FRET assay for live-cell high-throughput screening of the cardiac SERCA pump yields multiple classes of small-molecule allosteric modulators

We have used FRET-based biosensors in live cells, in a robust high-throughput screening (HTS) platform, to identify small-molecules that alter the structure and activity of the cardiac sarco/endoplasmic reticulum calcium ATPase (SERCA2a). Our primary aim is to discover drug-like small-molecule activators that improve SERCA’s function for the treatment of heart failure. We have previously demonstrated the use of an intramolecular FRET biosensor, based on human SERCA2a, by screening a small validation library using novel microplate readers that can detect the fluorescence lifetime or emission spectrum with high speed, precision, and resolution. Here we report results from a 50,000-compound screen using the same biosensor, with hit compounds functionally evaluated using Ca2+-ATPase and Ca2+-transport assays. We focused on 18 hit compounds, from which we identified eight structurally unique compounds and four compound classes as SERCA modulators, approximately half of which are activators and half are inhibitors. While both activators and inhibitors have therapeutic potential, the activators establish the basis for future testing in heart disease models and lead development, toward pharmaceutical therapy for heart failure.


Introduction
Sarco/endoplasmic reticulum calcium ATPase (SERCA), integrated in the sarcoplasmic reticulum (SR, muscle cells) or endoplasmic reticulum (ER, non-muscle cells) membrane in most mammalian cells, is integral for using Ca 2+ -dependent hydrolysis of ATP to fuel active transport of cytosolic Ca 2+ into the SR or ER.In muscle, the activity of SERCA1a (skeletal isoform) or SERCA2a (cardiac isoform) is essential for relaxation (diastole), restoring SR Ca 2+ following its release via Ca 2+ channels (ryanodine receptors, RyR) for muscle contraction (systole).Decreased SERCA activity and excessive RyR leak results in failure to maintain the high gradient of [Ca 2+ ] between the cytoplasm (mM) and the SR (sub-µM) during diastole (muscle relaxation) and are associated with heart failure (HF) in human and animal models 1 .Decreased SERCA activity has been attributed to multiple factors, including reduced SERCA gene expression, increased post-translational modi cations, and altered interaction with regulatory proteins 1 .Overall, decreased SERCA activity and increased Ca 2+ -leak lead to a pathophysiological state of the cardiac myocyte 2 (HF, cardiac hypertrophy, diabetic hypertrophy), skeletal myocyte (Brody's disease and myotonic dystrophy) and non-muscle cells (Darier's disease, diabetes, Alzheimer's disease) 3 .Altered SERCA interaction with regulatory proteins (regulins), such as phospholamban (PLB), have been linked to ventricular dysfunction and maladaptive remodeling in failing hearts 4 .Of the seven regulins discovered 5 , the dwarf open reading frame (DWORF) regulatory peptide is the only one known to activate musclespeci c SERCA activity, both by direct activation of SERCA 6,7 and by competing with PLB binding 8,9 , and to prevent HF in a mouse model of dilated cardiomyopathy 10 .Current therapeutic measures include betablockers, angiotensin-converting enzyme (ACE) inhibitors, angiotensin-receptor blockers (ARB), and mineralocorticoid receptor antagonists.However, these do not directly target proteins responsible for dysfunctional Ca 2+ cycling.Discovery of small molecules (potential drugs) that target speci c proteins is needed to exert improved control measures for positive therapeutic outcomes.
In the present study, we seek primarily SERCA2a activators to alleviate heart failure 11,12 .However, SERCA uncouplers are also of potential interest in other indications such as in nonshivering thermogenesis, enhancing metabolism and thus reducing obesity 13,14 .Targeted SERCA uncouplers or inhibitors may be useful for treatment of cancer or malaria 15 .
SERCA2a is a large transmembrane protein, with the phosphorylation (P) and nucleotide-binding (N) domains forming the catalytic site, coupled by the actuator (A) domain (Fig. 1A).Large (5-10 nm)   relative movements of these domains are coupled to Ca 2+ transport, as detected in living cells by an intramolecular FRET biosensor (Fig. 1A) 16 .The interaction of small molecules with SERCA can induce measurable structural changes, detectable by this biosensor, that often correlate with function, making this biosensor a powerful tool for HTS discovery of SERCA-binding compounds 16,17 .
In several previous early-stage drug discovery campaigns, we have focused on the SERCA regulator phospholamban (PLB) 18,19 , but we have also tested and validated FRET biosensor constructs of SERCA alone 16,17,20 , to detect binding of drug-like small-molecule compounds directly to SERCA.For this, we engineered a "two-color" SERCA (2CS, Fig. 1A) construct with fused eGFP and tagRFP uorescent protein tags to the cytoplasmic N-and A-domains of SERCA, which detect relative motions of these domains during the enzymatic cycle responsible for Ca 2+ -transport 17,20,21 .Using the intramolecular FRET measurement of 2CS, stably expressed in HEK293 cells, we previously validated this biosensor using the NIH Clinical Collection (NCC, 727 compounds) 17 and the LOPAC library (1280 compounds) 20 .Along with the uorescent biosensors, high-throughput is enabled by a FLT-PR ( uorescence lifetime plate reader) that scans a 1536-well plate with unprecedented precision and speed, determining FRET with 0.3% CV (30 times better than conventional intensity detection) in 2.5 min 18,22,23 , enabling a high-precision screen of 50,000-compounds in two days.This instrument is equipped with simultaneous FLT detection at two emission wavelengths (two-channel lifetime detection) 21 , a function used to lter out uorescent compounds that would be falsely identi ed as hit compounds.We have demonstrated the additional high throughput acquisition of uorescence emission spectra using a spectral unmixing plate reader (SUPR), which helps detect (and rule out) interfering compounds, based on changes in the line shape in the donoronly region of the spectrum 21 .As in our previous projects involving other drug targets, the next logical step is to screen a 50,000-compound DIVERSet library, a diverse collection of drug-like small molecules that has yielded effective hit compounds in previous drug discovery projects [22][23][24] .
Here we apply our HTS platform using FRET lifetime measurements of two-color human cardiac SERCA (2CS) in living cells (Fig. 1B) to identify hit compounds.To validate selected hit compounds, we then acquire concentration response curves (CRCs) using both FRET and functional assays (Ca 2+ -ATPase activity and Ca 2+ -transport).We hypothesize that the combination of using improved uorescence technology and screening a larger compound library (50K DIVERSet) will result in a larger and more diverse collection of hit compounds that more effectively regulate cardiac SERCA function, thus increasing the potential for discovering lead compounds for new heart failure therapeutics.

FLT HTS of 50K DIVERSet library
2CS expressed in live HEK-293 cells was incubated with 5 nL of compound at a nal [compound] of 10µM (from a DIVERSet library of 50,000 compounds) or DMSO (as a control), preloaded on 40 assay plates for 20 min prior to being read on the FLT-PR.FLT measurements were observed to be normally distributed, with a coe cient of variation (CV) of 0.4% across all 40 plates (Fig. 2A).Plate-by-plate CV varied by < 1% (Fig. 2A).
Compounds that signi cantly altered the structure of 2CS were determined by computing the change in lifetime (Δτ) for each compound compared to DMSO control sample (2CS plus DMSO), and the magnitude of this change was compared to the normal statistical uctuation of the biosensor by computing the robust z-score (Methods).The distinct FLT changes induced by the potential hit compounds (Fig. 2B, red) are illustrated by the normal distributions of compounds not affecting SERCA (Fig. 2C, blue) and the control (Fig. 2C, dark blue).A hit threshold was set at a robust z-score of ± 3, resulting in 2960 FLT hit compounds (Fig. 1B, step 1).Interference from uorescent compounds was removed (Fig. 1B, step 2) using both the similarity index (SI, detected by SUPR) 17 and two-channel lifetime detection 21 .We also eliminated compounds that affected the lifetime detected from a donor-only (1CS) sample; 295 compounds remained.
More FLT decreasers than increasers were found to fail these tests, in alignment with previous studies 18, 20,21,25 .FLT increasers are more advantageous for two additional reasons: (a) They offer greater reproducibility between repeats of a screen 18,20,21 .(b) Most previously identi ed SERCA modulators have been shown to be FLT increasers 16,17,20 .Therefore, we prioritized the 158 FLT increasers (termed "hit compounds" (Fig. 1B, step 3) for follow-up retesting and CRC evaluation.
FLT retests of select hits compounds with 2CS and null biosensor 158 hit compounds were retested using 2CS (Step 4 in Fig. 1B; see Fig. 3A and C) and a null biosensor construct (Step 4 in Fig. 1B; see Fig. 3B and D), which consisted of GFP and RFP connected by a 32residue unstructured exible linker peptide (G32R) 20 .The null biosensor was used to rule out compounds that alter FLT by directly binding to the uorescent proteins.A plot of the change in lifetime (Δτ) vs. the ΔR/G ratio (Fig. 3C and D) shows that the 2CS hits had little to no effects on the null biosensor.
Hit compounds that produced > 75 ps Δτ (76 compounds, Fig. 1B, step 4) (Fig. 3A) were targeted for further functional testing.We focused on 18 of these for the CRC testing, after the FLT data were subjected to the rst four steps of the screening funnel (Fig. 1B) and compound repurchasing availability was determined.None of these compounds were in the PAINS (Pan-Assay INterference compoundS) category 26 , nor were they redox agents or metal chelators.

Validation of hit compounds using FLT CRC
To further evaluate the 18 hit compounds, we determined the FLT response to compound concentrations ranging 0.78-100.µM(Step 5 in Fig. 1B).All 18 hit compounds (Table 1) decreased FRET (increased FLT) of the 2CS biosensor, suggesting that the compounds induced a structural change (Fig. 1, top) in the cytosolic headpiece region of SERCA2a.Compounds 1 and 4 showed a signi cant decrease in FRET at the lowest concentration, but no further effect at higher concentrations.The remaining 16 compounds decreased FRET with detectable EC 50 (Fig. 4-7B, Table 1).

Classi cation of compounds by physicochemical characteristics
The 18 hit compounds were subjected to cheminformatic analysis, to determine whether any of the compounds shared a common chemical scaffold.Compounds with a Tanimoto coe cient and maximum common substructure (MCS) 31 scores above 0.4 were binned as clusters, while those with scores below 0.4 were classi ed as singletons.The analysis yielded diverse scaffolds 31,32 of the hit compounds (Supplementary Fig. S1 and Table S1). 1) were found, and the remaining eight were unique compounds (singletons) (E-L in Table 1 and Supplementary Fig. S1 and Table S1).The three compounds in cluster A (Compounds 1, 2, and 3 in Table 1) have a common 5-(aryloxymethyl)oxazole-3carboxamide) 33 , while those in cluster B (Compounds 4 and 5) share a N-heteroaryl-N-alkylpiperazine.

Four clusters of compounds with multiple examples (A-D in Table
Cluster C (Compounds 9 and 10) is de ned by an amide linkage and Cluster D (Compounds 11, 12, and 13) by a piperidine scaffold.Clusters E-L (Compounds 6, 7, 8, 14, 15, 16, 17, and 18) contain a single compound (singleton) with no common scaffold with any other hit compound in this study.All of the hit compounds have physicochemical properties 34 , conducive of favorable drug disposition in vivo, including a low molecular weight (< 500), low cLogP (calculated partition coe cient for lipophilicity) values (< 5), low non-H rotatable bonds that describe the molecular exibility (< 10), low degree of possible hydrogen bond formation (total number of hydrogen bond acceptors and donors should be less than 8), and low total polar surface area (tPSA < 140 Å) (Supplementary Table S1).
Next we describe in more detail the nine activators (Fig. 4-6) that are grouped into two subcategories, and the nine inhibitors (Fig. 7) that are grouped into three classes and four subcategories.

SERCA Activators
In the rst category of activators, Compounds 2, 4, 7, 8, and 9 activated both Ca 2+ -ATPase and Ca 2+transport.Compound 7 (Fig. 4A) (singleton F) decreased FRET of 2CS in live cells so that EC 50 = 0.3 µM, indicating stabilization of the open conformation of SERCA, and accelerated Ca 2+ -ATPase to induce ΔV MAX = 14% and ΔV MID = 7% (Fig. 4C and Table 1).Compound 7 induced the highest increase in Ca 2+transport of all the compounds at both V MAX (24%) and V MID (19%), (Fig. 4D), which was greater than that of Ca 2+ -ATPase activity (Fig. 4D).The CR increased to 0.74 compared to control (0.66, Table 1).Saturation of CRC was not reached at the highest [compound] measured, so the functional EC 50 was not determined, therefore we determined C 10 , the compound concentration that increases function by 10%.At V MAX and V MID , C 10 was 25 µM for Ca 2+ -ATPase, 14 µM and 21 µM for Ca 2+ -transport (Table 1).This compound will be placed at high priority for future optimization by medicinal chemistry and testing in animal models.
Compound 8 (singleton G) (Fig. 5A) decreased FRET (EC 50 = 4.9 µM, Table 1) and increased Ca 2+ -ATPase at both V MAX (49%, the largest increase observed in the screen) and V MID (31%) (Table 1 and Fig. 5C).For Ca 2+ -transport, effects (Δ values) were lower (10% for V MAX and 7% for V MID , Fig. 5D), decreasing CR to 0.4 (Table 1).EC 50 values for Ca 2+ -ATPase were not signi cantly different at V MAX and V MID , and were ~ 2x greater than the values observed by FRET.C 10 was 4.3 µM (V MAX ) and 7.8 µM (V MID ), indicating signi cant ATPase activation at low dosage.C 10 for Ca-transport was 25 µM at V MAX , and was not determined at V MID .
In the second category of activators, Compounds 1 and 3 (cluster A), 5 (cluster B), and 6 (singleton E) activated Ca 2+ -ATPase at both V MAX and V MID , but induced divergent effects on Ca 2+ -transport.

SERCA Inhibitors
Compounds 10-18 all decreased Ca 2+ -ATPase and Ca 2+ -transport activities at both V MAX and V MID .
Compared with FRET EC 50 of the activators (0.3-7 µM), most of the inhibitors (Compounds 10-18) showed weaker a nity, with FRET EC 50 values in the range of 5-32 µM, but the maximum functional effects (e cacies) of the inhibitors tended to be greater (Table 1).
Compounds 11, 12, and 13 (cluster D) showed similar inhibition of both SERCA2a functions: Ca 2+ -ATPase was inhibited by 61%, 93%, and 81% at V MAX; by 59%, 90%, and 72% at V MID .Ca 2+ -transport was completely inhibited.Compared with Ca 2+ -ATPase, Ca 2+ -transport inhibition at V MID required slightly higher compound concentration as shown by the shift to the right of the red curve (Fig. 7D).
Compounds 14 (singleton H) and 16 (singleton J) induced mild inhibition of Ca 2+ -ATPase, but a considerably larger effect of moderate-to-strong inhibition of the Ca 2+ -transport.Ca 2+ -ATPase decreased by 13% and 8% at V MAX , 26% and 16% at V MID .Ca 2+ -transport was inhibited by 83% and 66% at V MAX , 62% and 48% at V MID .C 10 values ranged from 0.5 to 7µM, for Ca 2+ -ATPase at V MAX and V MID .

Discussion
We identi ed new compounds based on an increase in donor FLT, within a human cardiac 2CS biosensor expressed in live mammalian cells, indicating a decrease in FRET, implying that the actuator (A) and nucleotide-binding domains (N) of SERCAC2a moved further apart.We con rmed that these compounds affect SERCA activity using Ca 2+ -ATPase and Ca 2+ -transport assays with SERCA2a in native SR preparations, where we further categorized them as activators or inhibitors.We identi ed two subcategories of activators, whereby the compound either (1) activates both Ca 2+ -ATPase and Ca 2+transport activities (Compounds 2, 4, 7, 8, and 9) (Figs. 4 and 5) and ( 2) activates Ca 2+ -ATPase with divergent effects on Ca-transport (Compounds 1, 3, 5, and 6) (Fig. 6).We identi ed four subcategories of inhibitors based on the extent of Ca 2+ ATPase and Ca 2+ -transport decrease for cardiac SERCA2a (Table 1 and Fig. 7).
In general, the FRET EC 50 values were smaller (indicating higher potencies) for activators (Compounds 1-9; 0.3-7µM) than for inhibitors (Compounds 10-18; (3-32µM) (Table 1).The potencies observed by FRET and function are not precisely correlated, probably because the assays were performed on different types of samples (live cells vs puri ed proteins), measuring different properties (structure vs function).
Functional CRC assays showed that inhibitors tended to induce larger changes (indicating higher e cacies) than activators, in both Ca 2+ -ATPase activity and Ca 2+ -uptake.Also, most inhibitors induce a larger change in Ca 2+ -uptake than in Ca 2+ -ATPase, decreasing the coupling ratio.In general, the C 10 and EC 50 values were smaller, indicating greater potency, for inhibitors than for activators.
Effects of most activators were to reduce the CR, as they induced larger changes in the Ca 2+ -ATPase than in corresponding the Ca 2+ -uptake assay.The most notable exception is Compound 7, which activates Ca 2+ -transport even more than it activates Ca 2+ -ATPase, increasing CR.This compound will be a high priority as a lead compound for future efforts in medicinal chemistry and assays of physiological function.Ten compounds were binned into four clusters (A-D), while eight compounds were classi ed as singleton (E-L) (Table 1).Many compounds showed similar functional traits, suggesting that there are ligand-sensing sites in SERCA2a that recognize a range of scaffolds, or that the ligand-binding sites are close to each other, providing potentially powerful tools in the design of future compounds 36,37,38 .Compounds 2 (cluster A),  (10), and H-L (14-18) of inhibitors also induced similar functional effects.
There was little or no overlap in the hit compounds identi ed in our previous FRET HTS screen of the same (DIVERSet) library with another biosensor for tumor necrosis factor receptor 1 (TNFR1 22 .There was 81% overlap in the uorescent compounds detected in these two HTS screens, indicating that our FRET HTS screening methodologies independently and successfully removed the uorescently interfering compounds 21,24 .In another HTS study of the DIVERSet library, using a SERCA functional (Ca 2+ -ATPase) assay in the primary screen, we discovered several activators, several of which showed isoform speci city for either SERCA1a or SERCA2a 28 .However, there was negligible overlap between hit compounds identi ed in that ATPase HTS study and in the current study that used FRET in the primary screen.This observation highlights the value of complementary HTS assays for the same target.
SERCA activators are needed when cardiac relaxation is impaired, as in early-onset diastolic dysfunction that precedes systolic impairment in HF 1 , diabetic cardiomyopathy 39 , Alzheimer's disease 40 , or Duchenne muscular dystrophy (DMD) 41 .The stimulation of SERCA2a activity, as a novel therapeutic measure to relieve cardiac dysfunction in heart failure without arrhythmogenic effects, is a promising strategy to be used in combination with other rst-line therapeutic agents such as β-blockers and ACE inhibitors 42 .Until recently, only a few compounds were known to stimulate SERCA: CDN1163 (stimulates Ca 2+ transport) 43,19 , CP-154526 (increases the apparent Ca 2+ a nity of SERC2a) 44 , Ro 41-0960 (increases SERCA maximal activity in high Ca 2+ ) 44 , and istaroxime (stimulates SERCA activity) 45 .However, our recent HTS using Ca 2+ -ATPase activity as the target HTS assay identi ed ~ 19 new activators of SERCA 28 , and we identi ed nine activator compounds in the present study.A SERCA activator from our previous work shows promise as a therapeutic target for Alzheimer's disease, as it rescued memory function in a mouse model for the disease 40 as well as for DMD as it was shown to ameliorate dystrophic phenotypes in dystrophin-de cient mdx mice 41 .Of all these SERCA2a activators only istaroxime, a known Na + /K + transporting ATPase inhibitor as well as an inotropic/lusitropic agent acting to enhance SERCA2a activity, has been in phase IIb clinical trials for treatment of heart failure 45,46 .However, because of its unsuitability for human usage (poor gastrointestinal absorption, high clearance rate, and extensive metabolic transformation) 46 , istaroxime derivatives were designed from QSAR studies and a new promising class of SERCA2a activators has been identi ed 47,48,49 .
Compounds 1, 3, 5, and 6 induced small effects on the V MAX of Ca 2+ -ATPase (~ 10-25% increase) and induced a negative effect on the V MAX of Ca 2+ -transport (Fig. 6C and D), thus decreasing the CR, which is likely to increase heat output 13,52,54 .These effects are similar to that of SLN on SERCA1a (skeletal muscle), where SLN induces no observable effect on the V MAX of Ca-ATPase but reduces the V MAX of Ca 2+ -transport (SERCA1a uncoupling), thus reducing CR 50 .This results in futile cycling of SERCA1a and higher usage of ATP, resulting in increased non-shivering thermogenesis (NST) 50 .Another contributor to NST is Ca 2+ leak from SR to sarcoplasm through RyR channels, stimulating SERCA to re-sequester Ca 2+ into the SR, thus using more ATP and generating heat 51 .This increase in energy expenditure in muscle has been suggested as a potential therapeutic strategy for weight loss 50,52 .Thus, the SERCA uncouplers in this study may serve as the basis for further drug development targeting weight loss.
Over the past several decades (~ 60 years), research on the potential for small-molecule SERCA inhibitors as oncology therapeutics has yielded hundreds of SERCA inhibitors with varying potencies and e cacies 15 .Similarly, our discovery of new SERCA inhibitors with a range of potencies and e cacies is likely to be advantageous for treatment of non-cardiac applications 15,53 .
In the present study, the 2CS biosensor has been used to identify novel small-molecule effectors of SERCA that have diverse chemical scaffolds for binding to SERCA, resulting in an array of hit compounds that are activators and inhibitors.Most importantly, we discovered a potential lead compound (Compound 7) that activates Ca 2+ -uptake more than the Ca 2+ -ATPase, increasing the CR, so this will be a high priority for future efforts in medicinal chemistry and assays of physiological function.The enabling technology included three novel plate-readers, the FLT instrument used in the primary screen, and two spectral instruments that were used to remove interference of uorescent compounds, allowing us to focus on valid SERCA activators and inhibitors.In the future, hits from the present study will be evaluated in more functional detail, including studies on multiple SERCA isoforms and on intact muscles and animals.Medicinal chemistry will be applied to elucidate structure-activity relationships, with the goal of designing analogs with greater potency and speci city 22,54 .We will also expand our approach to much larger compound libraries, since our primary screening technology is capable of evaluating thousands of compounds per hour.

Molecular biology
A two-color intramolecular human SERCA2a (2CS) biosensor, based on human cardiac SERCA2a fused to green uorescent protein (eGFP) and red uorescent protein (tagRFP) was developed to detect structural changes that are related to the functional changes of SERCA 20 .Brie y, tagRFP was genetically fused to the N-terminus of SERCA and eGFP was inserted as an intrasequence tag before residue 509 in the nucleotide-binding domain (N-domain) 55,56 .A donor-only (1CS) biosensor was created in a similar manner as the 2CS biosensor with the exception of the construct containing only eGFP.The uorescent proteins fused to SERCA in 2CS and 1CS do not signi cantly affect SERCA activity, in membranes puri ed from HEK cells 18,20 .A null biosensor construct consisting of eGFP and tagRFP connected by a 32-residue unstructured exible linker peptide (G32R) was created as described previously 18,20 .All constructs were cloned into expression vectors containing the genes for antibiotic resistance to G418, puromycin, or blasticidin.

Compound handling
A DIVERSet 50,000 compound library was purchased from ChemBridge Corporation (San Diego, CA) at a 10 mM stock concentration for each compound.All compounds met the high quality standard of 100% identi cation by NMR and/or LC-MS and have a minimum purity of 85% and their identity veri ed using LC-MS/ELSD as con rmed by the ChemBridge Corporation.For the FRET HTS initial screens, the compound library was reformatted into 384 well Echo compatible plates using the Biomek FX (Beckman Coulter, Miami, FL) and 5 nL of either compound (columns 3-22 and 27-46) or DMSO (columns 1-2, 23-26, and 47-48) was dispensed into 1536 well black polystyrene assay plates (Greiner, Kremsmünste, Austria) using an Echo 550 liquid dispenser (Beckman Coulter) to yield a nal assay screening concentration of 10µM.The low auto uorescence and low interwell cross-talk of these plates made them advantageous for FRET measurements.Plates were heat sealed with a PlateLoc Thermal Microplate Sealer (Agilent, Santa Clara, CA) and stored at -20°C prior to use.The same methods were applied for subsequent FRET retesting of the reproducible hit compounds identi ed in the pilot screen, except that the [compound] was tested at 10µM and 30 µM in triplicate.
FRET CRC assay plates (0.78-100 µM compound range) with at least ten different compound concentrations were made by adding the appropriate volume of compound or DMSO into black 384 well plates (Greiner Bio-One) using a Mosquito HV (SPTLabTech, United Kingdom).Subsequent ATPase and Ca 2+ -transport CRC assay plates (0-50 µM compound range) with repurchased compounds were made in a similar manner using with the Echo 550 (Beckman Coulter) using either 384 well transparent plates (Greiner Bio-One) or black-walled plates with transparent bottoms (Greiner Bio-One), respectively.

HTS sample preparation and FRET measurements
On each day of screening, cells were harvested, washed three times with PBS, and centrifuged at 300g for 5min.Cells were ltered using a 70µm cell strainer and diluted to 1-2 x 10 6 cells/mL.Cell concentration and viability were assessed using the Cell countess (Invitrogen) and trypan blue assay.During assays, cells were constantly and gently stirred using a magnetic stir bar at room temperature, keeping the cells in suspension and evenly distributed to avoid clumping.Cells were dispensed at 5µL or 50µL per well into assay plates (dispensed into 40 assay plates, each containing 1536 wells) pre-plated with either compound or DMSO using a Multdrop Combi liquid dispenser (Thermo Fisher Scienti c, Pittsburg, PA) and sealed until needed.Because the kinetics of membrane permeability, diffusion, and/or binding of the compound to live cells may be compound-dependent, we tested two incubation times, 20 min and 120 min, for the FLT CRC.FRET EC 50 values determined from both incubations were similar, but the 120 min incubation yielded a more reproducible and sigmoidal curve.Plates containing eight-point concentration curves of three tool compounds (known SERCA effectors) were also included on the plates as positive controls for the HTS FRET assay.The FRET HTS screen was performed over two days with a custom HTS uorescence lifetime plate reader (FLT-PR) and spectral plate reader (SUPR) provided by Photonic Pharma LLC (Minneapolis, MN) 18 .
The same methods were applied for subsequent FRET retesting of the reproducible hit compounds identi ed in the pilot screen, except that the compound tested at 10µM and 30 µM [compound].158 hit compounds were picked from the library master plates and reloaded onto new assay plates for retesting with 2CS and a null biosensor.Then 18 hit compounds were selected and purchased from ChemBridge Corporation to determine CRC from FRET, ATPase, and Ca-transport assays using at least ten different concentrations by repeatedly scanning the 1536-well plates.

FRET HTS instrumentation and data analysis
A detailed description of the high-throughput uorescence lifetime plate reader (FLT-PR) and spectral unmixing plate reader (SUPR), manufactured by Fluorescence Innovations Inc and provided by Photonic Pharma, LLC was described previously 18,21 .Brie y, for lifetime measurement with the FLT-PR, the observed donor-uorescence waveform, F (t) was t by a convolution of the measured instrument response function (IRF) and a single-exponential decay to obtain the lifetime (τ) of the donor uorophore 57 in the absence (τ D ) and presence (τ DA ) of the acceptor as described in Eq. ( 1): 1 FRET e ciency (E) was determined as the fractional decrease of donor FLT in the absence and in the presence of acceptor as in Eq. ( 2): 2 E was determined in the presence and absence of compound and normalized relative to E of the DMSO control.For spectral detection with the SUPR, the observed uorescence emission spectrum F(λ) was t by least-squares minimization of a linear combination of component spectra for donor (D), acceptor (A), cellular auto uorescence (C) and water Raman (W) as described previously 17 .

HTS data analysis
FLT-PR data was used as the primary metric for agging potential hit compounds.After tting waveforms with a single exponential decay to quantify donor lifetime, the change in uorescence lifetime (Δ FLT) was computed by performing a moving average subtraction in the order the plate was scanned with a window size of half a plate row (24 columns).The reasons for this are twofold: 1) plate gradients are often observed due to heating of the digitizer during acquisition and 2) performing ΔFLT computations with DMSO controls alone can sometimes result in artifacts as a half of the DMSO wells are on the edge of plates, which occasionally exhibit artifacts due to processes needed for the preparation of the drug library being tested.As most compounds are likely to be non-hits, and therefore DMSO like, computation of a moving average is an effective alternative to solving both gradient issues and edge-effect distortion of the primary metric for hit selection, Δτ.As hit compounds from FLT-PR were to be further triaged with a secondary technique (using the spectral plate reader), a generous cutoff was set at a robust z-score of 3 on a plate-by-plate basis.The robust z-score was used, where the median (M) and median absolute deviation (MAD) are used in place of the mean and standard deviation (Eq.3), to best capture the most hits, as the standard z-score is more subject to strong outliers (compounds that fall outside of the de ned upper and lower limits) 18 .
To remove "false positive" uorescent compounds, the similarity index (SI 17 was computed by comparing a region (500-540nm) of the donor only spectrum (I (a) ) for each well to that of the plate-wide average DMSO spectrum (I (b) ) in the same wavelength band as described in Eq. 4 21 .Compounds that exceeded an SI robust z-score of 5 (corresponding to an SI of 2x10 − 4 ) were deemed likely uorescent compounds and removed from consideration.
4 Spectral (SUPR) data was processed similarly to FLT-PR data, with the ΔR/G ratio being computed by applying the same moving average lter on the initial measurement of the ratio of the acceptor amplitude over the donor amplitude as found by tting basis sets of the component spectra through least squares minimization.The hit threshold was also set using a robust z-score of 3.While the FLT-PR data and SUPR data showed a robust correlation, the FLT-PR data exhibited some strong outliers, presumably due to compounds directly modifying the donor lifetime.To eliminate these likely interfering compounds, correlation was enforced by eliminating compounds that exceed a robust z-score of 3 from the median value of the ratio of ΔFLT over the ΔR/G ratio metric.Additional interfering compounds were removed using two-channel lifetime detection 18 .

Cardiac SR preparation
Cardiac SR vesicles were isolated from fresh porcine left ventricular tissue using differential centrifugation of the homogenized tissue as previously described 58 .The SR vesicles were ash-frozen and stored at -80ºC until needed.
Enzymatic SERCA activity assays of FRET hit compounds Functional assays were performed using porcine cardiac SR vesicles 16 .An enzyme-coupled, NADH-linked ATPase assay was used to measure SERCA ATPase activity in 384-well microplates.Each well contained 50 mM MOPS (pH 7.0), 100 mM KCl, 1 mM EGTA, 0.2 mM NADH, 1 mM phosphoenol pyruvate, 10 IU/mL of pyruvate kinase, 10 IU/mL of lactate dehydrogenase, 7 µM of the calcium ionophore A23187 (Sigma), and CaCl 2 was added to set free [Ca 2+ ] to three different concentrations 59 .The Ca 2+ -ATPase were measured at V MAX (saturating, pCa 5.4), V MID (subsaturating, midpoint, pCa 6. were incubated for 20 min at room temperature before measurement of functional assays with each of the 18 hit compounds, because a shorter incubation time than the FRET live-cell assays achieved optimal responses.The assay was started upon the addition of MgATP, at a nal concentration of 5 mM (total volume to 80 µL), and absorbance was read in a SpectraMax Plus microplate spectrophotometer from Molecular Devices (Sunnyvale, CA) at 340nm.
Ca 2+ -transport assays of FRET hit compounds Ca 2+ -transport assays were performed with similar porcine SR samples as in the Ca 2+ -ATPase assays described above.The compound effect on the Ca 2+ -transport activity of SERCA2a was determined using an oxalate-supported assay in which the change in uorescence in a Ca-sensitive dye, Fluor-4, was determined as previously described 18 .A buffered solution containing 50 mM MOPS (pH 7.0), 100 mM KCl, 30 mg/mL sucrose, 1 mM EGTA, 10 mM potassium oxalate, 2 µM Fluo-4, 30 µg/mL porcine cardiac SR vesicles, CaCl 2 calculated to reach the free [Ca 2+ ] (pCa 8.0, 6.2, and 5.4), and compound (0.048 to 50µM) was dispensed into 384-well black walled, transparent bottomed plates (Greiner Bio-One) containing the tested small molecule and incubated at 22℃ for 20 minutes while covered and protected from light.To start the reaction, MgATP was added to a nal concentration of 5 mM, and the decrease in 485-nm excited uorescence of Fluo-4 was monitored at 520 nm for 15 min using a FLIPR Tetra (Molecular Devices, San Jose, CA).

Data analysis of FRET CRC assays of hit
FRET e ciency (E) (Eq.2) was determined as the fractional decrease of donor (1CS) lifetimes (τ D ) in the presence of acceptor (2CS) uorophore (τ DA ) due to FRET as described in Eq. 1 and normalized to DMSO controls.
Data analysis of Ca 2+ -ATPase and Ca 2+ -transport CRC assays SERCA2a ATPase (or Ca-transport) activity at pCa 8.0 was subtracted from pCa 5.4 and pCa 6.2 values.
The % effect ATPase (or Ca-transport transport) activity was normalized to the DMSO only (or 2CS in the absence of compound), and then were plotted against [Ca], and the curves were tted using the Hill function, where V is the initial ATPase rate (or uorescence rate), V MAX is the ATPase (or Ca 2+ -transport) at saturating [Ca 2+ ], and EC 50 or pK Ca , or V MID is the apparent Ca 2+ dissociation constant as described previously ATPase (or Ca 2+ -transport) at (Midpoint Ca 2+ ) 44 .These parameters and the [Ca 2+ ] at 10% (C 10 ) above or below baseline pCa (8.0) are reported in Table 1.

Cheminformatic analysis of hit compounds
An online interactive program was used to perform cheminformatics analysis 60 to determine whether the hit compounds had structural similarity by identifying common chemical scaffolds (core structural feature) using binning, multidimensional scaling (MDS), and compound similarity methods where the Tanimoto coe cient 31 and maximum common substructure 31 values were used to determine clustering (Supplementary Table S1).The physicochemical properties (for e.g.Lipinski Rule of 5) and bioactivity properties of the compounds were also used in the clustering analysis 34 .A cluster contained two or more compounds with similarity score > 0.4, while a unique compound with a similarity score < 0.4 was referred to as a singleton.

Statistical analysis
Analysis of two-group comparisons was done by a two-tailed unpaired Student's t-test (*p < 0.05) using the data analysis program Microsoft Excel (Santa Rosa, CA).Data are presented as mean ± SEM calculated from a minimum of three separate experiments (n = 3).

Figure 2 The human cardiac 2 -
Figure 2

Figure 4 A
Figure 4

Figure 5 A
Figure 5

Figure 6 A
Figure 6

Table 1
is available in Supplementary Files section.