A STIM2 splice variant negatively regulates store-operated calcium entry

Cellular homeostasis relies upon precise regulation of Ca2+ concentration. Stromal interaction molecule (STIM) proteins regulate store-operated calcium entry (SOCE) by sensing Ca2+ concentration in the ER and forming oligomers to trigger Ca2+ entry through plasma membrane-localized Orai1 channels. Here we characterize a STIM2 splice variant, STIM2.1, which retains an additional exon within the region encoding the channel-activating domain. Expression of STIM2.1 is ubiquitous but its abundance relative to the more common STIM2.2 variant is dependent upon cell type and highest in naive T cells. STIM2.1 knockdown increases SOCE in naive CD4+ T cells, whereas knockdown of STIM2.2 decreases SOCE. Conversely, overexpression of STIM2.1, but not STIM2.2, decreases SOCE, indicating its inhibitory role. STIM2.1 interaction with Orai1 is impaired and prevents Orai1 activation, but STIM2.1 shows increased affinity towards calmodulin. Our results imply STIM2.1 as an additional player tuning Orai1 activation in vivo.


Homology modelling and protein-protein docking analyses
The CAD domain (CRAC activation domain) was defined as the segment containing amino acids 342-448 of human STIM1. By definition, the CAD domain covers the entire region of SOAR (STIM1 Orai1activating region, amino acids 344-442). The X-ray crystallographic structure of the human STIM1 CAD/SOAR dimer was determined by 1 and was deposited in the Protein Data Bank as 3TEQ.pdb. To build a 3D model of the STIM2.2 CAD domain that has 78 % sequence identity with STIM1, we applied homology modeling. The reliability of the homology modeling method is based on the pioneering work by Sander and Schneider 2 who showed that two protein sequences that can be aligned over a stretch of 80 residues or more and show a sequence identity higher than 25% are extremely likely to share similar 3D structures. Thus, we adopted 3TEQ as the structural template 3 and used the homology modeling software MODELLER 9.13 4 . Due to the way of construction, the STIM2.2 model necessarily matches very closely the STIM1 crystal structure.
Modeling of the new splice variant STIM2.1 carrying the 8-residue insertion "VAASYLIQ" is much more complicated than for STIM2.2. As a word of caution, we point out that the exact conformation of this insertion and its precise positioning in STIM2.1 cannot be predicted with ultimate confidence with current computational resources and algorithms. To propose the most likely secondary structure of this 8residue insertion, we performed a similarity search against the full PDB database. The top hit structure is a 7-residue peptide VAAYLIQ of protein tyrosine phosphatase (PDB ID: 3RGQ) that adopts an -helical conformation but lacks the central serine residue. As an independent source of information, we submitted the entire sequence of human STIM2.1 to the established protein structure prediction servers PredictProtein 5 , PHD 6 and PSIPRED 7 . All servers predicted for this region an -helical conformation as most likely conformation. The confidence scores for the 8 positions are "66666766" (PredictProtein), "89999999" (PHD) and "87767899" (PSIPRED). In addition, we performed an 100 ns long self-guided Langevin dynamics simulation 8 for a 16 amino acid peptide with the sequence "AKDEVAASYLIQAEKI". This sequence is that of the 8-residue insertion flanked by 4-residue extensions at both N-terminal and C-terminal ends adopted from the STIM2.1 sequence. As initial conformation of the peptide, we used an unfolded and stretched conformation. During the 100 ns long simulation, the peptide folded and stayed a substantial time in alpha-helical conformation as shown in the figure below (Supplementary Figure 8). All these approaches suggest the 8-residue insertion to be helical. By applying the function of multiple templates (Supplementary Figure 9) of MODELLER to constrain amino acids 384-391 to an -helical structure of roughly two turns in length (see Figure 7), we generated a homology model of the STIM2.1 CAD domain where the 8-residue insertion of STIM2.1 simply leads to an extension of the respective helix (S1). We termed this STIM2.1 conformation as model 1. Note that the extension of the helix by 8 residues shifts and rotates the residues following this insertion with respect to the STIM1 crystal structure.
To account for possible ambiguities in that the available secondary structure predictions cannot guarantee the 8-residue insertion to adopt a helical structure, we considered three further possibilities for the STIM2.1 conformation where we assumed the 8-residue insertion to be only partly helical, or to completely adopt a loop conformation. For model 2, we built the model exactly according to the sequence alignment between the STIM1 and STIM2.1 CAD domains. In this case, the inserted 8 residues of STIM2.1 were modeled as a flexible loop. In model 3, the first half segment "VAAS" was constrained to be in helical structure and the other half "YLIQ" was modeled as loop. For model 4, "VAAS" was modeled as loop whereas "YLIQ" was constrained to be helical (Supplementary Figure 4).
All homology models of STIM2 and the X-ray structure of STIM1 were evaluated by the Swiss-Model Structure Assessment server (http://swissmodel.expasy.org/workspace/) after an initial energy minimization. This server evaluates the quality of protein structures by a Qmean6 score 9 . According to Supplementary Table 2, all models have Qmean6 score higher than 0.5. Surprisingly, the STIM2 model has even a slightly better score than the STIM1 X-ray crystal structure. We also performed structural evaluations by Procheck 10 . All our models have 99% of the residues in either the core region or in the allowed region of the Ramachandran plot. Note that STIM2.1 model 1 has the highest percentage of residues in the core region.
Protein-protein docking calculations were then performed with the aim of discriminating between the interaction types of STIM/Orai1 complexes. Docking was performed with three established docking packages DOT2 11 , FRODOCK 12 and ZDOCK 13 . The three programs treat biomolecules (proteins and nucleic acids) as rigid bodies and generate complex conformations by considering shape complementary (or vdW interaction) and electrostatic interactions between the biomolecules as well as desolvation of the binding interfaces. Before applying protein-protein docking in a predictive scenario to STIM/Orai1interactions, we performed a "redocking" benchmark for reproducing bound conformations of four protein complexes formed by SMRT/GPS2 14 Table 3 shows the RMSD deviation of the docking models with the most favorable docking scores obtained with the three docking packages (DOT2, FRODOCK and ZDOCK) from the correct X-ray or NMR conformations. All the best scored conformations show small RMSD CA values below 0.8 Å from the native structures.
Before docking of the STIM/Orai systems, the STIM1 CAD crystal structure, the STIM2.2 and STIM2.1 model 1 to model 4 homology models and the NMR structure of the C-terminal helix of Orai1 (amino acids 272-291, PDB ID: 2MAK) were first energy minimized by the Amber9 package 18 to relax some steric clashes between side chain atoms. For this, an implicit solvent model (generalized Born model) was used and all backbone atoms were harmonically restrained to the starting geometry during the energy minimization using a force constant of 8.0 kcal/Å-mol. During the protein-protein docking calculations with DOT2, FRODOCK and ZDOCK, the Orai1 C-terminal helix was set as mobile object whereas the STIM CAD monomer structures were set as stationary objects.
As mentioned in the main text, the ORAI C-terminal helix docked against the STIM proteins in different regions that we termed motif 1, motif 2, and motif 3. We point out that docking motif 1 identified for the Orai1:STIM1 interaction appears most plausible. We do not expect that the Orai1-helix will bind to STIM2.2/STIM2.1 using motifs 2 or 3. Generally, one expects that homologous protein pairs bind in similar orientations 19 . Supplementary Table 4 compares the docking scores for binding to motif 1 compared to the best-scoring motif for each case. For STIM1, all 3 docking packages predict motif 1 as the most favorable docking orientation. For STIM2.2, the DOT2 program assigns a slightly weaker binding affinity for motif 1 than the most favorable orientation. This suggests that in the case of STIM2, binding to motif 1 is reduced but still feasible. This would in fact match the experimental finding that STIM2.2 can still activate Orai1, but with a reduced efficiency. DOT2 further predicts that in the case of STIM2.1, docking to motif 1 (best predicted affinity is -14.105 kcal/mol) is 1.5 kcal/mol more unfavorable than the best docking pose (-15.519 kcal/mol). Also, all docking scores for STIM2.1 are all clearly lower than for STIM1 and STIM2.2. The results obtained with the two other software packages do not allow us to conclude on the efficiency of Orai1-stimulation by STIM2.2 vs. STIM2.1. However, we point that the DOT2 package applies -in our view -the most detailed model for electrostatic interactions. Thus, we believe that it is most suitable among these 3 packages to treat electrostatically dominated interactions as is the case here.