The lipid transfer function of RDGB at ER-PM contact sites is regulated by multiple interdomain interactions

In Drosophila photoreceptors, following Phospholipase C-β activation, the phosphatidylinositol transfer protein (PITP) RDGB, is required to maintain lipid homeostasis at endoplasmic reticulum (ER) plasma membrane (PM) membrane contact sites (MCS). Depletion or mis-localization of RDGB results in multiple defects in photoreceptors. Previously, interaction between the FFAT motif of RDGB with the integral ER protein dVAP-A was shown to be important for its localization at ER-PM MCS. Here, we report that in addition to FFAT motif, a large unstructured region (USR1) of RDGB is required to support the RDGB/dVAP-A interaction. However, interaction with dVAP-A alone is insufficient for accurate localization of RDGB: this also requires association of RDGB with apical PM, through its C-terminal LNS2 domain. Deletion of LNS2 domain results in complete mis-localisation of RDGB and also induces large mis-regulated interdomain movements abrogating RDGB function. Thus, multiple independent interactions between individual domains of RDGB supports its function at ER-PM MCS.


57
The close approximation of intracellular membranes without fusion between them is emerging as a 58 theme in cell biology (Gatta and Levine, 2017). Such apposition of membranes, referred to as 59 membrane contact sites (MCS) can occur between multiple cellular organelles; most frequently, the 60 endoplasmic reticulum (ER) which is the largest organelle, makes MCS with other cellular organelles 61 including the plasma membrane (PM) (Cohen et al., 2018). ER-PM contact sites have been described 62 in multiple eukaryotic cells, and are proposed to regulate a range of molecular process including 63 calcium influx and the exchange of lipids (Chen et al., 2019;Saheki and Camilli, 2017). Charged residues in the LNS2 domain of RDGB are required for membrane interaction 225 and localization 226 We investigated the membrane binding mechanism of the LNS2 domain using molecular dynamics 227 (MD) simulations (using the model generated as described in methods) and a 228 dipalmitoylphosphatidylcholine (DPPC) membrane (system 1-RDGB). The system was subjected to 229 six steps of equilibration at 50 ns per step followed by MD run for 100 ns (in replicates). The 230 minimum distance between RDGB and the membrane in each system was measured throughout the 231 100 ns run to check the closest distance between RDGB and the DPPC membrane. Findings from 232 this study show that the LNS2 domain is required for interaction of RDGB with apical PM. We 233 captured the movements of the LNS2 domain during the simulation and calculated its distance from 234 the membrane. As observed during the MD simulations, an alpha helical region of the LNS2 domain 235 containing two charged residues was seen to move closest to the membrane. These were two lysine 236 residues found at positions 1186 and 1187 of RDGB protein (K1186 and K1187). 237 The distance between K1186 and K1187 in the alpha helical region of the LNS2 domain and the 238 membrane was then checked. It was observed that in the simulation with system 1-RDGB, the 239 minimum distance between RDGB and the membrane is 2 Å which remains stable throughout the 240 simulation [Supplementary Figure S3A]. The distance between the hydrogen atoms (closest to the 241 lipid) of the residues K1186 and K1187 (LNS2 domain) and the DPPC molecule was 3 Å for upto 242 80% of the simulation after which it deviates to about 15 Å. Based on the above observation we 243 mutated these two lysine residues to alanine (RDGB KK/AA ) using the FOLDX protocol (referred in 244 methods). This mutated protein was then subjected to similar MD simulations as for wild type RDGB 245 (referred to as system 2-RDGB KK/AA ). For system 2-RDGB KK/AA the minimum distance between the 246 protein and the membrane at the start of the simulation was 2 Å which increased to 8 Å as the 247 simulation progressed [Supplementary Figure S3B and S3C]. The distance between the hydrogen 248 atoms (closest to the lipid) of the residues A1186 and A1187 (in system 2-RDGB KK/AA ) with DPPC 249 molecule remained more than 5 Å for upto 60% of the simulation afterwhich it deviated to about 30 250 Å [Supplementary Figure S3D,E and F]. We also performed simulation for the protein lacking the 251 LNS2 domain in the presence of membrane (system 3-RDGB LNS2Δ ) and observed that the protein 252 remained at a distance of 8 Å and more from the membrane throughout the simulation 253 [Supplementary Figure S3G] and did not form any stable interactions. The charged residues of LNS2 domain in system 1-RDGB moved at a distance to the membrane such that they could form van der 255 Waals interactions with each other while the distance between the protein and the membrane in 256 system 2-RDGB KK/AA and 3-RDGB LNS2Δ remained larger throughout the simulation after the system 257 stabilized [Video files 1,2 and 3]. 258 The study suggests that presence of LNS2 domain is important for interaction of the RBGB protein 259 with the membrane. Since the minimum distance between the protein and the membrane is smaller 260 in system 1(RDGB) as compared to system 2 (RDGB KK/AA ) and system 3 (RDGB LNS2 ), it is clear that 261 this domain is required for the protein to associate to the membrane. 262 To validate the results of our simulations, we mutated the K1186 and K1187 residues of RDGB in the 263  The LNS2 domain is essential to support RDGB function during in vivo signalling.

271
Given the indispensable role of LNS2 domain in localizing RDGB to the MCS, we tested if the domain 272 has a physiological role in supporting RDGB function in vivo. A key function of RDGB is to maintain 273 the electrical response to light in Drosophila photoreceptors (Harris and Stark, 1977;Yadav et al., 274 2015); this requirement manifests as a reduced electroretinogram (ERG) amplitude in rdgB 9 flies. 275 Upon testing the requirement of the LNS2 domain in supporting phototransduction, we found the 276 light response in RDGB LNS2Δ expressing photoreceptors to be as low as that in rdgB 9 [ Fig 5A, B]. 277 RDGB is also essential to support the levels of PIP2 at the apical PM. We measured apical PM PIP2 278 levels by quantifying the fluorescence of PH-PLCδ::GFP probe in the pseudopupil of the eye 279 (Chakrabarti et al., 2015). As previously reported (Yadav et al., 2015),we found that the resting level 280 of PIP2 at the apical PM of rdgB 9 was reduced and could be restored to wild type levels by 281 reconstitution with a wild type RDGB transgene [Fig 5C, D]. However, when rdgB 9 was 282 reconstituted with RDGB LNS2Δ (rdgB 9 ; GMR> rdgB LNS2Δ ), the PM PIP2 levels were found to be as low

The LNS2 domain restricts the movement of the DDHD domain 287
While the data above highlights the indispensable role of LNS2 domain in supporting RDGB 288 function in vivo, previous studies (Milligan et al., 1997;Yadav et al., 2015) have reported the 289 sufficiency of the PITPd in supporting RDGB function in vivo. We obtained similar results when we 290 tried to rescue rdgB 9 phenotypes by expressing RDGB (USR1-LNS2)Δ . RDGB (USR1-LNS2)Δ was able to 291 completely rescue the reduced light response phenotype of rdgB 9 mutants [ Fig 6A, B]. Similarly, 292 expression of RDGB (USR1-LNS2)Δ could partially restore the reduced basal PIP2 levels in rdgB 9 293 photoreceptors to wild type levels [ Fig 6C,D], with equivalent levels of PH-PLCδ::GFP probe 294 expression [ Fig 6E]. Importantly, while loss of the LNS2 domain completely abrogated in vivo RDGB 295 function, loss of all the domains collectively after the PITPd has only marginal effect on its function. to predict potential inter-domain interactions, a normal mode analysis (described in materials and 303 methods) was performed. Lowest frequency modes that describe the largest movements were 304 analysed in detail for all the three models. The presence of multiple domains in LTPs is hypothesized to enable their correct localization at MCS. How is RDGB accurately localized to ER-PM MCS? It has previously been reported (Yadav et al.,347 2018) that an interaction between the FFAT motif and the ER-resident protein dVAP-A is essential 348 for the normal localization and function of RDGB. In this study, surprisingly, we found that a RDGB 349 protein lacking all domains downstream of the FFAT motif was (i) mislocalized away from the base 350 of the rhabdomere and (ii) was unable to interact with dVAP-A despite the presence of an intact FFAT 351 motif, suggesting that additional regions of the RDGB protein are required to stabilize the 352 simulations in silico, we established that the unstructured region 1 (USR1), positioned C-terminal to 354 the FFAT motif but prior to the start of the DDHD domain is sufficient to stabilize the FFAT/dVAP-355 A interaction. These findings suggest that inter-domain interactions within the RDGB protein are 356 important for stabilizing the interaction between its FFAT motif and dVAP-A, thus anchoring it to 357 the ER side of the ER-PM contact site. Interestingly, a role for an intrinsically disordered region of 358 OSBP in regulating its function at ER-Golgi MCS has recently been proposed (Jamecna et al., 2019). 359 In the broader context, a large number of proteins that interact with dVAP-A have been 360 reported (Murphy and Levine, 2016); many of these are via interaction with the FFAT motif but 361 similar to the findings reported in this manuscript, they too may require other domains to stabilize 362 this interaction. This may offer additional modes of recruiting regulators of lipid transfer proteins to 363 the vicinity of the contact site.

365
Proteins localized to an ER-PM MCS also require a mechanism to tether them to the PM. In this 366 study, using sub-cellular fractionation assay, we found that RDGB is a membrane associated 367 protein. This membrane association is likely to be both with the ER and the apical PM of 368

photoreceptors. Disruption of RDGB/dVAP-A interaction only reduced membrane association of 369
RDGB partially, implying the presence of additional signals that allow RDGB to associate with 370 membranes. Using a series of C-terminal deletion constructs, we found that this key membrane 371 association signal is the C-terminal LNS2 domain. When the LNS2 domain is deleted (RDGB LNS2Δ ), 372 the majority of the protein is cytosolic and is mis-localized away from the ER-PM contact site into 373 the photoreceptor cell body. Upon expression of the LNS2 domain alone, it was found to be targeted 374 to the apical PM in photoreceptors. Lipid binding assay revealed that the LNS2 domain is able to bind 375 PM enriched acidic phospholipids such as PA and PS in vitro, suggesting a molecular signal from the 376 PM that engages in this interaction. The contribution of these signals to function in vivo is clearly 377 critical since RDGB LNS2Δ is unable to restore function when reconstituted in the rdgB 9 378 mutant. Interestingly, the protein-protein interaction of RDGB LNS2Δ with dVAP-A itself is unaffected, 379 implying that the interaction with dVAP-A and apical PM, are mutually exclusive properties of 380 RDGB. These findings suggest that the function of the LNS2 domain is a prerequisite to target RDGB 381 to the apical PM, and following this the USR1 assisted FFAT/dVAP-A interaction restricts its 382 localization specifically to the apical PM-ER junction. Based on these findings, the RDGB protein (711-900 aa) and LNS2 domain (1159-1200 aa). There is an unstructured region-USR1 (no 540 secondary structure predicted) between the FFAT-motif and DDHD domain (420-600 aa). 541 Secondary structure prediction shows that the PITPd contains alpha helices and beta sheets, the 542 DDHD domain is mostly alpha helical and the LNS2 domain contains alpha helices and beta sheets. 543 The region between the PITPd and DDHD domain is largely unstructured and lacks high homology 544 to any existing protein structure. In order to obtain a 3-D structure of the protein, homology-545 based modelling (MOELLER) ( Roy et al., 2010) was done [ Table 1]. The regions which have high 546 homology to known structures available in PDB database were modelled using the homology 547 modelling protocol in MODELLER. The regions that did not have template predicted with high 548 identity but had closely related protein FOLD (classified as per SCOP classification) structure were 549 modelled using fold prediction method while those regions of the model which had neither of the 550 two above mentioned templates were modelled ab-initio. The table lists all templates that were 551 obtained by either homology or fold prediction, while regions where template coverage was low were modelled using ab-initio methods. The final full-length model was energy minimized for 500 steps 553 of conjugate gradient cycles followed by steepest descent gradient for 500 steps. The initial steps of 554 minimization were stringent to avoid large structural deviations in the model while minimization. 555 Once the model reached a negative potential energy state with allowed Ramachandran plot values, 556 relaxed energy minimization steps were performed to further reach an energy minimum. The final 557 model had negative potential energy and is stable. The Ramachandran plot distribution of amino 558 acids in the model was checked using the PROCHECK analysis tool (Laskowski et al., 1993). To selected for further analysis. The protein-protein complex was analyzed for stable interactions based 580 on PPcheck evaluation and models with lowest overall energy and very low unfavorable interaction 581 was selected as the best model. Each of the best models selected was analyzed further using molecular 582 dynamics simulation for stability of the complex in solution. Desmond module of molecular 583 dynamics was used to perform 100 ns simulation for each protein in duplicates. OPLS_2005 force 584 field with standard NPT conditions was used. The protein complexes were solvated in an 585 orthorhombic box with periodic boundary conditions by adding TIP3P water molecules. The initial 586 equilibration was carried out using default protocol of restrained minimization followed by 587 molecular dynamics simulations for 100 ns. Based on the analysis described below we obtained a 588 RDGB model with K1186 and K1187 residues mutated to alanine using FOLDX (Schymkowitz et  membrane-protein systems were generated for RDGB protein (system 1), RDGB KK/AA (system 2) and 601 the RDGB LNS2Δ protein (system 3). The protein was placed at a distance of 10 nm along the Z-axis 602 from the membrane at the start of the simulation. The protein and the membrane system were 603 solvated in an orthogonal box with TIP3P water and neutralized with ions to get a final system with 604 net charge being zero. The simulations were performed with a 2 fs step size and the nearest neighbour 605 list was recorded every 20 ps. The temperature and pressure were maintained at 300K and 1 bar -1 606 respectively. The system was energy minimized using Gromacs charmm36m force field for 50 ns 607 followed by six steps of equilibration each for 50 ns. The final MD run was carried out for 100 ns 608 (with two replicates for each system with different initial velocities) using charmm36m force-field. 609 All the downstream analysis was carried out using different modules of Gromacs, PYMOL and VMD. 610

Protein-Membrane Interaction 611
The g_mindist command was used to calculate the minimum distance between the protein and the 612 bilayer throughout the simulations. 613

Interacting residues 614
The distance of residues that interact with membrane was calculated using the n_index and 615 g_minddist module of Gromacs. 616

Normal mode analysis to understand inter-domain movements:
620 Normal modes to understand inter-domain movements in the RDGB protein were calculated using 621 the ANM 2.1 server. ANM (Anisotropic network model) is a simple NMA tool for analysis of 622 vibrational motions in molecular systems (Eyal et al., 2015). Elastic Network methodology was used 623 and it helped to represent the system at the residue level. The macromolecule is represented as a 624 network of atoms. In the model each protein node is the Cα atom of a residue and the overall potential 625 is simply the sum of harmonic potentials between interacting nodes. The network included all 626 interactions within a cut-off distance (distance cut-off of 15 Å). This was the predetermined 627 parameter in the model. Information about the orientation of each interaction with respect to the 628 global coordinates system was considered along the force constant. The force constant was described 629 by Hessian matrix. Each element of the matrix is interaction between two nodes i and j (two C-alpha 630 atoms of two amino-acids). The distance between two nodes was added as a weight at each element   The table represents the details on templates used for modelling the RDGB protein using an 867 integrative modelling approach. The first column represents the different regions of RDGB 868 protein modelled individually using different templates. The second and third column 869 represents the available template and the protein name of the template. The fourth and fifth 870 column represent the confidence (in percentage) with which the template is selected for 871 modelling and percent identity of the template with the RDGB protein sequence respectively. 872 The last column indicates the method used for modelling. The well-structured domains of 873 the RDGB protein were modelled using the available templates, either by homology 874 modelling or fold prediction. For regions with low confidence or low sequence identity with 875 the template, ab-initio modelling protocol was used. The molecular dynamics simulation was performed using Gromacs for RDGB (video file 1), 879 RDGB KK/AA (video file 2) and RDGB LNS2 Δ (video file 3) protein in the presence of DPPC 880 membrane. The simulations were carried out for 100 ns (in replicates) and the movie file was 881 generated using the 'trjconv' command of Gromacs. The video files show the interaction 882 between protein and membrane during the entire course of simulation. The protein domains are colour coded as in Figure 1C while the membrane is represented as spheres in grey colour. 884 The residues K1186/A1186 and K1187/A1187 are represented in cyan colour (sphere 885 representation). YW motif is coloured black. The files were generated using PYMOL and can 886 be viewed using a video player with the repeat/loop mode ON. 887

Video file 1: Molecular dynamics of RDGB and DPPC membrane 888
The residues K1186 and K1187 represented in cyan spheres are seen to interact with the 889 membrane for 80% of the simulation. 890

Video file 2: Molecular dynamics of RDGB KK/AA and DPPC membrane 891
The residues A1186 and A1187 represented in cyan do not interact with membrane after the 892 system stabilizes and move to a distance greater than 10 Å as the simulation progresses and 893 stabilizes.