Dimerization deficiency of enigmatic retinitis pigmentosa-linked rhodopsin mutants

Retinitis pigmentosa (RP) is a blinding disease often associated with mutations in rhodopsin, a light-sensing G protein-coupled receptor and phospholipid scramblase. Most RP-associated mutations affect rhodopsin's activity or transport to disc membranes. Intriguingly, some mutations produce apparently normal rhodopsins that nevertheless cause disease. Here we show that three such enigmatic mutations—F45L, V209M and F220C—yield fully functional visual pigments that bind the 11-cis retinal chromophore, activate the G protein transducin, traffic to the light-sensitive photoreceptor compartment and scramble phospholipids. However, tests of scramblase activity show that unlike wild-type rhodopsin that functionally reconstitutes into liposomes as dimers or multimers, F45L, V209M and F220C rhodopsins behave as monomers. This result was confirmed in pull-down experiments. Our data suggest that the photoreceptor pathology associated with expression of these enigmatic RP-associated pigments arises from their unexpected inability to dimerize via transmembrane helices 1 and 5.


Supplementary Figure 8
Model of an opsin dimer with a TM4/TM5-TM4/TM5 interface. Opsin monomers (PDB 4J4Q) were docked using the HADDOCK web server by applying constraints to residues in TM3 and TM4. The opsin monomers are shown in black and magenta. Residues V209 and F220 are shown as CPK representations in green and yellow, respectively. The top and bottom views correspond to the exoplasmic and cytoplasmic side of the dimer, respectively.

Supplementary Figure 9.
The figure is shown on the next page.
A parallel rhodopsin dimer with TM5 at the interface. A. Docking statistics. The HADDOCK web server was used to dock opsin monomers (PDB 4J4Q) by applying constraints to residues in TM5 and TM6. The dimer models obtained were grouped into 10 clusters according to the parameters shown in the bar chart (normalized to the highest value of each parameter amongst the clusters; the clusters are numbered 1-11, but cluster-3 did not have a high enough score to be placed in the top 10). B. Dimer models. Representative dimer models (side and top views, monomers in grey and blue) from each of the clusters, presented in order of their score. Clusters 6 and 9 have similar parallel orientations of the constituent monomers and feature the residues of interest (V209 and F220) at the dimer interface (designated dimer interface 2). Cluster 6 had a higher score and was therefore chosen for further discussion in the main paper.

Supplementary Tables
Supplementary Table 1 Fit M is the molar mass of the functional scramblase deduced from the fit, using ε=0.472 nm as the cross-sectional radius of a phospholipid 1 . The standard error associated with the fit is indicated.
#Several combinations of dimer and higher order multimers could account for M~100,000 g/mol for the Ops sample. For example, a mixture of dimers (75%) and tetramers (25%) would yield M~100,000 g/mol. While it is also possible that a fraction of Ops is inactive as a scramblase, resulting in lower α values, we consider this to be unlikely as the protein is fully active in vitro as a GPCR ( Figure 3, main paper) and no aspect of our protein expression/purification procedure results in inactivation as can be seen from the fact that each of the three RP Mutants reconstitutes as a monomer, i.e. with close to the highest α value possible. $ The molar mass of an opsin monomer is 41,700 g/mol.

Supplementary Notes
Supplementary Note 1

Scramblase activity assay (analysis of fluorescence traces).
To measure lipid scrambling, large unilamellar vesicles (LUVs) are symmetrically reconstituted with a trace quantity of fluorescent NBD-phospholipid. The vesicles are treated with dithionite, a membrane-impermeant reagent that irreversibly eliminates the fluorescence of all NBDphospholipids in the outer leaflet 2 . As spontaneous translocation of phospholipids is not detectable on the time scale (<10 min) of our experiments 3 , dithionite treatment of protein-free LUVs is predicted to result in 50% reduction in fluorescence. For samples where every LUV has a scramblase, the predicted reduction in fluorescence on dithionite treatment is 100% as NBD-phospholipids can exchange between the inner and outer leaflet because of scramblase activity. Supplementary Fig. 2 shows examples of fluorescence reduction traces for vesicles reconstituted with different amounts of wild-type (WT) opsin, Ops* (a conformationally distinct state of the protein resulting from the point mutation M257Y) 4 , and the three RP-associated opsin mutants. We previously used a double-exponential function to fit the fluorescence reduction traces and found that the slow component of the fit was typically more than an order of magnitude slower than the fast component and represented a minor fraction of the total fluorescence change 4 . To fit the traces more reliably by reducing the number of fitting parameters, we now approximate the slow component as a variable or constant line as others have done previously 2 , thereby reducing the number of fitting parameters from 4 (for a double-exponential) to 3 or 2 (for a mono-exponential with a variable or fixed line, respectively). We find that all fluorescence reduction traces are well described by a combination of a mono-exponential decay function with a half-life of ~15 sec, and a linear component with a slow decay rate (Supplementary Table 1, columns 2 and 3). The half-life of the exponential function was essentially unchanged when we performed the fit using a constant line with a negative slope = 0.0001 sec -1 in place of a variable line (Supplementary  Table 1, compare columns 4 and 2).
The exponential decay is the main feature of the traces and it describes the dithionite reduction reaction. The half-life of the exponential decay is the same for protein-free liposomes and scramblase-active proteoliposomes, indicating that the dithionite reaction is the rate-limiting step of the assay. Thus, it is possible only to provide a lower limit for the rate of lipid scrambling. For 176-nm diameter vesicles ( Supplementary Fig. 4) that contain ~280,000 phospholipids and a single functional opsin scramblase, this indicates a transport rate of >10,000 lipids per second per scramblase. The molecular basis for the slow linear decay component of the fit is not known, but it is unlikely to be due to leakage of dithionite into the vesicles as NBD-glucose trapped within the vesicle lumen is stably protected from dithionite 4 . As the linear decay component is the same for both proteinfree liposomes and proteoliposomes, it is clearly not related to the scrambling process, and will not be considered further.
The half-life of fluorescence decay upon dithionite addition was slightly but significantly larger for vesicles reconstituted with the F45L protein compared with vesicles reconstituted with the other opsin variants (Supplementary Table 1 and Supplementary   Fig. 3). While it is not possible to decompose reliably the F45L traces to identify the fast and slow rate processes, this result nevertheless suggests that scrambling by F45L opsin may be impaired such that it occurs on a similar time scale as the dithionite reduction reaction.

Supplementary Note 2
Scramblase activity assay (protein dependence). The scramblase assay reports the extent of fluorescence reduction. For protein-free liposomes, the extent of reduction is 45.0 ± 0.8 % (mean ± SEM, n=6), identical to the value of 46.7 ± 0.9% that we reported previously 4 , and close to the expected value of 50% for LUVs that are symmetrically labeled with NBD-phospholipids.
For proteoliposomes, the extent of reduction depends on the protein to phospholipid ratio (PPR) of the sample as the amount of protein used for reconstitution dictates whether an individual vesicle in the sample will contain a functional scramblase. The maximum extent of reduction is expected to be 100%, corresponding to high PPR values where every vesicle should be equipped with a scramblase. However, we obtained a maximum of 82.5 ± 0.5% (n=14), identical to the result that we reported previously (82.6 ± 0.5%) 4 . This indicates that a fraction of the vesicles is consistently refractory to reconstitution, an observation that has been made by us and by others in a variety of proteoliposome reconstitution experiments [4][5][6][7][8][9] . The origin of the refractory population is not known, but as it corresponds to a highly reproducible fraction of the vesicles it may be an intrinsic property of the system. The refractory pool corresponds to 35% of the vesicle population, as 17.5% of the NBD-phospholipid fluorescence is inaccessible to dithionite.
The extent of fluorescence reduction exceeding the value obtained for protein-free liposomes is a measure of the fraction of vesicles that contains at least one functional scramblase. Assuming that opsin reconstitutes randomly into vesicles, this fraction (or equivalently p(≥1 scramblase), the probability that a particular vesicle in the ensemble possesses at least one functional scramblase) can be calculated using Poisson statistics. We previously described a simple version of this calculation, in which we did not consider the refractory pool and assumed that all vesicles were spheres with the same radius.
We now present an advanced calculation that takes into account the size distribution of the vesicles (Supplementary Fig. 4) and uses PPR* rather than the measured PPR to account for the refractory pool. Similar calculations have been described previously 10 .

Supplementary Note 3
Analysis of the functional reconstitution of scramblases. We obtained an analytical expression for p(≥1 scramblase) in terms of PPR* as follows. This expression was used to fit the experimental data enabling calculation of the molar mass of the functionally reconstituted scramblase.
1. We define µ = average number of functional scramblases per vesicle, i.e., µ = p/L, where p = total number of functional scramblases and L = total number of vesicles. Assuming that opsin reconstitutes randomly into vesicles, the fraction of vesicles that has zero functional scramblases is f = e -µ = e -p/L based on Poisson statistics. As the vesicles are not uniform in size, we note that for the subset of Lj vesicles of radius rj, fj = e -µ j = e -p j /L j.
2. Assuming that the probability of a scramblase inserting into a vesicle is proportional to the relative surface area of that vesicle, i.e. surface area of the vesicle ÷ total surface area of all vesicles 10 , then for the subset of Lj vesicles of radius rj, pj = p•(4πrj 2 •Lj)/A. Thus, µj = pj/Lj = 4πrj 2 (p/A) = 4πrj 2 z, where A is the total surface area of all vesicles in the sample and z=p/A.
3. Using the information from steps 1 and 2, and summing over all vesicles, we can write p(≥1 scramblase) = 1 -Σwjfj/Σwj, where wj is a weighting factor and j goes from 1 to N (the total number of vesicles). Assuming a Gaussian frequency distribution of vesicle sizes, with a mean radius € r and standard deviation σ ( Supplementary Fig. 4), and integrating from -∞ to +∞ (permissible when the minimum and maximum r values are much smaller and larger, respectively, than

4.
The term z is related to PPR as follows: (i) PPR = m/λ, where m is the mass of reconstituted protein in grams and λ is the amount of lipid in moles; (ii) p = mNA/M, where NA is Avogadro's number, and M is the molar mass of the functional scramblase; (iii) A = λNAπε 2 /2, where ε is the cross-sectional radius of a phospholipid in nm, and the factor 2 in the denominator is necessary because only half of the total lipids contribute to the outer surface of the vesicles; (iv) Thus, z = p/A = PPR•(2/Mπε 2 ).

6.
We used dynamic light scattering to measure the size distribution of the vesicles. Fitting the light scattering data to a Gaussian distribution yielded a mean radius € r = 88 nm with a standard deviation σ = 28 nm ( Supplementary Fig. 4).

Using
€ r = 88 nm, σ = 28 nm and ε = 0.47 nm 1 , equation 3 can be written as 1+784αx (Equation 4) 8. We obtained p(≥1 scramblase) by transforming end-point fluorescence reduction data from scramblase activity assays as follows: p(≥1 scramblase) = (F -Fo)/(Fmax -Fo), where F is the percentage fluorescence reduction for a particular sample 400 s after adding dithionite, Fo is the percentage reduction obtained with protein-free liposomes (~45%, see above) and Fmax is the maximum percentage reduction observed at high PPR values where all vesicles are expected to have at least one functional scramblase (~82.5%, see above). We also scaled our measured PPR values by a factor of 0.65 to account for the pool of vesicles that is refractory to reconstitution (the fraction of refractory vesicles is 0.35; see above); thus, x = PPR* = (measured PPR)÷0.65.

9.
Analysis of the data for Ops (WT) and Ops* (M257Y). Experimental data and fits are shown in Supplementary Fig. 5, and the corresponding fit constant α and the deduced molar mass of the functional scramblase are presented in Supplementary Table 2. Both proteins reconstitute minimally and predominantly as dimers (predicted molar mass of 83,700 g/mol). Data fitting (Supplementary Table 2) reveals a somewhat larger reconstituted unit for Ops(WT) compared with Ops*, consistent with the possibility that Ops (WT) may partly multimerize.