Talin folding as the tuning fork of cellular mechanotransduction

Mechanical Characterization of the Talin R3 Domain.

The talin R3 domain is the weakest of all 13-rod domains, thought to be the precursor of talin-based adhesions, given its low unfolding force and the presence of two vinculin-binding sites (16, 20, 23, 24). This low mechanical stability is due to the four-residue threonine belt on its hydrophobic core; the substitution of these residues in the mutant R3 IVVI increases its stability significantly (16, 20). Due to these different mechanical properties, we explore both the R3 WT and R3 IVVI domains perturbed by complex mechanical signals, which will allow us to compare and relate their responses with their mechanics. First, in order to establish the benchmarks for the study under complex force signals, we characterize the folding mechanics of both domains at constant force (Fig. 1).

Fig. 1 B and C shows representative recordings measured at the coexistence force (${\mathrm{F}}_{1/2}$; same occupation of the folded and unfolded states), which is 9 pN for R3 IVVI and 5 pN for R3 WT. Their folding dynamics are characterized by reversible transitions between the folded and unfolded states with an extension change of $\sim$20 nm, but the kinetics of R3 WT is nearly 10 times faster, compared to those of R3 IVVI (Fig. 1 C, Inset). From equilibrium trajectories such as these, we measure the relative occupation of the folded state (folding probability; Fig. 1D), and the folding (r_F) and unfolding (r_U) rates (Fig. 1E). The folding probability spans less than 2 pN, illustrating the exquisite force sensitivity of both domains. Due to this steep force dependency, the folding and unfolding rates have an exponential Bell-like dependency ($r_{\mathrm{U}/\mathrm{F}}=k_0^{\mathrm{U}/\mathrm{F}}\exp \left[\pm F{x}_{\mathrm{U}/\mathrm{F}}^{†}/kT\right]$; ref. 25), with similar distances to transition state from the folded and unfolded states. Overall, both domains exhibit two-state–like dynamics, occurring at different force ranges—between 4 pN and 6 pN (R3 WT), and 8 pN to 10 pN (R3 IVVI)—and kinetic rates on a very different timescale—at ${\mathrm{F}}_{1/2}$, $r_{F/U}\approx$12 ${\mathrm{s}}^{-1}$ (R3 WT), and $r_{F/U}\approx$1.5 ${\mathrm{s}}^{-1}$ (R3 IVVI).

Talin Discerns between External Mechanical Noise and Oscillatory Forces.

Motivated by the naturally noisy cellular environment, we first asked whether talin maintains a robust response when subject to a randomly fluctuating perturbation instead of a constant force. Applying forces that change rapidly is challenging, as force spectrometers typically change the force by displacing large mechanical probes. To this aim, we take advantage of our magnetic tape head tweezers design, able to change the force with a bandwidth greater than 10 kHz while maintaining intrinsic force-clamp conditions (Fig. 1A and ref. 26). We generate external mechanical noise as a broadband uncorrelated force signal, with an average of 9 pN and a peak-to-peak amplitude greater than 7 pN (SI Appendix, Fig. S1 and Supplementary Methods), and apply it to the talin R3 IVVI domain. This signal fluctuates around $F_{\mathrm{1}/\mathrm{2}}$ over a force range that significantly exceeds the folding regime of R3 IVVI. Importantly, this external mechanical noise is different from the intrinsic thermal noise provided by the heat bath, which we do not control in our experiment and that induces the stochastic folding and unfolding transitions we observe at constant force.

Fig. 2 A, Left shows the dynamics of R3 IVVI perturbed by external mechanical noise. Strikingly, despite the fine force sensitivity of talin and the broad force range over which the external mechanical noise fluctuates, we do not observe any effect on talin dynamics, which are statistically indistinguishable from those measured at the average force of 9 pN. The dwell-time histograms—obtained using a maximum-likelihood procedure (SI Appendix, Fig. S2)—are both exponential distributions with approximately the same time constant—$0.59\pm 0.02$ s (noise) versus $0.69\pm 0.02$ s (no noise) (Fig. 2 A, Right). This apparent insensitivity to external mechanical fluctuations is maintained at any other folding force. We observe no statistically significant differences (within a 95% CI) between the folding probabilities measured in the presence or absence of external mechanical noise, and the coexistence forces (${\mathrm{F}}_{1/2}$) obtained under both conditions overlap within error bars (SI Appendix, Fig. S3). This evidence suggests that talin has a robust mechanical response, filtering out the randomly fluctuating components, and responding only to the average force.

Given the ability of talin to remain indifferent to the external mechanical noise, we explore its response under a force signal that oscillates at a specific frequency and is submerged in external mechanical noise. We build this signal by adding a sinusoidal force $F\left(t\right)=F_0+A\sin \left(2\pi f_{\mathrm{\Omega }}\right)$ (with $F_0=9$ pN, $A=1$ pN, and $f_{\mathrm{\Omega }}=0.7$ Hz) to external mechanical noise with the same properties employed before. Hence, the amplitude of the correlated component is small compared to the background fluctuations (Fig. 2 B, Inset). By contrast, talin dynamics are now dramatically altered, and most transitions occur at regular intervals of $\sim$0.7 s, while the mechanical noise is ignored (Fig. 2 B, Left). These entrained folding dynamics shift the shape of the dwell-time histogram from an exponential to a peaked distribution (Fig. 2 B, Right). The first peak is centered at $\sim$0.7 s, which coincides with the half-period of $f_{\mathrm{\Omega }}$, while the second peak appears at 3 times that value, $\sim$2.1 s. Despite the dominant entrained dynamics, we observe an underlying exponential distribution, which reflects a minority of remnant stochastic transitions, occurring at a time constant compatible to a force of 9 pN (Fig. 2 B, Right, red dotted line). Hence, unlike random force fluctuations, talin detects force oscillations through its folding dynamics, which shift from stochastic—characterized by exponentially distributed dwell times—to a phase-locked response, reflected in peaked dwell-time distributions.

Mechanical Signal Detection by Talin is Frequency Dependent.

The talin R3 domain detects minimal force changes through its folding dynamics; differences in force of only 0.3 pN shift the folding probability by up to 30% (Fig. 1D). Since our data demonstrate that talin detects oscillatory forces, we seek to test whether such a response is also frequency sensitive. To this aim, we subject R3 IVVI to purely oscillating forces at different frequencies, $f_{\mathrm{\Omega }}=0.1$, 1, and 10 Hz, with a mean value of $F_0=9$ pN, and amplitude of $A=0.7$ pN.

Fig. 3 shows R3 IVVI folding dynamics under these three oscillatory force signals (Fig. 3, Left), with the associated dwell-time histograms (Fig. 3, Right). Under the slowest perturbation ($f_{\mathrm{\Omega }}=0.1$ Hz; Fig. 3A), we observe that, while the extension of the protein follows the oscillation, most transitions are stochastic, as the force changes much more slowly than the natural folding kinetics—0.1 Hz versus $\sim$1.5 Hz. The dwell-time histogram is an exponential distribution with a time constant compatible to a constant force of 9 pN ($r_{\mathrm{K}}=2.16\pm 0.15\hspace{0.17em}{\mathrm{s}}^{-1}$), and a marginal contribution of entrained transitions as a small peak at 5 s (red asterisk). At $f_{\mathrm{\Omega }}=1$ Hz (Fig. 2B), most folding transitions are synchronized with the perturbation and occur at regularly spaced time intervals. The dwell-time histogram has three peaks of diminishing amplitude and a small underlying exponential distribution. These peaks are centered at odd multiples of half of the driving period—0.5, 1.5, and 2.5 s—and represent harmonics of talin phase-locked response. Finally, at a driving frequency of 10 Hz (Fig. 3C), the stochastic behavior is recovered with the same dynamics as those measured at 9 pN. The dwell-time histogram is an exponential distribution with a time constant of $r_K=2.13\pm 0.10\hspace{0.17em}{\mathrm{s}}^{-1}$, which indicates that talin filters this fast oscillation. Only when calculating the dwell-time distribution with logarithmic binning, we observe a small contribution at 0.05 s corresponding to a minority of entrained transitions (Fig. 3 C, Inset). Hence, these data indicate a frequency-selective response of talin upon oscillating forces. Interestingly, talin folding dynamics segregate into phase-locked dynamics that respond to the oscillation with a frequency-dependent intensity, and a stochastic behavior that filters the signal and detects only the average of the perturbation.

Talin Response is Governed by the Mechanical Stability of the Domain.

The response of the R3 IVVI domain suggests that the entrained dynamics manifest optimally at frequencies of the order of the natural folding rates, which, at 9 pN, are $\sim$1.5 ${\mathrm{s}}^{-1}$ (Fig. 1E). Hence, it can be expected that proteins with faster kinetic rates will respond to higher frequencies and vice versa, providing a tuning mechanism based on the mechanical stability of the protein domain. In order to test this hypothesis, we measure the folding dynamics of the R3 WT domain subject to oscillatory forces given that, at the coexistence force $F_{1/2}=5$ pN, its kinetic rates are nearly 10 times faster ($\sim$12 ${\mathrm{s}}^{-1}$; Fig. 1E).

Fig. 4 shows single-molecule recordings of R3 WT perturbed with sinusoidal signals of 0.5 (Fig. 4A), 5 (Fig. 4B), and 50 Hz (Fig. 4C), an average of $F_0=5$ pN, and amplitude of $A=0.7$ pN. These data show an analogous response to that measured for R3 IVVI but shifted to higher frequencies. At 0.5 Hz, the dynamics are exponentially distributed, with a minor entrained contribution appreciable in log scale (Fig. 4 A, Inset). At 5 Hz, the transitions are entrained with the force signal, showing a peak at 0.1 s and an underlying exponential distribution with a time constant compatible to that measured at 5 pN ($r_K=18.0\pm 2.24\hspace{0.17em}{\mathrm{s}}^{-1}$). At 50 Hz, the dynamics are stochastic, with a rate of $r_K=11.0\pm 0.8\hspace{0.17em}{\mathrm{s}}^{-1}$ and a minimal contribution at 0.01 s appreciable with logarithmic binning (Fig. 4 C, Inset). These results suggest the generality of the frequency-dependent signal detection by talin folding dynamics, where the mechanical stability of the protein domain selects the responsive frequency regime. Additionally, we measured the folding probability of R3 WT in the presence and absence of external mechanical noise, observing no statistically significant difference (SI Appendix, Fig. S3). These data suggest that, similar to the mutant R3 IVVI, R3 WT filters out random mechanical fluctuations but responds to coherent force oscillations.

Stochastic Resonance in Talin Folding Dynamics.

Our observations show a phase-locked response of talin folding dynamics as a mechanism of mechanical signal detection. This behavior resembles the physical phenomenon of stochastic resonance, by which nonlinear systems exhibit an amplified response upon a weak input signal due to the presence of stochastic noise (27, 28). Stochastic resonance has been demonstrated in very diverse contexts, such as climate dynamics (29), optical systems (30), DNA hairpin dynamics (31), and quantum systems (32), since it only requires three fundamental ingredients: a bistable system, a weak periodic input, and an intrinsic source of noise. All three of these conditions are fulfilled by our system; the free energy landscape of the R3 domain is well approximated by a symmetric double-well potential (SI Appendix, Fig. S4), and the thermal bath provides the intrinsic noise. Importantly, this intrinsic thermal noise is different from the external mechanical noise we used in Fig. 2, which was applied as an external random force on talin.

Signal detection by stochastic resonance typically refers to an increase in the signal-to-noise ratio (SNR) of the system output, which is the metric usually employed to characterize this physical phenomenon (27). Indeed, the power spectrum of talin folding response under a periodic force perturbation (calculated on the idealized trace to remove the elastic contribution) exhibits a resonant peak at the driving frequency $f_{\mathrm{\Omega }}$ (SI Appendix, Fig. S5). However, we can also characterize stochastic resonance by the shape of the dwell-time distributions, which, as discussed before, include a combination of stochastic and resonant transitions. Hence, these distributions can be modeled as the combination of an exponential distribution (stochastic), and a series of Gaussian peaks (resonant),

where $r_K$ is the stochastic folding rate of talin (exponential contribution), $T_{\mathrm{\Omega }}=1/f_{\mathrm{\Omega }}$ is the period of the driving force, and $A_{\mathrm{s}}$ and $A_{\mathrm{r}}^{\left(n\right)}$ are the amplitudes of the stochastic and resonant contributions, respectively. When the frequency is in the responsive regime, we observe several peaks at odd multiples of half of the driving frequency ($\left(2n-1\right)T_{\mathrm{\Omega }}/2$; Fig. 3B). These contributions correspond to resonant transitions where one or more periods are missed but the transition occurs coherently with the driving signal at a future cycle. Hence, we can quantify statistically the fraction of resonant transitions $F_{\mathrm{R}}$ as the relative weight between the Gaussian distributions and the underlying exponential one. If the distribution is normalized ($\int f\left(t\right)dt=1$), $F_{\mathrm{R}}=1-A_{\mathrm{s}}/r_K$.

To demonstrate the response of talin over a broad range of frequencies, we carry out experiments on both the R3 IVVI and R3 WT domains under signals with $A=0.7$ pN and frequencies ranging from 0.05 Hz to 100 Hz. Fig. 5A shows $F_{\mathrm{R}}$ as a function of the driving frequency $f_{\mathrm{\Omega }}$. At frequencies above $r_{\mathrm{K}}$, the resonant peak can be buried in the shape of the exponential distribution, and it is not possible to directly determine its contribution. In such cases, we calculate the dwell-time histogram with logarithmic binning to separate fast timescales and estimate the relative weight from a logarithmic transform of Eq. 1 (SI Appendix, Figs. S6 and S7). Similarly, we use the SNR to characterize stochastic resonance, measured from the power spectrum of the folding/unfolding time series (SI Appendix, Fig. S5). For both R3 IVVI and R3 WT, we observe the same dependence of the SNR on the driving frequency as that obtained using the fraction of resonant transitions, which illustrates the equivalence of both quantities to characterize stochastic resonance (SI Appendix, Fig. S5).

Fig. 5A shows that talin operates as a mechanical band-pass filter, responding only over a narrow frequency range defined by the mechanical stability of the protein domain. For a bistable system, the optimal resonance frequency can be estimated as $r_K/2$ (27) (red dotted line), which agrees with the experimental data. This resonance frequency is associated with the height of the intrinsic free energy barrier that separates the folded and unfolded states. Oscillating forces tilt the protein landscape toward the folded and unfolded states cyclically, increasing and decreasing the barrier periodically. If this periodic forcing occurs on a timescale compatible with the thermal-activated transitions, the state switching synchronizes with the perturbation. If the driving frequency is much faster than the intrinsic rates, the system responds only to the average barrier, filtering the oscillation; if the oscillation is too slow, the transitions are much more frequent than the timescale imposed by the external forcing.

Molecular kinetic processes measured with single-molecule techniques can be altered by the tethering system, which occasionally leads to artifactual effects arising from the interplay between the molecule and the dynamics of the pulling apparatus (33). Hence, in order to test whether our observations are an intrinsic property of the talin protein or are influenced by the magnetic probe, we conduct control experiments on R3 IVVI using magnetic beads of a larger size (M-450 with 4.5-$\mathrm{\mu }$m diameter, compared to the standard M-270 with 2.8-$\mathrm{\mu }$m diameter). These beads require a different chemical strategy to anchor our molecular constructs (SI Appendix, Supplementary Methods and Fig. S8). The experiments with the larger M-450 beads show no appreciable difference than those measured with the smaller M-270, despite its significant difference in size; R3 IVVI filters external mechanical noise of large bandwidth and gets optimally entrained under signals with a frequency of $\sim$1 Hz (SI Appendix, Figs. S9–S11). Hence, these data suggest that our observations are not altered by the tethering used in our experiments, likely because the diffusion coefficient of the magnetic beads is much higher than that of the elastic and folding response of talin, which, hence, become rate limiting.

All of the experiments in Fig. 5A were done using signals of constant amplitude, $A=0.7$ pN, which is a moderate perturbation compared with the folding probability. In order to test the limit of signal detection, we measure the response of R3 IVVI at the resonance frequency $f_{\mathrm{\Omega }}=0.7$ Hz and under different amplitudes (Fig. 5B). When the force signal spans the whole folding probability range (A $>$ 1 pN), the resonant response is saturated, and nearly all transitions are entrained with the periodic forcing. Under signals as small as 0.3 pN, we still observe an effect of resonant transitions of nearly 15%. This amplitude-dependent response is unaffected when the signal is buried in external mechanical noise (gray squares), which indicates that these random mechanical fluctuations do not influence the resonant properties exhibited by talin. This behavior is also reproduced with the R3 WT domain (SI Appendix, Fig. S12), which suggests that both protein domains respond similarly to force signals that combine coherent oscillations and randomly fluctuating components. Although the amplitude of the periodic forcing enhances the level of signal detection, stochastic resonance does not arise simply from the magnitude of the oscillating forcing applied to the system. At an amplitude of $A=1.4$ pN, where the response at the resonance frequency is saturated, we obtain the same frequency response dependence as that at half that amplitude, indicating that the high selectivity of talin toward the driving force is amplitude independent (SI Appendix, Fig. S13).

Single-Molecule Experiments.

All of our experiments were carried out in our custom-made magnetic tape head tweezers setup, detailed in ref. 26. The fluid chambers were built and functionalized as described before (26, 49). Experiments are carried out in Hepes buffer with 10 mM ascorbic acid (pH 7.3). The data are acquired at rates between 1 kHz and 1.6 kHz with our custom-made software, written onto a binary file, and visualized in real time with Igor Pro 8 (Wavemetrics). Single-molecule recordings are smoothed with a Savitzki–Golay algorithm with a 101-point box size (R3 IVVI) or a 31-point box size (R3 WT). Recordings are idealized for analysis with a double-threshold algorithm to detect the occupations on the folded and unfolded states automatically. Further details on experimental analysis are described in SI Appendix.

Protein Expression and Purification.

Polyprotein constructs are engineered using BamHI, BglII, and KpnI restriction sites in pFN18a restriction vector, as described previously (25). Our protein construct contains the R3 IVVI or R3 WT mouse talin domain, followed by eight titin I91 domains, and flanked by an N-terminal HaloTag enzyme and a C-terminal AviTag for biotinylation.