Active amplification in insect ears: mechanics, models and molecules

Abstract

Active amplification in auditory systems is a unique and sophisticated mechanism that expends energy in amplifying the mechanical input to the auditory system, to increase its sensitivity and acuity. Although known for decades from vertebrates, active auditory amplification was only discovered in insects relatively recently. It was first discovered from two dipterans, mosquitoes and flies, who hear with their light and compliant antennae; only recently has it been observed in the stiffer and heavier tympanal ears of an orthopteran. The discovery of active amplification in two distinct insect lineages with independently evolved ears, suggests that the trait may be ancestral, and other insects may possess it as well. This opens up extensive research possibilities in the field of acoustic communication, not just in auditory biophysics, but also in behaviour and neurobiology. The scope of this review is to establish benchmarks for identifying the presence of active amplification in an auditory system and to review the evidence we currently have from different insect ears. I also review some of the models that have been posited to explain the mechanism, both from vertebrates and insects and then review the current mechanical, neurobiological and genetic evidence for each of these models.

Active versus passive sensing

Insects use their ears for a range of functions: to find resources, such as mates or hosts to parasitize, but also to avoid becoming a resource, to avoid parasites and predators (reviewed in Gerhardt and Huber 2002). Auditory function is thus essential and under strong selection. Auditory systems are therefore often finely honed to respond to the most salient sounds.

Research has uncovered many mechanisms that auditory systems use to ensure sensitivity. One broad class of adaptations are structural adaptations which use physical form to generate function. They include the resonance of field cricket tympana, which crickets exploit to be tuned to their conspecific song (Paton et al. 1977; Larsen et al. 1989). They also include the structure and geometry of the locust ear which allows locusts to discriminate different frequencies of sound (Robert 1989; Windmill et al. 2005; Malkin et al. 2014). The structural mechanics of the ear of the tiny parasitoid fly Ormia ochracea underlies its exceptional directional sensitivity (Miles et al. 1995; Robert et al. 1996, 1998). Similarly, the unique acoustical properties of field cricket trachea also underlie directional hearing (Michelsen et al. 1994; Michelsen and Löhe 1995). Each of these adaptations is based purely on the morphology and mechanical properties of the auditory system and is, therefore, termed passive.

‘Active’ sensing on the other hand requires an ongoing expenditure of physiological energy. One form of active sensing is behavioural (Nelson and MacIver 2006). Bats use sonar to navigate their environment and must produce echolocation calls which are expensive to produce at least at rest (Speakman et al. 1989). Electric fish expend energy to produce the electrical fields they use to detect their prey (Salazar et al. 2013). Among insects, some parasitoid wasps actively tap the leaves of the plant in which their hosts live and detect them using the returning vibrations (Fischer et al. 2001).

There is another form of active sensing as well. Energy can also be expended to increase the sensitivity of the sensory transduction machinery and indeed it is in several auditory systems (Martin et al. 2001; Nadrowski et al. 2004; Göpfert et al. 2005; Nadrowski et al. 2008). Active auditory amplification was initially discovered in vertebrates and only much later from dipterans (Drosophila melanogaster and mosquitoes) (Göpfert and Robert 2001, 2003). Only very recently, has it been observed in an orthopteran, another insect lineage known to have auditory capabilities (Mhatre and Robert 2013).

This discovery hints at the possibility that this active process is much more ancient than currently appreciated. There have been some previous attempts at identifying it within other members of the insect family. These attempts have had shortcomings that need to be addressed. This review aims to clarify the well-established benchmarks for identifying active auditory amplification for future investigations. To do this, I will explain (1) the expected phenomenology required to establish active amplification and the phenomenology observed in insect ears; (2) the different models proposed to explain the process; (3) the mechanical, neuronal and molecular evidence for the different models; and (4) the behavioural contexts in which auditory active amplification is used. The review will thus also serve to clearly establish the data, the controversies and most importantly, the gaps in our knowledge about this fascinating phenomenon.

Active auditory amplification and transduction

I will first describe the chain of events that occur within the insect auditory system that convert sound into a neuronal signal. I will do this to be able to point out the stage and manner in which the sub-cellular machinery associated with the mechanotransduction channels expends energy to amplify sound. More detailed models are presented subsequently in “Models of active amplification”.

Sound into stimulus receiver structure displacement

Auditory transduction is a subset within a larger category of sensory transduction—mechanotransduction. Sound is an oscillating mechanical disturbance that travels through a medium. The most peripheral organ of the auditory system, the stimulus receiver structure is adapted for being forced by this disturbance. The two most distinct categories of this stimulus receiver structure (SRS) in insects are tympana and antenna (Fig. 1a, b). Tympanal ears are displaced mainly by the pressure component of sound and tend to work at large distances from the sound source. Antennal ears, on the other hand, are displaced mainly by the particle velocity component of sound, i.e., the motion of air particles and typically work best close to the sound source. Nonetheless, the fundamental function of both is to convert sound into a mechanical displacement.

Fig. 1

The peripheral organs in insect auditory systems are structurally diverse and respond to both sound and particle velocity, but they all serve to convert sound into a force that they transmit to very similar mechanotransductory cells. a A tympanal ear of a tree cricket, for instance, responds to sound and is very different from, b the antennal ear of a male mosquito (fl flagellum, pd pedicel, adapted from Jackson et al. 2009). However, they are both subserved by chordotonal organs made up of similar (c) scolopidial units. Scolopidia in insects usually contain two or three mechanosensory neurons. Each neuron has a ciliated dendrite which is surrounded by a spindle-shaped space created by scolopale cell. The scolopale cell contains actin microfilaments, as well as microtubules and tropomyosin, and thus may be able to play an active role as well. However, it is generally believed that the active contributions to mechanosensation arise from the dendritic cilium. The dendritic cilium has a 9 × 2 + 0 organisation and is nonetheless motile. It undergoes an inflation known as the ciliary dilation within the scolopale space and this anatomical feature marks an important transition in ciliary structure. Below the ciliary dilation, we find the presence of axonemal dynein motors and above it these are absent. These axonemal dyneins are associated with axonemal microtubules. These motors are responsible for inter-tubule sliding, and thus cause the cilium to bend. Other dynein and kinesin motors associated with inter-flagellar transport are also known to be present within the cilium, and can processively move along the microtubule backbone. The ciliary dilation also marks an important transition in the expression of two families of mechanosensory TRP channels in the cilium. NompC, the TRP N channel is only found distal to the dilation. The two TRP V channels, Nan and Iav are only found proximally, the only part of the cilium that possesses axonemal dyneins (adapted from Todi et al. 2004; Kavlie and Albert 2014; Yack 2004)

Displacement into mechanosensory deformation

In the next stage of this chain of events, SRS displacement must be used to deform a mechanosensory cell. The form of deformation experienced by each cell (stretching, compression or bending) will be determined by the nature of its mechanical coupling to the SRS and other parts of the insect body. Mechanosensory cells in the auditory systems of insects are integrated into a structure called a chordotonal organ (Fig. 1c; Todi et al. 2004). In some insects such as the mosquitoes, flies and locusts, the chordotonal organ is directly attached to the SRS and mechanical coupling appears straightforward (Gray 1960; Göpfert and Robert 2001, 2002; Yack 2004). In other insects, such as crickets, katydids and cicadas, chordotonal organs are not directly attached to the tympanum but have some bridging structure between the SRS and mechanosensory cells (Yack 2004). In crickets, this is the tracheal wall (Young and Ball 1974), in katydids, the auditory vesicle (Rossler et al. 1994; Montealegre-Z et al. 2012) and in cicadas the tympanal apodeme (Young 1977). In each of these cases, coupling is complex and we do not understand its mechanics. Yet, given that these cells form the basis of sound perception, the mechanical coupling must be such that it can transform the sound-induced SRS deflection into a neural signal.

Deformation to transduction to active amplification

It is at this stage that active amplification is initiated. The essential basis of the active process is that motor molecules within the mechanosensory cell expend energy and generate forces that potentiate and enhance the imposed mechanical forces. Put another way, forces generated internally within the cell assist the externally applied sound-based forces in opening mechanotransduction ion channels. The exact mechanisms that subserve this active process are the subject of much debate and research, both in vertebrates (Ashmore et al. 2010) and invertebrates (Göpfert 2008; Kavlie and Albert 2014).

In mammals, the debate centres on the site of active amplification system; whether in the electromotile cell body of the mechanosensory cell (Nobili et al. 1998) or in a structure found on the top of the cells known as the hair bundle (LeMasurier and Gillespie 2005) or whether both contribute to the process (Maoiléidigh and Jülicher 2010; Peng and Ricci 2011). Active amplification based in the cell body is linked to the unusual motor protein prestin and is known only from birds and mammals (Liberman et al. 2002; Beurg et al. 2013; He et al. 2014). It is unlikely to be important in insect auditory systems (see Kavlie et al. 2014).

The active mechanism in other vertebrates is based in the hair bundle (Manley 2000, 2001). The hair bundle is a set of finger-like protrusions found on the top of vertebrate mechanosensory cells. The basic hair bundle contains a single long protrusion called the kinocilium, surrounded by shorter stereovilli (Manley and Ladher 2008). The kinocilium is a true cilium and has a microtubule core (Manley 2000) and has dynein at regular intervals along the cilium whose motor activity presents one of the possibilities for the amplification motor (Camalet et al. 2000; Kindt et al. 2012). Alternatively, the stereocilia have actin cores with associated myosin motors which may instead be responsible for amplification (Holt et al. 2002; Gillespie and Cyr 2004).

In insects, the cilium in the chordotonal organ is homologous to the kinocilium in the hair bundle (Manley and Ladher 2008). Several mutations in proteins expressed within the cilia affect auditory transduction, as well as active amplification. Hence, it is generally believed that a ciliary mechanism is used for active amplification (Senthilan et al. 2012; Kavlie and Albert 2014). SLC26, a paralogue of prestin, has been discovered in insects (Weber et al. 2003) but it does not seem to be crucial to auditory amplification (See Kavlie et al. 2014).

When the cilium is deformed by SRS displacement, mechanosensory ion channels are opened. The exact opening mechanism is also under debate (Christensen and Corey 2007). Whatever the mechanism, once ion channels open, they allow ions into the intracellular space setting off a chain of neurobiological events that eventually culminate in action potentials being evoked by the auditory signal. However, in addition to the external forces due to SRS displacement, in active amplification, we see the emergence of forces generated within the cell, specifically the cilium in insects. There are motor molecules within the cilium and their activity is thought to increase the deformation caused by external forces. Thus, active amplification is primarily a mechanical process. Forces generated by ciliary motors increase the opening probability of mechanosensory channels, specifically in response to very low sound levels at the mechanosensory cell’s best frequency (Nadrowski et al. 2008). In insects, this amplified frequency is usually the frequency of conspecific song (Göpfert and Robert 2001, 2003; Riabinina et al. 2011; Mhatre and Robert 2013).

Detecting active amplification

The signatures of amplification can also be detected in the neurological activity of the auditory system (Göpfert and Robert 2001; Robles and Ruggero 2001; Nadrowski et al. 2008; Warren et al. 2010). In addition, in insects, such activity can also be detected in the vibrations of the SRS (Göpfert and Robert 2001, 2003; Mhatre and Robert 2013). The allure of mechanical over neurological measurements is obvious. In insects, highly sensitive laser Doppler vibrometry can be used to make measurements of displacements down to the picometer level. This can be accomplished in an intact system, without making any physical contact with the organism during measurement. Auditory organs can be stimulated using both naturalistic free field sound (Göpfert and Robert 2001, 2003; Jackson and Robert 2006; Mhatre and Robert 2013) or if so required more controlled electrostatic force stimuli (Albert et al. 2007; Nadrowski et al. 2008). As a result of this combination of experimental amenability, a diversity of auditory form and adaptive function, and genetic amenability (at least in the Drosophila model system), auditory insects are quickly becoming extremely attractive model systems for studying active amplification.

Signatures of active amplification

A suite of mechanical features characteristic of active auditory amplification has been identified (Hudspeth 1989; Manley 2001; Duke and Jülicher 2007). These are common to vertebrates (Manley 2001; Robles and Ruggero 2001) and invertebrates (Göpfert and Robert 2007; Mhatre and Robert 2013). The main observed characteristics are (1) frequency-selective, compressive non-linearity, (2) sound level-dependent gain, (3) two-tone suppression accompanied by the generation of distortion products, (4) self-sustained oscillations of the SRS and (5) energy gain. I will tackle each characteristic in turn and review the evidence from different insect systems.

Frequency-selective compressive non-linearity

Transfer functions give us the relationship between a driving stimulus and the resulting vibration of a structure over a range of driving frequencies. In auditory systems, we typically measure the transfer function between SRS displacement amplitude in response to uniform sound level at different frequencies (Fig. 2). In a passive system, we expect that this transfer function will remain independent of the stimulation level (at least within the elastic limit of the system: see Fig. 6). However, in active auditory systems, it depends on stimulation level and is usually called the sensitivity function. In the canonical form, the SRS is most sharply tuned at the lowest stimulation levels and as the stimulus level is increased, this tuning grows less sharp (Fig. 2). This decrease in sharpness is also accompanied by a reduction in sensitivity (ratio of the displacement to stimulus level; Fig. 2). Active amplification is highly frequency selective. If the frequency of a stimulus is greatly different from the best frequency the SRS’s sensitivity at that frequency is independent of stimulus level (Fig. 2).

Fig. 2

The transfer functions of a a tree cricket and b Drosophila SRS in response to different stimulation levels in their passive state (OFF state and post mortem) show the behaviour of a passive oscillatory system. In this state, the amplitude of the stimulus is irrelevant and the system sensitivity remains constant. However, when they are being actively amplified the sensitivity of the two auditory systems is both stimulus level dependent as well as frequency dependent. Note the change in best frequency in the Drosophila antenna (a adapted from Mhatre and Robert 2013; and b from Göpfert and Robert 2003)

Frequency-selective compressive non-linearity has been observed only in the SRSs of Drosophila species, mosquitoes and a tree cricket (Göpfert and Robert 2001; Riabinina et al. 2011; Mhatre and Robert 2013). There are, however, subtle differences even between these three groups (Fig. 2). In both mosquitoes and tree crickets, the best frequency is independent of the stimulus level and the tuning sharpness of the ear decreases with increasing stimulus level (Göpfert and Robert 2001; Mhatre and Robert 2013). In all Drosophila species tested so far, the best frequency of the transfer function increases considerably with stimulus level (Göpfert and Robert 2002; Riabinina et al. 2011). In Drosophila melanogaster, the best frequency changes from <200 to 800 Hz for stimulus amplitude changes from 1 to 100 pN, representing a >400 % change (Nadrowski and Göpfert 2009a). Similarly large changes have been observed across the entire genus over antennal displacements of 0.2–20 μm (Riabinina et al. 2011).

Surprisingly, while the best frequency changes, the sharpness of the tuning of the Drosophila SRS (as indicated by the Q factor) does not (Göpfert and Robert 2002; Nadrowski and Göpfert 2009a). This is contrary to observations both from mosquitoes and tree crickets (Göpfert and Robert 2001; Mhatre and Robert 2013) as well as from other vertebrate systems (Robles and Ruggero 2001). These differences in the mechanics are significant and the possible sources of the differences will have be considered carefully.

Sound level-dependent gain

The behaviour of active systems can also be tested using single frequency stimuli, to measure the exact relationship between gain or sensitivity and stimulus level. It has been observed, that at best frequency, gain typically shows a power law dependence on stimulus level, i.e., the gain or the ratio of displacement amplitude (|x|) to driving force amplitude (|f|) varies with a negative exponent of force amplitude (Göpfert et al. 2006; Mhatre and Robert 2013). The specific dependence predicted by one theory (Duke and Jülicher 2007) is as follows:

$$ \frac{\left| x \right|}{\left| f \right|}\sim \left| f \right|^{ - 2/3} . $$

(1)

With tonal stimulation, the force applied to the SRS in any auditory system will be directly proportional to the sound level, hence, the stimulus amplitude can replace the |f| term. This specific relationship is interesting because it is predicted by a set of theoretical models suggested for the auditory amplification system, i.e., the ‘critical oscillators’ model (Duke and Jülicher 2007). These models describe the amplification mechanism using equations for oscillators poised near a parametric bifurcation point, i.e., such as near a Hopf bifurcation (Duke and Jülicher 2007; Hudspeth et al. 2010). There is, however, intense debate about whether critical oscillators are the best descriptors of the behaviour observed in actively amplified auditory systems (Hudspeth et al. 2010; Champneys et al. 2011; Szalai et al. 2013) and this debate will be covered in Sect. “Models of active amplification”.

Fig. 3

The gain in actively amplified system varies with stimulus level. a Gain in the tree cricket ear is frequency dependent and becomes linear at low levels away from best frequency. At very different frequency the gain is entirely linear. (The data presented here are from both increasing and decreasing stimulus levels and show some hysteresis like the mosquito data (Avitabile et al. 2010). (Data for BF and 7 kHz from Mhatre and Robert 2013, BF + 100 Hz are previously unreported). b The change in gain (G) in the tree cricket ear fits the form predicted by theory (G = C ₁•|P|^−2/3), but only when an additional term (C ₂) is included to account for the passive motion of the tympanal membrane. (BF best frequency and C ₁ and C ₂ are constants. Adapted from Mhatre and Robert 2013)

Whatever the exact classification of the underlying oscillator and bifurcation, data from Drosophila and tree crickets certainly support the power law dependence of gain. The gain predicted by Eq. 1 has been observed in Drosophila wild-type flies (Göpfert et al. 2006), as also in tree crickets (Mhatre and Robert 2013; Fig. 3). In both cases, modifications have been made to the equation to accommodate system-specific behaviour. In Drosophila, the best frequency of the antenna changes with stimulus level (Göpfert and Robert 2003). Hence, gain was measured at the best frequency at each stimulus level and was found to fit Eq. 1 within a certain range (Göpfert et al. 2006). In tree crickets, the best frequency does not change; however, the passive tympanal system has a gain that is comparable to that of the active system (Mhatre and Robert 2013). Hence, in tree crickets, the fit was evaluated to an equation modified to include a linear term for the passive system (Mhatre and Robert 2013; Fig. 3).

Two-tone suppression and DPOAE

Another well-known signature of active systems is two-tone suppression and the production of phantom tones. When an actively amplified system is stimulated by two tones simultaneously, its best frequency and a nearby tone, the system experiences a lower gain at best frequency, i.e., two-tone suppression. Additionally, the system vibrates at a series of new frequencies not present in the original stimulus, known as distortion products. These ‘phantom’ tones are easily observed in power spectrum representation of the SRS or hair bundle motion (Siegel 2008; Barral and Martin 2012). The additional tones thus produced bear a mathematical relationship with two stimulus tones and are known as distortion products (Jülicher et al. 2001; Fig. 4). The levels of the distortion products decay exponentially (Barral and Martin 2012), and hence they are observable above the spectrum noise floor only near the main tones.

Fig. 4

The response of a tree cricket tympanum when stimulated by two tones simultaneously, one at its best frequency (f ₁) and another 200 Hz below that (f ₂). a The spectrum of the tympanal response shows several distortion products which were not present in the original stimulus. These distortion products occur at specific frequency intervals and decay exponentially. The production of such distortion or phantom tones is accompanied by (b) two-tone suppression. Here, the gain at best frequency (f ₁) is reduced when it is paired with another nearby tone (f ₂) but only at low stimulation levels (figures adapted from Mhatre and Robert 2013)

Distortion products are thought to arise from both passive and active non-linearities in auditory systems (Siegel 2008; Barral and Martin 2012). Theoretical models and evidence from hair bundle measurements both suggest that distortion products arising out of the action of the active amplification system occur only close to the frequency of compressive non-linearity and are accompanied by two-tone suppression (Jülicher et al. 2001; Duke and Jülicher 2007; Barral and Martin 2012). Distortion products that arise from passive non-linear behaviour, in the absence of both compressive non-linearity and two-tone suppression, have been observed, indicating that by themselves distortion products are not a reliable indicator of active amplification (Barral and Martin 2012).

In insects, distortion products have been observed in katydids (Möckel et al. 2011), locusts (Kössl and Boyan 1998), moths (Coro and Kossl 1998), mosquitoes (Arthur et al. 2010) and also tree crickets (Mhatre and Robert 2013). While the distortion products in locusts and moths have been shown to be temperature dependent (Möckel et al. 2012) and physiologically vulnerable (Möckel et al. 2007), their association with two-tone suppression remains to be established. Similarly, because compressive non-linearity has also never been tested in the locust or katydid SRS, it is not yet possible to establish whether these result from active amplification behaviour. In moths, where stimulus level effects have been tested for, free field data does not support compressive non-linearity (Windmill et al. 2006). Similarly free field measurements from locusts do not support the presence of distortion products either (Moir et al. 2011).

The clearest evidence in this context is from tree crickets, where both distortion products and two-tone suppression have been observed at a frequency where the SRS shows compressive non-linearity (Mhatre and Robert 2013). Additionally, it has been established that these distortion products decay exponentially (Mhatre and Robert 2013). Among mosquitoes, distortion products but not two-tone suppression were measured using a neural preparation in Aedes aegyptyi (Arthur et al. 2010). These measurements were made close to the frequency range where A. aegypti mosquitoes are known to show compressively non-linear SRS vibrations (Göpfert et al. 1999) suggesting that the process may be linked to active auditory amplification.

Self-sustained oscillations

Self-sustained oscillations (SO) are the vibrational analogs of spontaneous otoacoustic emissions recorded from vertebrate auditory systems (Siegel 2008). They are also the most surprising and the most indicative of all active amplification signatures. In general, one expects that sensory systems are well adapted for being extremely sensitive. For instance, visual systems are optimised so that they can absorb as little as a single photon and produce a transduction event (Lillywhite 1977). Similarly, we expect that auditory systems should be designed to convert even the smallest deflection of their SRSs into a transduction event (Bialek 1987). A naïve expectation then is that the transfer of energy should be strictly unidirectional, from the environment into the sensory system. In actively amplified systems, however, the direction of energy transfer may be reversed; they produce self-generated oscillatory deflections of their sensory neurons and hence also of SRSs, in the complete absence of an external drive. An equivalent process in the visual system would be of eyes emitting photons!

Spontaneous oscillations are thought to be an artefact of the active amplification mechanism, which injects energy into the auditory system to improve its sensitivity (Martin et al. 2001; Göpfert et al. 2005; Nadrowski et al. 2008). Active amplification is like a feedback loop in which SRS deflection produced by an external stimulus is positively reinforced by the action of motor molecules within auditory mechanosensory neurons. Under certain conditions, this positive reinforcement is thought to become oversensitive and reinforces deflections produced by background thermal noise, thus generating self-sustained oscillations. The reversal of the normal flow of energy and our ability to measure this excess energy makes self-sustained oscillations the best indicators of an actively amplified system (Martin et al. 2001; Göpfert et al. 2005; Nadrowski et al. 2008).

Self sustained oscillations in insects have been observed in mosquitoes (Göpfert and Robert 2001), Drosophila (Göpfert and Robert 2003) and tree crickets (Mhatre and Robert 2013). Unlike vertebrate auditory systems, which respond to a range of frequencies, in these three, the active mechanism is tuned only to conspecific frequencies. Thus, self-sustained oscillations, and all the other signatures of active amplification are observed only within the same frequency range.

A feature that discriminates self-sustained oscillations from noise is the phenomenon of entrainment. Spontaneous oscillations at the initial frequency are reduced when a tone at a similar frequency is played to it. In presence of this new tone, the SRS oscillates only at the new driving frequency (Fig. 5). If the SRS oscillation was being produced simply by external noise, in the presence of the new tone the SRS would oscillate at both frequencies. However, a distant frequency does not produce a similar entrainment effect. The self-sustained oscillations of tree crickets’ (Mhatre and Robert 2013) and mosquitoes’ (Göpfert and Robert 2001) SRSs have been shown to be entrained by similar frequency tones. Thus, the evidence for active auditory amplification is quite strong in these insects.

Fig. 5

Spontaneous oscillations are the one of the best indicators of an actively amplified auditory system. However, it is possible to mistake system noise for such behaviour. Only spontaneous oscillations produced by actively amplified systems can be entrained by nearby tones. Noise affecting the measurement would not be similarly entrained. a In tree crickets we find that spontaneous oscillations (blue curve) can be entrained by tones 100 Hz away from the best frequency (red curve) but not by a tone over 1 kHz away (green curve) (adapted from Mhatre and Robert 2013). b It has been suggested that such oscillations are produced because the neurons in the auditory system behave like a non-linear oscillator poised at a bifurcation point such as a Hopf bifurcation. One example of such an oscillator is the van der Pol oscillator, in which damping is variable. The Ca²⁺ concentration (equivalent to the control parameter C) is thought to control the damping experienced by the oscillator. When C reaches a critical point (C _c), the damping is so low that the cell has a tendency to oscillate even in response to small perturbations. There are other types of bifurcations possible and indeed other types of non-linear oscillators that may produce similar behaviour and these are used only as an example (adapted from Duke and Jülicher 2007)

Active ears versus non-linear ears: energy gain

The presence of these signatures raises the question that if one observes all of the characteristics associated with active auditory amplification, are these sufficient to prove the presence of ‘active’ rather than passive behaviour? While the presence of all the signatures is strongly suggestive, to prove that the process is active, one has to establish that energy is being expended. As we have seen, self-sustained oscillations are generally considered a strong indicator of such energy expenditure. Yet strictly speaking, in themselves, they are insufficient proof (Martin et al. 2001). While the SRS may not be driven by sound stimuli, it is nonetheless being bombarded by air particles which are in Brownian motion. While the forces produced by air particles are tiny, SRSs have evolved to be compliant and to respond to small forces.

Hence, to show that the system is truly active and not driven simply by thermal motion, we must show that it violates the equipartition theorem (Martin et al. 2001; Göpfert et al. 2005). The equipartition theorem allows us to estimate the amount of motion observed in a structure at thermal equilibrium, i.e., the motion of a passive structure (Martin et al. 2001). The amount of energy within a passive system at thermal equilibrium is the same as the medium it is immersed in. The ambient energy within a medium is k_BT, where k_B is the Boltzmann constant and T is the temperature in Kelvin. Since the auditory system and the medium are at equilibrium, they must contain the same amount of energy. In an oscillatory system such as an insect SRS, this energy is equipartitioned, i.e., split equally between potential and kinetic energy. Therefore, we can start by considering potential energy in an oscillatory system with only one degree of freedom such as would describe the Drosophila antenna, in which,

$$ \frac{1}{2} Kx^{2} = \frac{1}{2}k_{\text{b}} T $$

(2)

where K is the static stiffness of the system and x is its displacement. For an oscillatory system this relationship is true only over a sufficiently long period of time, over which the system periodically exchanges kinetic with potential energy. This time-averaged quantity is the system’s fluctuation power,

$$ \frac{1}{2} K \langle X \rangle ^{2} = \frac{1}{2}k_{\text{b}} T $$

(3)

The equipartition theorem thus allows us to predict the total mechanical fluctuation power of a passive oscillatory system in thermal equilibrium in the absence of other stimulation. If the system is active, we expect that its power will be greater than this predicted value.

While this is theoretically straightforward, there are unfortunately two problems with this approach. The first is practical: we often simply do not know the static stiffness of the auditory system. The second is much more complicated: biological systems do not usually have a constant stiffness. Specifically in the case of the auditory systems, we know that the stiffness of mechanosensory cells can be very low, even negative, for small displacements and the same cells become stiffer when large displacements are applied (Fig. 6; Martin et al. 2000; Albert et al. 2007; Nadrowski et al. 2008). If one plots the displacement of a mechanosensory cell against the static force applied to it, we can obtain two plots, one from the steady state displacement and another from the initial displacement undergone by the cell (Fig. 6b; Albert et al. 2007). The slope of either plot is the systems’ stiffness. For many non-biological systems, within certain elastic limits, this relationship is linear, i.e., the stiffness is constant. Unfortunately, this is not the case for mechanosensory cells, at least for the initial displacement. This change in initial stiffness is thought to be brought about by mechanosensory channels. As the mechanosensory cell is deformed, some of the forces produced within the cell membrane are imparted to mechanosensory ion channels. Upon being forced, the channels open, leading to a step displacement and hence a change in the slope, i.e., in effective stiffness (Martin et al. 2000; Albert et al. 2007; Nadrowski et al. 2008). This non-linear behaviour depends on the number of channels in the cell and is limited to small deflections, since at larger deflections all the channels are likely to be open. However, this difference in stiffness at small and large deflections makes it difficult to establish whether self-sustained oscillations are produced through active contributions or simply because the displacements involved are small and mechanosensory cells are very compliant in this range (Martin et al. 2001).

Fig. 6

a The stiffness of a material is captured by its elastic or Young’s modulus (E). A curve of the stress (σ) is plotted against the strain (ε) produced in the system and the slope of this curve is E. If a force is applied to a test block of the material as shown in the figure the stress is defined as the force applied divided by the area is applied over (F/A). The strain is defined as the change in the length of the block divided by its initial length (Δl/L). Most materials used by engineers (blue curve) have two stress–strain regimes. In the elastic or linear regime, the stiffness remains constant; but in the plastic or non-linear regime, it changes as increasing forces are applied. In this region, since stiffness is variable, the resonance and transfer function of the system will also change. Biological materials (red curve) usually have two zones of non-linear behaviour: at very low stresses and again at very high stresses. These are passive material properties and do not require the active contribution of energy. They, nonetheless, produce non-linear behaviour. b Analogous data from force–displacement measurement in Drosophila antennae indeed show that the entire system as a whole does possess non-linear stiffness. In this system, this non-linearity is attributed to the opening of ion channels as the JO neurons are placed under stress (adapted from Nadrowski et al. 2008)

These confounding factors make it difficult to test directly whether a system is truly active or passive using the equipartition theorem. This has been addressed both in vertebrate and insect systems by comparing an ‘alive’ active auditory system with one that is recently dead, incapacitated or quiescent. The assumption in such a comparison is that the ion channels are still present and continue to contribute to non-linear stiffness but the motor molecules no longer expend energy and hence a difference in fluctuation power is detectable. In insects, self-sustained oscillations have been recorded only from mosquitoes, fruit flies and tree crickets and their absence have been demonstrated in dead or quiescent states (Göpfert and Robert 2001; Göpfert et al. 2005; Mhatre and Robert 2013). The spectra of all three suggest a marked increase in fluctuation power during self-sustained oscillations; however, an exact calculation has only been carried out only in Drosophila (Göpfert et al. 2005). In Drosophila, an expenditure of energy has also been shown using a model that shows that the energy dissipated by the antenna is greater than that delivered by the driving stimulus (Nadrowski et al. 2008).

It is important here to reiterate the difference between non-linear and active behaviour in auditory systems. There are several possible forms of non-linear behaviour in oscillatory systems, several parameters of the system might change during oscillations giving rise to such behaviour, e.g., boundary conditions, damping or stiffness. One very common form arises from a change in effective stiffness. As mentioned before, cells with mechanically gated ion channels, naturally have non-linear stiffness in the small deformation range. But such changes can occur as a result of large deformation as well (Fig. 6). Most materials have an elastic limit, and when they are extended beyond this limit, their stiffness can undergo rapid, sometimes reversible changes before the material is detectably deformed or broken (Fig. 6). None of these examples of non-linear behaviour require energy expenditure and indeed no amplification is achieved. Yet the systems can show some of the characteristics associated with active behaviour, e.g., distortion products with two-tone stimulation (Barral and Martin 2012) and changes in frequency response (Windmill et al. 2006).

Non-linear behaviour has indeed been observed in a few insect ears, notably moths (Coro and Kossl 1998; Windmill et al. 2006), katydids (Möckel et al. 2011) and locusts (Kössl and Boyan 1998). Indeed, these non-linearities may even have a physiological basis and they may be abolished or attenuated by death, yet they still do not imply active amplification. In one moth species, using a theoretical model, it has been suggested that the actual physical basis of this non-linearity is a change in effective stiffness (Windmill et al. 2006) rather than an active process. In all other species, particularly where only a limited set of non-linear behaviours have been observed, the physical basis of this non-linearity remains to be established.

Models of active amplification

It is beyond the scope of this review to provide a comprehensive review of all the models of active auditory amplification. Not only are there many models, but also models have been developed to different levels of mathematical and structural description, making a comprehensive review difficult. I will, however, make brief sketches of the broadest classes of models. I define these classes by considering how they differ in the principal roles they assign to the core actors in the active amplification process.

Electromotility model

The first distinction between the models, as mentioned before, is whether the site of motor activity is in the cell body of the mechanosensory cell or in the hair bundle or cilia. It is unlikely that the Drosophila homolog of prestin is important in amplification (see Kavlie et al. 2014), but I will nonetheless outline briefly the basic prestin-based cell body mechanism for the sake of completeness.

The cell bodies of the outer hair cells (OHCs) within the organ of corti in mammals are thought to be important contributors to active amplification via the motor protein prestin (Liberman et al. 2002). Prestin is a unique motor molecule—it does not use ATP, rather the protein is a solute carrier which changes conformation depending on whether its solute is bound or not (Homma and Dallos 2011a, b). Prestin molecules are embedded in the lateral membrane of OHCs (Belyantseva et al. 2000). When sonic forces are transmitted to OHC hair bundles, the motion of the hair bundle causes the OHC to depolarise. This change in cell potential causes the Cl⁻ ions associated with prestin to translocate, causing a larger conformational change which is reversed upon repolarisation (Homma and Dallos 2011a, b). Conformational changes in many prestin molecules together are thought to cause the OHC to cyclically shorten and lengthen, producing a cyclical force (Santos-Sacchi 2003). OHC motion, in turn, is thought to influence inner hair cell (IHC) motion via indirect mechanical coupling through the organ of Corti (Mammano and Ashmore 1993). It is this addition of the OHC forces to the direct forces experienced and transduced by the IHC that is thought to underlie electromotility-based auditory amplification (Ashmore et al. 2010).

Adaptation motors model

Apart from the electromotility model, most other models implicate the hair bundle, or in insects, the cilium. The adaptation motor model is the first of these models that we will consider. This model though initially developed for vertebrate systems (Hudspeth 2008) has since been adapted for insects systems. It is also the best explored model in terms of genetic data (Fig. 7; Nadrowski et al. 2008; Nadrowski and Göpfert 2009b; Effertz et al. 2012; Kavlie and Albert 2014; Liang et al. 2013).

Fig. 7

Schematics of four possible models of active amplification. a Adaptation motor model: in Drosophila, the JO is in the second antennal segment (A2) and the neurons within it rotate the third segment (A3) which carries the flagellum around its axis. In the schematic here, the A3 axis around which rotation takes is depicted by arrows. Only the three components of the JO neurons important to the model are depicted here. Several sets of JO neurons and hence ion channels, motors and cytoskeletal elements are involved in proper function but only one on either side is depicted for clarity. See text for model details (adapted from Nadrowski et al. 2008). b Dynein motor model: in this model the stereocilia are considered mechanosensory and are moved by external forces allowing influx of Ca²⁺ ions. The kinocilium is considered principally motor in function and its motions produce the internal forces that produce amplification. However, in insects, both functions would have to be integrated into the single cilium (adapted from Camalet et al. 2000). c Ion channel model: in this model, there are no motors; the internal forces are produced by conformational changes within the ion channel, i.e., by the ion channel going from the open to the closed state. In the complete model, the authors posit 6 states that the model transitions between (Choe et al. 1998). Here I have only depicted four for clarity (adapted from Hudspeth et al. 2000). d Integrate and twitch model: in this model the geometry of the mosquito JO is simplified from a curved to rectilinear shape. When the flagellum is moved, the mechanosensory ‘threads’ furthest from the pivot are excited the most and those nearest, the least. Each has a threshold level of distension, below which it does not fire and twitch. Thus, for small deflections of the flagellum, only the outer neurons twitch and only when the deflection increases do the central neurons begin to twitch. This differential recruitment forms the basis of the overall non-linear behaviour predicted by the model (adapted from Avitabile et al. 2011)

The insect adaptation motor model (Nadrowski et al. 2008) is based on the structure of the Drosophila SRS and Johnston’s organ (JO). The model assumes that the antennal SRS is coupled to two symmetric sets of mechanosensory neurons, one on either side of the SRS’s base (Fig. 7). Each JO neuron is then modelled as a unit containing a mechanosensitive ion channel connected to a compliant protein molecule, and to a motor that can travel along the ciliary cytoskeleton (Fig. 7; Nadrowski et al. 2008; Effertz et al. 2012; Liang et al. 2013). The function of the gating spring is to be held under tension so as to funnel forces from the SRS to the ion channel. This tension is achieved by the adaptation motors which translate along a cytoskeletal element and stop when they reach equilibrium tension.

However, if the SRS were now to be deflected from this initial position by some steady force, most ion channels on one side of the SRS would be opened and the gating springs on neither side would be at equilibrium tension. If the SRS experienced sound at this new position, the mechanosensory apparatus would no longer be able to respond optimally. Hence, a system that allows for adaptation to such a scenario exists. This system allows the motors to move, stretching and relaxing the gating springs until they reach predeflection tension at this new position. This is achieved by the motors sliding along the cytoskeleton (Fig. 7). Motors on both sides of the SRS move; on one side the tension is increased and on the other it is decreased. Thus, it is predicted that the existence of a gating spring and of motor-based adaptation can be inferred based on SRS deflection in response to force steps. One expects an initial softening of the SRS due to the opening of channels via a gating spring, with an eventual return to steady state stiffness due to the adaptation motors. Indeed, such behaviour has been observed in Drosophila (Albert et al. 2007; Nadrowski et al. 2008).

The process of adaptation itself, however, does not provide active auditory amplification. It has been speculated, however, that the same machinery and its dynamics may also underpin active amplification (Nadrowski et al. 2008). In close parallel with models developed for frog hair bundles (Nadrowski et al. 2004), a model was developed for Drosophila which posits negative feedback between the open probability of the ion channel and the force applied by the motor molecules to the transducer (Nadrowski et al. 2008). In the vertebrate system, this feedback is believed to be mediated by the intracellular Ca²⁺ concentration which is thought to be the mediator in insect systems as well (Kim et al. 2003; Kamikouchi et al. 2009; Effertz et al. 2011).

Imagine that the SRS is being pushed in the forward direction (Fig. 7). This will cause the channels in the front set of JO neurons to open due to high gating spring tension and those in the back to close because of low gating spring tension. The adaptation motors will then attempt to accommodate this change. Because of negative feedback, the motors in JO neurons in front set will attempt to close ion channels, decreasing tension. In the back set, the Ca²⁺ will be low so the motors will now attempt to reset so that channels return to the prestimulus opening probability, i.e., the tension will increase. The motors will be able to accomplish this at a particular speed intrinsic to the system. If the frequency of a periodic stimulus on the SRS, such as a sinusoidal sound wave, matches the motor speeds, the internal and external forces will potentiate each other, effectively giving rise to frequency dependent amplification.

It was found that this simple but realistic model was sufficient to capture many of the non-linear characteristics of the Drosophila auditory system: self-sustained oscillations, stimulus level-dependent sensitivity at best frequency, as also the generic behaviour expected from compound action potentials measured from the Johnston’s organ (Nadrowski et al. 2008).

Dynein motor feedback model

Another model developed for vertebrate systems views the auditory amplification system through the context of critical oscillators (Camalet et al. 2000; Duke and Jülicher 2007). This model shows how a hair bundle, or indeed cilia, may be able to tune themselves to a critical instability thus naturally giving rise to the behaviour described in “Active auditory amplification and transduction”. The model stems from the observation that when a collection of motors is coupled to an elastic element, then the entire structure tends to oscillate on its own at an intrinsic frequency (Ajdari et al. 1997; Camalet et al. 1999) forming the basis of the amplification system.

You will notice that the adaptation motor model also exploits such a preferred oscillatory model. It is the manner in which the motors are used that is different; here the motors in question are the axonemal dynein motors which are fixed to the ciliary microtubules. Unlike the adaptation motor model, here the motors that bring about amplification need not be coupled to the ion channels or indeed change gating tension (Camalet et al. 2000). This model does not implicate adaptation motors in the active amplification process. It incorporates the adaptation characteristics they provide, but they are not responsible for cycle by cycle amplification (Camalet et al. 2000). Hence, according to this model, amplification and transduction are not linked and may be separately mutagenized or manipulated.

The model proposes that the tendency of the kinocilium to oscillate as a result of dynein activity is controlled by a parameter—the Ca²⁺ concentration. At a particular Ca²⁺ concentration, the kinocilium is at a critical bifurcation point; if the system is either at or just above the bifurcation point, it has a tendency to be set into oscillatory limit cycles by small disturbances (Fig. 5: critical or supercritical). The distance from critical point determines both the amplitude and frequency of the kinociliary oscillations. Modelled hair bundles at the critical point show all the non-linear behaviours associated with auditory systems outlined in “Active auditory amplification and transduction” (Camalet et al. 2000). Not only did small forces set the hair bundle into self-sustained oscillation, but if the disruption was also applied at the same frequency as the kinocilium’s intrinsic frequency, then they displayed frequency-specific amplification and compressive non-linearity (Camalet et al. 2000).

However, when stimulated by sound, Ca²⁺ would be allowed in by the ion channels which are modelled as being present on the stereocilia. This increase in Ca²⁺ concentration drives the system away from the critical bifurcation point into the sub-critical region (Fig. 5b, Camalet et al. 2000). To return the system to the critical point, the crucial interaction was negative feedback, i.e., the incoming Ca²⁺ions decreasing the activity of the dynein motor proteins thus reducing the subsequent influx of ions (Camalet et al. 2000). Along with other mechanisms of Ca²⁺ removal, this led to a return to the critical bifurcation point. Thus, negative feedback was posited to be used by the cells to keep the Ca²⁺ concentration in the cell constant, and at a point at which the hair bundle is at a critical instability, i.e., at a Hopf bifurcation point. Thus, the hair bundle is thought to self-tune itself to an oscillatory instability (Camalet et al. 2000).

The frequency of oscillation is determined by the elastic characteristics of the cilium and the model suggested that it could be varied easily by changing the hair bundle geometry (Camalet et al. 2000). This allows the model to account for the wide frequency range observed in vertebrate systems. While the model was developed primarily for a vertebrate hair bundle, in principle, it should be generalizable to a single cilium-based auditory system where the ion channels would be present on the same cilium rather than in accessory stereocilia. However, only after a careful consideration of the geometry of insect cilia, the location and density of Ca²⁺ channels and dynein motors will a complete mathematical generalisation be possible.

Ion channel feedback model

The next model treated in this review also includes the now familiar list of three players: ion channels, motors and negative feedback. However, in this model, the motors are unusual. In this model, ATPase motor molecules are not required, except for adaptation. It is the ion channels that change conformation with Ca²⁺ efflux and it is their behaviour that is modulated by Ca²⁺-based feedback.

When the hair bundle is distorted by a sound wave and ion channels within the bundle open, there is an influx of Ca²⁺ ions into the hair bundle. The model posits that as the level of Ca²⁺ builds up, it causes ion-channel closure. This closure of the ion channels changes the tension on the gating spring, causing the bundle to move against the initial stimulus force direction (Fig. 7; Choe et al. 1998). Channel closure along with the action of ATP-hydrolysing Ca²⁺ pumps, causes the Ca²⁺ level in the hair bundle to drop again. This releases the ion channels from being held in the closed position. If the stimulus persists, the channels will be pushed back again into an open position in the stimulus direction. Thus, in this model, it is only the opening and closing of channels and the associated conformational changes driven by a temporally varying chemical gradient that drives the amplification process (Choe et al. 1998). The energy source here is not primarily ATP and in this aspect the ion channel model is similar to the electromotility model although located in the hair bundle rather than the cell body.

Unlike the other models, here negative feedback occurs directly between the intracellular Ca²⁺ concentration and ion channel itself, and does not involve the adaptation or even dynein motors (Choe et al. 1998). Feedback is modelled as taking place through a modulatory binding site on the ion channel itself, as in Ca²⁺ inactivation of voltage-gated Ca²⁺ ion channels (Choe et al. 1998). Like the motor model, this model undergoes a Hopf bifurcation and shows critical oscillator-like dynamics, hence, replicating the behaviours observed from actively amplified auditory systems (“Active auditory amplification and transduction”).

In this model, it is the temporal dynamics of Ca²⁺ influx and efflux from the hair bundle lumen and the cyclical binding of Ca²⁺ to the ion channels that will determine the timescales and hence intrinsic frequency of oscillation of the hair bundle (Choe et al. 1998). If the sound deflecting the hair bundle is persistent, and matches the intrinsic frequency of the Ca²⁺ binding unbinding, this would lead to amplification (Choe et al. 1998).

Like the dynein motor model, in this model adaptation and amplification are separable processes. However, amplification and transduction are not, since the same ion channel underlies both processes. Like the motor model, this model is at the moment appropriate only for hair bundles and a careful consideration of the Ca²⁺ spatio-temporal dynamics within the insect system will be required to test its generalizability.

Integrate and twitch thread model

The final model I will consider in this review is a geometric model of the mosquito hearing system (Fig. 7; Avitabile et al. 2010). This model does not explicitly consider the individual behaviours of the ion channels, motor molecules or the feedback between them. Instead, it models the whole response of the invertebrate mechanosensory unit to deformation (Avitabile et al. 2010). There are also other differences, for instance, the nature of the force produced by individual JO neurons in this model is impulsive. In previous models, while the individual motors may have produced impulsive forces, the force imparted by the collective behaviour of many motors within a single cell was more complex (Camalet et al. 2000; Nadrowski et al. 2008). The model also posits a positive feedback between deformation of the mechanosensory unit and the force the unit applies back to the SRS (Avitabile et al. 2010).

In this model, the mechanosensory units termed threads temporally integrate the deformation transmitted by the SRS. This integration is leaky and if they reach a threshold within a time window they spike and apply an impulsive force back to the SRS (Avitabile et al. 2010). After the thread has twitched, it has a refractory period during which it cannot respond. This model of threads is very similar to a ‘integrate and fire’ neuron (Avitabile et al. 2010; Champneys et al. 2011). Below a certain sound level threshold, the mechanosensory units are not activated; past the threshold, only some units are activated because of their geometric position (Fig. 7). As the motion of the SRS due to external forcing increases, more threads are recruited, increasing the force produced by the threads. This continues until saturation, i.e., until all the threads have been recruited and the force produced can no longer be increased. This ‘integrate and twitch’ model reproduces several of the behaviours observed in the auditory system it seeks to model (Avitabile et al. 2010; Champneys et al. 2011). The model captures self-sustained oscillations, the hysteretic change in compressive gain observed in the mosquito antenna below intrinsic frequency, as well as the double frequency compound action potential measured from the JO (Avitabile et al. 2010).

It is important to note here that this model considers the action of population of mechanosensory units, i.e., the active non-linear behaviour is emergent at the level of the collective (Avitabile et al. 2010; Champneys et al. 2011). The properties of a single mechanosensory unit are similar to those of a simple spike-generating ‘all or nothing’ neuron and do not have the properties set out in “Active auditory amplification and transduction” (Avitabile et al. 2011). In the other models, the non-linear behaviour is already present at the level of the single hair bundle (Choe et al. 1998; Camalet et al. 2000; Martin et al. 2000; Nadrowski et al. 2004) or the cilium of the insect mechanosensory neuron (Nadrowski et al. 2008).

In vertebrates, many of the signature behaviours have been observed directly from individual hair bundles, (Martin et al. 2000; Martin and Hudspeth 2001; Barral and Martin 2012) and also at the higher cochlear level (Robles and Ruggero 2001). In contrast, we do not know the behaviour of individual mechanosensory neurons in insects, and so far we have only measured the collective forces exerted by these cells on the SRS. If individual cells do indeed show the signatures determined in “Active auditory amplification and transduction”, the integrate and twitch model will have to be abandoned or modified. On the other hand, if the non-linear active behaviour emerges only at the level of the collective, one would have to consider an independent convergent origin for insect auditory amplification.

Debates about models

There are a great number of debates about the underlying models. The bifurcation paradigm has been criticised; it has been suggested that several bifurcation types other than the Hopf may explain the behaviour of auditory systems (Szalai et al. 2013). Additionally, it has been pointed out that systems that show self-sustained oscillations and hence are clearly past the bifurcation point can nonetheless produce at least some of the non-linear behaviours outlined in “Active auditory amplification and transduction” (Göpfert and Robert 2001, 2003; Mhatre and Robert 2013; Szalai et al. 2013). Indeed, what is clear is that whether the active amplifier is one or another kind of non-linear or ‘critical’ oscillator, whether it is poised at a Hopf or another type of bifurcation, there are certainly several models that reproduce the non-linear active behaviour of auditory systems. Only by confronting and refining these models with empirical data will we be able to discriminate between these alternatives.

Molecular and neuronal evidence for the different models

We know of several molecular actors important for the structural integrity, development and function of the mechanosensory units within the auditory chordotonal organ from Drosophila. These have been identified using both random and directed genetic methods (Todi et al. 2004; Kernan 2007; Senthilan et al. 2012). The genetic data in Drosophila are also complemented by calcium imaging based neuronal recordings (Kamikouchi et al. 2009; Effertz et al. 2011) and exquisite mechanistic measurements (Göpfert et al. 2006; Albert et al. 2007; Nadrowski et al. 2008; Senthilan et al. 2012). In comparison, only limited data are available from other species. An attempt has been made to identify the motors in mosquitoes (Warren et al. 2010), some extracellular neuronal recordings exist (Arthur et al. 2010; Warren et al. 2010) and we have so far made only mechanical measurements in tree crickets (Mhatre and Robert 2013). Despite the limitations of this dataset, we have some tantalising hints at the underlying mechanisms in insect active amplification. Here, I will briefly review the evidence in view of discriminating between these models and larger debates outlined in “Models of active amplification”. The main questions relate to the configuration and molecular identities of the different actors.

Mechanosensory ion channels and the gating spring: NompC, Nanchung and Inactive

NompC: One of the genes discovered in a screen for mechanotransduction in Drosophila larvae was for NompC (no mechanoreceptor potential C), so called because adult bristle organ mechanoreceptors had reduced potentials when this gene was mutated (Kernan et al. 1994). NompC was eventually identified as a member of the transient receptor potential (TRP) family of ion channels and is also known as TRPN1. NompC has been shown to be expressed in the distal portion of the cilium in a campaniform mechanoreceptor (Liang et al. 2013), and it has been implicated in the auditory mechanosensory process as well (Göpfert and Robert 2003; Göpfert et al. 2006; Effertz et al. 2011, 2012).

NompC mutants are extremely interesting from an auditory point of view. They show no sign of active auditory amplification (Göpfert and Robert 2003; Göpfert et al. 2006) and yet some residual sound-evoked potential is observed from the compound action potential (CAP) of the Johnston’s organ (JO) (Eberl et al. 2000). The Drosophila JO contains different subsets of neurons that are tuned to different signal characteristics; effectively each is a sensor for different stimuli (Kamikouchi et al. 2006; Yorozu et al. 2009; Kamikouchi et al. 2009). A subset known as AB is thought to be the principal sound sensor while CE neurons respond mainly to slow changing static stimuli such as wind and gravity (Kamikouchi et al. 2006, 2009). A study that supplemented vibrometric and CAP measurements with ratiometric calcium imaging, suggested that the residual potential recorded from NompC mutants may have come from the CE subset (Effertz et al. 2011).

Additional data from changes in antennal stiffness and CAPs from JO neurons indicated that the NompC channel may make up both the channel and a gating spring (Effertz et al. 2012). In the campaniform receptors, a filamentous structure connects the ciliary plasma membrane to axonemal microtubules and this connecting structure is absent in NompC mutants (Liang et al. 2013). Antibodies raised against the N-terminal ankyrin repeat end of the NompC channel colocalised with microtubules and it has been argued that this N-terminal end of NompC may be the gating spring (Liang et al. 2013). These data have been used to suggest that the NompC channel is either the main ion channel or a subunit of such a channel and/or the gating spring in the transduction machinery of the fly auditory amplifier (Effertz et al. 2012; Kavlie and Albert 2014).

One of the main difficulties in investigating the transduction machinery of the Drosophila JO is the inability to make recordings directly from the JO neurons. Much of the previous work was based on ratiometric Ca²⁺ imaging (Kamikouchi et al. 2009; Effertz et al. 2011) or CAPs (Eberl et al. 2000; Effertz et al. 2012). However, a recent paper showed that Drosophila JO neurons very likely formed gap junctions with giant fibre neurons in the brain (Lehnert et al. 2013) and patch clamp recordings from these neurons were found to be a very sensitive technique for interrogating the transducer mechanism (Lehnert et al. 2013). Flies with nompC mutations were found to produce lower level but distinct sub-threshold sound-evoked potentials (Lehnert et al. 2013) as known from before (Eberl et al. 2000). The findings that these potentials were evoked by the AB subset of JO neurons were, however, found to be more problematic (Lehnert et al. 2013). Put together with previous findings that nompC mutant flies show no active amplification, Lehnert et al. (2013) suggest that transduction and amplification are processes with different genetic substrates, thus undermining support for the adaptation motor model. Additionally, the response of nompC mutants to static stimuli was measured to establish their contribution to adaptation while controlling for their lower sensitivity. It was found that the antennae of nompC flies showed some adaptation albeit abnormal adaptation (Lehnert et al. 2013). These data contradict the suggestion that NompC is the main transducer channel and instead supports a role only in secondarily amplifying stimulus-evoked forces (Lehnert et al. 2013) and it also shows that adaptation and amplification may be unlinked as suggested by the dynein motor model.

Nanchung (Nan) and Inactive (Iav): Two further members of TRP family ion channels are expressed in the cilia of Drosophila JO neurons, Nanchung (Nan) (Kim et al. 2003) and Inactive (Iav) (Gong et al. 2004). The two proteins belong to the TRP V family and are thought to form a heteromultimeric ion channel since mutating either gene prevents the expression of both proteins in JO neurons (Gong et al. 2004). Flies with either mutation are also completely deaf (Kim et al. 2003; Gong et al. 2004). Additionally in nan and iav mutants, no signs of even a sub-threshold receptor potential could be measured from giant fibre neurons (Lehnert et al. 2013). Nan and Iav expression in the AB subset of JO neurons was crucial to normal transduction and transduction could not be rescued by expression in the CE subset (Lehnert et al. 2013). Surprisingly, while nan and iav mutants both showed no sign of transduction, previous work has shown that amplification seems to be present, and even enhanced in these mutants (Göpfert et al. 2006).

So, while previous interpretations considered Nan and Iav to be amplifying ion channels which controlled feedback gain (Göpfert et al. 2006; Kavlie et al. 2010), these new data suggest that they actually form the primary ion channels (Lehnert et al. 2013). Interestingly, both Nan and Iav channels are expressed only in the proximal part of the chordotonal neurons below the ciliary dilation (Lee et al. 2010) whereas NompC is only expressed in the distal half, above the ciliary dilation (Liang et al. 2013). In Drosophila chordotonal neurons, axonemal dyneins are only present proximally and thus colocalise with Nan and Iav channels (Yack 2004; Lee et al. 2008), lending further support to the possibility that it is the dynein motor model that is at work in insect active amplification.

Motors and feedback

Myosin VII A/Crinkled: There are several motor molecules that may be involved in active amplification. Myosin VII A or rather its Drosophila homolog, Crinkled, is a motor molecule known to be expressed within the Drosophila chordotonal organ (Todi et al. 2004; Boekhoff-Falk 2005). The expression of the molecule appears to be restricted to the scolopale cells of the JO (Boekhoff-Falk 2005). Mutants of this gene have abnormally structured scolopidia and are thought to be deaf based on electrophysiological data (Todi et al. 2004; Boekhoff-Falk 2005). However, whether deafness is a result of the protein’s role in active amplification or simply in organ development is not clear at this point.

Dynein: Dynein is another motor molecule well known to be associated with ciliary microtubules in Drosophila JO neurons (Eberl et al. 2000). The first data indicating that dynein may be the motor protein responsible for active amplification in insects came from mutants of the Touch insensitive larvae B (TilB) protein (Eberl et al. 2000; Kavlie et al. 2010). The JO neurons of tilB mutant flies are mostly normally structured, showing the same gross mechanical properties as wild-type flies (Kavlie and Albert 2014). However, within the JONs, there are no microtubule-associated dynein arms in the proximal part of the axoneme (Kavlie et al. 2010) and tilB mutants show no sign of active amplification (Göpfert and Robert 2003; Göpfert et al. 2006).

Subsequent work from mosquitoes showed that the frequency of self-sustained oscillations changed at a rate that was consistent with a change in dynein activity (Warren et al. 2010). However, these data are difficult to interpret. Other parameters such as the viscosity of the intra and extracellular medium also change with temperature, changing ciliary beat frequency (Humphries 2013). Thus, teasing apart the influence of the two processes remains difficult.

The most recent data have come from a forward genetic screen that uses a whole organ knockout strategy (Senthilan et al. 2012). Atonal is a transcription factor required for the formation of a JO in the second antennal segment of Drosophila antenna. Knocking out the atonal gene prevents JO formation (Jarman et al. 1994, 1995). This property of atonal mutants was exploited and the transcriptomes within the second antennal segment of flies with and without JOs were compared using microarrays (Senthilan et al. 2012). Mutants in the two transcripts (CG9492 and Dhc93AB) identified in the genetic screen that encoded dyneins showed severe auditory defects, namely their CAPs were eliminated and so was mechanical amplification (Senthilan et al. 2012), further supporting the hypothesis that dyneins are the motors proteins involved in active auditory amplification. It remains to be seen whether these axonemal dyneins which typically have a limited range of motion can support an adaptation motor model as has been speculated (Senthilan et al. 2012) or whether the dynein model captures insect amplification better.

Feedback: As seen from the models in “Models of active amplification”, either negative or positive feedback between JO neuron deformation and motor activity is capable of producing the well-known signatures of active amplification. Yet very little data are available in this domain especially in insect systems. In mosquitoes, where direct model fitting was carried out, the observed mechanics is well predicted by a positive feedback model (Avitabile et al. 2010; Champneys et al. 2011). This model predicts van der Pol-like dynamics for the mosquito and similar dynamics have been observed both in Drosophila (Stoop et al. 2006) and tree crickets (Mhatre and Robert 2013).

There are very little data from genetics, and these contradict the modelling. Calmodulin is a calcium-sensitive modifier of dynein activity and it affects transducer function and specifically adaptation in vertebrates (Walker and Hudspeth 1996). Calmodulin mutants were shown to have high levels of self-sustained activity consistent with negative feedback (Senthilan et al. 2012). Additionally, iav and nan mutants are reported to show higher self-sustained activity, which is also supportive of the idea of negative feedback, if they are the main transducer channels (Göpfert et al. 2006). Unfortunately, we do not yet know whether the adaptation in Drosophila calmodulin mutants is disrupted as it is in vertebrates (Walker and Hudspeth 1996) and so cannot yet choose between the different models using feedback either.

Context and conclusions

A great deal of progress has been made since the first discovery of active amplification in insect ears. We know much more about the mechanical behaviour of fly and mosquito antennae and these measurements have since formed an important basis for inferring the presence of active amplification in the tree cricket tympanal ear. Given the molecular and genetic tools available in Drosophila, a great deal of progress has been made in linking models to genes. Similarly new techniques in electrophysiology, imaging and genetics have expanded and will continue to expand the data it is possible to collect from the Drosophila model system. Perhaps, soon, even from other insect systems. Each of these lines of evidence will prove invaluable in elucidating the true model underlying the fascinating phenomenon of active amplification.

Unfortunately, while the physical and molecular mechanisms behind active amplification continue to be unravelled, the biological context of active amplification remains largely unexamined. Of course, a few studies have directly addressed the context of active amplification and these should be noted. The obvious function of active amplification is in creating sharp frequency selectivity in the auditory organ. For instance, in tree crickets it was found that the passive auditory system was effectively detuned near conspecific frequency and only when active amplification was switched on was the tree cricket ear ‘tuned’ to conspecific male song (Mhatre and Robert 2013). Indeed it was found that active amplification in the tree cricket ear could be switched on and off and it may follow a circadian rhythm matched to normal conspecific activity (Mhatre and Robert 2013). In several Drosophila species, the role of active amplification seems to be to shift the tuning of the antenna to match conspecific song since in the passive state it is higher than the song (Riabinina et al. 2011).

Similarly, in mosquitoes, the passive and active tuning appear well matched and active amplification generates a large increase in gain (Göpfert and Robert 2001). However, this increase in gain seems to be hysteretic and may depend on the whether the antenna was previously stimulated (Jackson and Robert 2006). Male mosquitoes were played sounds that mimicked a female flying past, first approaching then receding from the male. It was found that the gain of the male antenna when the female was receding was higher than when she was approaching, which the authors suggest is similar to a ‘snapping to attention’ moment (Jackson and Robert 2006).

Mosquitoes are also known to modify the frequency of their songs as they approach other mosquitoes (Gibson and Russell 2006; Cator et al. 2009). When males encounter other males, the frequency of their songs diverges and when they encounter females, they converge in frequency (Gibson and Russell 2006; Cator et al. 2009). However, this convergence and divergence does not occur at the level of the fundamental frequency but at a higher harmonic (Gibson and Russell 2006; Cator et al. 2009) and there has been some debate about the exact mechanism of this harmonic convergence (Robert 2009). One of the suggested mechanisms is that the non-linear and possibly active mechanism in mosquitoes produces a difference tone between the two wing-beat frequencies and that the male minimises this tone to produce harmonic convergence (Gibson and Russell 2006; Warren et al. 2009). This observation is certainly unusual and suggests a new and interesting biological context to non-linear auditory mechanics. One has to be careful, however; while such a difference tone is definitely a non-linear phenomenon, it may or may not be related to the active process. For instance, such difference tones were not observed while measuring tree cricket distortion products. Indeed, the classic active mechanism is not predicted to produce such a difference tone but only distortion products (Jülicher et al. 2001; Duke and Jülicher 2007). The exact origin of such a tone in the mechanical and neuronal measurement needs to be established. Perhaps the simpler possibility must be considered, that the frequency response of other species of mosquitoes was underestimated as was that of Aedes aegypti (Cator et al. 2009).

A great deal more remains to be explored. Not least, it will be crucial to establish the phylogeny of active amplification among insects. Orthoptera and diptera are only two lineages in the insect phylogeny (Trautwein et al. 2012). We are now uncertain of the status of the auditory mechanics of all other orthoptera, which have been the favourite insect model systems to investigate acoustic communication for a long time and this is to say nothing of all the other auditory insect families in between! Finding active processes in these two insect families holds out the promise that there is much more that is interesting that we are yet to discover, even in the ears of the humble insects.