Research
We study the physics of mechanical sensing in the inner ear. The auditory system reliably detects sounds that move hair bundles by less than a nanometer and responds within a millisecond. We ask how hair cells achieve remarkable sensitivity in the presence of noise, how they behave when mechanically coupled, and how the brain adjusts their sensitivity, protecting them from damage, via efferent neurons. We combine experiments on hair cells ex vivo, mostly from the bullfrog sacculus, with theoretical and computational methods from nonlinear dynamics, statistical physics, and information theory. We have also recently applied our theoretical methods to hearing in insects.
- Chaoticity in hair bundle dynamics
- Efferent input and critical self-tuning
- Hair bundle coupling
- Phase locking, phase slips, and noise
- Hearing in insects
Chaoticity in hair bundle dynamics
Models of the auditory system often describe it as poised near a bifurcation. These models capture many experimental phenomena, but they carry a tradeoff: being close to the critical point raises sensitivity but slows the response (critical slowing down), and the parameters must be tuned precisely, which is difficult for a biological system operating amid noise and changing conditions. Chaotic dynamics avoid this tradeoff. Nearby trajectories of a chaotic system diverge exponentially, so it is sensitive to small perturbations, and in models and experiments the system's response becomes faster as chaoticity increases.
Recordings of spontaneous and driven hair-bundle oscillations show a low-dimensional chaotic attractor. Under forcing below the bundle's own frequency, Poincaré maps show the motion passing from chaos through quasiperiodicity to entrainment as the forcing amplitude increases. By changing the viscosity and ionic composition of the surrounding fluid, we can tune the degree of chaos ex vivo. Bundles were most sensitive to weak stimuli in the weakly chaotic regime, and their response to a force step became faster as chaos increased; numerical simulations of a chaotic Hopf oscillator reproduce both results.
In theory, noise can make the nonisochronous Hopf oscillator chaotic, and this regime gives the highest sensitivity and a faster response. In a model of coupled hair cells, chaos in the individual cells enhances synchronization and weak-signal detection, and experiments on artificially coupled bundles agree. We have compared chaotic and critical descriptions of both auditory and vestibular detection and reviewed the evidence for chaos in hair-cell dynamics. Together, these results suggest that chaos may contribute to the sensitivity and temporal resolution of the auditory and vestibular systems.
Papers: Faber 2018, Sci. Rep. · Faber 2019, Sci. Rep. · Faber 2019, Chaos · Faber 2021, Phys. Rev. Res. · Faber 2024, Sci. Rep. · Faber 2024, Front. Neurol.
Efferent input and critical self-tuning
Hearing must detect sounds as quiet as 0 dB SPL, pick them out of louder backgrounds, and survive sounds as loud as 120 dB SPL. Our work on efferent control is framed by a model of two feedback processes that underlie the robustness and sensitivity of hair cells: a critical self-tuning mechanism intrinsic to the cell, which consumes metabolic energy to amplify mechanical stimuli, and a neural input process that protects the cell from damage by powerful stimuli. That neural input comes from efferent neurons, which carry signals from the brain to hair cells and are thought to act as a gain control mechanism that protects hair cells and fine-tunes detection.
We found that stimulating efferent neurons immediately raised the frequency of a hair bundle's spontaneous oscillations, and that stronger, longer or more frequent stimulation suppressed oscillations. Receptive-field maps showed that efferent activity reduces the bundle's mechanical sensitivity. Both oscillation frequency and sensitivity reduction grew with the strength of stimulation and reversed when it stopped, and blockers of the α9α10 nicotinic acetylcholine receptors, through which efferent neurons act on hair cells, abolished the effect on spontaneous oscillations.
Efferent activity also shapes how a bundle recovers from mechanical overstimulation. A high-amplitude deflection normally pushes a bundle out of its oscillatory regime; efferent stimulation during recovery often prevented this transition. Efferents therefore help determine the dynamical regime in which a hair bundle operates. Ongoing work examines how efferent activation changes the hair bundle's force–displacement relation and its energy balance.
Papers: Kao 2013, PLoS ONE · Lin 2020, J. Neurosci. · Lin 2022, eNeuro · Metzler-Winslow 2024, Biophys. J.
Hair bundle coupling
In vivo, hair bundles are connected by overlying membranes, so the dynamics of coupled nonlinear oscillators underlie the performance of the inner ear. We developed a technique for mechanically coupling two active hair bundles; coupling enhanced the coherence of their spontaneous oscillations and of their phase-locked response to sinusoidal stimuli, consistent with a model of two coupled nonisochronous oscillators, each displaying a supercritical Hopf bifurcation. Coupling groups of living hair cells with artificial membranes, we identified chimera states and frequency clustering, which a numerical model with heterogeneous parameters reproduces; the model is most sensitive to external signals when poised at the onset of synchronization, where these states are likely to form. Our numerical model of coupled hair cells suggests that the onset of chaos in the individual oscillators enhances synchronization and signal detection. Efferent activation decreases synchronization among coupled bundles, suggesting a mechanism by which neural modulation may reduce the sensitivity of the system.
Papers: Faber 2021, Chaos · Faber 2021, Phys. Rev. Res. · Roongthumskul 2021, Biophys. J. · Toderi 2026, Sci. Rep.
Phase locking, phase slips, and noise
Ex vivo, hair bundles of some species oscillate spontaneously. These noisy limit-cycle oscillations consume energy, synchronize readily to weak stimuli, and are one candidate mechanism of active amplification. A common description places the bundle near a Hopf bifurcation, where a change in a control parameter changes its amplification and frequency selectivity.
We measured how an oscillating bundle phase-locks to a stimulus across the physiological range of frequencies and amplitudes. A simple model that captures the measured Arnold tongue suggests that phase locking proceeds through a saddle-node bifurcation on an invariant circle at low amplitudes and through a supercritical Hopf bifurcation at higher amplitudes. Near the transition, the bundle's phase advances via phase slips, described by the stochastic Adler equation. Instead of oscillating steadily, hair bundles also sometimes exhibit “bursting” or “spiking” motions, and spikes evoked in quiescent bundles amplify their motion up to 100-fold. We have recently found regular, bursting and spiking oscillations in the amphibian papilla, one of the frog's auditory organs. Theoretical work with Arik Yochelis's group predicts that forcing which modulates the cell's internal control parameter changes the phase-locking transition and the shape of the traveling wave in the inner ear.
To drive bundles without loading them, we have attached magnetic beads to the stereocilia and deflected them with a magnetized probe; cubic magnetic nanoparticles moved by an electromagnet give noncontact control with sub-millisecond resolution. Driven this way, bundles mode-lock at frequencies well above their own, with overlapping regions of different mode-locking ratios between which a bundle flickers; an ensemble of such oscillators could detect signals well above the cells' characteristic frequencies.
Because these oscillations are noisy, we also study stochastic limit-cycle oscillators. With Alex Levine, we have examined how noise distorts the mean limit cycle, and shown in models that a state-dependent internal drive, like the bundle's, violates even a generalized fluctuation-dissipation theorem. We also develop methods to reduce and fit the many parameters of biophysical models using noisy recordings.
Papers: Roongthumskul 2011, Biophys. J. · Rowland 2011, Appl. Phys. Lett. · Fredrickson-Hemsing 2012, Phys. Rev. E · Roongthumskul 2013, Phys. Rev. Lett. · Shlomovitz 2013, Biophys. J. · Lee 2014, ACS Nano · Shlomovitz 2014, Interface Focus · Roongthumskul 2015, Biophys. J. · Edri 2016, EPL · Levy 2016, Sci. Rep. · Edri 2018, Phys. Rev. E · Sheth 2018, Phys. Rev. E · Sheth 2019, Phys. Rev. E · Sheth 2021, Phys. Rev. Res. · Marcinik 2024, Phys. Rev. Res. · Vaido 2025, Hear. Res. · Marcinik 2026, Phys. Rev. Res.
Hearing in insects
Insects rely on hearing to communicate, to identify and locate potential mates, and to avoid predators. Many species are too small to use the interaural time and intensity differences that vertebrates use to localize sound; instead, antennal ears detect the vector-valued velocity field of sound rather than its scalar pressure. With Joerg Albert's group, we apply our theoretical framework to mosquito hearing. Several disease-carrying mosquito species use distortion products to detect and locate mates, and the mechanical tuning of the male's flagellum is not aligned with that of its sensory neurons; in a generic model of acoustic sensing, combining these two schemes yields a detector with enhanced frequency selectivity and speed of response, able to detect transient, narrowband flight tones. We argue that velocity detection at small scales, together with the dipole-like sound emitted by flying insects, results in short ranges of detection and brief signals, and we propose mechanisms by which insects could still extract precise frequency and directional information. We have also explored masking strategies against mosquito hearing: masks that focus their acoustic power into one or a few frequencies perform best, and we propose that rapidly frequency-modulated masks disrupt information transfer most effectively.
Papers: Faber 2025, PNAS · Faber 2026, Phys. Rev. Res. · Faber 2026, Chaos