A synthetic peptide that prevents cAMP regulation in mammalian hyperpolarization-activated cyclic nucleotide-gated (HCN) channels

Binding of TRIP8b to the cyclic nucleotide binding domain (CNBD) of mammalian hyperpolarization-activated cyclic nucleotide-gated (HCN) channels prevents their regulation by cAMP. Since TRIP8b is expressed exclusively in the brain, we envisage that it can be used for orthogonal control of HCN channels beyond the central nervous system. To this end, we have identified by rational design a 40-aa long peptide (TRIP8bnano) that recapitulates affinity and gating effects of TRIP8b in HCN isoforms (hHCN1, mHCN2, rbHCN4) and in the cardiac current If in rabbit and mouse sinoatrial node cardiomyocytes. Guided by an NMR-derived structural model that identifies the key molecular interactions between TRIP8bnano and the HCN CNBD, we further designed a cell-penetrating peptide (TAT-TRIP8bnano) which successfully prevented β-adrenergic activation of mouse If leaving the stimulation of the L-type calcium current (ICaL) unaffected. TRIP8bnano represents a novel approach to selectively control HCN activation, which yields the promise of a more targeted pharmacology compared to pore blockers.


Introduction
Hyperpolarization-activated cyclic nucleotide-gated (HCN1-4) channels are the molecular correlate of the I f /I h current, which plays a key role in controlling several higher order physiological functions, including dendritic integration and intrinsic rhythmicity both in cardiac and neuronal cells (Robinson and Siegelbaum, 2003). Unique among the voltage-gated ion channel superfamily, HCN channels are modulated by the direct binding of cAMP to their cyclic nucleotide binding domain (CNBD). Binding of the cyclic nucleotide increases the channel open probability upon hyperpolarization via conformational changes in the CNBD that are propagated to the pore through the C-linker domain (DiFrancesco and Tortora, 1991;Wainger et al., 2001;Zagotta et al., 2003).
In addition to cAMP, HCN channels are regulated by TRIP8b, a brain-specific auxiliary (b) subunit, which modulates two independent features of the channel, namely trafficking and gating (Santoro et al., 2009;Zolles et al., 2009). For this dual regulation, TRIP8b binds HCN channels through two distinct sites: via the tetratricopeptide repeat (TPR) domain, which interacts with the last three amino acids (SNL) of HCN channels and regulates their trafficking; and via the TRIP8b core domain, which interacts with the CNBD and antagonizes the effect of cAMP on the voltage dependency of the channel (Santoro et al., 2011;Han et al., 2011;Hu et al., 2013).
Here, we focus our attention on the specific action of TRIP8b in preventing cAMP regulation of HCN channels. Given the brain-specific localization of TRIP8b, we posit that a TRIP8b-derived peptide drug, able to reproduce the effect of the full length protein on HCN channel gating, can be developed for orthogonal selective regulation of HCN in cells/tissues in which TRIP8b is not expressed. cAMP-dependent modulation of HCN channels underlies distinct roles of cAMP in heart rate regulation (DiFrancesco, 1993) and development of peripheral neuropathic pain (Emery et al., 2012;Herrmann et al., 2017), which can be dissected by using a TRIP8b-based tool. In this regard, peptide-based drugs (2-50 aa long) are emerging as a fascinating application area as they open new therapeutic possibilities with an advantage over small molecules in terms of specificity and affinity for the target (Fosgerau and Hoffmann, 2015;Henninot et al., 2018). To this end, we searched for the minimal peptide that binds to the CNBD and recapitulates the gating effect of full length TRIP8b in three HCN isoforms (HCN1, HCN2 and HCN4) and in the native I f current. In previous studies, we identified the core portion of TRIP8b (TRIP8b core, 80 aa long) that interacts with the HCN CNBD and prevents cAMP modulation in full length channels (Santoro et al., 2011;Hu et al., 2013;Saponaro et al., 2014). A recent paper (Lyman et al., 2017) reported an even shorter binding sequence of TRIP8b (37 aa). However, this peptide, which was identified by progressive truncation of TRIP8b core , failed to reproduce the binding affinity of the starting construct. Moreover, evidence for activity of this peptide on HCN currents is lacking. In the present study, we adopted a structure-driven rational design approach to engineer a 40-aa long peptide, TRIP8b nano , that efficiently prevents cAMP regulation of HCN channels. The rational design of this peptide, based on secondary structure predictions and on NMR data of TRIP8b core , was supported by an NMR-based 3D model structure of the complex formed by the TRIP8b nano peptide and CNBD of the human HCN2 channel isoform. This structural information identifies crucial interactions between the two partners and explains both direct (Han et al., 2011;DeBerg et al., 2015;Bankston et al., 2017) and indirect (allosteric) (Hu et al., 2013;Saponaro et al., 2014) modes of competition between TRIP8b and cAMP for binding to the CNBD. The evidence that TRIP8b nano establishes all relevant interactions with the CNBD is reflected by the finding that, contrary to shorter core sequences (Lyman et al., 2017), it binds to the isolated CNBD with the same affinity as TRIP8b core and acts with even higher efficacy than TRIP8b core in preventing cAMP modulation of full length HCN channels (Hu et al., 2013). In pacemaker myocytes of the sino-atrial node (SAN), TRIP8b nano equally prevented cAMP stimulation of native f-channels leading to a 30% reduction in spontaneus firing rate.
To develop TRIP8b nano as a membrane permeable drug, we linked it with the positively charged TAT sequence (Herce et al., 2014). TAT-TRIP8b nano was tested in SAN pacemaker myocytes where its addition to the extracellular buffer prevented adrenergic stimulation of the I f current leaving the activation of the L-type calcium current (I CaL ) unaffected. Our study opens the possibility of selective in vivo control of the cAMP-dependent facilitation of HCN channel opening, by local supply of TAT-TRIP8b nano peptide.

Results
We have previously shown that TRIP8b core (residues 223-303 of mouse TRIP8b splice variant 1a4, hereafter TRIP8b) interacts with two elements of the isolated CNBD protein fragment from HCN channels (residues 521-672 of human HCN2, hereafter CNBD): the C-helix and the N-bundle loop, a sequence connecting helix E' of the C-linker with helix A of the CNBD (Saponaro et al., 2014). Biochemical assays confirmed that each of these two elements, that is, the N-bundle loop and C-helix, is necessary but not sufficient for binding (Saponaro et al., 2014).
To understand the interaction in atomic detail, we used solution NMR spectroscopy to characterize the structural properties of the CNBD -TRIP8b core complex. However, the NMR spectra of TRIP8b core showed very few signals. In order to improve the quality of the NMR spectra, we reduced the length of the TRIP8b fragment by progressively removing residues at the N-and C-termini with no predicted secondary structure. The truncated peptides were then tested for CNBD binding activity by isothermal titration calorimetry (ITC). We thus identified a 40-aa peptide (TRIP8b nano , comprising residues 235-275 of TRIP8b, Figure 1A) with a binding K D of 1.5 ± 0.1 mM, a value similar to the K D of 1.2 ± 0.1 mM obtained with TRIP8b core ( Figure 1B). TRIP8b nano was therefore employed for all subsequent NMR experiments, resulting in a remarkable improvement in the spectral quality and sample stability.

Structural characterization of TRIP8b nano bound to CNBD
The comparison of the 1 H-15 N HSQC spectra of TRIP8b nano with and without CNBD bound shows that the peptide folds upon interaction with the CNBD. Thus, the 1 H-15 N HSQC spectrum of TRIP8bnano without CNBD shows a limited 1 H resonance dispersion, characteristic of unstructured proteins (Dyson and Wright, 2004), while a larger number of well-dispersed amide signals appear in the spectrum of the CNBD-bound form ( Figure 1C). Importantly, we were now able to assign the backbone chemical shift resonances of TRIP8b nano bound to the CNBD. The j and y dihedral angles obtained from the NMR assignment indicate that the peptide displays two a-helices (stretch L 238 -E 250 named helix N and stretch T 253 -R 269 named helix C) when bound to CNBD. The helices are separated by two amino acids; three and six residues at the N-and C-termini, respectively, are unstructured ( Figure 1D).

Structural characterization of CNBD bound to TRIP8b nano
NMR-analysis of the CNBD fragment bound to TRIP8b nano revealed that the interaction with the peptide does not affect the overall fold of the protein. Thus, the CNBD adopts the typical fold of the cAMP-free state, in line with previous evidence that this is the form bound by TRIP8b (Saponaro et al., 2014;DeBerg et al., 2015) More specifically, the secondary structure elements of the cAMP-free CNBD are all conserved in the TRIP8b nano -bound CNBD ( Figure 2). This finding generally agrees with a previous double electron-electron resonance (DEER) analysis of the CNBD -TRIP8b interaction, which showed that TRIP8b binds to a conformation largely similar to the cAMPfree state (DeBerg et al., 2015). Despite the overall agreement with the DEER study, the NMR data also reveal a new and unexpected feature of TRIP8b binding to the CNBD. Indeed, our results show that TRIP8b nano binding to the CNBD induces a well-defined secondary structure of the distal region of the C-helix ( Figure 2). This means that the distal region of the C-helix (residues 657-662), which is unstructured in the free form of the CNBD (Saponaro et al., 2014;Lee and MacKinnon, 2017), extends into a helical structure upon ligand binding irrespectively of whether the ligand is cAMP (Puljung and Zagotta, 2013;Saponaro et al., 2014;Lee and MacKinnon, 2017) or TRIP8b (Figure 2). In contrast, and very differently from cAMP, which directly contacts the P-helix in the Phosphate Binding cassette (PBC) and causes its folding (Saponaro et al., 2014;Lee and MacKinnon, 2017), the NMR data show that TRIP8b nano binding to the CNBD does not induce P-helix formation ( Figure 2).

Modeling the CNBD -TRIP8b nano complex
Despite the significant improvement in sample stability and NMR spectra quality achieved upon TRIP8b nano binding, we were still unable to assign the side chains of both proteins in the complex and thus could not solve the solution structure of the complex by the canonical NMR procedure. We therefore built a model (Figure 3) of the CNBD -TRIP8b nano complex by docking the two NMRderived structures described above using the Haddock program (a detailed description of how the respective structures were generated is provided in Materials and methods and Figure 3-source data 1).
In order to define the active residues (ambiguous interaction restraints) on the CNBD we used the chemical shift perturbation values as described in Figure 3-figure supplement 1. For TRIP8bnano , we defined as active a stretch of residues, E 239 -E 243 , previously identified as critical for the interaction (Santoro et al., 2011). Output clusters of this first molecular docking calculation (settings can be found in Materials and methods) were further screened for TRIP8b nano orientations in agreement with a previous DEER analysis, which placed TRIP8b residue A 248 closer to the proximal portion and TRIP8b residue A 261 closer to the distal portion of the CNBD C-helix (DeBerg et al., 2015). Remarkably, in all clusters selected in this way, residues E 264 or E 265 in TRIP8b were found to interact with residues K 665 or K 666 of the CNBD (Figure 3-figure supplement 2). This finding was notable, because we previously identified K 665 /K 666 as being critical for TRIP8b interaction in a biochemical binding assay (Saponaro et al., 2014). We thus proceeded to individually mutate each of these four positions, and test their effect on binding affinity through ITC. As expected, reverse charge mutations K 665 E or K 666 E (CNBD) as well as E 264 K or E 265 K (TRIP8b nano ) each strongly reduced the CNBD/ TRIP8b nano binding affinity ( Figure 3-figure supplement 3).
Based on these observations, we performed a second molecular docking calculation, including E 264 and E 265 as additional active residues for TRIP8b nano . This procedure resulted in the model shown in Figure 3, which represents the top-ranking cluster for energetic and scoring function (Figure 3-source data 2) and was fully validated by mutagenesis analysis as described below. Scrutiny of the model shows that TRIP8b nano binds to both the C-helix and the N-bundle loop ( Figure 3A).
Binding to the C-helix is mainly guided by electrostatic interactions between the negative charges on TRIP8b nano , and the positive charges on the CNBD ( Figure 3A). As shown in Figure 3B, the model highlights a double saline bridge (K 665 and K 666 of CNBD with E 265 and E 264 of TRIP8b nano ) in line with the ITC results described above (Figure 3-figure supplement 3). Of note, the contribution of residue R 662 to the binding is also consistent with previous experiments showing residual TRIP8b interaction in a CNBD deletion mutant ending at position 663 (Saponaro et al., 2014). Our modeling data suggest that, upon folding of the distal portion of the C-helix, the side chains of residues R 662 and K 665 face to the inside when contacting cAMP, but face to the outside when binding TRIP8b ( Figure 3-figure supplement 4). This indicates that cAMP and TRIP8b directly compete for the binding to the distal region of C-helix.
In addition to clarifying the role of residues in the distal portion of the CNBD C-helix, the model also highlights a second important cluster of electrostatic interactions, with R 650 in the proximal portion of the CNBD C-helix contacting E 240 and E 241 in helix N of TRIP8b nano ( Figure 3C). To confirm the contribution of these residues, we reversed charges and tested each residue mutation for binding in ITC. The results in Figure 3-figure supplement 3 show that R 650 E caused a more than sixfold reduction in binding affinity for TRIP8b nano , with smaller but significant effects seen also for E 240 R and E 241 R.
A third important contact highlighted by the model is the interaction between N 547 in the N-bundle loop of the CNBD and D 252 in the link between helix N and helix C of TRIP8b nano ( Figure 3D). We tested this potential interaction by disrupting the expected hydrogen bond between N 547 and the carboxyl group of the negative residue (D 252 ) in TRIP8b nano . The asparagine in CNBD was mutated into aspartate (N 547 D) to generate an electrostatic repulsion for D 252 , and the carboxyl group in D 252 of TRIP8b nano was removed by mutation into asparagine (D 252 N). As predicted, N 547 D greatly reduced binding to TRIP8b in ITC assays (  . Lower panel, binding curve obtained from data displayed in the upper panel. The peaks were integrated, normalized to TRIP8b peptide concentration, and plotted against the molar ratio (TRIP8b peptide/His 6 -MBP-CNBD). Solid line represents a nonlinear least-squares fit to a single-site binding model, yielding, in the present examples, a K D = 1.2 ± 0.1 mM for TRIP8b core and K D = 1.4 ± 0.1 mM for TRIP8b nano . (C) Evidence for TRIP8b nano folding upon CNBD binding based on the superimposition of the [ 1 H, 15 N] heteronuclear single quantum coherence (HSQC) NMR spectrum of CNBD-free TRIP8b nano (black) and CNBD-bound TRIP8b nano (red). The latter experiment was performed at the molar ratio (   (Saponaro et al., 2014), cAMP-bound CNBD (Zagotta et al., 2003) and cAMP-free CNBD bound to TRIP8b nano (this study). Secondary structure elements are indicated by arrows (b-strands) and cylinders (a-helices) and labeled. The loop between b 6 and b 7 constitutes the Phosphate Binding Cassette (PBC). The elements that fold upon binding of cAMP and TRIP8b nano are shown in red. (Bottom) Chemical Shift Index (CSI, calculated using TALOS+) plotted as a function of the residue number of CNBD bound to TRIP8b nano . Positive values represent helical propensity, while negative values represent strands. (B) Ribbon representation of the 10 lowest energy conformers of CNBD bound to TRIP8b nano used for in silico modeling of CNBD -TRIP8b nano complex. Secondary structure elements are coloured in light gray and labeled. Loop regions are colored in dark gray. The distal region of the C-helix (residues 657-662), which is unfolded in the free form of the CNBD (Saponaro et al., 2014) and folds upon TRIP8b nano binding, is coloured in red. The unfolded regions at the Nand C-termini of the construct (residues 521-532 and 663-672 respectively) are omitted for clarity. DOI: https://doi.org/10.7554/eLife.35753.003  confirm and extend our previous finding that the N-bundle loop contributes in a substantial manner to the binding of TRIP8b (Saponaro et al., 2014). To understand the mechanism for the allosteric effect of TRIP8b on cAMP binding, which has been postulated on the basis of electrophysiological and structural data (Hu et al., 2013;Saponaro et al., 2014), we further tested by ITC the affinity of the N 547 D CNBD mutant for cAMP. Somewhat surprisingly, the affinity of the mutant is much lower than that of the wt (N 547 D K D = 5.5 ± 0.4 mM (n = 3) vs. wt K D = 1.4 ± 0.1 mM (n = 3)). Moreover, we measured a reduced sensitivity to cAMP also in patch clamp experiments where addition of 5 mM cAMP caused a right shift in the V 1/2 of the mutant HCN2 channel of only 5 mV while the wt channel shifted by 12 mV (Figure 3-figure supplement 5). To exclude that the N 547 D mutation affects the overall structure of the CNBD, we performed the NMR ( 1 H-15 N HSQC spectrum) analysis of the N 547 D CNBD mutant. Our data show that the protein is appropriately folded (Figure 3-figure supplement 6). In conclusion, since the N-bundle loop does not directly contact any of the residues of the cAMP binding pocket, these findings underscore a previously unaddressed role of the N-bundle loop in allosterically modulating cAMP binding to the CNBD (see Discussion).

TRIP8b nano as a tool for the direct regulation of native HCN currents
Next, we asked whether the relatively short TRIP8b nano could be used to block cAMP-dependent modulation of HCN channels by delivering the peptide to full length channels. To this end, we dialyzed TRIP8b nano into the cytosol of HEK 293 T cells transfected either with HCN1, HCN2, or HCN4 channels. The peptide was added (10 mM) in the recording pipette together with a non-saturating concentration of cAMP (5 mM for HCN2, 1 mM for HCN4) expected to induce a~10 mV rightward shift in the half-activation potential (V 1/2 ) of the channels (Figure 4). No cAMP was added in the case of HCN1, because, in HEK 293 T cells, this isoform is already fully shifted to the right by the endogenous cAMP and does not respond further (Figure 4-figure supplement 1). Indeed, it is possible to induce a~10 mV left shift in HCN1 V 1/2 by introducing the mutation R 549 E that prevents cAMP binding to the CNBD (Figure 4-figure supplement 1). Figure 4A-C show representative current traces recorded at four given voltages, in control solution, cAMP, and cAMP +10 mM TRIP8b nano (HCN4 and HCN2) or +10 mM TRIP8b nano only (HCN1) in the patch pipette. Already from a visual comparison of the most positive voltage at which the current appears measurable, it is evident that TRIP8b nano counteracts the activating effect of cAMP on the voltage-dependent gating. In the case of HCN1, the effect of TRIP8bnano can be observed without added cAMP for the aforementioned reasons. Figure 4D-F show the mean channel activation curves obtained from the above and other experiments. Fitting the Boltzmann equation to the data (solid and dashed lines of Figure 4D-F, see Materials and methods for equation) yielded the half-activation potential values (V 1/2 ) plotted in Figure 4G. The addition of TRIP8b nano prevents the cAMP-induced right shift of about 13 mV in HCN4 (V 1/2 = -102.8, -89.2, À102.1 mV, for control, cAMP and cAMP + TRIP8b nano , respectively), and of about 11 mV in HCN2 (V 1/2 = -93.7, -83.5, À94.5 mV, for control, cAMP and cAMP + TRIP8b nano , respectively). In HCN1, TRIP8b nano induced a left shift in V 1/2 of about 10 mV (from -72.8 to À82 mV) which is comparable to that induced by the R 549 E mutation (from À72.7 to À80.4 mV) (Figure 4-figure supplement 1). Figure 4G also shows the result of a control experiment performed on HCN4 where 10 mM TRIP8b nano was added to the extracellular medium (TRIP8b nano bath) in order to test if the peptide was able to cross the cell membrane (current traces and activation curves not shown). The V 1/2 value, which is similar to that of cAMP alone (À91.7 vs. À89.2 mV), confirmed that TRIP8b nano peptide  , cAMP (filled circle) and cAMP +TRIP8b nano , or TRIP8b nano only in the case of HCN1 (filled triangle). HCN4, control = À102.8 ± 0.3 mV; HCN4 +1 mM cAMP = À89.2 ± 0.6 mV; HCN4 +1 mM cAMP +10 mM TRIP8b nano = À102.1 ± 0.6 mV, HCN4 +1 mM cAMP in the patch pipette +10 mM TRIP8b nano in the bath solution = À91.7 ± 0.3 mV; HCN2, control = À93.7 ± 0.3 mV; HCN2 +5 mM cAMP = À83.5 ± 0.3 mV; HCN2 +5 mM cAMP +10 mM TRIP8b nano = À94.5 ± 0.6 mV; HCN1, control = À72.8 ± 0.2 mV; HCN1 +10 mM TRIP8b nano = À82 ± 0.5 mV. Data are presented as mean ± SEM. Number of cells (N) ! 11. There is no significant difference between the controls and the addition of TRIP8b nano with (HCN4, HCN2) or without (HCN1) cAMP in the pipette. No significant difference was observed following the addition of TRIP8b nano in the bath. Statistical analysis performed with twoway ANOVA, followed by post-hoc Tukey test (p<0.001). DOI: https://doi.org/10.7554/eLife.35753.013 The following figure supplements are available for figure 4: affects channel gating only if added to the intracellular solution presumably because it does not diffuse through the cell membrane.
It is worth noting that TRIP8b nano prevents other related effects of cAMP activation in HCN channels, such as the acceleration of activation kinetics (Wainger et al., 2001) and, for HCN2 only, the increase in maximal current (Chen et al., 2007;Hu et al., 2013). For example, the activation kinetics (t on ) of HCN4 measured at À120 mV was: control = 2 ± 0.2 s, 1 mM cAMP = 1.2 ± 0.1 s, 1 mM cAMP + 10 mM TRIP8b nano = 2 ± 0.3 s (Figure 4-figure supplement 2). Moreover, Figure 4B clearly shows that TRIP8b nano fully prevented the increase in maximal current in HCN2.
Based on these results, we reckoned the peptide may be employed as a regulatory tool for native I f /I h currents. As proof of principle, we tested whether TRIP8b nano can modulate the frequency of action potential firing in SAN myocytes. In these cells, I f is key contributor of the diastolic depolarization phase of the pacemaker action potential cycle. Moreover, the autonomic nervous system modulates the frequency of action potential firing by changing intracellular cAMP levels, which in turn acts on f-HCN channel open probability (DiFrancesco, 1993). We thus recorded the native I f current in acutely isolated rabbit SAN myocytes with and without 10 mM TRIP8b nano in the pipette solution ( Figure 5A). Figure 5B shows that the averaged I f activation curve measured in presence of TRIP8bnano is significantly shifted to hyperpolarized voltages compared to the control. This indicates that the peptide is displacing the binding of endogenous cAMP to native HCN channels. Moreover, when the experiment was repeated in the presence of 1 mM cAMP, TRIP8b nano prevented the typical cAMP-dependent potentiation of the native I f current ( Figure 5B). In light of these results, we tested whether TRIP8b nano is also able to modulate cardiac automaticity by antagonizing basal cAMP. The data in Figure 5C show that TRIP8b nano indeed significantly decreased the rate of action potential firing in single SAN cells. Strikingly, the observed 30% decrease in action potential rate corresponds to the effect induced by physiological concentrations of acetylcholine (DiFrancesco et al., 1989).
To conclusively prove that the inhibition of the native I f current was specifically due to TRIP8b nano rather than caused by the dilution of the cellular content following whole cell configuration, we created a TAT version of TRIP8b nano (hereafter TAT-TRIP8b nano ). Indeed, the TAT sequence allows the entry of biomolecules into a cell via endocytosis and/or direct translocation across the plasma membrane, thus leaving the cytosolic content unaltered (Guidotti et al., 2017).
We therefore tested whether both TRIP8b nano and TAT-TRIP8b nano were able to selectively inhibit the b-adrenergic stimulation of I f current, while leaving the potentiation of L-type Ca 2+ current (I Ca,L ) unaltered. To this end, we recorded either the native I f or I Ca,L current from cardiomyocytes acutely isolated from mouse sinoatrial node (SAN) in the presence and in the absence of 10 mM TRIP8b nano or TAT-TRIP8b nano , before and after stimulation with 100 nM isoproterenol (ISO), a b-adrenergic receptor agonist ( Figure 6). Strikingly, TRIP8b nano prevented the isoproterenol-induced increase of I f current density, both when the peptide was added in the recording pipette solution ( Figure 6A and B), and when it was used in the TAT version added to the bath ( Figure 6A and C). The specificity of TRIP8b nano for I f current was confirmed by the absent inhibition of basal I Ca,L ( Figure 6D). In addition, we failed to record a significant difference in the isoproterenol-stimulated increase of the I CaL current density between the control condition and 10 mM TRIP8b nano ( Figure 6E) or TAT-TRIP8b nano ( Figure 6F) conditions. To test whether the TAT-TRIP8b nano effect described above was exclusively due to TRIP8b nano peptide, we repeated the experiments with a scrambled version of the peptide (TAT-(SCRAMBLED) TRIP8b nano ) to exclude that the effect could be due to the TAT sequence (Figure 6-figure supplement 1). We failed to observe a significant reduction in the responsiveness of I f to isoproterenol in the presence of TAT-(SCRAMBLED) TRIP8b nano confirming that prevention of cAMP induced f-current stimulation was specific of the TRIP8b nano sequence.

TRIP8b-CNBD complex
In this study, we have identified the minimal binding peptide that reproduces the effects of TRIP8b on HCN channel gating. The peptide is 40 aa long and binds the HCN CNBD with high affinity (K D = 1.4 mM). By solving the NMR structures of TRIP8b nano and HCN CNBD in the bound form, we generated a structural model of their complex. The model provides detailed information on this protein-protein interaction at the atomic level with implications on their physiological function. The data show that the minimal binding unit of TRIP8b, TRIP8b nano , folds in two helices upon binding and suggest that this region is intrinsically disordered when it is not bound. The model structurally validates previous indirect evidence, which suggested that TRIP8b binds to two discrete elements of the CNBD: the N-bundle loop and the C-helix (Saponaro et al., 2014). The complex forms by electrostatic interactions, which are spread throughout the contact surface. As a consequence of the interaction with TRIP8b nano , the C-helix of CNBD increases in length, a behavior previously observed in the case of cAMP binding (Puljung and Zagotta, 2013). This portion of C-helix includes the two residues R 662 and K 665 engaged in salt bridge formation with respectively E 250 /D 257 and E 264 of TRIP8bnano . It is important to note that these two cationic residues are also involved in cAMP binding (Zagotta et al., 2003;Zhou and Siegelbaum, 2007;Lolicato et al., 2011). The finding that TRIP8b and cAMP share the same binding sites on the C-helix provides a solid molecular explanation for Mean normalized I f current density recorded at À95 mV in absence and in presence of 10 mM TRIP8b nano in the patch pipette, before (open bars) and after Figure 6 continued on next page functional data, which imply a competition between the two regulators (Han et al., 2011;DeBerg et al., 2015;Bankston et al., 2017). Another study, however, has indicated that a direct competition model cannot fully explain the mutually antagonistic effect of the two ligands (Hu et al., 2013). Specifically, the fact that the inhibitory effect of TRIP8b on channel activity persists even at saturating cAMP concentrations advocated an allosteric component in the regulation mechanism. Our data, showing that the N 547 D mutation in the N-bundle loop controls cAMP affinity in the binding pocket support the conclusion that the N-bundle loop allosterically controls cAMP binding. This is not surprising, given its crucial role of mechanically transducing to the pore the cAMP binding event within the CNBD (Saponaro et al., 2014).
The structural model also explains why a previously identified peptide selected by Lyman et al. (2017) failed to reproduce the binding affinity of the TRIP8b core for the CNBD. This 37 aa long fragment is lacking one important contact residue, namely E 240, which, in our model, forms a salt bridge with R 650 of the CNBD. The loss of one crucial interaction is presumably the reason for the major decrease in affinity (about 20 times lower) reported for this peptide.

TRIP8b nano as a tool for modulating native I f currents
In functional assays, we showed that TRIP8b nano binds the HCN channel CNBD with high affinity and fully abolishes the cAMP effect in all tested isoforms (HCN1, 2 and 4).
Given the small size of the peptide (<5 kDa), TRIP8b nano is a good candidate for in vivo delivery into intact cells. As a proof of concept, we fused TRIP8b nano to an internalization sequence, the TAT peptide (YGRKKRRQRRRGG). This arginine-rich Cell Penetrating Peptide (CPP) from HIV has been used in several studies as a vehicle for the delivery of large molecules across the plasma membrane (Guidotti et al., 2017). In our case, the challenge was to construct a TAT-fusion protein, which would be efficiently delivered in the cell without compromising TRIP8b nano function. Indeed, covalent conjugation of a CPP may negatively affect both the function of the cargo, and the cell-penetrating efficacy of the CPP-peptide chimera (Kristensen et al., 2016). The design of the construct was greatly supported by the detailed knowledge of the electrostatic interactions with the target protein CNBD, provided by the NMR model structure. This structure suggested that the polycationic TAT sequence would be best linked to the N-terminus of TRIP8b nano to avoid interference with the cationic residues of CNBD, mainly located in the distal region of C-helix, which are crucial for the binding of the peptide. From test experiments with the TAT-TRIP8b nano peptide in SAN myocytes, we can conclude that this strategy was successful in that: (i) the peptide is efficiently delivered inside the cells; (ii) it is kept in its active conformation; (iii) the TAT sequence did not damage cell membranes and did not interfere with the basic features of I f and I Ca,L currents; (iv) the modification did not affect the proteolytic stability of the TRIP8b nano peptide at least in the time frame of our experiments (30 min to 1 hr). Representative examples of I Ca,L recordings at À10 mV in control conditions (top), in 10 mM TRIP8b nano dialyzed cell (middle) and in cells perfused with 10 mM TAT-TRIP8b nano (bottom), before (black trace) and after (red trace) application of ISO 100 nM. The voltage-clamp protocol used for recordings is shown above current traces. (E) Mean normalized I Ca,L current density recorded at À10 mV in absence and in presence of 10 mM TRIP8b nano in the patch pipette, before (open bars) and after (filled bars) 100 nM ISO perfusion. Data are presented as mean ± SEM. Number of cells (N) ! 8. Statistical analysis performed with two-way ANOVA test, followed by Sidak multiple comparisons test (***p<0.001; ****p<0.0001). (F) Mean normalized I Ca,L current density recorded at À10 mV in control solution or in the solution containing 10 mM TAT-TRIP8b nano , in absence (open bars) and in the presence (filled bars) of 100 nM ISO. Data are presented as mean ± SEM. Number of cells (N) ! 7. Statistical analysis performed with two-way ANOVA test, followed by Sidak multiple comparison test (**p<0.01; ***p<0.001). DOI: https://doi.org/10.7554/eLife.35753.017 The following figure supplement is available for figure 6: In conclusion, we successfully used the miniaturized TRIP8b nano peptide to selectively control native I f currents and the rate of spontaneous firing in SAN myocytes. Unlike channel blockers, which inhibit ionic currents, the peptide only interferes with the cAMP-based regulation of HCN channels, while leaving basal HCN functions unaltered. In addition and in contrast to even the most selective blockers, it is selective for HCN and it does not interfere with other cAMP-modulated channels present in the SAN, such as L-type Ca 2+ channels. Collectively, this makes TRIP8b nano a promising tool in targeted therapeutic interventions.

Materials and methods
Key resources Continued on next page

Constructs
The cDNA fragment encoding residues 235-275 (TRIP8b nano ) of mouse TRIP8b (splice variant 1a4) was cloned into pET-52b (EMD Millipore) downstream of a Strep (II) tag sequence, while the cDNA fragment encoding residues 521-672 of human HCN2 (HCN2 CNBD) was cloned, in a previous study, into a modified pET-24b downstream of a double His 6 -maltose-binding protein (MBP) (Saponaro et al., 2014). The cDNA encoding full-length human HCN1 channel and mouse TRIP8b (1a4) were cloned into the mammalian expression vector pcDNA 3.1 (Clontech Laboratories), while mouse HCN2 channel and rabbit HCN4 channel were cloned into the mammalian expression vector pCI (Promega). Mutations were generated by site-directed mutagenesis (QuikChange site-directed mutagenesis kit; Agilent Technologies) and confirmed by sequencing.

Preparation of proteins
The HCN2 CNBD WT and mutant proteins, as well as the TRIP8b core and TRIP8b nano proteins (WT and mutants) were produced and purified following the procedure previously described (Saponaro et al., 2014).
Structure calculation of the cAMP-free human HCN2 CNBD in complex with TRIP8b nano and vice versa NMR experiments were acquired on Bruker Avance III 950, 700 and 500 MHz NMR spectrometers equipped with a TXI-cryoprobe at 298 K. The acquired triple resonance NMR experiments for the assignment of backbone resonances of cAMP-free HCN2 CNBD (CNBD hereafter) in complex with TRIP8b nano and vice versa are summarized in Figure 3-source data 1. 15 N, 13 C', 13 C a , 13 C b , and H a chemical shifts were used to derive f and y dihedral angles by TALOS + program (Cornilescu et al., 1999) for both CNBD and TRIP8b nano . For TRIP8b nano , CYANA-2.1 structure calculation (Güntert and Buchner, 2015) was performed using 68 f and y dihedral angles and 40 backbone hydrogen bonds as input. For CNBD, CYANA-2.1 structure calculation was performed using 108 f and y dihedral angles, combined with the NOEs obtained in our previous determination of the cAMP-free form of the CNBD (Saponaro et al., 2014) for those regions not affected by the interaction with TRIP8b nano . The 10 conformers of TRIP8b nano and CNBD with the lowest residual target function values were subjected to restrained energy minimization with AMBER 12.0 (Case, 2012) (http://pyenmr.cerm.unifi.it/access/index/amps-nmr) and used as input in docking calculations.

Docking calculations
Docking calculations were performed with HADDOCK2.2 implemented in the WeNMR/West-Life GRID-enabled web portal (www.wenmr.eu). The docking calculations are driven by ambiguous interaction restraints (AIRs) between all residues involved in the intermolecular interactions (Dominguez et al., 2003). Active residues of the CNBD were defined as the surface exposed residues (at least 50% of solvent accessibility), which show chemical shift perturbation upon TRIP8b nano binding.
Passive residues of CNBD were defined as the residues close in space to active residues and with at least 50% solvent accessibility.
In the case of TRIP8b nano , the conserved stretch E 239 -E 243 , located in helix N, was defined as active region in a first docking calculation, while all the other solvent accessible residues of the peptide were defined as passive. This docking calculation generated several clusters. A post-docking filter step allowed us to select those clusters having an orientation of TRIP8b nano bound to CNBD in agreement with a DEER study on the CNBD -TRIP8b nano interaction (DeBerg et al., 2015). The selected clusters grouped in two classes on the basis of the orientation of helix N of TRIP8b nano (N) relative to CNBD (Figure 3-figure supplement 2. A second docking calculation was subsequently performed introducing also residues E 264 -E 265, located in helix C of TRIP8b nano as active residues. The active residues for CNBD were the same used for the first calculation. For this second HAD-DOCK calculation, 14 clusters were obtained and ranked according to their HADDOCK score. Among them only four clusters showed both an orientation of TRIP8b nano bound to CNBD in agreement with the DEER study (DeBerg et al., 2015) and the involvement of E 239 -E 243 stretch of TRIP8bnano in the binding to CNBD. These clusters were manually analyzed and subjected to a per-cluster re-analysis following the protocol reported in http://www.bonvinlab.org/software/haddock2.2/analysis/#reanal. From this analysis, it resulted that the top-ranking cluster, i.e. the one with the best energetic and scoring functions, has a conformation in agreement with mutagenesis experiments (Figure 3-figure supplement 3). Energy parameters (van der Waals energy, electrostatic energy, desolvation energy, and the penalty energy due to violation of restraints) for this complex model are reported in Figure 3-source data 2.
Both docking calculations were performed using 10 NMR conformers of both the CNBD and the TRIP8b nano structures calculated as described above. In the TRIP8b nano structures the unfolded Nand C-terminal regions were removed, while in the CNBD structures only the unfolded N-terminal region was removed. This is because the C-terminal region of the CNBD is known to comprise residues involved in TRIP8b nano binding (Saponaro et al., 2014). Flexible regions of the proteins were defined based on the active and passive residues plus two preceding and following residues. The residue solvent accessibility was calculated with the program NACCESS (Hubbard and Hornton, 1993). In the initial rigid body docking calculation phase, 5000 structures of the complex were generated, and the best 400 in terms of total intermolecular energy were further submitted to the semiflexible simulated annealing and a final refinement in water. Random removal of the restraints was turned off. The number of flexible refinement steps was increased from the default value of 500/ 500/1000/1000 to 2000/2000/2000/4000. The final 400 structures were then clustered using a cutoff of 5.0 Å of RMSD to take into consideration the smaller size of protein-peptide interface.
TRIP8b nano was added (10 mM) to the pipette solution. cAMP was added at different concentration to the pipette solution depending on the HCN isoform used: 0 mM for HCN1, 5 mM for HCN2 and 1 mM for HCN4.
Whole-cell measurements of HCN channels were performed using the following voltage clamp protocol depending on the HCN isoform measured: for HCN1, holding potential was -30 mV (1 s), with steps from -20 mV to -120 mV (10 mV interval, 3.5 s) and tail currents recorded at -40 mV (3 s); for HCN2, holding potential was -30 mV (1 s), with steps from -40 mV to -130 mV (10 mV interval, 5 s) and tail currents recorded at À40 mV (5 s); for HCN4, holding potential was -30 mV (1 s), steps from -30 mV to -165 mV (15 mV interval, 4.5 s) and tail currents were recorded at À40 mV (5 s). Current voltage relations and activation curves were obtained by the above activation and deactivation protocols and analyzed by the Boltzmann equation, see data analysis.

Isolation and electrophysiology of rabbit sinoatrial node cells
Animal protocols conformed to the guidelines of the care and use of laboratory animals established by Italian and European Directives (D. Lgs n˚2014/26, 2010/63/UE). New Zealand white female rabbits (0.8-1.2 kg) were anesthetized (xylazine 5 mg/Kg, i.m.), and euthanized with an overdose of sodium thiopental (i.v.); hearts were quickly removed, and the SAN region was isolated and cut in small pieces. Single SAN cardiomyocytes were isolated following an enzymatic and mechanical procedure as previously described (DiFrancesco et al., 1986). Following isolation, cells were maintained at 4˚C in Tyrode solution: 140 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl 2 , 1 mM MgCl 2 , 5.5 mM D-glucose, 5 mM HEPES-NaOH (pH 7.4).
For patch clamp experiments cells were placed in a chamber on an inverted microscope and experiments were performed in the whole-cell configuration at 35 ± 0.5˚C. The pipette solution contained: 10 mM NaCl, 130 mM KCl, 1 mM egtazic acid (EGTA), 0.5 mM MgCl 2 , and 5 mM HEPES-KOH buffer (pH 7.2). The I f current was recorded from single cells superfused with Tyrode solution with 1 mM BaCl 2 , and 2 mM MnCl 2 . I f activation curves were obtained using a two-step protocol in which test voltage steps (from À30 to À120 mV, 15 mV interval) were applied from a holding potential of À30 mV and were followed by a step to À125 mV. Test steps had variable durations so as to reach steady -state activation at all voltages. Analysis was performed with the Boltzmann equation (see data analysis).
In current-clamp studies, spontaneous action potentials were recorded from single cells superfused with Tyrode solution, and rate was measured from the interval between successive action potential. When indicated cAMP (1 mM) and/or nanoTRIP8b (10 mM) were added to the pipette solution.
Single SAN cells were isolated by manual agitation in KB solution at 35˚C for 30-50 s. Cellular automaticity was recovered by re-adapting the cells to a physiological extracellular Ca 2+ concentration by addition of a solution containing: 10 mM NaCl, 1.8 mM CaCl 2 and normal Tyrode solution containing BSA (1 mg/ml). The final storage solution contained: 100 mM NaCl, 35 mM KCl, 1.3 mM CaCl 2 , 0.7 mM MgCl 2 , 14 mM L-glutamic acid, (±) 2 mM D-b-OH-butyric acid, 2 mM KH 2 PO 4 , 2 mM taurine, 1 mg/ml BSA, (pH = 7.4). Cells were then stored at room temperature until use. All chemicals were from SIGMA (St Quentin Fallavier, France).
For electrophysiological recording, SAN cells in the storage solution were harvested in special custom-made recording plexiglas chambers with glass bottoms for proper cell attachment and mounted on the stage of an inverted microscope (Olympus IX71) and perfused with normal Tyrode solution. The recording temperature was 36˚C. We used the whole-cell variation of the patch-clamp technique to record cellular ionic currents, by employing a Multiclamp 700B (Axon Instruments Inc., Foster USA) patch clamp amplifier. Recording electrodes were fabricated from borosilicate glass, by employing a WZ DMZ-Universal microelectrode puller (Zeitz-Instruments Vertriebs GmbH, Martinsried, Germany). I f was recorded under standard whole-cell configuration during perfusion of standard Tyrode's containing 2 mM BaCl 2 to block I K1 . Patch-clamp pipettes were filled with an intracellular solution containing: 130 mM KCl, 10 mM NaCl, 1 mM EGTA, 0.5 mM MgCl 2 and 5 mM HEPES (pH 7.2).