Research ArticleCalcium signaling

Sequential activation of STIM1 links Ca2+ with luminal domain unfolding

See allHide authors and affiliations

Science Signaling  19 Nov 2019:
Vol. 12, Issue 608, eaax3194
DOI: 10.1126/scisignal.aax3194

Keeping STIM1 unSTIMulated

Upon ER Ca2+ depletion, the ER-localized Ca2+ sensor STIM1 activates the Orai family of plasma membrane–localized Ca2+ channels to replenish Ca2+ stores. Schober et al. described the conformational changes that occur during the initial steps of STIM1 activation. The authors used biochemical and electrophysiological analyses and molecular dynamics simulations to characterize constitutively active STIM1 mutants associated with tubular aggregate myopathy or cancer, as well as naturally occurring STIM1 variants. Their results indicated that although STIM1 was stabilized by the binding of a single Ca2+ ion, it could bind to multiple Ca2+ ions, a property that was disrupted by disease-associated mutations. Furthermore, these mutations caused STIM1 to adopt an unfolded conformation similar to that caused by Ca2+ depletion in wild-type STIM1.

Abstract

The stromal interaction molecule 1 (STIM1) has two important functions, Ca2+ sensing within the endoplasmic reticulum and activation of the store-operated Ca2+ channel Orai1, enabling plasma-membrane Ca2+ influx. We combined molecular dynamics (MD) simulations with live-cell recordings and determined the sequential Ca2+-dependent conformations of the luminal STIM1 domain upon activation. Furthermore, we identified the residues within the canonical and noncanonical EF-hand domains that can bind to multiple Ca2+ ions. In MD simulations, a single Ca2+ ion was sufficient to stabilize the luminal STIM1 complex. Ca2+ store depletion destabilized the two EF hands, triggering disassembly of the hydrophobic cleft that they form together with the stable SAM domain. Point mutations associated with tubular aggregate myopathy or cancer that targeted the canonical EF hand, and the hydrophobic cleft yielded constitutively clustered STIM1, which was associated with activation of Ca2+ entry through Orai1 channels. On the basis of our results, we present a model of STIM1 Ca2+ binding and refine the currently known initial steps of STIM1 activation on a molecular level.

INTRODUCTION

Store-operated Ca2+ channel entry (SOCE) is a ubiquitous cellular mechanism to generate long-lasting Ca2+ signals at restricted endoplasmic reticulum (ER) and plasma membrane (PM) junctions (1, 2). SOC influx is especially important in the immune system, because Ca2+ influx mediated by SOCs triggers T cell activation and controls cellular immune responses (3). In general, SOCE is activated by a stimulus that reduces the Ca2+ concentration within the ER, followed by a Ca2+ release–activated Ca2+ (CRAC) current (4).

SOCs are specifically localized at ER-PM junctions, because the highly Ca2+-selective PM channel Orai1 and the Ca2+ sensor protein stromal interaction molecule 1 (STIM1), located in the ER, bind directly to each other for activation (57). STIM1 is a finely tuned sensor that responds to Ca2+ levels in the range of 100 to 400 μM (5, 8). In cells under resting conditions, STIM1 dynamically and constitutively move along microtubules; in contrast, store depletion results in the redistribution of STIM1 into puncta at ER-PM junctions (9). Both STIM1 and its close homolog STIM2 are single-pass transmembrane (TM) proteins with their N termini residing in the ER lumen and their larger C-terminal domains within the cytosol. The ER luminal region, which functions as a Ca2+ sensor of intraluminal [Ca2+]ER, contains a Ca2+-sensing canonical EF hand, a stabilizing noncanonical EF hand, and a sterile α motif (SAM) domain, followed by an α-helical TM domain (10, 11). Upon a decrease in [Ca2+]ER, a Ca2+ ion dissociates from the canonical EF hand, which, together with a variable N-terminal STIM1 random coil sequence, destabilizes the EF hand–SAM complex (10, 12, 13). This destabilization triggers STIM1 oligomerization and the release of a larger cytosolic binding pocket for coupling to and activation of Orai channels (2, 14). High-resolution structures of STIM1 and STIM2 luminal domains under resting conditions have successfully been determined by nuclear magnetic resonance (NMR) (10, 11). In contrast, conformational transition structures within the luminal domain, which are established during or upon final STIM1 activation, remain elusive (15). In addition, the number of bound Ca2+ ions to STIM1 is currently under extensive debate (1619).

Here, we present structural aspects of how the luminal STIM1 domain stabilizes a monomer and describe STIM1 sequential structural activation within the ER upon Ca2+ depletion. A combination of MD simulations, live-cell experiments, and biochemical data suggests a dynamic model of Ca2+ sensing that integrates the interplay between the canonical and noncanonical EF hands and the SAM domain. Ca2+ binding of the luminal STIM1 domain and its switch into an active conformation controls the entire signaling cascade of SOCE, and mutations in this domain resulted in severe pathological phenotypes in humans. Loss-of-function mutations lead to immunodeficiency (Mendelian Inheritance in Man registration number; MIM number 612783) (20, 21), whereas gain-of-function mutations can cause the multisystemic Stormorken syndrome (MIM number 185070) and tubular aggregate myopathy (TAM) (MIM number 160565) (22, 23). Here, we characterized STIM1 mutants from a cancer database and evaluated TAM mutants to identify key residues that determine constitutive activity upon their mutation on a molecular level. We showed that several pathophysiological point mutations within the two EF hands disturb the binding of multiple Ca2+ ions, affect the hydrophobic interplay of the luminal domains, and lead to constitutive Ca2+ influx and Ca2+-dependent transcription factor activation.

RESULTS

A functional screen of STIM1 cancer database and TAM mutants identifies constitutively active mutants

The luminal STIM1 domain provides the initial trigger for store-operated Ca2+ entry by switching from a monomer to a dimeric or multimeric state (17). To identify key residues that control the primary step of activation in the luminal part of STIM1, we focused on gain-of-function mutations in the canonical EF hand (amino acids 63 to 96) and subsequently in the noncanonical EF hand (amino acids 97 to 128), which are wrapped around a single SAM domain. Several EF-hand mutations are responsible for TAM, specifically H72Q, D84G, H109R, and H109N (23, 24), which points at a central role of the EF hand in human diseases. Analysis of somatic mutations from 11,000 patients has identified the EF-hand region of STIM1 as a potential hotspot for mutations based on the three-dimensional (3D) structure (25). This approach focused on the evaluation of the mutated amino acid positions in the folded protein, which enabled the identification of driver mutations and study of structural aspects of so-called driver sites (25).

Nuclear translocation of the transcription factor NFAT [nuclear factor of activated T cells; tagged with a cyan fluorescent protein (CFP) fluorophore] served as our initial assay to identify constitutively active STIM1 mutants and constitutive Ca2+ entry. This fluorescence-based screening method, which has been used in the context of Orai1 channels (26, 27), uses the NFAT signaling cascade, which is activated by STIM1- and Orai1-mediated Ca2+ entry. A subsequent increase in the intracellular Ca2+ concentration activates calmodulin/calcineurin to initiate dephosphorylation and nuclear translocation of NFAT (2, 28). To search for previously uncharacterized STIM1 gain-of-function mutants, we compared previously described constitutively active TAM STIM1 mutants (Fig. 1A) to 11 cancer database mutants with mutations in the canonical EF hand (Fig. 1B). In addition, three canonical EF-hand natural variations of STIM1, identified in the International Genome Sample Resource, were investigated (Fig. 1C).

Fig. 1 STIM1 cancer database mutants result in nuclear localization of NFAT.

(A to C) Model of the STIM1 Ca2+-sensing domain [PDB code: 2 K60; (10)] with STIM1 single point mutants associated with tubular aggregate myopathy (TAM; green) (A), found in human tumors (orange) (B), and are natural STIM1 variations (blue) (C). Canonical EF hand (cEF), noncanonical EF hand (nEF), and SAM domain are shown in red, blue, and yellow, respectively. (D) Representative fluorescence images of cells coexpressing NFAT and wild-type (WT) STIM1 or STIM1-H72R, respectively. Scale bars, 10 μm. (E) Percentage of cells with nuclear NFAT localization measured 24 hours after cotransfection of YFP-STIM1 constructs and CFP-NFAT in 2 mM Ca2+ medium. Black, orange, green, blue, and red bars represent wild-type STIM1, cancer database mutants, tubular myopathy mutants, natural variants of STIM1, and the STIM1-D76A mutant as positive control, respectively. For each mutant, n = 80 to 120 cells per group were measured on three different days. An asterisk (*) indicates a significantly increased number of cells with nuclear NFAT localization (t test, P < 0.05). (F) Percentage of cells with nuclear NFAT localization measured 24 hours after cotransfection of the indicated YFP-STIM1 constructs and CFP-NFAT after exposure to 2 mM Ca2+ medium for 2 hours or to 0 mM Ca2+ medium for 24 hours (black bars). Light gray bars are from (E). For each group, n = 80 to 130 cells per group were measured on three different days. An asterisk (*) indicates a significantly different number of cells with nuclear NFAT localization in the presence or absence of Ca2+ (t test, P < 0.05). (G) Model of the STIM1 Ca2+-sensing domain with mutated positions of constitutively active STIM1 mutants highlighted. The color code is the same as in (A) to (C). The light blue circle and the red rectangle indicate the hydrophobic cleft and the Ca2+-binding loop within the cEF, respectively.

Cellular localization of NFAT was monitored 24 hours after coexpression with wild-type STIM1 or STIM1 canonical EF-hand mutants in human embryonic kidney (HEK)–293 cells (Fig. 1, D and E). The expression levels of NFAT and STIM1 proteins were determined by fluorescence intensity measurements to rule out the influence of different protein quantities on NFAT translocation (fig. S1, A and B). A constitutively active form of STIM1 with a D76A mutation that interferes with the Ca2+-binding affinity in the canonical EF hand (29) was used as a positive control. A STIM1-E87Q mutant from the cancer database yielded an increased number of cells (~60 to 70%) with nuclear NFAT localization similar to the positive control STIM1-D76A. In contrast, wild-type STIM1 coexpression led to only 10 to 15% nuclear NFAT-positive cells. Screening of all cancer database mutants, TAM mutants, and natural variants of STIM1 identified multiple mutations that statistically significantly enhanced the number of cells with strong nuclear NFAT localization (about 40 to 70% of cells), which suggests constitutively active STIM1 proteins. These mutations were H72R, H72Q, D76V, D78G, A79T, N80K, G81D, E87Q, L92P, and L96V (Fig. 1E). Compared to wild-type STIM1, coexpression of one of the natural variants of the STIM1 protein (STIM1-D77N) also led to a significantly higher proportion of NFAT-positive cells (~35 to 40%), although not as high as some cancer database mutants (Fig. 1E).

Because endogenous store-operated Ca2+ channels translocate NFAT to the nucleus with a t1/2 (half-time) of ~30 min (1), we limited the Ca2+ influx for constitutively active STIM1 mutants to 2 hours. In the presence of 2 mM extracellular Ca2+, the more restricted period of constitutive Ca2+ influx stimulated NFAT translocation as efficiently as after 24 hours (Fig. 1F). In contrast, in nominally Ca2+-free medium, NFAT remained in the cytosol upon overexpression with the most active STIM1 mutants (H72R, H72Q, N80K, and E87Q), as with wild-type STIM1 (Fig. 1F). In a similar approach, the D77N mutant, a natural variant of STIM1, did not mediate NFAT translocation upon 2 hours of Ca2+ treatment, narrowing down constitutively active mutants to those found in the cancer database and in patients with TAM (Fig. 1F and fig. S1C). The results of the NFAT screen were further verified in STIM1/Orai1 knockout cells (30). Here, because of the lack of endogenous Orai1 proteins, the expression of the STIM1 mutants did not significantly enhance NFAT nuclear translocation, although 2 mM extracellular Ca2+ was present for 24 hours (fig. S1D).

Evaluation of the locations of the constitutively active luminal STIM1 mutations in the NMR structure revealed two key domains: the Ca2+-binding loop and the hydrophobic cleft (Fig. 1G). The hydrophobic cleft is formed by the canonical and noncanonical EF hands. Seven constitutively active STIM1 mutations were identified in the Ca2+-binding loop of the canonical EF hand (D76V, D78G, A79T, N80K, E87Q, G81D, and D77N) (Fig. 1G, red rectangle). Specifically, five of these active mutations resulted in a charge transfer from negative (and electrostatically Ca2+ attracting) to noncharged or positively charged residues. Further, four active mutations are located within the hydrophobic cleft (H72R, H72Q, L92P, and L96V) (Fig. 1G, light blue circle). These four gain-of-function mutations located in the canonical EF-hand motif do not bind Ca2+ ions directly but are expected to play a role in stabilizing the compact luminal structure of STIM1 through interactions with the SAM domain (10). These mutations are less hydrophobic, helix breaking, or smaller in size (Fig. 1G and fig. S1E). In summary, constitutive activation of STIM1 is generated not only by mutations within the Ca2+-binding loop but also by those at residues that form the hydrophobic cleft.

Mutations in the STIM1 canonical EF hand that confer constitutive activation mimic the initial mechanistic steps of store-dependent activation

Partial unfolding and subsequent aggregation upon store depletion of the luminal STIM1 are the initial steps toward release of the cytosolic binding structure, called the SOAR/CAD [STIM-Orai activating region/CRAC-activating domain; CC2-CC3 (31, 32)], that can then couple to the Orai1 channel. Specifically, the cytosolic STIM1 structure exhibits a very tightly packed conformation in the resting state, which is maintained by a CC1 to SOAR/CAD clamp (15, 33). This clamp extends after store depletion and subsequently enables STIM1-Orai interaction at ER-PM junctions (34, 35). In confocal fluorescence experiments, this uncoupling was monitored by colocalization analysis of a yellow fluorescent protein (YFP)–STIM1 fragment with a truncation after CC1 (STIM1-CC1; amino acids 1 to 343) with a CFP-tagged CAD of STIM1 (amino acids 344 to 449) (fig. S2, A and B). Truncated wild-type STIM1 interacted with the CAD domain in the resting state, but to a lesser extent in store-depleted cells treated with the sarcoplasmic/endoplasmic reticulum calcium ATPase pump inhibitor thapsigargin. However, an analogous STIM1-CC1 truncation with an engineered H72R mutation did not colocalize with the CAD domain and was already in the active state under resting conditions (fig. S2, A and B). Thus, the H72R mutation leads to unfolding of the cytosolic STIM1 clamp.

To monitor the next step in STIM1 activation, we used fluorescence resonance energy transfer (FRET) to determine when Orai1-YFP was in close proximity to STIM1-CFP upon store depletion. Wild-type STIM1 yielded low FRET in the resting state, which indicates that STIM1 and Orai1 were not coupled. Store depletion increased FRET, which points to coupling of STIM1-CFP and Orai1-YFP. In contrast, STIM1 cancer database mutants already yielded maximum FRET levels under resting conditions that were unaffected by store depletion (Fig. 2A). This result indicates that STIM1-Orai1 binding was independent of store depletion for these mutants. To activate Ca2+ signaling for wild-type STIM1, we induced store depletion by buffering intracellular Ca2+ with EGTA in the patch pipette (Fig. 2B) or by treatment with thapsigargin, respectively (Fig. 2C). The positive mutants identified in the NFAT screen (H72R, H72Q, N80K, and E87Q) exhibited constitutively active currents by endogenous Orai (Fig. 2B) and significantly increased Ca2+ levels under basal conditions (Fig. 2C). Increased Ca2+ concentrations in the absence of store depletion are consistent with Ca2+/calmodulin/calcineurin-dependent activation (1, 2) of NFAT upon overexpression of the STIM1 mutants investigated. These data show that STIM1 cancer database mutants that interfere with Ca2+ binding or destabilize the hydrophobic cleft result in loss of Ca2+ and lead to the extension of the cytosolic clamp, Orai1 binding, and subsequent channel opening, which enables inward Ca2+ currents that increase cytosolic Ca2+ levels. In summary, these data show that the STIM1 cancer database mutants are fully active and explain the strong NFAT activation under basal conditions.

Fig. 2 STIM1 cancer database mutants induce fully active STIM1 coupling with Orai1 and Ca2+ signaling.

(A) Left: Time course of FRET experiments performed in cells coexpressing STIM1-CFP and Orai1-YFP, STIM1-H72R/Q, STIM1-N80K, and STIM1-E87Q. Store depletion was induced with 2 μM thapsigargin (TG). Right: FRET levels in the resting state were plotted for each mutant (n = 6 to 7 cells per group, from at least three independent transfections). An asterisk (*) indicates FRET values that differ significantly from that of the wild-type STIM1 in the resting state (t test, P < 0.05). (B) Left: Time courses of whole-cell patch-clamp recordings from cells overexpressing wild-type STIM1, STIM1-H72R, STIM1-H72Q, STIM1-N80K, and STIM1-E87Q. Store depletion was induced passively by an intracellular solution containing 20 mM EGTA. Right: Starting currents (n = 6 to 11 cells per group, from at least three independent transfections). An asterisk (*) indicates currents that differ significantly from that of wild-type STIM1 in the resting state (t test, P < 0.05). (C) Left: Time courses for Fura-2 AM measurements in cells overexpressing wild-type STIM1, STIM1-H72R, STIM1-H72Q, STIM1-N80K, and STIM1-E87Q. Store depletion was induced by 2 μM TG. Right: Cytosolic Ca2+ levels in the resting state (n = 16 to 22 cells per group from at least three independent transfections). An asterisk (*) indicates mutants that lead to significantly increased Ca2+ levels (t test, P < 0.05).

MD simulations identify multiple Ca2+-binding sites within the canonical and noncanonical EF hands of STIM1

We carried out molecular dynamics (MD) simulations to gain more insight into the Ca2+ store depletion mechanism that switches luminal STIM1 to the active state and to identify the initial conformational changes involved. MD simulations of the EF hand were conducted by taking advantage of the available NMR structure of the luminal domain of STIM1 [Ca2+-sensing domain, Protein Data Bank (PDB) code: 2 K60] (10) and by rebuilding the nonresolved TM α helix embedded in a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine (POPC) lipid bilayer (Fig. 3A). NMR structures of both STIM1 and STIM2 monomers have shown a single Ca2+ ion bound to the canonical EF hand (10, 11, 17). However, the latest model, which is based on a luminal STIM1 dimer stabilized by an artificial dimerization domain, suggests that five to six Ca2+-binding events are involved in the sensing mechanism of the domain (16).

Fig. 3 MD simulations of the luminal STIM1 protein reveal putative multiple Ca2+ ion binding.

(A) Ca2+ ion occupancy around the EF-SAM domain. Left: Snapshots with view of the EF-hand motif showing the occupancy of the Ca2+ ions during the last 150 ns of the simulations. Ca2+ ions surrounding the cEF hand, nEF hand, and the SAM domain are given in red, green, and gray beads, respectively, representing a total of 1000 superimposed conformations. Right: Rotation (180°) of the simulation box with view of the SAM domain. SAM domain, cEF hand, and nEF hand motifs are represented as orange, red, and violet ribbons, respectively. The rest of the protein is depicted as a gray transparent ribbon. (B) Ca2+-binding time series (500 ns) showing contact for each amino acid with several Ca2+ ions. Contact is defined as a distance of less than 3 Å between Ca2+ ion and Cα of the backbone from the residues. Each color indicates a specific Ca2+ ion binding to one or two residues at the same time. Continuous lines indicate stable binding, whereas discontinued lines indicate dynamical binding of Ca2+ ions. (C) Ca2+-binding probability plot showing high probabilities for Ca2+ binding of specific amino acids (red peaks), reaching 100% (inset: cumulative sum of the probabilities) at position 119. The probabilities were estimated from the last 400 ns of two MD simulations. (D) Left: RMSD (root mean square deviation) values for various domains of the luminal domain of wild-type STIM1 over simulation time in the presence of Ca2+. RMSD values for the overall protein, the cEF hand, the nEF hand, and the SAM domain are shown as thin gray, thick red, magenta, and yellow lines, respectively. Middle: Number of ions bound to the cEF hand (red histogram) and nEF hand (violet histogram). An ion was considered bound when in the vicinity of (meaning, at a distance of <3 Å from) any of the residues constituting either EF hand. Analyses were performed after 50 ns of simulation. Right: Snapshot of the luminal STIM1 with several Ca2+ ions bound. (E) Left: Same analysis as in (D) but in the absence of Ca2+ ions (replaced with K+ ions). Middle: Number of bound K+ ions is almost zero. Right: Snapshot of the luminal STIM1 structure in the absence of Ca2+ ions.

Initially, a strategy called “flooding” simulations (3639) was applied, in which 51 Ca2+ ions (corresponding to a nonphysiological free Ca2+ concentration of 30 mM) were added in the simulations to maximize the sampling of binding events and coverage of binding sites. In the simulations, the protein in the ER is surrounded by water and Cl ions, with Ca2+ as the only cation. After a short simulation time, highly preferential binding of Ca2+ ions to the canonical EF hand was observed (Fig. 3, A to C, and movie S1), but the noncanonical EF hand provided additional surface interaction sites for Ca2+ (Fig. 3, A to C). Preferential occupancy of Ca2+ ions was observed in a total of 1000 superimposed conformations, with the ions being clustered around the canonical EF-hand domain and also around the noncanonical EF hand and SAM domain (Fig. 3A). Within the first 100 ns of simulations, the STIM1 protein frequently exchanges Ca2+ ions with the reservoir to lastly reach an equilibrium of six to seven ions with persistent contact with the EF-hand domain. This equilibration period is the result of an interplay between electrostatic attraction of 19 acidic residues (12 of them belonging to the canonical EF hand) and Ca2+ ions present in the reservoir and the repulsion between Ca2+ ions loaded on the EF hand.

More detailed analysis of the Ca2+-binding events over time (500 ns) revealed specific positions within the EF hand involved in Ca2+ binding. Binding events were calculated by measuring a distance of <3 Å to the closest atom of negatively charged residues (Fig. 3, B and C, and fig. S3, A and B). Figure 3B illustrates a specific Ca2+ ion (represented in different colors) binding to a certain residue over a 500-ns simulation time. A similar plot is depicted in fig. S3A, showing up to two Ca2+ ions coordinated by neighboring amino acids (detailed representative snapshots in fig. S3, C and D). The first preferential binding region is formed by residues from Asp76 to Asp95 in the canonical EF hand and the second by the region from Glu111 to Asp119 in the noncanonical EF hand (Fig. 3C and fig. S3B). However, Ca2+ binding to the SAM domain was not observed (Fig. 3C and fig. S3B, inset). This finding correlates well with the work of Gudlur et al. (16) showing multiple Ca2+-binding events. Hence, our MD simulations predict that Ca2+ binding takes place at both the canonical and noncanonical EF hands of STIM1 and that multiple ions bind to the luminal STIM1 domain at the same time in the presence of supraphysiological Ca2+ concentrations.

These simulations also allowed us to investigate the stability of the designed luminal STIM1 domain and its conformational dependence on bound Ca2+ ions: In the presence of Ca2+, the luminal STIM1 structure remained in a stable resting conformation (Fig. 3D, left). Configurational stability of the luminal STIM1 was reached at around 50 to 60 ns and is reflected by constant root mean square deviation (RMSD) values over all residues of the canonical and noncanonical EF hands or SAM domain. Although no changes were observed in the presence of Ca2+, substantial intrinsic unfolding took place in the absence of Ca2+ (Fig. 3E, left). In the presence of Ca2+, the protein forms a tight conformation over the whole simulation time (500 ns), exhibiting about six to seven Ca2+ contacts to the canonical EF hand and around one to two Ca2+ contacts at the noncanonical EF hand (Fig. 3D, middle and right). Note that two to three ions were relatively stably bound to the Ca2+-binding loop of the canonical EF hand (Fig. 3B and movie S1). In the absence of Ca2+ and with K+ as the only cation present, the canonical and noncanonical EF hands of the protein unfolded rapidly because K+ ions cannot bind to the EF hand (Fig. 3E, middle and right). Comparison of the RMSD of Fig. 3E with RMSD analysis of specific residues that form the hydrophobic cleft revealed almost simultaneous unfolding of this region and the noncanonical and canonical EF hands (Fig. 4A). The SAM domain of the protein remained stable over the whole simulation time (Fig. 3E, left). In conclusion, we could simulate the behavior of the luminal STIM1 domain in the presence and absence of Ca2+ and display the initial unfolding of the EF hands.

Fig. 4 EF-SAM stability for various Ca2+ concentrations.

(A) RMSD values for individual residues within the hydrophobic cleft (black) for wild-type STIM1 in the absence of Ca2+ ions compared to the RMSD values of the cEF and nEF hands of Fig. 3E. (B) Snapshots of the simulations in the presence of Ca2+ ions only (left) and with both Ca2+ and Mg2+ ions (right). SAM, nEF hand, and cEF hand are shown as yellow, violet, and red ribbons, respectively. The rest of the protein is represented by gray ribbons. Acidic amino acids present in the nEF and cEF hands are shown as sticks using the same colors as for the domains they belong to. Red and blue translucent balls correspond to bound Ca2+ and Mg2+ ions, respectively. (C) Number of cations in contact with the cEF and nEF hands as a function of simulation time. Contact is defined as a distance less than 3 Å of the Ca2+ ion from Cα of the backbone. Contact distribution plots for the Ca2+ (red area) and Mg2+ (blue area) ions over the 500-ns simulation. A single bound Mg2+ ion is positioned at the start of the simulation to Asp77 and remained there during the simulations. (D) Binding probabilities of Ca2+ and Mg2+ ions to the EF hand domains of STIM1 under conditions containing both Ca2+ and Mg2+. Probabilities for binding of Ca2+ and Mg2+ ions are given in red and violet solid lines. Probabilities for the pure Ca2+ simulations are given in dashed red lines. Binding is defined as a Ca2+ ion being within 3-Å distance from the Cα of the backbone. The probabilities were estimated from the last 400 ns of the two simulation replicas. (E) Comparison of the resolved NMR structure to the MD simulations of the luminal STIM1 with a single bound Ca2+ ion. (F) Time course of the acidic residues that are in contact with the single Ca2+ ion. (G) RMSD values for the whole protein and for various domains of the wild-type STIM1 in the presence of K+ ions only and one Ca2+ ion bound to the cEF hand. Color code is the same as (A). (H) Root mean square fluctuations (RMSF) for each residue of the wild-type STIM1 in the presence of various concentrations of Ca2+. Systems with an excess amount of Ca2+, Ca2+/Mg2+, and one Ca2+ ion bound are shown as solid black, dashed black, and dotted black curves, respectively. Areas shaded in red, magenta, and yellow correspond to the cEF hand, nEF hand, and SAM domain, respectively. The peaks indicate more flexible regions, which are loop regions within the EF-SAM part. (I) Percentage of cells with nuclear NFAT localization, measured 24 hours after cotransfection of YFP-STIM1 constructs and CFP-NFAT in 2 mM Ca2+ medium. A black bar indicates wild-type STIM1, and light gray bars represent the indicated STIM1 mutants. STIM1-E111A-D112G and STIM1-E118A show no statistically significant difference compared to wild-type STIM1 (t test, P > 0.05).

A single Ca2+ ion stabilizes the EF-hand structure of STIM1

To rule out nonselective binding of divalent ions in these simulations, we examined the influence of Mg2+ ions on the binding properties of the EF hand. Physiological total ER Mg2+ concentrations are 14 to 18 mM and free Mg2+ of 1.5 to 2 mM (40). Because Mg2+ ions rarely interact with Ca2+-binding proteins, we substituted all cations with Mg2+ ions in the simulations, which corresponded to 150 mM free Mg2+. We carried out simulations (i) in Mg2+-only solution, and thus in the complete absence of Ca2+ ions, or (ii) with 50% Mg2+ and 50% Ca2+ ions (Fig. 4, B to D). These experiments showed that only a single Mg2+ ion remained bound to the luminal STIM1 when positioned at the start of the 150 mM Mg2+ simulation (Fig. 4, B to D, bound to residues 77 and 78). Hence, Mg2+ had a much lower affinity than Ca2+ ions and did not affect Ca2+ binding. In contrast to Ca2+ ions, Mg2+ ions do not exchange water molecules in their hydration shells as easily as Ca2+ ions. As a result, no adsorption on the protein or direct interaction of Mg2+ ions with residues of the STIM1 protein in MD simulations was observed. However, in these simulations, fewer Ca2+ ions bound to the EF hand as a consequence of a lower Ca2+ concentration because half of the original Ca2+ ions are substituted by Mg2+ ions in these simulations (Fig. 4D).

Multiple ion binding sites to the EF hand does not necessarily imply that all of the Ca2+-binding positions must be occupied to keep the structure stable. We investigated this hypothesis by simulating a single bound Ca2+ ion that was positioned within the Ca2+-binding loop of the canonical EF hand at the beginning of the simulation. One free Ca2+ ion in the simulation box would correspond to a Ca2+ concentration of 0.3 mM, which is within the physiological range of ER Ca2+ concentrations. This Ca2+ ion remained bound over the entire simulation time. The specific amino acids with which this single Ca2+ ion interacted were mostly Asp78 and Asp76 through their side chains (Fig. 4, E and F). A single bound Ca2+ ion was sufficient to keep a stable conformation because the luminal domains remained folded over the whole simulation time and only minor conformational rearrangements took place (Fig. 4G and movie S2). In all simulations, helix linkers (canonical EF hand, noncanonical EF hand, and SAM loop) were the only flexible parts according to the root mean square fluctuation (RMSF) values (Fig. 4H).

Next, we sought to identify the number of Ca2+-binding sites that were strictly required to keep STIM1 inactive under resting conditions. Therefore, we designed single or multiple point mutations within the STIM1 protein at those positions where Ca2+ binding was observable in MD simulations outside the canonical EF-hand loop and investigated whether disruption of these accessory Ca2+-binding positions would result in constitutive STIM1 activation, as assessed by NFAT activation. Substitution of the negatively charged surface residues at positions 90 and 94 with alanine residues (E90A or E94A), located C-terminally relative to the canonical EF loop, did not result in NFAT activation (Fig. 4I). Using STIM1-E111A-D112G and STIM1-E118A mutants, we additionally studied the second putative Ca2+-binding region in the noncanonical EF hand (amino acids 111 to 119), which can also coordinate one or two Ca2+ ions. However, these mutations also did not affect STIM1 activation and failed to promote nuclear translocation of NFAT (Fig. 4I). Consequently, substitution of one surface Ca2+ interaction site did not influence STIM1 Ca2+ sensing. In conclusion, STIM1 can bind multiple Ca2+ ions, not only in the Ca2+-binding loop but also in the noncanonical EF hand, whereas a single Ca2+ ion at the Ca2+-binding loop is sufficient to keep STIM1 in its resting-state configuration. Nevertheless, we cannot exclude that over a longer simulation time, STIM1 unfolds when the single Ca2+ ion leaves the binding loop because the calcium-free EF hand will start to unfold. Our MD simulations and cell assays correlate well and suggest that Ca2+ binding to the noncanonical STIM1 is not required to maintain the resting state of STIM1.

Constitutively active STIM1 mutants from the cancer database show rapid unfolding of the noncanonical EF hand

We sought to determine whether the STIM1 mutants with the greatest constitutive activation (H72R, H72Q, N80K, and E87Q) as determined by NFAT nuclear translocation (Fig. 1E) showed similar conformational rearrangements in MD simulations and live-cell experiments as wild-type STIM1 in a Ca2+-free environment. YFP-tagged wild-type STIM1 forms tubular structures in the resting state and puncta upon stimulation (34). In contrast, the selected STIM1 mutants already exhibited a clustered localization in the resting state, similar to fully stimulated wild-type STIM1 (Fig. 5A). The formation of puncta by STIM1 mutants suggests that these single point mutations can induce oligomerization of STIM1 proteins, mimicking the loss of bound Ca2+.

Fig. 5 Constitutively active STIM1 cancer database mutants induce partial unfolding of the EF-SAM domain.

(A) Representative images of cells expressing YFP-tagged wild-type STIM1, STIM1-H72R, STIM1-H72Q, STIM1-N80K, or STIM1-E87Q. Scale bars, 10 μm. (B) Left: RMSD evolutions as a function of time for various domains of the luminal domain of the STIM1-H72R mutant. The RMSD values for the overall protein, the cEF hand, the nEF hand, and the SAM domain are given as thin gray and thick red, magenta, and ocher lines, respectively. An increase in the RMSD value indicates more flexibility in the corresponding region. Right: Number of ions bound to the cEF hand (red histogram) and the nEF hand (magenta histogram). An ion was considered bound when in the vicinity of (meaning, at a distance of <3 Å from) any of the residues constituting either EF hand. Analyses were performed after 50 ns of simulation. (C) Top: Representative snapshot from the simulation of wild-type STIM1 with His72 represented by blue and green beads and their hydrogen atoms by white beads. Bound Ca2+ and the domains of STIM1 use the same colors as in Fig. 3A. Bottom: Representative snapshot from the simulation with the STIM1-H72R. (D) Analogous MD simulation and analysis as in Fig. 3D but for the mutant STIM1-E87Q. Left: RMSD of the STIM1-E87Q mutant in the presence of Ca2+. Right: Number of ions bound to the cEF hand (red histogram) and nEF hand (magenta histogram). (E) Top: Representative snapshot from the simulation of wild-type STIM1 with Glu87 represented by blue and green beads and their hydrogen atoms by white beads. Bottom: Representative snapshot from the simulation with the STIM1-E87Q mutant.

Introduction of these specific mutations to the modeled luminal STIM1 domain revealed new structural and mechanistic insights into STIM1 Ca2+ sensing. Simulations of the STIM1-H72R mutant in the presence of Ca2+ showed extensive unfolding of the noncanonical EF hand in the time course (Fig. 5B, left) and in a representative snapshot when compared to resting wild-type STIM1 (Fig. 5C). The increase in the RMSD of the noncanonical EF hand was comparable to that of wild-type STIM1 upon Ca2+ depletion, whereas the canonical EF hand and SAM domain showed steadily increasing RMSD. These results indicate that the simulation time is not sufficient to determine whether the luminal domain in the H72R mutant unfolds to a similar extent as that in wild-type STIM1 in the absence of Ca2+ (Fig. 5B). MD simulations of the wild-type STIM1 luminal domain reached an equilibrium state already around 50 to 60 ns as demonstrated by the stable values adopted by the RMSD values. In addition, the RMSD values for wild-type STIM1 oscillated around a specific value after having reached this plateau for the rest of the simulation (Fig. 3D). STIM1-H72R still showed binding of Ca2+ ions to the canonical and noncanonical EF hands, but with reduced Ca2+ binding to the loop region of Asp76, Asp77, and Asp78 (Fig. 5B, middle, and fig. S4A). This Asp76-Asp77-Asp78 segment in the canonical EF-hand loop is most critical to Ca2+ binding as shown in the single Ca2+ ion simulations (Fig. 4F).

Simulations of STIM1-E87Q, which has a mutation in the Ca2+-binding site of the canonical EF-hand loop, demonstrated a certain destabilization of the structure as indicated by steadily increasing RMSD over the whole simulation time in which the RMSD did not reach a plateau (Fig. 5D, left). However, over the simulated time, this destabilization did not reach the same extent as that of wild-type STIM1 under Ca2+-free conditions. A snapshot of STIM1-E87Q shows a partially unfolded configuration of the noncanonical EF hand compared to wild-type STIM1 (Fig. 5E). The canonical EF hand was stable throughout the whole simulation, suggesting that the canonical EF hand is more stable than the noncanonical one. A conformational change in the noncanonical EF hand might take longer than the simulation time of our experiments. Note that the canonical EF hand has one or two fewer Ca2+ contacts than wild-type STIM1 (Fig. 5D, middle, and fig. S4B), but not all of the Ca2+ ions dissociate. The lower number of Ca2+-binding events is likely due to the destabilization of the luminal STIM1 structure.

Simulations of additional cancer database mutants (N80K and L92P) showed extensive unfolding of the noncanonical EF hand as observed for STIM1-H72R (fig. S4, C and D). Hence, this domain is expected to be most prone to structural luminal STIM1 unfolding. In addition, all of these mutations show a steadily increasing destabilization of the whole luminal STIM1 protein over the entire simulation time.

Mutations associated with TAM destabilize the noncanonical EF hand and Ca2+ sensing

Because our MD simulations indicated that the noncanonical EF hand is most prone to unfolding in constitutively active STIM1 mutants, we extended our search for mutations in the noncanonical EF hand. Five TAM mutants, one additional STIM1 cancer database mutant, and one natural variant of the STIM1 protein (Fig. 6A) were investigated using our NFAT screen (Fig. 6B). Four of the five tubular myopathy mutants (STIM1-F108I, STIM1-H109N, STIM1-H109R, and STIM1-I115F) exhibited a significantly higher percentage of cells with nuclear NFAT localization in 2 mM Ca2+ medium after 2 or 24 hours (Fig. 6B). Overexpression of the natural STIM1-D100E variant resulted in a slight increase in NFAT activation observed exclusively after 24 hours incubation in 2 mM Ca2+ medium (Fig. 6B).

Fig. 6 Constitutive activity of TAM mutations within the noncanoncial EF hand of STIM1.

(A) Model of the luminal STIM1 domain structure with the mutated positions in cancer database mutants, tubular myopathy mutants, and natural variants within the nEF hand are highlighted. The color code is the same as in Fig. 1, A to C. (B) Percentage of cells with nuclear NFAT localization upon coexpression with wild-type STIM1 and STIM1 nEF-hand mutants after 24 hours (darker colored bars) or 2 hours (lighter colored bars) of extracellular 2 mM Ca2+. The color code is the same as in Fig. 1E. For each mutant, n = 80 to 120 cells were measured on three different days. An asterisk (*) indicates significantly increased number of cells with nuclear NFAT localization (t test, P < 0.05). (C) Left: Time courses for Fura-2 AM measurements in cells overexpressing wild-type STIM1, STIM1-F108I, and STIM1-I115F. Store depletion was induced by 2 μM TG. Right: Cytosolic Ca2+ levels in the resting state before TG stimulation (n = 16 to 22 cells per group from at least three independent transfections). An asterisk (*) indicates mutants whose expression led to significantly increased Ca2+ levels (t test, P < 0.05). (D) Stability of the different domains of STIM1 and number of Ca2+ ions bound to the cEF-hand and nEF-hand motifs for the STIM1-F108I mutant. Left: RMSD values as a function of the simulation time for various parts of the luminal domain of wild-type STIM1. The RMSD values for the whole protein, cEF hand, nEF hand, and SAM domain are given in gray, red, magenta, and ocher, respectively. Right: Number of ions bound to the cEF hand (red histogram) and nEF hand (magenta histogram). An ion was considered bound when in the vicinity of (meaning, at a distance of <3 Å from) any of the residues constituting either EF hand. Analyses were performed over the whole simulation. (E) Starting (left) and final (right) configurations from the simulations with the STIM1-F108I mutant. Ile108 is represented by gray beads, whereas residues Val85, Leu92, Leu120 , and Phe200 are represented by white sticks. Bound Ca2+ and the different domains of STIM1 follow the same color code as in Fig. 3A. (F) RMSD values of hydrophobic amino acid interactions. For wild-type STIM1 (two black traces), the following amino acid interactions showed the highest deviations: in simulation 1: 83 to 108, 83 to 115, 83 to 103, 83 to 200, and 108 to 192; in simulation 2: 103 to 114, 101 to 114, 101 to 115, 103 to 115, and 83 to 103; and for the STIM1-F108I mutant (red trace): 103 to 199, 103 to 200, 103 to 120, 101 to 199, and 101 to 120.

Cells expressing two of the mutants with the greatest constitutive activation, STIM1-F108I and STIM1-I115F, showed raised cytosolic Ca2+ levels in the unstimulated state, as demonstrated by Fura-2 acetoxymethylester (AM) imaging (Fig. 6C). Thus, even relatively conservative, hydrophobic mutations, such as F108I and I115F, resulted in increased NFAT nuclear translocation. These mutants indicate that the STIM1 luminal domain folding is sensitive to variations in the noncanonical EF hand, because single missense mutations destabilize the compact resting configuration of STIM1. Phe108 is located at the first position of the STIM1 noncanoncial EF-hand loop, whereas Asp residues are invariantly located in canonical EF-hand loops. Thus, a STIM1-F108D mutation would mimic the first invariant Asp residue of canonical EF-hand motifs such as in calmodulin. However, attempts to restore the Ca2+-binding affinity of the noncanonical EF hand in STIM1 by introduction of an F108D-G110D double mutant result in constitutively active STIM1 aggregates (10). Here, MD simulations of STIM1-F108I showed destabilization of the noncanonical EF hand, whereas the canonical EF hand and SAM domain remained stable (Fig. 6, D and E). Unfolding of the noncanonical EF hand due to the F108I mutation correlated well with the RMSD results of amino acid interactions of the hydrophobic cleft. Destabilization of the hydrophobic cleft substantially increased after 250 ns, which indicates unfolding of the hydrophobic cleft (Fig. 6F); this event occurred simultaneously with unfolding of the noncanonical EF hand (Fig. 6D, left). Analysis of Ca2+-binding events for the F108I mutant showed that Ca2+ binding was reduced particularly in the region of the canonical EF hand (Fig. 6D, right, and fig. S5A).

Noncanonical and canonical EF-hand mutants show reduced Ca2+-binding affinity

To evaluate the effects of TAM and cancer database mutations on the isolated Ca2+ binding and structural properties of EF-SAM in an in vitro system, we next expressed and purified the human EF-SAM domain from Escherichia coli. We introduced H72R, E87Q, and F108I mutations into the bacterial expression vector because they were localized in the EF-hand region of the domain (His72 is located in the canonical entering EF-hand helix, Glu87 is located in the canonical EF-hand loop, and Phe108 is located in the noncanonical EF-hand loop). Only the expressed EF-SAM mutant proteins F108I and E87Q yielded sufficient quantities for our analyses. We found that H72R tended to degrade readily.

First, we assessed Ca2+ binding and associated changes in secondary structure by monitoring far-ultraviolet (UV) circular dichroism (CD) spectra. As previously observed, Ca2+ binding caused an increase in α-helicity for wild-type EF-SAM, indicated by the more intense negative ellipticity at ~208 and 222 nm (Fig. 7A). To estimate Ca2+-binding affinity, we plotted the change in far-UV CD signal at 222 nm as a function of different CaCl2 concentrations. We fitted the resulting binding isotherm to the Hill equation to extract the apparent Ca2+ dissociation constant (KD) in the absence of a priori assumptions regarding the number of Ca2+-binding sites. Wild-type EF-SAM titrations yielded KD values between ~0.5 and 1.3 mM (Fig. 7B). The E87Q (Fig. 7, C and D) and F108I mutant (Fig. 7, E and F) EF-SAM proteins yielded far-UV CD spectra that indicated similar α-helical levels as wild type in the absence and presence of high Ca2+ (15 mM CaCl2) (Fig. 7, C and E). In contrast, the apparent Ca2+-binding affinities were systematically lower for E87Q compared to wild-type EF-SAM proteins (KD ~ 1.2 to 3.5 mM) (Fig. 7D). The F108I mutation resulted in even more pronounced weakening of the apparent EF-SAM Ca2+-binding affinity, showing significantly increased KD values of ~3.0 to 3.4 mM compared to wild type (Fig. 7, F and G). We globally fit the Hill coefficients in each group of titrations, thereby decreasing the uncertainties of all fitted parameters. The fitted Hill coefficients were similar for the wild-type, E87Q, and F108I proteins and ranged from 1.3 to 1.4 (Fig. 7, B, D, and F).

Fig. 7 Interrogation of EF-SAM biochemical properties.

(A to F) Ca2+-induced changes in secondary structure and apparent Ca2+-binding affinities of wild-type (A and B), E87Q (C and D), and F108I (E and F) EF-SAM proteins evaluated by far-UV CD. Data in (A), (C), and (E) are representative of n = 3 independent experiments for each protein variant. Data of n = 3 independent titrations for each protein variant are shown in (B), (D), and (E). (G) Summary of Hill plot–derived apparent KD values. (H) Far-UV CD–determined apparent thermal stabilities of the EF-SAM proteins in the presence of 5 mM CaCl2. The apparent Tm values derived from the Boltzmann sigmoidal fits are shown. Data in (H) are representative of n = 2 independent experiments for each protein variant. In (A), (C), and (E), far UV-CD spectra in the absence (black) and presence (red, green, and blue) of 15 mM CaCl2 are presented. In (B), (D), and (F), Ca2+-binding curves were derived from changes in the CD signal at 222 nm as a function of increasing CaCl2 concentration. Solid red lines are the best-fit Hill plot lines. The apparent KD values of the three separate titrations are shown for each protein. The globally fitted cooperativity coefficients (n) are also indicated. Statistics in (G) represent one-way analysis of variance (ANOVA), followed by Bonferroni post hoc test.

Having determined that the Ca2+-saturated secondary structures were similar for wild-type, E87Q, and F108I proteins, we next sought to evaluate the thermal stabilities of these proteins. We used changes in α-helicity as a function of temperature in the presence of 5 mM CaCl2 as an indicator of stability. On the basis of our Ca2+ titrations that showed that E87Q and wild-type EF-SAM attained maximal α-helicity by 5 mM CaCl2, whereas the F108I protein was not fully folded at this Ca2+ concentration, we expected F108I to be the most destabilized. The apparent midpoints of temperature denaturation (Tm) were extracted from the thermal melts using Boltzmann sigmoidal equation. The EF-SAM fragment with the F108I mutation showed a destabilization of ~9°C compared to the wild-type and E87Q versions, even in the presence of 5 mM CaCl2 (Fig. 7H). The comparative Tm values were 51.4°, 51.2°, and 42.5°C for wild-type, E87Q, and F108I proteins, respectively.

We next assessed whether differences in EF-SAM transition state energies could contribute to the constitutive activity observed for E87Q and F108I. We monitored kinetics of protein unfolding induced by 2.5 M urea in the presence of 5 mM CaCl2 at four different temperatures. The data for the wild-type and mutant proteins were reliably fitted to single exponential decays (fig. S5B). The temperature and denaturant dependence of protein unfolding and folding is nonlinear, thus making it difficult to reliably extract quantitative transition state energy values from a simple Arrhenius relationship (4143). However, over the identical temperature range examined (~16° to 34°C), E87Q and wild-type EF-SAM proteins exhibited similar temperature dependence in the observed unfolding rate constants, implying similar transition state energies (fig. S5C). The F108I EF-SAM proteins unfolded ~10- to 15-fold faster than the E87Q and wild-type EF-SAM proteins (fig. S5C), contributing to the destabilization observed for this mutant protein. Because of this faster unfolding, rate constants could not be acquired over the same temperature range for the F108I EF-SAM, and the transition state energy could not be qualitatively compared to wild-type and E87Q EF-SAM proteins.

These data, in line with MD simulations, provide structural insights into why the two STIM1 mutants are constitutively active. The luminal STIM1 domain is largely destabilized by the F108I mutation compared to the E87Q mutation and wild-type STIM1. Instead, the E87Q mutant is expected to be constitutively active because of a shifted KD for Ca2+ precisely at the critical Ca2+ concentration of the ER. These data show that the canonical and noncanonical EF hands act together to maintain a stable luminal structure strictly required for Ca2+ sensing within the ER and as essential parts in STIM1 activation.

DISCUSSION

We combined functional live-cell recordings, in vitro biochemical analyses, and in silico MD simulations to determine the conformational steps of STIM1 activation upon ER Ca2+ store depletion on a molecular level. Our results revealed that STIM1 EF hands and their hydrophobic interfaces to the SAM structure are critical to lock the protein in a resting state at high ER Ca2+ concentrations. We uncovered previously uncharacterized disease-related sites in these luminal STIM1 domains that induce unfolding and STIM1 activation even at resting ER Ca2+ levels. Screening and functional analysis of STIM1 cancer database mutants and comparison to tubular myopathy mutants revealed in total 11 constitutively active mutants. The STIM1 canonical EF-hand mutations H72R, D76V, D78G, A79T, N80K, and E87Q mainly affect the negatively charged surface potential of the Ca2+-binding loop. The STIM1 noncanonical EF-hand mutations L92P, F108I, H109N, H109R, and I115F affect the hydrophobic cleft, which is formed together with the canonical EF hand and SAM domain. These cancer database mutations are additions to the pool of previously identified engineered mutations (D76A, E87A, and F108D-G110D) that trigger STIM1 dimerization or multimerization (10). The two cancer database mutations E87Q and F108I reduced Ca2+ affinity. Biochemical experiments also revealed that the STIM1-F108I peptides were less temperature stable than those of wild-type STIM1. In addition, MD simulations of selected mutants showed destabilization of the EF-SAM domain interface mainly through unfolding of the noncanonical EF hand. This is in line with an intermediate transition state of a wild-type STIM1 monomer and a dimer upon store depletion, as has been suggested on the basis of 2D NMR results (44). Our MD simulations of STIM1 mutants predict that the noncanonical EF hand is the most prone to structural luminal STIM1 unfolding. This initial unfolding subsequently destabilizes the hydrophobic cleft. Destabilization of the luminal STIM1 experiments is a starting point for dimerization that we will target in future investigations. In wild-type STIM1, these concerted conformational transitions of the noncanonical and canonical EF hands upon Ca2+ store depletion occur within a short period of time, which indicates that they are the initial key events for STIM1 dimerization or oligomerization.

A key aspect of Ca2+ store–operated activation is the number of Ca2+ ions that control the conformational transitions of STIM1. The coordination of a single Ca2+ ion by canonical EF-hand motifs has been the focus of many studies and reviews (4547). Accordingly, the first quantitative studies of the Ca2+–EF hand interaction suggested a single Ca2+ ion bound to the STIM1 EF hand (17). In contrast, the data of Gudlur et al. (16) contradict the idea of single–Ca2+-ion sensing by the STIM1 EF hand and suggest that five to six Ca2+ ions are bound. Multiple ion binding sites of the EF hand would possibly require several amino acids coordinating multiple Ca2+ ions at the same time. We identified three Ca2+ ion coordinating positions including Asp76, Asp77, and Asp78; Asp82 and Asp84; and Glu86 and Glu87, which perfectly overlay with the residues that coordinate a single Ca2+ ion in the NMR structure of STIM1 (10). In addition, four Ca2+ ions are coordinated by residues Asp89 and Glu90; Glu94 and Asp 95; Glu111 and Asp 112; and Glu118 and Asp119. These potential binding sites in the noncanonical EF-hand positions overlap with the region of the 2NQ mutation, including a neutralization of 11 charged residues in the two EF-hand structures (16). Specifically, the STIM1-2NQ mutant includes the D77N, D82N, E86Q, D89N, E90Q, E94Q, D100N, E111Q, D112N, E118Q, and D119N mutations, results in largely reduced Ca2+ binding, and shows no conformational activation after Ca2+ depletion. These mutations overlap with the canonical and noncanonical EF structures that we identified for Ca2+ binding. MD simulations revealed a predominant binding to the canonical EF hand, whereas the noncanonical EF hand (amino acids 111 to 119) only bound one to two Ca2+ ions on average. Consistent with our findings, the previous 2NQ (611) mutant, which includes noncanonical EF-hand residues, exhibits a shift in Ca2+ sensitivity but still shows five to six Ca2+-binding events.

Flexibility of the negatively charged STIM1 side chains enable the Ca2+-binding loop to coordinate up to three ions at the same time at high Ca2+ concentrations, whereas at physiological ER Ca2+ concentrations, the canonical EF hand is expected to coordinate only a single Ca2+ ion. Note that the high number of available Ca2+ ions in the MD simulations was used to maximize potential Ca2+-binding sites. The Ca2+-binding events identified here fit well with the calculated binding of five to six Ca2+ ions under a lower set of Ca2+ concentrations (16). Nevertheless, we found that the number of bound Ca2+ ions depends critically on the number of freely moveable Ca2+ ions present in the simulations. Because our MD simulations determined the exact residues that coordinate the bound Ca2+ ions, we created a set of STIM1 point mutations at the corresponding sites to interfere with individual Ca2+ binding to STIM1. Single point mutations that substituted uncharged for charged residues resulted in constitutive STIM1 activation only when the mutations interfered with the Ca2+-binding loop region of the canonical EF hand. Hence, surface Ca2+-binding positions in the noncanonical EF hand do not seem to regulate the STIM1 activation mechanism.

The identification of the cancer database mutants as constitutively active corroborates a previous computational analysis, which suggested that this luminal STIM1 domain is a structural cancer driver site (25). Certain constitutively active STIM1 mutations result in TAM, but the affected cells do not undergo apoptosis in patients with TAM (23). In addition to our insights gained into the structural aspects of these mutations, further studies are required to determine a potential role of STIM1 mutants in cancer cell progression.

Our results identified a sequential mechanism of STIM1 activation upon ER Ca2+ store depletion. The resting state is mainly stabilized by a single bound Ca2+ ion, coordinated by the canonical EF hand, but it can bind up to three Ca2+ ions simultaneously. Several other bound Ca2+ ions interact with the surface of the luminal STIM1 transiently only at Ca2+ concentrations used to maximize Ca2+ binding in the MD simulation and without affecting the initial activation mechanism. These results provide a more detailed view of the Ca2+-binding pattern for STIM1. Using TAM and cancer database mutants, our study refines the sequential activation cascade of STIM1 rearrangements at the atomic level. We identified that upon store depletion, the noncanonical and canonical EF hands unfold, thus destabilizing the hydrophobic cleft formed together with the stable SAM domain. This key event is critical to the subsequent STIM1 activation steps of dimerization or oligomerization, CC1 decoupling from the SOAR structure, coupling to Orai1, and downstream signaling of Ca2+-dependent transcription factor activation.

MATERIALS AND METHODS

Plasmids

STIM1 (accession number NM_003156) was cloned into the pECFP-C1 and pEYFP-C1 expression vectors (Clontech). Point mutations in STIM1 constructs were introduced using the QuikChange site-directed mutagenesis kit (StrataGene). The integrity of all resulting mutants was confirmed by sequence analysis (Eurofins Genomics).

Transfections

HEK-293 cells were transfected with TransFectin (Bio-Rad) with 1 μg of STIM1 constructs or mutants. For FRET measurements, 1 μg of Orai1 was cotransfected. Rat basophilic leukemia (RBL) cells were electroporated with 7 μg of wild-type STIM1 or with STIM1 mutants. Cells were regularly tested for mycoplasm contamination.

Electrophysiological recordings

RBL cells were electroporated with 7 μg of mCherry-STIM1 wild type or mutants. Electrophysiological experiments were performed 24 to 34 hours after transfection, using the patch-clamp technique in whole-cell recording configurations at 21° to 25°C. A Ag/AgCl electrode was used as reference electrode. Voltage ramps were applied every 5 s from a holding potential of 0 mV, covering a range of −90 to 90 mV over 1 s. Experiments were run with 10 mM Ca2+ bath solution, and store-dependent activation was induced by buffering cytosolic Ca2+ with 20 mM EGTA. For passive store depletion, the internal pipette solution included 145 mM Cs methane sulfonate, 20 mM EGTA, 10 mM Hepes, 8 mM NaCl, and 3.5 mM MgCl2 (pH 7.2). Standard extracellular solution consisted of 145 mM NaCl, 10 mM Hepes, 10 mM CaCl2, 10 mM glucose, 5 mM CsCl, and 1 mM MgCl2 (pH 7.4). A liquid junction potential correction of +12 mV was applied, resulting from a Cl-based bath solution and a sulfonate-based pipette solution. Currents were leak corrected by subtracting the initial voltage ramps obtained shortly after break-in with no visible current activation from the measured currents or at the end of the experiment using La3+ (10 μM).

Fluorescence-based Ca2+ imaging

RBL cells were loaded with 1 μM Fura-2 AM for 20 min at 37°C in Ringer solution containing 145 mM NaCl, 5 mM KCl, 10 mM glucose, 10 mM Hepes, and 1 mM MgCl2 + 2 mM CaCl2 for 2 mM Ca2+. The cells were then washed three times, and the coverslips were mounted on an Axiovert 100 TV microscope (ZEISS), where fluorescence was recorded from individual cells, with excitation wavelengths of 340 and 380 nm and emission wavelength at 505 nm. SOCE was triggered using 1 μM thapsigargin in 0 mM Ca2+ Ringer solution. Changes in Ca2+ were monitored using the Fura-2 340/380 fluorescence ratio and calibrated according to the method established by Grynkiewicz et al. (48). The fluorescence microscope was equipped with a monochromator (T.I.L.L. Photonics) and corresponding filter sets and allowed detection of CFP/YFP/red fluorescent protein fluorescence.

Confocal FRET microscopy

Transfected HEK-293 cells were grown on coverslips for 24 hours and subsequently transferred to an extracellular solution consisting of 140 mM NaCl, 5 mM KCl, 1 mM MgCl2, 2 mM CaCl2, 10 mM glucose, and 10 mM Hepes buffer (adjusted to pH 7.4 with NaOH). A QLC100 Real-Time Confocal System (VisiTech Int.) connected to two Photometrics CoolSNAPHQ monochrome cameras (Roper Scientific) and a dual-port adapter (dichroic, 505lp; cyan emission filter, 485/30; yellow emission filter, 535/50; Chroma Technology Corp.) was used to record fluorescence images. This system was attached to an Axiovert 200 M microscope (ZEISS, Germany) in conjunction with two diode lasers (445 nm, 515 nm) (Visitron Systems). VisiView 2.1.1 software (Visitron Systems) was used for image acquisition and control of the confocal system. Illumination times for CFP/FRET and YFP images that were recorded consecutively with a minimum delay were about 800 ms. Image correction to address cross-talk and cross-excitation was performed before the calculation. To this end, appropriate cross-talk calibration factors were determined for each construct on each day of the FRET experiment. After threshold determination and background subtraction, the corrected FRET (Eapp) was calculated on a pixel-to-pixel basis with a custom-made software integrated into MATLAB 7.0.4 according to the method published by Zal and Gascoigne (49), with a microscope-specific constant G value of 2.0. All experiments were performed at room temperature.

Colocalization analysis

The same equipment as for confocal FRET microscopy was used. R value calculations for CFP and YFP overlay was performed to quantify colocalization.

Subcellular localization

The same technical equipment as for confocal FRET microscopy was used. ImageJ was used to analyze subcellular localization of transcription factors by intensity measurements of the cytosol and nucleus. Three different populations were distinguished with different nucleus/cytosol ratios: inactive (<0.85), homogenous (0.85 to 1.15), and active (>1.15).

All data are presented as means ± SEM for the indicated number of experiments. Statistical significance was determined by unpaired two-sided Student’s t tests for comparison of two groups (using OriginPro 2017). Statistical significance was set to P < 0.05.

Computational details

The structure for the luminal domain of STIM1 was obtained from the PDB ID code 2 K60 (10). Because STIM1 is a type I single-pass TM protein (50), the missing TM α helix (residues 201 to 236) was rebuilt in silico with abalone software (www.biomolecular-modeling.com/Abalone/). This minimal reconstructed version of STIM1 was then embedded within a phospholipidic bilayer composed of POPC lipids by means of the online input generator CHARMM-GUI (51). The upper and lower leaflets of the membrane both contained 120 lipids. The systems were solvated by roughly 20,100 water molecules. Last, for all simulations, the initial ionic strength was set to 150 mM.

MD setup and parameters

All systems were equilibrated using the built-in scripts (52) provided by CHARMM-GUI, where the systems were allowed to relax slowly by stepwise reduction of the constraints applied to the lipids and the protein. Regular MD without harmonic constraints was then performed on the systems. An equilibrated luminal STIM protein conformation taken after a 100-ns simulation in the presence of CaCl2 was used to build the mutants. For each system, two replicas were performed, each 0.5 μs long.

The GROMACS (53) software version 5.1.4 was used to perform all MD simulations, which were conducted at 310 K under 1 atm using the Nosé-Hoover thermostat (54, 55) and Parrinello-Rahman barostat (56, 57), respectively. Pressure coupling was applied semi-isotropically, decoupling the z dimension, normal to the membrane, from the x and y dimensions. Short-range interactions of Lennard-Jones were cut off at 1.2 nm. Long-range electrostatic interactions were treated using particle mesh Ewald (58) with a real-space cutoff of 1.2 nm. Covalent hydrogen bonds were constrained using LINCS algorithm (59). The leap-frog algorithm (60) with a time step of 2 fs was used to integrate Newton’s equation of motion.

CHARMM36 parameters were used for both protein (61) and lipids (62). The TIP3p (63) model was used for water molecules. Ions were treated using a rescaled charge paradigm (6467). In this model, the electronic polarization is included in a mean field by rescaling the ionic charges by a factor proportional to the dielectric constant of water. These procedures better reproduce the properties of the ion in water and its hydrated structure (68) and improve the interactions with the backbones of proteins and the free energy of binding of Ca2+ ions to Ca2+-signaling proteins (68) . Minimization and equilibration followed the CHARMM-GUI protocols in six steps (52, 69).

Visual molecular dynamics (VMD) (70) was used to visualize the simulations. VMD and MD analyses (71) were used to analyze the simulations. Figures were generated using Matplotlib (72). For the ionic affinity studies, the wild-type STIM1 was placed in the presence of the following cations: K+, Na+, Ca2+, and Mg2+. In all cases, Cl ions were always used as counterions. The physiological concentration of Ca2+ within the ER is in the micromolar range. However, trying to mimic such low concentration would require the simulation of a fairly huge system (73), mainly consisting of water molecules. In our setup, the presence of only one Ca2+ ion solvated in bulk water represented a concentration of 0.3 mM.

Because of the inherent problem of having to work with physiologically low concentration, we relied on a strategy using flooding simulations in which higher amounts of ligand molecules (corresponding to nonphysiological concentrations) are added to the simulations to maximize the sampling of binding events and coverage of binding sites (3638). Because possible multiple binding sites have been reported, we sought to explore which domains of the protein might be more occupied by Ca2+ ions (16). We aimed at investigating which are the regions more likely to be occupied by Ca2+ ions on the luminal domain of STIM1, the existence of probable multiple sites occupied by multiple ions, and how these bindings affect the dynamic and configuration of the luminal domain.

To this end, we removed the initial ion present in the NMR structure and put the luminal domain close to a reservoir containing Ca2+ ions. The system was prepared as such that no ions were in contact with the protein at the beginning of the simulations. The starting concentration of Ca2+ was set to 150 mM. However, because the protein’s binding sites for the ions and the membrane (74) also act as a buffer for Ca2+ ions, the number of free Ca2+ ions present in the simulation box dropped substantially during the simulation (only about 10 of 51 Ca2+ ions remained unbound within the bulk water). As a result, protein and membrane remained in contact with a reservoir of Ca2+ ions, which allowed both to exchange Ca2+ ions with the bulk water.

The final concentration of free Ca2+ ions after equilibration was calculated according toC=ncalciumVwater=NcalciumNa×Vwaterwhere C is the concentration, ncalcium the amount of Ca2+ in mole, Vwater is the volume occupied by bulk-like water molecules, Ncalcium is the number of free Ca2+ ions, and Na is the Avogadro constant.

The average x, y, and z dimensions of the simulation box were 8.6, 8.6, and 12.4 nm, respectively. However, calculation of the concentration required the thickness of the membrane to be taken into account, which was therefore subtracted from the z dimension. In this case, the distance between opposing choline moieties in the bilayer was used, which corresponds on average to 5 nm. As a result, after equilibration of the electrolytes, the concentration of free Ca2+ ions dropped to 30 mM.

MD simulations using classical force fields have been reported to be ill-suited to properly study binding energetics of divalent cations to structural motifs such as EF hands because the lack of polarization tends to favor ions with increased size (75). Furthermore, selectivity and sensitivity toward different divalent cations are fine-tuned by a competition between electrostatics, polarization, as well as flexibility and geometry of the binding pocket (75, 76). Still, studies (68, 75, 77) have shown that rescaling charges according to a continuum mode substantially improves binding free energies. However, such calculations are yet out of the scope of this study as we sought to explore the multiple binding sites aspects of the EF hands present in STIM1. Calculations performed by Jing et al. (75) predicted that without polarization explicitly taken into account, Mg2+ ions will always be favored over Ca2+. In our cases, we tried to alleviate such outcome as the protein was stripped from all ions at the beginning of our simulations. Stronger interaction between magnesium and water molecules allowed the Ca2+ ions to first occupy the sites on the EF hands. MD simulations using polarizable force fields tend to be more computationally expensive (78) compared to MD with regular force fields, thus limiting exploration of configurations adopted by wild type or mutants as well as the exchange of Ca2+ ions between binding sites. On the basis of the herein reported results and recent findings from Gudlur et al. (16), it would be interesting to explore the impact of a surplus of Ca2+ ions at the outskirt of the binding loop on the binding of Ca2+ ions within the binding pocket.

To investigate the stability of the different domains of the proteins, we calculated the RMSD for different segments of the luminal domain of STIM1, namely, the SAM domain and the canonical and noncanonical EF hands. These latter domains comprised residues 63 to 96, 97 to 128, and 130 to 200 for the canonical and noncanonical EF hands and SAM domains, respectively. To this end, the RMSD analysis tool implemented within VMD was used to compute the RMSD values for reach of these domains and plotted as a function of the simulation time.

The RMSD was not calculated over the positions of the Cα of the protein residues but over distances between each pair of residues constituting the hydrophobic cleft (71, 74, 75, 83, 85, 91, 92, 96, 101, 103, 108, 114, 115, 117, 120, 192, 195, 199, and 200). This RMSD is defined as followedRMSDDistance=1NiNδi2where N is the number of distances between the residues within the hydrophobic cleft and δ is the difference between the distances of two residues l, k forming the hydrophobic cleft at the beginning of the simulation (t = 0) and at time t = tδl,k=d(residuel,residuek)t=0d(residuel,residuek)t=t'dl,k(t=0)dl,k(t=t')

To identify the pair of residues that contributes most to this RMSD, we computed the SD from the initial distance value between the residues k and l, sl, k bySl,k=iN'(dl,k(t0)(dl,k(ti))2N'1where N′ is the number of frames and dl,k(t0) and dl,k(ti) are the distances between residues l, k at the beginning of the simulation and at time ti, respectively.

To explore the flexibility of the domains within the protein, we calculated the RMSF for the whole luminal domain for each residue. The RMSD is an average of the fluctuations of the atomic positions over all the residues, whereas the RMSF is an average over time performed for each residue iRMSFi=t(ri,t0ri,t)2where ri,t0 is the initial position of residue i and ri,t is the atomic positions of residue i at time t. The RMSF was calculated over the last 300 ns of each simulation.

Recombinant protein expression and purification

The human STIM1 EF-SAM protein (National Center for Biotechnology Information accession AFZ76986.1) was expressed and purified essentially as previously described (79, 80). Briefly, pET-28a plasmids harboring STIM1 EF-SAM (residues 58 to 201) were introduced in BL21(DE3) E. coli. Protein expression was induced with 0.2 mM isopropyl-β-d-thiogalactopyranoside and maintained for ~16 hours at 22°C and 190 rpm. The 6× His-tagged protein was purified using the Ni2+–nitriloacetic acid agarose beads under denaturing conditions, essentially as described by the manufacturer (HisPur, Thermo Fisher Scientific). EF-SAM protein was refolded overnight by dialysis (3500 molecular weight cutoff; Thermo Fisher Scientific) in 20 mM tris, 150 mM NaCl, and 5 mM CaCl2 (pH 8). The 6× His-tag was cleaved by adding 1 U of bovine thrombin (Calbiochem) per milligram of protein and incubating overnight. The digested protein mixture was separated using HiTrap Q FF anion exchange chromatography (GE Healthcare Inc.). A final overnight dialysis step was used to exchange the pure protein into experimental buffer. The Ca2+-depleted EF-SAM proteins were prepared by overnight incubation in 50 mM EDTA, followed by 20 × 20 × 20 × 20–fold exchange into Ca2+-free buffer by ultrafiltration. Protein concentrations were estimated using an extinction coefficient at 280 nm of 1.61 (mg ml−1)−1 cm−1. The E87Q, F108I, and H72R missense point mutations were introduced into the pET–28a-EF-SAM plasmids by polymerase chain reaction–mediated site-directed mutagenesis using the QuikChange kit according to the manufacturer’s guidelines (Agilent).

Far-UV CD spectroscopy

Far-UV CD experiments were performed using a Jasco J-810 Spectropolarimeter equipped with a PTC-423S Peltier temperature controller. Far-UV CD spectra were collected at 4°C using a 0.1-cm pathlength quartz cuvette. Spectra are an average of three acquisition scans acquired between 240 and 200 nm at 20 nm min−1 and using an 8-s response time.

Thermal melts were constructed by monitoring the change in CD signal at 222 nm between 4° and 80°C. The temperature ramp was set to 1°C min−1, and the data were collected using an 8-s response time and 1°C data pitch. The apparent midpoints of temperature denaturation (Tm) were estimated from Boltzmann sigmoidal equation fitted to the data.

To assess Ca2+ binding, we acquired spectra as indicated above in the presence of increasing amounts of CaCl2 (between 0 and 15 mM CaCl2). Changes in CD signal at 222 nm as a function CaCl2 concentration were then plotted and fitted using the Hill equationfb=[Ca2+]n KDn+[Ca2+]n [1],and ObsY=Isat×fb+I0×(1fb) [2]where fb is the fraction of protein bound, [Ca2+] is the molar concentration of Ca2+, KD is the dissociation constant, n is the Hill coefficient, ObsY is the observed CD signal at 222 nm, Isat is the CD signal at saturation, and I0 is the initial CD signal. All CD experiments were performed with 0.25 to 0.30 mg ml−1 protein samples buffered in 20 mM tris, 150 mM NaCl (pH 7.5), with or without CaCl2 as indicated.

Kinetics of EF-SAM protein unfolding were monitored as changes in intrinsic fluorescence using an excitation wavelength of 280 nm and an emission wavelength of 330 nm. Data were acquired on a Cary Eclipse spectrofluorimeter (Agilent Inc.), which was temperature-equilibrated with a Lauda E100 water bath (Brinkmann Inc.). Excitation and emission slit widths were set to 5 and 20 nm, respectively. Protein (0.2 μM final concentration) was added to temperature-equilibrated solutions (600 μl) of 2.5 M urea, 20 mM tris, 150 mM NaCl, and 5 mM CaCl2 (pH 7.5). Kinetics of unfolding data were fit to a single exponential decayy=Ae(kx)+mx+bwhere y is the fluorescence signal, A is the amplitude of fluorescence change, k is the unfolding rate constant (kunfolding), m is the slope of the unfolded baseline, b is the intercept of the unfolded baseline, and x is time (in seconds).

SUPPLEMENTARY MATERIALS

stke.sciencemag.org/cgi/content/full/12/608/eaax3194/DC1

Fig. S1. NFAT activation by constitutively active STIM1 mutants.

Fig. S2. Constitutive CAD dissociation for the STIM1-H72R mutant.

Fig. S3. Ca2+ ion binding to the EF-SAM domain.

Fig. S4. Ca2+-binding probability and RMSD values for the STIM1 canonical EF-hand mutants.

Fig. S5. Altered protein unfolding for the STIM1-F108I mutant.

Movie S1. Multiple Ca2+-binding events to the EF-hand domains of STIM1.

Movie S2. A single bound Ca2+ ion maintains the luminal STIM1 structure.

REFERENCES AND NOTES

Acknowledgments: We thank K. Groschner and I. Abfalter for scientific advice and for carefully proofreading the manuscript. Funding: We acknowledge the support by the Austrian Science Fund (FWF) through project P28701 to R. Schindl, P27263 to C.R., P32075 and P27872 to I.F., P28123 to M.F., P27641, P30567, and LIT-2018-5-SEE-111 to I.D., and BMWFW HSRSM (PromOpt2.0) to C.R.; by the Czech Research Infrastructure for Systems Biology C4SYS (LM2015055); and by a Memorandum of Agreement between the Institute of Microbiology, Czech Academy of Sciences, and the College of Biomedical Sciences, Larkin University and by the Natural Sciences and Engineering Research Council (NSERC 05239) to P.B.S. D.B. was supported by the EFRR project Interreg Austria–Czech Republic “Czech-Austrian Center for Supracellular Medical Research (CAC-SuMeR, no. ATCZ14)” and the Czech Science Foundation (19-20728Y). R. Schindl, C.R., and R.H.E. were funded in part through a European Cooperation in Science and Technology (COST) action (BM1406) and by the program Inter-COST (project LTC17069 to R.H.E.). Access to the National Grid Infrastructure Metacentrum, and provided computational resources are acknowledged. Author contributions: R. Schindl conceived ideas, directed the work, and designed the study. I.F., V.L., and T.S. designed and generated all plasmid constructs. R. Schindl performed initial analysis of tumor genomes. A.K. performed patch-clamp experiments. R. Schober performed fluorescence experiments. R. Schober and L.W. performed Ca2+ imaging experiments. D.B. and R.H.E. performed computational modeling and MD simulations, as well as analyzed theoretical data and predictions. J.Z., M.Z., and P.B.S. performed protein expression and purification and biochemical experiments. R. Schindl analyzed data with input from the other authors. R. Schindl and R. Schober wrote the manuscript with input of D.B., V.L., A.K., I.F., P.B.S., M.F., L.W., I.D., C.R., and R.H.E. All authors discussed the results and commented on the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. The plasmid vectors used in Fig. 6 and fig. S4 (B and C) are available from P.B.S. and require a material transfer agreement with University of Western Ontario.

Stay Connected to Science Signaling

Navigate This Article