Vertebrate photoreceptors contain a unique tetraspanin protein known as ‘retinal degeneration slow’ (RDS). Mutations in the RDS gene have been identified in a variety of human retinal degenerative diseases, and more than 70% of these mutations are located in the second intra-discal (D2) loop, highlighting the importance of this region. Here we examined the conformational and thermal stability properties of the D2 loop of RDS, as well as interactions with ROM–1, a non-glycosylated homolog of RDS. The RDS D2 loop was expressed in Escherichia coli as a fusion protein with maltose binding protein (MBP). The fusion protein, referred to as MBP–D2, was purified to homogeneity. Circular dichroism spectroscopy showed that the wild-type (WT) D2 loop consists of approximately 21% α–helix, approximately 20% β–sheet and approximately 59% random coil. D2 loop fusion proteins carrying disease-causing mutations in RDS (e.g. R172W, C214S, N244H/K) were also examined, and conformational changes were observed (compared to wild-type D2). In particular, the C150S, C214S and N244H proteins showed significant reductions in α–helicity. However, the thermal stability of the mutants was unchanged compared to wild-type, and all the mutants were capable of interacting with ROM–1, indicating that this functional aspect of the isolated D2 loop remained intact in the mutants despite the observed conformational changes. An I–TASSER model of the RDS D2 loop predicted a structure consistent with the circular dichroism experiments and the structure of the conserved region of the D2 loop of other tetraspanin family members. These results provide significant insight into the mechanism of RDS complex formation and the disease process underlying RDS-associated retinal degeneration.
Retinal degeneration slow (RDS) is a photoreceptor-specific glycoprotein that is required for the formation and maintenance of the outer segment (OS) rim region in rods and cones [1, 2]. The importance of RDS for OS morphogenesis is supported by the observation that rds−/− mice lack OSs  and exhibit no retinal function . As a member of the tetraspanin protein super-family, RDS exhibits several highly conserved structural characteristics, including four membrane-spanning domains, small (D1/EC1/LEL1) and large (D2/EC2/LEL2) extracellular/intra-discal loops, and cytoplasmic N- and C–termini [5, 6].
In general, tetraspanins form large functional membrane microdomains that are assembled via protein–protein interactions between tetraspanins and interactions with other proteins . RDS is known to interact with itself and its non-glycosylated homolog rod outer segment membrane protein 1 (ROM–1). Both in vivo and in vitro studies have demonstrated that non-covalent interactions between RDS and ROM–1 result in formation of homo- and hetero-tetrameric complexes [8, 9], and, in the OS, these tetramers are linked together through inter-molecular disulfide bonds to form octamers and higher-order oligomers that are crucial for disc rim formation [10-12].
RDS has two main functional domains that have been studied so far: the C–terminal domain and the large D2 loop. The cytoplasmic C–terminal domain is highly variable among tetraspanin proteins and is often very short. The RDS C–terminal domain is intrinsically disordered , and is thought to play a role in OS targeting , membrane fusion  and interaction with binding partners such as calmodulin and melanoregulin [16, 17].
The tetraspanin D2 loop, which is extracellular in most cases but either extracellular (cone) or intra-discal (rods) in the case of RDS and ROM–1, is divided into a highly conserved region and a hypervariable region [18, 19]. The D2 loop of RDS consists of 142 amino acids: the highly conserved region corresponds approximately to residues 125–167 and residues 250–263, while the intervening amino acids (168–249) comprise the hypervariable region [19, 20]. The D2 loop is known to play an important role in the protein–protein interactions that are necessary for disc formation and stabilization [8, 21, 22], and has specifically been mapped as the region responsible for interactions between molecules of RDS and interactions between RDS and ROM–1 . Tetraspanins exhibit highly conserved D2 loop cysteines that are involved in formation of intra-molecular disulfide bonds. RDS has six such cysteines (C165, C166, C213, C214, C222 and C250), and mutagenesis studies  suggest that mutation of any of these six cysteines results in a failure to form normal complexes, probably due to improper folding. RDS also contains a 7th cysteine in the D2 loop (C150), which is important for the formation of inter-molecular disulfide bonds and assembly of large complexes described above. In a heterologous expression system and transgenic animal model, RDS with a cysteine to serine substitution at position 150 (C150S RDS) was unable to form inter-molecular disulfide bonds [10, 22]. Animals that expressed only C150S RDS (no wild-type RDS) were incapable of forming OSs, and the mutation caused a dominant-negative degeneration in cones (i.e. even in the presence of wild-type RDS) [10, 23, 24], underscoring the importance of this cysteine in RDS function and the differential role of RDS in rods versus cones.
Mutations in RDS cause a variety of retinal degenerations, including rod-dominant autosomal dominant retinitis pigmentosa and multiple classes of cone-dominant macular dystrophy [25, 26]. Over 70% of these mutations are located in the D2 loop (http://www.retina-international.org/files/sci-news/rdsmut.htm), highlighting the functional importance of this region. Evidence from in vitro and animal studies has led to the hypothesis that mutations associated with rod defects are likely to be null alleles and result in disease via haploinsufficiency, while mutations causing cone-dominant disease are likely to be gain-of-function alleles [10, 24, 27-29]. However, the mechanism underlying the development of cell type-specific disease in the case of RDS mutations is not understood. The goal of the present study was to understand the effects of various disease-causing mutations on the structure of the D2 loop. We consider the mutations C214S and N244K, which cause autosomal dominant retinitis pigmentosa [30, 31], R172W and N244H, which cause macular dystrophy [26, 32], and C150S, which does not form inter-molecular disulfide bonds (Table 1). We show that these mutations do not alter the thermal stability of the D2 loop, but do result in substantial changes in the secondary structure of the D2 loop. Interestingly, the pattern of these changes does not correlate with the cell type affected by the mutations.
Table 1. Phenotypes of the RDS mutations characterized in this study
A fusion protein comprising MBP and the wild-type (WT) RDS D2 loop (Fig. 1A,B) was generated and used as a base to create several MBP–D2 loop mutants. Purified MBP protein migrated to a position corresponding to approximately 37 kDa, while the wild-type MBP–D2 fusion protein migrated to approximately 56 kDa, in line with the predicted size of 37 + 19 kDa (Fig. 1C). Western blot analysis confirmed that our polyclonal antibody against the RDS D2 loop recognized the fusion proteins but not MBP (Fig. 1D). MBP–D2 loop fusion proteins carrying the mutations C214S, C150S, R172W, N244H or N244K (Table 1) were also expressed and purified, and migrated equivalently to MBP–D2 (WT) (Fig. 1E). Occasionally, a small amount of free MBP was detected in the fusion protein preparations, which may reflect either partial degradation during the purification procedure or formation of an early truncated product. The small amounts of these degradation/truncation products represent < 10% of the total protein, and thus do not significantly affect the results reported.
MBP–D2 interacts with ROM–1
As the D2 loop is isolated from the rest of the RDS polypeptide when expressed as an MBP fusion protein, we wished to confirm that the fusion protein exhibits similar functional characteristics to the endogenous protein. As RDS interacts with itself and ROM–1 through non-covalent interactions in the D2 loop, we examined the interaction between MBP–D2 and ROM–1/RDS from freshly prepared retinal extract. Immunoprecipitation with antibodies against the C–terminus of ROM–1 (ROM1–CT) showed that MBP–D2 (WT) retained the ability to bind endogenous ROM–1 (Fig. 2A). Similarly, antibodies against the C–terminus of RDS (RDS–CT) also precipitated MBP–D2 (Fig. 2B). These results were confirmed when mutant MBP–D2 proteins were incubated with retinal extracts and immunoprecipitated using ROM1–CT antibody indicating that the mutations did not affect this protein–protein interaction (Fig. 2C), and suggesting that the MBP–D2 fusion proteins preserve the ability to interact with native proteins. Although our data suggesting that MBP–D2 C214S binds ROM–1 appear to contradict previous evidence showing that the full-length C214S protein does not have the ability to bind ROM–1, this discrepancy probably arises due to the nature of the MBP–D2 fusion proteins. The MBP–D2 proteins used here for study of the secondary structure are soluble (rather than membrane bound as full-length RDS is), and the C214S mutation probably induces aggregation of the full-length membrane-bound protein to a much greater extent than for the MBP fusion protein due to interactions of the mis-folded C214S D2 domain in the full-length protein with the membrane.
Secondary structures of MBP–D2 fusion proteins
CD spectra were obtained for MBP and the MBP fusion proteins (Fig. 3A,B). The CD spectra are typical of proteins containing mixed α–helical and β–sheet secondary structures. The mutant MBP–D2 proteins yield CD signals that vary by differing extents compared to the WT spectra (Fig. 3A), indicating conformational changes between the mutant and WT fusion proteins. As all of the MBP fusion proteins are readily purified using amylose resin (Fig. 1E), and are recognized by MBP antibody (Fig. 1D), we infer that the conformational changes occur in the D2 portion of the fusion proteins. To analyze the secondary structures of the WT and mutant D2 loop proteins, each of the MBP–D2 spectra were subtracted from an ‘MBP only’ spectrum (shown for MBP–D2 in Fig. 3B) to generate spectra corresponding to the D2 loop portion of the fusion protein (green line in Fig. 3B). The mean subtracted CD spectra (n =3–5) for the WT and mutant D2 loops were analyzed using two algorithms, SELCON3 and CONTILL (cdpro software), to calculate the secondary structure content of each protein. The distributions of α–helix and β–sheet in each D2 loop are indicated in Fig. 3C. The wild-type D2 loop was approximately 21% α–helix, approximately 20% β–sheet and approximately 59% random coil. Several of the mutants, including C150S, C214S and N244H, showed a significant decrease in α–helicity in comparison to wild-type D2. Interestingly, N244K and R172W did not show as large a conformational change compared to the WT as the other mutants. In most, but not all cases, a loss of α–helicity was reflected in an increase in the percentage of β–sheet.
Conformational changes in the D2 loop as a function of temperature
Given the observed structural changes for the D2 mutants relative to WT, we next examined the effects of the mutations on the stability of the D2 loop. To analyze thermal stability, we investigated changes in the CD signal over a wide temperature range (10–80 °C) at 220 nm. Changes in the molar ellipticity of wild-type D2 at 10 °C increments are shown in Fig. 4A. It is evident that, by 80 °C, the majority of secondary structure has been lost (Fig. 4A). Molar ellipticity is plotted as a function of temperature in Fig. 4B, and the mid-point of the transition (Tm) for denaturation of MBP alone was approximately 50 °C (blue, Fig. 4B). In contrast, MBP–D2 (WT) showed two separate broad transitions, one at approximately 50 °C and a second at approximately 60 °C (red, Fig. 4B). Given that the first transition coincides with denaturation of MBP, it is likely that the second transition is due to unfolding of the D2 loop. Notably, all of the mutants show similar thermal denaturation curves to the WT D2 loop (Fig. 4C). Thus, although several of the mutations affect the secondary structure of the D2 loop, the mutants retain a folded core structure that shows similar thermal stability to the WT D2 loop. Interestingly, the two unfolding transitions are more apparent for some of the mutants, including the R172W, N244H and N244K mutants. This may indicate loss of an intermediate in the unfolding pathway, resulting in more cooperative unfolding transitions. Further studies arw necessary to elucidate the unfolding pathways of WT and mutant D2 proteins.
We used computational methods, in this case the I–TASSER server , to predict and model the 3D structure of the D2 loop. First, the secondary structure was predicted using the programs Jpred 3  and GOR4 . Second, the 3D structure was predicted using I–TASSER (iterative threading assembly refinement) . Jpred 3 predicts α–helices between amino acid residues 123–141, 150–164 and 251–254 in wild-type and mutant D2 loops. Using GOR4, the wild-type D2 loop was predicted to contain approximately 25% α–helix, approximately 23% β–sheet and approximately 53% random coil. These predictions are consistent with the secondary structure content determined by CD spectroscopy. A representative model is shown in Fig. 5. To generate this model, distance constraints were used to keep the three highly conserved intra-molecular disulfide bonds within range. However, regardless of whether distance constraints were used, the high-scoring models (Z values > –1.5 ) consisted of a similar orientation of the first 60 N–terminal residues (120–180). This region is predicted to consist of three α–helical segments that include residues 121–145, 149–163 and 172–181 as helices 1 (green), 2 (yellow) and 3 (orange) (Fig. 5). Furthermore, I–TASSER-generated models of the D2 loop consistently showed that the C–terminal 10–15 residues of the D2 loop are oriented in close proximity to helix 1. This results in the N- and C–terminal residues extending in similar directions, as expected for anchoring of the D2 loop to the membrane. C150, R172 and N244 are represented as red and blue stick models in Fig. 5. C150 is located near the turn between helices 1 and 2, in a position that appears to be accessible for formation of an inter-molecular disulfide bond. In addition, R172 and N244 are located on the same face of the protein, suggesting that they may be located in the same inter-molecular interface with a potential binding partner. The remainder of the D2 structure is not reliably predicted by the computational methods used. Overall, we provide a conservative estimate of the major structural features of the D2 loop that are consistently predicted by computational methods. Such a model will be useful in future studies to determine structure–function correlations of the D2 loop of RDS.
Here we used CD spectroscopy and bioinformatics methods to analyze the RDS D2 loop sequence and generate models of the structure of this critical region of the protein. The only tetraspanin whose D2 loop has a solved crystal structure is CD81  (a cell-surface receptor for hepatitis C); however, comparative analysis and structural predictions of a wide array of tetraspanins have indicated that some features of the CD81 D2 loop crystal structure are likely to be common across members of the protein family . Specifically, the CD81 D2 loop exhibits two α–helices in the N–terminal portion of the D2 loop (termed A and B) and one α–helix in the–C-terminal portion of the D2 loop (termed E) that are predicted to be conserved throughout all tetraspanins [19, 36]. In addition, the CD81 D2 loop has two additional α–helices (C and D) that are predicted to be conserved in some tetraspanins . The tetraspanin D2 loop is thought to have a mushroom-like structure, with the highly conserved helices A and E (which are antiparallel) serving as the ‘stalk’ while the remainder of the D2 loop serves as the ‘cap’ . The majority of this cap region corresponds to the hypervariable region of the D2 loop and is thought to confer protein-specific function. Although the RDS D2 loop is significantly larger than the CD81 D2 loop (88 versus 142 amino acids), this difference is entirely in the hypervariable region. Our predicted model of the RDS D2 loop (Fig. 5) exhibits a secondary structure that correlates well with that of CD81, i.e. helices 1–3 in Fig. 5 (green, yellow and orange) correspond in approximate size and location with helices A–C of CD81, and our structure identifies two helical regions in the C–terminal portion of the D2 loop that are similar in size and location to helices D and E of CD81.
These similarities may have significant implications for the mechanism of RDS complex assembly. In common with most other tetraspanins (including RDS), CD81 forms homodimers. The crystal structure of dimerized CD81 D2 loops predicts that the two subunits are held together by interactions between the two opposing A helices and by interactions between several residues in the B helix and several residues in the E helix of the opposing subunit . However, our previous work has indicated that the region required for assembly of RDS homodimers is C165–N182 , a region corresponding to helix 3 (orange, Fig. 5) in RDS and approximately to helix C in CD81. ROM–1/RDS interactions have been mapped to a region incorporating Y140–N182, roughly corresponding to helices 2 and 3 (yellow and orange, Fig. 5). These data suggest that, while the structure of the conserved region of the RDS D2 loop is consistent with that known or predicted for other tetraspanins [19, 36], the orientation and assembly of RDS complexes in the membrane diverge from those of other family members. This may arise due to the different nature of RDS/ROM–1 complexes compared to other tetraspanin complexes. For many tetraspanins, CD81 included, the core unit is a homo- or heterodimer that anchors the rest of the tetraspanin microdomain [19, 36, 37]. In contrast, the core unit of RDS complexes appears to be tetrameric in nature [11, 12]. The difference between RDS complex assembly and that of other tetraspanins is further emphasized by the presence of C150 in RDS. RDS and ROM–1 alone among tetraspanins form D2 loop-mediated inter-molecular disulfide bonds as part of their normal complex assembly.
More than 70% of RDS disease-causing mutations occur in the D2 loop, and there are a wide variety of associated degenerative phenotypes. Many of these mutations have been examined in heterologous expression systems and animal models, leading to the hypothesis that mutations may be grouped into those that result in null alleles and are associated with rod-dominant disease (such as autosomal dominant retinitis pigmentosa), and those that result in gain-of-function alleles and are associated with cone-dominant disease (such as macular dystrophy). Although this hypothesis may have some utility in guiding our understanding of genotype/phenotype correlations, the data presented here, together with the variation in patient phenotypes, suggests that such a dichotomous interpretation is too simplistic. Although some patients exhibit disease that truly targets rods or cones, many others exhibit disease that targets both cell types, such as cone/rod or rod/cone dystrophy. Here we have shown that there is no clear correlation between the effects of mutations on secondary structure and the target cell type. R172W and N244H are both associated with macular dystrophy or cone/rod dystrophy (cone-targeted) [32, 38, 39], but our results indicate that the secondary structure of the R172W D2 loop is similar to that of the wild-type D2 loop, while the N244H D2 loop exhibits reduced α–helicity compared to WT. Similarly, the percentage of α–helix in the N244K D2 loop is similar to that of the wild-type D2 loop, but the percentage in the C214S D2 loop is reduced, even though both cause autosomal dominant retinitis pigmentosa or rod/cone dystrophy [30, 40]. The only other study that has examined the secondary structure of RDS  reported very little change in secondary structure in the case of the P216L autosomal dominant retinitis pigmentosa mutation , although that study examined the secondary structure of the entire RDS polypeptide (not just the D2 loop), so it is difficult to compare results. These results suggest that assessing each mutation independently may be a better way to approach understanding of RDS-associated disease, and it is likely that, for many mutations, disease arises due to a combination of molecular defects.
The majority of the mutants show an increase in β–sheet content relative to the wild-type protein, particularly C214S. Mutants with increased β–sheet content may show an increased propensity for protein aggregation under certain conditions, as a result of the exposure or extension of β–sheet edges that are normally protected in native β–sheet-rich proteins . Such a propensity for increased aggregation may disrupt heteromeric protein–protein interactions, leading to a loss of function. C214S shows the most dramatic shift in secondary structure of all the mutations studied here, and previous data strongly suggest that C214S is a null allele [29, 43]. Given this alteration in structure, combined with the accompanying ablation of a highly conserved disulfide bond, it is not surprising that it has been shown that full-length C214S RDS does not bind ROM–1 in vitro or in vivo [22, 29]. In addition, it has been shown that full-length C214S does not form proper complexes, probably due to misfolding of the tertiary RDS structure as a result of ablation of a critical intra-molecular disulfide bond, leading to protein degradation and haploinsufficiency [22, 29]. Although we show here that the thermal stability of MBP–D2 (C214S) is not different from that of MBP–D2 (WT), and that MBP–D2 (C214S) can bind ROM–1, the structural changes arising from the C214S mutation may significantly impair the ability of the entire RDS protein to fold properly and to embed in the membrane during synthesis in vivo, thus leading to aggregation and degradation. The N244H/K mutations occur in a predicted helical region (Fig. 5), and both result in substitution of an uncharged amino acid by a positively charged amino acid. Given the similarity in charge between histidine and lysine, and the fact that both are commonly found in α–helices, it is not clear why one (N244H) results in a decrease in D2 helicity but the other (N244K) does not. Furthermore, this pattern of alteration in secondary structure does not follow what we have observed in other experiments . In common with C214S, heterologously expressed N244K does not bind ROM–1 or form higher-order complexes (only aggregates), but, in contrast to C214S, the secondary structure of N244K is relatively unchanged. In accordance with this lack of correlation between changes in secondary structure and disease type or molecular characteristics, N244H exhibits normal complex assembly and ROM–1 binding  but shows a decrease in D2 loop α–helicity.
One of our most interesting results is the case of R172W, which results in substitution of a positively charged amino acid with a bulky hydrophobic amino acid. In our CD experiments, we used two different algorithms (CONTINLL and SELCON3) to predict the percentage of α–helix or β–sheets in the D2 loop. In most cases, the two algorithms were in fairly close agreement (small error bars in Fig. 3C); however, in the case of R172W, one algorithm predicted a substantial drop in the percentage of α–helix (16.0% in WT versus 6.9% in R172W with CONTILL) while the other predicted virtually no change (27% in WT versus 24% in R172W with SELCON3). The reason for this difference is not clear. R172W causes macular dystrophy and butterfly-shaped pattern dystrophy in patients, and dominant, early-onset, cone-specific degeneration in transgenic mice. We have observed that R172W RDS traffics properly to the OS, retains the ability to bind ROM–1, and forms higher-order oligomeric complexes, and have been unable to determine the mechanism underlying the development of disease.
Despite the usefulness of these results, it is important to recognize limitations associated with the use of fusion constructs comprising only a portion of the protein of interest. Use of only a portion of the protein means that overall protein folding and assembly will be different for MBP–D2 fusion proteins compared to full-length RDS. Furthermore, the fusion proteins we are using are soluble, rather than membrane-bound (as is the case with full-length RDS), such that mutations that cause aggregation or mis-folding of the full-length protein may have a less severe effect on the soluble MBP–D2 proteins. Although these facts may mean that the MBP system is not useful for the study of RDS complex assembly (as proper folding and insertion in the membrane are necessary for oligomerization), complex assembly has been studied in detail elsewhere [11, 12, 44]. A benefit of using the D2 loop alone for the CD experiments is that changes in secondary structure are localized to a specific region of the protein, a result that is difficult to achieve with the full-length protein.
In conclusion, the present study indicates that the RDS D2 loop comprises a mixture of α–helix, β–sheet and random coil. We show that, while the RDS D2 loop exhibits similar secondary and predicted tertiary structure to other tetraspanins in the highly conserved region, this region is not where RDS–RDS and RDS–ROM–1 interactions are predicted to occur, in contrast to other tetraspanins. This divergence highlights critical differences in the role of RDS/ROM–1 compared to other tetraspanins: RDS and ROM–1 may play a role in organizing a tetraspanin microdomain, but, unlike most other tetraspanins, their function is primarily structural, and, as a result, RDS/ROM–1 complexes are much larger and of a different type than other tetraspanin dimeric complexes, i.e. they contain inter-molecular disulfide bonds.
Fusion protein cloning
A MBP fusion protein was generated that encoded the predicted D2 loop of mouse RDS (Phe120–Asn256)  joined to a MBP header via a PreScission protease site . The RDS D2 loop cDNA was amplified by PCR from murine RDS cDNA, and the amplicon incorporated 5′ (BamHI) and 3′ (HindIII) restriction sites for sub-cloning. This fragment was inserted into the pMal–c2 vector (containing the PreScission site) (New England Biolabs, Ipswich, MA, USA) . The C150S, R172W, C214S, N244H and N244K mutations were created via site-directed mutagenesis using a QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA, USA). Authenticity of the constructs was confirmed by sequence analysis.
Expression and purification of the fusion protein
Escherichia coli BL21 (DE3) cells carrying the MBP–D2 fusion protein construct were grown in Luria broth (LB) medium containing ampicillin (50 μg·mL−1). Expression was induced and cells harvested as described previously [21, 46]. MBP and all MBP–D2 fusion proteins were purified using amylose resin (New England Biolabs) as described previously . The purified protein was dialyzed overnight in buffer (20 mm Tris/HCl pH 7.4 and 0.2 m NaCl). Protein concentration was determined from the absorbance at 280 nm using extinction coefficients of 66.5 mm−1·cm−1 for MBP, 95 mm−1·cm−1 for wild-type, C150S, C214S, N244H and N244K D2 loops, and 100.5 mm−1·cm−1 for the R172W D2 loop. The expression and purity of the MBP and MBP–D2 fusion proteins was assessed by Coomassie blue staining and western blotting using antibody against MBP (Santa Cruz Biotechnology, Santa Cruz, CA, USA) and antibody against RDS-D2 (generated in-house and characterized previously ).
SDS/PAGE and western blot analysis
SDS/PAGE and western blot analysis were performed as described previously . Blots were incubated with primary antibodies (MBP, 1 : 2500 dilution; RDS-D2, 1 : 1000 dilution) for 2 h at room temperature. After washing in TBST (tris-buffered saline with 0.1% tween-20, 4×15 minutes), blots were incubated with horseradish peroxidase-conjugated secondary antibodies (1 : 25 000 dilution; KPL, Gaithersburg, MD, USA). Blots were developed using Super Signal West Dura Extended Duration chemiluminescent substrate (Pierce, Rockford, IL, USA), and images were captured using a kodak imaging station 4000r and kodak mi Software version 4.0.3 (Carestream Health, Rochester, NY, USA).
Purified MBP or MBP fusion proteins (approximately 5 μg) were incubated at 4 °C with freshly prepared retinal extract (approximately 100 μg) for 3–4 h in the presence of 25 μL Protein A beads (Sigma-Aldrich, St Louis, MO, USA) conjugated with ROM1–CT or RDS–CT antibodies (1 : 100 dilution, generated in house and characterized previously ). After adsorption, the protein beads were washed three times with extraction buffer (50 mm Tris-HCl, pH 7.5, 100 mm NaCl, 5 mm EDTA, 1% Triton X-100, 0.05% SDS, 2.5% glycerol and 1.0 mm phenylmethyl-sulfonyl fluoride), and proteins were eluted with Laemmli sample buffer (4% SDS, 20% glycerol, 0.004% bromphenol blue, 0.125 m Tris HCl) under reducing conditions before being subjected to SDS/PAGE and western blot analysis.
Circular dichroism spectroscopy
The CD spectroscopy experiments were performed using a JASCO J715 spectropolarimeter with a PTC–348WI Peltier temperature controller (Jasco, Tokyo, Japan) as described previously [47, 48]. Protein samples were dialyzed against 20 mm Tris/HCl (pH 7.4), 200 mm NaCl at 4 °C overnight prior to CD spectropolarimetry analysis. Spectra were obtained for protein (at approximately 6.5 μm) with wavelengths ranging from 270 to 200 nm using a 0.1 cm cuvette path length at 25 °C, and with the following parameters for the spectra measurements: 1.0 nm bandwidth, 1 nm resolution, 16 s response time and five accumulations. The results are expressed as mean molar ellipticity. Protein secondary structural contents were calculated using the cdpro software package  (http://lamar.colostate.edu/~sreeram/CDPro). Two programs available in the CDPro package, CONTINLL and SELCON3, were run using the same 40-protein reference library . The values from the resulting calculations were averaged to yield the percentages of α–helix, β–sheet and random coil for each protein sample.
Thermal denaturation experiments
Thermal denaturation curves were obtained by monitoring the CD signal at 220 nm over a temperature range of 10–80 °C. Thermal denaturation experiments were performed using the same protein concentrations used above for the wavelength scan in 20 mm Tris/HCl (pH 7.4), 200 mm NaCl. The CD signal was detected at 220 nm with the following parameters: 10–80 °C temperature range, 1 nm bandwidth, 0.2 °C resolution and a response time of 16 s. Tm values were obtained as previously described .
Protein threading was performed using i–tasser software . The PDB file generated was used for 3D molecular modeling of the D2 loop using PyMOL (http://www.pymol.org/).
We are grateful to Lori Gwyn for her help at the initial stage of this work. We thank Ruby Rahman for providing excellent technical assistance with CD spectroscopy, and Seok Ho Kim for site-directed mutagenesis work. This study was supported by grants from the US National Institutes of Health (EY018656 and EY10609 to M.I.N., and EY018512 to S.M.C.), the Foundation Fighting Blindness (to M.I.N.), and the Oklahoma Center for the Advancement of Science and Technology (to M.I.N., K.K.R. and S.M.C.).