German Edition: DOI: 10.1002/ange.201612050Single-Molecule StudiesInternational Edition: DOI: 10.1002/anie.201612050

Observing Single-Molecule Dynamics at Millimolar ConcentrationsMarcel P. Goldschen-Ohm+, David S. White+, Vadim A. Klenchin, Baron Chanda,* andRandall H. Goldsmith*

Abstract: Single-molecule fluorescence microscopy is a power-ful tool for revealing chemical dynamics and molecularassociation mechanisms, but has been limited to low concen-trations of fluorescent species and is only suitable for studyinghigh affinity reactions. Here, we combine nanophotonic zero-mode waveguides (ZMWs) with fluorescence resonance energytransfer (FRET) to resolve single-molecule association dynam-ics at up to millimolar concentrations of fluorescent species.This approach extends the resolution of molecular dynamics to> 100-fold higher concentrations, enabling observations atconcentrations relevant to biological and chemical processes,and thus making single-molecule techniques applicable toa tremendous range of previously inaccessible moleculartargets. We deploy this approach to show that the binding ofcGMP to pacemaking ion channels is weakened by a slowerinternal conformational change.

Single-molecule fluorescence microscopy reveals details ofmolecular composition and dynamics otherwise hiddenbecause of averaging in ensemble measurements.[1] However,a frequent experimental compromise is the requirement of nmor lower concentrations of fluorescent species. This limitationoriginates from the diffraction limit of focused light, as thesmaller the probe volume can be made, the fewer moleculeswill contribute to the background signal.[1b] For example,confocal detection schemes with diffraction-limited excitationenable observation volumes as small as 0.1–0.2 fL (1 fL = 1 �10�15 L).[2] Thus, in both confocal and total internal reflection(TIRF) modalities, there is an inherent concentration limit of< 10 nm to detect the binding of a single fluorescently labeledsubstrate.[3] This “concentration barrier” is severely debilitat-ing, as many biological and chemical mechanisms requirehighly concentrated conditions in the mm to mm range toproceed. Metabolites (including ATP), neurotransmitters,

and amino acids are frequently present at cellular concen-trations of 100 mm and above,[4] thus preventing the applica-tion of single-molecule microscopy to examine binding inkinases, receptors, translation machinery, and the vast major-ity of enzymes[5] (Figure 1a). Access to high concentrations

becomes even more significant for new single-moleculeinvestigations of synthetic catalysts, most of which operateat substrate or ligand concentrations of mm and above.[6]

Increases in the highest attainable fluorophore concen-tration can be achieved through reduction in the observationvolume to below the diffraction limit. Stimulated emissiondepletion (STED) in the context of fluorescence correlationspectroscopy (FCS) permits detection volumes near 0.02 fL.[7]

Photoactivation[8] of or photobleaching[9] down to sparsesubsets of fluorophores transiently creates low concentrationsof active fluorescent species from a larger reservoir offluorophores, thereby enabling concentrations up to

[*] Dr. M. P. Goldschen-Ohm,[+] D. S. White,[+] Dr. V. A. Klenchin,Dr. B. ChandaDepartment of Neuroscience, University of Wisconsin-Madison1111 Highland Ave., Madison, WI 53705 (USA)E-mail: chanda@wisc.edu

D. S. White,[+] Dr. R. H. GoldsmithDepartment of Chemistry, University of Wisconsin-Madison1101 University Ave., Madison, WI 53706 (USA)E-mail: rhg@chem.wisc.edu

Dr. B. ChandaDepartment of Biomolecular ChemistryUniversity of Wisconsin-Madison420 Henry Mall, Madison, WI 53706 (USA)

[+] These authors contributed equally to this work.

Supporting information and the ORCID identification number(s) forthe author(s) of this article can be found under:http://dx.doi.org/10.1002/anie.201612050.

Figure 1. ZMW-FRET imaging. a) Histogram of over 30000 enzymeaffinities from the BRENDA database.[5] Concentration ranges accessi-ble to single-molecule resolution are indicated for several methods.b) Illustration of FRET between bound donor (fcGMP) and an acceptoron the CNBD. c) Experimental setup for ZMW-FRET microscopy.Fluorescence in the donor and acceptor channels from arrays ofZMWs were simultaneously imaged on two EMCCD cameras. Inset:schematic representation of a single ZMW with an immobilizedCNBD. The observation volume decays rapidly within about 25 nm ofthe surface. The red dashed circle represents the Fçrster radius, furtherreducing the effective observation volume. Thus, freely diffusingdonors unbound to the CNBD are not observed.

AngewandteChemieCommunications

1Angew. Chem. Int. Ed. 2017, 56, 1 – 5 � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

These are not the final page numbers! � �

10 mm.[10] The use of nanophotonic antennas has enableddetection of fixed individual molecules in up to 25 mm

concentrations of quenched fluorescent species,[11] with theadded benefit of plasmonically enhanced fluorescence.[12]

Imaging in the vicinity of the interface between a convexlens and a flat surface enables access to concentrations up to2 mm.[13] Indeed, a variety of chemical and photonic tools haveenabled access to concentrations up to the low mm range.[14]

ZMW, nanophotonic arrays of subwavelength holes ina metallic film (Figure 1c) provide subdiffraction-limitednearfield observation volumes as small as 20 zL (1 zL = 1 �10�21 L), far smaller than that achievable with TIRF or STED,such that single fluorophores can be resolved at up to low mm

concentrations.[2, 15] ZMWs have been successfully used toobserve molecular recognition processes at high nm to low mm

concentrations including translation events at individualribosomes,[16] dynamics of membrane-bound proteins,[17] andsingle-molecule electrochemistry,[18] and have enabled high-throughput single-molecule genomic sequencing.[19] Theyhave also been combined with plasmonic nanosized antennasto gain the advantages of fluorescence enhancement.[20]

Regardless, association processes that require concentrationsupwards of tens to hundreds of mm remain out of reach, thusrequiring new single-molecule methods. Here, we show thata combination of ZMWs and single-molecule FRET(smFRET) enables resolution of single-molecule molecularrecognition events at mm concentrations. This approachmerges ZMW’s subdiffraction-limited observation volumewith a detection volume defined by the Fçrster radius of theFRET pair on the order of 1 zL (Figure 1b,c).

Although smFRET alone enables observation of single-molecule binding dynamics at up to 10 mm,[21] access to higherconcentrations is limited by nonspecific adsorption andbackground signals from freely diffusing fluorophores.While both of these interferences exist in our dual ZMW-FRET method, their influence has been sharply reduced bythe volume restriction in zero-mode waveguides (ZMWs).Critically, our approach extends the resolution of singlefluorophore association by over 100-fold from low mm to lowmm concentrations, thus allowing the elucidation of previ-ously inaccessible biological and chemical mechanisms at thelevel of single molecules.

As validation, we report time-resolved single-moleculebinding events for fluorescently labeled cyclic guanosinemonophosphate (fcGMP)[22] to monomeric cyclic nucleotide-binding domains (CNBDs) from human hyperpolarizationand cyclic nucleotide-activated (HCN) channels. HCN chan-nels are critical for the regulation of heart and brain rhythms,but the mechanism by which cyclic nucleotide bindingmodifies channel gating remains unclear. Single-moleculebinding dynamics report on electrically silent and transientconformations energetically coupled to binding, and informon the forces by which they interconvert, thus providinga novel window into this process. Here, we used ZMW-FRETto directly observe single binding events of fcGMP.

CNBDs were specifically labeled with a FRET acceptorand immobilized within arrays of over 100 000 ZMWs (Fig-ure 1c, and see the Supporting Information). Two EMCCDcameras were used to simultaneously record the donor

(fcGMP) and acceptor fluorescence from about 1000 ZMWsat once, thereby making this approach feasible for high-throughput studies. Excitation alternated between the donorand acceptor pump wavelengths at a frame rate of 10 Hz,which allowed observation on interleaved frames of smFRETarising from donor binding (Figure 2a, middle) and acceptor

stability. Importantly, this method allows quantification of thenumber of fluorescently labeled proteins in a ZMW bycounting the number of acceptor bleaching steps, thusenabling the selection of ZMWs featuring single proteinsonly (Figure 2 a, bottom). Notably, although individual boundfcGMP molecules could not be resolved at a concentration of1 mm in the donor channel (Figure 2a, top), the smFRETsignal from single binding events was clearly visible in theacceptor channel prior to bleaching (Figure 2a, middle). Incomparison, direct observation of fcGMP binding withoutFRET was possible in ZMWs only at low mm concentrations(Figure 2b).

Specific binding at single molecules as reported bysmFRET was recorded at fcGMP concentrations from 1 mm

Figure 2. Single-molecule ligand binding at mm concentrations withZMW-FRET. a) Fluorescence time series for fcGMP binding events ata single CNBD with freely diffusing fcGMP at a concentration of 1 mm.Simultaneous emission from the donor (blue, fcGMP) and acceptor(red) upon interleaved donor (l = 532 nm) and acceptor (l = 640 nm)excitation (see the Supporting Information). The acceptor emission isoverlaid with the idealized time series (black). b) Fluorescence timeseries for fcGMP binding to CNBDs without an acceptor label (donoronly). Notably, the background from freely diffusing fcGMP in ZMWsoccludes resolution of single binding events at high mm concentrationsand above. The fluorescence time series in both (a) and (b) arebackground-subtracted, while the smFRET trace in (a) additionallyunderwent crosstalk subtraction and baseline correction by splinefitting (see the Supporting Information).

AngewandteChemieCommunications

2 www.angewandte.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim Angew. Chem. Int. Ed. 2017, 56, 1 – 5� �

These are not the final page numbers!

to 1 mm (Figure 3 a). The concentration-dependence of theequilibrium bound probability across all molecules indicatesan apparent affinity of about 10 mm (Figure 3b), similar toprevious bulk-averaged measurements.[23] Notably, bindingcurve saturation required fcGMP at concentrations of hun-dreds of mm.

Histograms of bound and unbound dwell times wereconstructed from idealized pooled data (Figure 4a, see alsothe Supporting Information). As expected for a bindingreaction, unbound lifetimes decreased with increasing fcGMPconcentration, whereas bound lifetimes were relatively con-centration-independent. Based on previous observations thatthe CNBD isomerizes between two conformations,[24] wecompared the likelihood of several kinetic models (Fig-

ure S2e). The preferred model had two unbound and twobound states, such that isomerization of the CNBD can occurboth with and without bound ligand (Figure 4b). This modelis consistent with X-ray crystal structures in which the C-helixcaps the bound ligand,[24, 25] and electron paramagneticresonance studies that suggest similar capping of the bindingsite also occurs in the CNBD without a ligand, therebytemporarily blocking access of fcGMP.[23] Furthermore, thisscheme is similar to that observed for the binding dynamics ofthe higher affinity ligand fcAMP.[24] In comparison to fcAMPat monomeric CNBDs, our high-concentration single-mole-cule studies reveal that the lower apparent affinity of fcGMPcompared to fcAMP is due not only to slower binding, butalso to a reduction in the probability that bound ligand willinduce a stabilizing isomerization of the CNBD that prolongsthe total lifetime of the bound state.

Figure 3. Single-molecule association dynamics of fcGMP at HCNCNBDs. a) FRET time series for fcGMP binding to single CNBDs (red)overlaid with idealized traces (black) at various concentrations offcGMP. Traces shown underwent corrections for the baseline withspline fitting and were both background and crosstalk subtracted (seeraw traces in Figure S4). Horizontal dashed lines indicate fluorescencelevels for bound, unbound, and bleached conditions. Triangles denotethe time of the acceptor bleach. b) Bound probability from the totaltime fraction spent bound for all molecules versus the fcGMPconcentration (circles) fit with Bmax/(1 +Kd/[fcGMP]), where Bmax = 0.83is the maximal bound probability and Kd =10 mm is the apparentdissociation constant (solid line). Prediction from the model in Fig-ure 4b normalized to Bmax is shown as a dashed line.

Figure 4. A dynamic model of fcGMP association at HCN CNBDs.a) Histograms of unbound and bound single-molecule dwell-timedistributions (gray) for events from all molecules combined overlaidwith monoexponential (blue dashed line) or biexponential (red line)maximum likelihood fits. The concentration of fcGMP for each pair ofhistograms is indicated on the left-most ordinate. Summary of fitparameters shown in Figure S2a. b) Kinetic model of fcGMP associa-tion dynamics.

AngewandteChemieCommunications

3Angew. Chem. Int. Ed. 2017, 56, 1 – 5 � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim www.angewandte.org

These are not the final page numbers! � �

Now that several methods allow access to elevatedconcentrations for single-molecule experiments, thus break-ing the “concentration barrier” to varying degrees, it is usefulto compare and contrast the advantages of these methods.One issue relevant to the investigation of biological structuresis access to the biomolecule. Our ZMW-FRET combination,similar to other ZMW geometries, entails the biomoleculebeing in a microenvironment with a high surface-to-volumeratio, although the 100–150 nm diameter of the aperture islarge compared to the CNBD, and the sidewalls and bottomof the ZMW are passivated.[26] This geometry is less restrictivethan antenna-based approaches that involve a nanosizedaperture[11, 16] but more restrictive than photoactivation[10] andFRET[21] approaches that do not have the same constraintsand may be more suitable for in vivo measurements. Ourapproach does not require specialized fluorophores, unlikephotoactivation-based approaches.[10] However, the temporalobservation window of our approach as well as FRET-basedapproaches[21] is limited by the photobleaching of theacceptor, whereas in ZMWs without FRET[2, 15] or otherapproaches where all fluorophores are continually replen-ished,[10, 11, 13, 16] the observation window will be longer,although more stable fluorophores[27] can likely extend theobservation window in ZMW-FRET. Finally, only ZMW-FRET is capable of reaching the biologically significant rangeof 100 mm–1mm.

In conclusion, we demonstrate single-molecule resolutionof binding events at up to mm concentration, more than twoorders of magnitude higher concentrations than were feasiblewith other methods. Our combined ZMW-FRET approachenables observation of molecular dynamics at relevantconcentrations for the majority of biological and chemicalassociation reactions that were previously inaccessible tosingle-molecule techniques.

Acknowledgements

This research was supported by funding from the NationalInstitutes of Health (GM084140, NS081293 to B.C. andT32GM007507 to D.S.W.) and the National Science Founda-tion (CHE-1254936 to R.H.G.). B.C. was also supported bya Romnes Faculty fellowship (WARF). We thank Dr. MikeSanguinetti for the wild-type HCN2 plasmid.

Conflict of interest

The authors declare no conflict of interest.

Keywords: FRET · kinetics · nucleotides ·single-molecule studies · zero-mode waveguide

[1] a) T. Ha, Nat. Methods 2014, 11, 1015 – 1018; b) W. E. Moerner,D. P. Fromm, Rev. Sci. Instrum. 2003, 74, 3597 – 3619.

[2] M. J. Levene, J. Korlach, S. W. Turner, M. Foquet, H. G. Craig-head, W. W. Webb, Science 2003, 299, 682 – 686.

[3] A. M. van Oijen, Curr. Opin. Biotechnol. 2011, 22, 75 – 80.[4] B. D. Bennett, E. H. Kimball, M. Gao, R. Osterhout, S. J.

Van Dien, J. D. Rabinowitz, Nat. Chem. Biol. 2009, 5, 593 – 599.[5] M. Scheer, A. Grote, A. Chang, I. Schomburg, C. Munaretto, M.

Rother, C. Sohngen, M. Stelzer, J. Thiele, D. Schomburg, NucleicAcids Res. 2011, 39, D670 – 676.

[6] a) T. Cordes, S. A. Blum, Nat. Chem. 2013, 5, 993 – 999; b) J. D.Ng, S. P. Upadhyay, A. N. Marquard, K. M. Lupo, D. A. Hinton,N. A. Padilla, D. M. Bates, R. H. Goldsmith, J. Am. Chem. Soc.2016, 138, 3876 – 3883; c) M. B. Roeffaers, G. De Cremer, H. Uji-i, B. Muls, B. F. Sels, P. A. Jacobs, F. C. De Schryver, D. E.De Vos, J. Hofkens, Proc. Natl. Acad. Sci. USA 2007, 104, 12603 –12609.

[7] L. Kastrup, H. Blom, C. Eggeling, S. W. Hell, Phys. Rev. Lett.2005, 94, 178104.

[8] S. Manley, J. M. Gillette, G. H. Patterson, H. Shroff, H. F. Hess,E. Betzig, J. Lippincott-Schwartz, Nat. Methods 2008, 5, 155 –157.

[9] M. Moertelmaier, M. Brameshuber, M. Linimeier, G. J. Sch�tz,H. Stockinger, Appl. Phys. Lett. 2005, 87, 263903.

[10] A. B. Loveland, S. Habuchi, J. C. Walter, A. M. van Oijen, Nat.Methods 2012, 9, 987 – 992.

[11] A. Puchkova, C. Vietz, E. Pibiri, B. Wunsch, M. Sanz Paz, G. P.Acuna, P. Tinnefeld, Nano. Lett. 2015, 15, 8354 – 8359.

[12] A. Kinkhabwala, Z. Yu, S. Fan, Y. Avlasevich, K. Mullen, W. E.Moerner, Nat. Photonics 2009, 3, 654 – 657.

[13] S. R. Leslie, A. P. Fields, A. E. Cohen, Anal. Chem. 2010, 82,6224 – 6229.

[14] P. Holzmeister, G. P. Acuna, D. Grohmann, P. Tinnefeld, Chem.Soc. Rev. 2014, 43, 1014 – 1028.

[15] P. Zhu, H. G. Craighead, Annu. Rev. Biophys. 2012, 41, 269 – 293.[16] S. Uemura, C. E. Aitken, J. Korlach, B. A. Flusberg, S. W.

Turner, J. D. Puglisi, Nature 2010, 464, 1012 – 1017.[17] S. M. Christensen, M. G. Triplet, C. Rhodes, J. S. Iwig, H.-L. Tu,

D. Stamou, J. T. Groves, Nano. Lett. 2016, 16, 2890 – 2895.[18] J. Zhao, L. P. Zaino III, P. W. Bohn, Faraday Discuss. 2013, 164,

57 – 69.[19] a) J. Korlach, K. P. Bjornson, B. P. Chaudhuri, R. L. Cicero, B. A.

Flusberg, J. J. Gray, D. Holden, R. Saxena, J. Wegener, S. W.Turner, Methods Enzymol. 2010, 472, 431 – 455; b) J. Larkin, M.Foquet, S. W. Turner, J. Korlach, M. Wanunu, Nano. Lett. 2014,14, 6023 – 6029.

[20] D. Punj, M. Mivelle, S. B. Moparthi, T. S. van Zanten, H.Rigneault, N. F. van Hulst, M. F. Garcia-Parajo, J. Wenger, Nat.Nanotechnol. 2013, 8, 512 – 516.

[21] M. Sugawa, S. Nishikawa, A. H. Iwane, V. Biju, T. Yanagida,Small 2010, 6, 346 – 350.

[22] C. Biskup, J. Kusch, E. Schulz, V. Nache, F. Schwede, F.Lehmann, V. Hagen, K. Benndorf, Nature 2007, 446, 440 – 443.

[23] H. A. DeBerg, P. S. Brzovic, G. E. Flynn, W. N. Zagotta, S. Stoll,J. Biol. Chem. 2016, 291, 371 – 381.

[24] M. P. Goldschen-Ohm, V. A. Klenchin, D. S. White, J. B. Cow-gill, Q. Cui, R. H. Goldsmith, B. Chanda, eLife 2016, 5, e20797.

[25] W. N. Zagotta, N. B. Olivier, K. D. Black, E. C. Young, R. Olson,E. Gouaux, Nature 2003, 425, 200 – 205.

[26] J. Korlach, P. J. Marks, R. L. Cicero, J. J. Gray, D. L. Murphy,D. B. Roitman, T. T. Pham, G. A. Otto, M. Foquet, S. W. Turner,Proc. Natl. Acad. Sci. USA 2008, 105, 1176 – 1181.

[27] Q. Zheng, M. F. Juette, S. Jockusch, M. R. Wasserman, Z. Zhou,R. B. Altman, S. C. Blanchard, Chem. Soc. Rev. 2014, 43, 1044 –1056.

Manuscript received: December 11, 2016Final Article published: && &&, &&&&

AngewandteChemieCommunications

4 www.angewandte.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim Angew. Chem. Int. Ed. 2017, 56, 1 – 5� �

These are not the final page numbers!

Communications

Single-Molecule Studies

M. P. Goldschen-Ohm, D. S. White,V. A. Klenchin, B. Chanda,*R. H. Goldsmith* &&&— &&&

Observing Single-Molecule Dynamics atMillimolar Concentrations

Dynamic duo : Nanophotonic zero-modewaveguides (ZMWs) have been com-bined with fluorescence resonance energytransfer (FRET) to resolve single-mole-cule association dynamics at concentra-

tions of fluorescent species of up tomillimolar. These concentrations are>100-fold higher than previously acces-sible with other methods, such as totalinternal reflection (TIRF).

AngewandteChemieCommunications

5Angew. Chem. Int. Ed. 2017, 56, 1 – 5 � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim www.angewandte.org

These are not the final page numbers! � �

Supporting Information

Observing Single-Molecule Dynamics at Millimolar ConcentrationsMarcel P. Goldschen-Ohm+, David S. White+, Vadim A. Klenchin, Baron Chanda,* andRandall H. Goldsmith*

anie_201612050_sm_miscellaneous_information.pdf

1

Further improvements

Several of the drawbacks of our ZMW-FRET can also be improved using recently reported

strategies. For example, use of more photostable FRET acceptors[1] can greatly augment

observation time windows. Use of DNA Origami nanoadapters[2] can significantly enhance the

fluorescence signal by placing biomolecules at the center of the ZMW and increasing experimental

throughput by streamlining deposition of biomolecules into ZMW’s. Thus, even more robust

access to single-molecule dynamics at high concentrations is within reach.

Methods

Protein expression, purification and labeling

The CNBD from human HCN2 channels was expressed in E. coli, biotinylated and labeled with a

maleimide derivative of DyLight 650 (DY650) at an introduced cysteine (E571C) as described in

detail previously[3]. For specific labeling at position E571C, two other accessible native cysteine

residues were mutated (C508A/C584S), which we have previously shown to have little effect on

the CNBDs ability to bind cyclic nucleotide[3].

Single-molecule imaging and analysis

Single-molecule imaging and analysis of fcGMP (Biolog) binding was performed as described

previously for fcAMP[3]. Briefly, CNBDs were deposited in ZMWs with diameters between 100-

150 nm (Pacific Biosciences) and imaged on an inverted microscope (Olympus IX-71) with two

512x512 EMCCD cameras (Andor iXon Ultra X-9899) under alternating 532 and 640 nm laser

excitation (Coherent) at 60 W/cm2 and 25 W/cm2 at the sample, respectively, and a frame rate of

10 Hz using Metamorph software (Molecular Devices). This set-up enabled simultaneous

recording of ~1000 ZMWs at a time. However we sparsely deposited the CNBDs onto the array

in order to reduce the probability of having more than one CNBD per ZMW. At our sample

loading, we found 25% overall occupancy of the 1000 wells. Of those ~250 wells, ~10% showed

two bleach steps suggesting two proteins per ZMW, and ~1% showed more than three bleachsteps

suggesting three or more proteins per ZMW. These observed probabilities, P(1) = 0.23, P(2) =

0.03, P(3)<0.003, are close to the probabilities suggested by Poisson statistics with a Poisson rate

parameter λ=0.35, P(1) = 0.25, P(2) = 0.04, P(3) = 0.005, where P(k) is the probably of an

occupancy of k in a ZMW according to the equation

!)(

k

ekP

kk

Fluorescence time series from single molecules were analyzed with custom software

written in Matlab (The MathWorks, Inc.). A trace was used for analysis if: (1) it featured only a

single fluorescently-labeled CNBD in the ZMW, as indicated by a single bleach step in the

acceptor channel (Fig. 2a, bottom), and (2) featured at least one event in the FRET channel,

indicating fcGMP binding to the CNBD. Traces featuring multiple bleach steps, large amplitude

fluctuations in the donor channel likely due to imperfect surface passivation and consequent

increased non-specific binding, and traces without any FRET channel fluctuations, indicating no

binding events, were all removed prior to analysis. All selected traces were then background

subtracted, and crosstalk was also subtracted from the smFRET traces, as described in more detail

below. Drifts in the baseline were corrected with spline fitting before the traces were idealized

with vbFRET[4]. After idealization, traces were once again visually inspected and single frame

2

events were removed from the idealized records as such events could often be attributed to noise

in simulated data. For analysis of dynamics we selected the subset of molecules with smFRET

binding signal-to-noise ratios >2.25 (Fig. S1e). Heterogeneity in signal-to-noise is likely due to

both variation in ZMW fabrication and location of CNBDs within each ZMW[2]. Overall,

approximately 11,000 ZMW’s satisfied the initial screen, and of those, approximately 4,000

ZMW’s passed the second signal-to-noise screen, consistent with Figure S1f. From these ZMW’s,

we recorded approximately 15,000 binding/ unbinding events. Hidden Markov modeling (HMM)

was performed with QuB[5] using a dead time of 200 ms.

In general, the signals in the donor (D) and acceptor (A) channels upon excitation at the

donor pump wavelength are given by

D = Dbg + Dfree + Dbound,

A = Abg + Acrosstalk + Adirect + FRETfree + FRETbound,

where Dbg and Abg are background levels including camera dark counts, and scattered or reflected

light, Dfree and Dbound are from directly excited freely diffusing and bound fcGMP, respectively,

Acrosstalk is crosstalk from the fraction of the donor emission spectrum that falls within the acceptor

channel, Adirect is due to weak direct excitation of the acceptor at the donor pump wavelength, and

FRETfree and FRETbound are FRET between freely diffusing and bound donors (fcGMP),

respectively, and the acceptor on the CNBD. Crosstalk was described by Acrosstalk = (Dfree +

Dbound), where = 0.08 was determined from the slope of a linear fit to a scatter plot of Ableached

versus Dbleached for each molecule, where bleached denotes a time average after acceptor bleaching

(Fig. S1a). This value is slightly higher than the = 0.04 estimated from the relative area of

fcGMPs emission spectrum within the donor and acceptor channels, likely due to non-ideal

transmission efficiencies along the optical path. Our analysis was relatively insensitive to small

changes in on this order. In the absence of donor and acceptor dyes, the background signals Abg

and Dbg varied between ZMWs, likely due to heterogeneity in the fabrication process. Empirically,

Abg and Dbg were found to be similar within a given ZMW under our experimental conditions.

Thus, we determined Abg in each individual ZMW by solving Ableached = Abg + (Dbleached - Dbg)

with Abg = Dbg. The distribution of Abg values across ZMWs is shown in Figure S1b.

Prior to acceptor bleaching, A exhibits two distinct intensity levels Aunbound and Abound

corresponding to conditions where the CNBD is unoccupied or occupied by fcGMP (Figs. 2a,

middle, 3a and S1c). Relative to Ableached, Aunbound also contains both Adirect and FRETfree,

whereas Abound includes all of those components plus FRETbound. The increase in the per molecule

mean Aunbound (and also Abound) with increasing concentrations of fcGMP comes from the

concentration-dependence of FRETfree, which arises from freely diffusing fcGMP near the

acceptor on the CNBD (Fig. S1c, d). The contribution of FRETfree was lower than predicted for a

bulk solution of freely diffusing donors and acceptors[6], because only a donor quenched by

multiple acceptors was considered, not multiple donors transferring to a single acceptor, and

because of limited access to the acceptor from regions occupied by the CNBD in our system. In

contrast, the mean amplitude of FRETbound = Abound - Aunbound across molecules was relatively

independent of fcGMP concentration (Fig. S1c), as expected. At low concentrations, Abg

constitutes the dominant source of background in the acceptor channel, whereas at high donor

concentrations, Acrosstalk becomes a significant contributor, as expected. Critically, the amplitude

of FRETbound was large in comparison to both FRETfree and the standard deviation of the subtracted

crosstalk signal Acrosstalk (Fig. S1c) indicating that on average, neither FRET from freely diffusing

fcGMP nor subtraction of crosstalk grossly distorted the binding signal even at 1 mM fcGMP.

3

Dwell time distributions from idealized records were fit with mono or biexponential

distributions by maximum likelihood, and the conditional probability that a given dwell time

would be observed within the associated observation time window defined by acceptor bleaching

was accounted for as described by Colquhoun and Sigworth [7] (Fig. 4a, S2a). Simulated data from

a uniform distribution of dwell times with amplitudes, noise, and acceptor lifetimes drawn from

gamma or log normal distributions describing the observed data (e.g. Fig. S2b), show that although

acceptor bleaching limited the observation of longer dwell times, this effect was relatively minor

over the majority of the unbound and bound dwell time distributions (Fig. S2c), and was also

accounted for in our kinetic model (Fig. 4b) as described below. Acceptor lifetimes were inversely

correlated with bound probability, suggesting that excitation due to FRET from bound donor

during frames with laser excitation at the donor pump contributed to bleaching (Fig. S2d). To

account for this, an irreversible transition to a bleached state was added to all states during HMM

model optimization. The bleach rate from either unbound (0.03 s-1) or bound (0.08 s-1) states were

determined by extrapolating the correlation between acceptor lifetime and bound probability to

bound probabilities of zero or one, respectively. This method ignores the comparatively small

concentration-dependence of the bleach rate due to FRET from freely diffusing donors. Models

were globally optimized for all molecules and fcGMP concentrations, and ranked by their Akaike

information criterion[8] (Fig. S2e).

Consistent with an isomerization of the unliganded CNBD, unbound dwell time

distributions were biexponential (Fig. 4a). In contrast, only a single exponential component was

resolvable in the bound time distributions. Although we cannot completely rule out the existence

of only a single bound state (e.g. model #2), the rate of entry into the state B2 in our preferred

model #3 is sufficiently slow that acceptor bleaching limits its observation (Fig. S2e), therefore

making it difficult to detect in the distribution of bound times. Regardless, both of the most likely

models (#2 and #3) agree that the reduced apparent affinity of fcGMP as compared to fcAMP

arises from both slower binding and either reduced or nonexistent probability to isomerize while

ligand is bound. Notably, the small number of events in dwell time distributions at 1 mM fcGMP

does not reflect the number of observed events at molecules exhibiting binding signals with

sufficiently high signal-to-noise ratios, but instead reflects the difficulty in observing fully

bracketed events (bound-unbound-bound or unbound-bound-unbound) at a concentration

significantly above Kd prior to acceptor bleaching. Thus, protein-ligand complexes with higher Kd

values will be even easier to analyze. Finally, significant improvements in data throughput and

model analysis could be obtained by utilizing longer-lived acceptor dyes[1].

4

Figure S1. Statistics for single-molecule association dynamics with ZMW-FRET.

a, Correlation between mean fluorescence intensity in donor and acceptor channels after acceptor

bleaching at various donor (fcGMP) concentrations overlaid with linear fit (black line). Each circle

represents an individual molecule. b, Distribution of camera background intensity levels (Abg)

across ZMWs (see methods). c, Normalized distribution of the average acceptor bleach step height

per concentration of fcGMP, d, Fluorescence intensities as a function of donor (fcGMP)

concentration. See methods for details. Briefly, Ableached, Aunbound and Abound are the mean

intensities in the acceptor channel during time periods where a donor is either unbound or bound,

or the acceptor has bleached, respectively (see Fig. 2a, middle and 3a). Triangles and circles are

the respective means from log normal distribution maximum likelihood fits to the observed

intensities across all molecules, and error bars are the 95% confidence limits in the distribution

means. Squares are the mean (error bars are standard deviation) across all molecules of the standard

deviation in the crosstalk from donor emission in the acceptor channel (see methods). e, The ratio

of FRET from freely diffusing or bound donors (fcGMP) as a function of donor concentration.

Symbols are means from a maximum likelihood fit to a log normal distribution, and error bars are

the 95% confidence limits in the distribution means. f, Histograms of signal-to-noise ratios in the

acceptor channel for various donor (fcGMP) concentrations. Histograms at each donor

concentration are overlaid with maximum likelihood fits to a gamma distribution. Vertical dashed

line indicates cutoff for our analysis.

5

Figure S2. Dwell times and HMM modeling of fcGMP association at single molecules.

a, Summary of maximum likelihood exponential fits to unbound (biexponential) and bound

(monoexponential) dwell times. Amplitudes for biexponential fast (solid) and slow (dashed)

components are shown in the bottom frame. b, Distributions of bleach times at various donor

(fcGMP) concentrations overlaid with maximum likelihood fits to exponential (dashed) or log

normal (smooth) distributions. Donor concentrations are indicated along the ordinate. c, Simulated

dwell times drawn from a uniform distribution using amplitude, noise and bleach times drawn

from distributions that describe the observed data. The reduced frequency of longer dwell times

reflects truncation of the observation window due to acceptor bleaching. d, Linear correlation

between acceptor lifetime and bound probability. e, Kinetic models for transition between unbound

(U*) and bound (B*) states explored with HMM (see methods). Not shown are irreversible

transitions to a bleached state that were allowed from each state as described in the methods. The

Akaike information criterion score for each model is given relative to the most likely model (AIC,

lower is better).

obs

exp

logn

b

0

300

0

800

0

800

Co

un

ts

0 50 100 150

Bleach Time (s)

0

600

1 µ

M1

0 µ

M1

00

µM

1 m

M

0.0 0.5 1.0

Bound Probability

10

20

30

Me

an

Ble

ach

Tim

e (

s)

-6 -5 -4 -3

log([fcGMP (M)])

0

10

20U

nb

ou

nd

t (

s)

-6 -5 -4 -3

log([fcGMP (M)])

-6 -5 -4 -3

log([fcGMP (M)])

0.0

0.2

0.4

0.6

0.8

1.0

Un

bo

un

d A

mp

litu

de

a

d

0

10

20

Bo

un

d t

(s)

0 10 20

Unbound Time (s)

0.0

0.2

0.4

0.6

0.8

1.0

No

rmaliz

ed

Co

unts

0 2010

Bound Time (s)

0.0

0.2

0.4

0.6

0.8

1.0

c

Norm

aliz

ed

Cou

nts

1310301001000

[fcGMP] (µM)e

U1 B1

0.4 x105 M-1s-1

0.69 s-1

U1 B1

1.0 x105 M-1s-1

U2

0.78 s-1

0.2

7 s

-1

0.3

6 s

-1

B2

U1 B1

U2

1.0 x105 M-1s-1

1.16 s-1

0.2

6 s

-1

0.3

5 s

-1

0.1

0 s

-1

0.2

2 s

-1

∆AIC = 7,196

∆AIC = 2,124

∆AIC = 0

Model #1

Model #2

Model #3

6

Figure S3. High throughput imaging and FRET channels. Fluorescence image of ~1,000

ZMWs in the acceptor channel during excitation at the acceptor pump wavelength in a solution

of 10 µM fcGMP. Approximately 250 fluorescently labeled CNBDs are identified in this ZMW

array. Boxed region is shown on an expanded scale below. Boxed region within expanded

region is shown to the right for both donor (top) and acceptor (bottom) channels. Arrow denotes

an individual ZMW containing an acceptor-labeled CNBD with bound donor (fcGMP).

7

Fig S4. Raw smFRET traces. FRET time series for fcGMP binding to single CNBDs at various

concentrations of fcGMP. Triangle denotes time of acceptor bleach. These traces did not undergo

the baseline correction with spine-fitting, background subtraction, or cross-talk subtraction that

were used to produce the analogous figures in Figure 2a (1 mM) or Figure 3a (1,10,100 μM).

8

References

[1] Q. Zheng, M. F. Juette, S. Jockusch, M. R. Wasserman, Z. Zhou, R. B. Altman, S. C.

Blanchard, Chem Soc Rev 2014, 43, 1044-1056.

[2] E. Pibiri, P. Holzmeister, B. Lalkens, G. P. Acuna, P. Tinnefeld, Nano Lett 2014, 14, 3499-

3503.

[3] M. P. Goldschen-Ohm, V. A. Klenchin, D. S. White, J. B. Cowgill, Q. Cui, R. H.

Goldsmith, B. Chanda, eLife 2016, 5.

[4] J. E. Bronson, J. Fei, J. M. Hofman, R. L. Gonzalez, Jr., C. H. Wiggins, Biophys J 2009,

97, 3196-3205.

[5] aC. Nicolai, F. Sachs, Biophysical Reviews and Letters 2013, 08, 191-211; bF. Qin, A.

Auerbach, F. Sachs, Biophys J 2000, 79, 1915-1927.

[6] J. B. Birks, Leite, J. Phys. B: At. Mol. Phys. 1970, 3, 513.

[7] D. Colquhoun, F. J. Sigworth, in Single-channel recording, 2nd ed. (Eds.: B. Sakmann, E.

Neher), Plenum Press, New York, 1995, pp. 483-587.

[8] H. Akaike, IEEE Transactions on Automatic Control 1974, 19, 716-723.