Allosteric regulators selectively prevent Ca2+-feedback of CaV and NaV channels

Calmodulin (CaM) serves as a pervasive regulatory subunit of CaV1, CaV2, and NaV1 channels, exploiting a functionally conserved carboxy-tail element to afford dynamic Ca2+-feedback of cellular excitability in neurons and cardiomyocytes. Yet this modularity counters functional adaptability, as global changes in ambient CaM indiscriminately alter its targets. Here, we demonstrate that two structurally unrelated proteins, SH3 and cysteine-rich domain (stac) and fibroblast growth factor homologous factors (fhf) selectively diminish Ca2+/CaM-regulation of CaV1 and NaV1 families, respectively. The two proteins operate on allosteric sites within upstream portions of respective channel carboxy-tails, distinct from the CaM-binding interface. Generalizing this mechanism, insertion of a short RxxK binding motif into CaV1.3 carboxy-tail confers synthetic switching of CaM regulation by Mona SH3 domain. Overall, our findings identify a general class of auxiliary proteins that modify Ca2+/CaM signaling to individual targets allowing spatial and temporal orchestration of feedback, and outline strategies for engineering Ca2+/CaM signaling to individual targets.

By leveraging synergistic insights from Ca V and Na V channels, we demonstrate that stac selectively diminishes Ca 2+ -regulation of Ca V 1. In-depth analysis shows that stac binds to a distinct channel interface from CaM and uses an allosteric mechanism to lock Ca V 1 into a high open probability (P O ) gating mode. We further localize a minimal motif that recapitulates stac modulation of Ca V 1 gating. Paralleling stac-effect on Ca V 1, fhf reduces CDI of Na V 1 with no effect on Ca V 1. In all, our findings point to a general class of auxiliary proteins that intercept CaM signaling to individual targets, allowing spatial and temporal orchestration of Ca 2+ -feedback.

Results
Stac selectively suppresses Ca 2+ -feedback of Ca V 1 channels We sought to determine stac effect on Ca V 1, Ca V 2, and Na V 1 channels in heterologous systems. Figure 1A shows baseline effects of stac on Ca V 1.2 (Campiglio et al., 2018;Polster et al., 2015;Wong King Yuen et al., 2017). Devoid of stac, Ca V 1.2 exhibits CaM-mediated CDI manifesting as enhanced decay of Ca 2+ (red) versus Ba 2+ current (black) when elicited using a step depolarization ( Figure 1A, middle subpanel). As Ba 2+ binds CaM poorly (Linse and Forsén, 1995), Ba 2+ -currents furnish a baseline measure of voltage-dependent inactivation (VDI) without CDI. Upon stac2 coexpression, CDI is diminished ( Figure 1A, right subpanel). To quantify steady-state extent of inactivation, we measured the fraction of peak Ca 2+ and Ba 2+ current remaining after 300 ms depolarization, r Ca and r Ba (Figure 1-figure supplement 1A). The strength of CDI is quantified as CDI 300 = 1 -r Ca /r Ba , the fractional Ca 2+ -dependent component of inactivation. Thus quantified, the population data confirm a reduction in CDI of Ca V 1.2 with stac2 (p=3.6 Â 10 À5 ; Figure 1B). Further analysis shows that both stac1 and stac3 isoforms also diminish CDI (p=2.0 Â 10 À5 and 7.1 Â 10 À5 , respectively, Figure 1B and  Left, cartoon schematic shows Ca V 1.2. Middle, exemplar current traces evoked in response to +10 mV voltagestep shows robust CDI (rose shaded area) evident as enhanced current decay with Ca 2+ (red) versus Ba 2+ (black) as the charge carrier. Right, stac2 abolishes CDI. Steady-state levels of inactivation are assessed as the fraction of peak current remaining following 300 ms depolarization (r Ca and r Ba ) and CDI = 1 -r Ca /r Ba . (B) Bar graph displays population data of CDI 300 for Ca V 1.2 in the absence and in the presence of stac1, stac2, or stac3. Dashed line shows baseline CDI in the absence of stac for comparison. Each bar, mean ±S.E.M. obtained from specified number of cells (n). (C-D) Stac isoforms suppress CDI of Ca V 1.3 S , the canonical short variant, as confirmed by both exemplar traces (C) and population data (D). Format as in (A) and (B). (E-F) Stac2 abolishes CDI of Ca V 1.4 43* assessed in response to +20 mV test pulse. Format as in (A) and (B). (G-H) Stac2 spares CDF of Ca V 2.1, as evaluated using a prepulse protocol. An isolated test pulse to 0 mV elicits Ca 2+ currents with biphasic activation Figure 1 continued on next page close homolog of Ca V 1.2, exhibits strong baseline CDI that is reduced on co-expression of stac1, stac2, and stac3 (p<1 Â 10 À5 ; Figure   . Encouraged by its pervasiveness, we considered whether stac alters Ca 2+ -dependent modulation of Ca V 2 isoforms that are abundant in the central nervous system. For Ca V 2.1, CaM elaborates both CDF and CDI (DeMaria et al., 2001;Lee et al., 2000). However, the Ca 2+ -sensitivity of CDI process is over 50-fold weaker than that of CDF, casting this negative feedback beyond physiological bounds . As such, we probed whether stac tunes CDF of Ca V 2.1 using a well-established prepulse protocol (DeMaria et al., 2001;Thomas and Lee, 2016). Figure 1G displays wildtype Ca V 2.1 currents in the absence of stac2. On presentation of an isolated test pulse to 0 mV, the activation of Ca 2+ current follows a biphasic response (gray trace). Following a brief voltage prepulse, however, the ensuing test pulse yields enhanced Ca 2+ -currents with monophasic activation reflecting CDF (red trace). Further quantification revealed no change in CDF of Ca V 2.1 following the addition of stac2 in both exemplar current recordings ( Figure 1G) and population data ( Figure 1H Here CDI is quantified by metric CDI 800 = 1 -r Ca /r Ba , measured following 800 ms of depolarization. Likewise, neuronal Ca V 2.3 also possesses robust CDI, which is spared with stac2 co-expression ( Lastly, we tested whether stac suppresses Ca 2+ -regulation of Na V 1, related to Ca V 1. Although all Na V 1 possess a CI module homologous to both Ca V 1 and Ca V 2 (Babitch, 1990), CDI that bears mechanistic similarity to Ca V has been identified only in Na V 1.4 (Ben- . Unlike Ca V , Na V channels do not convey Ca 2+ influx that triggers Ca 2+ -feedback. We used rapid photo-uncaging of Ca 2+ to produce a step-like increase in intracellular [Ca 2+ ] i , whose magnitude is simultaneously monitored via fluorescent indicators. Figure 1M displays baseline Ca 2+ -regulation of Na V 1.4. As CDI is kinetically slow in comparison to fast inactivation, we applied a train of step depolarizations evoked at 10 Hz to probe Ca 2+ -dependent effects . Without Ca 2+ -uncaging, peak Na V 1.4 currents remained steady (gray dots). In response to an~10 mM Ca 2+ step, the peak Na current declined rapidly revealing CDI (red envelope). Stac overexpression, however, does not disrupt Na V 1.4 CDI (Figure 1M-N; Figure 1-figure supplement 1G). Overall, these results show the specificity of stac in tuning Ca 2+ -regulation of Ca V 1 channels.

Stac interacts with Ca V 1 CI module to elicit CDI suppression
We sought to identify molecular mechanisms that underlie selective Ca V 1 modulation by stac. As the stac effect here is inferred based on overexpression analysis, we determined relative concentration requirements for stac binding to Ca V 1 holo-channel complexes within the milieu of living cells. We used live cell FRET 2-hybrid assay (Erickson et al., 2001) to probe the interaction of CFP-tagged stac3 with YFP-linked Ca V 1.3 S . As all three stac variants suppress the CDI of all Ca V 1 isoforms, we Figure 1 continued (gray, G). With a + 20 mV prepulse, channels are partially facilitated and the slow component of activation is reduced (red, G). The area between the two current traces (DQ), divided by t slow , yields facilitation (g). Bar graph plots, CDF = g Ca -g Ba H). Each bar, mean ±S.E.M from specified number of cells (n). (I-J) Stac2 spares CDI of Ca V 2.2 assessed in response to +30 mV test pulse. Here, CDI is evaluated following 800 ms of depolarization to accommodate slow inactivation kinetics, yielding CDI 800 . Format as in (A) and (B). (K-L) Stac2 spares CDI of Ca V 2.3. Format as in (A) and (B). (M-N) Stac2 spares CDI of Na V 1.4. Both in the absence and presence of stac, Na V 1.4 peak currents decline following a Ca 2+ step (rose fit) (M). Gray dots, peak currents before uncaging. CDI = 1average peak I Na of last three to four responses after Ca 2+ uncaging / peak I Na before uncaging. Bar graph plots maximal CDI observed with Ca 2+ steps > 5 mM (N). Each bar, mean ±S.E.M. DOI: https://doi.org/10.7554/eLife.35222.002 The following figure supplement is available for figure 1: chose Ca V 1.3 as YFP-tethered channels and a repertoire of YFP-tagged intracellular loop peptides are readily available for in-depth binding analysis . Stac3 was selected for its high potency in suppressing Ca V 1.3 CDI ( Figure 1D). Accordingly, we quantified FRET efficiency (E D ) between FRET pairs co-expressed in individual cells. By leveraging stochastic expression of the FRET pairs in cells, we obtained a saturating Langmuir relationship between E D and the free acceptor concentration (A free ) permitting estimation of relative binding affinities (K d,EFF ). Thus probed, we obtained a Ca V 1.3 holo-channel affinity for stac3 of K d,EFF = 1458 ± 251 D free units proportional to~47 nM (Figure 2A). By comparison, similar analysis of CaM binding to Ca V 1.3 showed K d,EFF = 700 D free units~22 nM . Consequently, stac's relatively high binding-affinity for Ca V 1.3 suggests that it may be a potent modulator even at low concentrations.
With holo-channel binding assured, we systematically scanned YFP-tagged Ca V 1.3 intracellular domains  for stac binding sites ( Figure 2B; Figure 2-figure supplement 1A). We found that stac3 binds well to the CI region (K d,EFF = 20697 ± 3023 D free~0 .67 ± 0.1 mM, Figure 2C). By contrast, analysis of the amino-terminus, intracellular loops between domains I and II  To further localize the putative binding loci, we subdivided the CI module into two: (1) a proximal CI segment (PCI) composed of dual vestigial EF hand and preIQ segments and (2) the IQ domain (IQ). The YFP-tagged PCI segment bound stac3 with approximately tenfold higher affinity (K d,EFF = 17725 ± 3990 D free~0 .58 ± 0.1 mM) than the downstream IQ domain (K d,EFF = 204739 ± 25465 D free~6 .67 ± 0.8 mM) ( Figure 2C-D). In all, systematic FRET analysis reveals that stac binds to Ca V 1 CI relying on upstream elements including the dual vestigial EF hand and preIQ domains, an interface distinct from that for CaM Minor and Findeisen, 2010).
To test for functional relevance of stac binding to the Ca V 1 CI module, we sought to confer stacsensitivity to a stac-insensitive channel via a chimeric approach. We turned to Ca V 2.3 that lacks strong stac-mediated CDI suppression, yet forms functional chimeras with Ca V 1 (Mori et al., 2008;Yang et al., 2014). We replaced the CI region of Ca V 2.3 with the corresponding segment from Ca V 1.3 (Ca V 2.3/1.3 CI). Devoid of stac, Ca V 2.3-1.3 CI channels exhibit CDI isolated by high intracellular buffering ( Stac uses an allosteric mechanism to suppresses CaM signaling Given that both CaM and stac share the CI module as an effector site, two disparate mechanistic possibilities may allow suppression of Ca 2+ -regulation. First, stac may competitively displace Ca 2+free CaM (apoCaM) from its preassociation site. Second, stac may supersede CaM signaling to the channel pore via an allosteric mechanism. Systematic FRET analysis suggests that stac preferentially binds upstream CI elements ( Figure 2D) while high-affinity CaM preassociation is supported via the IQ domain Minor and Findeisen, 2010), hinting that the two modulatory proteins may bind concurrently. To rule out competitive displacement of CaM preassociation, we covalently tethered CaM onto the Ca V 1.3 carboxy-tail using a poly-glycine linker (Ca V 1.3 S -CaM) (Mori et al., 2004;Yang et al., 2014). This maneuver preserves CDI ( Figure 3A left) and ensures a high local CaM concentration near Ca V 1 extending into the millimolar range, sufficient to protect the channel from a competitive inhibitor (Mori et al., 2004). Dominant-negative CaM 1234 with its Ca 2+binding sites disabled, typically displaces intact apoCaM from the CI module thereby resulting in a loss of CDI for wildtype channels ( . These findings suggest that stac does not need to dislodge CaM from its Ca V 1.3 carboxy-tail binding interface to exert functional modulation. To test this possibility, we undertook FRET 2-hybrid assay comparing binding of CFP-tagged CaM to YFP-tagged Ca V 1.3 CI in the presence and absence of unlabeled stac3. If stac3 were to competitively dislodge CaM, then this binding is predicted to be weakened. At baseline, CaM binds to Ca V 1.3 CI with a relative dissociation constant, K d,EFF~4 000 ± 291 D free units ( Figure 3E) (Ben . Upon co-expression of CaM 1234 , this baseline binding is weakened~11 fold, yielding a relative affinity of 47153 ± 4815 D free units ( Figure 3F). By contrast, co-expression of stac3 did not appreciably perturb CaM binding to the CI module with K d,EFF = 4182 ± 330 D free units ( Figure 3G). These results suggest that both stac and CaM act concurrently via distinct sites on the channel carboxy-tail, in contradiction with a competitive mechanism.
Elementary mechanisms underlying stac-regulation of Ca V 1 Beyond Ca 2+ -dependent regulation, apoCaM binding tunes the baseline activity of Ca V channels (Adams et al., 2014). Absent stac, Ca V 1 lacking prebound CaM adopts a low P O configuration (empty configuration, P O/E ) while apoCaM binding switches channels into a high P O mode (CaMbound configuration, P O/A ) (Adams et al., 2014). Ca 2+ /CaM divests this initial enhancement in P O and returns channels into a low P O gating mode (P O/E ) manifesting as CDI. The addition of stac as a regulatory agent enriches this modulatory scheme ( Figure 4A). Three distinct scenarios may underlie suppression of Ca 2+ -regulation by stac ( Figure 4B): (1) stac binding may pre-inhibit channels into the low P O configuration (P O/E ) akin to Ca 2+ -inactivated channels and prevent further Ca 2+ -modulation, (2) stac may obstruct Ca 2+ /CaM regulation while allowing apoCaM to change baseline P O , (3) stac binding may allosterically lock channels into a high P O mode irrespective of CaM-binding status. For Scenario 3, it is possible that the baseline P O of Ca V 1.3 with stac may be distinct from that observed with CaM-overexpression. These three scenarios may be distinguished at the single-molecule level by assessing Ca V 1 P O under various CaM-bound conditions using low-noise electrophysiology. We focused on Ca V 1.3 given the rich assortment of post-transcriptionally modified variants with distinct CaM binding affinities Liu et al., 2010;Singh et al., 2008). We focused on three variants, Ca V 1.3 S with high apoCaM affinity, and Ca V 1.3 MQDY and Ca V 1.3 L with low affinities. These variants possess distinct baseline P O and CDI and constitute a convenient platform to identify stac-dependent effects (Adams et al., 2014;Tan et al., 2011).
First, we analyzed Ca V 1.3 S in the presence and absence of stac ( Figure 4C-E) to determine whether stac may promote channel entry into a low P O gating configuration. Ca V 1.3 S is typically prebound to CaM at endogenous CaM concentrations given its high affinity (Adams et al., 2014). Ba 2+  is used as a charge carrier to estimate baseline behavior of channels without confounding effects of CDI. A slow voltage-ramp (shown between À50 and +40 mV) evokes stochastic channel openings that approximate near steady-state P O at each voltage. Stochastic records displayed in Figure 4C show channel openings as downward deflections to the open level (gray curves) and closures correspond to the zero-current portions of the trace. Robust openings are detected both in the presence and absence of stac ( Figure 4C). To estimate the steady-state P O -voltage relationship, we averaged many stochastic records to obtain a mean current that is divided into the open level and averaged over multiple patches. Ca V 1.3 S variant exhibits high P O in the absence of stac ( Figure 4D) (Adams et al., 2014). Upon stac2 co-expression, the open probability remains high with~10 mV hyperpolarizing shift in the voltage-dependence of activation ( Figure 4D). We scrutinized singlechannel trials to assess changes in gating modes. Figure 4-figure supplement 1 displays 10 sequential trials of Ca V 1.3 single-channel activity evoked by voltage-ramps introduced at 12 s intervals both in the presence and absence of stac. In the absence of stac, Ca V 1.3 S activity switches between epochs of high and low activity, as evident from the diary plot of average P O within individual trials ( P o ) (Figure 4-figure supplement 1B). Analysis of single-trial P o distribution reveals a bimodal distribution denoting discrete high and low P O gating modes ( Figure 4E). Upon stac overexpression, channel activity is high as evident from P o -diary plots ( Figure 4-figure supplement 1D) and single-trial P o distribution ( Figure 4E). In contradiction with Scenario 1, stac-bound channels are not pre-inhibited; rather, channels preferentially adopt a high P O mode.
To distinguish between the second and third mechanistic possibilities, we considered Ca V 1.3 variants with weakened apoCaM binding affinity that largely reside in the low P O configuration (Adams et al., 2014). Accordingly, we tested the baseline P O of Ca V 1.3 MQDY , an RNA-edited variant whose central isoleucine within the IQ domain is substituted to a methionine, Huang et al., 2012) and an alternative splice variant Ca V 1.3 L containing an autoinhibitory domain that competitively displaces CaM (Liu et al., 2010;Singh et al., 2008).  Figure 4K for Ca V 1.3 L ). CaM overexpression with both channel variants reveals the resurgence Figure 4 continued possibilities for stac binding to Ca V 1 and their functional outcomes. Scenario 1, stac uniformly suppresses P O of Ca V 1 (P O/E ) and abolishes CDI. Scenario 2, apoCaM tunes baseline P O of Ca V 1 despite concurrent stac binding. Stac, nonetheless, abrogates CDI. Scenario 3, stac uniformly upregulates the baseline P O of Ca V 1 and abolishes CDI (P O/A ). (C) Top, cartoon shows the canonical Ca V 1.3 S variant with a high apoCaM binding affinity. Single-channel analysis of recombinant Ca V 1.3 S in the absence (middle) and presence of stac (bottom). In both panels, the unitary Ba 2+ currents during voltage-ramp are shown between À50 mV and +40 mV (slanted gray lines, GHK fit). Robust Ca V 1.3 openings are detected in the absence and presence of stac. (D) Average single-channel P O -voltage relationship for Ca V 1.3 S obtained from multiple patches in the absence (gray) and presence of stac2 ( Figures 4D, G and J). These findings demonstrate that consistent with Scenario 3, stac-binding locks Ca V 1.3 channels in the high P O configuration and effectively decouples the channel pore from CaM-dependent conformational changes. Moreover, these results highlight the dominance of stac over CaM in Ca V 1 modulation.

U-domain constitutes a minimal motif for Ca V 1 CDI suppression
With elementary mechanisms discerned, we turned to identify stac motifs functionally critical for allosteric suppression of CaM-regulation. Structurally, stac isoforms share a modular architecture composed of a C1 domain linked to two SH3 domains via a largely unstructured linker segment (U-linker region) (Suzuki et al., 1996). As these modular subcomponents typically recognize distinct ligands, we reasoned that their molecular functions may be separable (Cohen et al., 1995;Colon-Gonzalez and Kazanietz, 2006). We trisected stac2 to assess whether individual subcomponents can recapitulate functional regulation. We focused initially on C1 and tandem SH3 domains as these segments were recently shown to be critical for Ca V 1.1 binding and triadic localization in skeletal muscle Wong King Yuen et al., 2017). Co-expression of either segment, however, only minimally perturbed CDI of Ca V 1. To localize functional segments within the U-linker, we undertook bioinformatic analysis to identify highly conserved regions. We performed multiple sequence alignment of complete sequences of 770 stac orthologs using the MUSCLE algorithm (Edgar, 2004) and subsequently computed an empirical measure for the degree of protein sequence conservation at each position. The degree of conservation is defined as the likelihood of observing the consensus residue at each sequence position divided by the number of distinct residues observed at this position. By this algorithm, perfectly conserved residues will yield a unitary value, whereas poorly conserved residues will have a lower score. We identified a 22-amino acid stretch, termed the U-domain ('unknown' domain), exhibiting high conservation ( Figure

U-domain modulates native Ca V 1 and reshapes cardiac action potentials
Having identified a minimal U-domain for CDI suppression, we sought to assess potential physiological consequences of stac regulation in cardiac myocytes. As stac expression is yet to be identified in myocytes, we first probed its presence using immunohistochemistry with stac1-and stac2-specific antibodies. To ensure that the two antibodies reliably probe the two isoforms, we first evaluated the ability to detect stac isoforms exogenously expressed in HEK293 cells (  endogenous stac2 with a similar molecular weight to that of recombinant stac2 in HEK293 cells (Figure 6-figure supplement 1G).
Given this baseline expression, we next considered potential effects of fluctuations in ambient stac levels. We synthesized the U-domain of stac2 as a peptide and delivered it via pipette dialysis to acutely elevate the myocyte's cytosolic concentration. We validated the synthesized peptide by testing its effect on recombinant Ca V 1.2 expressed in HEK293 cells ( Figure 6A). Following pipette dialysis of the U-peptide, CDI of Ca V 1.2 was reduced as evident from exemplar currents and population data ( Figure 6B-C; Figure 6- figure supplement 2A-B). Thus affirmed, we isolated ventricular myocytes from adult guinea pigs (aGPVMs) to probe changes in CDI of native Ca V channels and action potential duration in response to changes in stac levels ( Figure 6D). Devoid of U-domain peptide, endogenous Ca 2+ currents in ventricular myocytes displayed CDI, establishing baseline levels of CaM-regulation ( Figure 6E, Figure 6-figure supplement 2D). Pipette dialysis of U-peptide reduced CDI in myocytes ( Figure 6E-F, Figure 6-figure supplement 2E). The reduction in overall inactivation of Ca 2+ currents suggest that fluctuations in stac levels may markedly alter action potential waveforms. To test this possibility, we obtained current-clamp recordings of aGPVMs and compared action potential waveforms in the presence and absence of U-peptide. Figure 6G shows typical voltage profiles of action potentials in aGPVMs paced at 0.5 Hz. Waveforms are stable between traces and the mean action potential duration (APD 80 ), the duration of time when the action potential is at least 80% of its peak voltage, is 277.9 ± 31.37 ms (mean ±S.E.M., n = 6). Figure 6H displays the complement of the cumulative distribution of APD 80 . When the peptide is added to the internal solution, APD 80 is enhanced to 740.1 ± 105.49 ms (n = 6) ( Figure 6G-H). Thus, the U-peptide both alters the CDI of endogenous cardiac Ca V 1, prolongs APD, and may ultimately destabilize rhythmicity of the heart.
Fhf selectively abrogate CaM signaling to Na V 1 Encouraged by the selectivity of stac for Ca V 1, we sought to identify other regulatory proteins that may tune CaM-signaling to related channel families. However, recognizing such modulators amidst ion channel signalosomes is challenging. Given that stac interacts with Ca V 1 CI module via the PCI element, we reasoned that other Ca V and Na V interacting proteins that engage a similar interface may suppress CaM-feedback. Intriguingly, recent atomic structures show that fhf interacts with Na V 1 CI module via the PCI interface ( Figure 7A) . Yet, functionally, fhf isoforms are thought to modulate only voltage-dependent gating properties, with effects on Ca 2+ /CaM-regulation unknown (Goldfarb et al., 2007;Lou et al., 2005;Wang et al., 2012). To test whether fhf alters Na V CDI, we undertook quantitative Ca 2+ photo-uncaging of the skeletal muscle Na V 1.4 isoform. We focused here on fhf1b given its modest baseline expression in skeletal muscle and pathological enrichment in critical illness myopathies (Kraner et al., 2012). Figure 7B  suggesting that fhf may be a selective modulator of Na V 1.
Mechanistically, functional results along with atomic structures of Na V 1 CI bound to CaM and fhf yield insights on mechanisms for CDI suppression (Gabelli et al., 2014;Wang et al., 2012;Wang et al., 2014). Both fhf and CaM bind concurrently to Na V 1 CI Wang et al., 2014), with fhf binding triggering a conformational rearrangement of CaM ( Figure 7A) (Gabelli et al., 2014;Wang et al., 2012). To experimentally validate allostery, we followed our strategy with Ca V 1.3 and tethered CaM to Na V 1.4 carboxy-tail. Reassuringly Na V 1.4-CaM exhibits robust baseline CDI ( Figure 7F To garner a structural perspective, we turn to Na V 1.5 CI/fhf complex ( Figure 7A) as the atomistic basis of the stac/Ca V 1 CI interaction is unknown Wang et al., 2014;Wong King Yuen et al., 2017). Whereas the dual-vestigial EF hand segments of Na V 1.5 and Ca V 1.1 are similar ( Figure 7H-I), the fhf binding interface of Na V 1.5, including the preIQ loop diverges from corresponding segments of Ca V 1.1 and introduces a steric clash ( Figure 7I-J) Wu et al., 2016). Thus, by leveraging structurally distinct loci on the CI module, fhf selectively diminish CaM signaling to Na V channels. These findings point to a class of auxiliary proteins that selectively adjust Ca 2+ -dependent feedback to individual ion channel targets.  Figure 1A-B. Control data reproduced from Figure 1D for comparison. (F-G) Fhf1 suppresses CDI of Na V 1.4 tethered to CaM. Fusion of CaM protects Na V 1.4 from competitive inhibitors such as CaM 1234 (G). Format as in Figure 1M-N. (H) Structure of Ca V 1.1 upstream CI elements (blue) composed of dual vestigial EF hands and preIQ segments isolated from cryo-EM structure of Ca V 1.1 (PDBID, 5GJV). This domain is the primary interface for stac interaction in the Ca V 1 CI. (I) Structural overlay of upstream CI elements of Ca V 1.1 (PDBID, 5GJV) and Na V 1.5 (PDBID, 4DCK) shows highly conserved dual vestigial EF hand segments while the fhf binding site is structurally Figure 7 continued on next page Engineering synthetic modulation of Ca V channels As both stac and fhf tune Ca 2+ -feedback to individual Ca V and Na V targets by interacting with respective PCI segments, this mechanism furnishes a strategy to engineer synthetic channel modulators. We reasoned that introducing a short interaction motif into the PCI locus may permit inhibition of Ca V 1 Ca 2+ -feedback by a novel protein. We chose the well-characterized RxxK motif from SLP-76 for its small size and high-affinity interaction with SH3 domain of Mona (Harkiolaki et al., 2003) ( Figure 8A). Co-expression of Mona SH3 with wildtype Ca V 1.3 S demonstrated the persistence of CDI, confirming the suitability of these channels as a 'blank slate' to confer synthetic modulation (Figure 8B-C; Figure 8-figure supplement 1A). We replaced a 12-residue segment in the preIQ domain with the RxxK motif, as highlighted in Figure 8A, yielding Ca V 1.3 RxxK engineered channels. As this locus is situated upstream of the IQ domain, this maneuver spares apoCaM prebinding. These findings illustrate the versatility of the CI module as a regulatory hub and highlight the feasibility of developing synthetic modulators to tune Ca 2+ -feedback of ion channels.

Discussion
CaM is a dynamic regulator of Ca V 1, Ca V 2, and Na V 1, affording millisecond-precision Ca 2+ -feedback of channel activity. Our findings suggest that distinct auxiliary regulatory proteins tune CaM signaling to individual targets selectively. Stac prevents CaM signaling to Ca V 1, while fhf reduces signaling to Na V 1 ( Figure 8F). Parallel analysis of the two proteins delineates mechanisms and sets the stage for in-depth physiological analysis.
A few mechanistic nuances merit attention. First, stac binds to multiple Ca V 1 segments including (1) the II-III linker (Polster et al., 2018;Wong King Yuen et al., 2017), (2) the III-IV linker ( Figure 2D), and (3) the carboxy-tail ( Figure 2D) (Campiglio et al., 2018;Niu et al., 2018). Previous studies have shown that stac interaction with the II-III linker is important for Ca V 1 trafficking in skeletal muscle (Polster et al., 2018;Wong King Yuen et al., 2017). Chimeric analysis here suggests that stac interaction with the carboxy-tail is critical for tuning CDI. Prior analysis of Ca V 1.2 triadic localization in myotubes suggested that the channel IQ domain may be important for stac binding (Campiglio et al., 2018). However, FRET 2-hybrid assay indicates that stac interaction with the IQ is around tenfold weaker than with the PCI segment. Second, prior work also suggested that stacmediated reduction in CDI results from competitive displacement of CaM by stac (Campiglio et al., 2018). Functional experiments using Ca V 1 tethered to CaM, however, suggest that stac does not compete with CaM. Consistent with this scheme, FRET 2-hybrid analysis shows that CaM binding with the CI module is intact even in the presence of stac. Third, key domains within stac relevant for Ca V modulation remain controversial. Previous studies have identified the dual SH3 and C1 domains to be critical for stac effect on trafficking and coupling to RyR Linsley et al., 2017a;Linsley et al., 2017b;Polster et al., 2016), while the C1 has been proposed to be critical for modifying Ca V 1 CDI (Campiglio et al., 2018;Wong King Yuen et al., 2017). Our findings instead suggest that the U-domain in the stac2 linker region is sufficient to fully recapitulate reduction in Ca V 1 CDI. Notably, prior analysis of the C1 domain also included this linker (Wong King Yuen et al., 2017). Given these experimental findings, a simple possibility is that distinct subdomains within stac interact with disparate channel segments to support multifunctionality of stac. While the U-domain modifies channel inactivation, other subdomains may support plasmalemmal trafficking and conformational coupling. Defining a general class of auxiliary modulators of CaM signaling Although functionally divergent, Ca V 1, Ca V 2, and Na V 1 feature a modular CI element with a common CaM interaction fingerprint and subsequently, shared mechanistic basis for Ca 2+ -regulation. For all three families, apoCaM prebinds the CI module while Ca 2+ /CaM interaction switches channels between discrete high and low P O gating modes (Ben-Johny et al., 2015). How do allosteric regulators override CaM-signaling? First, stac and fhf use unique interfaces on the channel CI to selectively tune Ca 2+ -feedback. Second, stac locks Ca V 1 into a high P O gating mode irrespective of whether apoCaM or Ca 2+ /CaM is bound, effectively disengaging the pore from CaM-conformational changes. For Na V 1, despite fhf binding, CaM undergoes a profound Ca 2+ -dependent rearrangement Wang et al., 2014) suggesting that fhf does not prevent Ca 2+ binding to CaM or Ca 2+ /CaM interaction with effector interfaces. Instead, like stac and Ca V 1, fhf may override CaMdependent changes to Na V , akin to a clutch disengaging power transmission in mechanical systems. As fhf elicits a change in apoCaM conformation ( Figure 7A) (Gabelli et al., 2014;Wang et al., 2012), baseline gating of Na V may also be altered (Goldfarb et al., 2007;Lou et al., 2005). This parallelism between stac and fhf hints at a shared mechanism.
Ca 2+ -binding proteins (CaBPs) (Haeseleer et al., 2000) also suppress CaM signaling to Ca V 1 (Lee et al., 2002;Yang et al., 2006). Mechanistically, CaBPs exploit a mixed allosteric scheme -at low concentrations, they engage distinct interfaces from CaM but at higher concentrations displace CaM Oz et al., 2013;Yang et al., 2014). The existence of other regulatory proteins that curtail Ca 2+ -feedback points to a general class of auxiliary regulators of CaM-signaling to targets beyond Na V 1 and Ca V 1. Identifying such molecular players is critical to understand how CaM signaling is orchestrated.

Biological implications of stac modulation of Ca V 1
Stac1/2 isoforms are widely expressed in multiple brain regions, including both the hippocampus and the midbrain (Nelson et al., 2013;Suzuki et al., 1996). Our experiments hint at low basal stac2 expression in guinea pig ventricular cardiac myocytes, although previous studies have failed to detect stac2 in murine heart (Nelson et al., 2013). Further quantitative analysis will help establish ambient stac levels including species-specific differences and potential modulatory effects on cardiac function. Interestingly, endogenous Ca V 1 in both hippocampal and midbrain neurons Oliveria et al., 2012) as well as ventricular cardiac myocytes exhibit CDI. As all stac variants shunted CDI of Ca V 1 in HEK293, it is possible that stac function may be tightly regulated in native settings. One possibility is that stac abundance may be tuned developmentally (Suzuki et al., 1996), pathologically, or via interacting proteins (Satoh et al., 2006). For instance, the transcription factor, NFAT binds to an upstream promoter region of stac2 gene to upregulate stac2 expression in osteoclasts as well as during hypoxic conditions in neural stem cells (Jeong et al., 2018;Moreno et al., 2015). Physiologically, as Ca V 1 CDI is a potent homeostatic mechanism that prevents pathological Ca 2+ -overload (Dunlap, 2007), a low concentration regime of stac may be advantageous. By modulating a subpopulation of Ca V 1, stac may circumvent homeostatic requirements to amplify local Ca 2+ -signals via sustained Ca 2+ influx. The C1 and SH3 domains may serve as scaffolds to localize stac to specific signaling complexes Cohen et al., 1995;Colon-Gonzalez and Kazanietz, 2006). It is also possible that phosphorylation of stac may dynamically tune its function (Huttlin et al., 2010). Resolving these complexities may unveil mechanisms that tune Ca V function spatially and temporally.
In cardiac myocytes, CDI of Ca V 1 is a key factor for action potential duration (Limpitikul et al., 2014;Mahajan et al., 2008). Experimentally, this importance is inferred from prolongation of action potentials upon expression of mutant CaM 1234 (Alseikhan et al., 2002). Yet, constitutive CaM expression may yield nonspecific effects (Hall et al., 2013;Wang et al., 2007) that obscure the net contribution of Ca V 1 CDI . Acute elevation of the U-domain bypasses these ambiguities and confirms a key role for Ca V 1 CDI for cardiac action potentials. Pathophysiologically, differential expression of stac2 has been reported in right ventricular heart failure, hinting at a potential role in calcium remodeling during heart failure (di Salvo et al., 2015).
Post-transcriptional modification of Ca V 1.3 generates an assortment of variants with modified carboxy-termini (Bock et al., 2011;Huang et al., 2012). The apoCaM affinities of these variants are such that CaM fluctuations may redistribute channels between populations lacking or endowed with apoCaM , evoking concomitant changes in maximal P O and CDI of Ca V 1.3 (Adams et al., 2014). Stac uniformly locks these variants into a high P O configuration incapable of CDI, thereby supporting reliable and persistent Ca 2+ -influx in spite of CaM. Notably, functional effects of Ca V 1.3 alternative splicing have been shown to be cell-type specific suggesting that auxiliary regulators may tune channel properties (Scharinger et al., 2015). Fitting with these regulatory possibilities, disruption of stac modulation of Ca V 1.3 in Drosophila alters circadian rhythm (Hsu et al., 2018).

Biological implications of fhf modulation of Na V 1
Unlike canonical fibroblast growth factors, fhf lack a secretory signal sequence (Smallwood et al., 1996) and serve as intracellular proteins (Schoorlemmer and Goldfarb, 2001). Four distinct fhf isoforms have been identified with tissue-specific expression in neurons, cardiomyocytes, and skeletal muscle (Goldfarb, 2005;Kraner et al., 2012;Smallwood et al., 1996). Functionally, fhf isoforms promote Na V 1 trafficking and fast inactivation (Pablo and Pitt, 2016). More specifically, fhf adjust steady-state voltage-dependence of inactivation (Lou et al., 2005), elicit a kinetically distinct longterm inactivation (Dover et al., 2010), and modify resurgent current . Our present findings suggest that fhf1 also tunes CDI of Na V 1. Physiologically, Na V CDI may be prominent during repetitive activity, as excess Ca 2+ accumulation may inhibit Na currents. Thus, suppression of Na V 1 CDI by fhf may enhance repetitive firing. Interestingly, loss of fhf1 and/or fhf4 result in diminished firing properties of cerebellar Purkinje neurons (Bosch et al., 2015;Goldfarb et al., 2007), while loss of fhf2 reduces cardiac conduction (Park et al., 2016;Wang et al., 2011a). It is possible that loss of fhf may enhance net CDI thus contributing to diminished excitability in these cells. As mutations in fhf1 are associated with epileptic encephalopathy (For CENetDDD Study group ‡* et al., 2016) and cardiac conduction disorders (Hennessey et al., 2013) while mutations in fhf4 are linked to spinocerebellar ataxia (Brusse et al., 2006), resolving the dynamic interplay between CaM and fhf in tuning Na V 1 may be critical for understanding pathogenic mechanisms.

New strategy for synthetic ion channel modulation
Finally, our results highlight the possibility of engineering synthetic regulation to tune CaM signaling. While Ca V 1.3 is insensitive to Mona SH3, insertion of an RxxK motif (Harkiolaki et al., 2003) into the carboxy-tail preIQ segment allows latent modulation by Mona SH3. Given the structural similarity of the CI modules of Ca V 1, Ca V 2, and Na V 1, and sequence variability within the preIQ domain, emerging protein engineering methods may be used to screen for synthetic modulators of related ion channel families. As the ligand specificity of SH3 domains can be custom-engineered (Nguyen et al., 2000) and subcellular localization tuned via targeting motifs (Komatsu et al., 2010), a custom library of synthetic regulators may be developed to combinatorially modify kinetic properties of Ca V 1, Ca V 2, or Na V 1 channels with spatiotemporal specificity. Generalizing this approach may lead to the development of new tools to manipulate Ca 2+ signaling.

Cell culture and transfection of HEK293 cells
For whole-cell electrophysiology, single-channel electrophysiology, and immunohistochemistry, HEK293 cells (ATCC; mycoplasma tested negative) were cultured on glass coverslips in 10 cm dishes and transfected by a calcium phosphate method (Peterson et al., 1999) with the following amounts of DNA: 3 mg of SV40 T antigen to enhance expression, 2-8 mg of a 1 -subunit of Ca 2+ or Na + channel depending on expression, 8 mg from rat b 2A (Perez-Reyes et al., 1992) (M80545), 8 mg from rat a 2 d (Tomlinson et al., 1993) (NM012919.2), and 8 mg of the stac1, stac2, or stac3 variants indicated. For FRET two-hybrid experiments, cells were cultured on glass-bottom dishes and transfected with a standard polyethylenimine protocol (Lambert et al., 1996). Epifluorescence measurements were recorded 1-2 days after transfection.

Adult guinea pig ventricular myocyte isolation
Adult guinea pig ventricular myocytes (aGPVMs) were isolated from whole hearts of Hartley strain guinea pigs 3-4 weeks old (250-350 g). Guinea pigs were anesthetized via intraperitoneal injection with pentobarbital (35 mg/kg). Hearts were then excised, and single ventricular myocytes were isolated following a previously published protocol (Joshi-Mukherjee et al., 2013). Cells were plated on glass coverslips that were laminin (20 mg/mL) coated overnight at 4˚C.
Transfected HEK293 were immunostained following a similar protocol to that of aGPVM, but were not labelled with sarcomeric primary antibody and its respective secondary antibody.
Western blot aGPVMs and HEK293 cells were washed twice with PBS buffer. Cells were harvested with 1 mL 1x RIPA buffer (20-188, Sigma Aldrich) containing half a tablet of complete mini-EDTAfree protease inhibitor (11836170001, Sigma Aldrich) and incubated at 4˚C for 30 min. Samples were centrifuged at 15,000 RPM for 15 min, and the pellet was discarded. Then, 2-5 mg of proteins in the supernatant were heated at 37˚C for 30 min with 2x Laemmli sample buffer (S3401, Sigma Aldrich). Samples were loaded into 4-12% gradient gel (NP0335BOX, Invitrogen) with PageRuler plus prestained protein ladder (26619, Invitrogen) and run at 100 V for 2 hr at room temperature in running buffer: 1x NuPAGE MOPS SDS running buffer: 50 mM MOPS, 50 mM Tris base, 0.1% SDS, 1 mM EDTA, pH to 7.7. Proteins were transferred on ice from the gel to nitrocellulose membrane (10600003, GE Healthcare Life science) for 75 min at 10 V in transfer buffer: 24 mM Tris base, 192 mM glycine, 20% v/v methanol. Membrane was blocked with 5% (w/v) Blotting-Grade-Blocker (1706404, Bio-Rad) in 1x TRIS-buffered saline for 1 hr at 4˚C. Primary antibody for stac2 (1:250) was added to the blocking buffer with 0.1% (v/v) Tween 20 (1706404, Bio-Rad) and incubated overnight at 4˚C. Next day, the membrane was washed three times for 5 min each with TBS with 0.1% (v/v) Tween 20 (TBS-T). The secondary antibody (111-035-144, Jackson ImmunoResearch; 1:10,000) was added to the blocking buffer with 0.1% (v/v) Tween 20 and incubated for 1 hr. The membrane was washed again three times for 5 min each with TBS-T. Finally, western blots were developed with SuperSignal West Pico Chemiluminescent Substrate (34580, ThermoFischer) and images were collected on an Alpha InnoTech FluorChem HD2 imaging system.

Confocal optical imaging
Images of immunostained tissue slices and cells were captured with either an Olympus Fluorview FV300 confocal laser scanning microscope or an LSM780 (Carl Zeiss, Oberkochen, Germany) confocal microscope. For the FV300, we used Fluoview software (Olympus) with a PlanApo 403 or 603 oil objective (NA 1.40, PLAPO60XO3; Olympus). Argon laser (488 nm) was used to excite Alexa Fluor 488 (green), and Helium Neon (HeNe) Green Laser was used to excite Alexa Fluor 594 (red). Olympus optical filters used were 442/515 nm excitation splitter (FV-FCV), 570 nm emission splitter (FV-570CH), BA510 IF and BA530RIF for green emission channel, and 605 BP filter for red channel. Images were processed in ImageJ. Similar settings were used for the LSM780 setup.

Whole-cell electrophysiology
Whole-cell voltage-clamp recordings for HEK293 were collected at room temperature 1-2 days after transfection with Axopatch 200A (Axon Instruments). Glass pipettes (BF150-86-10, Sutter Instruments) were pulled with a horizontal puller (P-97; Sutter Instruments Company) and fire polished (Microforge, Narishige, Tokyo, Japan) to have 1-3 MW resistance. Recordings were low-pass filtered at 2 kHz and sampled at 10 kHz with P/8 leak subtraction and 70% series resistance and capacitance compensation. For recordings of Ca V 1.2 ( Figure 1A-B, Figure 5I-K, Figure 6B , or BaCl 2 5; adjusted to 300 mOsm with TEA-MeSO 3 and pH 7.4 with TEA-OH. At a holding potential of À80 mV, we used a family of test pulses from À30 mV to +50 mV with repetition intervals of 20 s. Custom MATLAB (Mathworks) software (https:// github.com/manubenjohny/WCDTY; copy archived at https://github.com/elifesciences-publications/ WCDTY) was used to determine peak current and fraction of peak current remaining after either 300 ms (r 300 ) or 800 ms (r 800 ) of depolarization.
For CDI recordings, we determined required sample size based on power analysis. Based on historical estimates of normal variation in CDI/CDF measurements, we computed the sample size required such that type I and type II errors are 5% to be 3.5. Thus, we obtained at least four independent measurements for all electrophysiological experiments.

Single-channel electrophysiology
Single-channel recordings were performed at room temperature using an on-cell configuration previously established in the laboratory (Tay et al., 2012) with the same setup as used for whole-cell electrophysiology. Glass pipettes were pulled and polished from ultra-thick-walled borosilicate glass (BF200-116-10, Sutter Instruments) and coated with sylgard to have 5-10 MW resistance. Recordings were filtered at 2-5 kHz. The pipette solution contained (in mM): TEA-MeSO 3 , 140; HEPES, 10; BaCl 2 40; adjusted to 300 mOsm with TEA-MeSO 3 and pH 7.4 with TEA-OH. The external solution contained (in mM): K glutamate, 132; KCl, 5; NaCl, 5; MgCl 2 , 3; EGTA, 2; HEPES, 10; adjusted to 300 mOsm with glucose and pH 7.4 with KOH. Cell-attached single-channel currents were measured during 200 ms voltage ramps between À80 and +70 mV (portions between À50 and 40 mV displayed and analyzed) as previously described. For each patch, we recorded 80-150 sweeps with a repetition interval of 12 s. Patches were analyzed as follows: (1) The leak for each sweep was fit and subtracted from each trace. (2) The unitary current relation, i(V), was fit to the open-channel current level using the following equation: i Þ where g is the single-channel conductance (~0.2 pA/mV), z is the apparent valence of permeation (~2.1), F is Faraday's constant, R is the gas constant, and T is the temperature in degrees Kelvin (assumed room temperature). These parameters were held constant for all patches, except for slight variations in the voltage-shift parameter V s~3 5 mV, as detailed below. (3) All leak-subtracted traces for each patch were averaged (and divided by the number of channels in the patch) to yield an I-V relation for that patch. As slight variability in V S was observed among patches, we calculated an average V S for each construct, V S,AVE . The data from each patch were then shifted slightly in voltage by an amount DV = V S,AVE -V S , with DV typically about ±5 mV. This maneuver allowed all patches for a given construct to share a common open-channel GHK relation. Thus shifted, the I-V relations obtained from different patches for each condition/construct were then averaged together. (4) P O at each voltage was determined by dividing the average I (determined in step three above) into the open-channel GHK relation. Channel number was determined by the maximal number of overlapping opening events upon application of the channel agonist Bay K8644 (5 mM) at the end of each recording. For modal analysis, a dashed line discriminator was chosen to be the average single-trial P O = 0.075 such that traces with average single-trial P O >0.075 were categorized as high P O while the remaining traces were considered to be low P O .

Quantitative calcium photo-uncaging
All Ca 2+ -uncaging experiments were conducted on a Nikon TE2000 inverted microscope with a Plan Fluor Apo 40 Â oil objective as previously described . Briefly, a classic Cairn UV flash photolysis system was used for Ca 2+ -uncaging with brief UV pulses of~1.5 ms in duration powered by a capacitor bank of up to 4000 mF charged to 200-290V. For concurrent Ca 2+ imaging, Fluo4FF and Alexa568 dyes were dialyzed via patch pipette and imaged using Argon laser excitation (514 nm). Background fluorescence for each cell was measured prior to pipette dialysis of dyes and subtracted subsequently. A field-stop aperture was used to isolate fluorescence from individual cells. Dual-color fluorescence emission was attained using a 545DCLP dichroic mirror, paired with a 545/ 40 BP filter for detecting Fluo4FF, and a 580LP filter for detecting Alexa568. Typically, uncaging experiments were conducted after~2 min of dialysis of internal solution. Welch's T-test was used to verify statistical significance between the population data.

FRET-two-hybrid assay
To collect a range of donor molecule (D free ) concentrations, HEK293 cells were transfected with combinations of DNA ratios. Cells were immersed in 2 mM Ca 2+ Tyrodes solution, which contained (in mM): NaCl, 138; KCl, 4; CaCl 2 , 2; MgCl 2 , 1; HEPES, 10; glucose, 10. Three-cube FRET fluorescence measurements were performed under resting Ca 2+ concentrations on an inverted fluorescence microscope. FRET efficiency (E A and E D ) was calculated for each cell (Erickson et al., 2001) and a binding curve, either E A = [D free ]/(K d,EFF + [D free ]) Á E A,max or E D = [A free ]/(K d,EFF + [A free ]), was fit to compute the effective dissociation constant (K d,EFF ).

Gordon Tomaselli Manu Ben-Johny
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. Ethics Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols of the Johns Hopkins University (GP15M172). The protocol was approved by the Committee on the Ethics of Animal Experiments of the Johns Hopkins University. All surgery was performed under sodium pentobarbital anesthesia, and every effort was made to minimize suffering. Data availability All data generated or analysed during this study are included in the manuscript and supporting files.