Organization and trade-off of spectro-temporal tuning properties of duration-tuned neurons in the mammalian inferior colliculus

Abstract

Neurons throughout the mammalian central auditory pathway respond selectively to stimulus frequency and amplitude, and some are also selective for stimulus duration. First found in the auditory midbrain or inferior colliculus (IC), these duration-tuned neurons (DTNs) provide a potential neural mechanism for encoding temporal features of sound. In this study, we investigated how having an additional neural response filter, one selective to the duration of an auditory stimulus, influences frequency tuning and neural organization by recording single-unit responses and measuring the dorsal-ventral position and spectral-temporal tuning properties of auditory DTNs from the IC of the awake big brown bat (Eptesicus fuscus). Like other IC neurons, DTNs were tonotopically organized and had either V-shaped, U-shaped, or O-shaped frequency tuning curves (excitatory frequency response areas). We hypothesized there would be an interaction between frequency and duration tuning in DTNs, as electrical engineering theory for resonant filters dictates a trade-off in spectral-temporal resolution: sharp tuning in the frequency domain results in poorer resolution in the time domain and vice versa. While the IC is a more complex signal analyzer than an electrical filter, a similar operational trade-off could exist in the responses of DTNs. Our data revealed two patterns of spectro-temporal sensitivity and spatial organization within the IC: DTNs with sharp frequency tuning and broad duration tuning were located in the dorsal IC, whereas cells with wide spectral tuning and narrow temporal tuning were found in the ventral IC.

the detection and analysis of temporal features of sound is a fundamental property of auditory systems (Griffin et al. 1960; Simmons et al. 1979; Moiseff and Konishi 1981; Faure and Hoy 2000). The duration of an acoustic signal contains important species-specific information and neural mechanisms of duration selectivity exist across a variety of vertebrates (Aubie et al. 2009, 2012). For example, human speech contains vocal elements called phonemes that vary in multiple acoustic parameters, including signal duration (Denes 1955). Temporal differences between phonemes can occur on timescales shorter than what is consciously processed (Liberman et al. 1967; Shannon et al. 1995), suggesting there must be a neural mechanism for computing and/or comparing auditory signal durations. In amphibians, signal duration is an important feature of mating calls (Narins and Capranica 1980; Akre et al. 2011; Jones et al. 2013) and frogs possess midbrain neurons with responses selective to the duration of an auditory stimulus (Potter 1965; Narins and Capranica 1980; Gooler and Feng 1992; Leary et al. 2008). Echolocating bats use acoustic information for orientation and prey detection, and they also possess duration-selective neurons (Galazyuk and Feng 1997; Fuzessery and Hall 1999; Faure et al. 2003; Mora and Kössl 2004; Luo et al. 2008). Indeed, neural tuning for sound duration appears to be a general feature of vertebrate hearing (Sayegh et al. 2011; Aubie et al. 2012). In this article, we investigate the relationship between spectral and temporal tuning in a subclass of midbrain auditory neurons selective to stimulus duration.

Duration-tuned neurons (DTNs) first appear at the level of the auditory midbrain (Casseday et al. 1994). The mammalian inferior colliculus (IC) is a complex integrative center that receives parallel synaptic inputs from lower brainstem auditory nuclei (Adams 1979; Covey et al. 1991; Casseday et al. 2002) and descending modulatory inputs from the auditory thalamus and cortex (Coleman and Clerici 1987; Yan and Suga 1996; Zhang and Suga 2000; Ma and Suga 2001). Below the level of the IC, most auditory neurons respond to any sound, regardless of its duration, as long as the signal amplitude is suprathreshold and the frequency falls within the excitatory tuning curve of the cell (Galambos and Davis 1943). In the same way that auditory neurons are tuned to stimulus frequency and have a characteristic frequency (CF), DTNs are tuned to stimulus duration and respond maximally at a characteristic or best duration (BD) and show diminished responses to sounds shorter and/or longer than BD. First discovered from the torus semicircularis of bullfrogs (Potter 1965) and subsequently from the IC of bats (Jen and Schlegel 1982; Casseday et al. 1994), DTNs have now been described from the auditory midbrain of several animals including chinchillas (Chen 1990), mice (Brand et al. 2000; Xia et al. 2000), guinea pigs (Wang et al. 2006), rats (Perez-Gonzalez et al. 2006), and a variety of echolocating bats (Pinheiro et al. 1991; Ehrlich et al. 1997; Fuzessery and Hall 1999; Faure et al. 2003; Mora and Kössl 2004; Luo et al. 2008). Auditory DTNs have also been described from the cortex of bats and cats (Galazyuk and Feng 1997; He et al. 1997; Ma and Suga 2001), and visual DTNs have been found in the cortex of cats (Duysens et al. 1996). The existence of DTNs in different classes of vertebrates and across different sensory modalities indicates that duration tuning is a general feature of sensory systems (Faure et al. 2003).

To date, research on auditory DTNs has focused primarily on the neural mechanisms that create their temporally selective responses (Casseday et al. 1994, 2000; Ehrlich et al. 1997; Fuzessery and Hall 1999; Hooper et al. 2002; Faure et al. 2003; Jen and Wu 2006; Aubie et al. 2009, 2012; Sayegh et al. 2012, 2014) and on the stability (tolerance) of duration tuning with changes in signal amplitude (Zhou and Jen 2001; Mora and Kössl 2004; Fremouw et al. 2005). We know that duration tuning is disrupted or abolished by blocking neural inhibition in the IC, which also suggests that duration tuning originates there (Casseday et al. 1994, 2000; Faure et al. 2003; Leary et al. 2008). Although detailed reports exist on the frequency tuning properties of IC neurons in general (Wu and Jen 1991; Casseday and Covey 1992; Haplea et al. 1994; Zhang and Suga 2000; Wu et al. 2004), much less is known about the frequency tuning characteristics of DTNs in particular. This report aims to fill this gap in our knowledge by describing the topographical organization and spectral-temporal tuning properties of DTNs from the IC. It also tests the hypothesis that DTNs experience an interaction or trade-off in their spectral-temporal resolution.

Knowledge of the interaction between spectral and temporal tuning allows for a more general understanding of how the mammalian central nervous system (CNS) encodes and analyzes acoustic information and will be essential for understanding the functional role(s) of DTNs in normal hearing. Recent work suggests that the temporal selectivity of a neural filter influences its spectral selectivity (Wu and Jen 2008a). For example, some neurons from the nuclei of the lateral lemniscus that are important for analyzing temporal features of sound have broader frequency tuning curves than other types of brainstem auditory neurons (Covey and Casseday 1991), suggesting that neural specializations for temporal processing arise at the expense of spectral processing. In the case of a resonant electrical filter, this trade-off happens because sampling sounds over a long duration adds more cycles of the signal to increase the accuracy of frequency estimation at the expense of temporal precision. Conversely, increasing the accuracy estimate of the start and/or end of a signal, and thus its duration, requires sacrificing frequency resolution at those time points (Blanchard 1941). Because the auditory system is also a signal analyzer, we hypothesized that its neural components may be subject to a time-frequency trade-off not unlike that of a resonant electrical filter. If true, then this hypothesis predicts that DTNs with narrow duration tuning will have broad frequency tuning and cells with wide temporal tuning will have sharp spectral tuning. The purpose of this study was to test this hypothesis by investigating the frequency and duration tuning characteristics of DTNs from the mammalian IC.

MATERIALS AND METHODS

Electrophysiological Recordings

Single unit recordings were conducted at the University of Washington (UW) and McMaster University (MU). Procedures conducted in Seattle were approved by the UW Laboratory Animal Care and Use Committee, while procedures conducted in Hamilton were approved by the MU Animal Research Ethics Board and were in accordance with the Canadian Council on Animal Care. Animals at both institutions were housed in outdoor husbandry facilities where lighting and temperature corresponded to ambient conditions, and food and water were available ad libitum.

Surgical preparation.

Neural recordings were obtained from the IC of 39 awake big brown bats (Eptesicus fuscus) of both sexes (23 from UW, 16 from MU) that were brought into the laboratory to acclimatize 1–3 days before surgery. Bats were anesthetized either by a combination of Metofane (methoxyflurane) inhalation (1–5 min) and a subcutaneous injection of a neuroleptic [0.3 ml 1:1 mixture of 0.025 mg/ml fentanyl citrate + 1.25 mg/ml Inapsine (droperidol) 19.1 mg/kg] or by isoflurane:oxygen inhalation (mixture 1–5%; flow 1–5 l/min). Anesthetized bats were then placed in a foam-lined restraint, molded to the shape of the body to hold the bat firmly yet comfortably while still allowing access to the head, within a stereotaxic alignment system fitted with a custom bite bar/mask for gaseous inhalation (David Kopf Instruments model 1900). The hair covering the skull was shaved, and the underlying skin was swabbed with 70–100% ethanol followed by Betadine disinfectant. Local anesthetic (0.2 ml bupivicaine; 5 mg/ml) was injected subcutaneously before making a midline incision in the scalp. The temporal muscles were reflected, the skull was scraped clean and swabbed with 70–100% ethanol, and a stainless steel post was affixed to the skull to ensure that the position of the bat's head could be precisely replicated between recording sessions. The post was glued to the skull overlying the dorsal surface of the cortex with cyanoacrylate gel adhesive (Zap Gel; Pacer Technology) or superglue (Henkel Lockite) and instantly cured with liquid acrylic hardener (Jet Liquid; Lang Dental). One end of a chlorided silver wire attached to the head post was placed under the temporal musculature and served as the reference electrode.

Stimulus generation.

Sound pulses were digitally synthesized with custom software controlling two signal processing boards from Tucker Davis Technologies (TDT: Apos II sampling rate = 357 kHz) that were optically interfaced to two digital-to-analog (D/A) converters (TDT: DA3–2). The output of each D/A was fed through a low-pass anti-aliasing filter (TDT FT6–2; f_c = 120 kHz) and one (TDT PA5; MU) or two programmable attenuators (TDT PA4; UW) before being mixed in a summer with equal weighting (TDT SM5) and fed through a manual attenuator (Leader LAT-45) before final amplification (Krohn-Hite model 7500). All stimuli were presented monaurally, contralateral to the IC being recorded, with a Brüel and Kjær (B&K) 1/4-in. condenser microphone (type 4939; protective grid on) modified for use as a loudspeaker with a transmitting adaptor (B&K type UA-9020) to correct for nonlinearities in the transfer function (Frederiksen 1977). The diaphragm of the loudspeaker was positioned ∼1 mm in front of the external auditory meatus. The output of the speaker, recorded with a B&K type 4138 1/8-in. condenser microphone (90° incidence; grid off) connected to a measuring amplifier (B&K type 2606) and band-pass filter (Krohn-Hite model 3500), was calibrated (B&K type 4231) and expressed in decibels sound pressure level (dB SPL re 20 μPa) equivalent to the peak amplitude of continuous tones of the same frequency (Stapells et al. 1982). The loudspeaker transfer function was flat ±6 dB from 28 to 118 kHz, and there was at least 30 dB attenuation at the ear opposite the source (Ehrlich et al. 1997). All stimuli had rise/fall times of 0.4 or 0.5 ms shaped with a square cosine function and were presented at a rate of 3 Hz.

Neural recordings.

Electrophysiological recordings began 1–4 days after surgery and took place in a double-walled sound-attenuating booth (Industrial Acoustics). Each bat was used in one to eight sessions lasting 4–8 h each on separate days. Recordings were terminated if the animal showed signs of discomfort. Between sessions, the electrode penetration site was covered with a piece of contact lens and/or Gelfoam covered in topical antibiotic (Neosporin or Polysporin).

Bats were given a subcutaneous injection of a neuroleptic [0.3 ml 1:1 mixture of fentanyl citrate 0.025 mg/ml + Inapsine (droperidol) 1.25 mg/ml; 19.1 mg/kg] before being placed in the foam-lined body restraint suspended within a custom stereotaxic frame (ASI Instruments) mounted atop a floating vibration table (TMC Micro-g). To prevent the bat from moving during recording, the head post was clamped with a micromanipulator (David Kopf Instruments) mounted on the stereotaxic frame. In the first recording session, a scalpel was used to make a small opening in the skull and dura mater overlying the IC for insertion of thin-wall borosilicate glass micropipette electrodes pulled to a tip diameter of ∼1.0 μm and filled with either 0.9% or 3 M NaCl. Electrode impedances ranged from 7 to 53 MΩ. Microelectrodes were visually aimed at the dorsal surface of the IC with a manual micromanipulator (ASI Instruments) and advanced into the brain with a stepping hydraulic micropositioner (David Kopf Instruments model 650). Action potentials were recorded with a Neuroprobe amplifier (A-M Systems model 1600) whose ×10 output was further amplified (×500–1,000) and band-pass filtered (TDT: PC1; high-pass f_c = 700 Hz; low-pass f_c = 3 kHz). Spike times were logged to a computer by passing the PC1 output to a spike discriminator (TDT: SD1) and event timer (TDT: ET1) synchronized to a timing generator (TDT: TG6). Spikes were visualized online as dot raster displays with custom software. All recordings were assumed to be from the central nucleus of the IC.

Data Analysis

We report acoustic thresholds from 149 DTNs and detailed spectral and temporal tuning measurements from 78 DTNs from a collective database. Single units were found by searching with short-duration (1–10 ms) pure tones and/or downward frequency-modulated (FM) sweeps. Upon isolating a unit, the acoustic threshold (dB SPL), CF (kHz), frequency tuning characteristics (V-, U-, or O-shaped frequency tuning), stimulus BD (ms), and duration-selective response class (short-pass or band-pass) were measured. Each cell was stimulated with a range of stimulus frequencies and durations spanning the CF and BD at +10, +20, and +30 dB above threshold, yielding 450 spike count functions [78 cells × 3 SPLs × 2 tuning parameters (frequency, duration) minus 18 cases (4 spectral, 14 temporal) where the cell was lost before data collection at +30 dB had completed]. Spectral tuning was measured with BD tones randomly varied in frequency steps of 250–1,000 Hz and spanning the excitatory spectral bandwidth of the cell. Temporal tuning was measured with CF tones randomly varied in duration steps of 1 ms and spanning the excitatory temporal bandwidth of the cell. Acoustic testing was conducted in blocks consisting of 10–20 stimulus repetitions per frequency or duration step. To minimize the effects of spontaneous activity, data were windowed so that spikes were counted only if they were evoked between stimulus onset and 100 ms after stimulus offset. Data analysis was automated with custom Matlab and Python scripts.

Measuring CF, BD, first spike latency, 50% tuning bandwidth, and Q_{10 dB} values.

The CF of each cell was defined as the stimulus frequency with the lowest acoustic threshold. The BD of each cell was defined as the stimulus duration evoking the highest spike count at +10 dB above threshold using a CF tone. Because stimulus durations were always presented as integer values varied in 1-ms steps, for the purposes of analysis and to remove the discretization of the data, we calculated BD to be the average of the shortest and longest stimulus durations with evoked spike counts ≥90% of the peak count at +10 dB re threshold (Fremouw et al. 2005). First spike latency (FSL) was measured in response to CF and BD tones presented at +10 dB re threshold. Our study only included DTNs with an average response probability of ≥0.5 spikes per stimulus, a criterion that ensured all neurons were healthy and responding consistently.

We measured the 50% excitatory frequency bandwidth (eFBW) and the 50% excitatory duration bandwidth (eDBW) from the frequency and duration tuning spike count functions, respectively, of each cell with a common technique (Fig. 1). First, we noted the stimulus value evoking the peak spike count and then found the lowest and highest stimulus frequencies or durations where the function dropped to ≤50% of the peak thus delimiting the lower and upper cut-offs of the 50% eFBW and 50% eDBW. Cut-offs were computed with two complementary methods (see Sayegh et al. 2012, for use of a similar technique). The inside-out method started at the peak of a tuning function and moved outward, toward the minimum and maximum stimulus values along the abscissa, noting the first data points where the function decreased to ≤50% of the peak. The outside-in method started at the minimum and maximum stimulus values and moved inward, toward the median value along the abscissa, noting the first data points where the function increased to ≥50% of the peak. Once the cut-offs of a tuning function were delineated, stimulus steps straddling the boundaries were interpolated to obtain a more precise estimate.

Fig. 1.

Measuring spectral and temporal excitatory bandwidths. A–D show the means ± SD spikes per stimulus at +10 dB above threshold in response to pure tones that were randomly varied in frequency (left column, stimulus duration = 1 ms) or randomly varied in duration (right column, stimulus frequency = 56 kHz). The spike count data within each column are identical. In A–D, the 50% excitatory bandwidth (BW) is illustrated with a gray box and the cut-off frequencies/durations are the edges of the box. A and B: spike count functions illustrating how the 50% spectral BW was measured with the inside-out (I/O; A) and outside-in (O/I; B) methods. A: with the use of the I/O method, the lower cut-off frequency where the spike count function first crossed the 50% criterion (dotted line) was 42.2 kHz, and the upper cut-off frequency was 50.5 kHz, yielding a spectral BW = 8.3 kHz. B: with the use of the O/I method on the same data, the lower cut-off frequency was 36.8 kHz and the upper cut-off frequency was 50.5 kHz, yielding a spectral BW = 13.7 kHz. C and D: spike count functions illustrating how the 50% temporal BW was measured with the I/O (A) and O/I (B) methods. C: with the use of the I/O method, the lower cut-off duration of the short-pass DTN was arbitrarily set to 1 ms (i.e., the shortest duration tested), and the upper cut-off duration where the spike count function first crossed the 50% criterion (dotted line) was 3.9 ms, yielding a temporal BW = 2.9 ms. D: with the use of the O/I method on the same data, the lower cut-off duration was again arbitrarily set to 1 ms (i.e., the shortest duration tested) and the upper cut-off duration was 5.9 ms, yielding a temporal BW = 4.9 ms. For calculation details, see materials and methods.

COMPARING THE INSIDE-OUT AND OUTSIDE-IN METHODS.

The inside-out (I/O) and outside-in (O/I) methods yielded the same excitatory bandwidths in 388 of 450 spike count functions (230 eFBWs and 220 eDBWs). Overall, the O/I method yielded significantly larger spectral (paired t = 3.871, P << 0.001) and temporal bandwidths (paired t = 5.209, P << 0.001) than the I/O method. The O/I method also yielded more variable bandwidth measures (spectral: O/I SD = 2,904.2 kHz, I/O SD = 2,694.7 kHz; temporal: O/I SD = 5.2 ms, I/O SD = 2.8 ms). Because the I/O method yielded more conservative tuning bandwidth estimates, for the remainder of this article we used it for measuring eFBWs and eDBWs and constructing excitatory response areas.

We also report spectral and temporal quality factors (Covey and Casseday 1991; Haplea et al. 1994; Brand et al. 2000). The Q value is a dimensionless measure of tuning sharpness, with higher values indicating that a filter is more sharply tuned (Capranica 1992). The Q value is calculated by dividing the center (resonant) frequency of a filter by its excitatory bandwidth at a given amplitude level. The Q_{10 dB} of the frequency tuning curve of the neuron was calculated as the CF divided by the eFBW at +10 dB re threshold, and the Q_{10 dB} of its duration tuning curve was calculated as the BD divided by the eDBW at +10 dB re threshold.

Construction and analysis of excitatory response areas.

The excitatory response area (eRA) of a cell provides a visual depiction of how its excitatory bandwidth changes with increasing SPL and can be used to categorize the shape of a tuning curve. We calculated an eRA for each DTN by measuring its excitatory bandwidth (see above) at +10, +20, and +30 dB re threshold (Kiang 1965; Ehret and Schreiner 2005). There is debate on how to define the boundaries of an eRA and a range of definitions have been employed, including the of use hand scored boundaries (Jen and Kamada 1982; Covey and Casseday 1991; Casseday and Covey 1992). Another approach defines eRA boundaries as the range of stimulus frequencies evoking spikes at the estimated spontaneous rate plus 20% of the peak rate (Sutter 2000). Building on this quantitative approach, we measured spectral and temporal excitatory bandwidths in our population of DTNs in 5% response intervals, yielding four bandwidth measures (i.e., from 35–50% of the peak response) at three SPLs (i.e., at +10, +20, and +30 dB re threshold). To be included in the analysis, a cell was required to have an average response probability of ≥0.5 spikes per stimulus at each suprathreshold level. This process yielded 74 excitatory frequency response areas (eFRAs) and 64 excitatory duration response areas (eDRAs) from 78 DTNs.

CATEGORIZING FREQUENCY TUNING CLASSES.

There is debate on the existence of discrete frequency tuning classes and on how to quantify tuning curve shapes (e.g., Javel 1994; Ramachandran et al. 1999; Casseday and Covey 1992; Sutter 2000; Palmer et al. 2013); however, there appears to be consensus that frequency tuning curves of central auditory neurons can be shaped by neural inhibition (Javel 1994; Palmer et al. 2013). Our study sought to investigate the relation between spectral and temporal tuning within the same cell; hence, we sacrificed collecting high-resolution frequency tuning curve data at high SPLs, where some of the more interesting and complicated frequency tuning curve shapes begin to emerge (e.g., slanted tuning curves; Sutter 2000; Palmer et al. 2013) to collect medium resolution duration tuning curve data at the same SPLs within the typical recording time available for a cell. We used a scheme similar to Ramachandran et al. (1999) and Casseday and Covey (1992) to categorize frequency tuning curve shapes. Cells with spectral bandwidths that broadened with increasing SPL were categorized as having V-shaped frequency tuning curves, cells with spectral bandwidths that broadened slightly but then remained constant with increasing SPL were categorized as having U-shaped frequency tuning curves, and cells with spectral bandwidths that first broadened but then narrowed and closed with increasing SPL were categorized as having O-shaped (closed) frequency tuning curves. This classification scheme does not conflict with recent work by Palmer et al. (2013) who demonstrated that the eFRAs of IC neurons are sculpted by the integration of afferent inputs (i.e., neural excitation and inhibition), resulting in continuous rather than discrete frequency tuning curve classes. The V-, U-, and O-shaped frequency tuning classes employed in our study seem equivalent to the V (including VN), N, and C classes used by Palmer et al. (2013).

The overall shape of the eFRA of a neuron was classified by how it changed with frequency and SPL. A cell was scored as having a V-shaped eFRA if its 50% tuning bandwidth increased by >1 kHz at each amplitude level, U-shaped if its bandwidth did not increase or decrease by ≥1 kHz at each amplitude level, and O-shaped if its bandwidth first increased but then decreased so that the cell had an upper threshold beyond which spikes (≥0.5 spikes per stimulus) were no longer evoked (i.e., the cell had a closed tuning curve). The final shape of a frequency tuning curve of a cell was based on the consistency of its eFRA scores from 35 to 50% of the peak response. This automated process yielded 62 robust eFRAs for 74 cells; the shapes of the frequency tuning curves for the remaining 12 cells were independently hand scored by two investigators.

CATEGORIZING DURATION TUNING CLASSES.

The shape of the eDRA of the cell was classified by how it changed with duration and SPL (Fig. 2). Short-pass DTNs fire maximally at BD and have spiking responses that eventually drop to ≤50% of the peak response for sounds longer but not shorter than BD. Band-pass DTNs also fire maximally at BD but have spiking responses that eventually drop to ≤50% of the peak response for sounds both shorter and longer than BD. This process yielded 51 robust eDRAs for 64 cells; the shapes of the duration tuning curves for the remaining 13 cells were independently hand scored by two investigators. Short-pass and band-pass cells were the only DTNs included because they are unambiguous examples of duration tuning (see Sayegh et al. 2011, for a review).

Fig. 2.

Short-pass and band-pass duration-tuned neurons (DTNs) from the inferior colliculus (IC) of the bat. A and B: short-pass DTN. C and D: band-pass DTN. A and C: poststimulus rastergrams illustrating the timing of action potentials in response to characteristic frequency (CF) tone pulses randomly varied in duration and presented at +10 dB above threshold. Stimulus duration illustrated with black bars. B and D: duration tuning curves showing the means ± SD spikes per stimulus in response to CF tone pulses that were randomly varied in duration and presented at three suprathreshold levels. Note how the duration tuning curves remain tolerant (stable) despite large changes in SPL. The CF and acoustic threshold for each cell are given in the legends. The calculated best durations (BDs) for UW223.02.07 and UW235.01.07 were 2.01 ms and 3.61 ms, respectively, and were derived from the mid-point duration value bracketing 90% of the peak spike count. The interpolated 50% temporal bandwidths measured at +10 dB (re threshold) were 1.97 ms and 3.39 ms, respectively. For calculation details, see materials and methods.

Statistical analysis.

All data are reported as the means ± SD and were measured from responses recorded at +10 dB re threshold. Cells were grouped by response class (i.e., V-, U-, and O-shaped frequency tuning; short-pass and band-pass duration tuning), except when testing for differences within tuning response classes. All statistical tests were calculated with IBM SPSS Statistics version 20. Independent samples t-tests were used to compare duration tuning response classes, unless Levene's test for equality of variances was found to be significant in which case Welch's t-test was used. Paired t-tests were used to compare excitatory bandwidths computed with the I/O and O/I methods. ANOVA tests were used to compare between eRA classes.

RESULTS

Topographical Organization of DTN Response Properties

Our first goal was to examine the spatial organization of DTNs within the IC. Tonotopy is well established in mammalian hearing and from the central auditory system of bats (Haplea et al. 1994; Covey and Casseday 1991; Covey et al. 1991; Grothe et al. 2001). As with other types of central auditory neurons, DTNs from the IC of E. fuscus showed a clear tonotopic organization evidenced by a strong positive correlation between recording electrode depth and CF (Fig. 3A; R² = 0.801, P << 0.001). Careful inspection of the data shows a compressive nonlinearity in the correlation function with an expanded representation of CFs between 25 and 35 kHz. This overrepresentation may not be surprising given the dominance of this frequency band in the search and approach echolocation calls of E. fuscus (Casseday and Covey 1992).

Fig. 3.

Topographical organization of DTN electrophysiological response properties. A: there was a strong positive correlation between recording electrode depth and neural CF demonstrating that DTNs were tonotopically organized within the IC. B: there was also a positive correlation between electrode depth and acoustic threshold. C: there was no correlation between electrode depth and first spike latency. There was a weak negative correlation between electrode depth and the calculated BD (D) and a similar weak negative correlation between CF and BD across the population of cells tested (E). In A–E, n = 78.

Acoustic thresholds of DTNs were also topographically organized, although the relation was not as strong as for tonotopy. There was a positive correlation between acoustic threshold and CF; DTNs with the lowest thresholds were found in the dorsal IC and were tuned to lower CFs than DTNs with higher thresholds located in the ventral IC and tuned to higher CFs (Fig. 3B; R² = 0.3773, P << 0.001).

A spatial organization of DTN response properties was less evident in the temporal domain. There was no correlation between recording electrode depth and neuronal FSL (Fig. 3C; R² << 0.001, P = 0.982), demonstrating that DTNs with similar response latencies were found across a wide range of spatial locations and CFs within the IC. There was a weak negative correlation between electrode depth and stimulus BD (Fig. 3D; R² = 0.0906, P = 0.008) and a similarly weak negative correlation between CF and BD (Fig. 3E; R² = 0.123, P = 0.002), indicating that cells tuned to longer durations were more likely to be found at shallower depths and were tuned to lower CFs.

A behavioral audiogram for E. fuscus measured by Koay et al. (1997) shows that hearing thresholds in the big brown bat were lowest in a narrow band of frequencies from 15 to 35 kHz, with a second sensitivity peak between 55 and 70 kHz (Fig. 4). Plotting the acoustic thresholds of our sample of DTNs on the audiogram reveals that cells with the lowest thresholds mirrored the bat's best behavioral sensitivity in the frequency range from 15 to 55 kHz. Interestingly, the majority of DTNs had CFs between 25 and 60 kHz, a spectral bandwidth important for target ranging and corresponding to the dominant frequencies in the fundamental FM element of the bat's search and approach phase calls (Surlykke 1992; Surlykke and Moss 2000). Few DTNs in our database had CFs between 50 and 120 kHz, a frequency range corresponding to the second harmonic of the bat's FM echolocation calls. Although we cannot exclude the possibility that we failed to record from DTNs tuned to higher CFs, our sample size suggests there are fewer DTNs tuned to the bat's second harmonic.

Fig. 4.

Comparison of neural and behavioural threshold tuning. Acoustic spiking thresholds in a large population of DTNs plotted as a function of CF and compared with the mean behavioral audiogram of E. fuscus (solid line) measured by Koay et al. (1997). Note the marked drop in hearing sensitivity ∼45 kHz in the audiogram. In the frequency range from 15 to 55 kHz, the most sensitive DTNs mirrored the bat's best behavioural sensitivity. The range of neural thresholds across the population of DTNs covers most of the dynamic range of hearing.

Frequency Tuning Classes of DTNs

The eFRA shapes for DTNs from the IC of E. fuscus correspond to a subset of frequency tuning classes reported for other types of central auditory neurons (e.g., Wu and Jen 1991; Casseday and Covey 1992; Sutter 2000; Palmer et al. 2013). The frequency tuning curves of DTNs were classified into one of three classes (Fig. 5). The most common eFRA (n = 37 cells; 50.0%) was from DTNs with V-shaped frequency tuning curves similar to primary afferents of the periphery auditory system (Kiang 1965) and from some neurons located in lower brainstem auditory nuclei (Rhode and Smith 1985; Covey and Casseday 1991; Covey et al. 1991; Haplea et al. 1994; Ramachandran et al. 1999). Figure 5A shows an example of a band-pass DTN with a V-shaped frequency tuning curve (left column). Its spectral Q_{10 dB} value was 8.483; hence, the cell was not sharply tuned in frequency. Also shown are the cell spike count functions in response to 4-ms BD tones randomly varied in frequency (middle column) and to 41-kHz CF tones randomly varied in duration (right column) and presented at +10, +20, and +30 dB above threshold. Unlike the eFRAs, note the consistency of the eDRA functions over a 20-dB range of SPLs. The temporal Q_{10 dB} value of the cell was 0.574; hence, it was also not sharply duration tuned.

Fig. 5.

Frequency and duration tuning in the IC of the bat. Shown are 3 DTNs with different frequency tuning curves, CFs, and duration tuning curves. Threshold frequency tuning curves (left column), eFRA functions showing the means ± SD spikes per stimulus in response to BD tones randomly varied in frequency (middle column), and duration tuning curves showing the means ± SD spikes per stimulus in response to CF tones randomly varied in duration and presented at three suprathreshold levels (right column). A: band-pass DTN with a broadband V-shaped frequency tuning curve. B: short-pass DTN with a narrowband U-shaped frequency tuning curve. C: short-pass DTN with an intermediate bandwidth O-shaped (closed) frequency tuning curve. To highlight differences in spectral and temporal tuning, the frequency (left column) and duration tuning curves (right column) of the 3 neurons were plotted on a common abscissa. Unlike frequency tuning, duration tuning was highly tolerant over a 20 dB change in SPL. A–C legends show the CF, acoustic threshold (Thresh), stimulus duration (Dur), calculated BD, and spectral and temporal Q_{10 dB} values of each cell.

The second most common eFRA (n = 30 cells; 40.5%) was from DTNs with U-shaped frequency tuning curves and nearly constant spectral bandwidths. Figure 5B (left column) shows an example of a short-pass DTN with level-tolerant frequency tuning. This cell responded with multiple spikes per stimulus but only over a narrow range of frequencies between 27 and 29 kHz (middle column). The sharp frequency tuning of the cell was confirmed by its large spectral Q_{10 dB} value of 55.56. In contrast, the cell was less selective in the time domain (right column) as evidenced by a temporal Q_{10 dB} value of 0.518. The eFRAs and eDRAs of this cell were tolerant over a 20 dB change in SPL.

The third and least common type of eFRA (n = 7 cells; 9.46%) was from DTNs with O-shaped frequency tuning curves. These nonmonotonic cells were interesting because they had both lower and upper acoustic thresholds; hence, the neuron eventually ceased responding as the tone amplitude was increased (Casseday and Covey 1992; Ehret and Schreiner 2005). Figure 5C shows an example of a short-pass DTN with a closed frequency tuning curve (left column), along with its spectral (middle column) and temporal (right column) spike count functions at +10, +20, and +30 dB re threshold. The spectral and temporal Q_{10 dB} values were 14.74 and 0.719, respectively; hence, this DTN displayed intermediate tuning sharpness in the frequency domain but was more narrowly tuned in the time domain.

Topographical Organization of Spectral and Temporal Tuning Classes

The spatial organization of spectral and temporal response properties in the IC may provide clues into the role(s) that DTNs serve in normal hearing, including the processing of species-specific vocalizations (Wu and Jen 2008b; Sayegh et al. 2011). Expanding on this notion, we expected that topographical trends in the IC may facilitate the processing and perception of echolocation calls by the bat. Search phase calls of E. fuscus consist of shallow FM chirps sweeping down in frequency from ca. 30 to 20 kHz over a duration of ca. 10–20 ms (Griffin et al. 1960; Simmons 1987). Because neurons with low CFs are located in the dorsal IC, we expected to find DTNs with U-shaped eFRAs (i.e., sharp frequency tuning) tuned to long BDs (i.e., broad duration tuning) at shallow electrode depths. As a hunting bat approaches an insect, it increases its maximum frequency and signal bandwidth while decreasing call duration. Therefore, in the ventral IC where cells are tuned to high CFs, we expected to find DTNs with V-shaped eFRAs (i.e., broad spectral tuning) tuned to short BDs (i.e., narrow temporal tuning).

To test these predictions, we grouped cells by their spectral and temporal response class and used factorial ANOVA to test for an effect of recording electrode depth, CF, acoustic threshold, and/or FSL. In the spectral domain, DTNs with U-shaped eFRAs were found in more dorsal areas of the IC compared with DTNs with V-shaped and O-shaped eFRAs (Fig. 6; F = 4.075, P = 0.021), but there was no effect of any other neural parameter (CF, F = 2.284, P = 0.109; BD, F = 0.294, P = 0.746; acoustic threshold, F = 1.352, P = 0.265; FSL, F = 0.654, P = 0.523). In the temporal domain, there was no difference in recording electrode depth between short-pass (n = 44; Fig. 6, A–C) and band-pass DTNs (n = 30; Fig. 6, D–F); however, there was a strong tendency for short-pass DTNs to be located deeper in the IC (t = 1.940, P = 0.056). Consistent with this trend, short-pass DTNs were tuned to higher CFs than band-pass DTNs (Welch's t = 2.408, P = 0.019). Not surprisingly, short-pass DTNs also had shorter BDs than band-pass DTNs (Welch's t = 4.486, P << 0.001), but there was no significant difference in acoustic threshold (t = 0.200, P = 0.842) or FSL (Welch's t = 1.195, P = 0.238).

Fig. 6.

Topographical organization of DTN spectral and temporal tuning classes. Histograms illustrating the distribution of recording electrode depths for the population of short-pass (A–C) and band-pass DTNs (D–F). Histograms illustrating the distribution of recording electrode depths for DTNs with V-shaped (A and D), U-shaped (B and E), and O-shaped (C and F) (closed) frequency tuning curves. Cells with U-shaped tuning curves were located in more dorsal areas of the IC compared with DTNs with V-shaped and O-shaped tuning. There was no significant difference in electrode depth between short-pass and band-pass DTNs, but there was a trend for short-pass DTNs to be found more ventrally in the IC. Cells with the narrowest spectral tuning (i.e., U-shaped frequency tuning curves) and broadest temporal tuning (i.e., band-pass DTNs) were located in more dorsal regions of the IC. The distribution of spectral and temporal response classes echoes the organizational trends of spectral and temporal bandwidths and tuning sharpness (see Figs. 7–9).

A plot of frequency and duration 50% bandwidths and tuning sharpness as a function of recording electrode depth suggests that DTNs experience a spectral-temporal trade-off in response sensitivity. In the frequency domain, there was a moderate-to-strong positive correlation between recording electrode depth and the 50% frequency tuning bandwidth (Fig. 7A; R² = 0.373, P << 0.001); DTNs with narrow spectral tuning were found in more dorsal areas of the IC compared with cells with broadband frequency tuning. The significance of this effect held, although the relationship was weaker, when tuning sharpness was examined with spectral Q_{10 dB} values that characterize the spectral bandwidth of a filter relative to its CF (Fig. 7B; R² = 0.055, P = 0.040). Not surprisingly, DTNs with U-shaped frequency tuning curves had narrower 50% spectral bandwidths (Fig. 7A; F = 6.238, P = 0.003) and larger spectral Q_{10 dB} values (Fig. 7C; F = 6.693, P = 0.002) than DTNs with V-shaped and O-shaped frequency tuning curves.

Fig. 7.

Topographical organization of DTN spectral and temporal tuning bandwidths and quality factors. A: relation between recording electrode depth and the 50% bandwidth of frequency tuning; DTNs located in the ventral IC were more broadly tuned in frequency than cells in the dorsal IC. B: there was a significant negative correlation between electrode depth and spectral Q_{10 dB} illustrating that DTNs in the dorsal IC had sharper frequency tuning curves. C: in contrast, there was a significant negative correlation between electrode depth and the 50% bandwidth of duration tuning; DTNs located in the dorsal IC had broader temporal tuning than cells in the ventral IC. D: when this relationship was reexamined by calculating the temporal Q_{10 dB}, no correlation was observed. Legends in A and C apply to B and D, respectively; n = 74 in A–D.

In the time domain, there was a moderate-to-weak negative correlation between recording electrode depth and the 50% duration tuning bandwidth (Fig. 7C; R² = 0.101, P = 0.005); short-pass DTNs were more narrowly duration tuned than band-pass DTNs (Fig. 7C; Welch's t = 4.432, P << 0.001). The relationship was no longer significant when the sharpness of duration tuning was examined with temporal Q_{10 dB} values that characterize the temporal bandwidth of a filter relative to its BD (Fig. 7D; R² = 0.003, P = 0.610). Moreover, the temporal Q_{10 dB} value of short-pass and band-pass DTNs were not significantly different (t = 1.285, P = 0.203).

Relation of CF and BD to Spectral and Temporal Tuning Bandwidths

Our second objective was to examine how the CF and BD of DTNs relates to their spectral and temporal bandwidths and tuning sharpness. In the frequency domain, there was a strong positive correlation between neuronal CF and the 50% frequency tuning bandwidth (Fig. 8A; R² = 0.400, P << 0.001); DTNs tuned to lower CFs were also more narrowly tuned in frequency. When tuning sharpness was examined with spectral Q_{10 dB} values, there was a weak but non-significant trend (Fig. 8B; R² = 0.041, P = 0.074). Interestingly, DTNs with the sharpest frequency selectivity were located at shallower electrode depths within the IC and had CFs matching the dominant frequency range of the fundamental FM acoustic element of the echolocation calls of E. fuscus. In the time domain, there was a strong positive correlation between the calculated BD and the 50% duration tuning bandwidth (Fig. 8C; R² = 0.518, P << 0.001); DTNs with long BDs had absolutely wider temporal bandwidths. When this relation was re-examined with temporal Q_{10 dB} values, no correlation was observed (Fig. 8D; R² = 0.002, P = 0.715).

Fig. 8.

Relation of CF and BD to spectral and temporal tuning bandwidths and quality factors. There was a strong positive correlation between CF and the 50% bandwidth of frequency tuning (A) and a weak negative trend between CF and spectral Q_{10 dB} (B). C: there was a strong positive correlation between the calculated BD and the 50% bandwidth of duration tuning. D: when this relationship was reexamined with temporal Q_{10 dB} values, no correlation was observed. In A–D, n = 78.

Relation Between Spectral and Temporal Tuning

Our final objective was to examine the interaction between frequency and duration tuning to see if DTNs experience a spectro-temporal trade-off in resolution analogous to that of resonant electrical filters. A plot of the 50% frequency tuning bandwidth as a function of the 50% duration tuning bandwidth revealed a weak but significant negative correlation (Fig. 9A; R² = 0.073, P = 0.017). Across the population of cells tested, DTNs with the narrowest temporal bandwidths were among those neurons with the widest spectral bandwidths and vice versa. When this relationship was reexamined with Q_{10 dB} values, no correlation was observed (Fig. 9B; R² = 0.001, P = 0.424).

Fig. 9.

Trade-off between spectral and temporal tuning in DTNs. The data within A and B are identical but the columns use different markers to highlight different spectral (left column) or temporal (right column) tuning classes. A: there was a negative correlation between the 50% bandwidth of duration tuning and the 50% bandwidth of frequency tuning, suggesting that DTNs experience a trade-off in spectral-temporal resolution. A, left: cells with U-shaped frequency tuning curves had smaller spectral bandwidths than DTNs with V-shaped and O-shaped tuning curves, but there was no difference in temporal bandwidth between the three frequency tuning classes. A, right: short-pass DTNs had narrower 50% temporal bandwidths than band-pass DTNs, but there was no difference in spectral bandwidth between the two duration tuning classes. B: there was no correlation between temporal Q_{10 dB} and spectral Q_{10 dB} values. B, left: cells with U-shaped frequency tuning curves had larger spectral Q_{10 dB} values than DTNs in temporal Q_{10 dB} values between the three frequency tuning classes. B, right: there were no difference in temporal or spectral Q_{10 dB} values between short-pass and band-pass DTNs. For statistical details, see Topographical Organization of Spectral and Temporal Tuning Classes and Fig. 7. In A and B, n = 74.

DISCUSSION

This is the first study to directly compare the spectral and temporal tuning properties of DTNs from the mammalian IC. Although frequency tuning and tonotopy are ubiquitous at all levels of the vertebrate central auditory pathway, DTNs are found only at and above the auditory midbrain (Casseday et al. 1994; Sayegh et al. 2011). While the responses of all auditory neurons, including DTNs, are selective for stimulus frequency and amplitude, DTNs have an additional neural filter for auditory processing, one that works in the time domain. Hence, the responses of DTNs can be viewed as level-dependent spectro-temporal filters. In echolocating bats, such filters could help to maximize the extraction of acoustic information from reflected echoes and/or trigger fixed action patterns associated with foraging (Casseday and Covey 1996). Indeed, the range of emitted signal durations was a good predictor of the range of neuronal BDs for DTNs in the IC of a variety of echolocating bats such as E. fuscus (Ehrlich et al. 1997), Myotis lucifugus (Galazyuk and Feng 1997), Antrozous pallidus (Fuzessery and Hall 1999), and Molossus molossus (Mora and Kössl 2004).

Topographical Organization of Spectral and Temporal Tuning

As expected from previous reports (e.g., Pinheiro et al. 1991; Haplea et al. 1994; Faure et al. 2003; Jen and Wu 2006; Sayegh et al. 2012), DTNs exhibited a robust tonotopic organization and had neural thresholds that increased with recording electrode depth in the IC and that covered the dynamic range of hearing (Figs. 3–4). The cells were also tuned to the frequencies and durations of big brown bat echolocation calls (Simmons 1987; Surlykke and Moss 2000; Fenton et al. 2011).

The frequency tuning bandwidths and spectral Q_{10 dB} values of DTNs were similar to other central auditory neurons not selective for stimulus duration in E. fuscus and other species of FM bats (Fig. 7, A and B; Grinnell 1963; Suga 1964; Jen and Kamada 1982; Casseday and Covey 1992; Haplea et al. 1994). The range of spectral Q_{10 dB} values for DTNs from the IC of E. fuscus was higher than the range for DTNs reported from the IC of the mouse (Brand et al. 2000), suggesting the ability to echolocate has naturally selected for sharper frequency tuning in bats. This idea is further supported by the evolution of an auditory fovea in bats that employ Doppler-shift compensation (Pollak and Casseday 1989; Fenton et al. 2012), where peripheral and central auditory neurons within the fovea can have Q_{10 dB} values ≤400, compared with cells outside the fovea with Q_{10 dB} values ≤20 (Suga et al. 1976; Pollak et al. 1986). While many DTNs in the IC of E. fuscus had small Q values, we noted that cells tuned to the frequencies of the fundamental FM acoustic element were those with the highest Q_{10 dB} values (Fig. 8B).

Our study found mixed evidence for the hypothesis that the temporal properties of DTNs are spatially organized within the IC. The negative correlations between electrode depth and BD (Fig. 3D) and CF and BD (Fig. 3E) are at odds with some previous reports, including this one by the present authors, that did not find evidence for a topographical organization of DTN temporal response properties within the IC of the bat (Pinheiro et al. 1991; Ehrlich et al. 1997; Faure et al. 2003; Luo et al. 2008). In contrast, two studies found a positive relation between CF and BD (Jen and Wu 2006; Wu and Jen 2006, 2008b), suggestive of a spatial map of duration tuning but organized in a direction opposite to the relation reported here (see Fig. 3D) and by Sayegh et al. (2012).

The inconsistency of this finding, both within and across laboratories, is puzzling. Perhaps DTNs are topographically organized along the medial-lateral and/or rostral-caudal axes of the IC? In the auditory midbrain of rats, Perez-Gonzalez et al. (2006) found that DTNs were more abundant in the external cortex and lateral areas of the IC. Unfortunately, their study did not report electrode depths so it was not possible to make direct comparisons with our results. Future studies employing three-dimensional stereotaxic measures and/or intracellular recording with staining are needed to resolve the question of whether the mammalian IC contains an organized map of duration selectivity and/or if particular subdivisions of the IC are specialized for processing signal duration.

Frequency Tuning and Implications for Auditory Processing

Notwithstanding the eloquent paper by Palmer et al. (2013), we found three classes of eFRAs for DTNs from the IC of E. fuscus. V-shaped frequency tuning curves were most common among DTNs. Neurons with V-shaped spectral tuning are typical of primary auditory afferents (Kiang 1965) and neurons in the cochlear nucleus (Rhode and Smith 1985), superior olivary complex (Covey et al. 1991), lateral lemniscus (Covey and Casseday 1991), and, to a lesser extent, the IC (Haplea et al. 1994). Primary afferents inherit their V-shaped tuning from the vibratory mechanics of the cochlea: as sound amplitude increases, this causes larger displacements of the basilar membrane and stronger shearing forces on the hair cell stereocilia within the Organ of Corti, resulting in receptor depolarization, the generation of action potentials in a wide array of afferents, and a decrease in tuning sharpness (Robles and Ruggero 2001). Therefore, it is not surprising that many auditory neurons have V-shaped frequency tuning curves.

Duration-tuned neurons with U-shaped eFRAs were the second most common spectral tuning class. In FM bats and other mammals, U-shaped tuning originates within the central auditory system through sideband inhibition that causes a narrowing of frequency selectivity (Galazyuk and Feng 1997; Ramachandran et al. 1999; Palmer et al. 2013). Level tolerant cells have also been described as “filter” neurons because they have narrow (selective) frequency tuning with a constant bandwidth over a large range of SPLs (in some cases up to +90 dB re: threshold; Casseday and Covey 1992). Although not tested for duration selectivity, the filter neurons reported by Casseday and Covey (1992) from the IC of E. fuscus were clustered in isofrequency lamina tuned to the fundamental FM acoustic element of the bat's calls, suggesting their importance in processing echolocation information. Similar clustering patterns have been reported for filter neurons from the auditory cortex of M. molossus and Macrotus waterhousii (Macías et al. 2013). Cells with U-shaped frequency tuning curves may play a role in the constancy of auditory perception during changes in stimulus amplitude (Comalli and Altshuler 1976; Suga and Tsuzuki 1985; Casseday and Covey 1996). For example, in human speech sound amplitude can convey affective information (e.g., yelling indicates anger/excitement), whereas semantic information is generally amplitude independent (Breazeal and Aryananda 2002). Echolocating animals in particular may benefit from level tolerant perception because echo amplitude decreases rapidly with target distance (Kick and Simmons 1984; Macías et al. 2013).

The least common frequency tuning class was DTNs with O-shaped eFRAs. One explanation for their low number is that they are simply more difficult to find owing to their restricted range of acoustic excitability. Alternatively, if DTNs with closed frequency tuning curves were more common in IC regions other than the central nucleus, then it is possible that our electrode sampling regime underestimated their true proportion. A recent study by Palmer et al. (2013) summarizing >20 yr of recording from the IC of the guinea pig found that neurons with closed tuning curves comprised <6% of the sampled population, a proportion that closely agrees with our results for DTNs. Echolocating animals may benefit from DTNs with closed frequency tuning curves because these cells could facilitate the analysis of auditory scenes (Moss and Surlykke 2010). For example, echolocating bats face the challenge of detecting faint echoes that are similar in frequency and duration to their emitted vocalizations (Simmons and Stein 1980). A typical auditory neuron with appropriate frequency tuning would respond to both the pulse and echo if its acoustic threshold was lower than the amplitude of the received echo; however, the responses of a DTN with appropriate spectro-temporal tuning and an O-shaped eFRA would be highly selective for the echo if the lower and upper acoustic thresholds of the cell were below than the amplitude of the emitted call but bracketed the amplitude of the received echo. In other words, populations of DTNs with closed frequency tuning curves could be highly selective and respond only to the loud outgoing call or the fainter returning echo but not both. These cells could also be helpful in discriminating echoes of the bat's own call from those of other bats and in highly cluttered acoustic environments (Moss and Surlykke 2010).

Spectral and Temporal Tuning Bandwidths and Sharpness

Studies investigating the selectivity of auditory neurons typically report tuning bandwidths or quality factors. To characterize the spectral-temporal selectivity of DTNs, we reported both. The 50% response bandwidth is a reliable and unbiased estimate of the lower and upper cut-off values of a spectral or temporal tuning curve. Moreover, its value is independent of differences in the maximum firing rate between cells.

The quality factor (Q value) is commonly used to report spectral tuning sharpness (Covey and Casseday 1991; Capranica 1992; Haplea et al. 1994; Brand et al. 2000). This measure works well because tuning in the frequency domain is based on the accepted notion that, owing to the vibratory mechanics of the cochlea and the asymmetry in the envelope of the basilar membrane traveling wave, spectral bandwidths of primary auditory afferents increase with SPL and CF (Oxenham and Shera 2003; Ruggero and Temchin 2005; Temchin et al. 2008). To the best of our knowledge, our study is the first to use a Q value in the temporal domain. We originally thought Q values would be useful in quantifying temporal tuning sharpness because the 50% temporal bandwidth increased with neuronal BD (but not SPL) in a manner analogous to the increase in 50% spectral bandwidth with CF (Fig. 8; Fremouw et al. 2005; Jen and Wu 2006). Although we found a positive correlation between neuronal BD and the 50% duration tuning bandwidth, there was no relation between BD and temporal Q_{10 dB}. Moreover, the range of temporal Q_{10 dB} values in our population of DTNs was quite narrow. In hindsight, these findings were not surprising. Primary afferents inherit their frequency selectivity from the mechanics of the Organ of Corti, but there is no mechanism linking the sharpness of duration tuning to the mechanics of the cochlea nor is there a compelling reason to characterize the BD of a neural temporal filter relative to its bandwidth. We conclude that frequency tuning bandwidths and spectral Q values are useful and appropriate for characterizing neural filter properties in the frequency domain and that duration tuning bandwidths but not temporal Q values are useful for characterizing neural filter properties in the time domain.

Relation Between Spectral and Temporal Tuning

Our study found that the spectral and temporal tuning bandwidths of DTNs from the IC of the bat showed opposite patterns of spatial organization: cells with the widest spectral bandwidths tended to be found in the ventral IC, whereas cells with widest temporal bandwidths tended to be found in the dorsal IC (Fig. 7, A and C). The organization of duration tuning classes mirrored these trends: band-pass DTNs with wide temporal tuning curves were located in the dorsal IC, whilst short-pass DTNs with narrow temporal tuning curves were distributed more in the ventral IC (Fig. 6; see also Fuzessery and Hall 1999). A direct comparison of spectral and temporal tuning properties revealed that DTNs with the smallest temporal bandwidths were among those cells with the widest spectral bandwidths and vice versa (Fig. 9). Altogether, these data suggest that DTNs in the IC of the bat experience a spectral-temporal trade-off in tuning bandwidth resolution.

Any trade-off that exists is suggestive of a physical constraint on the capabilities of the auditory system as a signal analyzer. A resonant electrical circuit is a closed system and the duration of ringing in the filter response depends on its center (resonant) frequency and spectral bandwidth (Blanchard 1941). Frequency tuning in the peripheral auditory system results directly from the mechano-electrochemical transduction processes but can also be inherited and modified by neurons located in higher auditory centers; however, duration selectivity is an emergent neural property that first appears at the level of the IC. There is evidence that auditory brainstem neurons have response properties that effectively facilitate independent processing of spectral and temporal information. For example, neurons in the columnar region of the ventral nucleus of the lateral lemniscus (VNLLc) are broadly tuned in frequency and always respond with a single spike tightly time-locked to stimulus onset (Covey and Casseday 1991). It would seem that frequency analysis by VNLLc neurons has been abandoned to create a robust representation of signal timing, an important aspect of temporal processing and a vital part of the computational circuits proposed for creating duration selective responses (Fuzessery and Hall 1999; Casseday et al. 2000; Faure et al. 2003; Aubie et al. 2009, 2012). While our study found evidence of a trade-off in the resolution of DTN spectral and temporal tuning bandwidths (Fig. 9A), future studies are needed to determine its underlying cause.

GRANTS

This research was supported by a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada (to P. A. Faure) and by operating grants from the National Institute on Deafness and Other Communication Disorders (to J. H. Casseday and E. Covey). R. Sayegh was supported by an Ontario Graduate Scholarship. The McMaster Bat Laboratory is also supported by infrastructure grants from the Canada Foundation for Innovation and the Ontario Innovation Trust.

AUTHOR CONTRIBUTIONS

E.C. and P.A.F conception design research; J.A.M., T.F., R.S., and P.A.F. performed experiments; J.A.M., F.F., R.S. and P.A.F analyzed data; J.A.M., F. F., T.F., R.S., E.C., and P.A.F. interpreted results; J.A.M. and P.A.F. prepared figures; J.A.M. and P.A.F. drafted manuscript; J.A.M., F. F., T.F., R.S., E.C., and P.A.F. edited and revised manuscript; J.A.M., F. F., T.F., R.S., E.C., and P.A.F. approved final version of manuscript.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).