 10 min. The first exon of the androgen receptor gene (AR) was amplified using a fluorescently labelled forward primer FAM-5 0 -CGCGAAGTGATCCAGAACCC-3 0 and reverse primer 5 0 -GTTGCTGTT CCTCATCCAGG-3 0 in digested and undigested DNA templates. The polymorphic CAG repeats in the AR gene were used to identify parental chromosomes. Methylation differences were analysed based on peak heights in the PCR products by Genemapper version 3.7 (Applied Biosystems, Foster City, CA).
Linkage analysis. Linkage analysis for Family I was performed according to the previously described methods 30 . X-chromosomal microsatellite markers were PCRamplified using panel 28 of ABI Prism Linkage Mapping Sets-MD10 Version 2.5 (Applied Biosystems, Foster City, CA). An additional 21 polymorphic microsatellite markers were used for fine mapping on chromosome Xp22 between DXS8051 and DXS1068 (Supplementary Table 3). All fluorescently labelled primers were typed on an ABI Prism 3100 genetic analyzer (Applied Biosystems, Foster City, CA) with Genemapper version 3.7. For genetic mapping, the MLINK program was used to calculate two-point linkage analysis with the following model: X-linked recessive with high penetrance, disease and normal allele frequencies were set at 0.01 and 0.99. Maximum logarithm of the odds (LOD) score using the deduced genotyping of unavailable members was also shown in the bracket. Linkage analysis of the X-chromosome using a panel of microsatellite markers defined a 21.6-Mb critical region from markers DXS7108 to DXS1067, with a maximum LOD score of 3.31 ( Supplementary Fig. 1a and Supplementary Table 3).
Mutation analysis of candidate genes. Candidate genes PIR, TRAPPC2, PHEX, Cxorf15, AP1S2, CA5B, PDK3, PIGA, PRRG1, SMPX, were selected because of their inferred biological relevance for connective tissue functions. HCCS, MSL3, PRPS2 on the telomeric end of the critical region were also studied. Primer pairs used for amplification of entire coding regions are available on request.
Comparative Genomic Hybridization. CGH was performed on test DNA (affected male, IV17) and reference DNA (unaffected male, IV10) for the determination of copy number variants (Macrogen Inc., Seoul, Korea). DNAs were independently labelled with fluorescent dyes, co-hybridized to a NimbleGen Human CGH 385K chromosome X Tiling array, and scanned using a 2-mm scanner. Log2-ratio values of the probe signal intensities (Cy3/Cy5) were calculated and plotted versus genomic position using Roche NimbleGen NimbleScan software. Data are displayed in Roche NimbleGen SignalMap software. CGH did not show any imbalances in the critical region.
Next-generation sequencing. We performed targeted resequencing by NGS of the 18.6-MB linkage region between DXS1224 and DXS1067 interval on affected male (IV17) and unaffected male (IV10) genomic DNAs (Macrogen Inc., Seoul, Korea). DNA was captured on a customized NimbleGen 2.1 array (Roche NimbleGen, Madison, WI) with a capturing capacity of 33 Mb. The targeted region corresponded to positions 13,235,460 to 31,798,001 bp on Chromosome Xp22 according to the UCSC hg19 Assembly. Capture efficiency varied across the target, with a mean read depth of 121.1X. The captured library was subsequently sequenced using the Illumina platform Genome Analyzer II X (GAIIX) in a single-end 76 bp configuration. Sequence reads were mapped against UCSC hg19 using BWA software (http://bio-bwa.sourceforge.net/). The SNPs and Indels were detected by SAMTOOLS (http://samtools.sourceforge.net/) and annotated by SIFT (http://sift.jcvi.org/) (Supplementary Table 4). We found 25 variants in coding sequences. After excluding known SNPs in dbSNP Build 135, 1000 Genomes and HAPMAP, and variants present in the unaffected relative, two candidates remained: c.1376A4G (p.N459S) in MBTPS2 and c.532G4A (p.V178M) in FAM48B1.
Mutation confirmation and segregation studies. PCR amplification and Sanger sequencing were performed to confirm the MBTPS2 mutation in the proband and his parents. Primer pairs and PCR conditions for the amplification of the coding exons of MBTPS2 including the intron-exon boundaries are available on request. Restriction enzyme digestion with BsgI was used for cosegregation study and normal control screening. The identified mutation was analysed with ClustalX for conservation analysis. Alamut, Polyphen2, SIFT, Panther and align GVGD were used for prediction of mutation function.
Genome-wide linkage analysis. For the five living members of Family II (Supplementary Fig. 1) a genome-wide linkage scan was performed using the HumanCytoSNP-12v2 BeadChip (Illumina, San Diego, CA) according to the supplier's instructions. Raw SNP call data were processed with the Genotyping Analysis Module in GenomeStudio 1.6.3 (Illumina, San Diego, CA). Homozygosity mapping was performed with the Allegro program, under the assumption of a recessive mode of inheritance in absence of consanguinity loops.
X-inactivation assay. X-inactivation analysis was performed on gDNA from the mother of P1/II according to Beever et al 31 . PCR products were analysed on an ABI NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11920 ARTICLE 3130 Capillary DNA Sequencer (Applied Biosystems) and the methylation status was analysed by Genemapper version 4.0 (Applied Biosystems) based on the peak heights of the PCR products.
Exome enrichment and high-throughput sequencing. Genomic DNA was extracted from whole blood using the MagNA Pure Compact System (Roche Life Science, Indianapolis, IN, USA). DNA library preparation was performed with the TruSeq DNA Sample Preparation Kit (Illumina, San Diego, CA) according to the manual instructions. Enrichment of X-chromosomal genes was done by using a custom designed NimbleGen Sequence Capture Microarray (solid array with 6962 targets on the X chromosome or 1.4 Mb). The captured library was sequenced on a MiSeq sequencer (Illumina, v1) with 2 Â 150 cycles of paired-end sequencing.
Data analysis. Raw data processing, sequence read alignment from FASTQ to BAM format and variant calling to generate VCF files were performed with the MiSeqReporter Software (Illumina, San Diego, CA). Aligned BAM files with removed duplicated reads were further technically analysed using the SeqMonk program version v0.25.0 (http://www.bioinformatics.babraham.ac.uk/projects/ seqmonk). Variant filtering was performed by GeneTalk (https://www.gene-talk.de/).
Mutation confirmation and segregation study. Mutation screening of MBTPS2 and segregation analysis of the mutation within the family was performed by fluorescent bidirectional sequencing of genomic amplified PCR products on an ABI 3100 automated sequence detection system (Applied Biosystems, Foster City, CA, USA). PCR conditions and primer sequences are available from the authors on request.
Bioinformatics analysis aimed to predict pathogenicity of the identified L505F variant was performed with Provean, Panther, MutationTaster and PolyPhen2. Conservation analysis of the protein variant was performed by multiple sequence alignment of MBTPS2 with ClustalX, ClustalW and BOXSHADE programs (Fig. 2e).
Urinary pyridinoline analysis. Total urinary pyridinoline crosslinks, LP and HP, were measured in the affected individuals and obligate carriers of Family I and Family II, as well as in an MBTPS2 R429H IFAP patient and unaffected carrier (Supplementary Table 2) by HPLC as described 32 .
Luciferase assay. CHO-M19 cells, a deleted Mbtps2 orthologous cell line, were used for transfection of reporter constructs. Cells were maintained in Ham's F-12 medium supplemented with 10% fetal bovine serum at 37°C with 5% CO 2 . We performed luciferase assays by transient transfection of CHO-M19 cells either with p5xATF6 or pSRE reporters as previously described 5,7 . The p5xATF6 and pSRE reporters contain either an ER-stress response element or a sterol regulatory element, respectively, two elements that are known targets of S2P. These assays are an indirect measurement of the transcriptional activity of S2P cleavage products on these two elements. For the p5xATF6 reporter, cells were set up at a density of 1.5 Â 10 5 cells per well in triplicate in 12-well culture plates. Cells were transfected with 1 mg of expression plasmid containing normal, mutant (N459S, R429H and L505F) MBTPS2 or no insert, together with 800 ng of p5xATF6-GL3 and 50 ng of pRL-SV40 as a transfection control. The total amount of DNA was adjusted to 2 mg per dish by addition of pdEYFP-C1amp. At 24 h post-transfection, cells were treated with 2 mg ml À 1 of tunicamycin dissolved in DMSO to induce stress. Untreated cells received 0.2% DMSO. After incubation for 24 h, the firefly and renilla luciferase activities were measured using a dual luciferase assay system according to the manufacturer's protocol (E1910 Promega kit, Madison, WI). For the pSRE-reporter, cells were set up at a density of 1.5 Â 10 5 cells per well in triplicate in a 12-well plate. On day 2, the cells were switched to medium containing 1:1 Ham's F12:DMEM, 5% lipoprotein-deficient serum and 20 mM sodium oleate for 7 h. The transfected plasmids were either wild-type S2P or one of the three mutants (p.N459S, p.R429H, p.H171F) and the pSRE reporter. Cells were then switched to medium containing 1:1 Ham's F12:DMEM, 5% lipoproteindeficient serum, 50 mM compactin and 50 mM mevalonate supplemented with sterol (10 mg ml À 1 cholesterol, 1 mg ml À 1 25-hydroxycholesterol). Untreated cells were treated with 0.2% ethanol (0.2% w/v). The firefly and renilla luciferase activities were measured after 16 h of incubation. The ratio between firefly luciferase and renilla luciferase luminescence from different constructs and control was analysed by one-way analysis of variance (ANOVA). Specific pairwise comparison between luciferase luminescence in each group used the least significant difference test. One-way ANOVA and least significant difference were included in SPSS software version 22.
For type I collagen biochemical studies, normal control and proband primary fibroblast and osteoblast (see below) cultures were labelled for 16 h in serum-free DMEM containing 437.5 mCi ml À 1 L-[2,3,4,5-3 H]-proline. Procollagens were precipitated with ammonium sulfate, pepsin digested and electrophoresed on 6% SDS-urea polyacrylamide gels under nonreducing conditions 33 . Procollagen secretion by control and proband fibroblasts (Fig. 5b) was determined in 24-h conditioned media using the Sircol collagen assay (Biocolor Ltd, Carrickfergus, UK) according to the manufacturer's specifications. Collagen protease sensitivity was analysed as previously described 34 . For protease sensitivity assays, type I collagen from the medium of cultured proband fibroblasts labelled overnight with 437.5 mCi ml À 1 L-[2,3,4,5-3 H]-proline was digested for 1, 3 and 5 min with trypsin and chymotrypsin (100 and 250 mg ml À 1 final concentration, respectively) at 37°C without prior cooling. Digests were stopped with 1 mg ml À 1 soybean trypsin inhibitor and analysed by SDS-urea PAGE.
Differential scanning calorimetry (DSC) was performed in buffer containing 0.2 M sodium phosphate and 0.5 M glycerol at pH 7.4, from 10 to 55°C in a Nano III DSC instrument (Calorimetry Sciences Corporation, Lindon, UT), as previously described 35 . Amino-acid composition was analysed by high pressure liquid chromatography (AIBiotech, Richmond, VA).
Collagen extracellular matrix deposition was analysed as previously described 36 . In brief, post-confluent proband and normal control fibroblasts were cultured for 2 weeks in DMEM containing 10% FBS with 100 mg ml À 1 ascorbic acid. After 14 days, fibroblasts were labelled 24 h with 406.25 mCi ml À 1 L-[2,3,4,5-3 H]-proline and the media and extracellular matrix was harvested. Collagens were sequentially extracted from the matrix with 150 mM NaCl, 0.5 N acetic acid and then 50 mg ml À 1 pepsin.
Mass spectral analysis of bone tissue collagen. Demineralized bone was digested with bacterial collagenase and the resulting collagen-derived peptides were separated by reverse-phase HPLC, as previously described 32 . Electrospray MS was performed on in-gel trypsin digests and individual HPLC column fractions using an LTQ XL ion-trap mass spectrometer equipped with in-line liquid chromatography (Thermo Scientific, Waltham, MA) using a C4 5-mm capillary column (300 Â 150 mm; Higgins Analytical RS-15M3-W045) eluted at 4.5 ml min À 1 . The LC mobile phase consisted of buffer A (0.1% formic acid in MilliQ water) and buffer B (0.1% formic acid in 3:1 acetonitrile:n-propanol v/v). The LC sample stream was introduced into the mass spectrometer by electrospray ionization with a spray voltage of 4 kV. Proteome Discoverer search software (Thermo Scientific) was used for peptide identification using the NCBI protein database. Proline and lysine modifications were examined manually by scrolling or averaging the full scan over several minutes so that all of the post-translational variations of a given peptide appeared together in the full scan.
Study oversight. This study was conducted according to the Declaration of Helsinki for Human Rights and approved by Swiss Ethics and by the Institutional Review Board of the Faculty of Medicine, Chulalongkorn University. Written informed consent was obtained from the patients or their parents.
Data availability. The data that support the findings of this study are available from the corresponding author on request. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11920