Structure/activity relationship of thapsigargin inhibition on the purified Golgi/secretory pathway Ca2+/Mn2+-transport ATPase (SPCA1a)

The Golgi/secretory pathway Ca2+/Mn2+-transport ATPase (SPCA1a) is implicated in breast cancer and Hailey-Hailey disease. Here, we purified recombinant human SPCA1a from Saccharomyces cerevisiae and measured Ca2+-dependent ATPase activity following reconstitution in proteoliposomes. The purified SPCA1a displays a higher apparent Ca2+ affinity and a lower maximal turnover rate than the purified sarco(endo)plasmic reticulum Ca2+-ATPase (SERCA1a). The lipids cholesteryl hemisuccinate, linoleamide/oleamide, and phosphatidylethanolamine inhibit and phosphatidic acid and sphingomyelin enhance SPCA1a activity. Moreover, SPCA1a is blocked by micromolar concentrations of the commonly used SERCA1a inhibitors thapsigargin (Tg), cyclopiazonic acid, and 2,5-di-tert-butylhydroquinone. Because tissue-specific targeting of SERCA2b by Tg analogues is considered for prostate cancer therapy, the inhibition of SPCA1a by Tg might represent an off-target risk. We assessed the structure-activity relationship (SAR) of Tg for SPCA1a by in silico modeling, site-directed mutagenesis, and measuring the potency of a series of Tg analogues. These indicate that Tg and the analogues are bound via the Tg scaffold but with lower affinity to the same homologous cavity as on the membrane surface of SERCA1a. The lower Tg affinity may depend on a more flexible binding cavity in SPCA1a, with low contributions of the Tg O-3, O-8, and O-10 chains to the binding energy. Conversely, the protein interaction of the Tg O-2 side chain with SPCA1a appears comparable with that of SERCA1a. These differences define a SAR of Tg for SPCA1a distinct from that of SERCA1a, indicating that Tg analogues with a higher specificity for SPCA1a can probably be developed.

As an important secondary messenger, the partitioning of intracellular Ca 2ϩ controls a wide range of cellular responses, including contraction, secretion, cell growth, differentiation, proliferation, and cell death (1,2). A sub-micromolar basal cytosolic Ca 2ϩ concentration range is strictly maintained by intracellular Ca 2ϩ pumps, whereas Ca 2ϩ channel activity produces confined spatial and temporal cytosolic Ca 2ϩ signals up to 10 -100 M that control various cellular activities. Ca 2ϩ transporters belonging to the P2-type Ca 2ϩ transport ATPases maintain the low housekeeping cytosolic Ca 2ϩ levels and reduce the cytosolic Ca 2ϩ concentration when a Ca 2ϩ -signaling event ceases. The P2-type ATPases encompass the plasma membrane Ca 2ϩ -ATPase, SERCA, 2 and SPCA Ca 2ϩ pumps that perform active Ca 2ϩ transport across the plasma membrane, the intracellular membranes of the sarco(endo)plasmic reticulum, and the Golgi/secretory pathway, respectively (3). SERCA and SPCA belong to the subfamily P2A and share around 30% sequence identity and over 40% similarity (4).
In humans, the SPCA proteins are encoded by two genes (ATP2C1 and ATP2C2) (4 -7). SPCA1 is the housekeeping isoform, which by transporting Ca 2ϩ and Mn 2ϩ into the Golgi lumen supports essential Golgi/secretory pathway functions. Ca 2ϩ /Mn 2ϩ transport is required for proper protein folding, enzymatic activities (e.g. glycosylation), sorting, and/or trafficking of proteins (8). In conjunction with SERCA, this ATPase also provides Ca 2ϩ -storage pools for subsequent Ca 2ϩ -release events (4,9,10). A total of four SPCA1 splice variants exist, SPCA1a-d, differing only at the C terminus. Although SPCA1a is considered to be the housekeeping variant, the differential tissue/subcellular distributions of the SPCA1 isoforms remain largely unexplored (11). The more restricted expression pattern of SPCA2 in cultured hippocampal neurons, colon, secretory acini, and luminal epithelial cells of mouse mammary tissue suggests that the second isoform may play a more specialized function, for instance in secretion (6,(12)(13)(14).
In humans, SPCA1 haploinsufficiency, due to heterozygous loss-of-function mutations, causes Hailey-Hailey disease (OMIM169600), a heritable autosomal-dominant skin disorder characterized by recurrent skin vesicles and erosions that typically arise in adulthood (15,16). In addition, SPCA1a and/or SPCA2 proteins are up-regulated in mammary gland during lactation and in various breast cancer sub-types thereby contributing to a pathological Ca 2ϩ dyshomeostasis (17,18). Independent of its transport activity, SPCA2 interacts with and activates the plasma membrane Ca 2ϩ channel Orai1 via the N and C termini (18), leading to store-independent Ca 2ϩ entry (18) and subsequent Ca 2ϩ transfer into the secretory pathway (19). During lactation, this system contributes to the cellular uptake of Ca 2ϩ in mammary gland epithelial cells for the subsequent release of Ca 2ϩ into the milk (20).
In general, SPCA1 isoforms display a higher Ca 2ϩ affinity but a lower maximal turnover rate than SERCA isoforms (21), but so far, a detailed functional analysis of SPCA isoforms has remained difficult due to the confounding presence of other abundant housekeeping Ca 2ϩ -ATPases in cellular membrane fractions. In addition, although several potent SERCA inhibitors (thapsigargin (Tg), cyclopiazonic acid (CPA), and 2,5-ditert-butylhydroquinone (BHQ)) are widely used to selectively inhibit endoplasmic reticulum Ca 2ϩ transport (4,22), selective pharmacological tools for SPCA inhibition are lacking. Because of the close sequence homology between SERCA and SPCA isoforms, several of the so-called highly selective SERCA inhibitors are also reported to inhibit SPCA in isolated membrane fractions of various organisms, albeit at much higher concentrations (4,(23)(24)(25).
The inhibition of SPCA1a by Tg may give rise to some concerns in connection with the clinical use of Tg analogues that target SERCA for prostate cancer therapy. In this therapeutic approach, a non-toxic, cell-impermeable pro-drug is administered that consists of a Tg analogue, which is coupled to a peptide that is specifically cleaved by the prostate-specific antigen (PSA) protease (26,27). Extracellular cleavage of the peptide by local PSA in or near the prostate cancer releases the toxic Tg analogue Leu-8ADT, which is taken up by neighboring PSAexcreting cancer cells inducing targeted cell death (27). The estimated 3 orders of magnitude difference in the binding affinity of Tg for SERCA and SPCA may be considered sufficiently discriminatory to enable the use of Tg analogues for selective SERCA inhibition in vitro and in vivo. However, it remains unclear whether this is also true for other Tg analogues.
Toward the design of Tg analogues for clinical therapy related to SPCA dysfunction (see "Discussion"), we wanted to understand the off-target effect of Tg on SPCA1a, the closest relative of SERCA, by determining the structure-activation relationship (SAR) of Tg for SPCA1a. Moreover, due to the reported micromolar affinity of Tg to SPCA1 (24), one might consider Tg as a possible lead compound for the development of an SPCA1a inhibitor. Insights in the SAR of Tg for SPCA1a may aid the design of specific SPCA1a inhibitors as powerful pharmacological tools.
In this study, we purified the human SPCA1a isoform from a yeast recombinant expression system and reconstituted it into lipid vesicles. This purified system eliminates the influence of other membrane-bound contaminants or Ca 2ϩ -ATPases, allowing the biochemical characterization of SPCA1a relative to purified and reconstituted SERCA1a. In particular, we demonstrate that SPCA1a activity is sensitive to the lipid environment, and we show that Tg and cholesteric compound sensitivity of SPCA1a may depend on the same or the overlapping binding pocket formed by transmembrane helices M3, M5, and M7. Finally, we have established the SAR of Tg for SPCA1a to facilitate the design of more selective SPCA inhibitors.

Affinity purification of recombinant human SPCA1a from Saccharomyces cerevisiae
The relatively low expression level of SPCA1a in native tissues, typically in a background of high SERCA expression, hampers biochemical analysis of SPCA1a. As an alternative approach, we purified His 8 -tagged human SPCA1a via Ni-NTA affinity chromatography from S. cerevisiae, which serves as an excellent model system to produce and purify P-type transport ATPases (8). SDS-PAGE analysis reveals that SPCA1a monomer and dimer bands were highly enriched in the elution fractions (Fig. 1A). Using mass spectrometry, we observed a molecular mass of 101.8 kDa, close to the predicted molecular mass of 101.7 kDa and high purity (Ͼ95%) of the human SPCA1a (Fig.  1B). The concentrated His-tagged SPCA1a in the elution fraction was further demonstrated by immunoblotting with SPCA1 ( Fig. 1C) and His tag (Fig. 1D)-specific antibodies.

SPCA1a reconstitution into proteoliposomes restores Ca 2؉ -dependent ATPase activity
We found that purified and lipid-free SPCA1a became reversibly inactivated in detergent solution (Fig. 2E). This we considered to be a result of either a removal of specific lipid component(s) essential for activity or loss of physical membrane constraints originally imposed by the membrane lipids. To restore a functional lipid environment, we reconstituted as described under "Experimental procedures" the proteoliposomes from detergent-solubilized SPCA1a by addition of defined lipid components, followed by removal of detergent with Bio-beads. For this, 600 g of phosphatidylcholine (PC) was first mixed with 200 g of purified and detergent-solubilized SPCA1a (i.e. the protein was solubilized with detergent at a lipid/protein ratio of 3:1, w/w). The efficiency of SPCA1a incorporation into liposomes was compared for three types of detergents, i.e. Triton X-100 (PC/Triton X-100 ϭ 8:1, 6:1, 5:1, 4:1 (w/w)), n-dodecyl ␤-D-maltoside (DDM) (PC/DDM ϭ 1:1 (mol/mol)), and octaethylene glycol monododecyl ether (C12E8) (PC/C12E8 ϭ 1:2 (w/w)), in line with already established protocols for SERCA1a (28,29). Reconstitution with Triton X-100 rendered the most efficient incorporation of SPCA1a into proteoliposomes, especially at a high lipid to detergent ratio ( Fig. 2A). Importantly, we were able to recover the Ca 2ϩ -dependent ATPase activity of the purified SPCA1a reconstituted with Triton X-100 or DDM but not with C12E8. Moreover, we determined that a lipid/detergent ratio of 20:1 generated the most active SPCA1a in proteoliposomes ( Fig.   Figure 2. Optimization of SPCA1a reconstitution into proteoliposomes. A, SDS-polyacrylamide gel stained with SPCA1a proteoliposomes generated with the detergents Triton X-100, DDM, or C12E8. Same volume of samples was loaded on the gel after reconstitution. The lipid/Triton X-100 ratios (w/w) are indicated above the lanes. B, comparison of the SPCA1a ATPase activity following reconstitution with a different (w/w) ratio of lipids/Triton X-100. C and D, 20% weight percent of the indicated lipid(s) were supplemented to PC for SPCA1a reconstitution, and the maximal Ca 2ϩ -dependent ATPase activity was determined at 1 M free Ca 2ϩ concentration (n Ն 3). The ATPase activities were normalized to the activity when only PC was used. Comparison of the specific SPCA1a ATPase activities in the presence of indicated phospholipids (C), CL, SM, or the combination of CL and SM (1:1 molar ratio) (D). E, comparison of the specific ATPase activities between SERCA1a and SPCA1a in proteoliposomes, and SPCA1a in lipid-free, DDM-solubilized state (n ϭ 3). The DDM-solubilized SPCA1a was purified without supplement of PC in the buffers. F, structure of CHEMS. G, dose-responses of CHEMS on reconstituted SERCA1a and SPCA1a ATPase activities. P, purified SPCA1a; L, ladder; PA, phosphatidic acid; PE, phosphatidylethanolamine; PS, phosphatidylserine; PI, phosphatidylinositol; CL, cholesterol; SM, sphingomyelin; CHEMS, cholesteryl hemisuccinate. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. 2B). Therefore, all further reconstitutions were carried out with a lipid/protein/detergent weight ratio of 60:20:3.

Lipids phosphatidic acid, sphingomyelin, cholesterol, and linoleamide/oleamide modulate SPCA1a activity
Membrane proteins are functionally adapted to the local lipid environment of their cellular membranes. We therefore analyzed the impact of different lipid types on the Ca 2ϩ -dependent ATPase activity of SPCA1a, which in human colon adenocarcinoma cells is sensitive to cholesterol (CL) and is preferentially associated with lipid rafts (30). For reconstitution, various phospholipids (Fig. 2C), CL, sphingomyelin (SM), or a combination of the latter two (SM ϩ CL) (Fig. 2D) were supplied together with PC, and SPCA1a ATPase activity was assessed. Compared with PC alone, the presence of phosphatidic acid (PA), SM, or SM/CL slightly but significantly enhanced activity, whereas phosphatidylethanolamine (PE) lowered the SPCA1a ATPase activity. In further experiments, PA was routinely added during SPCA1a reconstitution, as has been done previously for SERCA1a reconstitution (29), although we cannot exclude that PA may be a low-abundance lipid in the native Golgi membrane (31,32).
Next, we directly compared the biochemical properties of purified and reconstituted SPCA1a (4:1 PC/PA ratio) and SERCA1a, the skeletal muscle Ca 2ϩ -transport ATPase, which was purified and reconstituted according to a well established protocol (29) (with a 9:1 PC/PA ratio as in Ref. 33) (Fig. 2E). Compared with SERCA1a, SPCA1a displays a higher apparent Ca 2ϩ affinity (K m of 0.125 Ϯ 0.004 and 0.088 Ϯ 0.005 M, respectively) and a 5-fold lower turnover rate (V max of 12.29 Ϯ 0.13 and 2.54 Ϯ 0.04 nmol of P i /g/min), in line with previous reports (21).
Although we previously reported that SPCA1a and SERCA1a respond differently to CL (30), we observed here little functional effects of incorporating 20% CL upon reconstitution. However, by replacing CL with cholesteryl hemisuccinate (CHEMS), a more soluble cholesterol derivative (Fig. 2F), we found that it affected both the reconstituted SPCA1a or SERCA1a activities (Fig. 2G). SERCA1a displayed a concentration-dependent inhibition by CHEMS with an IC 50 (defined as the concentration that leads to 50% inhibition of activity) of 8.8 Ϯ 0.66 M and a maximal inhibition of 80% at around 100 M. For SPCA1a, a slight and not significant activation up to 106% was observed at lower concentrations, whereas higher concentrations above 6 M impaired SPCA1a to a maximal inhibition of 47% at 300 M, in line with the previous observation that SPCA1a and SERCA1a respond differently to CL (30).
Because two endogenous lipids oleamide and linoleamide have recently been reported to be inhibitors of SPCA2 (34), we tested here whether these lipids also modify SPCA1a or SERCA1a activity (Fig. 3, A and B). Up to 1 mM, oleamide only partially inhibited SPCA1a and SERCA1a (Fig. 3A), whereas complete inhibition was previously reported for SPCA2 (34). In contrast to SERCA1a, linoleamide fully inhibited SPCA1a with an IC 50 of 30 M (Fig. 3B), which is 8-fold higher than the value that has been reported for SPCA2 (34), although a side by side SPCA1a/2 comparison could not be performed.

Several SERCA inhibitors block SPCA1a activity at micromolar concentration
To further validate the reconstituted SPCA1a system, we verified whether bisphenol (35,36) and the SERCA inhibitors Tg, CPA, and BHQ also inhibit SPCA1a (24,25). We confirmed that all compounds completely inhibit SERCA1a and SPCA1a, but we show that the compounds are effective in different concentration ranges for SERCA1a and SPCA1a (Table 1 and Fig. 3, C and D). For SERCA1a, the IC 50 value of Tg falls below 10 nM, whereas higher (sub-M) IC 50 values are observed for CPA and BHQ. 2,2Ј-Methylenebis(6-tert-butyl-4-methylphenol) (bisphenol, BP) displays the lowest potency with an IC 50 above 1 M. For SPCA1a, the IC 50 values of Tg and CPA are at least 3 orders of magnitude higher than for SERCA1a, whereas the IC 50 value of BHQ is only 20 times higher. The IC 50 value of BP is comparable for the two pumps, indicating that unlike what was suggested by a previous study (35), BP cannot be considered as a more selective inhibitor of SPCA1a than of SERCA1a. Remarkably, the slopes of the dose-response curves for BP (SERCA1a) and CPA (SPCA1a) are higher than 2 (2.9 Ϯ 0.1 and 2.3 Ϯ 0.1, respectively). This was also observed in earlier studies (25,35) and may reflect differences in the cooperativity (37). In conclusion, the SERCA inhibitors Tg, CPA, and BHQ also inhibit SPCA1a, although at much higher concentrations.

Tg scaffold together with the O-2 moiety are mainly responsible for the inhibition of SPCA1a
Of the so-called specific SERCA inhibitors, Tg has the highest affinity for SPCA1a. Insights into the SAR of Tg on SPCA1a might be helpful to aid the development of Tg analogues with either low or high SPCA1a affinity. We therefore explored the SAR of Tg for SPCA1a and compared it with the well established SAR of Tg for SERCA1a (38).
First, we compared the Tg-binding pocket in the SERCA1a-Tg structure and in an SPCA1a homology model in the same E2

Inhibition of SPCA1a by thapsigargin
conformation. In SERCA1a, the non-flexible tricyclic guaianolide nucleus of Tg provides a scaffold for the optimal position of side chains within the Tg-binding pocket formed by transmembrane helices M3, M5, and M7 (Fig. 4B). The lipophilic groups at O-3, C-4, O-8, and O-10 of Tg (Fig. 4A) critically determine the hydrophobic interaction and provide complementarity to the binding cavity in SERCA1a (39,40). In the SERCA1a-Tg structure (41), the O-8 moiety of Tg orients to the interior of M3 and M5, whereas the O-2 chain is located in the interface between the transmembrane region and the membrane (42). By comparing the residues in the Tg-binding pocket of SERCA1a ( Fig. 4B) and the homology model of SPCA1a (Fig. 4C), we anticipate that the Tg-binding affinity, mechanism of inhibition, and/or positioning of Tg are affected by several amino acid substitutions, including the homologous SERCA1a/SPCA1a substitutions labeled in Fig. 4, B and C (Phe-256/Leu-265, Val-263/Tyr-272, Ile-765/Leu-707, Ile-829/Leu-771, and Phe-834/ Ile-776), which might result in an altered SAR of Tg for SPCA1a as compared with SERCA1a.
To establish the SAR of Tg for SPCA1a, we measured the IC 50 values of a series of Tg analogues with various side chain substitutions. First, the acyl groups O-2, O-3, O-8, and O-10 ( Fig. 4A) were one by one replaced by -OH or -H, which significantly increases the IC 50 of Tg in SERCA1a (39,40,43). Remarkably, only the removal of the Tg O-2 chain led to a significantly higher IC 50 value in SPCA1a, suggesting that of all tested side groups O-2 contributes the most to the inhibitory effect of Tg in SPCA1a (Table 2). Based on the calculation of the free energy change for binding caused by the removal of a side chain, the O-2 contribution is comparable in SPCA1a and SERCA1a, whereas O-3, O-8, and O-10 contribute much less in SPCA1a ( Table 2). The absence of O-3, O-8, and O-10 contributions at least partially explains the lower Tg affinity in SPCA1a than in SERCA1a.
Besides the side-chain contribution by O-2, the Tg scaffold might be mainly responsible for the inhibitory effect of Tg in SPCA1a. Indeed, like in SERCA1a, the conversion of the 7,11gem diol in Tg into an epoxide led to a dramatic reduction of the inhibitory effect in SPCA1a (Table 2), presumably by the loss of hydrogen donors in the molecule or by the changed conformation of the tricyclic nucleus due to the conversion of the stereochemistry at C-11. Of note, some Tg analogues require a longer pre-incubation time with SERCA1a to obtain the full inhibitory potential, which can be explained by a time-dependent induced fit of Tg into the SERCA1a pocket (44). However, the weaker inhibitory potential of the Tg analogues in which the octanoyloxy group has been substituted with a hydrogen atom (⌬O-2), and Tg-epoxide did not significantly increase after a longer pre-incubation time with SPCA1a (Table 2).

Only the O-2 moiety of Tg binds to a spatially restricted pocket in SPCA1a
In the further exploration of the poor contribution of the side chains at O-3, O-8, and O-10 to the affinity of Tg to SPCA1a, we considered the possibility that it might arise from a more spacious Tg-binding cavity than in SERCA, which in a more flexible way would accommodate these Tg side groups. This hypothesis was tested by replacing the O-2, O-3, O-8, and O-10 moieties with more bulky groups (Table 3). Replacing O-3 and O-10 with a slightly bulkier chain (HzL09012012 and HzL12012012, respectively) is well tolerated, whereas a more bulky biphenyl group on O-3 (HzL170809) or O-10 (HzL271009) led to a 7-and 49-fold higher IC 50 , respectively (Table 3). However, the higher IC 50 was significantly reduced after a longer pre-incubation time, indicating that a sufficiently large threedimensional space is potentially available in the Tg pocket near O-3 and O-10 to allow accommodation of biphenyl groups ( Table 3). The large but flexible Leu-8ADT substitution on O-8 led to a 19-fold reduction of the potency in SPCA1a most likely as a result of steric hindrance. However, overnight incubation of Leu-8ADT led to a lower IC 50 , indicative of an induced fit effect and sufficient space to accommodate a larger O-8 chain (Table 3).
Of all tested Tg side groups, O-2 contributes most to the inhibition of SPCA1a. To determine the SAR of Tg in SPCA1a, we therefore focused on the role of O-2 by testing the functional impact of various substitutions of the O-2 group ( Table 4). The IC 50 remains unaffected when the O-2 octanoyl group was substituted with a benzoyl (HzL20072015) or naphthalenecarboxyl group (HzL18072015). Thus, the difference between HzL20072015, HzL18072015, and the ϪH substitution of nortrilobolide (⌬O-2) indicates that the O-2 chain facilitates a hydrophobic interaction in the SPCA1a-binding pocket. However, a substitution of the O-2 moiety with a phenylbenzoyl group (HzL17072015) led to a significantly higher IC 50 than with smaller substituents (HzL20072015 and HzL18072015), which was independent of the pre-incubation time. This reveals that the local three-dimensional space to accommodate a bulky O-2 group is restricted in SPCA1a, although the O-2 acyl group is larger in one dimension.
Therefore, we explored the replacement of the O-2 chain with several longer but flexible acyl groups (Table 4). We observed little effect on the IC 50 by elongating the O-2 chain with two carbons (JBH04012016), whereas addition of four carbons (JBH09022016) increased the IC 50 3.4-fold (Table 4). Of interest, the inhibition of SPCA1a was completely abolished by introducing a more bulky end group (HzL24042015). The inhibitory effect of this compound is unaffected by the preincubation time.
Thus, strict spatial restrictions around O-2 are likely present, because only thin flexible O-2 chains are allowed, whereas more voluminous groups lower or prohibit the affinity for SPCA1a.

Inhibition of SPCA1a by thapsigargin SPCA1a inhibition by Tg and cholesteric compounds involves the same or overlapping binding pocket
To explore the functional role of amino acid replacements in SPCA1a on Tg binding and confirm that Tg binds to the same pocket in SPCA1a and SERCA1a, we generated and purified three SPCA1a mutants in the Tg-binding pocket in which we substituted one or more of the SPCA1a-specific residues into the corresponding SERCA1a amino acids (L265F, Y272V, and L265F/Y272V/L776F) (Fig. 5A). Phe-256 is a critical residue for Tg central scaffold binding in SERCA1a (45)  Compared with wild-type SPCA1a (WT), the more SERCAlike L265F and L265F/Y272V/L776F mutants are more sensitive to Tg, whereas Y272V behaves similarly as WT (Fig. 5B). This is consistent with the view that as in SERCA1a, Tg binds to a conserved pocket in the transmembrane region formed by M3, M5, and M7. Furthermore, the L265F substitution in SPCA1a significantly enhanced the binding affinity of Tg, possibly by promoting a direct interaction between Phe-265 and the Tg scaffold as is the case for SERCA1a (39). This result is also consistent with the view that binding of the Tg scaffold in SPCA1a may closely resemble the interaction in SERCA1a. Finally, the three SPCA1a mutants also displayed a more potent and stronger inhibition by CHEMS, a cholesterol analogue (Fig.  5C), suggesting that cholesteric compounds may bind at the same or overlapping site as Tg.

Modeling confirms the importance of the Tg scaffold in the binding to SPCA1a
To further explore how Tg might be accommodated in SPCA1a, we generated a homology model of SPCA1a and performed an in silico docking of Tg into the M3, M5, and M7 pocket, while allowing flexibility of 10 amino acid side chains in the pocket. To validate our modeling approach, we first docked Tg into an E2 crystal structure of SERCA1a without Tg (PDB code 3W5C (46)) and compared the results with the 1IWO crystal structure of SERCA1a with bound Tg (Fig. 5D) (41). Among the best 10 docking poses (Chemscore scoring function of GOLD), two major populations were recognized (Fig. 5, E and F, and Table 5). Model 1 (Fig. 5E) corresponds to a Tg  (41)) with important residues shown as blue sticks. The same view of the SPCA1a homology model is shown in C, and the residues that are different between SPCA1a and SERCA1a are labeled. a Tg-epoxide has a ␤-epoxide group formed between C7 and C11; ON is overnight incubation with compound. ⌬O-3, Ϫ8, and Ϫ10 indicate that the acyl group at O-X has been replaced with a proton, and ⌬O-2 means that the octanoyloxy group has been replaced with a proton to give nortrilobolide. position that is highly similar to the Tg position in the 1IWO crystal structure (Fig. 5D), with the O-8 group that is oriented toward the space flanked by M3 and M5 helices. Of the 10 highest ranked scores, six docking poses, including the five best belong to model 1 (Fig. 5E), validating the docking approach. The other four solutions belong to model 2 (Fig. 5F), which contains docking poses displaying a remarkable flipped orientation of Tg into the pocket, i.e. with the O-2 group of Tg facing M3 and M5. Next, we docked Tg onto the SPCA1a homology model, and again two major populations were recognized with opposite orientations of O-2 and O-8 (Fig. 5, G-I). Model 1 (Fig. 5G) displays a similar Tg orientation as in the SERCA1a crystal structure, except that the O-2 chain extends to the outside of the pump. Model 2 (Fig. 5, H and I) (Fig. 5K). The strict positioning of the Tg scaffold in SPCA1a may indicate the presence of spatial restrictions and/or critical interactions between the scaffold and the protein. The scaffold is in close proximity of Leu-265 on M3, which might explain the affinity-enhancing effect of the L265F substitution (Fig. 5B).
Thus, our docking analysis confirms the contribution of the Tg scaffold to the SPCA1a, and it points to two possible but opposite orientations of Tg inside the pocket formed by M3, M5, and M7.

Discussion
In this study, we purified for the first time the human Golgi/ secretory pathway Ca 2ϩ /Mn 2ϩ -ATPase SPCA1a from a yeast overexpression model. The one-step affinity purification protocol generates hSPCA1a with over 95% purity offering opportunities for structure/function analysis and inhibitor screening. We demonstrate that SPCA1a is highly sensitive to the lipid environment and that several SERCA inhibitors, including Tg, also block SPCA1a activity, although at higher concentrations only. Because of the wide use of Tg in cellular studies and prospects for the use of Tg analogues in prostate cancer therapy, we focused on deciphering the SAR of Tg for SPCA1a. We observed differences in the relative contribution of Tg side chains in the inhibition of SERCA1a versus SPCA1a, which might be exploited for the future design of more selective SPCA1a or SERCA1a inhibitors.

Activity of SPCA1a depends on the membrane environment
Ca 2ϩ transporters are tightly regulated by their lipid environment. For instance, the plasma membrane Ca 2ϩ -ATPase con-

values of Tg analogues with modified O-3, O-8, or O-10 side chains on SPCA1a
Columns provide information on compound name, the structure of the side-chain modification, and the IC 50 value on SPCA1a. StudentЈs t test were performed to compare between IC 50 of each analogue and IC 50 Tg , and before/after overnight (ON) incubation. * P Ͻ 0.05, *** P Ͻ 0.001.

Table 4 IC 50 values of Tg analogues with modified O-2 side chain on SPCA1a
Columns provide information on compound name, the structure of the side-chain modification and the IC 50 value on SPCA1a. StudentЈs t test were performed to compare between IC 50 of each analogue and IC 50 Tg , and before/after overnight (ON) incubation. * P Ͻ 0.05, *** P Ͻ 0.001.
tains two lipid-binding sites and can be stimulated by PS, PA, phosphatidylinositol bisphosphate, and cardiolipin (47). SERCA pumps are inhibited by sphingosine (48) and CL (49), whereas SPCA1a associates with CL-rich microdomains of human colon adenocarcinoma cells (30). Also, SPCA2 is inhibited by two endogenous signaling lipids oleamide and linole-amide (34). Here, we demonstrate that SPCA1a is inactive in a lipid-free detergent-solubilized state, but the activity restores upon reconstitution into a PC-rich lipid environment. In proteoliposomes, SPCA1a displays a higher apparent Ca 2ϩ affinity and lower maximal ATPase activity than SERCA1a. The K m value of the purified and reconstituted SPCA1a falls within the

Inhibition of SPCA1a by thapsigargin
reported range for SPCA1 in cell membrane fractions (0.04 -0.26 M (8, 21, 50)). The variation in the reported K m values presumably relates to slight differences in the free Ca 2ϩ concentrations, the choice of the SPCA1 splice variant, and the selected cellular expression system, which may affect the regulation of SPCA1.
Moreover, SPCA1a is sensitive to the SERCA inhibitors Tg, CPA, and BHQ but also to bisphenol and CHEMS. SPCA1a was previously considered as Tg-insensitive, whereas a few studies reported a Tg sensitivity in the range between 0.1 and 30 M (13,24,36). Here, we report that purified SPCA1a in reconstituted proteoliposomes displays an IC 50 for Tg of 7.7 M, which is four times lower than the reported IC 50 value of 28 M in microsomal fractions (24). The variations in the reported Tg sensitivities of SPCA1 may be due to differences in the experimental systems (membrane lipid composition and presence of other proteins or detergent). Thus, overall, the properties of the purified protein correspond well with SPCA1a in a cellular membrane context.
The inclusion of specific lipids in the proteoliposomes impacts on the activity of SPCA1a, with PA and SM leading to stimulation, whereas PE is inhibitory. PA is typically added during the reconstitution of SERCA1a to prevent aggregation (51). Because the stimulatory effect of PA was not observed with other charged lipids, it is also possible that PA serves as a regulatory lipid of SPCA1a. Although PA is a low-abundant lipid in the Golgi/secretory pathway, it can be locally formed by phospholipase D as a signaling lipid that may regulate SPCA1a activity (52,53). Moreover, we showed that SPCA1a preferentially accumulates in lipid rafts (30), which is in line with the stimulatory effect of SM on the ATPase activity of SPCA1a. The cholesteric compound CHEMS only moderately inhibits SPCA1a, while exhibiting a more potent inhibitory effect on SERCA1a, in line with a different impact of CL on SPCA1a and SERCA1a as suggested previously (30). Although our reconstitution conditions were selected to promote the proper orientation of the protein in the membrane (i.e. the large hydrophilic cytosolic domain facing the outside of the vesicles) (51), we cannot fully rule out the possibility that the functional effects of the lipids may relate to an impact on the distribution between inwardfacing and outward-facing enzymes in the proteoliposomes.
Of interest, the potency and efficacy of the CHEMS inhibition is higher in SPCA1a mutants with a more SERCA1a-like Tg pocket. The changes between the Tg-binding site of SERCA1a and SPCA1a might also explain the different sensitivity of SPCA1a and SERCA1a to cholesteric compounds. Although we cannot rule out that these mutations might exert an allosteric effect via an unrelated binding site for cholesteric compounds, it is reasonable to propose that the cholesteric compound binds and overlaps the Tg pocket. Indeed, it was already previously proposed that the SERCA1a Tg pocket might be a CL regulation site (39,54). In the SERCA1a crystal structure 2EAU, the Tg-binding site was occupied by a lipid (55). Moreover, a fluorescence quenching study suggested the existence of hydrophobic non-annular binding sites for CL in SERCA1a (56), whereas a computational docking analysis modeled the steroid compound Alisol B, which targets SERCA pumps and enhances autophagy, in the Tg site (57). However, a more recent molecular dynamics simulation rather indicates that little CL may occupy the Tg-binding site (58), suggesting that the CL effect on SERCA1a may also be indirect via the impact on the membrane properties.

Inhibition of SPCA1a and SERCA1a by Tg is largely determined by the Tg scaffold
Of all tested pharmacological compounds, we observed the most potent inhibition of SPCA1a with Tg, a plant-derived sesquiterpene lactone that is generally accepted as a highly selective SERCA1a inhibitor. Over the years, structures of the Tg-SERCA1a complex (59 -62), supported by extensive functional studies (38 -40, 42, 43, 63-66), rendered a comprehensive pharmacophore model that explains the sub-nanomolar affinity of SERCA1a for Tg. The availability of purified SPCA1a and a library of Tg analogues now offered an opportunity to determine the SAR of Tg on SPCA1a, which might be helpful for the design of more specific SERCA1a or SPCA1a inhibitors.
In accordance with the view that Tg targets the same binding pocket in SPCA1a as in SERCA1a, amino acid substitutions in the predicted M3/M5 pocket of SPCA1a impact the affinity for Tg. According to our in silico docking, the Tg scaffold is aligned on the same plane irrespective of the side chain orientations. It was previously demonstrated that an F256L mutation in SERCA1a led to a 4-fold reduction in the Tg sensitivity (67), whereas our complementary experiment shows that the L265F substitution in SPCA1a significantly enhanced SPCA1a's affinity for Tg. Thus, the Phe-256/Leu-265 substitution at least in part explains the lower Tg affinity for SPCA1a than SERCA1a. The Tg scaffold might possibly undergo -stacking with the L265F residue in the SPCA1a mutant. The importance of the Tg scaffold for inhibition is supported by findings with the 7,11epoxide, which show a major reduction in the binding energy for both SPCA1a (Table 2) and SERCA1a (38,39), pointing to possible similarities in the mechanism of inhibition in SPCA1a and SERCA1a.

Tg side groups differently contribute to the inhibition of SPCA1a versus SERCA1a
We further approached the SAR of Tg for SPCA1a by analyzing the IC 50   which is different for SERCA1a (Table 2) (40,43). Moreover, substitution by slightly bulkier groups at these sites on SPCA1a is well tolerated, indicating that the Tg pocket in SPCA1a is spacious near O-3, O-8, and O-10 and that only few functional contacts exist between the side chains of Tg and residues of the protein. The removal of O-2, the longest side chain of Tg, leads to a higher IC 50 value in SPCA1a, suggesting that besides the relevance of the Tg scaffold, O-2 also contributes to the inhibition of SPCA1a. The analysis of Tg analogues with various O-2 substitutions further indicates that the O-2 chain is accommodated in a narrow hydrophobic cleft, which is restricted in volume. Noteworthy, the change in binding energy of the ⌬O-2 compound is remarkably similar for SPCA1a and SERCA1a possibly pointing to a comparable binding mechanism. By in silico docking analysis, two major docking solutions of Tg O-2 in SPCA1a were predicted. Of interest, the docking model 1 predicts a similar Tg orientation in SPCA1a as observed in the solved SERCA1a-Tg complex, which also might point to a similar inhibitory mechanism in SPCA1a and SERCA1a. This model may explain why a prolonged incubation of Leu-8ADT shows a more potent inhibition in SPCA1a (Table  3), because a similar effect of Leu-8ADT was observed on SERCA1a (44). Alternatively, in line with O-2 being accommodated into a tight hydrophobic cleft, the Tg O-2 chain in the SPCA1a model 2 protrudes into the M3/M5 region. Such a flipped orientation of Tg is also predicted for SERCA1a, although this is not experimentally observed in the solved SERCA1a-Tg complex, which may question the validity of such an orientation.

Toward the design of more specific SPCA1a inhibitors
We demonstrate that the SAR of Tg differs for SPCA1a and SERCA1a, which is mainly attributed to amino acid changes in the Tg pocket that prevent most interactions of the Tg side chains in SPCA1a, but it still allows the Tg scaffold to be accommodated in the pocket. The different SAR of Tg for SERCA1a and SPCA1a provides a strong rationale to consider the Tg scaffold as a lead compound for the development of more specific SPCA1a inhibitors. Solving the crystal structure of SPCA1a in the presence of Tg will greatly facilitate a rational design approach for a Tg-based SPCA1a inhibitor with higher potency and specificity, but so far, crystallization of SPCA1a turns out to be difficult in our hands.
In the absence of detailed structural information, we need to test additional Tg analogues with substitutions of the O-2, O-3, O-8, and O-10 chains to develop Tg analogues with a higher SPCA1a/SERCA1a specificity. Tg analogues with O-2 acyl chains that are slightly shorter would be interesting for establishing the optimal fit of O-2 within the SPCA1a Tg pocket. Moreover, our results indicate that single or combined bulky substitutions at O-3, O-8, and O-10 might lower the potency in SERCA1a, while having little impact on SPCA1a inhibition, favoring a higher SPCA1a/SERCA1a specificity. The more spacious environment near O-3 and O-10 indicates that bulkier side groups on both positions might lead to a tighter fit of Tg within the pocket.
A specific SPCA1a inhibitor might not only be a powerful tool to dissect the cellular and pathological role of SPCA1a in the Ca 2ϩ (dys)homeostasis of various diseases, it might also be of therapeutic interest. Gain of SPCA1a and SPCA2 function is observed in breast cancer, because SPCA1a and/or SPCA2 proteins are up-regulated in various breast cancer sub-types contributing to a pathological Ca 2ϩ dyshomeostasis (17,18). Knockdown of endogenous SPCA2 prevents Orai1 activation, which reduces the transforming phenotype of MCF-7 cells in soft agar and impairs tumor generation upon injection in nude mice (18). Although this mainly depends on the direct interaction of SPCA2 with Orai1 leading to constitutive cytosolic Ca 2ϩ influx (18), SPCA2 and Orai1 are likely mechanistically coupled, suggesting that also a higher Golgi Ca 2ϩ uptake is contributing to the pathological Ca 2ϩ dyshomeostasis (19). In line with this view, knockdown of SPCA1a in the basal-type breast cancer cell line MDA-MB-231 reduces, independent of Orai1, the processing of insulin-like growth factor receptor IGF1R (17), which might counteract its mitogenic and anti-apoptotic actions. Of further interest, the absence of PMR1, the SPCA1a orthologue in lower eukaryotes such as yeast, flies, and Caenorhabditis elegans, prevents the elevation of cytosolic Ca 2ϩ and cell death caused by ␣-synuclein overexpression, a major toxic effector in Parkinson disease (68). Finally, SPCA1a mutations lead to a dominant skin disorder, Hailey-Hailey disease, which is generally considered to be caused by loss-offunction mutations that cause haploinsufficiency (15,16). However, it was recently proposed that dominant disease mutations in ion pumps might introduce a gating defect resulting in ion leakage, which would represent a gain-of-function effect (69). In mice, heterozygous loss of SPCA1a does not trigger a similar skin disease, but it increases the risk of developing squamous cell tumors (70), further highlighting the importance of a balanced SPCA1a activity.
In conclusion, we demonstrate that Tg and cholesteric compounds in SPCA target an overlapping binding site on SPCA1a, and the differences in the SAR of Tg for SPCA1a and SERCA1a might allow the development of more specific inhibitors of SPCA1a.

Human SPCA1a expression in yeast
The recombinant human SPCA1a was cloned in the vector pTV001E with a C-terminal His 8 tag (CATCATCACCAT-CACCATCACCAT) via Gateway cloning. Mutations were introduced via Q5 site-directed mutagenesis (New England Biolabs). pTV001-SPCA1a wild-type (WT) or mutant vectors were transformed into the S. cerevisiae strain BY4741a (his3⌬1; leu2⌬0;met15⌬0;ura3⌬0;MATa) as described previously (71). The yeast clone was inoculated in 10 ml of minimal medium (MM) deprived of Leu for overnight growth at 30°C and gradually expanded into 2 liters MM-Leu culture over 3 days. On day 4, the yeast pellet was transferred into 25 liters of yeast peptone dextrose medium in a 50-liter Cell Bag (GE Healthcare), for 30 h of growth on a shaking system supplied with 30% O 2 , 70% N 2 air mixture. Yeast cells were harvested by centrifugation (5,000 ϫ g, 5 min, 4°C) and washed with milli-Q water, and the pellets were weighted, flash-frozen, and stored at Ϫ20°C.

Inhibition of SPCA1a by thapsigargin Yeast membrane isolation
The yeast membrane was isolated based on the protocol described in Ref. 72. Briefly, the yeast pellet was suspended in lysis buffer (50 mM Tris-HCl, pH 7.5, 1 mM EDTA, 0.6 M sorbitol, 250 M PMSF, 1 mM DTT, SIGMAFAST TM protease inhibitor mixture) and subjected twice to high pressure homogenization at 140,000 kPa (High Pressure Homogenizer, Avestin). The membrane fractions were collected by differential centrifugation at 1,000 and 16,000 ϫ g, and they were suspended in solubilization buffer (10 mM imidazole, pH 7.5, 50 mM NaH 2 PO 4 , pH 7.5, 500 mM NaCl, 20% glycerol, 250 M PMSF, 1 mM DTT, SIGMAFAST TM ). Protein concentration was determined by the Bradford assay. The yeast membrane fraction was aliquoted, flash-frozen in liquid N 2 , and stored at Ϫ80°C.

Purification of SPCA1a and SERCA1a by affinity chromatography
SERCA1a was purified from rabbit fast twitch skeletal muscle as described before (29), whereas a purification protocol for SPCA1a was optimized as detailed below. The His 8 -labeled SPCA1a was purified via nickel-affinity chromatography using nickel-charged agarose beads coupled to Ni-NTA. One column volume (CV) of nickel-Sepharose high performance beads (GE Healthcare) was pre-equilibrated with 6 CV of equilibration buffer (50 mM NaH 2 PO 4 , pH 7.5, 75 mM imidazole, pH 7.5, 500 mM NaCl, 20% glycerol, 5 mM 2-mercaptoethanol (2-ME), 250 M PMSF, SIGMAFAST TM , 20 mg/ml DDM (Isogen Life Science, SOL-grade)). The yeast membrane proteins were solubilized in a triple weight amount of DDM detergent solution (200 mg/ml DDM, 250 M PMSF, SIGMAFAST TM protease inhibitor mixture, 5 mM 2-ME), and non-soluble material was removed by ultracentrifugation (160,000 ϫ g, 45 min, 4°C). After binding of the supernatant to the column for 1 h at 4°C, the liquid was drained off. The beads were washed with 6 CV of three consecutive washing buffers composed of binding buffer with a stepwise reduction in the NaCl and DDM concentrations: from 500 to 50 and 50 mM NaCl combined with 5, 1, and 0.25 mg/ml DDM, respectively. SPCA1a became eluted in the final equilibration buffer with 50 mM NaCl and 0.25 mg/ml DDM. The washing and elution buffers were supplemented with PC (Avanti Polar Lipids) at the ratio 1:0.67 (DDM/PC). Afterward, the buffer component of purified SPCA1a was exchanged into 20 mM imidazole, pH 7.5, 100 mM KCl, 20% glycerol, 0.25 mg/ml DDM, 0.167 mg/ml PC, 5 mM 2-ME, SIG-MAFAST TM via PD Mantrap G-25 (GE Healthcare) column. Aggregates were removed by centrifugation at 20,000 ϫ g for 20 min at 4°C. The supernatant was collected and flash-frozen in liquid N 2 and stored at Ϫ80°C. The purified SPCA1a was analyzed by MALDI-TOF mass spectrometry (4800 Proteomics Analyzer, AB Sciex) in the linear ion mode in the presence of ␣-cyano-4-hydroxycinnamic acid (5 mg/ml).

Reconstitution of the SPCA1a and SERCA1a Ca 2؉ -ATPases
The 550-g lipid mixture dissolved with chloroform in a glass tube was dried in a vacuum desiccator under a N 2 stream and overnight in a vacuum desiccator. Liposome buffer (150 l of 20 mM imidazole, pH 7, 100 mM NaCl, 10 mM MgCl 2 , and 1 mM DTT) was added and incubated at 37°C water bath for 1 h with short vortex periods in between. To this (unless otherwise specified) 30 g of Triton X-100 was added at room temperature, and the mixture was vortexed for 3 min followed by addition of purified SPCA1a (200 g of protein at a concentration 0.667 mg/ml). Detergent was extracted to form proteoliposomes by treatment with excess Bio-Beads SM-2 (Bio-Rad) for 2.5 h.
The proteoliposomes were purified by sucrose gradient centrifugation (20 and 50% sucrose in 20 mM imidazole, pH 7, 0.2% azide, 100 mM NaCl) at 150,000 ϫ g, 75 min, and 4°C from which a 450-l proteoliposome fraction was collected from the turbid layer present between the two sucrose fractions. This sample was flash-frozen in liquid N 2 and stored at Ϫ80°C. The reconstitution of SERCA1a was performed according to Ref. 29. PC, PA, and PE from chicken egg, PS and SM from porcine brain, and phosphatidylinositol from bovine liver were purchased from Avanti Polar Lipids, whereas CL was purchased from Sigma.

Protein electrophoresis and immunoblotting
Purified or reconstituted SPCA1a or SERCA1a was separated on NuPAGE 4 -12% BisTris polyacrylamide gel (200 V, 35 min, MES buffer) (Thermo Fisher Scientific) and visualized with Imperial Protein Stain (Thermo Fisher Scientific). SeeBlue Plus2 pre-stained protein standard (Thermo Fisher Scientific) was used as a molecular weight marker. The concentration of purified protein was determined on a denaturing gel by comparison with 0.125 to 2 g/l bovine serum albumin (BSA) standards (BSA Standard Ampules, Pierce). For immunoblotting, primary antibodies against the His tag (1:2,000 dilution, penta-His, Qiagen) or SPCA1a (1:50,000 dilution, Frodo, homemade) were used. Horseradish peroxidase-linked secondary antibodies (Cell Signaling Technology) were used for enhanced chemiluminescent detection (Bio-Rad ChemiDoc). MagicMark (Thermo Fisher Scientific) was used as the molecular weight marker.

ATPase assay
The Ca 2ϩ -dependent ATPase activity assay was performed on reconstituted SPCA1a or SERCA1a samples according to the Baginsky protocol as described previously (73). The reactions were performed for 15 min (SERCA1a, 150 ng in 50 l) or 30 min SPCA1a (300 ng in 50 l). The results were fitted with a Hill function.
The dose response of the ATPase to various compounds was performed by measuring the ATPase activity at 1 M free Ca 2ϩ concentration with increasing concentrations of compounds to be tested in the reaction mix. Reactions with Tg (200 nM for SERCA1a and 500 M for SPCA1a) samples were used as blanks to correct for irrelevant color development. The data were normalized as the percentage of inhibition compared with that obtained in the absence of inhibitor and fitted with a logistic curve. The CHEMS, CPA, BHQ, BP, and oleamide were purchased from Sigma. Linoleamide was purchased from Enzo Life Sciences, Tg was purchased from Alomone Labs. All compounds were dissolved in dimethyl sulfoxide (DMSO, Sigma).

In silico modeling
The SPCA1a homology model was generated with the homology model module in MOE 2014.09 (74) based on the SERCA1a E2 crystal structure with Tg bound (PDB code 1IWO (41)). Subsequently, the structure of the homology model was energy-minimized using Amber99 (75) force field in the energy minimization module of MOE. Tg was docked in the SPCA1a homology model or SERCA1a crystal structure via GOLD version 5.2 (Genetic Optimization for Ligand Docking (76,77)). Tg was docked in the SERCA1a E2 crystal structure without any inhibitors (PDB code 3W5C (46)) and compared with the crystal structure 1IWO with Tg bound. We specified the center of the Tg-binding site (x, 33.422; y, Ϫ18.503; and z, 82.4683) and selected all atoms in a radius of 11 Å. 10 residues in the Tgbinding site (Glu-255, Phe-256, Gln-259, Leu-260, Val-263, Ile-765, Val-769, Ile-829, Phe-834, and Met-838) were selected to allow for receptor flexibility through side-chain torsion rotations. For each of these residues, flexibility was limited to a library of selected side chain conformers extracted from published crystal structures. Docking conformations for Tg were generated by performing 20 runs using standard Genetic Algorithm parameters. The docking results were scored and sorted using the Chemscore scoring function of GOLD.
The docking poses for Tg or the analogues were generated in the SPCA1a E2 homology model using the GOLD version 5.2 docking software. Here, we specified the center of the binding site (x, Ϫ2.8375; y, Ϫ22.0812; and z, 12.2) and selected all atoms in a radius of 10 Å. 10 residues in the binding site (Leu-264, Leu-265, Leu-269, Tyr-272, Leu-707, Ile-711, Ile-770, Leu-771, Ile-776, and Ile-779) were selected to allow for receptor flexibility through side-chain torsion rotations. To generate docking conformations, 100 runs were performed for each ligand using Genetic Algorithm parameters. The docking results were scored and sorted using the Chemscore scoring function of GOLD.

Synthesis of Tg analogues
The synthesis of the Tg analogues is described in supplemental Materials and methods.

Statistical analysis
Values are provided as mean Ϯ S.E. Hill or logistic fitting was performed via OriginPro 9 software. Statistical significance was calculated by one-way t test and indicated by asterisks: *, p Ͻ 0.05; **, p Ͻ 0.01; and ***, p Ͻ 0.001.