Biochemical basis for dominant inheritance, variable penetrance and maternal effects in RBP4 congenital eye disease

SUMMARY

Gestational vitamin A (retinol) deficiency poses a risk for ocular birth defects and blindness. We identified missense mutations in RBP4, encoding serum retinol binding protein (RBP), in three families with eye malformations of differing severity. The mutant phenotypes exhibit dominant inheritance but incomplete penetrance. Maternal inheritance significantly increases the probability of phenotypic expression. RBP normally delivers retinol from hepatic stores to peripheral tissues, including the placenta and fetal eye. The disease mutations greatly reduce retinol binding to RBP yet paradoxically increase RBP affinity for its cell surface receptor, STRA6. By occupying STRA6 nonproductively, the dominant-negative proteins are predicted to disrupt vitamin A delivery from wild-type proteins within the fetus but also, in the case of maternal transmission, at the placenta. These findings establish a previously uncharacterized mode of maternal inheritance, distinct from imprinting and oocyte-derived mRNA, and define a group of hereditary disorders plausibly modulated by dietary vitamin A levels.

INTRODUCTION

Congenital eye malformations – including microphthalmia, anophthalmia and coloboma (MAC) disease – affect 2 in 10,000 births, and are an important cause of childhood blindness (Morrison et al., 2002). The severity depends on timing and the extent that growth and morphogenesis of the developing eye is disrupted (Graw, 2010). Anophthalmia, or total absence of eyes, is the most extreme form. Microphthalmia (small eyes) and coloboma (ventronasal notch-like defects in the iris and/or retina, arising from incomplete closure of the choroid fissure, Onwochei et al., 2000) are less severe. These can occur as uni- or bilateral birth defects, and may coexist in an individual or pedigree. Most cases are isolated but one-third are associated with systemic birth defects. Few genetic causes have been identified (Williamson and FitzPatrick, 2014). Loss-of-function SOX2 mutations account for 10% of bilateral anophthalmia (Fantes et al., 2003), whereas mutations in RX, CHX10, BCOR, HCCS and PAX6 transcription factors explain other monogenic cases. Signaling pathways mediated by BMP4, GDF6 and SHH may also be genetically disrupted. Finally, environmental agents such as poor maternal nutrition are potential risk factors (Hornby et al., 2003).

Vitamin A is an essential, fat-soluble nutrient for embryonic development, tissue homeostasis and physiology. Its most widely recognized function is to supply the visual cycle with 11-cis-retinal (vitamin A aldehyde) for generation of the light-sensitive visual pigment rhodopsin (Lamb and Pugh, 2004). Consequently, vitamin A deficiency (VAD) first manifests as night blindness (nyctalopia), a reversible loss of visual adaptation to dark environments (Dowling and Wald, 1958). Vitamin A is also required for epithelial, reproductive and immune health. At the molecular level, vitamin A is a substrate for synthesis of retinoic acid (RA), a potent signaling molecule needed for vertebrate organogenesis, including eye development (Duester, 2009; Niederreither and Dolle, 2008). Nutritional studies have long associated maternal vitamin A deficiency with eye malformations, as well as urogenital, diaphragmatic, cardiovascular and pulmonary defects (Hale, 1933; See and Clagett-Dame, 2009; Wilson et al., 1953). Recently, genetic links were established between retinoid signaling defects and MAC disease. Loss-of-function mutations in STRA6, encoding the membrane receptor for serum retinol binding protein (RBP), cause autosomal recessive anophthalmia or Matthew-Wood syndrome (OMIM 601186), characterized by structural eye defects, diaphragmatic hernias, cardiac malformations and pulmonary hypoplasia (Golzio et al., 2007; Pasutto et al., 2007; Casey et al., 2011; Chassaing et al., 2009). Likewise, mutations in ALDH1A3, encoding retinaldehyde dehydrogenase, account for a subset of recessive MAC cases (unpublished data; Fares-Taie et al., 2013; Yahyavi et al., 2013).

Here we show mutations in the serum RBP gene underlie an autosomal dominant form of MAC that is transmitted with incomplete penetrance and a unique maternal parent-of-origin effect (Sturtevant, 1923). We further show that the unliganded mutant RBPs bind STRA6 with much greater affinity than wild-type, and consequently disrupt delivery of vitamin A to target cells, consistent with a dominant-negative effect. These results shed light on the maternal-fetal nutritional interface, genetic susceptibility to vitamin A deficiency, and the etiology of eye malformations.

RESULTS

Autosomal dominant MAC disease with reduced penetrance and a maternal effect

A seven-generation pedigree (Family 1) was identified through two probands with anophthalmia (Figure 1A). The penetrance of eye disease is incomplete (P = 0.4), based on 54 informative meioses (Figure S1A). Carriers have phenotypes ranging from normal to microphthalmia to complete absence of the eyes (Figure 1B and Table S1). Several individuals have iris and/or chorioretinal colobomas. Transmission is skewed. Nearly all affected individuals (10 of 11) inherited the trait from their mother, such that maternal penetrance is significantly greater than paternal penetrance (P_mat = 0.7, P_pat = 0.1, Figure S1B). In the only instance of paternal transmission, one of two monozygous twins (VI-2) was affected.

Figure 1. Familial MAC disease.

(A) Family 1 pedigree with two probands (arrows). Eleven family members have microphthalmia or coloboma (gray symbols), or clinical anophthalmia (black symbols). (B) Anterior eye and fundus photographs of family members with iris or chorioretinal colobomas (VI-2, VII-2 and VII-3), microphthalmia (III-12 and VII-2) or bilateral clinical anophthalmia (VII-5), and orbital MRI views of VII-5. The T2w coronal MRI shows extraocular muscles (red arrowheads) but absent eye globes. The T2wFS axial image shows bilateral hyperintense orbital cysts (yellow arrowheads). onh, optic nerve head; crc, chorioretinal coloboma; L, left; R, right. See also Table S1. (C) Genetic mapping of MAC disease. top, Multipoint LOD plot of autosomes, based on affected individuals and obligate carriers. bottom, Expanded linkage analysis favors chromosome 10 localization. See also Figures S1 and S2.

A new MAC locus on chromosome 10q23

We first excluded 23 loci associated with MAC in humans or vertebrate models (Table S2) by comparing haplotypes of the two probands. We then examined available family members and performed genome-wide multipoint linkage analysis (Figure S1C,D). We applied a simple autosomal dominant (AD) model, scoring only affecteds and obligate carriers. This analysis suggested three candidate regions: 1q41, 10q23 and 19p13, with peak LOD scores >2 (Figure 1C). To rank these regions, we included at-risk unaffected family members and applied AD models with uniform (P_global = 0.4) or sex-specific (P_mat = 0.7, P_pat = 0.1) penetrance. This indicated a chromosome 10q23 localization with peak LOD score of 3.01 (Figure 1C). The 8.2 Mb nonrecombinant interval contains 81 genes (Figure S2). Given the importance of vitamin A in eye development (Warkany and Schraffenberger, 1946) and eye malformations associated with STRA6 and ALDH1A3 mutations, we tested genes in the critical region with roles in vitamin A transport (RBP4) and RA metabolism (CYP26A1 and C1).

Dominant RBP4 mutations in three unrelated MAC families

RBP4 encodes serum RBP (Kanai et al., 1968) and contains six exons (Figure 2A). Exon screening revealed a missense mutation (c.223G>A, p.A75T) that cosegregated with the disease trait (Figure 2B) and was not found in >11,330 control chromosomes. We then screened a cohort of 75 unrelated MAC samples, and discovered mutations in two cases, a male with bilateral anophthalmia and neurodevelopmental delay (Family 2), and a female with left microphthalmia and coloboma (Family 3). They share a single missense allele (c.217G>A, p.A73T) on two distinct haplotypes, indicating recurrence of the mutation, with maternal transmission in both families (Figure S3).

Figure 2. RBP4 mutations in three independent families with congenital eye malformations.

(A) Map of 9.4 kb RBP4 gene, with signal sequence (gray) and mature protein (black) coding regions, and MAC mutations (red box). (B). Sequence chromatograms showing heterozygous missense mutations, with maternal transmission in each pedigree. (C) Primary structure of translated RBP with ala-to-thr substitutions (red) in the mature polypeptide (yellow bar), and two alleles associated with recessive nyctalopia (gray). Note that A73T and A75T in the primary translation product correspond to A55T and A57T following cleavage of the signal sequence (gray bar, SS). Cyan coils, α-helix; blue arrows A-H, β-strands forming the β-barrel. (D) Ribbon diagrams showing positions of dominant (red) and recessive (gray) substitutions. Eight anti-parallel strands (dark blue) form the ligand pocket. Three loops (green) at the calyx opening contact transthyretin. The N-terminus is relatively unconstrained. (E) Alignment showing conservation of alanines 55 and 57 among vertebrates. See also Figure S3.

p.A73T and p.A75T alter the retinol-binding interface

RBP mobilizes all-trans retinol from liver stores to target tissues, including the retinal pigment epithelium and placenta (D'Ambrosio et al., 2011). As the archetypal lipocalin (Newcomer and Ong, 2000), RBP folds as a β-barrel with a central hydrophobic ligand cavity (Figures 2D and S4) (Cowan et al., 1990; Zanotti et al, 1993). Both mutations substitute threonine for alanine, in codons 73 and 75 of β-strand C (Figure 2C), corresponding to residues 55 and 57 in the mature polypeptide. These alanines face the ligand pocket (Figure 2D), contact carbons C4 and C3 of the retinol p-ionone ring, respectively (Cowan et al, 1990), and are completely conserved among vertebrates (Figure 2E).

Two previously reported RBP4 mutations, p.I59N and p.G93D, were associated with recessive night blindness in compound heterozygous sisters (Biesalski et al., 1999). These correspond to I41N and G75D in β-strands B and D of the mature protein, after signal peptide cleavage. These residues also interact with side groups of the β-ionone ring, and biochemical data suggest G75D and I41N proteins bind retinol poorly (Folli et al., 2005). Molecular modeling shows that A55T and A57T proteins can accommodate retinol, under increased strain due to steric, hydrophilic and H-bonding effects of the threonine side chain (Figure S3). To understand the allelic heterogeneity and pathogenic basis of MAC disease, we systematically compared properties of wild-type (WT) and mutant RBPs.

A55T and A57T proteins are secreted as stable 21 kD monomers

RBP is constitutively expressed by hepatocytes, retained in the endoplasmic reticulum (ER) and secreted into the bloodstream as holo-RBP (Muto et al., 1972; Soprano, 1994), stabilized by three disulfide bonds (Selvaraj et al., 2008). We first evaluated how missense mutations affect RBP synthesis and secretion in transfected HeLa cells (Melhus et al., 1992) by Western analysis, using an N-terminal hemagglutinin (HA) tag (Figure 3A). The size (21 kD) and abundance of A55T and A57T proteins in 48-hr conditioned media (CM) were indistinguishable from WT (Figure 3B). In contrast, G75D and I41N proteins migrated as 42 kD dimers, or larger multimers (I41N), linked by intermolecular disulfide bonds (Figures 3B and S4). We confirmed this result using glutaraldehyde cross-linked CM, and compared intracellular RBP levels using WT^KDEL as an ER retention control (Figure 3C). Intracellular G75D and, to a greater extent, I41N were elevated, suggesting a partial secretion defect, with no evidence of ER stress (Figure S4B). We conclude A55T and A57T are secreted as stable 21 kD monomers, whereas G75D and I41N misfold in the ER, aggregate, and exhibit increased cellular retention.

Figure 3. A55T and A57T proteins are secreted as stable RBP monomers and interact with transthyretin.

(A) Test of RBP^HA synthesis, secretion and integrity. (B) HA Western analysis of conditioned media, electrophoresed under native or denaturing conditions, before or after crosslinking. (C) Western blot of cell lysates, with α-tubulin loading control. (D) RBP-TTR binding assay in tissue culture. Brown circles, TTR homotetramers; blue bars, RBP monomers. (E) Western blot of HA immunoprecipitates probed in sequence with TTR and HA antibodies. (F) SPR analysis of RBP-TTR binding in vitro. top. Sensorgrams show a TTR concentration series interacting with apo his-RBPs on a biotin capture chip. middle. Steady state isotherms for apo and holo^HARBP binding to TTR. bottom. Histogram of K_d values. Error bars give the SEM for nonlinear regression. See also Figure S5.

A55T and A57T complex normally with transthyretin

Under normal conditions, holo-RBP and transthyretin (TTR), a 60 kD homotetramer (Heller and Horwitz, 1974), are co-secreted in a 1 -to-1 molar ratio as a 76 kD complex (Kanai et al., 1968). Its large size prevents renal filtration, allowing RBP to remain in circulation (Soprano, 1994; van Bennekum et al., 2001). In coimmunoprecipitation experiments (Figure 3DE), human TTR interacted strongly with WT, A55T and A57T proteins, but poorly with G75D or I41N. Similar results were obtained for bovine TTR, present in the media supplement. To quantitatively assess the RBP-TTR interaction, we performed reciprocal surface plasmon resonance (SPR) assays with purified transthyretin and recombinant RBP^HA or his-RBP (Figure 3F). Wild-type holo-RBP bound TTR with 2–3 fold greater affinity than apo-RBP, giving mean steady state K_d values of 0.9 and 2.2 µM respectively, similar to previous reports (Folli et al., 2010; Malpeli et al., 1996). The affinity of A55T and A57T mutant RBPs was similar or slightly lower than WT in buffered saline (HBS). However, inclusion of nonionic surfactant (0.005% Tween) significantly reduced holo A55T affinity for TTR, presumably by removing retinol (p< 0.001, unpaired t-test, df = 10, see below).

The in vitro behavior of G75D and I41N proteins is consistent with the absence of immunodetectable serum RBP in p.G93D/p.I59N compound heterozygotes and reduction of RBP in the p.I59N/+ parent, in the setting of normal TTR levels (Biesalski et al., 1999). Conversely, RBP and TTR levels in p.A75T/+ (Family 1: VI-2, VI-3 and VI-7) and p.A73T/+ (Family 3: II-2) carriers were within normal range (Table S3).

WT and A57T proteins coexist in p.A75T/+ carrier plasma

To assess the ratio of allotypes in vivo, total RBP was purified from obligate carrier VI-2 plasma (Figure S4), digested with trypsin and analyzed by mass spectrometry (Figure 4). The predicted WT and A57T peptides encompassing amino acid 57 differ by 30 Da. Consequently, we identified MALDI-TOF peaks in the 3,100 to 3,220 m/z range corresponding to WT and A57T tryptic peptides, with a 2-to-1 intensity ratio (Figure 4C). These were verified by MS² analysis (not shown) and parallel MS of RBP^HA controls. Since the peptides ionize with equal efficiency (Figure 4D), we conclude that A57T constitutes one-third of circulating RBP in p.A75T/+ heterozygotes.

Figure 4. Mass spectrometry of RBP proteotypes in p.A75T/+ carrier plasma.

(A) Tryptic peptides encompassing residue 57. Modified peptides (asterisks) arise from alkylation of methionine 53 (♦). (B) MALDI-TOF spectrum of RBP from control human plasma, indicating the critical m/z region (red box). The Y-axis (ions detected) reflects relative intensity. (C) Expanded view of control (top) and carrier (bottom) spectra from 3,100 to 3,250 m/z. Single-ionization peaks corresponding to WT (red lines) and A57T (green lines) proteins are marked. (D) MALDI-TOF spectra for recombinant RBP^HA. The invariant 3,223.3 m/z peak (human keratin, a common contaminant) serves as an internal standard. See also Figures S5 and S6.

Because both allotypes were present in carrier plasma, genomic imprinting is unlikely to explain the skewed transmission of the MAC disease (Figure S1B). This conclusion is supported by RT-PCR analysis of F1 mice, which showed comparable levels of allelic Rbp4 mRNA transcripts in adult and fetal tissues (Figure S4C).

In principle, the unequal ratio of allotypes could be explained by a difference in renal filtration. Under normal circumstances, RBP dissociates from TTR when retinol is delivered to tissues (Malpeli et al., 1996). Most of the resulting apo-RBP is filtered and metabolized by the kidney, but trace amounts are detected in urine, at 1% of serum levels (Raila et al., 2005), and are assumed to represent the RBP content of the glomerular ultrafiltrate proportionally. To test this hypothesis, we evaluated RBP allotypes in p.A75T/+ carrier urine by mass spectrometry, but found no evidence for increased urinary elimination of A57T relative to WT (Figure S6).

A55T and A57T proteins bind vitamin A poorly

We tested retinol-binding properties of mutant RBPs using two assays, double radioisotope labeling and fluorescence enhancement. HeLa cells expressing WT or mutant RBP^HA were exposed to ³⁵S-met/cys and ³H-retinol, and the ³H/³⁵S ratio was determined for RBP^HA immunopurified from conditioned media (Figure 5AB). We observed a dramatic reduction in retinol binding, as predicted by molecular modeling (Figure S3E). A55T bound negligible ³H-retinol whereas A57T bound 16% of WT levels. G75D and I41N mutants also bound very little vitamin A, as expected given their misfolded structures. RBP activity is evidently more sensitive to a threonine substitution at position 55 than 57, consistent with X-ray data placing retinol closer to Ala55 (3.6Å) than Ala57 (4Å) (Cowan et al., 1990).

Figure 5. A55T and A57T proteins bind retinol poorly in a mixed aqueous-lipid environment.

(A) ³H-retinol binding assay. (B) left³H-retinol binding data normalized to WT. Error bars give the SEM for three parallel assays. right. Autoradiogram showing secreted ³⁵S-RBP^HA in conditioned media. (C) left. in vitro retinol binding profiles for WT and mutant RBP^HA measured by fluorescence in PBS ± 0.1% α-L-phosphatidylcholine (PC), with 380 nM protein. right. Histogram showing similar K_d values in PBS. (D) left. Normalized retinol binding curves in PBS with 0 to 50% ethanol. right. Increased sensitivity of mutant RBPs in an amphipathic environment, measured by loss of retinol fluorescence after exposure to detergent micelles (1% Tx-100, 0.5% DOC) in PBS.

Retinol fluorescence intensity increases 15-fold when it occupies the RBP ligand pocket (Cogan et al., 1976). Accordingly, we added 1 to 5000 nM retinol to apo-RBP^HA, purified under native conditions, and measured fluorescence (ex 330 nm, em 460 nm) in phosphate-buffered saline (PBS) (Figure 5C, filled symbols). Surprisingly, A55T and A57T both bound retinol well in this assay, with affinities similar to WT (K_d ~ 80 nM). These results are consistent with SPR analysis of holo and apo forms interacting with transthyretin in HBS (Figure 3F), but differ sharply from the radioisotope data showing the mutants bind little or no vitamin A (Figure 5B).

Wild-type holo-RBP is relatively resistant to temperature, pH extremes and nonpolar solvents (Cogan et al., 1976; Raz et al., 1970), but sensitive to low ionic strength (Peterson, 1971). Our disparate findings may be reconciled if A55T and A57T substitutions destabilize RBP contacts with retinol, particularly under adverse environmental conditions, increasing the probability that ligand is released to the solvent. Whereas the initial fluorescence assay was performed in PBS (Figure 5C, closed symbols), our ³H-retinol binding assay involved sequential washes in PBS containing 1% Triton X-100 and 0.5% deoxycholate (Figure 5A). We therefore systematically tested retinol binding in nonpolar and amphipathic environments (Figure 5D), including a dispersion of phosphatidylcholine (PC) vesicles (Figure 5C, open symbols), to more closely approach in vivo conditions. Within the ER, bloodstream and tissue interstitial space, RBP is continuously exposed to phospholipid membranes and lipoprotein particles (van Meer et al., 2008). Indeed, retinol-binding activity of the mutant proteins was hypersensitive to ethanol, detergents and phospholipid vesicles, following an A55T > A57T > WT allelic series. Almost no retinol was bound by A55T in 0.1% PC (K_d > 30 µM).

Our in vitro data predict that RBP4 heterozygotes may have reduced circulating retinol. Indeed, three p.A75T/+ obligate carriers had fasting serum vitamin A levels below the lower normal limit (Table S3), ranging from 50–60% of the reference mean, and plasma retinol fluorescence was reduced (Figure S5B).

Increased binding of A55T and A57T proteins to the STRA6 receptor

STRA6, or stimulated by retinoic acid 6 (Bouillet et al., 1997), is the transmembrane receptor for RBP that mediates cellular uptake of vitamin A (Kawaguchi et al., 2007). At target tissues, holo-RBP binds STRA6 extracellular loop 6 with high affinity (Kawaguchi et al., 2008). Following transfer of vitamin A into cells, apo-RBP dissociates from the receptor, allowing a new holo-RBP molecule to dock (Kawaguchi et al., 2007).

To examine binding of A55T and A57T proteins to STRA6, we performed two sets of experiments. We first applied ³⁵S-labeled apo-WT, holo-WT, A55T or A57T RBP in parallel to HEK293T cells transfected with STRA6^myc or control expression vectors and measured ³⁵S-RBP bound after one hour (Figure 6A–C). In this assay, apo-RBP had 3-fold lower steady-state binding than holo-RBP. More dramatically, STRA6+ cells bound 4 to 7 times more mutant apo-RBP than WT holo-RBP (p<0.002, unpaired t-tests, df = 4). These findings, and the mass spectroscopy data (Figure 4), suggest that competition may occur between mutant and wild-type RBP molecules at STRA6 receptors in vivo. To explore this possibility, we mixed 8- to 250-fold excess unlabeled holo-WT with ³⁵S-labeled RBP^HA in parallel assays. In each case, unlabeled WT competitor displaced much less mutant ³⁵S-RBP than expected if the binding affinities were equivalent.

Figure 6. A55T and A57T proteins bind the STRA6 membrane receptor with greater affinity than WT.

(A) STRA6 radioligand binding assay. (B) STRA6^myc expression in HEK293T cultures. left. Fluorescence micrographs of transfected cells immunostained for myc (green) with nuclear counterstain (blue). Scale bars, 40 µm. right. Western blot simultaneously probed with antibodies to human STRA6 (72 kD) and α-tubulin. (C) Histogram showing binding of 5 pM ³⁵S-labeled WT, A55T or A57T RBP^HA proteins in the absence (black) or presence (gray) of 8- or 250-fold excess unlabeled (cold) holo WT competitor. Error bars give the SEM for three parallel assays. (D) Quantitative equilibrium analysis of RBP-STRA6 interaction by immunoassay. left. Binding isotherms and reciprocal plots of apo-RBP^HA ELISA data. Relative RBP levels are given in cps (counts per sec) emitted light, after subtracting nonspecific binding to control cells. right. Histogram of K_d values. The mutant RBPs bind to the receptor with 30–40 fold greater affinity than WT. (E) Kinetic analysis of the RBP-STRA6 interaction. Release of bound apo-RBP^HA to the media over time, from saturated STRA6+ cells at 25°C. (F) Histogram comparing STRA6 association (k_on) and dissociation (k_off) rate constants calculated from binding data. The A55T and A57T mutations greatly increase the on rate for RBP binding to STRA6. See also Figure S7 and Table S4.

To characterize the STRA6-RBP interaction more precisely, we determined the binding affinity (K_d) and rate constant for the approach to equilibrium of mutant and WT RBPs, using a sensitive ELISA method (Figure S7) to measure RBP^HA bound to cells and released into the media. These assays show that the mutant proteins have a 30–40 fold greater affinity for STRA6 than wild-type (Figure 6D), with K_d values of 1.9 nM (A55T) and 1.5 nM (A57T) compared to 59 nM (WT, p <0.001, unpaired t-tests, df = 6). In principle, two kinetic mechanisms can explain this striking result, which is central to disease pathogenesis – either the mutant RBP-STRA6 complex dissociates more slowly or the mutant RBPs bind the receptor more rapidly. To distinguish these possibilities, we measured the release of RBP from STRA6+ and control cells at 25°C and 37°C (Figure 6E) and calculated forward (k_on) and reverse (k_off) rate constants. As these data show, the major consequence of the mutations is to increase k_on by 25–50 fold (p< 0.001, unpaired t-tests, df = 42), with no significant change in k_off (Figure 6F and Table S4). The pathogenic RBPs thus bind STRA6 with much higher affinity than wild-type, yet carry little or no vitamin A.

DISCUSSION

Here, we identify RBP4 mutations as the cause for autosomal dominant MAC with incomplete penetrance and skewed maternal transmission. These findings demonstrate a new mode of inheritance in mammals, whereby phenotypic expression is governed by maternal genotype. Our conclusions are supported by linkage analysis, the discovery of independent alleles, evolutionary conservation, the established role of vitamin A in eye morphogenesis, and convergent biochemical, functional, modeling and clinical data which prove A55T and A57T proteins have impaired retinol binding, but resist renal filtration and interact strongly with STRA6. Together, these data provide a simple but elegant mechanism for disease pathogenesis.

A unified disease model

A55T and A57T RBPs act as dominant-negative proteins, most likely by blocking vitamin A delivery at the STRA6 receptor (Figure 7A). Mutant and WT proteins coexist in plasma (Figure 4) and are therefore both secreted. Following translation, A55T and A57T proteins may transiently bind vitamin A in the hepatocyte ER, but if so, are likely to lose a significant fraction of their retinol content in the amphipathic environments of the ER-Golgi compartment and bloodstream (Figure 5). They are otherwise stable and partner with TTR (Figure 3). At the target cell, mutant RBPs bind STRA6 receptors more avidly than wild-type (Figure 6), with faster association kinetics, increased affinity and thus longer net occupancy, creating a molecular restriction point. Consequently, delivery of vitamin A from holo-RBP is disrupted.

Figure 7. Model for disease pathogenesis, dominant inheritance and maternal effect on penetrance.

(A) RBP life cycle in wild-type (top) and heterozygous (bottom) individuals. In mutation carriers, A55T or A57T are co-secreted with WT proteins from the liver and/or extraembryonic tissues (yolk sac). Each RBP circulates in the maternal or fetal bloodstream in a stable complex bound to TTR, but most of the mutant proteins lack retinol. Upon reaching target tissues, the mutant RBPs bind STRA6 receptors with much higher affinity than WT, acting as dominant-negative particles that block vitamin A delivery. (B) Basis for maternal inheritance. Skewed penetrance arises from functional “bottlenecks” that occur at sequential RBP-STRA6 interaction sites in the placenta and fetal eye. Disruption of vitamin A transfer at both levels, coupled with low maternal dietary retinoids (orange and red lines), predispose the fetus to MAC disease when the trait is maternally transmitted. VAD, vitamin A deficiency.

When the RBP4 mutation is transmitted from the mother, this bottleneck effect is iterated twice – first, at the placenta, involving maternal-derived RBP, and later at the developing eye primordia, involving fetal-derived RBP (Figure 7B). In this setting, retinol delivery to fetal tissues may be dramatically reduced – and penetrance of eye phenotypes increased – compared to paternal transmission of the same mutation, creating a maternal inheritance pattern that resembles genomic imprinting, but does not involve chromatin or DNA modification. This model is supported by data showing that STRA6 is localized in the placenta and fetal eye (Bouillet et al., 1997; Kawaguchi et al., 2007) and that maternal RBP does not cross the placental barrier (Quadro et al., 2004) in mice. Furthermore, RBP is expressed in extraembryonic tissues that directly participate in retinol transfer across the maternal-fetal interface, including the visceral yolk sac (Johansson et al., 1997; Sapin et al., 1997; Soprano et al., 1986; Ward et al., 1997). Recently, STRA6 has been shown to mediate retinol efflux from cells as well as influx, loading extracellular apo-RBP with cytoplasmic vitamin A (Kawaguchi et al., 2012). This bidirectional mode may be critical during early development, as RBP originating from the visceral yolk sac can, in principle, ferry retinol stepwise between different STRA6+ cells. Because the mutant RBPs are predicted to disrupt STRA6 docking on both sides, this relay mechanism may be highly sensitive to dominant-negative effects. The labyrinthine zone of the chorioallantoic placenta, for example, is a major site of maternal-fetal exchange that strongly expresses STRA6, but not RBP (Bouillet et al., 1997; Johansson et al., 1997).

When transmitted from the father, the RBP4 mutation can only disrupt vitamin A transfer beyond the placenta. Consequently, the severity of fetal VAD, and the genetic penetrance from males, should be comparatively low. Clinical phenotypes may only be expressed when vitamin A supplied to the placenta is diminished, notably in twin gestation (individual VI-2), where retinol input is divided between two embryos.

Structural basis for enhanced STRA6 binding

RBP is the archetypal lipocalin, an ancient protein family represented in nearly all life forms, including mammals, invertebrates, fungi and eubacteriae (Flower, 1996; Newcomer and Ong, 2000). Its ligand pocket is formed by eight anti-parallel beta strands (A-H) with alternating hydrophilic and hydrophobic amino acids, the latter stabilizing retinol. The orientation of the A-B loop, specifically G34-L35-F36-L37, is the only major structural difference between apo- and holo-RBP crystals at neutral pH (Zanotti et al., 1993). Threonine substitutions at Ala55 or Ala57, conserved sites deep within the pocket, impair retinol binding and, paradoxically, enhance STRA6 binding. Because these sites are located in the interior of the protein and thus unlikely to contact STRA6, the mutations must increase receptor binding indirectly, by altering RBP conformation. The striking decrease in K_d is driven by a large increase in the association rate constant (k_on) with no apparent change in dissociation kinetics (k_off). While relatively unusual (Anderson et al., 1998) a small number of protein-receptor affinity mutations are known to specifically affect k_on (Lahti et al., 2011; Lengyel et al., 2007). Our findings strongly suggest that RBP-STRA6 docking involves a conformational adaptation of RBP and that this initial step, rather than diffusion, limits the binding reaction, consistent with a selected-fit model (Weikl and von Deuster, 2009). Complementary changes in STRA6 folding may further stabilize the ligand-receptor complex.

The RBP lipocalin undergoes reversible transformation to a molten globule state as pH or solvent polarity is reduced (Calderone et al., 2003; Greene et al., 2006). This cooperative unfolding has been proposed to occur naturally in the local acidic environment at the cell surface, favoring retinol release (Bychkova et al., 1998) and may be potentiated by interaction with STRA6. We propose that the A55T and A57T mutations, by altering the shape, polarity and hydrophilicity of the retinol pocket, lower the activation energy for this transition. Consequently, a significant fraction of the mutant RBP population may exist in a partially melted state under normal physiological conditions. These molecules, which may resemble wild-type intermediates in the RBP-STRA6 binding reaction, presumably account for the enhanced retinol release observed in the presence of organic solvents, surfactants or phospholipid vesicles. Indeed, retinol dissociates from the mutant RBPs with biphasic kinetics, in PBS following addition of Tx-100 and DOC, consistent with the existence of ≥2 discrete holo conformations (Figure 5D).

Despite their increased forward reaction rates, the mutant apo RBPs appear to undock normally from STRA6 (Figure 6E). Likewise, mutant apo and holo RBPs bind TTR with an intrinsic affinity similar to wild-type (Figure 3) and are thus retained in carrier plasma (Figure 4). Indeed, the enhanced binding of mutant RBP to STRA6+ cell surfaces may in part explain the 1:2 ratio of mutant-to-WT protein in carrier plasma, which cannot be accounted for by unequal urinary loss (Figure S6). RBP normally contacts TTR via three external loops (Figure 2D), which form the opening to the retinol pocket, and the C-terminus (Newcomer and Ong, 2000). Although the structural details are not known, these same features are likely to mediate the interaction between RBP and STRA6, allowing retinol to exit (Kawaguchi et al., 2008). For steric reasons, the RBP-TTR complex must dissociate before receptor binding can occur. This step is driven by a 20-fold difference in the affinity of RBP for STRA6 versus TTR. The alanine substitutions affect RBP binding to former but not the latter. Structural studies of the mutant proteins may shed light on the conformational steps necessary for STRA6 docking and dissociation, and vitamin A release.

RBP4 mutations, diet and vitamin A physiology

Among organs, the eye is most frequently affected in animal models of vitamin A deficiency (Hale, 1933; See and Clagett-Dame, 2009; Warkany and Schraffenberger, 1946; Wilson et al., 1953). Our findings are consistent with this pattern. Despite a global reduction in vitamin A available to the embryo, phenotypes in Family 1 are limited to the eye. Likewise, in humans, total loss of RBP4 is only associated with night blindness, retinal dystrophy and chorioretinal coloboma (Biesalski et al., 1999; Cukras et al., 2012). Given the central role of retinoids in light perception, this unique sensitivity is striking and may reflect an evolutionary origin of RA signaling in the visual system (Campo-Paysaa et al., 2008; Drager et al., 2000).

In addition to retinol, other forms of vitamin A (principally retinyl esters) are delivered to the placenta via chylomicron lipoprotein particles (D'Ambrosio et al., 2011; Wassef and Quadro, 2011). Indeed, 25% of postprandial retinoids, including retinyl esters (RE) and α/β-carotenoids, travel directly to extrahepatic tissues from intestinal enterocytes via this parallel system, with no involvement of RBP (Goodman et al., 1965). However, because chylomicron RE are rapidly cleared (Berr, 1992), RBP accounts for 95–99% of circulating retinoids in the fasting state (Soprano, 1994). Accordingly, the extent and timing of maternal RE consumption during pregnancy, along with other genetic and/or environmental modifiers, may account for the variable penetrance. For women carrying an RBP4 mutation, careful dietary supplementation with extra vitamin A in divided doses may be indicated to minimize risk of congenital eye malformations in offspring.

Recent nutritional studies showed that Rbp4 −/− mouse pups born from Rbp4 −/− dams are normal when mothers were fed diets replete with retinyl esters (Quadro et al., 2005). However, these pups developed microphthalmia or anophthalmia in the absence of dietary retinoids, and the severity was correlated with maternal vitamin A status. In our study, phenotypes were roughly correlated with biochemical effects (Figure 5). Thus, both affected males in Family 2 (A55T) had neurodevelopmental delay in addition to anophthalmia. The discovery of genes that modify plasma retinoid levels, apart from RBP4 and TTR (Mondul et al., 2011), may shed more light on this disease.

Nutritional mechanism for maternal inheritance of human genetic disease

Maternally skewed inheritance has been reported for other birth defects, including congenital heart disease (Burn et al., 1998; Nora and Nora, 1987), but the molecular basis is unknown. One study of scoliosis identified gestational hypoxia as an environmental factor that disrupts FGF signaling and somitogenesis, increasing penetrance of Notch pathway defects (Sparrow et al., 2012). Genetic vitamin A deficiency has been previously suggested as a potential factor for eye malformations (Hornby et al., 2003). Dominant-negative RBP4 alleles provide a further example of gene × environment effects. Unlike other modes of maternal inheritance, e.g. transmission of ooplasmic mRNA, mitochondrial DNA mutations or genomic imprinting, these alleles affect fetal and maternal metabolism at a functional level. The sex-specific penetrance has a physiological basis. Our findings highlight the importance of maternal-fetal nutrition and may apply broadly to congenital disease.

EXPERIMENTAL PROCEDURES

Clinical data

Human studies were approved by the University of Michigan (UM), UC Davis and Einstein Medical Center Institutional Review Boards, and informed consent was obtained from all subjects. Eye exams, fundus photography and magnetic resonance imaging (MRI) were performed at UM (Table S1). Blood tests for retinol, RBP and transthyretin (prealbumin) were performed on carrier samples collected after a 12 hr fast (Table S3). Details are provided in the Extended Clinical Description. Mass spectrometry of clinical samples is detailed in the Extended Experimental Procedures.

Genetic Analysis

Family 1 genotypes were determined at 51 simple sequence length polymorphism (SSLP) and 6070 single nucleotide polymorphism (SNP) loci using blood, saliva or buccal DNA. SNPs were assessed using the HL12 BeadChip platform and BeadStudio software (Illumina, San Diego, CA). Genetic mapping was performed in three steps. Exclusion tests were performed by comparing probands using SSLP markers flanking 23 candidate loci (Table S2). Multipoint linkage analysis was then performed on a core pedigree consisting of all living affected individuals, obligate carriers and spouses (n = 20) using MERLIN v1.1.2 (Abecasis et al., 2002). Finally, linkage analysis was extended to include all collected (n = 33, Figures S1C and D) and nodal family members. LOD scores from two subpedigrees (Figure S1D) were summed, discarding duplicate phenotypic information (Bellenguez et al., 2009) and applying an AD inheritance model with uniform or sex-specific penetrance, estimated from the pedigree.

To identify RBP4 coding variants, we screened 75 unrelated MAC probands and 307 controls (NINDS panel) by PCR Sanger sequencing (Table S5), and queried the EVS exome variant database. Chromosome 10q haplotypes of Families 2 and 3 were compared using the Omni1-Quad SNP platform (Illumina).

RBP secretion and TTR interaction assays

Parallel HeLa cultures were transfected with pUS2-RBP^HA vectors expressing wild-type, mutant (A55T, A57T, G73D, I41N) or ER retention (WT^KDEL) human RBP proteins with an N-terminal hemagglutinin (HA) epitope, or control plasmid (Table S6). After 48 hrs, conditioned media (CM) and cell lysates were electrophoresed through native or denaturing polyacrylamide gels and compared by HA Western analysis. To evaluate RBP multimerization, CM was crosslinked in 0.5% (v/v) glutaraldehyde for 30 min, boiled with or without 2 mM βME (2-mercaptoethanol), and electrophoresed. To assess RBP-TTR binding in culture, HeLa cells were cotransfected with pUS2-TTR^myc and WT or mutant pUS2-RBP^HA plasmids. Secreted RBP^HA complexes were immunopurified from CM with anti-HA agarose beads (Sigma, St Louis, MO), washed in PBS, and tested for TTR by Western analysis. To fully assess RBP-TTR binding in vitro, reciprocal surface plasmon resonance (SPR) assays were performed using a Biacore T100 system (GE Healthcare, Piscataway, NJ) and biotin capture chip with human plasma TTR (Sigma) and HA- or polyhistidine-tagged RBP purified from HeLa CM. Molecular cloning, cell culture, protein biochemistry, SPR and data analysis are detailed in the Extended Experimental Procedures.

Retinol binding assays

To assess retinol binding to RBP in culture, HeLa cells were transfected with WT or mutant pUS2-RBP^HA plasmids in Dulbecco’s Modified Eagle media (DMEM) containing 10% delipidated fetal bovine serum (FBS), metabolically labeled with 6 µCi/ml ³⁵S-methionine and -cysteine in serum-free DMEM for 1 hr, and exposed to 8.25 µCi/ml ³H-retinol (NEN Perkin-Elmer, Waltham, MA) for an additional 3 hrs. Secretion of ³⁵S-labeled RBP in the CM was assessed by gel electrophoresis and autoradiography. Radiolabeled RBP^HA was immunopurified from CM using anti-HA agarose beads, washed 3 times in PBS containing 1% Triton X-100, 0.5% sodium deoxycholate (DOC), and eluted in 2% sodium dodecyl sulfate (SDS). The ³H/³⁵S isotope ratio was measured by liquid scintillation counting (LSC) and normalized to WT.

For in vitro titration assays, recombinant apo RBP^HA was immunopurified from serum-free HeLa CM, eluted with HA peptide (Anaspec, Fremont, CA) and dialyzed into PBS. Homogeneity was verified by gel electrophoresis. Equal amounts of WT, A55T or A57T RBP^HA proteins were loaded with 0 to 5 µM fresh all-trans retinol for 1 hr in PBS. Binding was quantified by retinol fluorescence (330 nm excitation, 460 nm emission) enhancement (Cogan et al., 1976) using a microplate reader. To assess binding in nonpolar or amphipathic conditions, parallel assays were performed in PBS with 0 to 50% ethanol; 1% Triton X-100, 0.5% DOC for 0 to 90 min; or 0.1% α-L-phosphatidyl-choline vesicles dispersed in 5% n-butanol.

STRA6-RBP binding

Radioligand assay

Immunopurified ³⁵S-labeled apo WT, holo WT, A55T or A57T RBP^HA (5 pM at 1.2 × 10⁷ cpm/pmol specific activity) was added, with or without an 8- or 250-fold excess unlabeled holo-WT competitor, to paired HEK293T cultures, transfected with pUS2-STRA6^myc or pUS2 vector plasmid DNA. After 1 hr at 37°C, the cells were gently washed with prewarmed PBS (Kawaguchi et al., 2007) and the amount of bound ³⁵S was determined by LSC. Receptor-specific binding was calculated by subtracting the vector control. STRA6^myc expression was verified by myc immunofluorescence and STRA6 Western analysis.

Equilibrium and kinetic analysis

STRA6+ or control HEK293T cells were plated on poly-D-lysine (PDL) coated dishes and incubated with CM containing 0–80 µg/mL A55T, A57T or WT apo RBP^HA and 0.5% bovine serum albumin (BSA) for 90 min at 37°C. For K_d analysis, monolayers were washed with ice-cold Hanks balanced salt solution (HBSS) and bound RBP was eluted with 25 mM glycine HBSS pH 3. For kinetic analysis, monolayers were washed with DMEM 0.5% BSA, and dissociation was followed at 25°C or 37°C by sampling the media at time points from 0 to 90 min. The concentration of RBP^HA was determined by an enzyme-linked immunosorbant assay (ELISA) (Figure S7). The immunoassay, saturation binding and kinetic methods, and quantitative analysis are detailed in the Extended Experimental Procedures.