Lipid bilayers determine allostery but not intrinsic affinity of cAMP to pacemaker channels Authors: Vinay Idikuda, Susovan Roy Chowdhury, Audrey Chinn, Yongchang Chang, Suhaila Rahman, Qian Ren, Huan Bao, Ziao Fu, Randall H. Goldsmith, Baron Chanda Published: 2026 ABSTRACT The binding of cyclic adenosine monophosphate (cAMP) to hyperpolarization-activated cyclic nucleotide-gated (HCN) ion channels regulates cardiac pacemaking but key aspects of the mechanism of ligand-dependent regulation remain unresolved. Here, we examine the role of the lipid environment by reconstituting purified human HCN channels into lipid nanodiscs and measuring successive cAMP binding to single HCN channels using nanophotonic waveguides. Regardless of nanodisc size or lipid composition, cAMP molecules bind cooperatively to HCN channels in lipid bilayers, unlike channels solubilized in detergents. The affinity of the first ligand remains unchanged across conditions, indicating that the bilayer selectively alters higher-order ligation states. Cryo-EM structures of apo- and holo-HCN channels reveal additional lipid densities that are weak or absent in detergent-solubilized preparations. Together, these findings show that the lipid bilayer is both necessary and sufficient to induce cooperative ligand binding in HCN channels, thereby enhancing their sensitivity to gating stimuli. Introduction HCN, or pacemaker channels, are a distinct subclass of voltage-gated ion channels that are activated by membrane hyperpolarization and modulated by the intracellular second messenger cAMP1–5. By integrating electrical and chemical signals, HCN channels play a critical role in regulating the rhythmic firing of cardiac and neuronal action potentials6,7. Each subunit of a tetrameric HCN channel contains a single cAMP binding site in its intracellular cyclic nucleotide-binding domain (CNBD). The binding of cAMP allosterically modulates channel gating, enhancing macroscopic current and accelerating activation kinetics8. Despite the essential role of cAMP in pacemaking, the mechanisms underlying cAMP-dependent gating of HCN channels remain controversial. Electrophysiology and patch-clamp fluorometry studies suggest that HCN channels function as dimer-of-dimers, with higher binding affinities for the first and third ligation states compared to the unbound or doubly-bound states9–15. In contrast, cAMP binding to isolated CNBDs promotes tetramerization rather than dimerization suggesting a canonical positive cooperativity16. This type of discrepancy is neither surprising nor uncommon for two reasons. First, ensemble measurements cannot uniquely determine the microscopic equilibrium constants in a complex binding reaction17. As a result, very different allosteric schemes can generate essentially indistinguishable macroscopic binding curves, a limitation known as the parameter identifiability problem17,18. Second, most of these measurements were carried out in heterologous expression systems without isolating HCN channels from the surrounding cellular milieu9–15. Under these conditions, the channels are overexpressed to high densities, making it difficult to rule out contributions from other cellular cofactors, including accessory subunits and lipids. Tetramerization studies on purified CNBDs lacking the transmembrane domains do not provide any insight into the role of the lipid membrane16. The problem of parameter identifiability can be addressed by directly measuring the microscopic equilibrium constants associated with each ligation event using single-molecule measurements17,18. However, due to the increased background from free fluorescent ligands, conventional total internal reflection fluorescence (TIRF) microscopes cannot be used to detect individual ligand binding events at concentrations above 10 nM19–21. Zero-mode waveguides (ZMWs) reduce the excitation volume by three orders of magnitude compared to a TIRF microscope, enabling the study of fluorescent ligand binding dynamics at micromolar concentrations22–27. Using such a system, White et al. recently discovered that purified and detergent-solubilized HCN channels bind all four cAMP molecules independently28,29. This observed independent ligand-binding behavior in detergent-solubilized samples raises questions about the roles of lipid membranes and other cofactors present in cellular systems. High-resolution structures of detergent-solubilized HCN channels reveal some lipid densities corresponding to bound lipids, and functional studies demonstrate that specific lipids modulate the function of these channels30,31. In addition, a vast majority of the surrounding lipids, known as solvating lipids, interact transiently with integral membrane proteins32–34. They are either replaced by detergents upon solubilization or not resolved in the high-resolution structures. For instance, cholesterol, which is a key component of metazoan membranes, is not resolved in structures of many membrane proteins. In addition to the lipid membrane, a number of interacting proteins are known to modulate the function of HCN channels35 and it is important to determine whether these non-lipid interactors are necessary for cooperativity. To address these questions, we use single-molecule methods to measure ligand-binding dynamics of isolated HCN1 and HCN4 channels in lipid nanodiscs of varying compositions and compare them with those obtained in detergent-solubilized protein. To determine the effect of lipid membrane on protein structure, we also obtained high-resolution structures of apo- and holo-HCN1 channels. Our single-molecule studies reveal that lipid bilayers do not alter the binding affinity of the first ligand but, unlike detergents, promote the binding of subsequent ligands to the other subunits of the same channel. These findings provide specific insights into how lipid bilayers influence the allosteric regulation of ion channels. Results HCN1 reconstituted in soy lipid nanodiscs exhibits increased ligand binding activity To investigate whether lipid bilayers influence the ligand-binding dynamics of HCN1 channels, we isolated and purified heterologously expressed HCN1 channels using glyco-diosgenin (GDN) detergent. A portion of these purified channels was reconstituted into soy polar lipid nanodiscs by mixing the HCN1 protein with the scaffold protein MSP2N2 and phospholipids in a 1:2.5:200 ratio (Fig. 1A, Table S1). Reconstitution was validated through size exclusion chromatography and SDS-PAGE analysis (Fig. 1B, Fig. S1), with tetrameric assembly confirmed via GFP photobleaching (Fig. S2A–D).Fig. 1Ligand-binding activity in nanodisc-reconstituted and detergent-solubilized HCN1 channels.A Schematic of HCN1 reconstitution into lipid nanodiscs using MSP2N2 as scaffold protein (1:2.5:200 ratio of HCN1, MSP2N2, and soy polar lipids). B (left) Size exclusion chromatography profile of HCN1 purified in detergent (blue) and HCN1 purified in detergent and reconstituted into soy polar lipid nanodiscs with MSP2N2 (black). (right) Coomassie blue stained SDS-PAGE gel showing various SEC fractions of HCN1 at ~75 kDa before reconstitution (lanes 2–4) and both HCN1 and MSP bands after reconstitution (lane 5). SEC profiles of detergent and nanodisc samples were reproducible from multiple purifications (n ≥ 3). C Cartoon (not to scale) of single-molecule pulldown of HCN1 onto a ZMW. ZMWs passivated with mPEG silane doped with biotin-PEG-silane and coated with streptavidin. Biotinylated GFP pulled down GFP-tagged HCN1. Representative FT traces of 100 nM fcAMP (D) and 250 nM fcAMP (E) binding to HCN1 in a detergent micelle. Dashed lines, ligated state of the channel. F Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN1 in detergent micelles. G, H Representative FT traces of 100 nM and 250 nM fcAMP binding to HCN1 in soy polar lipid nanodiscs. I Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN1 in soy polar lipid nanodiscs. The idealized fluorescence time trajectories were represented using dynamic heatmaps. Each row corresponds to an individual molecule. The x-axis corresponds to time. The width of the column is dynamic and depends on how long each molecule spends in a particular state. The color of the heatmap indicates the ligation state according to the color scale. J, K Ratios of observed state occupancy (P(k)obs) and expected (P(k)exp) values of liganded state occupancies from binomial fits across all trajectories for HCN1 in soy polar lipid nanodiscs. J or detergent micelles K at different fcAMP concentrations. (SPL: n = 60, 233 and 195 for 50 nM, 100 nM and 250 nM fcAMP respectively; Detergent: n = 183, 203 and 91 for 50 nM, 100 nM and 250 nM fcAMP respectively. ‘n’ represents the total number of molecules. See Tables S2 and S3). Red dashed line represents concordance between observed and expected values; error bars (red) represent standard error. Data points are offset for clarity; each gray panel represents a different ligation state (U: Unbound, B1: 1st bound, B2: 2nd bound, B3: 3rd bound, B4: 4th bound states). L Reconstructed binding curve for HCN1 in soy polar lipid nanodiscs (black) and detergent (red) generated using Adair’s equation for four binding sites39. Figure 1A, C were created in BioRender. Chanda, B. (2023) BioRender.com/n03m505. Ligand-binding activity in nanodisc-reconstituted and detergent-solubilized HCN1 channels. A Schematic of HCN1 reconstitution into lipid nanodiscs using MSP2N2 as scaffold protein (1:2.5:200 ratio of HCN1, MSP2N2, and soy polar lipids). B (left) Size exclusion chromatography profile of HCN1 purified in detergent (blue) and HCN1 purified in detergent and reconstituted into soy polar lipid nanodiscs with MSP2N2 (black). (right) Coomassie blue stained SDS-PAGE gel showing various SEC fractions of HCN1 at ~75 kDa before reconstitution (lanes 2–4) and both HCN1 and MSP bands after reconstitution (lane 5). SEC profiles of detergent and nanodisc samples were reproducible from multiple purifications (n ≥ 3). C Cartoon (not to scale) of single-molecule pulldown of HCN1 onto a ZMW. ZMWs passivated with mPEG silane doped with biotin-PEG-silane and coated with streptavidin. Biotinylated GFP pulled down GFP-tagged HCN1. Representative FT traces of 100 nM fcAMP (D) and 250 nM fcAMP (E) binding to HCN1 in a detergent micelle. Dashed lines, ligated state of the channel. F Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN1 in detergent micelles. G, H Representative FT traces of 100 nM and 250 nM fcAMP binding to HCN1 in soy polar lipid nanodiscs. I Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN1 in soy polar lipid nanodiscs. The idealized fluorescence time trajectories were represented using dynamic heatmaps. Each row corresponds to an individual molecule. The x-axis corresponds to time. The width of the column is dynamic and depends on how long each molecule spends in a particular state. The color of the heatmap indicates the ligation state according to the color scale. J, K Ratios of observed state occupancy (P(k)obs) and expected (P(k)exp) values of liganded state occupancies from binomial fits across all trajectories for HCN1 in soy polar lipid nanodiscs. J or detergent micelles K at different fcAMP concentrations. (SPL: n = 60, 233 and 195 for 50 nM, 100 nM and 250 nM fcAMP respectively; Detergent: n = 183, 203 and 91 for 50 nM, 100 nM and 250 nM fcAMP respectively. ‘n’ represents the total number of molecules. See Tables S2 and S3). Red dashed line represents concordance between observed and expected values; error bars (red) represent standard error. Data points are offset for clarity; each gray panel represents a different ligation state (U: Unbound, B1: 1st bound, B2: 2nd bound, B3: 3rd bound, B4: 4th bound states). L Reconstructed binding curve for HCN1 in soy polar lipid nanodiscs (black) and detergent (red) generated using Adair’s equation for four binding sites39. Figure 1A, C were created in BioRender. Chanda, B. (2023) BioRender.com/n03m505. We measured ligand binding by immobilizing purified channels—either in GDN micelles or lipid nanodiscs—onto ZMWs and introducing the fluorescent ligand fcAMP (DY-547-cAMP) (Fig. 1C). To ensure single-occupancy of ZMWs, protein concentration was adjusted based on Poisson dilution statistics as previously described36. Ligand binding events were measured across a range of fcAMP concentrations using micromirror TIRF microscopy37. Fluorescence intensity-time (FT) traces were analyzed using AutoDISC29,38 to identify bound and unbound states (Fig. 1D–I, Fig. S3). Control experiments confirmed the absence of nonspecific binding, as evidenced by no detectable ligand binding in the absence of HCN channels and by the competitive displacement of fcAMP with excess non-fluorescent cAMP at the end of each experiment (Fig. S4). Our results revealed that channels in nanodiscs exhibited a higher frequency of multi-ligand bound states (three or four fcAMP molecules bound) compared to channels in detergent micelles, particularly at lower fcAMP concentrations (Fig. 1D–I). These dynamic heatmaps reveal, across molecules and concentrations, the prevalence and persistence of higher‑order ligation states (3rd and 4th bound states) and molecule‑to‑molecule heterogeneity that are not readily apparent from single FT traces. To quantify the higher frequency of multiligand states, we normalized the time spent by fcAMP in each ligation state (state occupancy) to the expected value from a binomial distribution under the assumption of independent and identical binding sites. When ligand binding is independent, the ratio of observed to expected occupancies is 1. For HCN1 channels in nanodiscs, occupancy probabilities deviated significantly from binomial predictions, especially in high-occupancy states across all concentrations tested (Fig. 1J, Fig. S7B, Table S4). For instance, at 50 nM fcAMP, the fourth bound state showed a deviation exceeding 25-fold. This effect diminished at higher fcAMP concentrations, only 2–5-fold at saturating concentrations. In contrast, channels in detergent micelles displayed binding distributions consistent with independent binding (Fig. 1K, Fig. S7A, Table S5), as shown previously28. Using a sequential binding model, we calculated bound probabilities for each ligation state at various concentrations and generated ensemble binding curves from the single-molecule data. A bound probability of 1.0 indicates complete occupancy of all binding sites by fcAMP. From these data, the microscopic rate constants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{\mbox{on}}}$$\end{document}kon and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{\mbox{off}}}$$\end{document}koff) for each sequential binding step were obtained, and the macroscopic binding curves were reconstructed by using Adair equation39 (see “Methods”). HCN1 channels in nanodiscs showed a three-fold increase in apparent macroscopic ligand binding affinity compared to those in detergent micelles (Fig. 1L, black vs. red). cAMP binding exhibits cooperative behavior in nanodiscs of different sizes and varying lipid compositions We wondered whether the cooperativity observed in HCN1 channels reconstituted in soy polar lipids was influenced by specific lipid species. Phospholipids are key regulators of many ion channels in biological membranes40–45, and recent studies31 show that negatively charged phospholipids modulate HCN channel activity. To test this, we compared two lipid conditions: a native lipid composition (porcine brain polar lipid extract) and a simple lipid system without any added cholesterol (POPC/POPE in a 3:1 ratio). HCN channels exhibited robust binding activity at all concentrations in both brain polar lipid and POPC/POPE nanodiscs (Fig. 2A, B, Fig. S5). Using idealized FT traces, we determined the state occupancies for each ligation state across the concentration range and normalized them to the expected values based on an independent binding model. In both brain polar lipids (Fig. 2F and Fig. S7C, Table S6) and POPC/POPE (Fig. 2G, Fig. S8A, Table S7), the HCN1 state occupancies deviated from the expected values, particularly at higher ligation states, mirroring trends observed in soy polar lipids that show non-independent binding.Fig. 2HCN1 channels in lipid nanodiscs exhibit cooperative ligand binding.Dynamic heatmap of three independent FT traces of 50 nM, 100 nM, and 250 nM fcAMP binding to HCN1 reconstituted in porcine brain polar lipid nanodiscs (A) and POPC/POPE lipid nanodiscs (B). HCN1 reconstituted in 11 nm (spNW11) (C) and 30 nm (spNW30) (D) circular nanodiscs of brain polar lipids. E Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN1 purified in native HEK cell nanodiscs. Ratios of P(k)obs and P(k)exp for liganded state occupancies from binomial fits across all trajectories for (F) HCN1 in brain polar lipid nanodiscs (BPL: n = 107, 173 and 59 for 50 nM, 100 nM and 250 nM fcAMP respectively, Table S8), G 3:1 ratio of POPC/POPE (PC/PE: n = 80, 60 and 85 for 50 nM, 100 nM and 250 nM fcAMP respectively, Table S9), H HCN1 reconstituted in 11 nm (spNW11: n = 143, 174 and 119 for 100 nM, 250 nM and 750 nM fcAMP respectively, Table S10), I 30 nm (spNW30 n = 159, 150 and 46 for 100 nM, 250 nM and 750 nM fcAMP respectively, table S11) circular nanodiscs of brain polar lipids, J HCN1 purified in native HEK cell nanodiscs (Native ND: n = 67 and 91 for 100 nM and 250 nM fcAMP respectively, Table S12) at different fcAMP concentrations. Error bars represent the standard error. ‘n’ represents the total number of molecules. Labeling and organization of F–J is similar to the J, K panels in Fig. 1. HCN1 channels in lipid nanodiscs exhibit cooperative ligand binding. Dynamic heatmap of three independent FT traces of 50 nM, 100 nM, and 250 nM fcAMP binding to HCN1 reconstituted in porcine brain polar lipid nanodiscs (A) and POPC/POPE lipid nanodiscs (B). HCN1 reconstituted in 11 nm (spNW11) (C) and 30 nm (spNW30) (D) circular nanodiscs of brain polar lipids. E Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN1 purified in native HEK cell nanodiscs. Ratios of P(k)obs and P(k)exp for liganded state occupancies from binomial fits across all trajectories for (F) HCN1 in brain polar lipid nanodiscs (BPL: n = 107, 173 and 59 for 50 nM, 100 nM and 250 nM fcAMP respectively, Table S8), G 3:1 ratio of POPC/POPE (PC/PE: n = 80, 60 and 85 for 50 nM, 100 nM and 250 nM fcAMP respectively, Table S9), H HCN1 reconstituted in 11 nm (spNW11: n = 143, 174 and 119 for 100 nM, 250 nM and 750 nM fcAMP respectively, Table S10), I 30 nm (spNW30 n = 159, 150 and 46 for 100 nM, 250 nM and 750 nM fcAMP respectively, table S11) circular nanodiscs of brain polar lipids, J HCN1 purified in native HEK cell nanodiscs (Native ND: n = 67 and 91 for 100 nM and 250 nM fcAMP respectively, Table S12) at different fcAMP concentrations. Error bars represent the standard error. ‘n’ represents the total number of molecules. Labeling and organization of F–J is similar to the J, K panels in Fig. 1. Among the three lipid compositions tested, the highest occupancy of three- and four-ligand-bound states was observed in the presence of soy-polar lipids. Since soy lipid membranes are enriched with anionic lipids (e.g., PI 18.4% w/w and PA 6.9% w/w compared to PI 4% w/w and PA 0.8% w/w in brain polar lipids) relative to brain lipids and neutral PC: PE lipids, our data suggest that anionic lipids further enhance the cooperative network that drives ligand binding in these channels. However, note that anionic lipids are not essential, as channels reconstituted in PC:PE also show cooperative binding unlike detergent samples. These findings are intriguing and warrant further investigation, especially given recent studies indicating that anionic lipids influence the state-dependent activity of pacemaker ion channels34,46. These findings also shed light on the role of cholesterol, a major component of mammalian membranes. In our experiments, cholesterol hemisuccinate (CHS) was present throughout solubilization and purification, so the detergent-solubilized sample contained some cholesterol. For nanodisc samples, this detergent-purified protein was used without additional cholesterol supplementation, so the nanodisc-reconstituted channels contained either similar or lower levels of CHS than in the detergent condition. Nevertheless, cAMP binding cooperativity was observed only in nanodiscs, suggesting that cholesterol is not required for cooperative ligand binding. Recent findings indicate that the structure of certain membrane proteins is influenced by nanodisc size. While MSP2N2 nanodiscs have an average diameter of 15 nm, their size distribution is broad and can be affected by the embedded protein. To examine whether nanodisc size impacts the cooperativity of ligand binding, we utilized circular nanodiscs with uniform sizes across a wide temperature range and minimal dependence on the reconstituted protein. Spy-catcher and spy-tag MSPs were purified to generate circular nanodiscs of 11 nm and 30 nm using soy polar lipids (Fig. S1C and F–I). Dynamic heatmap representations of FT traces reveal that both 11-nm and 30-nm circular nanodiscs exhibit robust ligand-binding activity (Fig. 2C, D and Fig. S6). Even at the lowest tested concentration (50 nM fcAMP), higher-order ligation states were observed, contrasting with detergent-solubilized samples. Plots comparing observed and expected state occupancies for unbound and various bound states deviated from the binomial distribution, indicating non-independent binding (Fig. 2H, I, Fig. S8B, C, and Tables S13, S14). Reconstituting membrane proteins in nanodiscs requires solubilizing both the target protein and lipids in detergent—in this case, LMNG and CHAPS, respectively. Although we use BioBeads to remove most, if not all, of the detergent, we wondered whether exposure to detergent during purification might irreversibly alter the channel gating energetics. The other possibility is that the detergent is not fully removed, and some remains in the nanodisc preparations. To avoid these possibilities, we employed detergent-free nanodiscs (DeFrNDs), which circumvent the need for detergent-mediated solubilization and reconstitution steps47. These nanodiscs are a distinct class of engineered membrane scaffold peptides derived from apolipoprotein A that facilitate the direct extraction of cell membranes as nanoscale membrane discoids. The HCN1-encapsulated discoids were further isolated by affinity purification using a Twin-strep tag. HCN1 channels extracted into DeFrNDs directly from HEK cells were immobilized using streptavidin inside the ZMW wells. These HCN1 in DeFrNDs also exhibited robust ligand-binding activity, as evidenced by a high binding activity (Fig. 2E) compared to the detergent-solubilized samples, and we also observed that higher ligation states were much more favored compared to the independent binding model (Fig. 2J, Fig. S14, and Table S15). Lipid bilayers but not detergents also influence binding cooperativity in the cardiac pacemaker isoform To determine whether this effect is generalizable to other HCN channels, we expressed and purified a C-terminal truncated, GFP-tagged human ortholog of HCN4 (Fig. S1J–L), which is approximately four times more sensitive to cAMP than HCN148. These channels were solubilized in GDN (Glyco-diosgenin), with a portion reconstituted in soy polar lipid nanodiscs using MSP2N2 as the MSP. Dynamic heatmap representations of FT traces showed that HCN4 ligand-binding activity in detergent micelles was lower than in soy polar lipids (Fig. 3A, B, Fig. S12). In detergent micelles, the ratios of observed to expected occupancies for various ligation states were close to one, consistent with independent ligand-binding events (Fig. 3C). By contrast, observed occupancies in soy polar lipids deviated from binomial distribution (Fig. 3D) where the higher ligation states are favored, resembling HCN1 channels reconstituted in nanodiscs (Fig. S13A, B and Tables S18, S19).Fig. 3Ligand-binding cooperativity in HCN4 isoform.Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN4 in GDN detergent micelles (A) and soy polar lipid nanodiscs (B). Ratios of P(k)obs and P(k)exp for liganded state occupancies from binomial fits across all trajectories for HCN4 reconstituted in (C) detergent micelles (Detergent: n = 251 and 136 for 100 nM and 250 nM fcAMP respectively, Table S16) and (D) soy polar lipid nanodiscs at (SPL: n = 158 and 102 for 100 nM and 250 nM fcAMP respectively, Table S17) 100 nM and 250 nM fcAMP. Error bars represent the standard error. Labeling and organization of C, D is similar to the J, K panels in Fig. 1. Ligand-binding cooperativity in HCN4 isoform. Dynamic heatmap of three independent FT traces of 100 nM and 250 nM fcAMP binding to HCN4 in GDN detergent micelles (A) and soy polar lipid nanodiscs (B). Ratios of P(k)obs and P(k)exp for liganded state occupancies from binomial fits across all trajectories for HCN4 reconstituted in (C) detergent micelles (Detergent: n = 251 and 136 for 100 nM and 250 nM fcAMP respectively, Table S16) and (D) soy polar lipid nanodiscs at (SPL: n = 158 and 102 for 100 nM and 250 nM fcAMP respectively, Table S17) 100 nM and 250 nM fcAMP. Error bars represent the standard error. Labeling and organization of C, D is similar to the J, K panels in Fig. 1. Lipid bilayers promote higher-order ligation states but not intrinsic affinity To determine whether the differences in the ligand binding dynamics between detergents and bilayer mimetics arise due to changes in the intrinsic agonist binding affinity, we analyzed unbound-bound-unbound (UBU) dwell time distributions for singly bound states, avoiding stacked events which correspond to multiple ligand binding (Fig. 4A)28. UBU dwell times correspond to the binding of the first ligand, and the analysis of this bound dwell time gives us an estimate of initial ligand binding affinity without confounding effects of allosteric regulation.Fig. 4Lipid bilayers alter stepwise association constants for higher-order ligation steps.A Typical FT trace (blue) showing ligand binding dynamics superimposed with idealized trace (black). The events which correspond to unbound-bound-unbound dwell time sequences are marked with tan shading. Dwell time distributions of isolated UBU events for HCN1 in (B) brain polar lipid nanodiscs (n = 173) and C POPC/POPE nanodiscs (n = 60) at (D) and native HEK cell nanodiscs (n = 67) at 100 nM fcAMP. Distributions overlaid with maximum likelihood estimates for monoexponential (blue dashed) and bi-exponential (red) distributions. Error bars denote standard deviation. Individual time constants and amplitudes in Tables S20, S21, S22. ‘n’ represents the number of molecules under observation. Error bars represent standard error and the center of the error bars represents the mean. E Amplitude weighted mean dwell time of the first bound state at various fcAMP concentrations obtained by fitting UBU dwell times with a biexponential decay. Error bars represent the standard error. F Schematic of sequential model for ligand binding. White squares unbound (U); squares with gray circles, bound (B) ligand. G Association constants (Ka) for each ligation state of HCN1 in Soy polar lipid nanodiscs (n = 945), POPC/POPE nanodiscs (n = 372), brain polar lipid nanodiscs (n = 898), and detergent micelles (n = 815) calculated using global fitting of the unconstrained sequential binding model across all concentrations from ‘n’ number of molecules. H Ka values for all tested conditions showing positive cooperativity for HCN1 in both 11 nm and 30 nm nanodiscs (n = 743 and 461, respectively), and independent binding in detergent micelles (n = 815). I Ka values for all tested conditions showing positive cooperativity for HCN1 in native lipid nanodiscs (n = 671) and independent binding in detergent micelles (n = 815). (J) Ka values for all tested conditions showing positive cooperativity for HCN4 in soy lipid nanodiscs (n = 498). and independent binding in detergent micelles (n = 517). Lipid bilayers alter stepwise association constants for higher-order ligation steps. A Typical FT trace (blue) showing ligand binding dynamics superimposed with idealized trace (black). The events which correspond to unbound-bound-unbound dwell time sequences are marked with tan shading. Dwell time distributions of isolated UBU events for HCN1 in (B) brain polar lipid nanodiscs (n = 173) and C POPC/POPE nanodiscs (n = 60) at (D) and native HEK cell nanodiscs (n = 67) at 100 nM fcAMP. Distributions overlaid with maximum likelihood estimates for monoexponential (blue dashed) and bi-exponential (red) distributions. Error bars denote standard deviation. Individual time constants and amplitudes in Tables S20, S21, S22. ‘n’ represents the number of molecules under observation. Error bars represent standard error and the center of the error bars represents the mean. E Amplitude weighted mean dwell time of the first bound state at various fcAMP concentrations obtained by fitting UBU dwell times with a biexponential decay. Error bars represent the standard error. F Schematic of sequential model for ligand binding. White squares unbound (U); squares with gray circles, bound (B) ligand. G Association constants (Ka) for each ligation state of HCN1 in Soy polar lipid nanodiscs (n = 945), POPC/POPE nanodiscs (n = 372), brain polar lipid nanodiscs (n = 898), and detergent micelles (n = 815) calculated using global fitting of the unconstrained sequential binding model across all concentrations from ‘n’ number of molecules. H Ka values for all tested conditions showing positive cooperativity for HCN1 in both 11 nm and 30 nm nanodiscs (n = 743 and 461, respectively), and independent binding in detergent micelles (n = 815). I Ka values for all tested conditions showing positive cooperativity for HCN1 in native lipid nanodiscs (n = 671) and independent binding in detergent micelles (n = 815). (J) Ka values for all tested conditions showing positive cooperativity for HCN4 in soy lipid nanodiscs (n = 498). and independent binding in detergent micelles (n = 517). The maximum likelihood estimations of UBU dwell time distributions for HCN1 were well described by two exponentials at all fcAMP concentrations (Fig. 4B–D, Fig. S9A, B, Fig. S10A–D, and Tables S20–26). This revealed that each subunit undergoes isomerization to a secondary conformation or flip state28. Importantly, these biexponential fits were consistent across detergent and nanodisc samples, and the weighted time constant was similar in all conditions (Fig. 4E, Fig. S10E and Tables S20–S26). The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{on}}$$\end{document}kon values for the first fcAMP binding derived from BUB dwell-time analysis are statistically similar across the different HCN1 conditions (Fig. S11B). These results suggest that the lipid environment does not alter the intrinsic affinity of fcAMP for its binding site or the time spent in the initial bound state. Instead, lipid bilayer influences downstream allosteric transitions and the energetics of higher-order ligand-binding occupancies. Next, we calculated the bound open probabilities and estimated the microscopic rate constants from the idealized FT trajectories by fitting them to an unconstrained sequential binding model as described previously (see Methods) (Fig. 4F). A comparison of association constants reveals that, unlike detergent-solubilized HCN1 channels (Fig. 4G, Table S29), the ligand-binding affinity for HCN1 channels reconstituted in bilayer mimetics with different lipid compositions increases as the ligand occupancy increases (Fig. 4G, Tables S30, S31, S32). This type of positive cooperativity was observed even in nanodiscs of various sizes (Fig. 4H, Tables S33 and S34) and in HCN1 channels extracted directly from membranes without detergent solubilization (Fig. 4I, Table S35). The cardiac pacemaker ion channel, HCN4, also showed similar differences between detergent and nanodisc reconstituted ion channels (Fig. 4J, Table S36 and S37, Fig. S13). Structural analysis of detergent and nanodisc reconstituted HCN channels Recent studies on prokaryotic pentameric ion channels have shown that the structures of channels reconstituted into nanodiscs differ significantly from those in detergent49–51. To determine whether reconstitution into lipid bilayers influences the structure of HCN channels, we used single particle cryo-EM to solve structures of human HCN1 in soy phospholipid nanodiscs in both the apo and cAMP-bound states at resolutions of 2.5 and 2.6 Å, respectively (Fig. 5A, B, E, F and Fig. S15–19). Superposition of our nanodisc models with apo (PDB: 5U6O) and cAMP-bound (PDB: 5U6P) models in detergent showed that in both states, the nanodisc structures were highly similar to the detergent structures52,53 (Fig. 5C, G), with root mean square deviations of less than 1 Å in most regions (Figs. S19–20). Minor deviations can be observed in flexible loop regions due to the lower resolution of these regions and are unlikely to account for the observed functional differences. In addition, our structures are similar to HCN1 structures in PC:PE:PS nanodiscs as reported by Kim et al.53. Thus, we conclude that protein structures of unbound and bound HCN1 channels are not altered upon reconstitution into lipid membranes. It is still possible that intermediate state structures with different ligand occupancies are modified by the lipid environment and that may account for some of the observed differences.Fig. 5Cryo-EM structures of apo and cAMP-bound hHCN1 in soy polar lipid nanodiscs.A Side view of apo hHCN1 cryo-EM map. B Top view of apo hHCN1 cryo-EM map. C Superimposed models of apo hHCN1 in nanodiscs (blue) and detergent (PDB: 5U6O, orange). D Superimposed lipid densities from apo hHCN1 in nanodiscs (blue) and detergent (EMD-8511, orange) cryo-EM structures. E Side view of cAMP-bound hHCN1 cryo-EM map. F Top view of cAMP-bound hHCN1 cryo-EM map. G Superimposed models of cAMP-bound hHCN1 in nanodiscs (purple) and detergent (PDB: 5U6P, cyan). H Superimposed lipid densities from cAMP-bound hHCN1 in nanodiscs (purple) and detergent (EMD-8512, cyan). Cryo-EM structures of apo and cAMP-bound hHCN1 in soy polar lipid nanodiscs. A Side view of apo hHCN1 cryo-EM map. B Top view of apo hHCN1 cryo-EM map. C Superimposed models of apo hHCN1 in nanodiscs (blue) and detergent (PDB: 5U6O, orange). D Superimposed lipid densities from apo hHCN1 in nanodiscs (blue) and detergent (EMD-8511, orange) cryo-EM structures. E Side view of cAMP-bound hHCN1 cryo-EM map. F Top view of cAMP-bound hHCN1 cryo-EM map. G Superimposed models of cAMP-bound hHCN1 in nanodiscs (purple) and detergent (PDB: 5U6P, cyan). H Superimposed lipid densities from cAMP-bound hHCN1 in nanodiscs (purple) and detergent (EMD-8512, cyan). Another way the energetics of ligand binding can be influenced is through changes in the interaction between the HCN channel and its surrounding molecular environment. These changes can result from bound lipids that form stable interactions with the protein’s transmembrane region, or from solvating lipids, which are dynamic but still influence the energetics. Strikingly, the largest difference between detergent and nanodisc structures is observed in the nonprotein densities surrounding the transmembrane domain (TMD). Maps for both the apo- and cAMP-bound structures showed strong lipid densities in the TMD, many of which are absent or weaker in maps of detergent-solubilized HCN1 (Fig. 5D, H, Supplementary Movie 1 and Supplementary Movie 2). To enable proper comparison given the difference in resolution between the two maps, we low-pass filtered the nanodisc map to 3.5 Å resolution to match the 3.51 Å resolution of the detergent map (Fig. S21). Even at comparable resolution, the additional non-protein densities remain clearly visible in the nanodisc reconstructions. Furthermore, when we compare the 3.16 Å structure of HCN1 in mixed micelles containing both lipids and detergents, we also see more non-protein densities than in the detergent-only HCN structures. Collectively, these data suggest that bound lipids are present in nanodisc structures but absent in the detergent structures. These additional bound lipids may interact differently with various ligation states, thereby altering the energetics of gating transitions and leading to more cooperative binding. Discussion In this study, we investigate the role of lipid membranes in agonist binding to ion channels by reconstituting HCN1 and HCN4 channels into nanodiscs containing either purified lipids or native membranes, and comparing them with detergent-solubilized channels. To resolve the nature of ligand-binding cooperativity11,12,28,54, we directly measured the microscopic equilibrium constants associated with each of the four cAMP binding events using single-molecule methods. Our studies support three main conclusions. First, ligand-binding cooperativity is not an intrinsic property of the HCN channel but instead depends on the membrane environment. Second, lipid bilayers are both necessary and sufficient to make the ligand-binding process cooperative. Notably, this behavior is observed across nanodiscs of various sizes and different lipid compositions, including in bilayers composed of simple PC:PE lipids. Third, lipid membranes specifically alter transitions associated with higher-order ligation states but not the binding of the first ligand. Although the mechanism by which lipid bilayers make ligand binding to the cytosolic domains of an ion channel cooperative remains to be determined, our structural studies offer some clues. Compared with the detergent-solubilized channel structures, both apo- and holo-HCN1 structures in nanodiscs show numerous additional lipid densities absent in detergent. The hallmark of positive cooperativity is the suppression of intermediate states. One possibility is that these additional lipid molecules may facilitate cooperative transitions by destabilizing the intermediate states in which these interactions are disrupted. Nevertheless, further investigations will be needed to establish whether these bound lipids play a causal role in ligand-binding cooperativity. Cholesterol constitutes a major fraction of the membrane lipids in mammalian cells; however, it does not appear to be a key determinant for ligand-binding cooperativity. Cholesterol was present during detergent solubilization but was not included during reconstitution. We also did not observe any densities corresponding to bound cholesterol in our structures. Thus, cholesterol cannot account for the observed increase in cooperative binding in nanodisc samples. Although HCN1 in PC: PE nanodiscs lacking PIP2 or anionic phospholipids also exhibit cooperative binding, our measurements in membranes enriched in anionic lipids suggest that these lipids may further modulate the strength of the cooperative network. This would be consistent with their role in modulating HCN channel function30,31,55 but detailed studies with more defined preparations are needed to understand the mechanisms of this modulation. Thus, our studies do not rule out the possibility that lipids and other proteinaceous cofactors56–58 that modulate HCN channel function may further regulate ligand-binding cooperativity. Although both lipid bilayers and detergent micelles are widely used in biophysical studies, they differ in many ways. For instance, the lateral pressure profile of a lipid bilayer is likely to be very different than that of a detergent micelle59. Similarly, the thickness of the hydrophobic section of a lipid bilayer is more well-defined compared to a micelle. Other mesoscopic properties, such as membrane rigidity, may also contribute to the observed differences, but unless we can systematically measure and control these parameters, it will be difficult to tease out their contributions to protein allostery. We should note that a recent study reached a similar conclusion regarding cAMP binding to HCN2 channels expressed in cells54. Given that those measurements were carried out in unpurified preparations, these authors conclude that the observed cooperative ligand binding is due to the surrounding molecular environment rather than an intrinsic property of the channel. Our studies under controlled conditions allow us to specifically establish that this difference in ligand-binding cooperativity is due to the surrounding lipid membrane rather than to any cofactors or accessory subunits that typically interact with HCN channels in more native preparations. Another difference in our study is that the use of nanophotonic waveguides enables us to resolve successive ligand-binding events over a broader concentration range with a much higher signal-to-noise ratio than confocal or TIRF microscopy19–21. Our observation that lipid bilayers directly promote intersubunit cooperativity in integral membrane proteins is consistent with some of the emerging evidence in the field. For example, single-molecule study on β2-adrenergic receptors show that receptors reconstituted in nanodiscs transition more slowly between active and inactive states compared to detergent-solubilized receptors60,61. Similarly, 19F NMR studies on human α2-adrenergic receptors demonstrate that the mesoscopic properties of the membrane, rather than the lipid composition itself, influence the distribution of conformational states62. In addition, lipid bilayers have been shown to strengthen pairwise interaction networks within the intramembrane protease GlpG63. Our findings further suggest that these membrane-mediated effects extend beyond monomeric proteins to multi-subunit assemblies and even to extramembrane domains. The ability of lipid bilayers to promote cooperative binding may have important physiological implications. Positive cooperativity amplifies small changes in signaling inputs while limiting the dynamic range. For HCN channels, enhancing cooperativity without affecting intrinsic cAMP affinity allows lipid membranes to modulate signal amplification in pacemaker cells while maintaining responsiveness to physiologically relevant cAMP concentrations. Future studies may determine whether changes in lipid composition associated with aging, metabolic alterations, or disease modify or disrupt the integration of voltage- and cAMP-dependent signaling pathways. Methods Constructs used to express proteins used in single-molecule imaging The human ortholog of HCN4 (full length) was synthesized in the pTwist vector (Twist Bioscience) and was subcloned into the pEG BacMam vector71. From the full-length hHCN4 construct, amino acids 782-1062 (total 283 amino acids) were deleted for biochemical stability, N-terminus was tagged with eGFP followed by a 3 C protease cleavage site. A twin-strep tag was introduced in the N-terminus for affinity purification to generate twinstrep-GFP-hHCN4Δ282C, this construct is HCN4SM but is referred as HCN4 from here on in the manuscript. The construct was verified by Sanger sequencing. HCN1SM was the same construct used in a previous study referred as HCN1 in this manuscript 6. Circular nanodiscs were made from spNW11 and spNW30 constructs in pET28 (generously gifted by Huan Bao laboratory), and purification of spNW11 and spNW30 was done as described in the original work64. Cell culture and baculovirus generation HCN1 and HCN4 constructs described above were used to generate Baculovirus. Both constructs in the pEG vector were transformed into chemically competent DH10Bac E.coli cells (Invitrogen) as per the manufacturer’s instructions to generate bacmid DNA. One microgram of the resultant bacmid was then transfected into 1 × 106 cells sf9 cells (Thermo Fisher Scientific) using Cellfectin-II (Invitrogen) reagent, incubated for 72–96 h to generate the P1 virus. The supernatant containing the P1 virus was collected 96 h post-transfection, sterile filtered, and stored at 4 C. The resultant P1 virus was further amplified by infecting sf9 cells at a dilution of 1:1000 to generate the P2 virus. The P2 virus was collected 96 hours post-transduction, sterile filtered, supplemented with 2.5% FBS (Sigma), and stored at 4 °C until further use. This P2 Baculovirus was used for expressing the HCN1-SM and HCN4-SM in free-style HEK suspension cells (Thermo Fisher Scientific). An 800 ml suspension of free-style HEK cells was grown to a density of 2.5 × 106 cells/ml for viral transduction. The cells were transduced with 4% respective P2 baculovirus. Ten to 12 h post-transduction, 10 mM sodium butyrate (Sigma Aldrich) was added to boost the channel expression, and the flask was transferred to 30 °C for another 48 h. The culture was harvested 60 h post-transduction by centrifugation at 300 g for 20 min the resultant cell pellet was flash-frozen in liquid nitrogen and stored at −80 °C until further use. Purification of HCN channels for single-molecule experiments Protein purification for the HCN1 detergent sample was done as reported previously28. For the nanodisc reconstitution, the same protocol was adopted with a few changes to facilitate reconstitution as described here. Frozen cell pellet corresponding to 200 ml cell culture of either HCN1 or HCN4 was thawed on ice and lysed with 50 mL hypotonic lysis buffer (20 mM KCl, 10 mM Tris, protease inhibitor cocktail (Sigma Aldrich, P8340), pH 8.0) and sonicated at 4 °C 5 times for 10 s each. The lysate was spun at 50,000 × g for 45 min to pellet the membrane fraction. The resultant membrane fraction was homogenized on ice with 20 ml 2X solubilization buffer (600 mM KCl, 80 mM Tris, 4 mM DTT, 20% glycerol, protease inhibitor cocktail (P8340, Sigma Aldrich)). After homogenization, 20 ml of 2X detergent solution– 20 mM Lauryl Maltose Neopentyl Glycol (LMNG, Anatrace) and 8 mM Cholesteryl Hemi Succinate (CHS, Anatrace) was added to solubilize the membrane for 1.5 h at 4 °C. Later, the solubilized membranes were spun down at 50,000 × g for 45 min and the supernatant was incubated with pre-equilibrated Strep-Tactin Sepharose (IBA life sciences, 2-1201-025) for 2.5 h. Then, the resin was packed onto a column and washed with 5 column volumes of wash buffer - 300 mM KCl, 20 mM Tris, 2 mM DTT, 5% glycerol, 0.5 mM LMNG, 0.1 mM CHS, (either 164.5 μM Soy Polar Lipids/ 197.5 μM Brain polar lipids or 182 μM, 3:1 POPC/POPE), pH = 8.0). All the lipids were purchased from Avanti Polar Lipids as 25 mg/ml solution in chloroform. The protein was eluted in 5 mL elution buffer (wash buffer + 10 mM d-desthiobiotin (Sigma)) as 1 ml aliquots. The eluted protein concentration was measured using a Denovix spectrophotometer and then the protein sample was concentrated to 4 mg/mL using a concentrator (cut-off: 100 kDa, Amicon Ultra, Millipore). Nanodisc reconstitution for single-molecule experiments The day before reconstitution, lipids in chloroform (Avanti Polar Lipids) were dried under argon and then desiccated overnight under a vacuum at 4 °C. On the day of reconstitution, lipids were dissolved in reconstitution buffer (500 mM NaCl, 50 mM HEPES, 33 mM CHAPS, pH 7.4) to a concentration of 10 mg/ml and sonicated for 30 min immediately before use. Reconstitution was done by mixing purified protein, membrane scaffolding protein (MSP), and appropriate lipid species at corresponding molar ratios (Table S1). The mixture was incubated for 30 min before adding 15% (v/v) Bio-Beads SM2 resin (Bio-Rad) to remove the detergent. The reconstitution mix along with Bio-Beads was incubated overnight at 4 °C on a nutator. The next day, Bio-Beads were removed by brief centrifugation for 5 min at 2000 × g. The clear reconstitution solution was collected, concentrated to 300 µl total volume, and further purified in size exclusion chromatography (SEC) using a Superose 6 Increase 10/300 GL column (Cytiva Life Sciences, 29091596) in an SEC buffer (200 mM KCl, 20 mM Tris, 2 mM DTT, pH = 8.0). Representative SEC profiles and Coomassie blue-stained SDS-PAGE gels are depicted in Fig. S1. For the HCN4 detergent sample, all the steps until the elution were the same as described above, but the final SEC buffer contained 300 mM KCl, 20 mM Tris, 2 mM DTT, 5% glycerol, 0.1 mM GDN (Anatrace). Only the peak fractions from SEC (Fig. S1) were collected and used for single- molecule experiments. Native nanodisc preparation Free-style HEK suspension cells expressing HCN1 were lysed using a Branson cell disruptor. Crude membranes were isolated from cell lysate by ultracentrifugation at 184,000 × g for 1 h and resuspended in reconstitution buffer (50 mM Tris-HCl (pH 8), 300 mM NaCl, 10% glycerol, 2 mM DTT). Native nanodiscs were formed by incubation of crude membranes (~10 mg/ml) with the DeFrMSP peptide Hex18A at 1 mg/ml (4 °C, O/N). Hex18A is a peptide that was custom synthesized from GenScript by Huan Bao lab. The peptide sequence is as follows DWLKAFYDKVAEKLKEAF with functional N-terminal modification of Hexanoic acid47. HCN1 native nanodiscs were isolated using Streptactin affinity resin (IBA Life Sciences) and eluted in reconstitution buffer plus 5 mM desthiobiotin. Samples were further purified on a Superdex S200 3.2/300 column in reconstitution buffer and used for single-molecule experiments. CryoEM protein preparation Flag-HCN1-Avitag-TGP-10His channels, the human HCN1 1-636 residues with N-terminal flag tag and a C-terminal AviTag and the thermostable green protein (TGP) and 10X His tag, were expressed in yeast (P. pastoris strain SMD1163H, Invitrogen, #C17500). The yeast cultures were grown for 48 h post-induction, and the resultant cell pellet expressing HCN1 was disrupted by milling (Retsch MM400). The ground powder was resuspended in a buffer containing 50 mM Tris, 300 mM KCl, 10% Glycerol, 1 mM EDTA with protease inhibitor cocktail (Sigma) and homogenized. The homogenate was centrifuged at 100,000 × g for 40 min at 4 °C. The supernatant was discarded, and the membrane pellet was resuspended in 2X solubilization buffer (80 mM Tris, 600 mM KCl, 8 mM DTT, 2x protease inhibitor cocktail (Sigma Aldrich, P8340), 10% glycerol, pH 8.0), flash frozen in liquid nitrogen and stored at −80 °C until further use. On the day of preparation, yeast cell membranes equivalent to 5 g of pellet were thawed on ice for 30 min on ice. The membrane fraction was then solubilized by adding an equal volume of detergent solution containing 10 mM LMNG and 2 mM CHS into the membrane solution and incubating for 2 h at 4 °C on a rotator. The solubilized membrane was centrifuged at 20,000 × g for 40 min at 4 °C. The supernatant was added to 2 mL Ni-NTA resin (QIAGEN) that had been pre-equilibrated with 200 mM KCl, 20 mM Tris, 2 mM DTT, 1 mM LMNG, 0.2 mM CHS, 10 mM imidazole, pH 8.0, and incubated for 2 h at 4 °C on a rotator. After batch binding, the resin and supernatant mixture was transferred to a gravity column, and flow-through was discarded. The resin was washed with 20 bed volumes of wash buffer (200 mM KCl, 20 mM Tris, 2 mM DTT, 1 mM LMNG, 0.2 mM CHS, and 30 mM imidazole, pH 8.0). Protein was eluted with 15 bed volume of elution buffer (200 mM KCl, 20 mM Tris, 2 mM DTT, 0.1 mM LMNG, 0.02 mM CHS, and 250 mM imidazole, pH 8). The elution was concentrated using a 100 kDa-cutoff concentrator and further purified by size exclusion chromatography (SEC) using a Superose 6 Increase 10/300 GL column (Cytiva Life Sciences, 29091596) in SEC buffer A (200 mM KCl, 20 mM Tris, 2 mM DTT, and 0.01% GDN, pH 8). The peak fraction from SEC was collected and concentrated to 700 µl using a 100 kDa concentrator, and the protein concentration was measured using a Denovix spectrophotometer to determine the protein yield. For reconstitution in nanodiscs, dried soy polar lipids (Avanti) were solubilized in 20 mM Tris, 150 mM KCl, and 2% CHAPS by sonicating to yield a 20 mM lipid solution. SEC purified HCN1, MSP2N2, and soy polar lipids were combined at a molar ratio of 1:2:200 and incubated on ice for 30 min before adding 10% (w/v) Bio-Beads SM2 (Bio-Rad) to remove detergent. The nanodiscs were incubated with Bio-Beads for 2 h at 4 °C on a rotator then transferred to fresh Bio-Beads for overnight incubation. The next day, the nanodisc sample was removed from the Bio-Beads, centrifuged at 13,000 × g for 10 min, and incubated for 2 h with 1 mL anti-FLAG resin (Genescript Anti-DYKDDDDK G1 Affinity Resin, Cat#L00432) pre-equilibrated with SEC buffer B (20 mM Tris, 150 mM KCl, and 2 mM TCEP, pH 8) at 4 °C on a rotator. The resin was then transferred to a gravity column at room temperature and the flow-through was discarded. Resin was washed with 20 bed volumes of room temperature wash buffer containing 20 mM Tris, 150 mM KCl, and 2 mM DTT, pH 8. Protein was eluted from the anti-FLAG resin using wash buffer containing 300 μg/mL FLAG peptide (Sigma Cat #F3290). Three 1 mL elutions were collected, incubating for 10 m at room temperature before collecting each elution. Elution fractions containing protein were pooled and concentrated to 300 μL. Eluate was further purified by SEC in SEC buffer B. CryoEM data acquisition and analysis The HCN1 nanodisc sample was concentrated to 6.5 mg/mL. For the cAMP-bound state, the sample was spiked with 500 µM cAMP 30 m prior to freezing. 3 μL of purified HCN1 nanodisc sample was applied to plasma-cleaned UltrAuFoil R2/2 200 mesh Au grids. Grids were loaded in a Leica EM GP2 Automatic Plunge Freezer at 10 °C with 90% humidity and back-blotted for 8 s before plunge freezing into liquid ethane. CryoEM imaging was performed in the Electron Microscopy Core of the University of Missouri with ThermoScientific Titan Krios G4 Cryo-TEM. The imaging parameters are listed in the Table S38. The detailed flowcharts for analysis are shown in Fig. S15 (apo) and Fig. S17 (cAMP-bound). This resulted in final maps with 2.5 and 2.6 Å resolution for the apo and cAMP-bound structures, respectively. To calibrate the pixel size of the final reconstruction map, we performed an internal fitting procedure using the crystal structure of the HCN1 CNBD (PDB: 3U0Z). The model was rigid-body docked into our cryo-EM map using ChimeraX. We then systematically adjusted the voxel size of the map and re-evaluated the fit quality at each step. Starting from an initial pixel size of 0.904 Å, we found that a voxel size of 0.9525 Å yielded the optimal model-to-map agreement, as assessed by visual inspection and cross-correlation in ChimeraX. The model building began by fitting the AlphaFold model AF-A0A669KB45-F1-model_v4 to the density in ChimeraX 1.6, followed by manual adjustments in Coot v0.9.8.93. After validation of the monomer, the tetramer was built with Chimera 1.17.3 with symmetry expansion to match the C4 symmetry in the map. The tetramer was further refined in real space with Phenix, and the resulting model was subsequently adjusted to minimize clashes. The final model was validated in the PDB validation server (https://validate-rcsb-1.wwpdb.org/). Fabrication of zero-mode waveguides ZMW fabrication was done at The Center for Nanophase Materials Sciences, Oak Ridge National Laboratory. Briefly, Cover glasses (Fisher Scientific cat. no.12-548 C, 130–170 μm thickness) were cleaned with 2% Hellmanex (Hellma), HPLC grade Acetone (Sigma), HPLC grade Ethanol (Sigma), and 3 M KOH by sonicating for 15 mins in each solution, intervened by a water wash. Cleaned cover glasses were plasma cleaned (Gatan Solarus 950 plasma cleaner) for 10 min. Cover glasses were coated with aluminum using either sputter coating at a rate of 4 nm/min with a DC Magnetron sputtering system or thermally evaporated aluminum at a rate of 2 Å/s with a JEOL dual-source E-beam evaporator, resulting in a total thickness of 100 nm. In DC sputtering, an argon plasma is used to initiate the deposition process. Excited argon ions strike the target material, causing atoms to be ejected from the target. These atoms then deposit onto the substrate, which is placed below the target source. On the other hand, the E-beam evaporator utilizes an electron beam to heat metals until they reach a vapor phase. Substrates are positioned above the electron beam guns, and when the shutter is opened, the vaporized metal is deposited onto the substrates. The deposition thickness was checked with a surface profilometer and confirmed to be 100 ± 5 nm. Aluminum-coated cover glasses were then spin-coated with positive-tone electron beam resist ZEP 520 A (45 s at 3000 rpm). The cover glasses were baked at 180 °C for 2 min. After that, ZMWs featuring 120, 150, and 180 nm dots were patterned on the surface using 100 kV accelerating voltage and 2 nA beam current using the JEOL JBX-8100FS electron beam lithography system. Following exposure, the ZMWs were developed using a 30-sec treatment of xylenes, followed by a wash with 2-propanol, and dried with nitrogen gas. Developed cover glasses were mounted on a silicon wafer using a thin layer of Krytox oil applied on the back of the cover glasses, and 100 nm aluminum was dry etched for 1 min using OXFORD Plasmalab System 100 reactive ion etcher with a mixture of 30 standard cubic centimeters (sccm) chlorine (Cl2) and 10 sscm boron trichloride (BCl3) gasses at 50 °C. Finally, the samples were plasma cleaned in PVA TePla IoN Wave 10 using O2-Ar plasma to remove any excess photoresist. To ensure the quality, ZMWs from different batches were chosen at random and imaged under Zeiss Merlin FE-Scanning Electron Microscopy (SEM). We note that the hole sizes were within the margin of error ±7 nm (Fig. S2E). Sample and chamber preparation for single-molecule imaging Cover glasses and ZMWs used for single-molecule imaging were cleaned and passivated as described previously28. Briefly, ZMWs were plasma cleaned, treated with 2% polyvinyl phosphonic acid (Polysciences), washed with water, and dried under argon gas. Later the ZMW was passivated with mPEG-silane (Laysan Bio) and doped with biotin-PEG silane (Laysan Bio) overnight. On the following day a small 50 µl silicone-gasketed microfluidic chamber (Grace Bio labs) was attached to the ZMW for single-molecule imaging. Non-specific binding was assessed by adding 500 nM fcAMP to the unlabeled surface (Fig. S2G) and was found to be negligible. Single-molecule pulldown was done to immobilize the protein. Briefly, the chamber was sequentially incubated with 10% BSA (Sigma Aldrich) for 30 min (additional layer of passivation), 1 µM streptavidin (Prospec) for 10 min, 10 nM biotinylated GFP nanobody (Chromtek) for 10 min, ~225–250 pM of GFP-tagged-HCN1 and HCN4 for 10 min. After 10 min, the chambers were washed to remove excess freely diffusing protein and empty nanodiscs. Briefly, to achieve predominantly single occupancy of ZMWs, we loaded GFP-tagged HCN channels under Poisson conditions. The number of channels per ZMW follows P(j)=e−λλj/j! where λ is the mean occupancy and j is the number of channels in the same well. Before adding fcAMP, we estimated λ from the fraction of GFP positive ZMWs in the field of view and titrated the protein concentration to target λ ≈ 0.2−0.3 (~25% labeled ZMWs), (typically around 150–200 pM of purified tetrameric protein) which limits multioccupancy to P(j ≥ 2) = 1 − (1+λ) e−λ ≤∼4. Immobilization was for 10 min followed by a wash to remove unbound protein. Between each step, the chamber was washed with 1X Tris-buffered saline. All the single-molecule experiments were done in an imaging buffer (1X Tris-buffered saline supplemented with 1% BSA, 2 mM Dithiothreitol (DTT, Sigma-Aldrich) and oxygen scavengers including 2 mM Trolox (Sigma-Aldrich) and 2.5 mM protocatechuic acid (Sigma-Aldrich) and 250 nM protocatechuate 3, 4-dioxygenase from Pseudomonas sp. (Sigma-Aldrich) were added just before imaging. If the sample being imaged was a protein in the detergent micelle, then the buffer contained 100 μM Glycodiosgenin (GDN). The chamber was then added with various concentrations of 8-(2-[DY-547}-aminoethylthio) adenosine-3’, 5’-cyclic monophosphate (fcAMP; Biolog). The solutions in the chamber were replenished every 20 min to avoid evaporation. We have observed that the protein in nanodiscs is more stable than in detergents, we can record activity from the protein in nanodisc (even after a week of storage at 4 °C) but the protein in detergent micelle is only stable for a day and quickly loses activity beyond that. To be able to directly compare the results between the conditions, we have performed experiments on both detergent and nanodisc samples on the same day as purification (within 12 h of SEC) as described previously28. For the imaging, only a small region is exposed under the laser excitation at a time. For every imaging we move to a different region which was not previously exposed to laser excitation. Additionally, we also refrain from exposing the sample to room light throughout the experiment. Single-molecule imaging Imaging was done on an inverted microscope (Mad City Labs) on a high NA 60X oil immersion objective lens (Olympus) under a micromirror TIRF excitation field generated either with a 488 nm or 561 nm laser (OBIS, coherent) to excite the GFP and fcAMP respectively. The subsequent fluorescence emission was recorded on a 512 × 512 EMCCD camera (Andor iXON Ultra) at a frame rate of 10 Hz. Data collection was accomplished using Micromanager 2.0. The use of micromirror TIRF eliminated most of the stray excitation light. The emission was passed through a dichroic mirror (T565lpxr Chroma), and then a band-pass filter (Chroma ET550/25 nm) for GFP and a band-pass filter (Semrock bright line 593/40 nm) for fcAMP were in the corresponding emission pathways. The micromirror TIRF is set up for a 160x magnification and 234 × 234μm field of view, allowing us to view the ~720 ZMWs at once. Diameter of the input laser beam was optimized to 5 mm, which gives a 100 mm illumination spot on the objective, and the laser power measured using an optical power meter (Newport, 1918-R) at the objective was 9.6 mW and 7 mW for 488 nm and 561 nm lasers, respectively. The micromirror TIRF setup was equipped with a TIRF-lock that continuously monitors the position of the TIRF exit beam and adjusts the Nano-positioner for any drift in the Z-axis. ZMWs were labeled sparsely to reduce the probability of having more than one per ZMW. To circumvent protein aggregation/clustering we take many measures. First, we refrain from concentrating our purified protein fractions from size exclusion chromatography and use dilute protein samples where the final protein concentration is 225–250 pM. Second, to minimize aggregation, we also purify the protein and perform single-molecule experiments on the same day to ensure that we do not store the protein at 4 °C for long periods of time. Third, to rule out the possibility that the channels are not aggregating, we perform photobleaching experiments under identical Poisson dilution conditions to ensure that most of them are from a single tetramer. These steps ensure that we rarely observe more than 4 binding steps on our measurements. ZMWs that are labeled were identified by exciting the GFP molecules, these GFP molecules were photo-bleached. To guarantee full photobleaching of GFP, we extended the 488-nm illumination as required. Absence of spectral bleed-through was verified by comparing background fluorescence intensities in ZMW wells containing protein versus empty wells, which exhibited no measurable differences (Fig. S2F). Later, fcAMP molecules were excited with the 561 nm laser to measure the binding activity. The binding data were recorded from each molecule for a minimum of 240 s with 100 ms exposure time. We note that all our data were collected using 150 nm ZMW wells. Single-molecule data that was collected was analyzed exactly as described in the previous work28with an exception that AutoDISC38 (instead of DISC)29 was used to idealize the traces for the unsupervised selection of objective criteria during the data analysis. Briefly, single-molecule traces were extracted and analyzed using the DISC software for the colocalization and image projection to obtain time-dependent fluorescence intensity changes, the traces were then idealized in autoDISC. All the idealized traces were visually inspected to eliminate molecules with more than 4 bound states and traces with low signal-to-noise where state assignment was difficult. Alongside the representative fluorescence–time (FT) traces, we show dynamic heatmaps in which each row is an independent molecule, the x‑axis corresponds to time. The color of the heatmap indicates the ligation state (0 to 4) according to the color scale. For each represented condition and concentration, we have plotted three idealized traces as heatmaps. The data analysis was repeated to remove any single-frame events arising from rapid or non-specific binding to assess whether they influenced the distribution. From theiIdealized fluorescence-time traces, we extracted the time spent by fcAMP in each ligation state (state occupancy). These state occupancies were normalized to the expected value obtained from a binomial distribution assuming independent and identical binding sites, each with a probability of ligand binding P, the probability of x binding sites being occupied simultaneously is given by the equation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Pr \left(x\right)=\frac{4!}{x!\left(4-x\right)!}{p}^{x}{(1-p)}^{4-x}$$\end{document}Prx=4!x!4−x!px(1−p)4−x The ratio of Observed over Expected occupancies is 1 when the ligand binding is completely independent. All data analysis was performed using custom MATLAB (MathWorks) scripts. HMM, analysis of all the data sets was performed using QuB as described before29. Briefly, for each set of conditions, the sequential model (Fig. S11A) was globally optimized using maximum idealized point likelihood rate estimation (MIP) across all fcAMP concentrations. The goodness of fit was measured by the Bayesian Information Criterion (BIC). BIC = k X ln (n) - 2 x LL. From these fits, the microscopic rate constants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{\mbox{on}}}$$\end{document}kon and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{\mbox{off}}}$$\end{document}koff) for each sequential binding step were obtained directly from the optimized QuB models. Stepwise association constants were then defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{a}({M}^{-1})=K={k}_{{\mbox{on}}}/{k}_{{\mbox{off}}}$$\end{document}Ka(M−1)=K=kon/koff and these values were used in the following Adair equation to reconstruct the macroscopic binding curves.2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f=\frac{1}{4}\left(\frac{4{K}_{1}\left[X\right]+12{K}_{1}{K}_{2}{\left[X\right]}^{2}+12{K}_{1}{K}_{2}{K}_{3}{\left[X\right]}^{3}+4{K}_{1}{K}_{2}{K}_{3}{K}_{4}{\left[X\right]}^{4}}{1+4{K}_{1}\left[X\right]+6{K}_{1}{K}_{2}{\left[X\right]}^{2}+4{K}_{1}{K}_{2}{K}_{3}{\left[X\right]}^{3}+{K}_{1}{K}_{2}{K}_{3}{K}_{4}{\left[X\right]}^{4}}\right)$$\end{document}f=144K1X+12K1K2X2+12K1K2K3X3+4K1K2K3K4X41+4K1X+6K1K2X2+4K1K2K3X3+K1K2K3K4X4Where ‘f’ is the bound probability; K1, K2, K3, and K4 are stepwise association constants (M-1) for 1st, 2nd, 3rd and 4th bound states, respectively; and [X] is the fcAMP concentration. Note that, a bound probability of 1.0 corresponds to fcAMP occupying all four binding sites. Additionally, removal of single-frame events yielded stepwise association constants that were essentially identical, indicating that rapid events beyond the instrument’s temporal resolution do not substantially influence the measured association constants (Fig. S10F). UBU dwell times were obtained by isolating unbound-B1-unbound transitions across different fcAMP concentrations and were plotted and fitted with biexponential decay. The individual time constants and amplitudes obtained from the bi-exponential fit were used to calculate the weighted time constant <τ> using the following equation3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ < {{{\rm{\tau }}}} > =\frac{{{A}_{1}\tau }_{1}+{{A}_{2}\tau }_{2}}{{A}_{1}+{A}_{2}}$$\end{document}<τ>=A1τ1+A2τ2A1+A2Where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{1}$$\end{document}τ1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{2}$$\end{document}τ2 are the individual time constant of the bi-exponential fit, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}A1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${and}$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}A2 are amplitude of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{1}$$\end{document}τ1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{2}$$\end{document}τ2 respectively. BUB dwell times were obtained by isolating B1-Unbound-B1 transitions at different ligand concentrations. To obtain the time constants, BUB dwell times were fitted with a single exponential decay. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{on}}$$\end{document}kon was calculated using the following equation4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,{k}_{{on}}=\frac{1}{{\tau }_{1}*[C]*4}$$\end{document}kon=1τ1*[C]*4Where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{1}$$\end{document}τ1 is the time constant of the monoexponential fit of BUB dwell times, and [C] is the concentration of fcAMP and 4 is the statistical factor which accounts for total number of unoccupied binding sites. A two-sample Kolmogorov–Smirnov test was performed to assess the statistical significance of the differences in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{on}}$$\end{document}kon measured under various lipid conditions relative to the detergent. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supp Movie 1 Supp Movie 2 Reporting Summary Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supp Movie 1 Supp Movie 2 Reporting Summary Transparent Peer Review file Source data source data source data