Animals and human beings rely on accurate information about their external environment and internal state for proper behavioral reactions. This vital requirement has led to a large variety of highly sophisticated sensory systems . A common feature, though, is the step-by-step conversion of the incoming signal through multiple sequential transformations. In auditory systems, for example, air-pressure fluctuations induce oscillations of mechanical resonators such as the eardrums, basilar membranes, and hair sensilla [2,3,4,5]. These oscillations cause the opening of mechanosensory ion channels in auditory receptor cells [6,7,8]. The resulting electrical currents change the cells' membrane potentials. This, in turn, activates voltage-dependent ion channels that eventually trigger action potentials, which are passed to higher brain areas for further information processing (Figure 1). Each processing step induces a transformation of the stimulus representation that may include rectification, saturation, and temporal filtering. In the mammalian ear, this processing sequence is extended by nonlinear mechanical amplification and feedback , which influence the individual processing steps. Similar multi-step sequences of biophysical or biochemical transduction processes underlie the proper function of all sensory and many other signaling systems.
We here show that it is possible to extract fine temporal details of individual processes within such signal-processing chains from observing the output activity alone. This progress results from a new method that extends an experimental strategy well known from measuring threshold curves in neurobiology  or applying equivalence criteria in psychophysics : varying stimulus parameters such that the investigated pathway, cell, or system stays at a constant level of output activity. The key to the new method is to compare different stimuli within these measured iso-response sets in such a way that single processing steps can be dissociated. A cascade model is used as a mathematical framework to infer the salient features of the individual processes. This allows us to quantitatively characterize the signal-processing dynamics even under in vivo conditions.
Unlike many classical approaches of systems identification, the method is not based on temporal correlations between the input and output; hence, the time resolution of the method is not limited by the output precision of the system under study. In a spike-based analysis of neural response properties, this allows us to assess the dynamical features of the involved processes with considerably higher resolution than suggested by the spike jitter.
A particularly fine temporal resolution is needed to analyze signal processing in auditory systems that solve complex tasks such as sound localization, echolocation, and acoustic communication [12,13,14,15]. Here, even single receptor cells display extraordinary sub-millisecond precision [14,16,17], with the underlying signal-processing steps involving yet shorter time scales. How these individual steps operate over short times and eventually allow such remarkable precision is largely unknown because of the high vulnerability of the auditory periphery. This calls for methods based on neurophysiological measurements from a remote downstream location such as the auditory nerve, so that the mechanical structures of the ear remain intact.
As a suitable model system to study signal processing in the ear, we chose the auditory periphery of the locust (Locusta migratoria). Its anatomy is well characterized , and the auditory nerve is easily accessible for electrophysiological recordings. The nerve contains the axons of the receptor cells. These can be roughly divided into two groups according to their frequency of maximum sensitivity, which lies near 5 kHz for low-frequency receptor cells and around 15 kHz for high-frequency receptor cells. The mechanical structure of the locust system is simpler than that of mammals, as the receptor cells are directly attached to the tympanic membrane, the animal's eardrum. Also, in contrast to the signal amplification in the vertebrate cochlea, there are no known feedback loops, a circumstance which facilitates the modeling. General features of mechanoreceptors, on the other hand, are surprisingly similar across species and are also shared by hair cells in the mammalian inner ear .