1Dipartimento di Fisica, Università di Genova, 16146 Genova, Italy
(present address: SISSA, Trieste, Italy)
²Department of Physiology, University of Bristol, Bristol  BS8 1TD, U.K.
(present address: Department of Physiology, UCL, Gower Street, London, U.K.)
³Physiological Laboratory, Cambridge University, Cambridge  CB2 3EG, U.K.
⁴Istituto di Cibernetica e Biofisica, C.N.R., 16149 Genova, Italy

Journal of Neuroscience 15, 7757-7768 (1995).
Copyright © 1995 Society for Neuroscience

Note: The Figures have not yet been inserted in this web version.

[ Abstract | Introduction | What is sensory transduction? | Molecular mechanisms |
| Photoreceptor | Olfactory receptor | Auditory and vestibular receptors |
| Signal processing aspects | Summary | References ]

Sensory transduction shares common features in widely different sensory modalities. The purpose of this article is to examine the similarities and differences in the underlying mechanisms of transduction in the sensory receptor cells for vision, olfaction and hearing. One of the major differences between the systems relates to the nature of the stimulus. In both the visual and olfactory systems a "quantal" mechanism of detection is possible, because the absorption of a photon or the binding of an odorant molecule provides an energy change significantly greater than the thermal noise in the receptor molecule. In hearing, on the other hand, the energy of a phonon is far lower, and detection occurs by a "classical" mechanism. For vertebrate photoreceptors and olfactory receptor cells, sensory transduction employs a G-protein cascade that is remarkably similar in the two cases, and that is closely homologous to other G-protein signalling cascades. For auditory and vestibular hair cells, transduction operates via a mechanism of direct coupling of the stimulus to ion channels, in a manner reminiscent of the direct gating of post-synaptic ion channels in various synaptic mechanisms. The three classes of sensory receptor cell share similarities in their mechanisms of adaptation, and it appears in each case that cytoplasmic calcium concentration plays a major role in adaptation.

[Key words: sensory transduction, sensory adaptation, photoreceptor, olfactory receptor cell, hair cell, vision, olfaction, hearing]

Many sensory systems can detect extremely faint signals. In vertebrates, the sensory hair cells of the inner ear can detect displacements of atomic dimensions in the tips of the stereocilia (Crawford & Fettiplace, 1985; Hudspeth & Markin, 1994); rod photoreceptors can detect individual photons (Hecht, Shlaer & Pirenne, 1942; Baylor, Lamb & Yau, 1979; Baylor, Nunn & Schnapf, 1984); and recently it has been shown that olfactory receptor cells appear to be able to detect individual molecules of odorant (Menini, Picco & Firestein, 1995). These three different sensory modalities (hearing, vision and olfaction) employ receptor cells with radically different cellular designs specialized for very different physical stimuli. But despite the great differences, it is possible to discern common motifs underlying mechanotransduction, phototransduction and olfactory transduction.

The purpose of this article is to offer a unifying view of transduction in vertebrate sensory receptor cells, with emphasis on photoreceptors, olfactory receptor cells and auditory receptors. We shall show that there are common engineering objectives, in the design of sensory receptors, which can be appreciated in the light of recent discoveries about the cellular basis of sensory transduction. We shall deal with the features of the physical stimuli that are detected, and then outline the two main classes of molecular mechanisms utilized: direct and indirect transduction. Finally, we consider the link between signal processing and sensory transduction and address amplification and extraction of signals as well as adaptation and the significance of noise.

All animals have evolved sensory receptors, which provide the central nervous system with information about events in the physical world. These receptors consume metabolic energy and are designed so that a physical or chemical stimulus will trigger an electrical response in the cell membrane: a process known as sensory transduction.

It is helpful first to consider some basic properties of the physical stimuli encountered by the sensory receptors of vision, olfaction and hearing. In vision, the stimulus (light) is an electromagnetic wave in the frequency range 4-7.5 ´ 10¹⁴ Hz (corresponding to wavelengths between 750 and 400 nm). In olfaction, the stimulus (an odor) is a particular chemical substance carried in the air. In human hearing, the stimulus is a sound pressure wave in the frequency range 40 Hz to 20 kHz (some mammalian species, such as echo-locating bats, detect frequencies above 100 kHz).

The enormous difference in vibrational frequency between visual and acoustic stimuli implies a quantal mechanism for the detection of light, but a "classical" mechanism for the detection of sound (Block, 1992). A photon of wavelength 500 nm (with quantal energy hn = 4 ´ 10^-19 Joule or 57 kcal/mol) has approximately 10 orders of magnitude more energy than an acoustic phonon, i.e. the elementary energy of an acoustic wave, at a frequency of 10 kHz (with quantal energy hn = 2 ´ 10^-19 Joule or 3 ´ 10^-6 cal/mol). Detection of such stimuli is made against the thermal noise kT associated with each degree of freedom of the transduction mechanism. At room temperature (300 K), kT is approximately 4 ´ 10^-21 Joule or 0.57 kcal/mol. This value is much smaller than the quantal energy in vision, but is much larger than the quantal energy in hearing. This implies that the basic limitation in hearing is set by thermal noise rather than by quantization of the stimulus. In the case of olfaction, the binding of a molecule of odorant to a receptor releases a binding energy E, which is related to the dissociation constant K_d of binding by E = -kT ln K_d (Block, 1992). The dissociation constants for binding of odorants to receptor proteins are not known but for reasonable values of K_d between 10^-8 and 10^-3, E is expected to range from 3.9 to 10.5 kcal/mol.

Figure 1 shows receptor cells used in vision, olfaction, and hearing. In vision, an imaging system (the eye) delivers the visual stimulus to a large array of rod (which number about 100 million in the human eye) and cone (which number about 5 million in the human eye) photoreceptors, with each receptor sampling a small region of the visual environment. The rods operate only at low light levels, and provide vision of very high sensitivity but at low spatial and temporal resolution (Fig. 1A). The cones provide vision at much higher resolution, and function over a million-fold range of intensities, from moonlight to sunlight conditions. About 100,000 cones are concentrated in the fovea, to provide very high spatial resolution. Humans possess cones of three spectral sensitivities, which underlie their tri-chromatic colour vision.

In olfaction, odorants sniffed in the vapor phase by air-breathing animals are brought via the olfactory mucus to olfactory receptor cells (Fig. 1B). Odorants are usually small molecules, with a molecular weight of less than 1000 Da. They excite the olfactory receptor cell by binding to specific receptor proteins in the cilial membrane. Recently, a large family of protein molecules with seven-transmembrane segments, closely homologous to the visual receptor protein rhodopsin, has been identified (Buck & Axel, 1991). This family appears to comprise at least several hundred different protein species which are presumed to represent olfactory receptor molecules specific for many different odorants.

In hearing, sound enters the ear canal, is transmitted through the middle ear where it induces vibrations of the basilar membrane within the fluid-filled cochlea. These vibrations are detected by the mechanical displacement of auditory receptors, the hair cells, attached to the basilar membrane (Fig. 1C). In lower vertebrates the vibration pattern does not depend on the sound frequency, and frequency selectivity devolves from hair cell acting as a tuned filter. In mammals there is a division of hair cell function: inner hair cells are the actual transducers, while the outer hair cells act as mechanical motors. The pattern of basilar membrane vibration is actively controlled by the outer hair cells, and this ensures that each auditory frequency displaces only a localized population of inner hair cells. Mammalian inner hair cells are not individually tuned; their frequency selectivity is determined by the concerted action of many outer hair cells.

The input stimulus for auditory receptors can therefore be considered either as the sound pressure wave which enters the ear, or as the mechanical displacement experienced by the hair cells. In practice, displacements of the hair cell bundle can only be measured reliably in isolated hair cells or in hair cell epithelia, but not in the intact animal.

The sensitivity, S, of transduction is defined as the incremental response per unit intensity of stimulus; that is, S = dr/dI, where r is the response and I is the stimulus intensity. In the case of the photoreceptor and the olfactory receptor, S may be expressed in quantal terms (e.g. in pA/photopigment molecule isomerized or pA/odorant molecule bound respectively), provided that it is possible to estimate the number of quantal events induced by the test stimulus. In the auditory system, S might be expressed the response amplitude per unit of sound pressure (in mV/Pa) in an intact preparation, or alternatively as the response per unit deflection of the tips of the stereocilia (in mV/nm) in an isolated cell preparation.

The term dynamic range refers to the range of stimulus intensities over which a given sensory receptor is able to respond. It is convenient to define the dynamic range as the ratio of the stimulus intensities that elicit 95% and 5% of the maximal response; this is typically specified in log₁₀ units. Because of the exisence of adaptation, though, it is important to indicate whether dynamic range is measured with a transient or a steady-state stimulus, since the steady-state dynamic range may be much greater than the transient dynamic range. The operating point refers to the stimulus which elicits a 50% response.

The term adaptation refers to a decrease in receptor sensitivity that occurs in the presence of a maintained stimulus. The exact nature of the adaptational changes varies widely between different types of sensory receptors. Two extremes are represented by the hair cell and the photoreceptor. In turtle hair cells the application of a steady deflection of the hair bundle simply shifts the operating point of the transduction machinery, with little change in sensitivity (Crawford, Evans & Fettiplace, 1989). In mammalian hair cells, on the other hand, relatively little adaptation of the transduction process takes place (Kros, Rusch & Richardson, 1992).

A very different situation occurs in photoreceptors, where the application of a bright steady light to a cone cell may result in a decrease in sensitivity (a desensitization) of more than 10,000-fold (Burkhardt, 1994). In photoreceptors, the desensitization follows Weber’s Law, which specifies that sensitivity declines in inverse proportion to the background intensity:

where S_steady is the sensitivity in the presence of a steady stimulus of intensity I_steady, and where S₀ is the sensitivity in the absence of steady stimulation and I_½ is the steady intensity that half-desensitizes the response. A particular feature of desensitization in photoreceptors is the accelerated response kinetics caused by changes in the underlying biochemical machinery. Hence in the presence of a maintained stimulus, the receptor cell trades off reduced sensitivity for improved time response (i.e. increased bandwidth). Adaptation is also accompanied by a relaxation (or sag) in the cell’s response to a prolonged stimulus.

The "special" senses comprise vision, hearing and vestibular sensation, and olfaction and taste. In these senses, the receptor cells are specialized neurons, which have evolved elaborate structures designed specifically for the transduction of stimuli into neural signals. In photoreceptors, transduction occurs in the outer segment (Fig. 1A) and is initiated by the absorption of a photon by any one of the numerous rhodopsin molecules embedded in the stack of disk membranes. Olfactory transduction (Fig. 1B) takes place in the fine cilia which extend from the cell’s dendrite; these cilia contain a diversity of receptor molecules that bind odorants. In hair cells of the auditory and vestibular systems (Fig. 1C), transduction occurs in the bundle of stereocilia at the apical pole.

All sensory receptor cells share the feature that a stimulus causes modulation of the flow of a receptor current that circulates in a local loop (Fig. 1). Modulation of the current is achieved by the gating of ion channels located in the plasma membrane of the specialized sensory region of the cell. In olfactory receptor cells, the altered receptor current triggers action potentials that are conducted along the cell’s axon. In vertebrate photoreceptors and in hair cells, the altered current gives rise to a graded change in membrane potential, which is transmitted across a synapse only a short distance from the transducing region.

The molecular mechanisms of transduction found in different sensory receptors can be divided into two distinct classes: direct and indirect activation. In photoreceptors and many chemoreceptors, the sensory stimulus modulates the flow of receptor current by an indirect (or second messenger) pathway. This pathway typically involves a G protein cascade of reactions, and leads to modulation of the cytoplasmic concentration of a chemical messenger (for example, a cyclic nucleotide) and in turn to the modulation of ion channel gating. With directly activated transduction, exemplified by mechanoreceptors, the stimulus leads directly to the gating of ion channels, without the intervention of a second messenger.

G protein cascades are characterized by the presence in the transducing membrane of three species of protein molecule: a receptor protein (R), a G protein (G) and an effector protein (E); (Lefkowitz et al., 1986; Ross, 1989; Stryer, 1991) (Figs 1D, E). Activation of the cascade by a stimulus proceeds according to: Stimulus ® R* ® G* ® E*, where the asterisks indicate activated forms. The occurrence of these reactions depends crucially on intermolecular collisions resulting from lateral diffusion of the protein molecules within the membrane (Lamb & Pugh, 1992). A single molecule of R* can repeatedly (or catalytically) activate molecules of G protein to G* (by triggering the exchange of a GTP for a GDP on G). Each activated G* continues diffusing until it contacts a molecule of E, whereupon the two proteins bind to form G*–E, which represents the activated effector, E*.

An important feature of G protein transduction is its potential to provide very high amplification, through the cascading of two stages of gain. The first stage is provided by the activated receptor protein R*, which catalytically activates the G protein. A second stage of gain is provided by the activated effector protein E*, which is typically an enzyme that catalyzes the synthesis (or destruction) of a cytoplasmic messenger molecule.

G protein cascades form a closely homologous set of cellular signalling mechanisms, utilized not only in sensory receptors but also in many hormonal and synaptic systems, where the mechanism was first elucidated (Lefkowitz et al., 1986). The corresponding proteins bear a remarkable similarity to each other, even in systems as apparently diverse as the photoreceptor and the b -adrenergic mechanism. The receptor proteins (R) are members of a family, sharing homologous amino acid sequences and a common molecular structure comprising seven-transmembrane segments (Lefkowitz et al., 1986; Ross, 1989 Buck & Axel, 1991; Stryer, 1991; Reed, 1992). The G proteins, too, are also closely related, each comprising a hetero-trimer in which the -subunit binds GTP to bring about activation. One of the main differences between cascades in various cell types relates to the effector protein, E. In the olfactory receptor cell (as well as in the b -adrenergic system), E is adenylyl cyclase, the enzyme that synthesizes cAMP (Fig. 1E). In the photoreceptor, E is the phosphodiesterase (PDE) that hydrolyzes cGMP (Fig. 1D).

Modulation of the cyclic nucleotide concentration within photoreceptors and olfactory receptor cells modulates the opening of cyclic nucleotide-gated ion channels. The physiological and molecular properties of these channels are broadly similar in photoreceptors and olfactory receptor cells (Yau & Baylor, 1989; Torre & Menini, 1994; Zufall, Firestein & Shepherd, 1994; Menini, 1995). The genes encoding the cyclic nucleotide-gated channels have been cloned, and their amino acid sequences are highly homologous (Kaupp, 1991).

In photoreceptors, the average density of these channels in the outer segment cell membrane is of the order of 600/mm² (Karpen, Loney & Baylor, 1992), so that each rod contains at least 100,000 channels. In olfactory receptor cells, the density of cAMP-gated channels in the ciliary membrane is about 2500/mm² (Kurahashi & Kaneko, 1991). The photoreceptor channels are characterized by a very rapid flickering between open and closed states (Torre, Straforini, Sesti & Lamb, 1992; Sesti, Straforini, Lamb & Torre, 1994). Their mean open time is less than 40 m s, and their single channel conductance in the absence of divalent cations has been estimated as 25 pS (Yau & Baylor, 1989; Taylor & Baylor, 1995) or as greater than 60 pS (Torre et al., 1992; Sesti et al., 1994). However, physiological concentrations of extracellular Ca²⁺ or Mg²⁺ have a very powerful blocking effect, so that under normal conditions the single channel conductance is effectively reduced to a much smaller value. The opening of each ionic channel requires the binding of at least three molecules of cyclic nucleotide. In salamanders, half maximal activation of the channels occurs at a cGMP concentration of about 10 m M in rod photoreceptors, and at a cAMP concentration of about 20 mM in olfactory receptor cells.

These channels are permeable to divalent ions but are only weakly selective for the passage of different monovalent cations. The entry of calcium through the channels plays a critical role in adaptation in photoreceptors, and appears likely also to do so in olfactory receptor cells.

Because of the high amplification of the enzymatic cascades in phototransduction and chemotransduction, and the high density of cyclic nucleotide-gated channels in the membrane, transduction in these sensory neurons involves the synthesis or removal of many thousands of molecules of cyclic nucleotide and the opening or closure of many thousands of ion channels.

"Direct activation" refers to those systems, exemplified by hair cells, where the transduction channel is gated by the stimulus, without intervention of a second messenger. Direct activation allows fast gating of the channel and is suited for high frequency auditory stimuli. It is believed that the activation energy is supplied by a deformation of the channel, but the precise mechanism of coupling of the sensory stimulus (hair cell displacement) to gating the channel is unknown. There is evidence that the coupling arises from linkages between the channel and the cytoskeleton (as in the case of stretch activated channels in muscle spindle fibers), or from a specialized extracellular protein (as in the case of the hair cell; Fig. 1C). However, in the one case where an identified mechanosensory protein has been sequenced and expressed (a major conductance channel in a bacterial membrane) the stimulus may be coupled to the channel through in-plane forces in the membrane itself (Hamill & McBride, 1994). As yet there is no sequence information for a positively identified vertebrate mechanosensory channel.

Mechanosensory channels are cation-selective. Stretch-activated channels have been studied using patch clamp electrodes for which single-channel conductances of between 16–100 pS have been recorded. In hair cells the conductance of a single transducing channel appears to be about 100 pS.

The rod photoreceptor employs an archetypal version of G protein signalling (McNaughton, 1991; Stryer, 1991; Pugh & Lamb, 1993). A photon activates the receptor protein rhodopsin R to R*, by isomerizing the attached retinal chromophore from its bent 11-cis form to the straight all-trans form; R* then activates the cascade in the manner described above. A single R* causes activation of the G protein at a rate of about 1000-5000 G*/s at room temperature (Stryer, 1991; Lamb & Pugh, 1992). The activated effector protein, E* (the cyclic nucleotide phosphodiesterase, PDE*), hydrolyzes cGMP at a rate of 1000-4000 molecules/s. Therefore light triggers a decrease in cGMP concentration, leading to the closure of cGMP-gated channels in the plasma membrane and hence to a reduction in the cell’s circulating current (Fig. 2A). Recently, the molecular steps in the G protein cascade of phototransduction have been analyzed quantitatively, and a simple expression for the predicted time course of the rising phase of the light response has been derived (Lamb & Pugh, 1992; Pugh & Lamb, 1993).

Despite the much lower sensitivity and faster response kinetics of cone photoreceptors, their transduction mechanism is virtually identical to that of rods. The corresponding proteins are very similar to each other, and they display closely equivalent kinetic parameters (Pugh & Lamb, 1993; Hestrin & Korenbrot, 1990). In some respects the cone is comparable to a light-adapted rod, exhibiting a desensitized and accelerated response. There are some major differences, though: In particular, the cone is much noisier than the rod (see below), and its response does not saturate at high intensities of steady illumination, in the way that the rod’s response does (Baylor & Hodgkin, 1974; Burkhardt, 1994). The molecular basis for these differences in response properties between rods and cones is not yet fully understood.

A typical family of rod photoresponses is illustrated in Fig. 2A. These recordings were obtained from a newt rod, using the suction pipette technique, in response to a series of flashes of increasing intensity. Prior to the flashes, a steady "dark current" of 30 pA circulated, with current flowing into the outer segment and out of the inner segment (Fig. 1A). In response to illumination, this circulating current was reduced in a graded manner: Dim flashes elicited small reductions in circulating current, while bright flashes caused its complete and rapid elimination.

The relationship between response amplitude r and light intensity I (Fig. 3D) is most simply described by an exponential saturation (Baylor, Hodgkin & Lamb, 1974; Lamb, McNaughton & Yau, 1981):

where k is a constant at any fixed time. This relation shows that for brief flashes the dynamic range of the photoreceptor (as defined above) is 1.8 log₁₀ units, or a 60-fold range on intensities.

Comparison of results from a variety of species (ranging from mammals to lower vertebrates) shows that the underlying parameters of the transduction cascade appear to be closely similar across species. Inter-species differences in response kinetics and amplification can be accounted for largely on the basis of differences in outer segment dimensions and differences in body temperature (Pugh & Lamb, 1993).

The dark-adapted rod photoreceptor achieves the ultimate in sensitivity, responding reliably to the absorption of individual photons. Fig. 2D illustrates the response of a toad rod to presentation of a series of very dim flashes of nominally identical intensity. Because the mean number of photons absorbed per flash was very small (about 0.5), quantal effects are significant (see above) and a histogram of the distribution of response amplitudes shows clear evidence for quantization (not illustrated; see Baylor, Lamb & Yau, 1979b). The response to absorption of a single photon has an amplitude of just over 1 pA (about 5% of the dark current) and a time-to-peak of about 1 s, in a toad rod at room temperature. This indicates that the absorption of a single photon by any one of the 2 billion rhodopsin molecules in the rod causes the closure of one twentieth of all the channels present in its outer segment – a spectacular feat of amplification. The quantal response has a smooth time-course, indicating that it involves the activity of a large number of channels. Furthermore, the amplitude and shape of the quantal responses is reproducible, varying by less than 25% of the mean.

The light-evoked signals in photoreceptors must be detected in the presence of any noise in the cell, and this noise can be very large. Although there is intrinsically a large amount of "channel noise" in these cells, the great majority of this high frequency noise is filtered out by the cell’s electrical time constant. Hence, the physiologically significant noise arises predominantly from the "slow" processes associated with the transduction process.

In rods, one such source of noise is the spontaneous occurrence of photon-like events even in complete darkness. Baylor, Matthews & Yau (1980) presented evidence that these events arise from thermal isomerization of rhodopsin molecules, i.e., through the occasional isomerization of R to R* by thermal energy. Despite the occurrence of these isomerizations, each rhodopsin molecule is extremely stable, with a mean thermal life-time of about 3000 years (based on a 50 s mean interval between dark isomerizations and a total of 2 billion rhodopsins/rod). These quantal events in darkness appear to be highly significant for the overall visual system, and they probably set the limit to our visual performance at low light levels. A second source of noise is the so-called "continuous" component, which is likely to be caused by fluctuations in the enzymatic cascade of transduction (Baylor, Matthews & Yau, 1980). A third noise source is apparent during the period of "dark adaptation" following exposure of the rod to an intense light which isomerizes (or "bleaches") more than about 0.1% of the photopigment. These fluctuations do not represent thermal isomerization of the pigment, but instead appear to arise from activation of the cascade by the photoproducts of R* (Lamb, 1980, 1981; Leibrock, Reuter & Lamb, 1994).

In cones, the dark noise is much larger than in rods, with a peak-to-peak amplitude of about 20% of the maximal voltage response (Lamb & Simon, 1976); this value applies in the case of cones which are not electrically coupled to their neighbours. On the other hand, the single-photon response in cones is much smaller and faster than that in rods, and corresponds to suppression of only about 0.1% of the circulating current. From these figures it can be calculated that the noise in turtle cones is equivalent to a steady light intensity of about 2000 photoisomerizations/s (Lamb & Simon, 1977). This intensity of equivalent background in a turtle cone is about 100,000 times greater than the equivalent background in a toad rod resulting from thermal activation of rhodopsin.

When a steady light is presented to a rod, a significant fraction of the circulating current is initially suppressed, but after some seconds the circulating current partly recovers, as a consequence of light adaptation (Fig. 3A). The mechanism underlying light adaptation involves a light-induced drop in intracellular calcium concentration (Torre, Matthews & Lamb, 1986; Matthews, Murphy, Fain & Lamb, 1988; Nakatani & Yau, 1988; Matthews, 1991; Kaupp & Koch, 1992). In darkness calcium enters the cell via the cGMP-gated channels, and is extruded via sodium-calcium exchange; the light-induced closure of channels then reduces the calcium influx, and causes a reduction in cytoplasmic calcium concentration (Yau & Nakatani, 1985; Ratto, Payne, Owen & Tsien, 1988; Lagnado, Cervetto & McNaughton, 1992; Gray-Keller & Detwiler, 1994; McCarthy, Younger & Owen, 1994).

Intracellular calcium affects several important steps in phototransduction (Fig. 1D). A decrease in intracellular calcium concentration stimulates the guanylate cyclase (Koch & Stryer, 1988; Koch, 1991) through the mediation of a 23 kDa protein designated GCAP (guanylyl cyclase activating protein, Gorczyca et al, 1994). Changes in calcium concentration also affect the gain of one or more stages of the enzymatic cascade (Torre et al, 1986), and this has been shown to occur at least in part through the action of a separate 23 kDa protein called recoverin or S-modulin, which controls the rate of rhodopsin phosphorylation (Kawamura & Murakami, 1991; Stryer, 1991; Kawamura, 1993a,b). Furthermore lowered calcium concentration has also been reported to mediate a separate effect on rhodopsin gain (Lagnado & Baylor, 1994), and to increase the channel affinity for cGMP (Hsu & Molday, 1993). These multiple pathways, in which calcium regulates several different biochemical steps, provide a complex negative-feedback control system. The overall effect of this intricate system is to provide an automatic gain control for transduction, which gives rise to photoreceptor light adaptation.

Light adaptation results in desensitization of the photoreceptor in accordance with Weber’s law, as described in Equation 1 (Baylor & Hodgkin, 1974). Such adaptation has been shown to occur in a wide variety of rod and cone photoreceptors, from lower vertebrates to mammals (Fain & Matthews, 1990). This automatic gain control extends the rod’s steady state dynamic range at least a ten-fold beyond that measured at the peak (Fig. 3D). In cones, the effects of light adaptation are manifested over a much wider range of intensities, and the dynamic range may be extended by many orders of magnitude (Burkhardt, 1994).

Olfactory transduction is initiated by the binding of odorant molecules to receptor proteins in olfactory receptor cells. These proteins belong to the superfamily of receptors coupled to G-proteins (Buck & Axel, 1991). As described in "G-protein cascades", the activated receptor protein R* triggers a G-protein cascade and ultimately causes the activation of the effector protein adenylyl cyclase that synthesizes cAMP (Lamb & Pugh, 1992; Reed, 1992). The increased cAMP concentration leads to the opening of ion channels in the ciliary membrane, resulting in an odorant-induced current carried by Na⁺ and Ca²⁺ (Nakamura & Gold, 1987; Kurahashi, 1989). Recently, it has been shown that the odorant-induced current is composed not only of the cationic component through the cAMP-gated channels, but also by an inward anionic component carried by Cl^- (Kleene & Gesteland, 1991; Kurahashi & Yau, 1993), presumably activated by the Ca²⁺ influx through the cAMP-gated channels. The role of the Ca²⁺-activated chloride current in olfactory transduction has not yet been established.

Many basic properties of olfactory receptor cells still need to be clarified. A large number of receptor proteins exist (Buck & Axel, 1991), but little is known about their odor specificity. For example, it is not known whether a single cell expresses only one type of receptor protein (Ressler, Sullivan & Buck, 1994; Shepherd, 1994), or even what the membrane density of receptors might be.

A family of current responses, obtained by briefly exposing an olfactory receptor cell to a series of exposures of increasing odorant concentration, is shown in Fig. 2B (Firestein, Picco & Menini, 1993). The odor-induced current has a slow time course and greater concentrations of odorant induced progressively larger currents (Firestein & Werblin, 1989). By comparing the response families of Figs 2A and B it is evident that the odorant-induced current has a time-course broadly resembling that of the light-suppressed current in photoreceptors. This may not be surprising, since the transduction mechanism in olfactory receptor cells and photoreceptors involves a similar enzymatic cascade. The relation between odorant concentration and the amplitude of the induced current at the peak of the response has been described by the equation:

where c is the odorant concentration, C_½ the odorant concentration evoking half of the maximal response and n the Hill coefficient, which indicates the degree of cooperativity in the response-concentration relation (Firestein, Picco & Menini, 1993). The experimental data were always well-fitted by sigmoid curves (n>1), indicating that the amplitude of the response rises more steeply than linearly as a function of odorant concentration, and that therefore the relation between stimulus intensity and response is steeper than in photoreceptors. Accordingly, the dynamic range of olfactory receptor cells is smaller than that of photoreceptors: about 1.2 log units at the response peak (Fig. 3E), compared with 1.8 log units for a rod (Fig. 3D). Different olfactory receptor cells respond to the same odorant with different sensitivity. In salamander the value of K_½ for cineole varies between approximately 1 m M and 100 m M (Firestein, Picco & Menini, 1993).

When an olfactory receptor cell is exposed to high odorant concentrations for 40 s, the induced current exhibits an initial transient rise, but within a few seconds decreases to a much lower plateau level (Firestein, Shepherd & Werblin, 1990; Kurahashi & Shibuya, 1990), characterized by pronounced fluctuations (Fig. 3B); the plateau response and its fluctuations disappear upon removal of the odorant (Menini, Picco & Firestein, 1995).

This partial recovery of response level in olfactory receptor cells appears qualitatively similar to the recovery of the circulating current in photoreceptors during steady illumination. However, it is not yet known whether olfactory receptor cells share the other features of adaptation in photoreceptors; i.e. whether a change in sensitivity and dynamic range occurs in the presence of a steady odorant concentration. The tendency for the response to recover during prolonged stimulation has been shown to be reduced when extracellular calcium is lowered (Kurahashi & Shibuya, 1990). Although this is suggestive of a role for calcium in adaptation of olfactory receptor cells, the underlying mechanisms have not yet been clarified.

Fig. 2E illustrates current responses elicited by a series of identical brief exposures to a low concentration of odorant. These responses exhibit considerable variability in their amplitude: in some of the trials there was no apparent response, in others a small response of about 0.4 pA, and occasionally larger responses of 2–4 pA, and even one response of 9.5 pA. Analysis of the amplitude distribution of these events suggests the existence of a quantal response in the range 0.3–1 pA. It seems likely that such a quantal event would be caused by the binding of a single molecule of odorant to a receptor protein (Menini, Picco & Firestein, 1995).

The structures of the inner ear are specialised to detect acoustic frequency modulations in air pressure (in the cochlea), angular acceleration of the head (in the semicircular canals) or the gravity vector (as in the sacculus and utriculus). Their common sensory element is the hair cell, a polarised neuro-epithelial cell which bears at its apical surface a bundle of about 100 rod-like stereocilia which can be deflected by a mechanical stimulus. The geometry of the structures surrounding the hair cells ensures that the external physical stimulus is delivered to their cell as a displacement stimulus. Incoming sound initiates a travelling wave in the basilar membrane of the mammalian cochlea whose amplitude peaks at a position dependent upon the sound frequency. The stereocilia of hair cells located at the site of peak excursion of the basilar membrane are then deflected as a result of the geometry of the organ of Corti, the hair cell bearing epithelium which rides on the basilar membrane.

The apparatus of the mammalian cochlea thus transforms the input acoustic wave before its detection by the hair cells. As described below, this transformation includes both low-noise amplification and spectral decomposition of the incoming sound. To accomplish this, the mammalian cochlea relies on the concerted action of many hair cells (Dallos, 1992; Ashmore, 1994). The organ of Corti within the cochlea contains two classes of hair cell: inner hair cells, which act as sensory cells for the auditory nerve, and outer hair cells which are now considered to be motor cells involved in a local positive-feedback loop to provide enhancement of the travelling wave in the basilar membrane.

At auditory threshold, the peak displacement of the basilar membrane is about 0.3 nm (Sellick, Patuzzi & Johnstone, 1982). This movement includes the amplification provided by the outer hair cells. Without the operation of the cochlear amplifier, the peak displacement would be about 100-fold less, (i.e. approximately 0.003 nm)

The nature of the cochlear amplification reflects the properties of the outer hair cell (Ashmore & Kolston, 1994). As well as being a displacement sensor, with a mechano-electric transducer associated with the stereocilia, the outer hair cell has a specialised basolateral membrane which produces a longitudinal force when the membrane potential changes. This force generator works at frequencies well into the acoustic range (Ashmore, 1987; Santos Sacchi, 1992; Dallos & Evans, 1995), and the available evidence suggests that it can produce sufficient force to determine the basilar membrane mechanics (Ruggero & Rich, 1991; Mammano & Ashmore, 1993).

In the case of the mammalian cochlea, the frequency selectivity of the system (the ability to select sounds from a complex input signal) is intertwined with amplification. The effect of the outer hair cell force generator is to reduce the viscous damping of the fluid on the basilar membrane. As a result, each longitudinal section of the cochlea acts like a lightly-damped resonator, so that the envelope of the travelling wave becomes both larger in amplitude as well as more localised. Thus the frequency selectivity of the cochlea improves as a consequence of its amplification.

This particular design copes with the high-frequency demands of mammalian hearing. Other designs, adapted to a lower acoustic frequency range, have evolved in other vertebrates, where the hair cells, (instead of being non-frequency-selective), can themselves act as selective electrical filters to a single frequency of stereocilial displacement. The properties of these cells have been well-documented, especially in frogs (Hudspeth & Lewis, 1988) and turtles (Art & Fettiplace, 1987). Because of the presence of voltage- and time-dependent Ca²⁺ and K⁺ conductances in such hair cells, the membrane impedance behaves like an electrical resonator tuned to a specific frequency.

The electrically-tuned hair cells of lower vertebrates cannot be used for signal detection at very high frequencies. Since electrical tuning depends on the interaction of ion channels in the cell basolateral membrane (Ashmore & Attwell, 1985; Art & Fettiplace, 1987), the appropriate time constant for ion channel gating limits th highest frequencies that can be extracted by this mechanism to 2 kHz, some 3-4 octaves below the upper limit found in many mammals.

In mammals, where the tuning depends upon the macroscopic mechanics of the cochlea, the frequency limit is set in principle by the time constant of mechano-electrical transduction and the response of the force generator provided by the outer hair cells. Recent experiments indicate that the response of this force generator extends at least as high as 20 kHz (Dallos & Evans, 1995).

Transduction in hair cells is very fast. The kinetics of transduction have not been determined in detail, but the processes are likely to occur on a time-scale of the order of microseconds (Corey & Hudspeth, 1979; Crawford, Evans & Fettiplace, 1989), and this ensures that hair cells can respond to acoustic stimuli. The upper speed limit of the kinetics is set by the time-scale of protein conformation changes, and their gating time of 1 µs would set an upper bound of frequency detection at about 150 kHz, a limit approached in some bat species. Hair cell transduction is not subject to the frequency limitations of photoreceptors and olfactory receptors, because in hair cells the link between the mechanical displacement stimulus and the channel gating is simple and direct, without any intervening biochemical cascade.

The transducer channels in the hair cell membrane are located near the tips of the stereocilia (Jaramillo & Hudspeth, 1991). From evidence derived mainly from lower vertebrate species, these channels are cation-selective (Holton & Hudspeth, 1986; Roberts, Jacobs & Hudspeth, 1990) and may well be the only ion channel type in the apical membrane although in mammalian species there appears to be purinergically gated ion channels present as well, (Housley, Greenwood & Ashmore, 1993). Electron microscopy reveals that the tips of the stereocilia are associated with only a few linkage proteins ("tip links") which course between adjacent stereocilia (Pickles, Comis & Osborne, 1984). Under the simplest model, the mechanosensitivity of the channels arises from deflection of the stereocilial bundle which changes the tension in the tip link and feeds energy directly into the transducer channel thus leading to its opening (Pickles & Corey, 1992, Hudspeth & Markin, 1994). As a consequence the mechanosensory channels are opened and closed at the frequency of the stereocilial displacements.

Experiments on single hair cells allow measurement of the peak mechano-electrical transducer current i elicited by a stereocilial displacement x so that the dependence of i(x) on x can be determined (Corey & Hudspeth, 1983; Holton & Hudspeth, 1986; Crawford, Evans & Fettiplace, 1989). This relationship has been found to be sigmoidal, (Fig 3F), and can be described by an equation of the form:

where x₀, a and b are constants. In turtle hair cells, a = 3b = 10 m m^-1 when x is measured in microns (Crawford, Evans & Fettiplace, 1989). In the case of mammalian inner hair cells, the point where the stereocilia are at rest, the neutral set point x₀, corresponds to i(x₀)/i_max »   0.15 although for outer hair cells, i(x₀)/i_max »   0.5. In the latter case, the cell is operating near the position which corresponds to both maximum sensitivity and greatest linearity. This neutral position seems to be determined for outer hair cells by the mechanical bias arising from the surrounding structures of the organ of Corti (Dallos & Corey, 1991) whereas for the inner hair cells and other hair cell systems the neutral position is determined in part by intrinsic forces within the stereocilial bundle. There is now substantial evidence for a myosin-based motor, responsive to calcium entering the transducer channel (Assad & Corey, 1992, Solc et al, 1994), which retensions the tip link to ensure the operating point of the curve is appropriate.

Equation 4 is a non-linear relation between i(x) and bundle displacement even for relatively small displacements near auditory threshold. This non-linearity is believed to affect mechanical feedback of the outer hair cells and is likely to generate many of the non-linear properties of the auditory system which include distortion products (Robles, Ruggero & Rich, 1991; Jaramillo, Markin & Hudspeth, 1993) and possibly otoacoustic-emissions, where sound is emitted from the cochlea.

In mammals, the auditory system responds to sounds from auditory threshold (corresponding to a sound wave with a pressure amplitude of 20 m Pa) up to 5 log units above this level. Simultaneously, the system preserves its ability to discriminate frequencies which may differ by only 0.3% At the receptor cell level, the hair cell sensitivity varies by about an order of magnitude between different species . In mammals the maximum sensitivity of the hair cell is about 1 mV/nm displacement of the stereocilia (or approximately 10 pA/nm, in terms of current). This figure is determined by the maximal conductance of the transducer (about 5 nS) and the input electrical properties of the basolateral cell membrane (Kros, Rüsch & Richardson, 1992). The high sensitivity means that large signals can saturate the receptor’s response., as in the other sensory systems, the wide range of sensory coding employs gain-controlling mechanisms which operate not only at the hair cell but also act to limit the sound input delivered accessory structures surrounding the cells.

In lower vertebrates there is a pronounced adaptation of the transducer current when the stereocilial bundle is displaced by a fixed distance (Eatock, Corey & Hudspeth., 1987; Crawford, Evans & Fettiplace, 1989), as shown in Fig. 2C and 3F. This is equivalent to alteration of a and b in the i(x) curve. Where the response to a sustained displacement has been measured (e.g. in frog and turtle hair cells) the effect is a resetting of the neutral set point without any significant reduction of the dynamic range. The mechanism proposed to mediate this effect is again an entry of calcium through mechanosensory channels which retensions the linkage to the ion channels (Assad, & Corey, 1992, Hudspeth & Gillespie, 1994). Although such adaptation effects are prominent in hair cells of lower vertebrates, adaptation in mammalian hair cells, which can be observed in the transducer currents shown in Fig. 2C, is less pronounced. In mammals the synapse between the inner hair cell and the afferent fibre is likely to be responsible for the observed reduction of firing rate in the auditory nerve when prolonged tone stimuli are delivered to the ear.

The existence of substantial electrical noise in hair cells means that recordings usually have to be averaged from multiple repetitions of the stimulus in order that the response can be determined. The sources of noise can be divided into those which arise from (a) the thermal fluctuations associated with the Brownian motion of the stereocilial bundle, (b) spontaneous fluctuations in the opening of the transduction channels at the apical pole of the cell, and (c) spontaneous fluctuations in the opening and closing of ion channels in the basolateral membrane of the receptor. There is relatively little quantitative information about the relative contribution of each source. Of these, mechanism (a) appears to dominate in cells with free-standing stereocilia, causing wideband fluctuations comparable in amplitude with the signal at auditory threshold. Source (b) is measurable (Holton & Hudspeth, 1986). Indeed, the single channel conductance is estimated in this way to be about 100pS (Crawford, Evans & Fettiplace, 1991). The weighting given to noise source (b) relative to noise source (c) will depend on the precise type of hair cell under consideration. Since the membrane time constant of hair cells typically lies between 0.1 and 1 msec, the voltage noise arising from all sources of fluctuation in the cell will be filtered above 2 kHz.

The estimate of the signal-to-noise ratio for cochlear transduction, present at the inner hair cell synapse, depends on detailed assumptions about cochlear mechanics. In particular, it depends on the number of mechanical degrees of freedom (for example, the numbers of hair bundles) involved in cochlear processing. For a given input frequency, and assuming purely passive cochlear mechanics, the signal-to-noise ratio at the inner hair cells has been calculated to lie between 0.02 in the case where the cell bundles are individually free to be stimulated thermally and 11 in the case where the bundles in different cells are coupled together, thus reducing the number of independent degrees of mechanical freedom (Harris, 1967). This calculation has not been carried out for a cochlea with active mechanical amplification.

Thus, in the auditory system, the accessory structures need to be designed to minimize the effects of the intrinsic thermal noise, which may exceed the input signal. Some mechanical coupling between the cells in the cochlea needs to be present to ensure that the acoustic signal is not drowned in thermal noise. But, as with electrical coupling between photoreceptors, which leads to a compromised spatial resolution, lowered noise in the cochlea is obtained at the expense of reduced frequency resolution.

Artificial as well as natural sensors are designed to detect specific kinds of stimuli (such as sound or light) and are specialized for the detection of particular sensory modalities to appreciate how. In the case of both artificial and natural sensors, an optimization of performance has been achieved and it is useful to analyze sensory transduction in terms of signal-processing and the signal-to-noise ratio.

The electrical transduction of input signals to photoreceptors and olfactory receptors is relatively slow, as a result of the trade-off of temporal response for amplification in the G-protein cascade. Thus, in order for these receptor cells to achieve very high gain, it is necessary for them to reduce bandwidth. In addition to allowing a higher gain, the lower bandwidth has the advantage of increasing the signal to noise ratio within the transduction process. Associated with the slow time-course of their responses, photoreceptors and olfactory receptor cells have relatively long membrane time constants, of about 10–50 msec in the rods and cones of lower vertebrates (Bader, MacLeish & Schwartz, 1979; Perry & McNaughton, 1991). In this way it is possible to filter out noise arising from fast channel fluctuations, as well as fast noise within the G-protein cascade. In the presence of a maintained stimulus, adaptation occurs in the photoreceptor (and apparently also in the olfactory receptor cell). The gain of transduction is reduced, while the bandwidth of the response is increased, thus reducing the signal to noise ratio. In dark-adapted monkey rods, the dominant component of noise (the "continuous" component) has a standard deviation of about 0.17 pA over the bandwidth of the flash response, while the amplitude of the single-photon response is about 0.7 pA (Baylor, Nunn & Schnapf, 1984). The ratio of these two parameters can be used to estimate the signal-to-noise ratio of the dark-adapted maximally sensitive monkey rod as about 4. Because of this large signal-to-noise ratio, the single-photon event is reliably detectable at the level of the dark-adapted monkey rod photoreceptor. In the presence of steady illumination however detection by the light-adapted rod of a light flash has to be performed in the presence of a random stream of photon events arising from the background light. Detection of a very weak signal under these circumstances is inevitably a post-receptoral cross-correlation process, with (for example) a retinal ganglion cell collecting and correlating signals and noise from a large number of photoreceptors.

One mechanism which has evolved in the retina to cope with photoreceptor noise is electrical coupling between photoreceptors of the same kind. For stimuli of wide spatial extent, this electrical coupling reduces the absolute noise level without affecting the sensitivity, and therefore increases the signal-to-noise ratio (Lamb & Simon, 1976).

Auditory hair cells are specialized for responding to signals with a much greater frequency range (40-20,000 Hz) than the G-protein-coupled rods and cones (0-50 Hz). In these cells the membrane time constant is of necessity much shorter (as low as 0.1 msec), and it is not therefore possible to filter out high-frequency fluctuations without losing the signal of interest.

Sensory transduction is not a passive process, but requires the modulation of free energy which has been stored in an electrochemical gradient or in chemical bonds. A rod photoreceptor, upon absorption of single photon, produces an electrical signal of about 1 pA with an integration time of 0.7 s in cold-blooded animals, or of 0.2 s in mammalians. For a driving voltage of about -40 mV, these charge movements correspond to an expenditure of free energy of about 10^-14 Joule, indicating that during phototransduction the rod utilizes the free energy of a photon (about 10^-19 Joule, see "The sensory world: quantum versus classical") to modulate an energy change about 5 orders of magnitude larger. The olfactory receptor cell appears to respond to the binding of an odorant molecule with a with a current signal of about 0.5 pA, with a driving voltage and integration time similar to that of a rod photoreceptor. Therefore assuming binding energies of 10^-19 or 10^-20 Joule (see above, and Block, 1992), the energy modulation associated with chemotransduction is similarly of at least 5 orders of magnitude. A comparable calculation for hair cells is more difficult to make, but it is nevertheless clear that in this case too transduction is associated with a large energy modulation.

In photoreceptors, olfactory receptor cells and auditory receptor cells, transduction has evolved to optimize performance in terms of sensitivity, signal-to-noise ratio and the ability to extend the dynamic range. In photoreceptors and olfactory receptor cells, the molecular mechanisms underlying transduction involve closely similar enzymatic cascades, whereas in auditory receptor cells transduction proceeds by a completely different mechanism involving the direct gating of mechanosensory channels, which achieve very rapid transduction. The ion channels in the sensory membrane of these three types of receptor cells are all cation selective; they do not discriminate sharply between different alkali cations and are permeable to divalent ions. The entry of Ca²⁺ through these channels is an essential feature of transduction in all three sensory receptors, and appears to be involved in regulating the adaptational state of transduction. Each of the three sensory receptors consumes metabolic energy at a significant rate, and signal transduction involves the modulation of this energy consumption.

Nature’s solution to the problem of translating various aspects of the physical world into electrical signals that are amenable to further processing by the nervous system has led to protein-based processes that exhibit much more sophistication than the transducing systems encountered in man-made photomultipliers and microphones. Biological transducers are much more than devices in which physical processes lead to the extraction of photoelectrons or the creation of electron/hole pairs. In addition, they are subject to the demanding constraint of taking care of their own integrity and neural connectivity throughout the life of their host. They deserve to be called "smart sensors".

We acknowledge the support of the EC grant Esprit Basic Research Action SSS 6961.