A cuvette of doubly labelled peptide returns one FRET efficiency. If half the molecules are compact and half extended, the reading lands between the two, and nothing in that number separates the mixture from a homogeneous population sitting at the intermediate distance. Ha and co-workers made the case when they first pushed FRET down to a single donor and a single acceptor linked by a short DNA molecule and imaged by near-field scanning optical microscopy: in contrast to ensemble measurements, dynamic events on a molecular scale are observable in single-pair FRET measurements because they are not cancelled out by random averaging [1]. That experiment read transfer efficiency from the photodestruction dynamics of the donor or acceptor, and modified the classical ensemble equations for single-molecule use [1]. What it did not yet deliver is the object that makes the method useful: a distribution.
The histogram, and what its width is made of
Deniz and colleagues supplied that, collecting photon bursts from single donor–acceptor labelled macromolecules diffusing through a confocal spot and using DNA as a rigid spacer to build constructs of varying donor–acceptor spacing, so both the mean and the distribution width of FRET efficiencies could be measured as a function of distance [2]. The mean single-pair efficiencies qualitatively follow the distance dependence predicted by Förster theory, the conformational demonstration being substrate and product subpopulations resolved before and after a restriction endonuclease cleavage reaction, and the denaturation of a DNA hairpin [2].
The trap is reading every broad peak as heterogeneity. An efficiency from a finite photon count has non-zero spread even for a homogeneous population. Antonik and colleagues built probability distribution analysis to put a floor under this: their theoretical distributions are calculated explicitly including crosstalk, stochastic variations and background, and represent the minimum width that a FRET distribution must have, so that an unambiguous distinction is made between shot-noise distributions and distributions broadened by heterogeneities [3]. The fit's χ2 surface suggested an ultimate level of precision with FRET of less than 1% of the Förster radius; against donor–acceptor-labelled DNA the distributions were unambiguously broader than shot noise could explain, and a Gaussian distance model recovered 5 Å inhomogeneities attributed to dye motions [3]. That figure is dye motion on a DNA duplex, not conformational exchange in a peptide. Excess width over that floor licenses you to look for structure; it is not evidence of two conformers.
Dye pair, Förster radius, and the window you can see
In Hellenkamp and colleagues' multi-laboratory benchmark the reference duplexes carried Atto 550 with Atto 647N, R0 = 62.6 Å, and Atto 550 with Alexa Fluor 647, R0 = 68.0 Å [4]. Zaer and Lerner put the practical span of the FRET-based ruler at inter-dye distances of 3–10 nm, with protein-induced fluorescence enhancement reporting on proximity between a dye and a protein surface over a shorter 0–3 nm range [5]. Because efficiency runs as r-6 against R0, the sensitive band is narrow: at half R0 efficiency is already about 0.98 and at 1.5 R0 about 0.08, so an excursion confined to either shoulder barely moves E. A peptide whose end-to-end distance travels between 1 and 3 nm is compressed into the top of the curve by a 6 nm R0 pair.
Accuracy across that window is bounded by quantities unrelated to your peptide. The benchmark's error analysis gave an estimate of 7% for the determined R0, which the authors call relatively large and attribute mainly to uncertainty in κ2; the total error for a single smFRET-based distance between two freely rotating point dipoles came out at less than 6 Å for 35 Å < RDA < 80 Å; and twenty laboratories measuring the same dye-labelled DNA duplexes by a unified, straightforward method obtained FRET efficiencies with standard deviations between ±0.02 and ±0.05 [4]. That is a rigid nucleic acid standard measured under a protocol agreed in advance.
What polyproline did to the spectroscopic ruler
The cautionary result was measured on peptides: Schuler and colleagues recorded FRET efficiency distributions for donor and acceptor dyes attached to the ends of freely diffusing polyproline molecules of various lengths [6]. The observed mean efficiencies agreed with those from ensemble lifetime measurements but differed considerably from the values expected from Förster theory with polyproline treated as a rigid rod: at donor–acceptor distances much less than R0 the observed efficiencies were lower than predicted, whereas at distances comparable to and greater than R0 they were much higher [6]. Two possible contributions to the former are offered — incomplete orientational averaging during the donor lifetime and, because of the large size of the dyes, breakdown of the point-dipole approximation assumed in Förster theory — while Langevin molecular dynamics suggests the differences for the longer polyproline peptides can be explained by chain bending, which considerably shortens the donor–acceptor distances [6].
Best and colleagues took polyproline-20 further. NMR showed that in water approximately 30% of the molecules contain internal cis prolines, whereas none are detectable in trifluoroethanol; simulations suggest the all-trans form is relatively stiff, with persistence lengths of 9–13 nm using different established force fields, and that kinks arising from internal cis prolines are primarily responsible for the higher mean FRET efficiency in water [7]. The efficiency histograms and donor lifetime distributions were explained by multiple species with populations determined by NMR, and a similar simulation approach suggests that the flexibility of the chromophore linkers is largely responsible for the previously unexplained high value of R0 required to fit the data in the classic Stryer and Haugland study [7]. A backbone carrying a slow isomerisation degree of freedom generates static subpopulations indistinguishable in a histogram from conformational states, and here the assignment came from NMR rather than from the FRET data. The linker, meanwhile, is part of the ruler and not an attachment detail.
Double labelling and the purification it costs
Two spectrally distinct dyes must end up at two defined positions. Ratner and colleagues state the obstacle for engineered cysteines plainly: the challenge is to develop methods for selective modification of one of two reactive sulfhydryl groups in a protein molecule [8]. Their route, on Escherichia coli adenylate kinase, screened single-cysteine mutants for the rate of reaction of each engineered cysteine with DTNB as a reference reagent and selected site pairs whose rate-constant ratio was high, above 80; the more reactive cysteine took the donor in a first cycle, the second probe went onto the less reactive site under unfolding conditions in a second, and the yield of pure, site-specific doubly labelled product was 70–90% [8]. That is a globular-protein procedure and its unfolding step does not transfer to every construct, but the arithmetic holds wherever a stochastic double-labelling reaction runs: unlabelled, donor-only, acceptor-only and doubly labelled species leave together, and only the last is the measurement.
Roy, Hohng and Ha note that smFRET is relatively insensitive to incomplete labelling: if the donor is missing the molecule is simply not observed, and if the acceptor is missing this donor-only species shows up as a zero-FRET population [9]. Alternating-laser excitation sorts species directly: Hohlbein, Craggs and Cordes describe ALEX as enabling the sorting of fluorescently labelled species based on the number and type of fluorophores present, and as a convenient route to the correction factors necessary for determining accurate molecular distances across the entire FRET range [10]. A donor-only population you can identify is a nuisance; one you cannot is a low-FRET peak you will report as an extended conformer.
Diffusing through a confocal spot, or tethered under TIRF
In the freely diffusing experiment each molecule contributes one burst as it crosses the focus, and the diffusion time td is typically milliseconds. Kalinin and colleagues report that dynamic probability distribution analysis quantitatively recovers FRET efficiencies of the interconverting states and relaxation times of dynamics on the timescale of td, with a dynamic range of about ±1 order of magnitude with respect to td, and name the factors limiting its accuracy: brightness variations, shortening of the observation time due to diffusion, and a contribution of multimolecular events [11]. Exchange much slower than the burst is never seen twice in the same molecule. Zaer and Lerner put the consequence for histogram reading directly — each centrally distributed sub-population represents either a single conformation that stayed unchanged within milliseconds, or multiple conformations that interconvert much faster than milliseconds and are therefore an averaged-out sub-population [5].
Immobilisation buys the trajectory. The Roy, Hohng and Ha guide focuses on immobilised molecules precisely because they allow measurements of single-molecule reaction trajectories from 1 ms to many minutes, with time resolution limited by the frame rate of the CCD camera, in the best case 1 ms [9]. A tethered molecule yields dwell times and kinetics directly; a burst histogram yields them only through a model of the exchange. Below the burst, photon statistics still carry information: Nettels and colleagues probed an unfolded protein by FRET using the statistics of photon emission from single molecules, and found global reconfiguration of the chain on a timescale of approximately 50 ns, slowing concomitant with chain collapse under folding conditions [12]. That is a protein, and 50 ns is a chain reconfiguration time rather than a rate between defined conformers, but it marks the fast end of the accessible range.
What tethering can do, and the controls that test it
The passivation layer is a reagent. Roy, Hohng and Ha describe suppressing non-specific binding with an additional passivating agent, the most popular being polyethylene glycol: a precleaned surface-activated slide is aminosilanised and reacted with NHS-ester-modified PEG which also includes a small fraction of biotin-PEG-NHS ester for specific tethering; and strong interaction between denatured protein and linear PEG can be eliminated using branched PEG instead [9]. The warning is in that clause: the passivation layer can bind the unfolded state whose population you are measuring.
The direct control is to run the same construct both ways. Talaga and colleagues did this on a peptide — a disulfide-crosslinked GCN4 two-stranded coiled coil with a fluorescent donor and acceptor at the N-termini of its two chains and a cysteine disulfide near the C-terminus — comparing end-to-end distance distributions for the peptide freely diffusing in solution and deposited onto aminopropyl silanised glass [13]. As urea concentration increased the mean end-to-end distance shifted to longer distances both in free solution and on the modified surface, and the distribution widths indicated millisecond conformational fluctuations that under all three conditions gave non-exponential correlations on the 1–100 ms timescale, with a urea-sensitive component of the correlation decay corresponding to that measured by bulk relaxation kinetics [13]. The surface preserved the denaturant dependence for that peptide on that silane, which is what the paired measurement is for and not a licence to skip it.
The stronger control removes surface contact altogether. Okumus and colleagues encapsulated DNA or RNA molecules inside 100–200 nm diameter unilamellar vesicles immobilised on a supported lipid bilayer via biotin–streptavidin linkages, confirmed encapsulation by fluorescence correlation spectroscopy, nuclease digestion and external buffer exchange, and saw no direct binding of DNA or RNA on it even at concentrations exceeding 100 nM [14]. Inside vesicles the natural form of the hairpin ribozyme exhibited 50-fold variation in both folding and unfolding rates in 0.5 mM Mg2+, identical to what was observed from molecules tethered directly on the surface, which the authors take as strongly indicating that the heterogeneity does not arise as an artefact of surface attachment but is intrinsic to the nature of the molecules [14]. That settled it for one ribozyme; the same encapsulation is the experiment for yours.
Photophysics that impersonates dynamics
An acceptor that goes dark for a few hundred milliseconds and returns produces a trajectory indistinguishable from a genuine excursion to a low-FRET state. Rasnik, McKinney and Ha note that photobleaching and blinking of fluorophores pose fundamental limitations on the information content of single-molecule fluorescence measurements, and that photoinduced blinking of Cy5 has hampered many previous investigations using this popular fluorophore; they report that Trolox with the enzymatic oxygen-scavenging system eliminates Cy5 blinking, dramatically reduces photobleaching and improves signal linearity at high excitation rates [15].
The scavenging system is not a free choice. Aitken, Marshall and Puglisi compared the protocatechuic acid/protocatechuate-3,4-dioxygenase system with the glucose oxidase/catalase system, observed lower dissolved oxygen concentrations with the former under standardised conditions, and measured increased initial lifetimes of single Cy3, Cy5 and Alexa488 fluorophores [16]. The additive result matters at the bench: biological reducing agents increase both the frequency and duration of blinking events of Cy5, an effect that scales with reducing potential, while the antioxidants ascorbic acid and n-propyl gallate increased the stability of Cy3 and Alexa488 [16]. A reducing agent left in the buffer for an unrelated reason introduces the exact artefact you are trying to exclude.
Brightness can also change with no distance changing. Hwang, Kim and Myong calibrated protein-induced fluorescence enhancement, the photophysical phenomenon whereby the intensity of a fluorophore increases upon proximal binding of a protein, and showed a marked distance dependence below the 4 nm range, resolving as small as a single base pair distance at the extreme vicinity of the fluorophore where the enhancement is maximised; they note that FRET is insensitive to distance changes in the 0–3 nm range, and calibrated on a finely spaced DNA ladder carrying a BamHI site, on RNA translocation by the DExH enzyme RIG-I, and on RecA filaments on single-stranded DNA [17]. In smFRET the same effect is a confound: efficiency is a ratio of intensities, so anything brightening the donor moves apparent E, and a donor-only construct titrated with the same partner separates a distance change from an enhancement.
Gamma and beta, and what an accurate efficiency requires
Raw detector counts are not FRET efficiencies. Four correction factors stand between them, and the benchmark names each with its source. Leakage of donor emission into the acceptor detection channel, α, is determined using the FRET efficiency of the donor-only population; the fraction of direct excitation of the acceptor by the donor-excitation laser, δ, from the stoichiometry of the acceptor-only population; γ normalises the effective fluorescence quantum yields and detection efficiencies of the acceptor and donor; and β normalises the excitation intensities and cross-sections of the acceptor and donor [4]. In the fully corrected expressions β appears in the stoichiometry, not the efficiency [4]: β makes the stoichiometry axis quantitative for species sorting, γ makes the efficiency axis quantitative for distance.
Most intensity-based smFRET studies have so far characterised only relative changes in efficiency rather than corrected values [4]; there are three routes to γ. If at least two species with different inter-dye distances are present they can be used to obtain a global γ-correction; if one species with substantial distance fluctuations is present, a single-species γ-correction may be possible; and the detection correction factor can also be determined on a molecule-by-molecule basis, provided the acceptor photobleaches before the donor [4]. Roy, Hohng and Ha give that last route in working form: γ is calculated as the ratio of the change in acceptor intensity to the change in donor intensity upon acceptor photobleaching, γ = ΔIA/ΔID [9]. Alternating-laser excitation supplies the factors as a by-product of the excitation scheme [10]. An uncorrected efficiency is a proximity ratio: it ranks two states correctly and tracks a transition in time, but yields no distance you can put in a structural model, and two instruments will not agree on it.
Set against the TR-FRET guide
The companion TR-FRET guide describes an ensemble proximity method: a lanthanide donor, a time gate opening after prompt background has decayed, a two-wavelength ratio from a whole well, every molecule folded into one population average. smFRET gives up that throughput entirely — sample dilute enough that one molecule occupies the observation volume at a time, one labelled construct per site pair, a purification campaign behind each — and buys back what an ensemble ratio cannot produce: the shape of the distribution and the timescale on which it interconverts. Whether a peptide and a partner come close is the plate method's question. How many conformers exist and how fast they exchange is not a question averaging can answer.