Single-molecule nanopore sensing of actin dynamics and drug binding

Actin is a key protein in the dynamic processes within the eukaryotic cell. To date, methods exploring the molecular state of actin are limited to insights gained from structural approaches, providing a snapshot of protein folding, or methods that require chemical modifications compromising actin monomer thermostability. Nanopore sensing permits label-free investigation of native proteins and is ideally suited to study proteins such as actin that require specialised buffers and cofactors. Using nanopores we determined the state of actin at the macromolecular level (filamentous or globular) and in its monomeric form bound to inhibitors. We revealed urea-dependent and voltage-dependent transitional states and observed unfolding process within which sub-populations of transient actin oligomers are visible. We detected, in real-time, drug-binding and filament-growth events at the single-molecule level. This enabled us to calculate binding stoichiometries and to propose a model for protein dynamics using unmodified, native actin molecules, demostrating the promise of nanopores sensing for in-depth understanding of protein folding landscapes and for drug discovery.


Introduction
Actin is a ubiquitous and highly conserved ATPase found in all eukaryotes and is involved in a myriad of cellular processes including the formation of canonical eukaryotic cytoskeletal structures, cell division and cell movement 1 . Actin can also play a significant role in human diseases with rare point mutations in the actin molecule leading to aberrant aggregation and pathologies such as actin-accumulation myopathy 2 . There is also an emerging interest in the actin molecule serving as a potential drug target to stem disease. This includes targeting divergent pathogens that rely on their own actin dynamics for infectivity, such as the malaria parasite 3 , or other pathogens that utilise host cell actin for infection, such as bacteria 4 . In the human cell, actin has hundreds of binding partners, many of which facilitate the dynamic interplay between monomeric G-and filamentous F-actin forms 1 . G-actin forms a stable globular monomeric protein with intrinsic flexibility and ATPase activity [5][6][7] , which spontaneously forms filamentous F-actin above a critical concentration.
Understanding actin dynamics, how the molecule is folded, and how it interacts with other co-factors including drugs is therefore of keen interest and advances in technologies that can explore these actin dynamics offer significant potential for the discovery of therapeutics that target host cellular processes or pathogen infection.
To date, the protein dynamics and drug binding studies of native actin have been a complex and challenging problem. For most proteins, folding is an intrinsic property of the amino acid sequence, wherein hydrophobic residues are buried to prevent aggregation, and hydrophilic residues are exposed to the aqueous environment of the cell cytoplasm. Actin does not follow this canonical folding pathway and instead relies on the speciesspecific chaperonin containing TCP-1 (CCT) complex. The CCT complex facilitates the addition of the co-factors ATP and divalent metal ion to fold actin into a native, polymerisation competent form 8 without which the molecule is only partially functional 9 . Methods such as atomic force microscopy (AFM) 10 , cryo-electron microscopy (EM) 7 , crystallography 11 and tomography 12 have improved our understanding of actin in its G-and F-states. However, these non-time resolved approaches provide only static information and are yet to elucidate a protein folding pathway and characterise the intermediate actin folding states. While valuable information can be obtained using fluorescently labelled actin, usually by the addition of fluorophores such as N-(1-pyrene)iodoacetamide (pyrene) to the C-terminus, these chemical modifications alter the thermostability of the protein. In result, altered monomoer thermostability introduces substantial challenges in correlating polymerisation kinetics and free energy values to a native, unlabeled molecule 13,14 . Limitations also persist when studying drug binding and its mechanisms 15,16 . For example, the pyrenylation of actin is a common way to study drug binding, but this fluorophore only labels a minority of actin molecules and provides a readout at the level of the population, masking sub-population changes in polymerisation kinetics. Furthermore, very little information about how compounds specifically interact with actin species can be inferred from this method. Precise details of drug inhibition or stimulation and its mechanisms have traditionally relied on the structural characterisation 17,18 , which has been instrumental in identifying the binding sites of various drugs but is time-consuming and as with other techniques provides only a static image of a protein in a non-aqueous environment.
Many of the limitations above can be addressed by using single-molecule nanopore sensing. In recent years, nanopores have been used for label-free biosensing of some of life's fundamental building blocks such as nucleic acids 19 and proteins 20,21 . In a typical experiment, analytes are electrophoretically or electroosmotically translocated through a nanopore by an externally applied electric field. The translocation events lead to a characteristic temporal change of the measured ionic current passing through the nanopore. From these changes in the ionic current, one can extract information of the analyte molecular properties such as size, charge, conformational states and interactions with other biomolecules 22 . Both biological and solid-state nanopores have been used to study protein folding at the single-molecule level, revealing the conformational change and dynamics during protein unfolding [23][24][25][26] , and have also been used to observe macromolecular changes of proteins 27 . Quartz nanopipettes, a sub-class of solid-state nanopores, are low-cost and straightforward to fabricate, circumventing the technical barrier of using conventional and expensive solidstate nanopores or biological nanopores.
Here, we use quartz nanopipettes to measure real-time kinetics of the actin molecule in monomeric and polymerised state and its interaction with actin-binding drugs. Using this platform, we can distinguish between the two different macromolecular actin forms, G-and F-actin. We use the system to observe the dynamics of the unfolding of native actin using urea denaturation via measurement of thousands of single-molecule events, and we show that it is possible to observe the F-actin growth in real-time. Critically, we are able to distinguish between the binding of two drugs, Latrunculin B and Swinholide A, that prevent filament formation Fig. 1 Nanopore detection of actin macromolecular states. a Schematic of experimental set-up of protein detection using nanopipette-based nanopore. The nanopipette and external bath were filled with a monomeric ADP-actin calcium buffer. An Ag/AgCl working electrode was inserted into the pipette, and an Ag/AgCl reference/ground electrode was fixed in the bath where the protein was introduced. Under external electric field, actin molecules in different states were translocated from the bath to the inside of the pipette through a conical nanopore at the pipette tip. Right insert: a typical SEM image showing that the nanopipette tapers down at the tip to form a nanopore with the diameter of 25 nm (scale bar 200 nm, top; 50 nm, bottom. b Top: nanopore translocations of three different actin species (from left to right): G-actin monomer (42 kDa, PDB: 1NWK, 800 nM), drug-induced actin dimer (PDB: 1YXQ, 500 nM) and F-actin (up to 200 actin subunits, PDB: 3G37, 800 nM). Bottom: scatterplots of current blockades vs dwell times for different actin species at 250 mV. Corresponding molecular structures are shown in the inset. c Box and whisker plots showing peak current and dwell time (median line and interquartile range). All data in this figure was recorded at 250 mV, sampled at 1 MHz and low-pass filtered at 50 kHz.
To demonstrate the spatial and temporal resolution of our nanopipettes, we compared the translocation characteristics of three different actin species: monomeric (G)-actin, drug-induced dimer and filamentous (F)actin, at an applied voltage of 250 mV (Fig. 1). Two-dimensional scatterplots of dwell time ( " ) vs peak current combined with the box plots (Figs. 1b and c) show we observed a cluster of events along with outliers outside the confidence interval for three species. These outliers are likely due to the transient adsorption of proteins to the nanopore surface or due to transient oligomer formation during their translocation 30 . There was an apparent expansion in dwell-time distribution when comparing G-actin to F-actin, although the mean values were close (Fig. 1c). The distribution rather than the mean value of dwell time provides key information about different features such as the molecule's electrophoretic and electroosmotic properties or its interactions with the nanopore itself. It is noteworthy that this information relating to filament length is generally masked in ensemble averaging methods that are typically used for filament formation, such as the pyrene fluorescence assay (Fig. S3). In addition, we could also observe a distinct increase in the mean dwell time of actin dimers compared to monomers, which is likely due to a change in their respective electrophoretic mobilities.
The peak current is an essential parameter to estimate the spatial resolution of the nanopipettes, as the current blockade transients, ∆ % , are dependent on the excluded ionic volume, , occupied by individual molecules translocated through the nanopore (eq. 1) 25  (1) where is the solution conductivity, is the applied voltage, @ is the diameter and BCC is the effective length of a nanopore, E is the diameter and E is the length of a protein molecule. The BCC of the nanopipette was determined to be 110 ± 15 nm using 1 kb dsDNA as a standard, as shown in Fig. S9 and Method S1. ( E / @ , E / BCC ) is a correction factor that primarily depends on the relative dimension between molecules themselves and the nanopore amongst others. In terms of small spherical proteins or particles, the excluded volume can be estimated as ≈ ∆ % • BCC L /( ) (model 1) using eq. 1. For long linear molecules E ≫ BCC such as double-strand DNA (dsDNA) and polymeric or filamentous proteins, eq. 1 can be simplified as ∆ % ≈ • E / BCC (model 2), where E is the mean atomic volume per unit length of the molecule.
Model 1 was used for the analysis of the actin dimers, in which the mean peak current is 113.7 ± 15.8 pA, nearly double that of actin monomers (64.8 ± 6.7 pA). The change in current was consistent with the difference in volume between actin in these two states. For F-actin, we observed a larger mean peak current and broader distribution. This was expected since F-actin is made up of monomeric actin subunits. In the actin polymerisation process, ATP-bound monomers assemble to filaments with a 37 nm helical repeat and a 5 -9 nm diameter 31 . Polymerisation resulted in an approximately ten-fold increase in the mean current blockade of 298.5 ± 105.8 pA when compared to actin monomers. This long-tail distribution and variety of blockade currents are likely due to the range of potential actin filament lengths possible, ranging from several subunits to micron-scale. These results demonstrate that we can use a combination of spatial and temporal analysis to reliably define the protein states (from monomer to dimer and filament) of protein-protein assemblies and their kinetics with high resolution.

Actin unfolding using urea as a denaturant
Having established a system for studying dynamics involving the actin monomer in solution, we next sought to explore protein folding at the single-molecule level. Actin folding remains complex and poorly understood, partly due to the vast number of folding intermediates that may exist during folding pathways of any polypeptide, and also due to the intrinsic flexibility of the protein itself, which exists in a complex protein folding landscape 32 . To assess how the nanopore signal is dependent on the protein state, we performed a nanopore analysis of actin monomers exposed to 0 M to 6 M urea and bias of 250 mV. Before the measurements, the nanopipettes were tested and shown to be compatible with the chemical denaturant urea ( Fig. S8), and the solution conductivity for each buffer was measured (Table S1).
The excluded volume increased with increasing urea concentration (Fig. 2). The distribution is not a standard Gaussian distribution as part of the low translocation signals were cut off by the low-pass filter, which we also see for the normalised current blockade (∆ / O ) shown in Fig. S10. This trend was attributed to the increase in the exposed protein surface to the solution. Unfolding increases the solvent interaction with the protein, thus increasing the effective size of the protein and therefore contributing to a greater blockade amplitude 23 . The plot of mean excluded volume vs urea concentrations exhibits a sigmoidal shape and increases from 30.3 ± 2.5 nm 3 to 44.0 ± 3.7 nm 3 and corresponds to a two-state transition from folded to unfolded actin (Fig. 2b). By contrast, an opposite trend was observed for the translocation time whereby the dwell time decreases from 80.6 ± 6.4 µs to 58.1 ± 3.7 µs with increasing urea concentration ( Fig. 2c) indicating the unfolded, linear actin molecules translocate faster than the folded, globular ones. We attribute this increase in translocation speed of unfolded actin to the change of charge distribution that results from the conformational changes, which in turn is associated with electrophoretic mobilities in nanopore translocation.  Fig. S11). It should be noted that this value is buffer-and environment-specific and is, therefore, an estimate of free energy unfolding using this nanopore platform. Despite this, these data suggest that this single-molecule statistical approach is able to describe well the dynamic changes in protein conformations measured from translocation events and can, therefore, be considered a complementary approach to traditional methods, such as fluorescence.

Voltage-dependence on protein conformation
The applied voltage can often play a critical role in the translocation of proteins by affecting protein mobility and conformation 23,35 . For example, the transport of proteins through the nanopore is governed by bulk diffusion, EP and EO flow and therefore the large electric field generated at the tip of the nanopipette can significantly alter the velocity and direction of the molecule and even the conformation. The effective velocity of protein transport can be described as 29 : where = Y Z , is the solution viscosity, is the strength of the electric field, E and @ is the zeta potential of the protein and walls of the nanopore, respectively. We minimised the electroosmotic component using  Fig. S13.
Importantly, the ratio of = e / d can be used to both evaluate the structural changes of the protein during translocation but more importantly relate this to the degree of asymmetry and anisotropy in the protein 39 . We can estimate the RG value of the folded monomer from the crystal structure of G-actin 40 , with a value of approximately 2.45 nm, and of the unfolded protein using a model for denatured proteins, providing an estimate of around 7.0 nm 41 . We can then use these to calculate the ratio between the e and d , with of 0.775 for a globular protein with uniform density. As increases, the protein increases to an ellipsoid. Even in harsh denaturing conditions, unfolded protein can form a native-like organised structure rather than a disordered linear conformation 42 . We calculated values for folded actin ranging from 0.799 ± 0.015 at 150 mV to 0.942 ± 0.057 at 350 mV (Fig. 3e), suggesting the protein is approximately defined as an oblate ellipsoid, in agreement with the stretching of proteins with a dipole moment under an electric field 25 . for unfolded actin also exhibits a similar trend increasing from 1.460 ± 0.040 to 1.658 ± 0.097. These values are more consistent with a prolate ellipsoid. Based on these observations, the localised electric field generated at the tip of nanopipettes alters the shape or conformation of translocating proteins due to their heterogeneous charge distribution. We, therefore, suggest that low voltages should be applied during nanopore experiments to understand protein-protein or protein-drug interactions. The altered behaviour at higher voltages can be used as an advantage, enabling a better understanding of some electrostatic properties such as dipole moment, net charges and molecular conductivity. Alternatively, high electric fields can be used as a method to unfold proteins without a denaturant present to complement chemical-induced unfolding.

Actin polymerisation in real-time
Despite a plethora of actin structures, understanding the structural flexibility of actin and its polymerisation has progressed slowly 1 . Using nanopore sensing, we were able to record actin polymerisation in real-time (Fig.   4a). The ionic current of ATP-activated actin monomers was recorded over time, and translocation events with increasing peak amplitude and capture rates were observed. This noticeable increase in peak amplitudes, shown in Fig. 4b, is sensitive to molecular volumes and conformations, indicative of filament formation. We are, therefore, able to see single-molecule events and observe the distribution of dynamic interactions without averaging the population. The distribution of peak current in Fig. S14 shows time-dependent multiple populations, indicative of a concomitant increase in both the proportion and degree of ATP-actin polymerisation. Moreover, the persistence of the lowest population (105.6 ± 16.8 pA) likely represents monomeric actin, demonstrating treadmilling of the actin filament. By measuring the IV curves before and after nanopore experiments (Fig. S15), we can be certain these changes do not originate from the interaction between the analytes and the nanopore and thus are directly related to filament formation.
Given that nanopore sensing can visually provide rich information obtained from the electro-behaviour of molecules during the transient translocation, this time-resolved method can quantitively read out the realtime properties of the analytes or their interactions. The degree of actin polymerisation can be monitored by both the filamentous fraction and event frequency (Fig. 4c). The initial fraction was around 17 % due to early actin nucleation events under high salt and ATP activation. Once polymerisation was induced, the proportion of filament increased over time and approached a maximum of 75 %, suggesting around 25 % of actin monomers are undergoing nucleotide exchange as treadmilling occurs. As expected, we see a linear increase in capture rate, a parameter which is directly related to the size, mobility and conformation of analytes. Actin filaments have higher net charge and mobility, contributing to the increasing capture events in such a drift regime. We can, therefore, use capture rate as a readout for actin polymerisation. This enables us to visualise polymerisation of native protein on a single molecule level, of critical importance to in the study of complex physiological and pathological dynamics such as protein aggregation and polymerisation.  Fig. S14a). Three shadow areas and protein models (from left to right) represent actin monomers, oligomers and filaments, respectively.

Actin drug binding
Given our ability to resolve unfolding actin monomers, we next sought to characterise two actin-binding drugs that have defined modes of interaction with actin as determined by X-ray crystallography. Swinholide A, a marine toxin, which binds two actin monomers in an orientation that prevents the interactions required for filament formation 44 . Latrunculins, a family of plant toxins, bind to the nucleotide-binding pocket and prevent filament formation by blocking nucleotide exchange 17,45 .
Translocation events were recorded in the presence of either Swinholide A or Latrunculin B and compared with the monomer (Figs. 5a, S16). We observed a significant difference in the scatter plots of actin bound with Latrunculin B and Swinholide A, with much deeper current blockades and longer dwell times when binding to Swinholide A (Fig. 5b). Upon addition of excess Latrunculin B, the peak current of translocation events became more uniform at different voltages (59.1 ± 5.4 pA compared with native monomers of 64.8 ± 6.7 pA at 250 mV,  Next, we compared the charge distributions for actin alone and actin with either Latrunculin B or Swinholide A (Fig 5c). The equivalent charge, calculated by the integration of individual events, is a comprehensive parameter that includes both dwell time and peak current. As this charge profile increases, the effective charge of molecules can be considered to increase accordingly, related to the spatial excluded volume in high ionic-strength environment and molecular electrokinetics 19 . We found a charge profile of 2.5 ± 1.2 fAs for actin with Latrunculin B, 3.6 ± 2.0 fAs for monomeric actin, and 6.9 ± 2.9 fAs for actin with Swinholide A. This shift and wider distribution towards actin dimers, combined with the increase in capture frequency, is indicative of well-defined dimerisation in our measurements. The electrokinetic properties of actin monomers and dimers, including dwell time and capture rate, are shown in Fig. S17 and Fig. 5d. These properties suggest that binding with Swinholide A results in a dimeric state with a consistently higher electrophoretic mobility and net charge.
This is in agreement with the crystal structure, illustrating there is a limited burial of side-chain residues and instead an overall significant increase in surface exposure to surrounding solvent molecules.
Given the noticeable differences between monomer and Swinholide A-bound actin, the drug binding efficiency of actin with Swinholide A was measured by determining the dimer fraction as a function of time (Fig. 5e). The binding was quantified at 200 mV to minimise the potential effects of changes in protein conformation caused by the electric field. Before the addition of Swinholide A, the measurement was run for at least one hour to ensure that protein flux was stable. Upon addition of two equivalents of Swinholide A, an increase occurred in the peak current distribution, as shown in Fig. 5e, consistent with the formation of actin dimers upon drug binding. The dimer fraction plateaued at around 90 % after 30 min. This can be seen in the time-dependent binding curves, Fig. 5f. Curves at different drug concentrations (with molar ratios 1:1, 1:0.5, 1:0.2, Actin: Swinholide A) were fit using a positive, cooperative model 46 . In this model, Swinholide A takes a positive effect on actin dimerisation, showing an exponential accumulation curve of the dimer fraction. The plateau can be defined as the saturated concentration of actin dimer and its corresponding rate constant can be extracted using eq. S6. Given that actin dimers have a higher flux than monomers of the same concentration (Fig. 5d), at a low drug binding ratio (e.g. 1:0.2), the calculated saturated concentration (48 %) of dimers is larger than real stoichiometric ratio (33 %). Furthermore, since the detection of dimers has a lag behind their in-situ formation due to the free diffusion and the drift by electric fields, these results reflect the apparent kinetics of protein association. This is the first time, to our knowledge, that real-time drug-binding has been visualised on a singlemolecule level with native actin molecules, highlighting the potential of quartz nanopipettes for targeted drug discovery.

Discussion
Here Unlike the chemical denaturation, this conformational change is not a two-state pathway, but rather a gradual deformation or stretching the protein along the axis of the electric field 23 . Application of various electric fields has potentials to manipulate molecular conformation of macromolecules such as protein and nucleic acid in order to provide insight into the electrostatic behaviour of biological molecules.
This study provides great promise for using nanopores for drug discovery and drug binding kinetic assays.
Building on previous nanopore work that has shown the ability to study the effect of drug binding on protein aggregation 49 we applied the technology further to look at chemical and voltage induced protein unfolding and at drug binding to a monomeric species. Here, we have been able to study the effect of two different actinbinding drugs in solution and distinguish them from the monomer alone. The changes observed in charge of the actin monomer bound to Latrunculin B are distinct (Fig. 4c), despite negligible differences observed between the crystal structures of the drug-bound versus non-bound forms. The ability to resolve Swinholide Ainduced changes in real-time illustrates the power of this technique for use in drug discovery. Not only can we see a distinct difference in protein behaviour, but we are also able to propose a positive cooperativity model whereby the binding of one actin molecule to Swinholide A increases its affinity to the second actin in the dimer resulting in an enhanced binding affinity. We can record these minor changes in a direct and highthroughput manner. As we look forward, efforts integrating computational analysis and classification of changes seen with a singular protein in the presence of different compounds may enable high throughput single-molecule drug screening and simultaneous prediction of the mode of interaction to be dually possible.
The sensitivity of quartz nanopipettes to both macromolecular and protein unfolding changes provide a promising outlook for studying the dynamic process of filament formation across diverse molecules using native proteins. This label-free method would work well with visualising protein aggregation, such as alphasynuclein in real-time, which may find clinical significance in understanding the mechanism of Parkinson's and Alzheimer's diseases 50 . Furthermore, the high sensitivity achieved suggests the technique may provide a

Fabrication of Nanopipettes
Quartz capillaries (Intracel Ltd, UK) with an outer diameter of 1.0 mm and an inner diameter of 0.5 mm with an inner filament were cleaned using the plasma cleaner and then pulled by a laser-based pipette puller (Sutter

Nanopore measurements and data processing
Before performing translocation experiments, 1 M KCl buffer was added into the nanopipette using a microfilm needle (MF34G, World Precision Instruments, UK). Electrodes (Ag/AgCl) were inserted into the nanopipette (working electrode) and the external bath (reference/ground electrode), respectively. Voltages from 150 mV to 350 mV were applied, and the ionic current was recorded as a function of time by Chimera amplifier VC100 (Chimera Instruments) with a sampling rate of 4.17 MHz. and a low-pass kHz digital Bessel filter of 50 kHz and analysed using custom-written MATLAB code by Edel Group (Fig. S5 and S6) unless otherwise stated. Power spectral density (PSD) plots for this low-noise configuration are shown in Fig. S7, at different applied voltages and low-pass filters. The data was filtered with a 10 kHz-100 kHz digital Bessel filter and analysed using custom-written MATLAB code by Edel Group. Protein fluxes (`) were extracted from a single-exponential decay fitting of the distribution of interval time between two successive translocation events ( ) as previously reported 52 . All data collected at different voltages were obtained from the same nanopipette and error bars represent one standard deviation of at least three independent experiments with different nanopipettes unless stated otherwise. Traces shown were collected at 1 MHz and low-pass filtered to 50 kHz.