Using sinusoidal AM stimuli, temporally selective neurons in the anuran IC show either low-pass, high-pass, band-pass or band-suppression characteristics (Rose and Capranica, 1983; review: Rose and Gooler, 2007). In this stimulus, each modulation cycle can be considered to be a sound pulse; as AM rate increases, holding stimulus duration constant, pulse duration and rise/fall times decrease, and the number of pulses increases. Selectivity for AM rate, therefore, can be a consequence of sensitivities to any one or a combination of these temporal attributes. For example, cells that respond phasically to sound pulses show high-pass or band-pass selectivity for AM rate; below the best AM rate, the response level decreases with AM rate because pulse number also decreases. Many band-pass neurons, however, fail to respond to slow AM rates, suggesting that this AM selectivity class is mechanistically heterogeneous. A clearer picture of the temporal selectivities of cells emerges from also evaluating responses to stimuli in which pulse rate is varied while holding pulse duration, shape and number constant. Neurons that show selectivity for pulse rate largely comprise two groups. “Long-interval” cells respond well to slow PRs, but weakly, if at all, to fast rates (Alder and Rose, 2000). Neurons of this type appear to be particularly common in the midbrain auditory region (laminar N.) of the aquatic frog Xenopus laevis (Elliott et al., 2011). Other neurons respond best to mid or fast PRs, in a strongly band-pass fashion in some cases, and are known as “interval-counting neurons” (Alder and Rose, 1998; Edwards et al., 2002). Interval-counting neurons respond after a particular number of sound pulses have occurred with optimal timing; stimuli are ineffective if the interval between onsets of successive pulses alternates between values that are substantially shorter or longer that the optimal interval (Edwards et al., 2002, 2007). Thus the responses of these cells reflect the number of consecutive correct intervals, not the number of pulses that have occurred in a particular time window. Remarkably, a single interval that is too long can reset the counting process. As mentioned earlier, some interval-counting neurons show band-suppression characteristics when tested with sinusoidal AM stimuli. These cells are a subset of the interval-counting population and respond to slow AM rates, or tone bursts of sufficient duration, apparently because these stimuli elicit appropriately timed patterns of presynaptic spikes (Edwards and Rose, 2003; Leary et al., 2008). Band-suppression neurons also have been recorded in the IC of mice (Geis and Borst, 2009), but it is unclear whether they show interval-counting properties.

Advances in performing whole-cell patch recording in vivo (Ferster and Jagadeesh, 1992; Rose and Fortune, 1996) paved the way for investigating processes that mediate transformations in the representations of temporal information (i.e., the mechanisms that underlie temporal computations). More generally, this methodology has led to a renaissance in elucidating the integrative mechanisms that underlie response selectivity in central nervous systems. To make these recordings, a patch-type pipette is used to make a gigaohm seal onto a cell; the membrane that is invaginated into the pipette tip is then ruptured to enable recording of the transmembrane potential of the cell.

Whole-cell patch recordings from interval-counting neurons have revealed a complex interplay of inhibition and excitation (Edwards et al., 2007). In the most selective cases, individual sound pulses, presented at slow rates, elicit primarily IPSPs (Figures 1, 2A). As additional pulses are presented at the optimal rate, however, excitation progressively overcomes inhibition (Figures 2A–C) and spikes are elicited if a sufficient number of pulses (5 in the case shown) are presented. Thus, stimuli in which successive pulses have intervals that are alternately shorter or longer than optimal are ineffective, and elicit primarily inhibition. Remarkably, and consistent with the behavioral results mentioned earlier (Schwartz et al., 2010), a single long interval (2–3 times optimal) that is embedded in a series of optimally timed pulses can completely reset the interval-counting process (Figure 2C). The rate-dependent enhancement of excitation, which plays a critical role in the interval selectivity and counting properties of these neurons, appears to be reset by a long interval (arrows, Figure 2C); the first pulse following the gap elicits a small EPSP and large IPSP, shifting the balance once again in favor of the inhibition. The activity-dependent enhancement of excitation in interval-counting neurons stands in marked contrast to that seen in long-interval cells, which we turn to next.

Long-interval neurons respond well to pulses that are presented individually or at slow rates. Fast pulse rate stimuli, however, elicit a phasic onset response, or, in the most selective cases, no spikes at all (Alder and Rose, 2000; Edwards et al., 2008). Whole-cell recordings from long-interval cells have revealed that individual pulses trigger suprathreshold depolarizations (EPSPs) and phasic inhibition (Figure 3), consistent with a model proposed by Grothe (1994). In cells that do not respond to fast PRs, inhibition appears to precede excitation. At fast PRs, the inhibition triggered by a pulse appears to overlap temporally with the longer-latency excitation from the preceding pulse, thereby preventing excitation from triggering spikes. For neurons that show phasic responses to the onset of fast PR stimuli, inhibition appears to follow excitation. In addition, rate-dependent depression of excitation may contribute to the phasic, onset response properties of these cells to fast PR stimuli.

Peripherally, the duration of a tone burst is coded in the duration of activity of auditory-nerve fibers, with some units showing adaptation of firing rates during the stimulus (Megela and Capranica, 1981). However, in the anuran IC there is evidence of a transformation in this representation, such that neurons show short-pass, band-pass or long-pass duration selectivity (Narins and Capranica, 1980; Gooler and Feng, 1992). Similarly, duration selectivity is also present in the IC of bats (Casseday et al., 1994; Fuzessery and Hall, 1999; Macias et al., 2011), and mice (Brand et al., 2000), and appears to be a general feature of vertebrate auditory systems (Aubie et al., 2012). Several models have been proposed for how duration selectivity might be generated (Figure 4A). Stimulus onset might elicit delayed excitation that summates with “off excitation” (Figure 4A, left) over a particular range of stimulus durations. Alternatively, duration selectivity might result from interplay between short-latency, tonic inhibition and delayed, phasic excitation that is either subthreshold or suprathreshold (Figure 4A middle and right panels, respectively); in the first case, post-inhibitory rebound summates with the delayed, subthreshold excitation to trigger spikes. For short-pass or band-pass cells in anurans and bats, responses occur after the end of the stimulus; thus response latency increases with tone burst duration (Faure et al., 2003; Leary et al., 2008). In bats, blockade of inhibition decreases response latency and broadens the range of durations over which the neuron responds, suggesting that short-latency, tonic inhibition plays an important role in generating short-pass and band-pass duration selectivity (Casseday et al., 1994).

Whole-cell recordings from short-pass duration-selective neurons in the anuran IC (Leary et al., 2008) support the hypothesis that tone bursts elicit short-latency inhibition, followed by delayed excitation (Figure 4B). For the cell shown, each stimulus elicited an initial hyperpolarization, followed by a depolarization that, for tone-burst durations of 10–30 ms, was sufficient for reliably eliciting spikes. The short-latency nature of the inhibition was particularly evident in the responses to 300 Hz tone bursts, 20 ms in duration, which elicited little excitation (gray, lower trace, Figure 4B), The duration of hyperpolarization in most cases increases with tone burst duration, suggesting that short-pass neurons receive tonic inhibition. The delayed excitation, however, appears to be phasic (i.e., relatively invariant in time course and magnitude across a wide range of tone burst durations). It is presently unclear whether rebound from inhibition might sum with delayed excitation to augment responses for short- or mid-duration tone bursts. Band-pass duration selectivity also appears to result from interplay between short-latency, tonic inhibition and delayed excitation, however the strength of the excitation appears to increase non-linearly with tone burst duration until the best duration is reached.

Two types of long-pass duration selectivity have been identified (Leary et al., 2008). One class of long-pass duration-selective neurons also responds well to a series of sound pulses, and show interval-counting properties. These neurons show band-suppression characteristics for sinusoidal AM stimuli. It appears that long-duration tone bursts elicit a temporal pattern of spikes in the presynaptic excitatory inputs that closely match that of a series of pulses. The other class of duration long-pass cells responds to long-duration tone bursts, but not to pulses presented at fast rates (i.e., these neurons do not show interval-counting properties). The mechanism underlying this type of long-pass duration selectivity is not well understood, however our working hypothesis is that it may result from the integration of phasic inhibition and tonic excitation. In this model, short-duration tone bursts elicit primarily inhibition. As tone burst duration increases, however, inhibition wanes and tonic excitation promotes depolarization, which, over time, is sufficient for triggering spikes.

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.