Animal ears are exquisitely adapted to capture sound energy and perform signal analysis. Studying the ear of the locust, we show how frequency signal analysis can be performed solely by using the structural features of the tympanum. Incident sound waves generate mechanical vibrational waves that travel across the tympanum. These waves shoal in a tsunami-like fashion, resulting in energy localization that focuses vibrations onto the mechanosensory neurons in a frequency-dependent manner. Using finite element analysis, we demonstrate that two mechanical properties of the locust tympanum, distributed thickness and tension, are necessary and sufficient to generate frequency-dependent energy localization.
All auditory systems convert airborne sounds into vibrations of specialized anatomical structures. This motion is then transduced into nerve impulses for processing by the central nervous system. The mammalian ear has distinct locations for each of these tasks. The eardrum receives acoustic energy that is coupled via the middle and inner ear to hair cells in the cochlea, the site of transduction. In the mammalian ear, low energy density sound signals are conducted through the ossicles in the middle ear, an impedance conversion device that greatly facilitates the ingress of acoustic energy into the cochlea. The capture of acoustic energy by the fluid-filled ears of mammals thus requires elaborate structural mechanics . Insects exhibit auditory systems that are structurally simpler but still achieve complex signal analysis [2,3]. In locusts, both acousto-mechanical conversion and information processing are readily performed by the tympanum; a membrane endowed with unconventional structural and material specializations , that performs limited discretization of sound waves into distinct frequency components . The histological architecture and micromechanics that endow the microscale ears of insects with high sensitivity and frequency selectivity are still poorly understood, yet they constitute valuable evolutionary solutions to long-standing problems of sound detection.
The locust tympanal membrane (TM; figure 1) serves as a good model to explore frequency analysis in microscale auditory systems; it is experimentally amenable and has identifiable structural features that enable computational modelling. The kidney-shaped membrane (2.5 × 1.5 mm in size) is supported by a rigid cuticular rim around its perimeter, and presents two distinct regions; a thin and transparent membrane (delineated in red, figure 1a), and a thick membrane (delineated in green) where the folded body (FB) and the pyriform vesicle (PV) are located. The FB and the PV are connected on the inner surface to Müller's organ, which contains about 120 mechanoreceptor cells distributed in four distinct groupings (a–d) (a, b, c: connected to the FB and d connected to the PV) shown in figure 1b . The auditory nerve, connected to Müller's organ, in turn conveys acoustic information to the central nervous system [7,8]. The TM therefore performs the complete mechanical function of coupling airborne acoustic energy to the nervous system. The locust has an interesting mechanical response to impinging sound waves. Low frequencies (less than 10 kHz) elicit travelling flexural waves of the TM that cause mechanical deflection at the both the FB and PV. By contrast, higher sound frequencies (more than 10 kHz) only deflect the PV . Phenomenologically, this frequency response is achieved by waves travelling in a frequency-dependent manner to different mechanoreceptors attached to the two regions .
Understanding the vibrational dynamics of the TM is critical to understanding how the ear collects acoustic energy, and then distributes it within the TM structure to produce mechanical oscillations at different mechanoreceptor sites. Here, we characterize the geometry and vibrations of the TM, using enhanced methods, derive analytic models and predictions for TM behaviour and, using finite element analysis (FEA), we identify the TM features required for energy localization and frequency discretization, namely a heterogeneous thickness distribution and tension.
2. The locust tympanum as a membrane or stiff plate
Two distinct physical properties strongly influence the mechanical behaviour of homogeneous membranes; tension and stiffness . The locust TM is known to be under tension, as demonstrated by the fact that a linear cut readily relaxes into an oval opening . Also, changes in the pressure in the air sac behind the TM modulate the frequency response of the TM and changing the sensitivity of receptor cells , presumably through alteration of membrane tension. The membrane is likely to have non-negligible bending stiffness, especially in the thick region around the PV and FB. Measuring membrane tension and stiffness directly using local probe methods (atomic force microscopy) has proved very unreliable, owing to the uncontrollable adhesion of the probe to the sample during measurement (T. R. McDonagh & D. Robert 2010, personal observation).
Hence, the influence of tension and stiffness on flexural wave velocity were explored using theoretical predictions of vibration behaviour. Predictions were derived for two contrasting models of TM dynamics:
(1) Membrane model: a membrane with uniform in-plane tension and negligible stiffness.
(2) Stiff plate model: a membrane with isotropic stiffness and negligible tension.
Using wave-equations for each model, we derive an equation for the propagation velocity of the travelling flexural wave which described dependence on tension T, thickness h, density μ, Young's modulus E, Poisson's ratio ν and angular frequency ω (see the electronic supplementary material, note S1 for derivations) . 2.1
Therefore, if the locust membrane is a tension-dominated system, the velocity of the travelling flexural wave will decrease with increasing thickness of the membrane, i.e. travelling flexural waves will slow down as they approach the PV. If the stiff plate model dominates, the model predicts that the velocity of the wave will increase as it approaches the PV. Measuring the change in travelling flexural wave velocity during propagation allows us an experimental way to distinguish between the two possibilities.
3. Material and methods
Experiments were performed on seven adult male or female Schistocerca gregaria Forskål. The animals (supplied by Blades Biological, Cowden, UK) were kept in standard conditions (temperature: 24–26°C, relative humidity: 60–70%). Animals remained alive throughout the experiment but were anaesthetized by injection of glutamate in saline solution (approx. 10% by volume) to prevent abdominal breathing, which causes large and slow movements of the TM resulting in temporary loss of signal from the laser Doppler vibrometer (LDV). Preliminary experimental work showed that the use of glutamate had no observable effect on the membrane response. To allow optical access to the tympanal ear during laser vibrometry measurements, the wings were removed. The animals were fixed, ventrum down, to a horizontal brass bar and held firm by a looped wire around the animal and the bar. The brass bar was orientated such that the animal could be held at a slight angle from the perpendicular to the laser beam to fully expose the tympanal ear. The intact right ear of each animal was examined (figure 1c).
3.2. Coloured Newton ring interferometry
Reflected illumination creates optical interference between light reflected from the top and bottom surfaces of the TM. Using coloured Newton ring interferometry allows us to accurately visually identify the distribution of thin regions of the membrane. Note that membrane thickness quantification using Newton rings is not possible here as it requires knowledge of the local refractive index. The technique clearly reveals the geometric distribution of both the thin concentric region of the membrane and the opaque thick region (figure 1c). These images also informed the selection of regions to investigate with the focused ion beam (FIB) milling.
3.3. Focused ion beam milling
Accelerated gallium ions were used to etch the TM with high precision at five selected locations. Freshly excised TMs were mounted, gold coated by plasma deposition and placed under vacuum (less than 2.5 μbar). A FIB milling station (FEI FIB201, OR, USA) was used to mill small trenches from which TM thickness could be measured.
3.4. Membrane response to acoustic stimulus
Real-time velocity measurements of the vibrating TM were performed using time-resolved laser Doppler vibrometry. No beads or reflective coating was necessary to obtain adequate laser reflection. The surface velocity was measured at many (minimum of ≈400 points) scan points across the tympanum in a radial arrangement about the PV. Previous studies showed that travelling flexural waves above 12 kHz had shorter wavelengths about the PV [5,9]. Average velocity was calculated from 100 measurements made at each scan point. The ear was stimulated by sound at a single frequency for four cycles (5, 10, 15, 20 and 25 kHz at 80 dB SPL (200 mPa)). Acoustic signals were generated by a waveform generator (Agilent 33120A, CA, USA) amplified (Sony TAFE570, Tokyo, Japan) and fed to a Mylar leaf high dynamic range loudspeaker (ESS AMT-1, CA, USA) positioned 30 cm from the ear. The SPL was measured using a calibrated one-eighth inch precision pressure microphone (Bruel & Kjaer 4138, Nærum, Denmark) and preamplifier (Bruel & Kjaer 2633). This reference microphone was positioned at approximately 10 mm from the tympanum. High coherence between microphone signal and mechanical response was achieved across the tympanum (more than 0.85; shown in the electronic supplementary material, figure S1). The transverse membrane velocity was measured optically using a microscanning LDV (Polytec PSV-300-F, Waldbronn, Germany) with an OFV-056 scanning head with close-up attachment. The laser spot of diameter ≈5 μm can be positioned with an accuracy of ≈1 μm. Laser vibrometry experiments were carried out on a vibration isolation table (TMC 784–443–12R, MA, USA) inside an acoustic isolation booth (IAC 1204A, New York, NY, USA) at room temperature 24–26°C and relative humidity of 40–62%.
Polytec software (v. 8.5, Waldbronn, Germany) was used to create out-of-plane spatial plots of velocity. Contour polar plots of kinetic energy (KE) density were created using POLAR3D program in Matlab 7.5 (The MathWorks Inc, MA, USA). KE density, β, was calculated for each scan point using equation (3.1), 3.1where, V is the out-of-plane velocity and μ the density with a literature value of 1300 kg m−3 . The thickness, h, was estimated using a published thickness map which is more conservative than our measurements . To accurately measure the velocity of the travelling flexural waves across the membrane, the transverse velocity was recorded with high spatial resolution along a straight line through the FB and PV. These velocity data were integrated with respect to time to calculate displacement profiles along the scan line, using a custom-made program written in Lab VIEW v. 8.0 (National Instruments, TX, USA). An automated Gaussian curve-fitting program was written in Matlab 7.5 to track the position of the positive displacement peak of the travelling flexural wave at each time sample. The velocity of the travelling flexural wave along the scan line was then calculated by differentiating the peak position with respect to time.
3.5. Finite element analysis
A time-dependent model of the TM was developed using COMSOL (COMSOL 4.3, Stockholm, Sweden). Our approach to creating the TM geometry was to use sectioning-derived thickness measurements from literature  as well as FIB thickness measurements to create a simplified three-dimensional geometry (figure 1i). The periphery of the TM was fully constrained as this was considered to be a suitable simulation of the sclerotized cuticle ridge. In effect, velocity measurements on the ridge are negligible. To ensure that the geometry was discretize with sufficient fidelity, a mesh sensitivity study was conducted and it was observed that 64 k elements were sufficient to capture the response of the membrane. An increase in mesh element number to 100 and 200 k resulted in deflection differences of less than 1%, thus indicating a stable solution independent of mesh density. Similarly, a time stepping sensitivity study was carried out.
The prestrain in the model was captured by initially applying an in-plane displacement. A sinusoidal load (four cycles, as per experiment) was then applied to the model membrane surface to simulate the acoustic surface loading (at a pressure of 200 mPa, and at 5, 10, 15, 20 and 25 kHz).
Given the absence of published material data for insect tympana, for a first approximation, the membrane was modelled as an isotropic linear-elastic solid requiring only the density, Poisson's ratio and Young's modulus as material parameters. A density of 1300 kg m−3 was taken from literature . The Poisson ratio was found to have little effect for values between 0.1 and 0.4, but has been shown by Aernouts et al.  to have an effect on membrane mechanics for values more than 0.4. The commonly used value of 0.3 was used for the Poisson ratio. As discussed previously, the real TM is known to be under prestrain, which needs to be considered for representative modelling.
Two quantities that are difficult to measure, but are required for modelling; prestrain and Young's modulus. In effect, these two quantities are somewhat intractable as membrane stiffness is directly proportional to the tensile force, an effect known as ‘stress stiffening’ [17,18]. As a consequence, a membrane with low intrinsic stiffness and high prestrain behaves similarly to a membrane of high intrinsic stiffness and low prestrain. Therefore, without knowing one parameter, the other is also unknown; thus far, the measurement of either has proved very challenging. Here, a Young's modulus of 20 MPa was chosen as it has been used in other tympanic models . The strain was then chosen such that the membrane deflections matched experimental values, resulting in strain of 1%, a value known to be modest compared with other known biological tissues . Rayleigh damping was used for all the FEA models with damping parameter values of α and β taken from literature . For frequency-independent models α = 260 s−1 and β = 37 µs. For frequency-dependent damping models, α = 260 s−1 and β ranged linearly from 37 to 74 µs for 5 and 25 kHz, respectively. TM damping was formulated such that damping forces were directly proportional to the frequency of the acoustic stimulus. This assumption is based upon the principle of frequency-dependent damping at a liquid–solid interface [21,22].
4.1. Membrane microstructure
A key mechanical property of microphone diaphragms and TMs alike is membrane thickness. Previous measurements using a microtome method found that membrane thickness within an individual TM varies considerably, ranging from 0.6 to 20 μm . We used two approaches to study its thickness distribution in more detail; colour Newton ring interferometry (figure 1c–e) and FIB milling (figure 1f–g). The locust membrane presents a central thick region, containing neuronal attachment sites, and a surrounding roughly semicircular thin region (figure 1). TM thickness ranges from approximately 100 nm in the thin region to 15 μm at the PV. At position IV, at the opaque region near the PV (figure 1c,e), the membrane is 5 μm thick, exhibiting internal cavities with bridging links between the internal and external surfaces (figure 1g). Some local FIB redeposition artefacts are notable (figure 1g; see the electronic supplementary material, figure S2), probably resulting from liquids evaporating during cutting, suggesting that these cavities are fluid filled. Light microscopy imaging (figure 1d–e) also reveals that the white region (zone IV–V) has a compartmental structure (figure 1e), supporting the presence of fluid-filled chambers. The PV (figure 1c position V, (e) top right corner) is 15 μm thick at its apex; making it distinctly thicker than the surrounding membrane.
4.2. Tympanal response to acoustic stimuli and energy localization
The TM's mechanical response was tested using tonal stimuli at different frequencies; 5, 10, 15, 10 and 25 kHz (example frequency response shown in the electronic supplementary material, figure S3). For all frequencies, the initial TM displacement exhibits a semi-concentric travelling flexural wave developing in the thin region (figure 2a see the electronic supplementary material, movie S1). The travelling flexural wave subsequently converges on the PV and the thick region. The laser Doppler vibrometry measurements reveal that the travelling flexural wave slows down as it propagates towards the PV (figure 3a) and its wavelength shortens, causing the deflection to increase in magnitude (figure 3b,c). This is similar to wave shoaling, which occurs as sea waves approach shallower water at the shore .
To accurately quantify the velocity of the travelling flexural wave, we measured TM vibration displacement (normal to the surface) at high spatial resolution along a line in the established direction of wave propagation (figure 3b). The peak of the travelling flexural wave was spatially tracked allowing the measurement of its propagation velocity. The velocity of the travelling flexural wave (figure 3a) is relatively large in the thin region, where it first develops, and is notably frequency dependent. As the travelling flexural wave propagates towards the PV, propagation velocity decreases. This slowing down behaviour is observed at all frequencies, except for 5–10 kHz, supporting the membrane hypothesis. For 15 kHz and above, wave velocity reduces to zero at the PV where propagation terminates. At 10 kHz, wave velocity slows down but is not zero. At 5 kHz, however, the travelling wave traverses the PV without significant velocity reduction. This experimental analysis shows that in response to sound pressure, the TM's restorative force is principally owing to membrane tension rather than stiffness. Altogether, this evidence indicates that tension is a key feature of tympanal function. It thus becomes apparent that real TMs most probably have a specific bending stiffness while also being under tension, as is the case for the FEA models that have both intrinsic stiffness and an applied tension.
An additional and important effect arising from the behaviour of TM travelling waves is the localization of KE. In effect, the density of KE rises as the 15 kHz wave travels across the TM (figure 2b). Quantitatively, KE rises from KE = 2 ± 0.5 at stimulus onset to 12 ± 0.5 nJm−2 at the end of the stimulus. After one full oscillation period, the KE is highly localized exactly at the PV. At 15 kHz, the shoaling-like behaviour results in the energy density at the PV to be two orders of magnitude higher than at any point on the thin membrane (figure 4). A similar spatial pattern of energy localization also takes place at frequencies above 15 kHz (figure 4).
4.3. Finite element analysis
A simplified geometry based upon the real TM was derived, which captures the main TM structural characteristics: a thin semicircular region, a surrounding area of average thickness, the PV and the thick FB region (figure 1h–i). Material properties were based on known tympana (see Material and methods). The FEA shows that a simple pretensioned, isotropic and linear-elastic membrane is sufficient to capture all the main features of the real TM, namely the generation of a travelling flexural wave, frequency decomposition and KE localization.
As in the real TM, during the initial part of the sound stimulation (figure 2c, φ < 180°), a semi-concentric wave is produced in the thinnest region of the FEA membrane, for all frequencies tested. The wave then travels across the membrane and converges around the PV (figure 2c; see the electronic supplementary material, movie S2). Flexural waves produced by lower frequencies (less than 10 kHz) travel past the PV and reach the FB, similar in response to the real TM. When elicited by higher frequencies, flexural waves are reduced in amplitude (figure 3e). Thus, frequency decomposition is possible with this simulation and, as with the real locust TM, the ratio of maximum PV to FB displacement increases for increasing frequency (figure 5). The simulation is also able to capture the localization of KE taking place at the PV, and with a topographical structure similar to the real TM (figure 2d).
To the authors knowledge, pretension has as yet not been included in animal tympanum models [19,24–28] even though there is evidence of pretension existing in tympanic membranes [11,29,30]. Recent research indicates that ‘prestrain plays a critical role in the mechanics of thin biological membranes’ [31,32]. In our computational models, the inclusion of membrane tension reveals that tension is a requirement for the presence of both flexural waves and KE localization. Without the inclusion of membrane tension, the membrane behaves as a conventional vibrating plate without flexural wave propagation, frequency decomposition or KE localization (figure 5).
The locust TM is structurally heterogeneous, a feature that distinguishes it from conventional microphone membranes. Collectively, data from spatially resolved Doppler vibrometry and FEA models suggest that variation in membrane thickness and tension produce two major effects. The gradient in compliance across the TM surface causes the thinner parts of the membrane to be more readily deflected by incoming sound energy. The semicircular shape of this compliant part of the membrane causes the flexural wave to travel and converge towards its centre, causing an effective focusing of the acoustic energy captured by the membrane (figure 2a–d). The increase in membrane thickness along the direction of travel of the flexural wave generates the second behaviour. As in the locust TM the restoring force is dominated by tension, the flexural wave formed slows down as the membrane thickens, as shown in equation (2.1). The reduction in this propagation velocity directly affects other wave parameters; the length of the flexural wave shortens and its amplitude increases. The consequence of the spatial convergence and shoaling of the flexural wave is a large deflection at the PV, which is located at the point of convergence (figure 2a,c). The described mechanism is biologically relevant because it localizes acoustic vibrational energy on to mechanoreceptors near the PV, and therefore enhances the ear's sensitivity to sound. Two features define the locust TM; concentric shape of thickness gradient and tension-dominated mechanics. These features, when combined, improve hearing sensitivity. Interestingly, a similar distribution of thickness has also been observed in mammalian tympana [33–35].
Location-specific frequency decomposition is a purely mechanical effect of the wavelength of the travelling flexural wave interacting with the particular structure of the TM. At high frequencies, the wavelength of the travelling flexural wave is smaller than the dimensions of the TM, therefore flexural wave velocity will be dependent on the thickness at a smaller spatial scale, while wave speed and shape will be susceptible to local thickness gradients. By contrast, for long wavelengths, the effective thickness experienced by the flexural wave is the average thickness over that larger distance, thus making it impervious to local thickness gradients. We propose that the long wavelength of the travelling flexural wave at 5 kHz averages out local changes in membrane thickness and produces a nearly constant velocity. By a similar argument, the strong frequency dependence of travelling flexural wave velocity at the start of propagation could occur because higher frequencies have smaller wavelengths, and hence are more sensitive to the local changes in membrane thickness.
The TM is structurally complicated; FEA enables the incorporation of structural complexity into functional models testing the conjecture that material properties, concentric thickness gradients and tension together produce the observed mechanical response.
However, in formulating the FEA model, certain simplifying assumptions are necessary, and here we unfold their implications. For instance, TMs are usually modelled in an isotropic linear-elastic regime [19,36–38] even though they are probably anisotropic and viscoelastic . Given the difficulty of dynamic material characterization, using an isotropic linear-elastic regime often proves to be an acceptable first approximation, as validated by the present locust TM model. Ignoring the contribution of tension [19,26], these models only indirectly account for the behaviour by way of the stress-stiffening effect . The evidence presented, using experimental, mathematical and FEA approaches, suggests that tension is a crucial factor in the generation of the full range of mechanical behaviour.
Another assumption pertains to the damping behaviour of the liquid-filled chambers observed in some regions of the TM (figure 1g). Previous work has suggested that the damping experienced by flexural waves at a liquid–solid interface increases with frequency [21,22]. Here, damping is found to enable the membrane to terminate flexural waves and in a frequency-dependent manner. Yet, models where the damping forces are independent of frequency can still explain frequency-dependent localization (figure 5). Such models can behave like the locust TM when the thickness of the thick region is scaled by a factor of three or more. This adaptation is sufficient to produce frequency localization even in the absence of frequency-dependent damping (figure 5). Notably such a threefold change results an increase in membrane stiffness by a factor of 27, as flexural stiffness is proportional to the cube of thickness. Such stiffening would thus reduce the deflection magnitude of the membrane for a given pressure, likely reducing the mechanical sensitivity of the hearing organ.
The liquid-filled partitions surrounding the PV (figure 1g) seem to be implicated in the mechanism of frequency discrimination. This assertion is supported by experiments on excised membranes that do not show the frequency-dependent travelling waves observed in in situ membranes . We conjecture that this change in mechanical response is caused by the collapse of the fluid–solid interface, generating substantial changes in TM damping properties [40–42]. The FEA results show that the structural mechanism underlying this response is robust and can be readily generalized. There is therefore an opportunity for designing a new type of sound sensor based on the principles at work in the locust TM.
Conventionally, microphones are designed to prevent the formation of travelling flexural waves across their diaphragm. Such mechanical response is deemed undesirable, as it alters the sought-after flat frequency response. The energy localization observed in the locust, and the proposed mechanism, may be of interest to a different microphone design as it would allow transduction to take place at a location of maximum deflection. Further along, an array of such membranes can be imagined that would serve as a sensitive mechanical frequency analyser, powered by sound energy only. Such solution to acoustic frequency analysis constitutes an alternative to conventional digital signal processing and to the mechanisms found in the mammalian cochlea.
N.M. is supported by a Marie Curie fellowship. D.R. is supported by the Royal Society of London and the Wolfson Foundation. The project was financially supported by a grant from the Biotechnology and Biological Sciences Research Council UK.
R.M., T.R.M., N.M. and D.R. wrote the paper. T.R.M. conducted thin film interferometry, vibrometry and analysis of results, and formulated mathematical models. N.M. conducted FIB milling and analysis, and analysed vibrometry data. T.S.S. conducted FIB milling and analysis. R.M. conducted FEA modelling and analysis. D.R. oversaw research.
- Received September 20, 2013.
- Accepted October 14, 2013.
- © 2013 The Author(s) Published by the Royal Society. All rights reserved.