Fluorescence activation mechanism and imaging of drug permeation with new sensors for smoking-cessation ligands

Nicotinic partial agonists provide an accepted aid for smoking cessation and thus contribute to decreasing tobacco-related disease. Improved drugs constitute a continued area of study. However, there remains no reductionist method to examine the cellular and subcellular pharmacokinetic properties of these compounds in living cells. Here, we developed new intensity-based drug-sensing fluorescent reporters (iDrugSnFRs) for the nicotinic partial agonists dianicline, cytisine, and two cytisine derivatives – 10-fluorocytisine and 9-bromo-10-ethylcytisine. We report the first atomic-scale structures of liganded periplasmic binding protein-based biosensors, accelerating development of iDrugSnFRs and also explaining the activation mechanism. The nicotinic iDrugSnFRs detect their drug partners in solution, as well as at the plasma membrane (PM) and in the endoplasmic reticulum (ER) of cell lines and mouse hippocampal neurons. At the PM, the speed of solution changes limits the growth and decay rates of the fluorescence response in almost all cases. In contrast, we found that rates of membrane crossing differ among these nicotinic drugs by >30-fold. The new nicotinic iDrugSnFRs provide insight into the real-time pharmacokinetic properties of nicotinic agonists and provide a methodology whereby iDrugSnFRs can inform both pharmaceutical neuroscience and addiction neuroscience.


Introduction
Smoking cessation is an important goal to help decrease the burden, both individual and societal, of tobacco-related disease. The addictive tobacco alkaloid nicotine itself, via transdermal patches and other devices, remains available for people trying to quit smoking; but nicotine replacement therapy has distressingly low rates of success. Therefore, various research projects are continuing with the aim of developing more effective ligands for nicotinic acetylcholine receptors (nAChRs).
Membrane permeation is interesting for investigating and treating nicotine addiction in at least two ways. Firstly, note criterion #1 above. For uncharged molecules, the conventional metric for membrane permeability is logP, where P is the octanol-water partition coefficient. For weak bases including most orally available neural drugs, logP must be corrected to account for the fraction of uncharged (deprotonated) molecules at the pH of interest, usually pH 7.4; the resulting metric, termed logD pH7.4 , is always less positive than logP. Enhancing the membrane permeability of cytisine analogs and probing nAChR subtype selectivity was addressed via direct functionalization of cytisine within the pyridone ring (Rego Campello et al., 2018). Two of the resulting derivatives, 10-fluorocytisine and 9-bromo-10-ethylcytisine, have cytisine-like EC 50 for the α4β2 nAChRs, but more positive calculated logD pH7.4 values, suggesting greater membrane permeability at the nearly neutral pH of the blood, brain, and cytoplasm (Blom et al., 2019). Estimates of logD pH7.4 are inexact, extrapolated, or rely on algorithmic calculations whose results differ over 2 log units for individual molecules (Pieńko et al., 2016). These estimates have unknown applicability to biological membranes at the logD pH7.4 values < 0 that characterize varenicline, dianicline, and the cytisine analogs.
Secondly, nicotine dependence involves one or more 'inside-out' mechanisms. Nicotine itself (logD pH7.4 0.99) enters the endoplasmic reticulum (ER), binds to nascent nAChRs, becomes a pharmacological chaperone for the nAChRs, and eventually causes selective upregulation of these receptors on the plasma membrane (PM) (Henderson and Lester, 2015). For this reason, it is especially important to understand permeation into the ER.
These two neuroscience aspects of nicotinic ligands -pharmaceutical science and addiction science -call for direct measurements of drug movements in living cells (Video 1). We previously explored the subcellular pharmacokinetics of nicotine and varenicline in immortalized cell lines and cultured neurons using the iDrugSnFRs iNicSnFR3a and iNicSnFR3b to visualize that these nicotinic agonists enter the ER within seconds of drug application and exit equally rapidly from the ER upon extracellular washing . That nicotine diffuses across cellular membranes in seconds has been suspected for decades: nicotine crosses six PMs to enter the brain within 20 s, providing a 'buzz.' Video 1. Genetically encoded fluorescent biosensors show how drugs cross membranes in real time. https://elifesciences.org/articles/74648/figures#video1 That varenicline becomes trapped in acidic vesicles suggests appreciable membrane permeation but may also underlie unwanted effects (Govind et al., 2017;Le Houezec, 2003). We sought to generate and apply additional intensity-based drug-sensing fluorescent reporters (iDrugSnFRs) for candidate smoking-cessation drugs: dianicline, cytisine, 10-fluorocytisine, and 9-bromo-10-ethylcytisine. We hypothesized that a family of newly developed iDrugSnFRs would enable quantifiable fluorescence signals that compare the differences in permeation among these compounds.

Generation of additional nicotinic iDrugSnFRs: Structural tactic
To generate iDrugSnFRs for cytisine and dianicline, we followed two converging tactics. In the 'structure-based' tactic, we obtained the first structural data for OpuBC-based SnFRs bound by nicotinic ligands (nicotine and varenicline) (Figure 1, Supplementary file 1). Crystals of iNicSnFR3adt in the presence of 10 mM nicotine diffracted to 2.95 Å resolution (PDB 7S7U). Overall, the liganded periplasmic binding protein (PBP) domain of iNicSnFR3adt adopts a closed conformation ( Figure 1A). In the binding pocket between the top and bottom lobes of the PBP, we observed an 'avocado'-shaped electron density in the nicotine binding site, enclosed by several aromatic residues ( Figure 1B). The combination of protonation/deprotonation and the rotatable bond of nicotine (Elmore and Dougherty, 2000) vitiate unambiguously localizing it within the binding pocket.
We obtained an unambiguous ligand placement for iNicSnFR3adt in the presence of 10 mM varenicline in the same crystallization condition. Crystals of iNicSnFR3adt with varenicline bound were isomorphous to those of the nicotine-bound crystals and diffracted to 3.2 Å resolution (PDB 7S7T). While the protein structure ( Figure 1D) is identical to that of the nicotine bound structure ( Figure 1A), the rigidity and additional ring of varenicline allowed us to unambiguously localize it in the binding pocket. Varenicline is enclosed by the same aromatic residues as nicotine, forming cation-π interactions with Tyr65 and Tyr357, in addition to other interactions with the pocket residues ( Figure 1E).
In the full iDrugSnFR, in the apo state, the Glu78 in Linker 1 approaches within ~2.5 Å of the oxygen of the tyrosine fluorophore ( Figure 1E1; PDB 7S7V). Figure 1E2 provides structural details confirming the hypothesis (Barnett et al., 2017;Nasu et al., 2021) that in the liganded state Glu78 has moved away, presumably allowing the fluorescent tyrosinate to form (Video 2). We term this mechanism the 'candle snuffer'.

Generation of additional nicotinic iDrugSnFRs: Mutational tactic
In the mutational tactic, we screened each drug shown in Figure 1-figure supplement 2 against a panel of biosensors that included iNicSnFR3a and iNicSnFR3b  and iAChSnFR  as well as intermediate constructs from their development process. From this screen, we chose sensors with the lowest EC 50 for each drug as our starting protein for iDrugSnFR evolution.
Because the candle snuffer mechanism explains several details of the agonist-and pH sensitivity of both iNicSnFR3a and iSketSnFR (see 'Discussion'), we presume that it represents a general mechanism for OpuBC-cpGFP SnFRs. We did not mutate residues that lie (in 3D space) between the binding site and linkers.

Specificity and thermodynamics of nicotinic iDrugSnFRs
We characterized the specificity of purified iDrugSnFRs for their drug partners versus a panel of related nicotinic agonists (Table 2, Figure 3). The newly developed iDrugSnFRs showed some sensitivity to related nicotinic agonists. iDianiSnFR had the greatest fidelity for its drug partner but also showed an increased EC 50 (15 µM) as a purified protein versus its EC 50 in lysate (6.7 µM), possibly indicating decreased stability in a purified form. iCytSnFR, iCyt_F_SnFR, and iCyt_BrEt_SnFR showed a greater level of promiscuity for the compounds comprising the nicotinic agonist panel. Of note, iCytSnFR, iCyt_F_SnFR, and iCyt_BrEt_SnFR have an exceptionally low (60-90 nM) EC 50 for varenicline. The newly developed iDrugSnFRs showed negligible binding to choline or the neurotransmitter acetylcholine, leading one to expect minimal endogenous interference during future in vivo experiments. We also performed dose-response experiments with iDianiSnFR, iCytSnFR, iCyt_F_SnFR, and iCyt_ BrEt_SnFR against a panel of nine endogenous molecules, including neurotransmitters (Figure 3figure supplement 1). iDianiSnFR showed no response to any of the nine selected compounds above background. iCytSnFR, iCyt_F_SnFR, and iCyt_BrEt_SnFR showed no response above background for seven of the compounds. However, they exhibited a ΔF/F 0 of 0.25-0.8 to dopamine at 316 µM/1 mM and a ΔF/F 0 of 0.8-1.5 to serotonin (5-HT) at 316 µM/1 mM. In terms of S-slope, the relevant metric for most cellular or in vivo experiments, the SnFRs are at least 250-fold more sensitive to their eponymous partners than to other molecules we have tested.
To examine the thermodynamics of the iDrugSnFR:drug interaction, we conducted isothermal titration calorimetry (ITC) binding experiments ( Figure 4). The experimentally determined K D of each iDrugSnFR:drug pair using ITC was within a factor of 1.5 from the experimentally determined EC 50 for fluorescence in E. coli lysate or purified protein (Table 3). We infer that the EC 50 for fluorescence is dominated by the overall binding of the ligand for all the iDrugSnFRs.

Kinetics of nicotinic agonist iDrugSnFRs: Stopped-flow
In a stopped-flow apparatus, we measured the fluorescence changes of iDrugSnFRs with millisecond resolution during multiple 1 s trials and an independent 100 s trial. The stopped-flow data revealed that iDrugSnFRs do not have pseudofirst-order kinetic behaviors typical of two-state binding interactions. Time courses of iDianiSnFR (both over 1 s and 100 s) were best fitted by double exponential equations. Most of the fluorescence change occurs within the first 0.1 s of mixing ( Figure 5A), with only minor additional increase by 100 s. site residues. (C) Overall conformation of iNicSnFR3adt with varenicline bound (PDB 7S7T). (D) iNicSnFR3adt binding site with varenicline present. (E) Aspects of the PBP-Linker1-cpGFP interface, emphasizing contacts that change upon ligand binding. The Phe76-Pro77-Glu78 cluster (in Linker 1) lies 11-16 Å from position 43, which defines the outer rim of the ligand site (B); therefore, the cluster makes no direct contact with the ligand site. (E1) In the apo conformation, Glu78 acts as a candle snuffer that prevents fluorescence by the chromophore (PDB 7S7V). (E2) In the liganded conformation (PDB 7S7T), the Phe76-Pro77-Glu78 cluster moves Glu78 at least 14 Å away from the fluorophore. Pro77 is flanked by Phe76 and Pro396 (in the top lobe of the PBP moiety). The presumably deprotonated Glu78 forms salt bridges with Lys97 and Arg99, both facing outward on the β6 strand of the original GFP (within the original Phe165-Lys-Ile-Arg-His sequence).
The online version of this article includes the following figure supplement(s) for figure 1:   Changes in fluorescence from iCytSnFR during the first 1 s of mixing fit well to a single exponential ( Figure 5B) and have close to pseudo-first-order kinetics (i.e., the observed rate of fluorescence change is nearly linear with drug concentration). As with iDianiSnFR, most of the fluorescence change occurs within the first second, with additional fluorescent increase continuing over the next minute ( Figure 5B, right panel).
Like iDianiSnFR, iCyt_F_SnFR fluorescence changes are best fit by a double exponential ( Figure 5C), but the time course of fluorescence change is significantly slower. Fluorescence gradually increases throughout the recording period and beyond. This information was considered in later in vitro and ex vivo experiments. iCyt_BrEt_SnFR fits well to a single exponential ( Figure 5D) for the first 1 s of data collection, but like the other sensors, continues to increase its fluorescence over longer periods.
We plotted the k obs (s -1 ) obtained in the 1 s stopped-flow experiments versus concentration ( Figure 5E-H) (see also Supplementary file 2). The aberrations from ideal first-order kinetics vitiate generation of definitive k off and k on values but we can approximate a K max and K D from our fitting procedures. Our stopped-flow experiments reinforced previous observations (Unger et al., 2020) that the kinetics of iDrugSnFR binding involve complexities beyond a simple first-order kinetic model governing two binding partners.

Kinetics of nicotinic agonist iDrugSnFRs: Millisecond microperfusion
We also studied iCytSnFR_PM expressed in HEK293T cells during fluorescence responses to ACh, cytisine, or varenicline in a microperfusion apparatus that exchanged solutions near the cell on a millisecond time scale (Materials and methods). This system directly measures the decay of the response when ligand is suddenly removed. The rank order of the iCytSnFR steady-state sensitivities is varenicline > cytisine > ACh. The time constant for decay decreased with increasing steady-state EC 50 of the ligands, as though more tightly binding ligands dissociate more slowly ( Figure 6A).
We measured the decay waveforms after drug pulses at concentrations ≥ the EC 50 of the steadystate response to maximize the ΔF/F 0 signal/noise ratio ( Figure 6A-D). Because the decay phases are measured in zero [ligand], one expects that the decay rate constant(s) (k off ) for an iDrugSnFR do not depend on the pulsed ligand concentration. Decay of the ACh response followed a single exponential time course ( Figure 6B). The values of the k off for 30, 100, and 200 µM ACh did not differ significantly (ANOVA, p=0.62, degrees of freedom [df] = 2 (model), 20 [error]). We pooled them to obtain a mean k off of 1.9 ± 0.1 s -1 (mean ± SEM, n = 23 areas [50 cells]). The corresponding time constant τ 0ff was 530 ± 30 ms. Hence, the temporal resolution of the CytSnFR_PM sensor for changes in the ACh concentration was in the subsecond range.
The decay of the cytisine and varenicline response was biphasic ( Figure 6C and D): two exponential decay terms with an additional constant component fitted the cytisine decay significantly better than a single exponential term (F-test, p<0.05). As expected, neither the faster decay rate constants (kf off ) (ANOVA, p=0.30, df = 3,32) nor the slower decay rate constants (ks off ) (ANOVA, p=0.54, df = 3,31) differed among the tested cytisine concentrations (5-15 µM). The kf off and ks off for 5-15 µM cytisine were 0.61 ± 0.04 s -1 (n = 36 areas, 105 cells) and 0.146 ± 0.006 s -1 (n = 35 areas, n = 103 cells), respectively. The corresponding decay time constants (τf 0ff , τs 0ff ) were 1.8 ± 0.1 s and 6.9 ± 0.2 s. Therefore, the temporal resolution of CytSnFR_PM sensor for cytisine was <10 s, adequate for the temporal resolution of the live-cell experiments presented below. Interestingly, the decay waveform of the varenicline response was much slower than that for cytisine or ACh ( Figure 6A and D). We pulsed 2 µM varenicline, >> the EC 50 of the steady-state response of the isolated protein (60 ± 10 nM) ( Figure 6D). The values of the kf off and ks off were 0.9 ± 0.2 s -1 and 0.0065 ± 0.0002 s -1 , respectively (n = 4 areas [nine cells]). The slower component dominated the decay phase, with a fractional amplitude of 85% ± 1%. Thus, the temporal resolution of the iCytSnFR_PM sensor for varenicline was in the minute range. In the live-cell experiments described below, it would not be possible to resolve the differences between varenicline at the PM and in the ER. The relatively high affinity of iCytisineSnFR for varenicline, which presumably arises in part from the increased lifetime of the varenicline-iDrugSnFR complex, has drawbacks. The temporal resolution of iNicSnFR3a and iNicSnFR3b, which bind varenicline ~100-fold less tightly, is appropriate for subcellular experiments . The previous experiments showing ER entry of varenicline used iNicSnFR3a and iNicSnFR3b . For additional microperfusion data and analyses, see Appendix 2-figures 1-3.

Characterization of nicotinic iDrugSnFRS in HeLa cells and primary mouse hippocampal culture
We examined the subcellular pharmacokinetics of the nicotinic agonists in mammalian cell lines and primary mouse hippocampal neurons. The nicotinic iDrugSnFRs were targeted to the PM (iDrugSnFR_ PM) or the ER (iDrugSnFR_ER) as previously described Shivange et al., 2019). We then performed a dose-response experiment using wide-field fluorescence imaging with each iDrugSnFR and its drug partner, sampling a range of concentrations covering a log scale surrounding the EC 50 as determined for the purified protein (Figures 7 and 8,. iDianiSnFR showed a robust response to dianicline at the PM and the ER in HeLa cells across a range of concentrations (3.125-100 µM), and the speed was nearly limited by solution exchanges; there was a clear return to baseline fluorescence upon washout on the order of seconds after each drug application. At 100 µM, the PM and ER have a ΔF/F 0 of ~1.2, but at lower concentrations, the ER displayed 30-75% of the signal detected at the PM, which may indicate a difference in membrane crossing ( Figure 7A). Imaging in primary mouse hippocampal neurons demonstrated a similar trend ( Figure 8A).
Cytisine showed slower entry into and exit from the ER of HeLa cells. The iCytSnFR_PM construct detected cytisine at concentrations from 0.078 to 80 µM and demonstrated a return to baseline fluorescence upon washout on the order of seconds after each drug application, reaching a maximum ΔF/ F 0 of ~2 at concentrations above 5 µM ( Figure 7B). In contrast to the _PM construct, the iCytSnFR_ER construct only detected cytisine with a ΔF/F 0 above the buffer control in the range of concentrations from 1.25 to 80 µM with a ΔF/F 0 , which was 25-50% of the maximum ΔF/F 0 detected at the PM. Additionally, in the range of detectable concentrations of cytisine, the washout of cytisine was much slower than solution changes ( Figure 7B). The incomplete washout persists even after several minutes and corresponds with previous suggestions that cytisine has low membrane permeability, as evidenced by its low brain penetration (Rollema et al., 2010). Table 2. Intensity-based drug-sensing fluorescent reporter (iDrugSnFR) dose-response relations versus a selected panel of nicotinic agonists. ND, not determined. *, ** EC 50 and ΔF max /F 0 could not be determined from the data ( Figure 3). Therefore, the upper limit to the S-slope is estimated from the data at the foot of the dose-response relation. In primary mouse hippocampal neurons, iCytSnFR detection of cytisine exhibited the same kinetic trends seen in HeLa cell experiments ( Figure 8B). During cytisine application (60 s) from 0.078 to 80 µM, the iCytSnFR_PM fluorescence nearly reached a plateau, and during the washout (90-180 s), the fluorescence decayed back to baseline, though the decay slowed after removal of higher [cytisine]. The _PM construct reached a maximum ΔF/F 0 of ~1.25 at 80 µM, which was approximately 60% of the signal observed in HeLa cell experiments ( Figure 8B). The iCytSnFR_ER detection of cytisine in the ER reflected the trends seen in HeLa cells, with incomplete cytisine wash-in phases and prolonged cytisine washout phases. One observable difference was that the maximum ΔF/F 0 (~1.25) of iCytSnFR_ER  Table 2. Dashed lines indicate dose-response relations that did not approach saturation for the concentration ranges tested; therefore, EC 50 and ∆F max /F 0 could not be determined. (A) iDianiSnFR shows preference for dianicline, with some promiscuity for other nicotinic agonists. (B) iCytSnFR, (C) iCyt_F_SnFR, and (D) iCyt_BrEt_SnFR bind their drug partner, but also respond to other nicotinic agonists. Ch, choline; ACh, acetylcholine; Cyt, cytisine; Diani, dianicline; Nic, nicotine; Var, varenicline; 10FC, 10-fluorocytisine; 9Br10EtC, 9-bromo-10-ethylcytisine.
The online version of this article includes the following figure supplement(s) for figure 3:  . Isothermal titration calorimetry traces, fits, and thermodynamic data. Top row: exemplar heat traces of iCytSnFR, iCyt_F_SnFR, iCyt_BrEt_ SnFR, and iDianiSnFR paired with their drug partners obtained by isothermal calorimetry. The heats for iCytSnFR, iCyt_F_SnFR, and iCyt_BrEt_SnFR were exothermic, while that for iDianiSnFR was endothermic. Middle row: the resulting fits for each iDrugSnFR:drug pair from the integrated heats comprising each series of injections. Bottom row: energy calculations. All iDrugSnFRs show exergonic reactions, but the relative enthalpic and entropic contributions vary among iDrugSnFRs. Data are from three separate runs, mean ± SEM. iDrugSnFR, intensity-based drug-sensing fluorescent reporter.     reached a similar maximum to that of iCytSnFR_PM in neurons, which was not observed in HeLa cell experiments ( Figure 7B). In preliminary HeLa cell experiments with varenicline applied to iCytSnFR, we found much slower kinetics that differed little between the _ER and _PM constructs (data not shown). These findings, which vitiated the use of the iCytSnFR-varencline pair in the cellular experiments, are consistent with the markedly slow kinetics of varenicline-iCytSnFR interactions in the microperfusion experiments (see above).
iCyt_F_SnFR targeted to the PM and ER showed characteristics similar to iCytSnFR in HeLa cells. The _PM construct detected 10-fluorocytisine across a range of concentrations with a return to baseline fluorescence between applications, while the _ER construct detected 10-fluorocytisine with ΔF/ F 0 values that were only 25-33% of those detected at the PM ( Figure 7C). Similar to the iCytSnFR_ ER detection of cytisine, the iCyt_F_SnFR_ER detection of 10-fluorocytisine was much slower than  (200 µM) response (n = 1 area, 3 cells) was monophasic with a single time constant (τ 0ff ) of 0.61 ± 0.02 s (± SE, n = 86 frames, sampling rate of 9.8 frames/s). The red line is a fit to the sum of a negative exponential component (R 2 of 0.98). (C) The decay of the cytisine (15 µM) response (n = 1 area, four cells) was biphasic with time constants (τf off , τs off ) of 1.9 ± 0.2 and 6.6 ± 0.5 s (n = 149 frames, sampling rate of 5 frames/s). The red line is a fit to the sum of two negative exponential components and a constant (R 2 of 0.996). It was significantly better than that of the sum of a single negative exponential term and a constant (F-test, p<0.05). The relative amplitude of the slower decay component (A s /(A s + A f ), where A s is amplitude of the slower component of decay in units of ∆F/F 0 and A f is amplitude of the faster component) was 61%. Inset: neither rate constant changed significantly over the [cytisine] range from 5 to 15 μM. Dashed lines give the average over this range. (D) The decay of the varenicline (2 µM) response (n = 1 area, three cells) was also biphasic with a τf off and τs off of 9 ± 1 s and 150 ± 10 s (n = 178 frames, sampling rate of 1 frame/s), respectively. The A s /(A s + A f ) was 83%. The red line is a fit to the sum of two negative exponential terms and a constant (R 2 of 0.994), and it was significantly better than that to the sum of a single negative exponential term and a constant (F-test, p<0.05).
solution changes and did not return to baseline between applications, though the washout occurs on the order of minutes, rather than tens of minutes as with iCytSnFR_ER ( Figure 7C). The difference in PM and ER detection of 10-fluorocytisine again shows decreased membrane permeability into HeLa cells compared to other drugs we have examined with other iDrugSnFRs. Overall, the detection of 10-fluorocytisine with iCyt_F_SnFR in primary hippocampal culture resembled our data with iCyt_F_ SnFR in HeLa cells. Nevertheless, there were distinct differences ( Figure 8C), such as a decreased maximum ΔF/F 0 in the iCyt_F_SnFR_PM construct and a similar maximum ΔF/F 0 of ~1 for both the _ER and _PM constructs. Additionally, the decay of the iCyt_F_SnFR responses lasted tens of minutes, resembling the iCytSnFR_ER data in primary hippocampal culture.   9-Bromo-10-ethylcytisine showed a kinetic profile resembling dianicline. iCyt_BrEt_SnFR_PM responses to 9-bromo-10-ethylcytisine (0.1-31.6 µM) were nearly limited by solution exchanges with a return to baseline fluorescence on the order of seconds, and a maximum ΔF/F 0 of ~1.5 at 31.6 µM. iCyt_BrEt_SnFR_ER also detected 9-bromo-10-ethylcytisine over this range of concentrations and returned to baseline fluorescence between applications ( Figure 7D). ΔF/F 0 values for iCyt_BrEt_ SnFR_ER were 50-75% of the ΔF/F 0 values detected by iCyt_BrEt_SnFR_PM, which indicated that 9-bromo-10-ethylcytisine crossed into and out of cells readily ( Figure 7D). Imaging in primary mouse hippocampal neurons revealed the same trend ( Figure 8D). To more fully examine the membrane-crossing properties of the nicotinic agonists, we recorded the fluorescence waveforms for several drugs at concentrations between 0.1 and 3.16 µM with much longer application times and washout times than in the above experiments (Figure 7-figure  supplement 1). With these conditions, the fluorescence signals suggested complete washout of each nicotinic agonist from the ER of HeLa cells. However, it is noteworthy that even when applied at concentrations as low as 0.1 µM and 0.316 µM, cytisine and 10-fluorocytisine require washout times of several minutes from the ER. In contrast, the iDrugSnFR localized to the PM shows a rapid return to baseline after drug application.
Because the data of Figures 7B and 8B indicated that iCytSnFR_PM functions as expected from stopped-flow and millisecond perfusion, we applied additional experiments to ensure that our observations of drug entry and exit from the ER were not the result of idiosyncratic biosensor function or folding in the ER. iCytSnFR and iCyt_F_SnFR both bind nicotine in the same concentration range as cytisine (though with lower ΔF/F 0 ). After transfection of _PM and _ER constructs for each sensor into HeLa cells, we performed time-resolved imaging for pulses of 0.1-31.6 µM nicotine (Figure 7-figure  supplement 2). These nicotine waveforms resembled those already published with iNicSnFR3a and iNicSnFR3b , confirming that iCytSnFR_ER functions as expected with a more permeant nicotinic drug. Thus, the slower kinetics for iCytSnFR_ER with cytisine and iCyt_F_SnFR_ER with 10-fluorocytisine arise because these drugs cross membranes more slowly.
To examine localization of the _PM and _ER constructs at higher optical resolution, we imaged HeLa cells and primary mouse hippocampal culture using a spinning disk laser scanning inverted confocal microscope. As previously observed, ER-targeted iDrugSnFR was retained in the ER (Figure 7 (most likely as part of the cellular membrane trafficking system; Figure 7-figure supplement 3, Figure 8-figure supplement 1).
Several complexities in the HeLa cell and neuron experiments imposed uncertainties on our kinetic analyses. These complexities include the limitations of solution changes, diffusion within cytoplasm, unknown mixing at the surface facing the coverslip, and corrections for baseline drift due to bleaching. We restrict the quantitative comparisons to the estimate that cytisine and 10-fluorocytisine cross the membrane >30-fold more slowly than the other drugs tested.

Discussion
Membrane permeation of molecules with low logD pH7.4 The experiments show, to our knowledge, the first time-resolved measurements of membrane permeation for drugs in the logD pH7.4 range less than -1. Most orally available drugs have logD pH7.4 values between 2 and 4 (Smith et al., 2012). Cytisine, varenicline, dianicline, and the cytisine analogs studied here have calculated membrane partition coefficients some 3-6 orders of magnitude lower. These values and their order vary according to the algorithm, partially because of uncertainties in predicting pK a (Pieńko et al., 2016); here, we provide values calculated by Chemicalize (see Materials and methods): 10-fluorocytisine, -2.70; cytisine, -2.64; dianicline, -1.29; varenicline, -1.27; 9-bromo-10-ethylcytisine, -1.13. It is remarkable that drugs with such low calculated partition coefficients do cross membranes on a time scale of seconds (9-bromo-10-ethylcytisine, varenicline, dianicline) to minutes (10-fluorocytisine, cytisine). According to some (but not all) algorithms, the calculated logD pH7.4 values fall in the same two classes as the measured kinetics of membrane permeability: 10-fluorocytisine and cytisine are the slowest, and only these two agonists have logD pH7.4 values < -2. These observations support previous work suggesting that differences among chemical properties of nicotinic partial agonists correlate with drug permeation into the cerebrospinal fluid (CSF) after peripheral administration in mice (Rollema et al., 2010).

The iDrugSnFR paradigm
The iDrugSnFRs are sensitive enough to allow experiments near the experimentally determined (or otherwise projected) concentration in the human blood and CSF (Astroug et al., 2010;Jeong et al., 2018;Rollema et al., 2010). The iDrugSnFRs have the advantage that they measure free aqueous ligand concentration ('activity'), as sensed by nAChRs. Targeting sequences provide for visualization within the lumen of organelles -here, the ER.
The experiments do not use radiolabeled drugs, in vivo microdialysis or other experiments on live animals, or mass spectrometry-liquid chromatography instruments. Once protein design has given an entry into a class of iDrugSnFRs, straightforward optimization at the binding site produces the desired, selective iDrugSnFRs for individual molecules. For drugs that bind at orthosteric cholinergic sites (both nicotinic and muscarinic), we anticipate that a collection of tens, rather than hundreds, of iDrugSnFRs will suffice to detect all present and future ligands. The experiments use standard, modest-power fluorescence microscopes. Cultured cell lines yield data comparable to cultured neurons.
We comment on varenicline. None of the biosensors in Table 1 were evolved to bind varenicline; yet it binds to some iDrugSnFRs with nanomolar EC 50 . Only iDianiSnFR, which lacks His68, binds varenicline with EC 50 > 10 μM. Even tighter binding has been achieved with varenicline derivatives at mutated ligand-gated channels (Magnus et al., 2019). On the one hand, the cellular experiments described here cannot use iDrugSnFR pairs with dissociation rate constants less than ~0.1 s -1 , corresponding to an EC 50 of less than ~100 nM. On the other hand, all known neural drugs leave the human body and brain much more slowly, with rates determined primarily by metabolism; even 'fast' nicotine metabolizers display time constants of ~1200 s (Dempsey et al., 2004). Highly sensitive, tightly binding, reagentless iDrugSnFRs will be used in studies on personal pharmacokinetics in biofluids.
This study presents a general step forward in understanding the structure-function relations of iDrugSnFRs. The chromophore in the cpGFP moiety of most present iDrugSnFRs (this paper, iGluSnFR, iSeroSnFR) contains a tyrosine in an extended π system (Ormö et al., 1996;Tsien, 1998). The photophysics of the chromophore depends strongly on the surrounding water molecules and side chains (Brejc et al., 1997;Tsien, 1998). We found that Glu78 in Linker 1 changes its orientation: in the liganded state, it interacts with two positively charged residues (Lys97 and Arg99) on the surface of the cpGFP; and in the apo state, Glu78 has moved ~14 Å to form a hydrogen bonding interaction with the tyrosine moiety of the chromophore. Presumably the liganded state of iNicSnFR3adt allows for a water molecule to hydrogen bond with the hydroxy group of the chromophore, promoting its fluorescence; but this water molecule is replaced by protonated Glu78 in the unliganded state, which leads to nonfluorescent state of cpGFP, as suggested by Nasu et al., 2021. While we cannot resolve the protonation-deprotonation event, the available functional data show good support for its occurrence, as follows. (1) The apo form of the iDrugSnFR increases its F 0 by 10-fold per pH unit , as though when deprotonated, Glu78 leaves the 'candle snuffer' position and moves to make the salt bridges with Lys97 and Arg99. (2) The EC 50 for the ligand decreases by 10-fold per pH unit , as though the conformation of the linker that forms the salt-bridge form is also the closed, liganded, fluorescent form of the PBP. Other observations favor the crucial role of the Glu78-chromophore interaction. (3) Only glutamate functions in position 78 of iSeroSnFR (Unger et al., 2020). (4) The mTurquoise variant in iGluSnFR, which has a tryptophan chromophore, requires entirely different linkers (Marvin et al., 2018).

Challenges at the intersection of pharmaceutical science and nicotine addiction science
Our measurements show that nicotinic ligands with logD pH7.4 < ~-2 cross membranes much more slowly than do ligands with logD pH7.4 > ~-2. These measurements have two, possibly opposing, implications for future smoking-cessation drugs. On the one hand, α4β2 agonists that enter the ER, like nicotine and varenicline, upregulate nAChRs (Turner et al., 2011), which may be necessary and sufficient for addiction (Henderson and Lester, 2015); and maintenance of upregulation by varenicline may help to explain its suboptimal quit rate. On the other hand, ligands that do not enter the ER are also unlikely to enter the brain and therefore unlikely to be useful for smoking cessation.
Smoking-cessation drugs must also contend with other ER-based processes.
(1) Most drug metabolism takes place in the ER and (2) upregulation occurs at a sustained agonist concentration in the ER some hundredfold lower than the extracellular concentrations that transiently activate α4β2 nAChRs (Kuryatov et al., 2005).
Given these challenges, further progress may be possible now that we have two types of real-time, living cellular preparations. (1) For decades, cellular preparations have been available to measure nAChR pharmacodynamics and upregulation. (2) Now, the iDrugSnFRs present a paradigm to measure cellular and subcellular pharmacokinetics. The iDrugSnFR paradigm will be useful beyond the explicit case of nicotine addiction, with application to other exogenous neural drugs.

Crystallography
The gene encoding the full-length biosensor iNicSnFR3a was previously cloned into a bacterial expression vector . To improve crystallization, we deleted the N-terminal HA tag and the N-terminal Myc tag, forming the constructs with the suffix 'dt.' These deletions were carried out with the Q5 Site-Directed Mutagenesis Kit (New England Biolabs, Ipswich, MA). All proteins were overexpressed in E. coli BL21-gold (DE3) cells (Agilent Technologies, Santa Clara, CA) using ZYM-5052 autoinduction media (Studier, 2005). Cells were collected by centrifugation and stored at -80°C until use.
For purification, frozen cell pellets were resuspended in lysis buffer containing 100 mM NaCl, 20 mM Tris, pH 7.5, 20 mM imidazole, pH 7.5, 5 mM β-mercaptoethanol (BME), lysozyme, DNase, and protease inhibitor tablet. The resuspended cells were lysed by freezing and thawing using liquid nitrogen and a room temperature water bath for three cycles. Intact cells and cell debris were removed by centrifugation at ~20,000 × g for 40 min at 4°C. The supernatant was collected and loaded onto a prewashed Ni NTA column with wash buffer at 4°C. Ni NTA wash buffer contained 100 mM NaCl, 20 mM Tris, pH 7.5, 30 mM imidazole, pH 7.5, and 5 mM BME. Elution was achieved using the same buffer with 300 mM imidazole, pH 7.5. The eluted sample was further purified by size-exclusion chromatography using HiLoad 16/60 Superdex 200 in the same buffer without imidazole and BME. Peak fractions were collected and concentrated to ~50 mg/ml with Amicon Ultra 15 filter unit (Millipore, Burlington, MA) with 10 kDa cutoff.
X-ray datasets were collected at Stanford Synchrotron Radiation Laboratory beamline 12-2 and Advanced Light Source beamline 5.0.2 using Pilatus 6M detectors. All datasets were processed and integrated with XDS (Kabsch, 2010) and scaled with Aimless (Winn et al., 2011). For iNicSnFR3adt, molecular replacement was carried out using domains of the unliganded structure (PDB ID 6EFR) with Phaser in Phenix (Adams et al., 2010). The experimental phase information of KI-soaked crystals of iNicSnFR3adt was obtained with MR-SAD using AutoSol in Phenix (Adams et al., 2010). Molecular replacements of the remaining structures were carried out with the refined model of iNicSnFR3adt. Iterative refinement and model building cycles for all structures were carried out separately with phenix. refine in Phenix (Adams et al., 2010) and Coot (Emsley et al., 2010).

Directed evolution of iDrugSnFR proteins using bacterial-expressed protein assays
Starting with iAChSnFR and intermediate biosensor constructs of that sensor, we constructed and optimized iDrugSnFRs for each drug partner during iterative rounds of SSM as previously described Shivange et al., 2019). We utilized the 22-codon procedure including a mixture of three primers, creating 22 unique codons encoding the 20 canonical amino acids (Kille et al., 2013). The 22-codon procedure yields an estimated >96% residue coverage for a collection of 96 randomly chosen clones.
A Tecan Spark M10 96-well fluorescence plate reader (Tecan, Männedorf, Switzerland) was used to measure baseline and drug-induced fluorescence (F 0 and ΔF, respectively). Bacterial lysates were tested with excitation at 485 nm and emission at 535 nm. Lysates were also measured against choline to evaluate potential endogenous intracellular binding. Promising clones were amplified and sequenced. The optimally responding construct in each round of SSM was used as a template for the next round of SSM.
S-slope allows for comparison between iDrugSnFRs with differing ΔF max /F 0 values  at the beginning of the dose-response relation, which is usually the pharmacologically relevant range. With lysates or purified protein, which allow complete dose-response relations, the Hill coefficient is near 1.0. We therefore calculated in units of µM -1 .

Measurements of purified iDrugSnFRs
Biosensors selected for further study were purified using a His 6 sequence using an ÄKTA Start FPLC (GE Healthcare, Chicago, IL) as previously described . Performance of protein quantification and dose-response relations for drug-sensor partners was also as previously described . Where appropriate, we corrected for depletion of the ligand by binding with the equation, where K D is the ligand-sensor equilibrium dissociation constant (we assume that K D = EC 50 ), [S] is the iDrugSnFR protein concentration (typically 100 nM), and [L] is the nominal ligand concentration.

Stopped-flow kinetic analysis
Kinetics were determined by mixing equal volumes of 0.2 µM iDrugSnFR protein (in 3× PBS, pH 7.0) with varying concentrations of cognate ligand in an Applied Photophysics (Surrey, UK) SX20 stoppedflow fluorimeter with 490 nm LED excitation and 510 nm long-pass filter at room temperature (22°C). 'Mixing shots' were repeated five times and averaged (except for 100 s experiments, which were collected only once). Standard deviations are not included on the plots, but are nearly the same size as the data markers. The first 3 ms of data were ignored because of mixing artifacts and account for the dead time of the instrument.
Data were plotted and time courses were fitted, when possible, to a single exponential, and k obs was plotted as a function of [ligand]. The linear portion of that graph was fit, with the slope reporting k 1 and the y-intercept reporting k -1 . When the time course did not fit well to a single rising exponential, it was fitted to the sum of two increasing exponentials, and the first rise (k obs1 ) was treated as above to determine k 1 and k -1 .

Expression in mammalian cells
We constructed two variants of each iDrugSnFR for expression in mammalian cells. The PM (suffix _PM) and ER (suffix _ER) variants were constructed by circular polymerase extension cloning (Quan and Tian, 2009). To create the _PM constructs, we cloned the bacterial constructs into pCMV(MinDis), a variant of pDisplay (Thermo Fisher Scientific, Waltham, MA) lacking the hemagglutinin tag (Marvin et al., 2013). To generate the _ER constructs, we replaced the 14 C-terminal amino acids (QVDE-QKLISEEDLN, including the Myc tag) with an ER retention motif, QTAEKDEL .
We transfected the iDrugSnFR cDNA constructs into HeLa and HEK293T cells. Cell lines were purchased from ATCC (Manassas, VA) and cultured according to ATCC protocols. We purchased new aliquots of the cell lines listed above at 6-month intervals to ensure reproducibility. Mycoplasma contamination was assayed at 6-month intervals and was negative over the course of these experiments. For chemical transfection, we utilized either Lipofectamine 2000 or Lipofectamine 3000 (Thermo Fisher Scientific) following the manufacturer's recommended protocol. Cells were incubated in the transfection medium for 24 hr and imaged 24-48 hr after transfection.

Millisecond timescale microperfusion
HEK293T cells were imaged using a Nikon (Tokyo, Japan) DIAPHOT 300 with a Zeiss ×63 objective (1.5 NA). Because the ligand concentration after micro-iontophoretic drug application  is unknown, we applied drugs with a laminar-flow microperfusion (model SS-77B Fast- Step perfusion system; Warner Instruments, Holliston, MA). In an array of three square glass capillaries (600 µ i.d.), the center capillary contained vehicle (Hanks buffered salt solution [HBSS]) plus drug, while the two outer capillaries contained vehicle only. Vehicle also flowed from a separate input connected to the bath perfusion system. Solution exchange, measured by loading the center capillary with dye, had a time constant of 90 ± 20 ms (n = 6 trials).
We used Fiji ImageJ and Origin Pro 2018 (OriginLab, Northampton, MA) to fit the rise and decay of the iCytSnFR_PM drug response to the sum of one or two exponential components. An F-test determined whether two exponential components fit the data significantly better than 1 (p<0.05). Statistical comparisons between groups were carried out using ANOVA.

AAV production and transduction in primary mouse hippocampal neuronal culture
The adeno-associated virus plasmid vector AAV9-hSyn was described previously (Challis et al., 2019). Virus was purified using the AAVpro Purification Kit (TakaraBio USA). Mouse embryo dissection and culture were previously described . About 4 days after dissection, we transduced the _ER construct at an MOI of 0.5-5 × 10 4 ; and separately, the _PM construct was transduced at an MOI of 0.5-1 × 10 5 . Neurons were imaged ~2-3 weeks post-transduction.

Time-resolved fluorescence measurements in live mammalian cells and primary mouse hippocampal neuronal culture
Time-resolved dose-response imaging was performed on a modified Olympus IX-81 microscope (Olympus Microscopes, Tokyo, Japan), in widefield epifluorescence mode using a ×40 lens. Images were acquired at 2-4 frames/s with a back-illuminated EMCCD camera (iXon DU-897, Andor Technology USA, South Windsor, CT), controlled by Andor IQ3 software. Fluorescence measurements at λ ex = 470 nm and the epifluorescence cube were as previously described Srinivasan et al., 2011).
Solutions were delivered from elevated reservoirs by gravity flow, via solenoid valves (Automate Scientific, Berkeley, CA), then through tubing fed into a manifold, at a rate of 1-2 ml/min. The vehicle was HBSS. Other details have been described Srinivasan et al., 2011). Data analysis procedures included subtraction of 'blank' (extracellular) areas and corrections for baseline drifts using Origin Pro 2018.

Spinning disk confocal fluorescence images
HeLa cells and mouse primary hippocampal culture were transfected or transduced as described above. Live-cell images were collected using a Nikon Ti-E spinning disk laser scanning confocal inverted microscope equipped with ×100 objective, 1.49 NA (oil), 120 μm WD. The laser wavelength was 488 nm at 15% power. Dishes were imaged in a custom incubator (Okolab, Ottaviano, Italy) at 37°C and 5% CO 2 . Initial images were taken in HBSS. To add drug, we doubled the bath volume by adding HBSS containing drug using a hand-held pipette. The final drug concentrations were dianicline, 15 µM; cytisine, 10 µM; 10-fluorocytisine, 10 µM; and 9-bromo-10-ethylcytisine, 7.5 µM.

Plasmid availability
We will deposit plasmids with the following cDNAs at Addgene: We will deposit the following plasmids at Addgene:

Ethics
Generation of primary mouse hippocampal culture was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to the approved institutional animal care and use committee (IACUC) protocols (IA19-1386) of California Institute of Technology. Dissections were performed after euthanasia of the pregnant mouse and every effort was made to minimize suffering.

Appendix 1 iDrugSnFR structural transition
Because all our iDrugSnFRs derive from the Thermoanaerobacter sp. X513 choline/betaine binding protein, we asked whether the design process affected conformational changes between the nonfluorescent and fluorescent states. Principal component analysis showed that component 1 accounted for 75% of the total conformational change, which was associated with a hinge-like motion. Component 2 accounted for 17% of the total conformational change, which was associated with the overall conformational difference between the PBPs of iDrugSnFRs and the other published betaine and choline binding proteins. The liganded and apo biosensor structures partitioned into two groups (Figure 1). Component 1 accounted for the Venus-flytrap conformational change and was represented best by the structural alignment of varenicline bound to iNicSnFR3adt and the apo iNicSnFR3adt (Figure 1-figure supplement 1). As in other betaine or choline binding protein structures in the Protein Data Bank, the liganded conformations differed little from each other, and the conformational change between liganded and apo states related to the Venus-flytrap mechanism. Component 2 accounted for a small fraction of the conformational differences between the native structures and optimized biosensor. This small difference was also observed when the choline-bound iAChSnFRdtbp and Bacillus subtilis structures were aligned (RMSD of 1 Å) (Fan, 2020).

Appendix 2
Further details of kinetic experiments: 'Methods and results' Millisecond timescale microperfusion with iCytSnFR_PM Cell culture The iCytSnFR_PM sensor was expressed in HEK293T cells because the high level of protein expression provided by this cell line improved the fluorescent signal/noise ratio. The cells were plated at a density of 1.25 × 10 5 cells per dish on gelatin-coated glass coverslips attached to the bottom of a 35 mm culture dish and incubated at 37°C. 24 hr after plating, the cells were transfected with iCytSnFR_PM DNA (200 ng per dish) using Lipofectamine 3000 (Thermo Fisher Scientific). The transfection medium was removed after a 24 hr incubation and replaced with fresh growth medium. We incubated the transfected cells for another 24-72 hr to obtain sufficient iCytSnFR_PM membrane protein density for fluorescent imaging.

Fluorescent microscope
A blue LED (488 nm wavelength) provided epifluorescent illumination. The bandwidth and peak of the excitation and emission spectra were 460-500 nm and 480 nm (excitation), and 515-550 nm and 535 nm (emission), respectively (Cat. #41001 FITC filter cube). Immediately prior to imaging, we removed the culture dishes from the incubator and rinsed them three times with HBSS warmed to 37°C. After rinsing, we pried the coverslip off the bottom of the dish with a single-edged razor blade and resealed it to the bottom of a plastic perfusion chamber (model RC-25 diamond shape, Warner Instruments) with silicone grease. At this point, the chamber and coverslip were mounted on a motorized microscope stage and perfused continuously with HBSS at 18°C from a gravity-fed, bath perfusion system to keep the cells oxygenated. The fluid level in the chamber was regulated by suction from a syringe needle.

Drug microperfusion
The major components of the microperfusion system were (1) a parallel array of three, square glass capillaries (600 µm i.d.) that delivered the ligand to the cells, (2) a stepping motor that translated the capillary array horizontally above the coverslip surface, and (3) a controller that determined the motion and step size of the motor. Both the array and stepping motor were mounted on a motorized micromanipulator (model MP-285, Sutter Instrument Co., Novato, CA) that allowed us to position it in the chamber. Fluid flowed continuously from all three capillaries into the chamber during the experiment, as well as vehicle from a separate input connected to the bath perfusion system. The center capillary in the array contained vehicle plus drug, while the two outer capillaries contained vehicle only. Outflow from the end capillaries kept ligand flow from the center capillary confined to a laminar stream. We applied drugs to the cells in 5-30 s step-like pulses. The array was positioned so that outflow from an end capillary bathed the cells in the field of view with vehicle initially. To apply the ligand to the cells, we stepped the entire array horizontally by 700 µm so that vehicle plus drug from the center capillary bathed the cells. Ligand application was terminated by returning the array to its initial position.

Data acquisition
A computer equipped with the Clampex v.9 software and a Digidata 1200 series interface (Molecular Devices, San Jose, CA) controlled both the timing of ligand application and initiated image acquisition. A camera (model ORCA-3G, Hamamatsu, Hamamatsu City, Japan) attached to a microscope port recorded the images. A second computer equipped with the HCImage V3.0 software and a digital camera interface (Hamamatsu) controlled the camera operation, set the parameters of image acquisition, and stored the recorded image sequences. We recorded image sequences of 170-200 frames at rates of 1-8.9 frames/s to visualize the time course of the iCytSnFR_PM response. Event markers in the image file marked the timing of ligand application. We started acquiring images 5-10 s before applying the ligand to establish a baseline fluorescence and continued for 5-180 s after applying drug to record the decay of the response.
Millisecond timescale microperfusion with iCytSnFR_PM (results) Biphasic decay components of the cytisine response

Growth phase of the cytisine response
We visualized the time course of the growth phase of the iCytSnFR_PM response to 1-15 µM cytisine using the Fast-Step microperfusion system (Appendix 2-figure 1). The growth phases of responses to 30 s applications of 1-4 µM cytisine were biphasic and fitted best by the sum of two terms describing an exponential rise to a maximum (Appendix 2- figure 1A). The range of the faster rate constants (kf on s) for 1-4 µM cytisine was 0.41-1.22 s -1 (n = 40 areas [138 cells]), and the range for the slower rate constants was 0.01-0.1 s -1 (n = 37 areas [128 cells]). The faster component dominated the rising phase of the 1-4 µM cytisine response. The relative amplitude of the faster component A f /(A f + A s ) was 77% ± 1%, n = 37 areas (128 cells). It was not affected significantly by cytisine concentration in the 1-4 µM range (ANOVA, df = 3 model, 33 error, p=0.08). The responses to 5-10 s applications of 5-15 µM cytisine reached steady state within a 5 s application. They were fitted adequately by a single negative exponential rise to a maximum (Appendix 2- figure 1B). The single rate constant for the rising phase in individual areas ranged from 0.86 to 2.65 s -1 (n = 37 areas [133 cells]). We pooled the fast rate constants for the 1-4 µM cytisine responses with the single rate constants for the 5-15 µM responses to obtain an overall [cytisine]-kf on relation for the 1-15 µM cytisine concentration range (Appendix 2- figure 2A). Linear least-squares regression showed that the [cytisine]-kf on relation over this concentration range was approximately linear (r = 0.98, n = 8 concentrations, p<0.05) with a slope and intercept of 1.5 ± 0.1 10 5 (Ms) -1 and 0.43 ± 0.05 s -1 (± SE), respectively. Nevertheless, the [cytisine]-kf on data from the microperfusion experiments deviated from linearity at cytisine concentrations ≥ 8 µM, suggesting that [cytisine]-kf on relation for the microperfusion data was more hyperbolic than linear. Using the slower decay rate constant for cytisine (0.146 ± 0.006 s -1 ) as the value for the kf on at 0 µM cytisine, a hyperbolic relation fitted the cytisine-kf on relation for 0-10 µM cytisine significantly better than a straight line (Appendix 2- figure 2B, F-test, p<0.05).
In contrast to the faster rate constant of the rising phase kf on , the slower rate constant ks on decreased significantly as the cytisine concentration increased from 1 to 4 µM (r = 0.78, n = 4 concentrations, p<0.05, Appendix 2- figure 2D). The slope and intercept of a straight line fit to the [cytisine]-ks on data using least-squares regression were -1.5 ± 0.06 × 10 4 M -1 s -1 and 0.09 ± 0.2 s -1 (± SE, n = 4 concentrations), respectively.

Comments on a kinetic model
Our subsecond data with cytisine at iCytSnFR are the most complete, comprising both stopped-flow and HEK293T cell microperfusion. We therefore discuss the cytisine-iCytSnFR kinetics.
Recent literature on kinetics of PBPs emphasizes a modified induced-fit model with the additional possibility that the apo PBP can also undergo spontaneous activation (termed the 'closed state' in the SBP literature; de Boer et al., 2019;Gouridis et al., 2015). Such a scheme, shown in Appendix 2figure 3, resembles the three-state model we and colleagues developed to account for iSeroSnFR (Unger et al., 2020).
We simulated the scheme using the MATLAB Simbiology toolbox. The following rate constants account for the stopped-flow and millisecond perfusion data within a factor of 3: k bind2 , 0.5 × 10 6 / M -1 s -1 ; k unbind2 , 0.4/ s; k iso(+) , 0.001/ s; k iso(-) , 0.01/ s; the apo fluorescent state has 0.1 times the brightness of the bound fluorescent state.
The three-state model predicts the experimental observation (Appendix 2- figure 2D) that the rate constant of the slower component of the kinetics decreases as the ligand concentration increases. For the cytisine-iCytSnFR case reported in this paper, we conclude that the apo, fluorescent state is less bright than the bound state (shown by the different colors of the cpGFP moiety).
Interestingly, the three-state model fitted the kinetic data best if we assumed that there is a population of higher-sensitivity iDrugSnFRs in HEK293T cells, with an EC 50 at least 10 times less than we observed with the stopped-flow and HeLa cell data. A more complex, 'square' four-state model, comprising both ligand binding and protein conformational changes, has been applied to equilibrium measurements of cGaMP (Barnett et al., 2017). Full kinetic predictions of the four-state model are available (Lancet and Pecht, 1976).
The fragmentary kinetic data for acetylcholine at iCytSnFR_PM suggest equilibrium and rate constants in the same broad range as for cytisine. However, the kinetic data for varenicline suggest that ligand unbinding dominates the decay phase, with a rate constant <0.01 s -1 . As a caution, recent data show that mechanisms at PBPs (part of the larger class of substrate binding proteins [SBPs]) can change fundamentally with even a single mutation (Nguyen et al., 2018). We therefore wish to avoid generalizing past the single iDrugSnFR we consider here.
Appendix 2-figure 1. Rising phase of the iCytSnFR_PM response to cytisine in HEK293T cells. (A) Example of the biphasic rising phase of a 1 µM cytisine response in an individual area (black trace, mean of four cells). Cytisine was applied for 30 s. The fast and slow time constants of the rising phase (τf 0n , τs 0n ) were 2.15 ± 0.05 s and 17 ± 3 s (n = 151 frames, sampling rate of 5 frames/s), respectively. The red line is a fit to the sum of two declining exponentials (R 2 of 0.998). It was significantly better than that to a single negative exponential rise to maximum component and a constant term (F-test, p<0.05). The A s /(A s + A f ) was 20%. (B) Example of the rising phase of a response to a 5 s application of 5 µM cytisine in an individual area (mean of 10 cells, two replicates). The response appeared to be monophasic with a single time constant (τ 0n ) of 0.76 ± 0.04 s (n = 43 frames, sampling rate of 9.8 frames/s). The red line is a fit to the sum of a negative exponential rise to a maximum, and constant, term (R 2 of 0.98).
(iDrugSnFR) exists in an apo nonfluorescent state and an apo fluorescent state; these states interconvert with time constants of tens of seconds k( iso(+) , k iso(-) ). Cytisine binds to the apo fluorescent state (k bind2 ), inducing an additional fluorescent state on a briefer time scale. The initial fluorescence increase represents the binding-induced increase, and the slower increase is governed by partial re-equilibration of the two apo states. Upon removal of cytisine after just a few seconds of perfusion (Figure 6), the fluorescence decay represents the dissociation of cytisine (k unbind2 ). This scheme resembles the model we and colleagues developed to account for iSeroSnFR (Unger et al., 2020). For the iDrugSnFRs reported in this paper, we conclude that the apo, fluorescent state is less bright than the bound state (shown by the different colors of the cpGFP moiety).