Spectral and spatial tuning of onset and offset response functions in auditory cortical fields A1 and CL of rhesus macaques

In the current study, we examine the relationship between the tuning of neural responses evoked by the onset and offset of acoustic stimuli in the primary auditory cortex, as well as a higher-order auditory area—the caudolateral belt field—in awake rhesus macaques. In these areas, the relationship between onset and offset response profiles in frequency and space domains formed a continuum, ranging from highly overlapping to highly nonoverlapping.

Abstract

The mammalian auditory cortex is necessary for spectral and spatial processing of acoustic stimuli. Most physiological studies of single neurons in the auditory cortex have focused on the onset and sustained portions of evoked responses, but there have been far fewer studies on the relationship between onset and offset responses. In the current study, we compared spectral and spatial tuning of onset and offset responses of neurons in primary auditory cortex (A1) and the caudolateral (CL) belt area of awake macaque monkeys. Several different metrics were used to determine the relationship between onset and offset response profiles in both frequency and space domains. In the frequency domain, a substantial proportion of neurons in A1 and CL displayed highly dissimilar best stimuli for onset- and offset-evoked responses, although even for these neurons, there was usually a large overlap in the range of frequencies that elicited onset, and offset responses and distributions of tuning overlap metrics were mostly unimodal. In the spatial domain, the vast majority of neurons displayed very similar best locations for onset- and offset-evoked responses, along with unimodal distributions of all tuning overlap metrics considered. Finally, for both spectral and spatial tuning, a slightly larger fraction of neurons in A1 displayed nonoverlapping onset and offset response profiles, relative to CL, which supports hierarchical differences in the processing of sounds in the two areas. However, these differences are small compared with differences in proportions of simple cells (low overlap) and complex cells (high overlap) in primary and secondary visual areas.

NEW & NOTEWORTHY In the current study, we examine the relationship between the tuning of neural responses evoked by the onset and offset of acoustic stimuli in the primary auditory cortex, as well as a higher-order auditory area—the caudolateral belt field—in awake rhesus macaques. In these areas, the relationship between onset and offset response profiles in frequency and space domains formed a continuum, ranging from highly overlapping to highly nonoverlapping.

in their classic experiments, Hubel and Wiesel (1962) identified two classes of cells in primary visual cortex (V1): simple cells and complex cells. V1 neurons respond to stimulus onset (ON) as well as stimulus offset (OFF), and a key distinguishing criteria between these two classes was the difference in the relationship of the tuning of ON and OFF responses. For simple cells, ON and OFF subregions of the receptive field are spatially segregated, whereas in complex cells, they are coextensive. Subsequent studies have used other criteria, such as the response linearity of the cells, to classify V1 neurons and have reported bimodal distributions of neurons by those criteria (Chen et al. 2009; Skottun et al. 1991), supporting the existence of discrete populations of simple and complex cells in V1.

OFF responses in the auditory system are less well characterized. The existence of offset-evoked responses has been reported in various mammalian species at different levels along the auditory pathway, including the cochlear nucleus (Suga 1964; Young and Brownell 1976), superior olivary complex (Kulesza et al. 2003), inferior colliculus (Fuzessery and Hall 1999; Grinell 1973; Kasai et al. 2012; Lesser et al. 1990; Suga 1964), medial geniculate body (Aitken and Prain 1974; Anderson and Linden 2016; He 2001, 2002), and auditory cortex (Engle and Recanzone 2013; Recanzone 2000a; Tian et al. 2013). The paucity of data on OFF responses in auditory cortex is likely due to the effects of anesthesia on OFF responses of neurons across different levels in the central auditory pathway (Cheung et al. 2001; Gaese and Oswald 2001; Moshitch et al. 2006; Sutter and Schreiner 1991; Young and Brownell 1976). Frequently, although the presence of OFF responses was reported, these responses were not characterized extensively, due to low percentages of offset-evoked responses.

Under nonbarbiturate anesthetic or in awake animals, studies have found that substantial proportions of primary auditory cortex (A1) neurons exhibit OFF responses (Qin et al. 2007; Tian et al. 2013). Furthermore, a number of such studies reported that for many of these neurons, ON and OFF responses have different best frequencies and exhibit nonoverlap in their spectral tuning (Fishman and Steinschneider 2009; Moshitch et al. 2006; Qin et al. 2003, 2007; Scholl et al. 2010; Tian et al. 2013). Based on the relationship between ON and OFF spectral tuning, Tian et al. (2013) proposed that neurons in auditory cortex fall into two different classes, which they referred to as type-S and type-C cells, analogous to simple and complex cells in the visual system, with ON/OFF receptive-field subregions that are nonoverlapping in type-S but overlapping in type-C cells.

In addition to A1 neurons, OFF responses have previously been reported in belt areas of the auditory cortex (Engle and Recanzone 2013; Recanzone 2000a), including the caudolateral (CL) belt field (Engle and Recanzone 2013). Both A1 and CL contain neurons that exhibit frequency selectivity (Rauschecker et al. 1995; Recanzone 2000b; Schreiner et al. 2011). However, OFF responses of neurons in higher-order auditory cortical areas have been examined to an even lesser extent than those of neurons in A1. Thus it is unknown whether neurons in belt areas exhibit temporally dynamic frequency tuning, such as previously described in A1 of macaques (Fishman and Steinschneider 2009; Tian et al. 2013).

Auditory cortical areas A1 and CL also contain neurons tuned to specific spatial locations (Engle and Recanzone 2013; Juarez-Salinas et al. 2010; Recanzone 2000b; Recanzone et al. 2000; Woods et al. 2006). Among the few studies that have previously examined OFF responses in auditory cortex neurons in anesthetized and awake animals, a majority has focused on comparing frequency tuning of ON and OFF responses (Fishman and Steinschneider 2009; Moshitch et al. 2006; Qin et al. 2007; Scholl et al. 2010; Tian et al. 2013); other properties, including spatial tuning, are largely unstudied. A study in anesthetized ferrets (Hartley et al. 2011) reported differential binaural tuning of ON and OFF responses in A1. However, studies investigating spatial tuning of OFF responses have not previously been performed in awake animals. Furthermore, it is unknown whether neurons in either core or belt areas of auditory cortex in primates exhibit spatial tuning that changes over the time course of the neural response.

Whereas it is known that auditory cortex neurons are tuned to spatial, spectral, and temporal aspects of acoustic stimuli, it is unclear to what extent tuning properties are preserved across different phases of the neural response. Studies of the tuning of OFF responses in auditory cortex are very limited, particularly in awake animals and even more so in the case of primates. Given that there is some evidence for differences in frequency tuning of ON and OFF responses (Fishman and Steinschneider 2009; Moshitch et al. 2006; Qin et al. 2007; Scholl et al. 2010; Tian et al. 2013), we were interested in examining whether the selectivity of auditory cortex neurons to other attributes of sound also exhibits such differences between the tuning of their ON and OFF responses. Therefore, one reason to investigate spatial tuning of OFF responses was to extend our understanding of ON/OFF response tuning relationships beyond the spectral domain. Frequency tuning in the auditory system and retinotopic tuning in the visual system may be considered analogous in that they both arise from the organization of their respective sensory epithelia. For this reason, parallels have been drawn between A1 neurons that exhibit nonoverlapping/overlapping ON vs. OFF tuning in tonotopic space and simple/complex cells in V1 (Tian et al. 2013) that encode two-dimensional, retinotopic space. In addition, auditory neurons are tuned to two-dimensional space, although this does not arise directly from the functional organization of the auditory sensory epithelium and is instead computed centrally. If ON/OFF tuning relationships arise from circuits specialized to perform certain operations, such as spatial processing, which are shared across different cortical areas, then the spatial tuning of auditory cortical neurons could be expected to share spatial tuning properties with visual cortical neurons.

In the current study, we were interested in determining the extent of overlap between the ON and OFF responses as a function of both frequency and spatial location to extend studies examining the relationship between ON and OFF response tuning (Fishman and Steinschneider 2009; Moshitch et al. 2006; Qin et al. 2003, 2007; Scholl et al. 2010; Tian et al. 2013). Given the broad frequency and spatial selectivity of auditory cortical neurons compared with the spatial selectivity of visual cortical neurons, we investigated the level of variation in the amount of overlap between ON and OFF response tuning using several different metrics (Fig. 1, B and C). Furthermore, we were interested in seeing which types of metrics could reveal different classes of neurons, as opposed to a single continuum across any particular metric. We therefore recorded responses of single neurons to tone and noise stimuli in the A1 and the CL belt area of rhesus macaques, allowing us to make comparisons of ON and OFF response functions of neurons with respect to spectral and spatial tuning across different levels of the auditory cortical processing hierarchy in the absence of any effects of anesthesia. This relationship between the two cortical areas was of interest, as neurons in A1 have sharper spectral tuning than those in CL (Recanzone 2000a), and CL neurons have sharper spatial tuning than those in A1 (Recanzone et al. 2000; Woods et al. 2006).

Fig. 1.

Simulation illustrating how metrics of tuning similarity change with the extent of overlap in ON and OFF tuning response profiles. A: illustration of normalization method. Since OFF firing rates are often lower than ON firing rates in our dataset, in many cases, the raw OFF response profile may be entirely contained within the raw ON response profile. In some cases, the peak of the OFF tuning curve is aligned with 1 of the flanks of the ON tuning curve (top). In other cases, OFF tuning curve with lower firing rates may be closely aligned in shape with the ON tuning curve (bottom). Without normalization, the exact same value is obtained for tuning overlap fraction (TOF) in both cases. With our normalization method, the first case produces a lower TOF value than the second case and would therefore be ranked as having less overlap between ON and OFF response profiles compared with the second case (see materials and methods for more details). B: simulated ON (black) and OFF (gray) tuning curves that overlap to different extents (little overlap to almost complete overlap, from left to right). C: means and SD of each metric, across 1,000 simulated pairs for the different extents of tuning curve overlap specified in B. Error bars indicate SD of the mean. We used 7 different similarity metrics (see materials and methods). For all 7 metrics, higher values indicate greater overlap between ON and OFF tuning curves. Distributions that were deliberately generated to possess only low and high tuning overlap extents were significantly nonunimodal for all metrics (Hartigan's dip test statistics, P < 0.001).

MATERIALS AND METHODS

Animals.

All procedures used in these experiments were reviewed and approved by the Institutional Animal Care and Use Committee of the University of California, Davis (U.C. Davis), and were in accordance with the Association for Assessment and Accreditation of Laboratory Animal Care and the Society for Neuroscience guidelines for the use of animals in neuroscience research. Data used in this study have been reported previously (Engle and Recanzone 2013; Juarez-Salinas et al. 2010; Recanzone 2000a; Woods et al. 2006), and further detail on data collection procedures can be found in those publications. The experiments were performed on three adult male rhesus macaque monkeys (F, G, and L), aged 5–12 yr, weighing 7–12 kg during the course of the study. All animals used in this study were reared in the California National Primate Center at U.C. Davis and did not have any history of auditory impairment, ear infections, prolonged noise exposure, or treatment with ototoxic medications. The monkeys were fluid regulated, following U.C. Davis guidelines, to motivate behavioral performance.

Acoustic stimulus presentation.

The experiments were carried out inside a double-walled anechoic chamber, lined with sound-absorbent foam, with the monkey seated in an acoustically transparent primate chair. Stimuli comprised noise bursts to measure spatial tuning characteristics and tones to measure frequency tuning characteristics generated using a Tucker-Davis Technologies (Alachua, FL) system. Noise-burst stimuli consisted of 200 ms duration (5 ms rise/fall), “unfrozen” Gaussian noise at 55 and 75 dB sound-pressure level (SPL), presented from a speaker at 1 of 16 different locations, which were spaced by 22.5° and spanned 360° in azimuth at head level. No significant differences were found between the 55- and 75-dB SPL groups, so those data were combined for our analyses. Neural responses were tested at each speaker location for 12 repetitions. In experiments to assess frequency tuning, stimuli consisted of 200 ms (5 ms rise/fall) tone frequencies of 0.5, 1, 2, 3, 4, 5, 6, 7, 8, 10, 11, 12, 14, 16, and 18 kHz, presented at 65 dB SPL. Twelve repetitions of each frequency were presented in a randomly interleaved fashion. Tone stimuli were presented from the speaker directly opposite to the ear, contralateral to the recorded hemisphere.

Behavioral task.

To ensure that the monkeys remained alert throughout the recording session, they were engaged in performing a simple task. Monkeys initiated a trial by depressing a lever, and an acoustic stimulus (S1) was presented. After three to eight more stimuli with different parameters (intensity, tone frequency, spatial location, etc.) were presented (interstimulus interval of 750–800 ms), the test stimulus (S2) was presented. For sessions in which spatial tuning was measured, the S2 was a stimulus with the same intensity and location as the immediately, previously presented S1, i.e., two identical stimuli in succession. For sessions in which frequency tuning was measured, S2 was a stimulus presented from a location in front of the monkey (±30° in azimuth), instead of the speaker located at 90° on the contralateral side, which was used for the remaining stimuli. Monkeys were required to release the lever within 800 ms of S2 offset to receive a fluid reward, after which a 3- to 5-s delay was incorporated to allow the monkey to finish swallowing before initiation of the next trial. If the monkey did not release the lever in the required time, then fluid reinforcement was not provided, and a brief time-out followed instead.

The behavioral task was performed by monkeys F and L. Monkey G could not be trained on the task, so this animal was required to remain quietly seated in the chair while stimuli were presented and was provided with a reward every three to eight stimuli to match the reward scheme used for the remaining monkeys. Analysis of other response properties in this monkey did not differ from those in the other monkeys (Juarez-Salinas et al. 2010; Miller and Recanzone 2009; Recanzone 2008; Woods et al. 2006) nor were any noticed in this report, so all data were pooled.

Electrophysiological recording procedures.

The monkeys were surgically implanted with a restraining head post and a recording cylinder to allow for orthogonal penetration of the superior surface of the superior temporal gyrus [see Recanzone (2000a) for further details]. Recordings were made in the left hemisphere for all monkeys. During recording sessions, a tungsten microelectrode (2–4 MΩ; FHC, Bowdoin, ME) was inserted through a guide tube (penetrating the brain by ∼3 mm) and then advanced in depth using a hydraulic microdrive (Narishige, Tokyo, Japan). Neural activity was monitored on an oscilloscope and via a speaker by an experimenter seated outside of the sound-isolation chamber, while search stimuli (broad-band noise bursts, tones, band-passed noise, clicks) were presented. When driven activity was observed, single units were isolated using a time-amplitude window discriminator (Bak Electronics, Umatilla, FL). The time stamps of isolated units were recorded with a 1-ms resolution. The position of each recording site within A1 or CL was assigned based on location within the recording cylinder and physiological characteristics, such as tonotopic gradient and frequency tuning [see Recanzone (2000a)]. The location of electrode tracks was later verified histologically.

Data analysis.

Time stamps of isolated single units were exported to MATLAB (MathWorks, Natick, MA) and analyzed using custom scripts. Only data from S1 stimulus presentations are included in this report. For each neuron, peristimulus time histograms (PSTHs) were constructed for each stimulus by averaging spike counts across all repetitions of the stimulus, using a 1-ms bin width. PSTHs were smoothed with a 5-ms moving average for visualization purposes, but unsmoothed PSTHs were used for all calculations described. Responses were considered to be significant and were included in the remaining analyses if the firing rate in the time window under consideration exceeded the baseline firing rate (in a comparable time window) by at least 2 SD. We also tested an alternative method of detecting significant responses—Wilcoxon signed-rank for paired data—to compare onset and offset period firing rates with baseline firing rates (Tian et al. 2013). This did not result in any major differences in relative proportions of neurons exhibiting ON and/or OFF responses, although there were some differences in the total number of neurons found to exhibit a significant response. For the results shown, we have used the threshold based on SD, since significant responses detected using this method were in better agreement with visual inspection of the PSTHs, both for cells with low or high firing baseline levels. Peak firing rate was determined from the PSTH as the maximum spike count contained within a 50-ms sliding window (shifted in 1 ms increments), divided by the same time interval (50 ms). The stimulus that evoked the highest peak firing rate was defined as the best stimulus (best location or best frequency). Peak firing rates and best stimuli were computed separately for the onset period and for the offset period. For each neuron, the PSTH for the best ON stimulus and the PSTH for the best OFF stimulus were obtained, and these were then used to determine population average PSTHs in response to the best ON stimulus and the best OFF stimulus. ON and OFF responses were measured in a 50-ms time window following stimulus onset and offset, respectively. For each neuron, the time window for measurement of ON responses was shifted relative to stimulus onset by a period equivalent to the response latency of the neuron, which was the first bin within the peak firing rate window of the best stimulus PSTH, containing a firing rate >2 SD above the baseline firing rate. The time window for measurement of OFF responses was shifted relative to stimulus offset by the same amount. This was consistent with visual inspection of the PSTHs. Additionally, analyses were performed using time windows of 0–100 ms following stimulus onset and offset, and no appreciable differences from the results reported here were noted. We quantified the strength of OFF responses relative to sustained period activity (OFF/SUS) as the absolute value of the ratio of offset period activity to sustained period activity (101–150 ms) after subtracting baseline activity. Spontaneous activity was not subtracted for the calculation of tuning similarity metrics for the data shown. However, all analyses were also performed after correcting for spontaneous activity [following the method of Tian et al. (2013)], and all results obtained were similar to those reported here.

Spatial and spectral tuning curves were constructed using the peak firing rate to each stimulus. For each neuron, separate tuning curves were generated for the onset period and the offset period. Since the goal was to compare the shapes of the response profiles of onset- and offset-evoked responses, for all analyses, ON and OFF tuning curves were normalized to avoid artifacts in the calculation of similarity metrics due to the lower firing rates evoked by stimulus offset compared with stimulus onset. Tuning curves were normalized to lie within the range of zero to one. For n tested stimuli, normalized ON tuning curves were calculated as the following

$$\frac{\left\{\text{ON}1,\text{ON2},\dots ,\text{ONn}\right\}-\min \left\{\text{ON}1,\text{ON2},\dots ,\text{ONn}\right\}}{\max \left(\left\{\text{ON}1,\text{ON2},\dots ,\text{ONn}\right\}\right)-\min \left(\left\{\text{ON}1,\text{ON2},\dots ,\text{ONn}\right\}\right)}$$

Normalized OFF tuning curves were calculated as the following

$$\frac{\left\{\text{OFF}1,\text{OFF}2,\dots \text{OFFn}\right\}-\min \left(\left\{\text{OFF}1,\text{OFF}2,\dots \text{OFFn}\right\}\right)}{\max \left(\left\{\text{OFF}1,\text{OFF}2,\dots \text{OFFn}\right\}\right)-\min \left(\left\{\text{OFF}1,\text{OFF}2,\dots \text{OFFn}\right\}\right)}$$

The bandwidth of spectral and spatial tuning curves was calculated, respectively, as the range in octaves and the degrees in space, spanned by stimuli that evoked >50% of the best stimulus response. For multipeaked tuning curves, bandwidth was calculated by summing the bandwidths (range in octaves and the degrees in space, spanned by stimuli that evoked >50% of the best stimulus response) of each of the individual peaks of the tuning curves.

Similarity metrics.

Measures of similarity comparing onset period responses with offset period responses were calculated based on spatial and spectral tuning curves for all units that exhibited significant ON and OFF responses. Normalized tuning curves were used in the calculation of all similarity metrics. Best stimulus difference was calculated for spatial tuning as the absolute difference in ON and OFF best location (in degrees) and for spectral tuning as the absolute difference in onset and offset best frequency (in octaves). Furthermore, we used seven metrics of similarity to assess the relationship of ON and OFF tuning of auditory cortex neurons, and these are described below.

The tuning overlap fraction of the ON and OFF response profiles was examined. For each stimulus tested, the ON response and OFF response value was compared (calculated as described above). The larger of those two values for each stimulus was then summed to obtain the “total combined area.” The smaller of those two values for each stimulus was then summed to obtain the “overlap area.” The tuning overlap fraction was then calculated as the overlap area divided by the total combined area.

$$\text{for}n\text{tested}\text{stimuli},\text{tuning}\text{overlap}\text{fraction}=\frac{{\displaystyle {\sum }_{i=1}^n\min \left(\text{ONi},\text{OFFi}\right)}}{{\displaystyle {\sum }_{i=1}^n\max \left(\text{ONi},\text{OFFi}\right)}}$$

For each neuron, cumulative distribution functions were obtained for onset and offset response tuning curves. Cumulative distributions corresponding to ON and OFF tuning curves were each normalized to sum = 1. Cumulative profile distance was then measured as the maximum difference between the two cumulative distribution functions.

For the next three metrics, ON and OFF tuning curves were each considered to be vectors containing as many elements as stimuli tested, and the following metrics were computed between the two vectors, as per their SDs: Euclidean distance, Pearson's correlation, and Spearman's rank correlation. Before calculation of “Euclidean distance,” the vectors corresponding to ON and OFF tuning curves were each normalized by dividing by their respective Euclidean lengths.

All metrics of dissimilarity (cumulative response profile distance, Euclidean distance) were multiplied by −1 for consistency of data visualization.

Relative OFF response magnitude was calculated by computing the magnitude of the OFF response to the best ON stimulus and dividing that value by the magnitude of the OFF response to the best OFF stimulus.

Relative ON response magnitude was calculated by computing the magnitude of the ON response to the OFF best stimulus and dividing that value by the magnitude of the ON response to the best ON stimulus.

We validated the tuning overlap metrics used in this study by applying them to simulated data. We simulated ON and OFF tuning curves as pairs of Gaussian curves with a variance of 1 and means that were separated by varying extents, ranging from 0.0 to 2.5 SD. Overlap extent was divided into five groups based on the separation between means of the tuning curves. One-thousand pairs of tuning curves were randomly generated in each of the five groups, and the seven similarity metrics used in this study were then computed for each simulated pair. Furthermore, we deliberately generated a population of 1,000 pairs of tuning curves in which 50% of the population belonged to the “high-overlap” group (0.0–0.5 SD), and 50% of the population belonged to “low-overlap” group (2.0–2.5 SD), and intermediate values were not included. The resulting distributions for all seven metrics were then evaluated using Hartigan's dip test statistic.

Statistical analyses were performed in MATLAB version 8.1.0 (MathWorks) and R version 3.2.0 (R Development Core Team 2015; The R Project for Statistical Computing, https://www.r-project.org/). For each parameter, normality of the distribution was assessed using the Shapiro-Wilk test. Whenever assumptions of normality were not met, statistical analyses were conducted using nonparametric tests. Cumulative distributions were compared using Kolmogorov-Smirnov test for comparison of two samples and Kruskal-Wallis test for comparison of multiple samples. A nonparametric multivariate ANOVA (MANOVA) was used to assess for differences between A1 and CL neurons, across all similarity metrics computed. Differences were considered to be statistically significant when P < 0.05.

RESULTS

We recorded responses of single neurons in A1 and the CL belt area to tone and noise stimuli in three awake rhesus macaques. Table 1 shows the number of units recorded in each area and the fractions of those units displaying a significant response following stimulus onset and stimulus offset. Only neurons that responded to both ON and OFF were included for all following analyses. As is clear from Table 1, the vast majority of neurons were responsive during both the ON and OFF periods to tone stimuli in A1, as well as CL. In contrast, for noise stimuli, only ∼60% and ∼80% of responsive neurons had both ON and OFF responses in A1 and CL, respectively. A smaller fraction of cells had only ON responses (39% and 16% in A1 and CL, respectively).

Table 1.

Percentages of A1 and CL neurons displaying ON and OFF responses

	ON Only	OFF Only	ON + OFF	OFF/SUS >0.5	OFF/SUS >1.0
A1–tone, n = 207/231	0% (0/207)	0% (0/207)	100% (207/207)	91.3% (189/207)	73.4% (152/207)
CL–tone, n = 118/131	0% (0/118)	0% (0/118)	100% (118/118)	93.2% (110/118)	66.9% (79/118)
A1–noise, n = 166/177	38.6% (64/166)	1.8% (3/166)	59.6% (99/166)	72.7% (72/99)	36.4% (36/99)
CL–noise, n = 110/123	16.4% (18/110)	3.6% (4/110)	80% (88/110)	96.6% (85/88)	65.9% (58/88)

Total numbers of neurons evoking a significant response following stimulus onset and offset are listed above. Percentages are given for neurons that display a significant response to at least 1 tone stimulus among tested frequencies (0.5–18 kHz) or at least 1 speaker location among tested azimuthal locations (0–360°). Additionally, OFF responses are quantified relative to sustained period activity (see materials and methods). Among neurons that exhibited significant OFF responses relative to baseline, percentages are provided for neurons that met a criterion of the OFF/SUS ratio > 0.5 (Recanzone 2000a) or the OFF/SUS ratio >1.

ON/OFF tuning similarity metrics were computed based on normalized tuning curves. In some cases, one of the raw response profiles may be contained almost entirely within the other raw response profile, due to differences in the firing rates between the two response profiles (Fig. 1A). In some cases, the peak of one of the two tuning curves may be aligned with a flank of the other (Fig. 1A). In other cases, ON and OFF response profiles may be closely aligned in shape, although one might have lower firing rates than the other (Fig. 1A). The differentiation between such differences can be achieved by the normalization method described (see materials and methods; Fig. 1A).

Two of the metrics used—relative OFF response magnitude and relative ON response magnitude—by definition, require normalization. The measures of cumulative profile distance and Euclidean distance produce highly similar values, with or without normalization, which may be expected, since each of these metrics involves its own normalization step [Pearson's correlation coefficients (r) for values obtained with and without normalization: cumulative profile distance r = 0.89; Euclidean distance r = 0.87]. Two other metrics—Pearson's correlation and Spearman's correlation—produce identical values, with or without normalization. We found that normalization was particularly important for capturing shifts in tuning using the metric of “tuning overlap fraction.” When the tuning overlap fraction was compared with the Pearson's correlation overlap metric (which is identical, with or without normalization), the r value was 0.28, which indicates that neurons that were being considered high overlap or low overlap using Pearson's correlation metric were not similarly categorized using tuning-overlap fraction. However, after normalization of tuning curves, the two metrics were much more strongly correlated, with an r value of 0.79. Given these considerations, we expect that normalization of ON and OFF tuning curves aided in effectively quantifying the extent of overlap between ON and OFF response profiles.

We validated the tuning overlap metrics used in this study by applying them to simulated data. Figure 1 provides a graphical representation of tuning similarity metrics for five pairs of simulated tuning curves that exhibit different extents of overlap in their ON and OFF response profiles, which span the range of largely nonoverlapping to almost completely overlapping (Fig. 1B). Means and SDs of each of the seven tuning overlap metrics, for different extents of tuning-curve overlap, were calculated across all simulated pairs within each group (Fig. 1). It is clear from these simulations that each of the metrics provides a continuous range of values that closely mimic the similarity of the two tuning curves, as assessed visually.

To verify that these metrics would capture whether the neurons were distributed in two different classes, i.e., high overlap vs. low overlap, we simulated a population of 1,000 tuning-curve pairs in which 50% of the population belonged to the high-overlap group (0.0–0.5 SD), and 50% of the population belonged to low-overlap group (2.0–2.5 SD). The resulting distributions for all seven similarity metrics were significantly nonunimodal (Hartigan's dip test statistic, P < 0.001), verifying that these metrics could accurately reveal nonunimodal distributions of neurons.

The tuning functions of example neurons from our dataset are shown as a function of frequency (Fig. 2) and spatial location (Fig. 3). Figure 2 shows the tuning functions for tone stimuli with the best ON response and best OFF response, with the peak response shown in Fig. 2, A–D, for four individual neurons. The corresponding PSTHs are shown in Fig. 2, E–H. In some cases, spectral and temporal profiles for ON and OFF responses were highly overlapping (Fig. 2, A, C, E, and G). In other cases, ON and OFF responses were nonoverlapping (Fig. 2, B, D, F, and H). These examples are representative of the sample of neurons from which we recorded in both A1 and CL, as they included neurons with similar ON and OFF spectral and temporal profiles, as well as those with different ON and OFF spectral and temporal response profiles. A similar observation was made with respect to spatial tuning of neural responses in A1 and CL to broad-band noise stimuli (Fig. 3, A, C, E, and G); some neurons displayed similar spatial tuning profiles and response time course for noise stimuli at best ON locations vs. best OFF locations, whereas others displayed different spatial and temporal response profiles for ON and OFF subregions within their receptive fields (Fig. 3, B, D, F, and H).

Fig. 2.

Representative responses of neurons in A1 and CL to tone stimuli. Tuning curves and peristimulus time histograms (PSTHs) are shown for ON (black) and OFF (gray) responses. A and B: spectral tuning curves of 2 different A1 neurons. C and D: spectral tuning curves of 2 different CL neurons. E–H: PSTHs in response to the best ON frequency (black circles in A–D) and the best OFF frequency (gray circles in A–D) for the neurons in A and B. In both A1 and CL, the population included neurons displaying similar spectral and temporal profiles for ON and OFF responses (A, C, E, G), as well as those with different ON and OFF spectral and temporal response profiles (B, D, F, H). Dashed, vertical lines indicate time of stimulus offset.

Fig. 3.

Representative responses of neurons in A1 and CL to broad-band noise stimuli. Tuning curves and peristimulus time histograms (PSTHs) are shown for ON (black) and OFF (gray) responses. A and B: spatial tuning curves of 2 different A1 neurons. C and D: spatial tuning curves of 2 different CL neurons. E–H: PSTHs in response to the best ON location (black circles in A–D) and the best OFF location (gray circles in A–D) for the neurons in A and B. In both A1 and CL, the population included neurons displaying similar spatial and temporal profiles for ON and OFF responses (A, C, E, G), as well as those with different ON and OFF spatial and temporal response profiles (B, D, F, H). Dashed, vertical lines indicate time of stimulus offset.

Onset- and offset-evoked firing rates.

We examined the time course of the population average response of auditory cortex neurons in response to ON and OFF best stimuli (Fig. 4). These plots show the average firing rate for the stimulus that elicited the best ON response and the best OFF response measured across all neurons (Fig. 4, A–D) with significant ON and OFF responses and when restricted to neurons that satisfy a criterion of OFF/SUS > 1 (see materials and methods; Fig. 4, E–H). Similar onset- and offset-evoked firing patterns were observed in both A1 and CL neurons in response to ON and OFF best frequencies. On average, onset best frequencies elicited a strong transient response following stimulus onset and little to no response following stimulus offset. In contrast, the average population response to best OFF frequencies contained a transient component following stimulus onset as well as offset, although the onset component was weaker and the offset component stronger than that evoked by the best ON stimulus (Fig. 4, A, B, E, and F). In comparison, differences in the population response to ON and OFF best location in A1 as well as CL neurons were smaller in magnitude compared with those seen for frequency tuning (Fig. 4, C, D, G, and H).

Fig. 4.

Population temporal response profiles. A–D: population average peristimulus time histograms in response to the ON and OFF best stimuli for A1 neurons (A: tones, C: noise) and for CL neurons (B: tones, D: noise). E–H: same as A–D, after restricting the dataset to neurons with OFF/SUS ratio > 1. Conventions as in previous figures. Dashed, vertical lines indicate time of stimulus offset. Differences in the mean response to ON and OFF best locations were smaller in magnitude relative to differences in the mean response to ON and OFF best frequencies.

The average population response was greater following the onset of the stimulus compared with the offset of the stimulus, but from these population averages, it is difficult to know if this is a trend that is consistent across neurons or if different subpopulations of neurons have greater OFF responses compared with ON responses. We therefore compared the magnitude of the response evoked by the ON and OFF best stimulus in individual auditory cortex neurons (Fig. 5, A–D). Best ON and best OFF firing rates of individual neurons in response to tone and noise stimuli exhibited a strong positive correlation in A1 and CL. In a majority of A1 and CL neurons, the best OFF stimulus evoked lower response magnitudes compared with the best ON stimulus (sign test; A1 tone P < 0.001, CL tone P < 0.001, A1 noise P < 0.001, CL noise P < 0.001), giving rise to most points being below the unity line [percentages of OFF < ON: A1 tone 78.3% (162/207), CL tone 85.6% (101/118), A1 noise 82.8% (82/99), CL noise 93.2% (82/88)]. Summary data for best ON and OFF stimulus-evoked firing rates are listed in Table 2.

Fig. 5.

Comparison of ON and OFF period firing rates. A and B: scatterplots showing the firing rates of individual neurons in response to the best ON frequency and the best OFF frequency in A1 (A) and CL (B). C and D: scatterplots showing the firing rates of individual neurons in response to the best ON location and the best OFF location in A1 (C) and CL (D). The unity lines (gray, dashed) are shown for reference. In both A1 and CL, for both tone and noise stimuli, a majority of neurons exhibit a lower firing rate in response to the best OFF stimulus compared with the best ON stimulus. In all cases, ON- and OFF-evoked firing rates were significantly positively correlated.

Table 2.

Best ON and OFF stimulus evoked firing rates of A1 and CL neurons

	ON Mean	ON SD	ON Median	ON Interquartile Range	OFF Mean	OFF SD	OFF Median	OFF Interquartile Range
A1–tone	66.24	49.92	54.55	60.00	37.90	32.93	28.00	36.33
CL–tone	61.68	41.40	54.77	53.49	39.15	29.39	34.85	40.00
A1–noise	51.24	42.46	38.33	49.34	27.55	23.44	20.00	25.74
CL–noise	47.42	28.99	38.74	38.13	28.46	19.34	23.20	21.88

Spectral tuning properties of ON and OFF responses.

We measured the spectral tuning bandwidths and best frequencies separately for ON- and OFF-evoked responses across our sample of single neurons in A1 and CL. Figure 6A shows the cumulative distributions of ON and OFF tuning bandwidths in A1. The OFF tuning bandwidths were right shifted relative to the ON tuning bandwidths, showing that OFF responses were more broadly tuned than ON responses in A1 (mean: ON 1.50 octaves, OFF 1.72 octaves; median: ON 1.36 octaves, OFF 1.66 octaves; P = 0.009, Kolmogorov-Smirnov test; P = 0.014, Wilcoxon rank-sum test). A similar trend was seen in CL (for ON and OFF bandwidths), although this was not statistically significant (mean: ON 1.54 octaves, OFF 1.75 octaves; median: ON 1.56 octaves, OFF 1.68 octaves; P = 0.117, Kolmogorov-Smirnov test; P = 0.145, Wilcoxon rank-sum test). Figure 6B shows the difference between the ON and OFF best frequencies for A1 and CL. This analysis revealed that ON and OFF best frequencies differed by ∼1 octave, on average, in A1 (mean: 1.1 octaves; median: 0.8 octave) and CL (mean: 1.2 octaves; median: 0.8 octave). These distributions of best frequency difference in A1 and CL were not significantly different.

Fig. 6.

Spectral tuning characteristics of ON and OFF receptive field subregions of A1 and CL neurons. A: bandwidth (full-width at half-height) of onset period and offset period tuning curves. B: best stimulus difference (absolute difference in ON and OFF best frequency). C and D: scatterplots showing the relationship between spectral tuning bandwidths of onset period and offset period tuning curves for individual neurons in A1 (C) and CL (D). E and F: pseudocolor plot showing the relationship between ON and OFF best frequencies in A1 (E) and CL (F). Warmer colors indicate higher numbers of neurons. G and H: pseudocolor plot of best frequency difference for neurons with different bandwidths in A1 (G) and CL (H). Warmer colors indicate higher numbers of neurons. Spectral separation between ON and OFF tuning curve peaks tended to be small relative to the width of the tuning curves.

We next examined the relationship between tuning bandwidth of ON and OFF subregions of the spectral receptive fields of individual neurons. Scatterplots of the ON and OFF tuning bandwidth are shown for A1 (Fig. 6C) and CL (Fig. 6D). ON and OFF spectral tuning bandwidths displayed only a weak positive correlation, with a slight tendency toward larger OFF bandwidths compared with ON bandwidths in A1 (53.6% OFF > ON, 45.4% OFF < ON, 1% OFF = ON; r = 0.142, P = 0.041) and CL (54.2% OFF > ON, 44.1% OFF < ON, 1.7% OFF = ON; r = 0.286, P = 0.002).

One possibility is that ON and OFF responses differ only in tuning bandwidth; however, it is also possible that the best frequencies could be the same or systematically different. We therefore compared the best frequencies of ON and OFF responses of individual neurons in A1 (Fig. 6E) and CL (Fig. 6F). We found that OFF best frequencies were not systematically shifted toward either higher or lower values relative to ON best frequencies in either A1 neurons, which showed a weak positive correlation between ON and OFF best frequencies (36.7% OFF > ON, 39.6% OFF < ON, 23.7% OFF = ON; r = 0.167, P = 0.016), or CL neurons, which showed a trend toward weak positive correlation between ON and OFF best frequencies (38.1% OFF > ON, 34.8% OFF < ON, 27.1% OFF = ON; r = 0.165, P = 0.073). Finally, if ON and OFF spectral tuning profiles were completely segregated, then we would expect that separation between the peaks of the tuning profiles (best frequency difference) would be large relative to the sum of ON and OFF half-bandwidths (half-width at half-height). However, for individual neurons in both A1 and CL, best frequency difference was low relative to the sum of half-bandwidths of ON and OFF tuning curves (Fig. 6, G and H), suggesting that although ON and OFF best frequencies may differ, there is some overlap in the range of frequencies that elicit ON and OFF responses in the same neurons.

Spatial tuning properties of ON and OFF responses.

We performed a similar analysis on the spatial tuning bandwidths and best locations separately for ON- and OFF-evoked responses of single neurons in A1 and CL. Spatial tuning bandwidths (Fig. 7A) displayed the opposite trend to that seen for spectral tuning; OFF tuning bandwidths tended to be more narrowly tuned than ON tuning bandwidths. This was significant in A1 (mean: ON 151.25°, OFF 122.15°; median: ON 157.50°, OFF 123.75°; P < 0.001, Kolmogorov-Smirnov test; P < 0.001, Wilcoxon rank-sum test) but not in CL (mean: ON 114.03°, OFF 111.99°; median: ON 112.50°, OFF 112.50°; P = 0.595, Kolmogorov-Smirnov test; P = 0.792, Wilcoxon rank-sum test). In fact, the distribution of A1 OFF tuning bandwidths was very similar to distributions of ON or OFF tuning in CL (P = 0.36, Kruskal-Wallis test), which were, as expected, more narrowly tuned than A1 ON responses (P < 0.001, Kruskal-Wallis test). In A1, mean best location difference was 58.6°, and median best location difference was 45° (Fig. 7B). Best location differences tended to be smaller in CL, with a mean best location difference of 44° and median best location difference of 22.5°; however, the difference in the distributions of best location difference in A1 and CL did not reach significance (P = 0.224, Kolmogorov-Smirnov test; P = 0.063, Wilcoxon rank-sum test). ON and OFF spatial tuning bandwidths displayed a slight trend toward weak positive correlation in A1, where more neurons had narrow OFF spatial tuning bandwidths compared with ON spatial tuning bandwidths (Fig. 7C; 25.3% OFF > ON, 66.7% OFF < ON, 8% OFF = ON; r = 0.176, P = 0.082). In CL, ON and OFF spatial tuning bandwidths were weakly, positively correlated, where OFF spatial tuning bandwidths tended to be narrower than ON spatial tuning bandwidths in many neurons, although these constituted a smaller proportion in CL than in A1 (Fig. 7D; 38.6% OFF > ON, 46.6% OFF < ON, 14.8% OFF = ON; r = 0.294, P = 0.006). ON and OFF best locations were not correlated in A1 (Fig. 7E; circular correlation coefficient ρ = −0.036, P = 0.736; 31% OFF > ON, 37.1% OFF < ON, 31.9% OFF = ON) but positively correlated in CL (Fig. 7F; ρ = 0.441, P < 0.001; 24.7% OFF > ON, 36.2% OFF < ON, 39.1% OFF = ON). Best location differences in A1 and CL (Fig. 7, G and H) were small relative to the sum of ON and OFF spatial tuning half-bandwidths of the corresponding neurons, suggesting overlap in the range of spatial locations eliciting ON and OFF responses, although best stimuli may differ between the two.

Fig. 7.

Spatial tuning characteristics of ON and OFF receptive field subregions of A1 and CL neurons. A: bandwidth (full-width at half-height) of onset period and offset period tuning curves. B: best stimulus difference (absolute difference in ON and OFF best location). C and D: scatterplots showing the relationship between spatial tuning bandwidths of onset period and offset period tuning curves for individual neurons in A1 (C) and CL (D). E and F: pseudocolor plot showing the relationship between ON and OFF best locations in A1 (E) and CL (F). Warmer colors indicate higher numbers of neurons. G and H: pseudocolor plot of best location difference for neurons with different bandwidths in A1 (G) and CL (H). Warmer colors indicate higher numbers of neurons. Spatial separation between ON and OFF tuning curve peaks tended to be small relative to the width of the tuning curves.

Relationship between ON and OFF spectral tuning.

We characterized the extent of spectral segregation of ON and OFF responses to tones in individual neurons using seven continuous variables (5 metrics of similarity and 2 metrics of dissimilarity; see materials and methods). Metrics of dissimilarity were multiplied by −1. Therefore, for all metrics, the lowest values indicate highly nonoverlapping ON and OFF subregions within the receptive field, and the highest values indicate highly overlapping ON and OFF subregions within the receptive fields (see Fig. 1, B and C). Metrics of ON/OFF spectral tuning overlap for the example neurons shown in Fig. 2, A–D, are provided in Table 3. Means, SDs, medians, and interquartile ranges for neurons in A1 and CL are provided in Tables 4 and 5, respectively, for all metrics. Linear correlation coefficients for each pair of metrics are provided in Table 6.

Table 3.

Spectral tuning overlap of ON and OFF responses for example neurons in A1 and CL

	A1		CL
	A	B	C	D
Tuning overlap fraction	0.46	0.33	0.68	0.33
Cumulative profile distance	0.17	0.33	−0.090	−0.49
Euclidean distance	−0.63	−1.1	−0.42	−1.1
Pearson's correlation	0.70	−0.28	0.60	−0.16
Spearman's correlation	−0.79	−0.26	0.55	−0.069
Relative OFF response magnitude	0.60	0.20	0.70	0.21
Relative ON response magnitude	0.71	0.00	0.83	0.083

Values in columns A–D correspond to the example neurons shown in Fig. 2, A–D, respectively.

Table 4.

Summary data for spectral tuning overlap of ON and OFF responses in primary auditory cortex (A1)

	Mean	SD	Median	Interquartile Range
Tuning overlap fraction	0.38	0.15	0.39	0.18
Cumulative profile distance	−0.28	0.15	−0.24	0.15
Euclidean distance	−0.83	0.23	−0.81	0.28
Pearson's correlation	0.18	0.36	0.19	0.55
Spearman's correlation	0.11	0.32	0.14	0.46
Relative OFF response magnitude	0.52	0.36	0.48	0.68
Relative ON response magnitude	0.51	0.38	0.43	0.80

Table 5.

Summary data for spectral tuning overlap of ON and OFF responses in the caudolateral belt area (CL)

	Mean	SD	Median	Interquartile Range
Tuning overlap fraction	0.44	0.13	0.46	0.16
Cumulative profile distance	−0.25	0.13	−0.22	0.17
Euclidean distance	−0.75	0.22	−0.73	0.29
Pearson's correlation	0.28	0.33	0.29	0.55
Spearman's correlation	0.20	0.29	0.19	0.41
Relative OFF response magnitude	0.58	0.38	0.65	0.75
Relative ON response magnitude	0.58	0.35	0.59	0.75

Table 6.

Correlation coefficients of ON/OFF spectral tuning similarity measures for neurons in the primary auditory cortex (A1) and the caudolateral belt area (CL)

A1
CL	Tuning Overlap Fraction	Cumulative Profile Distance	Euclidean Distance	Pearson's Correlation	Spearman's Correlation	Relative OFF Response Magnitude	Relative ON Response Magnitude
Tuning overlap fraction		0.79	1.0	0.69	0.59	0.61	0.63
Cumulative profile distance	0.79		0.78	0.53	0.46	0.51	0.45
Euclidean distance	0.95	0.78		0.80	0.66	0.68	0.70
Pearson's correlation	0.71	0.53	0.83		0.85	0.67	0.72
Spearman's correlation	0.57	0.46	0.64	0.78		0.47	0.49
Relative OFF response magnitude	0.60	0.42	0.69	0.66	0.32		0.57
Relative ON response magnitude	0.67	0.49	0.74	0.74	0.41	0.63

Values for neurons in A1 are shown above the main diagonal, and values for neurons in CL are shown below the main diagonal. P < 0.001 in all cases.

The metrics of tuning overlap fraction, cumulative profile distance, Euclidean distance, Pearson's correlation, and Spearman's correlation all displayed unimodal distributions, which were statistically verified (Fig. 8; Hartigan's dip test statistic, P > 0.05). This was true for both A1 and CL neurons. However, the distribution of relative OFF response magnitude appeared more bimodal (Fig. 8). The nonunimodality of this distribution was statistically significant for both A1 (Hartigan's dip test statistic, P < 0.001) and CL (Hartigan's dip test statistic, P = 0.007). Finally, of all seven metrics, relative ON response magnitude displayed the most clearly bimodal distribution for the population of neurons (Fig. 8) in both A1 (Hartigan's dip test statistic, P < 0.001) and CL (Hartigan's dip test statistic, P = 0.003).

Fig. 8.

Distribution of the metrics of spectral tuning overlap of ON and OFF subregions of neurons in A1 (red) and CL (blue). On the main diagonal, histograms are shown for tuning overlap fraction, cumulative profile distance, Euclidean distance, Pearson's correlation, Spearman's correlation, relative OFF response magnitude, and relative ON response magnitude. The relationship between each pair of variables is shown in scatterplots off of the main diagonal. All metrics were significantly, positively correlated (P < 0.05) to various extents (correlation coefficients listed in Table 6). Only the last 2 variables—relative OFF response magnitude and relative ON response magnitude—appeared bimodal; the nonunimodality of these 2 metrics was statistically significant for both A1 and CL (Hartigan's dip test statistic for nonunimodality, P < 0.05). Across all variables combined, there was a small but significant difference in the distributions of spectral tuning overlap between neurons in A1 and CL (nonparametric MANOVA, P = 0.010), with A1 possessing a slightly larger fraction of neurons with dissimilar tuning of ON and OFF responses relative to CL.

Thus in both A1 and CL, the metrics of ON and OFF spectral tuning similarity, which were computed based on all data points on the respective tuning functions (tuning overlap fraction, cumulative profile distance, Euclidean distance, Pearson's correlation, and Spearman's correlation), displayed a continuum of values ranging from low to high. Thus there was considerable overlap in the range of frequencies that evoked tone ON and OFF responses in auditory cortex neurons. However, when only the responses evoked by the best OFF frequency and best ON frequency were taken into consideration (relative OFF response magnitude and relative ON response magnitude), A1 and CL neurons could be divided into two categories: units with similar responses to ON and OFF best stimuli and those with disparate responses to ON and OFF best stimuli.

For all computed tuning metrics, there was a large overlap in the distributions of values for A1 and CL units. However, for some metrics (particularly Euclidean distance, Pearson's correlation, relative OFF response magnitude, and relative ON response magnitude), A1 and CL distributions appeared to be skewed in opposite directions.

To assess the extent of the difference in A1 compared with CL, we first used the relative ON response magnitude to split the range of tuning similarity values observed, since this was the metric that showed the most bimodal distribution. Neurons with relative ON response magnitude ≥0.5 were categorized as having similar ON and OFF tuning, whereas those with relative ON response magnitude <0.5 were categorized as having dissimilar ON and OFF tuning. With the use of this criterion, in A1, 54.6% (113/207) neurons displayed dissimilar tuning of ON and OFF responses, whereas 45.4% (94/207) neurons displayed similar tuning. In CL, 44.9% (53/118) displayed dissimilar tuning, and 55.1% (65/118) of neurons displayed similar tuning of ON and OFF responses. Thus there was a modest difference in the proportions of neurons with similar or different ON and OFF tuning between A1 and CL when these classifications were made based on the metric of relative ON response magnitude (P = 0.107, Fisher's exact test). For all metrics, CL distributions had higher mean and median values (Table 5) compared with A1 distributions (Table 4), suggesting a shift toward higher ON/OFF spectral tuning overlap. We tested the dataset with a nonparametric MANOVA to assess whether there were differences in ON and OFF tuning relationships between A1 and CL populations across all measured variables and found that there was a statistically significant difference (P = 0.010, nonparametric, permutation-based MANOVA, 1,000 permutations).

To verify that our results were not driven by the choice of the analysis window used to measure responses, we repeated the analysis using a time window of 0–100 ms following stimulus onset and offset to measure ON and OFF responses. We performed a nonparametric MANOVA to compare the distributions of spectral overlap that resulted from this approach compared with the distributions reported here. In A1, this analysis resulted in P = 0.05, which just missed the criterion for significance (P < 0.05); however, none of the observed effects (unimodality/bimodality of distributions; greater overlap in CL vs. A1 populations) were altered by any differences in the distributions caused by using these different time windows.

In CL, the distributions using either time window produced distributions that were not significantly different (P = 0.51, nonparametric, permutation-based MANOVA, 1,000 permutations).

Relationship between ON and OFF spatial tuning.

We performed a similar analysis of ON and OFF response spatial tuning for A1 and CL neurons in response to noise stimuli, using equivalent metrics to those described above for spectral tuning (see materials and methods). Measures of ON/OFF spatial tuning overlap for the example neurons shown in Fig. 3, A–D, are provided in Table 7. Summary data are provided in Tables 8 and 9. Linear correlation coefficients between tuning similarity metrics are listed in Table 10. In the case of spatial tuning, distributions of all seven metrics were unimodally distributed for A1 as well as CL, which we verified statistically (Fig. 9; Hartigan's dip test statistic, P > 0.05, in all cases).

Table 7.

Spatial tuning overlap of ON and OFF responses for example neurons in A1 and CL

	A1		CL
	A	B	C	D
Tuning overlap fraction	0.6980	0.4476	0.5603	0.2837
Cumulative profile distance	−0.1208	−0.3154	−0.0821	−0.4982
Euclidean distance	−0.3076	−0.7591	−0.4348	−1.025
Pearson's correlation	0.8899	0.2779	0.8324	−0.3469
Spearman's correlation	0.8206	0.1765	0.8294	−0.2971
Relative OFF response magnitude	0.8499	0.8269	1.000	0.2531
Relative ON response magnitude	0.8730	0.000	1.000	0.000

Values in columns A–D correspond to the example neurons shown in Fig. 3, A–D, respectively.

Table 8.

Summary data for spatial tuning overlap of ON and OFF responses in primary auditory cortex (A1)

	Mean	SD	Median	Interquartile Range
Tuning overlap fraction	0.55	0.10	0.56	0.17
Cumulative profile distance	−0.18	0.09	−0.16	0.13
Euclidean distance	−0.57	0.16	−0.56	0.24
Pearson's correlation	0.44	0.33	0.54	0.47
Spearman's correlation	0.40	0.33	0.50	0.47
Relative OFF response magnitude	0.67	0.29	0.67	0.49
Relative ON response magnitude	0.71	0.29	0.80	0.45

Table 9.

Summary data for spatial tuning overlap of ON and OFF responses in the caudolateral belt area (CL)

	Mean	SD	Median	Interquartile Range
Tuning overlap fraction	0.58	0.13	0.60	0.16
Cumulative profile distance	−0.16	0.10	−0.15	0.081
Euclidean distance	−0.52	0.20	−0.48	0.22
Pearson's correlation	0.58	0.36	0.70	0.31
Spearman's correlation	0.52	0.36	0.62	0.45
Relative OFF response magnitude	0.72	0.28	0.80	0.44
Relative ON response magnitude	0.74	0.28	0.83	0.43

Table 10.

Correlation coefficients of ON/OFF spatial tuning similarity measures for neurons in the primary auditory cortex (A1) and the caudolateral belt area (CL)

A1
CL	Tuning Overlap Fraction	Cumulative Profile Distance	Euclidean Distance	Pearson's Correlation	Spearman's Correlation	Relative ON Response Magnitude	Relative OFF Response Magnitude
Tuning overlap fraction		0.65	0.93	0.79	0.78	0.56	0.50
Cumulative profile distance	0.76		0.72	0.67	0.66	0.44	0.46
Euclidean distance	0.96	0.81		0.93	0.90	0.64	0.63
Pearson's correlation	0.85	0.74	0.94		0.96	0.67	0.64
Spearman's correlation	0.84	0.76	0.92	0.96		0.58	0.57
Relative OFF response magnitude	0.63	0.54	0.72	0.73	0.67		0.37
Relative ON response magnitude	0.73	0.64	0.80	0.83	0.75	0.71

Values for neurons in A1 are shown above the main diagonal, and values for neurons in CL are shown below the main diagonal. P < 0.001 in all cases.

Fig. 9.

Distribution of the metrics of spatial tuning overlap of ON and OFF subregions of neurons in A1 (red) and CL (blue). On the main diagonal, histograms are shown for tuning overlap fraction, cumulative profile distance, Euclidean distance, Pearson's correlation, Spearman's correlation, relative OFF response magnitude, and relative ON response magnitude. The relationship between each pair of variables is shown in scatterplots off of the main diagonal. All metrics were significantly, positively correlated (P < 0.05) to various extents (correlation coefficients listed in Table 10). None of the variables displayed bimodal distributions (Hartigan's dip test statistic for nonunimodality, P > 0.05, in all cases). Across all variables combined, there was a small but significant difference in the distributions of spatial tuning overlap between neurons in A1 and CL (nonparametric MANOVA, P = 0.018), with A1 possessing a slightly larger fraction of neurons with dissimilar tuning of ON and OFF responses relative to CL.

Similar to spectral tuning relationships, in both A1 and CL neurons, the metrics of ON and OFF spatial tuning similarity, which were computed based on all data points on the respective tuning functions (tuning overlap fraction, cumulative profile distance, Euclidean distance, Pearson's correlation, and Spearman's correlation), displayed a continuous range of values, corresponding to low-to-high spatial overlap in ON/OFF receptive field subregions. Furthermore, unlike spectral tuning relationships, the distributions of remaining two metrics (relative OFF response magnitude and relative ON response magnitude) were also highly unimodal and did not provide any indication of distinct categories of neurons based on spatial tuning relationships of ON and OFF responses. All seven distributions were skewed toward high values of ON/OFF similarity metrics for neurons in both areas, indicating a high level of overlap in the tuning of ON and OFF subregions in the spatial receptive fields of A1 and CL neurons.

When neurons were classified based on relative ON response magnitude (a metric that was bimodally distributed in the frequency domain), there was no difference in the proportions of neurons with similar or different ON/OFF spatial tuning between A1 and CL (P = 1.0, Fisher's exact test), even to the modest extent observed for spectral tuning. In A1, 21.2% (21/99) neurons displayed dissimilar tuning (relative ON response magnitude <0.5) of ON and OFF responses, whereas 78.8% (78/99) neurons displayed similar tuning (relative ON response magnitude ≥0.5). In CL, nearly identical distributions were noted, where 21.6% of neurons (19/88) displayed dissimilar tuning, and 78.4% (69/88) of neurons displayed similar tuning of ON and OFF responses.

However, as in the case of spectral tuning, mean and median values of CL distributions for spatial tuning overlap (Table 9) were higher than those of A1 distributions (Table 8). We tested the dataset with a nonparametric MANOVA to check for differences in ON and OFF tuning relationships between A1 and CL populations, across all measured variables, resulting in P = 0.018 (nonparametric, permutation-based MANOVA, 1,000 permutations). This suggests that there is a small but significant shift toward higher ON/OFF spatial tuning overlap in CL compared with A1. When the analysis was repeated using a time window of 0–100 ms, following stimulus onset and offset to measure ON and OFF responses, there were no significant differences resulting from this approach compared with the distributions of spatial tuning overlap reported here in either A1 or CL (A1: P = 0.99; CL: P = 0.14, nonparametric, permutation-based MANOVA, 1,000 permutations).

Relationship between OFF responsiveness and tuning overlap.

As in a majority of previous studies that have compared ON and OFF response tuning in auditory cortex (see discussion), all neurons that exhibited significant responses relative to baseline, during both ON and OFF response periods, were included in the analyses. However, we also analyzed the data after applying stricter criteria for the inclusion of OFF responses, based on the quantification of the strength of OFF responses relative to sustained period activity using the OFF/SUS ratio. This was important to verify that the results were not dominated by the inclusion of long-latency responses or slow, decaying, sustained activity rather than true OFF responses. When the dataset was restricted to only the neurons that exhibit OFF/SUS ratios >1, the distributions of all seven overlap metrics still displayed shapes similar to those found when all neurons were included. For spectral tuning (A1: 152/207 cells, CL: 79/118) in A1 and CL, all tuning metrics were unimodally distributed except for relative OFF response magnitude and relative ON response magnitude (Hartigan's dip test statistic for nonunimodality, P < 0.01, in all cases). For spatial tuning (A1: 36/99 cells, CL: 58/88) in A1 and CL, all seven tuning metrics were unimodally distributed (Hartigan's dip test statistic for nonunimodality, P > 0.05, in all cases).

In all cases, OFF/SUS ratios formed unimodal, continuous distributions of values. We examined the relationship between the strength of OFF responsiveness, as assessed using the OFF/SUS ratio, and the tuning overlap metrics described above. When correlations with overlap metrics were present, they were weak (correlation coefficients = −0.2 to 0.2). Results for comparisons with two of the metrics—relative OFF response magnitude and relative ON response magnitude—are shown for all A1 and CL neurons in the spectral domain (Fig. 10A) and the spatial domain (Fig. 10B) that displayed significant ON and OFF responses relative to baseline. In the spectral domain, A1 neurons displayed no significant correlation of the OFF/SUS ratio with either overlap metric (A1, relative OFF response magnitude: r = −0.005, P = 0.94; relative ON response magnitude: r = −0.075, P = 0.29), whereas CL neurons displayed a weak, positive correlation of the OFF/SUS ratio with both overlap metrics (CL, relative OFF response magnitude: r = −0.248, P = 0.007; relative ON response magnitude: r = −0.267, P = 0.004). In the spatial domain, correlations between the OFF/SUS ratio and tuning overlap were not significant in any case (A1, relative OFF response magnitude: r = 0.116, P = 0.25, relative ON response magnitude: r = 0.099, P = 0.33; CL, relative OFF response magnitude: r = 0.074, P = 0.50, relative ON response magnitude: r −0.1087, P = 0.31). Given these results, it is unlikely that the observed results were driven by the inclusion of long-latency responses or slow, decaying, sustained activity rather than true OFF responses.

Fig. 10.

Relationship between OFF responsiveness and tuning overlap indices in A1 (red circles) and CL (blue crosses). A: scatterplots showing the relationship between the OFF/SUS ratio and relative OFF response magnitude (top) and relative ON response magnitude (bottom) in the spectral domain. For A1 neurons, the OFF/SUS ratio was not significantly correlated with either metric. For CL neurons, the OFF/SUS ratio shows a weak, negative correlation with both metrics. Marginal histograms show the distributions of each individual quantity. B: scatterplots showing the relationship between the OFF/SUS ratio and relative OFF response magnitude (top) and relative ON response magnitude (bottom) in the spatial domain. No significant correlations were present between the OFF/SUS ratio and either of these 2 tuning overlap metrics. Marginal histograms show the distributions of each individual quantity.

DISCUSSION

In the current study, we examined ON and OFF responses of neurons in the rhesus macaque auditory cortex. To our knowledge, this is the first study to examine ON and OFF tuning of neurons across core and belt auditory cortical fields of primates. Furthermore, unlike previous studies in primates, we compared ON and OFF spatial tuning, in addition to spectral tuning. We found that the majority of neurons in both A1 and CL responded to both the onset and offset of tone and noise stimuli. We compared the tuning profiles of ON and OFF receptive field subregions for neurons in A1 and CL. Across both A1 and CL, we found that the relationship between ON and OFF response profiles ranged from highly similar to highly dissimilar for both spectral and spatial tuning. In the spectral domain, when ignoring all responses other than those to the best ON and OFF frequencies, bimodal distributions were seen, suggesting the segregation of ON and OFF receptive field subregions by this metric of the peak response. However, distributions of similarity metrics taking into account responses to all tested stimuli were unimodal and skewed toward values indicating higher ON/OFF overlap. Thus ON and OFF best stimuli were distinct for a substantial proportion of neurons, although there was a fair amount of overlap in the respective tuning profiles, even in those cases. By contrast, in the spatial domain, distributions of all computed similarity metrics exhibited highly unimodal distributions, indicating similar ON and OFF best stimuli and highly overlapping ON and OFF tuning profiles. Overall distributions of similarity metrics were comparable in A1 and CL, although there was a small but significant shift toward more similar ON and OFF receptive field subregions in CL compared with A1.

Prevalence of OFF responses in auditory cortex.

As in previous studies, which have reported low OFF response firing rates relative to ON response firing rates in A1 (Engle and Recanzone 2013; Recanzone 2000a; Tian et al. 2013) and CL (Engle and Recanzone 2013), we observed that for noise and tone stimuli, even the best OFF stimulus elicited firing rates lower than the best ON stimulus in both A1 and CL. Almost all OFF-responsive neurons were also ON responsive. These data are consistent with neurophysiological and psychophysical evidence for the more dominant representation of stimulus onset compared with stimulus offset (Malone et al. 2015; Phillips et al. 2002; Qin et al. 2007, Scholl et al. 2010).

However, although they had lower magnitudes than ON responses, OFF responses were highly prevalent in our sample. As might be expected, our estimates of OFF-responsive neurons exceeded those from previous reports in auditory cortex of anesthetized animals [bats, 4–16% (Ostwald 1984); cat, 43% (Volkov and Galazjuk 1991)–28% (Moshitch et al. 2006); ferret, 25% (Hartley et al. 2011)]. Additionally, we found that OFF responses were displayed by a majority of auditory cortex neurons; our estimates exceeded those of previous reports in awake cats [59% (Qin et al. 2007)] and monkeys [30% (Recanzone 2000a) and 11% (Pfingst and O'Connor 1981)]; however, our results were in line with findings by Tian et al. (2013), performed in awake rhesus monkeys (>90% of A1 neurons displayed an OFF response).

A previous study from our laboratory (Recanzone 2000a) reported that ∼35% of A1 neurons in awake monkeys exhibit an OFF response to noise and tone stimuli. However, in that study, OFF responses were measured by presenting tone stimuli close to the best frequency, as determined from the ON response. As we found in the current study, ON and OFF best frequencies often differ for A1 neurons. In addition, noise stimuli were always presented from the 90° contralateral location and not the OFF best location. Thus lower percentages of OFF responses in previous studies on auditory cortex could be due to the use of nonoptimal stimuli that did not take into account OFF response spectral and spatial tuning.

Discrepancies in percentages of OFF responses might also be due to differences in the definition of OFF responses [see also Recanzone (2000a)]. Tian et al. (2013) reported that >90% of neurons in A1 of awake rhesus monkeys displayed an OFF response and ascribed this high proportion to their use of bandpass noise stimuli rather than tones. Additionally, they reported that even a small increase in the bandwidth of bandpass noise used as the stimulus led to a significant increase in OFF responses. In the current study, we did not examine the effect of tone vs. noise stimuli on ON and OFF firing rates of individual neurons. However, we observed that both tone and broad-band noise stimuli evoked an OFF response in the majority of neurons, in both A1 and CL. It is possible that the different methods used in the two studies to determine a significant response in our study [current study, threshold based on SDs compared with Tian et al. (2013), nonparametric statistical hypothesis test for matched pairs] could result in differences in proportions of ON and OFF responses detected. However, we also tested our data using the same method as Tian et al. (2013) but did not see any major differences in the proportions of “ON only,” “OFF only,” and “ON + OFF” neurons (see materials and methods for further details).

A majority of studies that have compared ON and OFF response tuning in auditory cortex (Fishman and Steinschneider 2009; Hartley et al. 2011; Qin 2007; Tian et al. 2013) have evaluated OFF response firing rates in a time window following stimulus offset that were significant relative to baseline. Other studies of OFF responses in subcortical structures have used qualitative methods, sometimes in combination with significance relative to baseline, to identify cells with true OFF responses (Kasai 2012; Lesser et al. 1990). Some studies (He 2001, 2002; Scholl et al. 2010) varied stimulus duration to distinguish true OFF responses from long-latency responses or slow, decaying, sustained activity; however, in our study, stimulus durations were fixed, so we could not adopt this method. Very few studies (Anderson and Linden 2016; Recanzone 2000a) have used quantitative criteria to identify true OFF responses relative to sustained activity. We have incorporated an index of OFF responsiveness (OFF/SUS), calculated as the absolute value of the ratio of offset period activity to sustained period activity (101–150 ms) after subtracting baseline activity. This is a slight modification of the ratio used by Recanzone (2000a) for identifying excitatory OFF responses. To be able to compare results with previous studies, we have presented data for all neurons that fulfilled the condition of significance relative to baseline; however, for key findings, we have reported statistics separately after restricting the dataset to neurons with OFF/SUS > 1. For these analyses that were repeated with the restricted dataset, we found that distributions of ON/OFF tuning overlap metrics were similar to those obtained for the entire population of A1 and CL neurons.

Tuning of OFF vs. ON responses.

Among previous studies that have reported responses to stimulus offset in A1, only a few have characterized tuning of OFF responses, and these have mostly focused on tuning in the spectral domain. Scholl et al. (2010) characterized OFF responses in A1 of anesthetized rats and reported that frequency response areas for ON and OFF responses were predominantly nonoverlapping. Furthermore, OFF response characteristic frequencies were, on average, shifted 1–2 octaves higher than corresponding ON response characteristic frequencies. Tian et al. (2013) reported differences between ON and OFF characteristic frequencies of ∼1.3 octaves for type-S cells but no difference between the two for type-C cells in macaque A1. In our experiments, we found that the mean difference between ON and OFF best frequencies was ∼1 octave in both A1 and CL of macaque monkeys; however, OFF tuning was not systematically shifted in the direction of either higher or lower sound frequencies. Fishman and Steinschneider (2009) examined frequency response functions of multiunit activity in awake macaque monkeys and found that on average, OFF best frequencies differed from ON best frequencies by 0.88 octave and could deviate by as much as 4 octaves. Shifts occurred to similar extents toward higher as well as lower frequencies relative to ON best frequencies. The lack of consistency in the relationship between ON and OFF response tuning was also reported by Qin et al. (2007) in awake cats. These authors reported differences between ON and OFF spectral receptive fields (>1 octaves, in many cases) in >50% of cat A1 neurons.

We found that OFF responses were, on average, more broadly spectrally tuned than ON responses in A1, with a similar trend in CL, although mean bandwidths of both and ON and OFF responses were within the 1- to 2-octave range. This is consistent with previous studies that have found that spectral tuning of neurons in A1 is most narrow at early times following the stimulus and tends to get broader at later times [anesthetized cat, Moshitch et al. (2006)]. Tian et al. (2013) reported that type-C cells were more broadly tuned than type-S cells when ON responses of the two groups were compared; however, they did not directly compare OFF and ON response bandwidths.

Studies on tuning of OFF responses with respect to auditory space are even more limited. In anesthetized ferret A1, Hartley et al. (2011) found that ON and OFF responses displayed complementary tuning for interaural timing differences, as well as interaural level differences, in almost all cells exhibiting both ON and OFF binaural sensitivity. However, these neurons formed a relatively small fraction (22%) of their entire sample. Our measurements of spatial tuning of macaque A1 and CL neurons showed that the median best location difference in A1 was 45° in A1 and 22.5° in CL, which, respectively, corresponds to two and one speaker locations apart in our experimental apparatus (see materials and methods). It is possible that testing with a higher resolution in stimulus locations could reveal a higher percentage of more subtle differences in the best location of ON vs. OFF responses. Regardless, it is clear that there is a high level of overlap in ON and OFF spatial tuning, both of which had mean bandwidths >100°. Unlike the case for spectral tuning, OFF responses were more narrowly tuned for space compared with ON responses.

Classification of auditory cortex cells.

In visual cortex of monkeys and cats, cells with ON and OFF responses have been classified as simple and complex cells, most often based on the criteria originally proposed by Hubel and Wiesel (1962), including spatial segregation of ON and OFF receptive field subregions (Hubel and Wiesel 1962, 1968; Martinez et al. 2005; Schiller et al. 1976), based on response linearity to drifting gratings (Cumming and Parker 1999; Movshon et al. 1978) or a combination both techniques [Dean and Tolhurst (1983); Skottun et al. (1991); see Mechler and Ringach (2002) for review]. A number of these studies have reported bimodal distributions of spatial-overlap metrics (Hubel and Wiesel 1962, 1968; Martinez et al. 2005), as well the response modulation ratio [Priebe et al. (2004); see Chen et al. (2009) for review]. Dean and Tolhurst (1983) found that although an individual metric of overlap may not be bimodally distributed, a combination of metrics plotted against each other could reveal clusters of neurons that could be used to classify neurons. Other studies have argued that simple and complex cells are not discrete classes but rather, represent two ends of a continuum, despite the fact that some metrics of overlap display bimodality (Mata and Ringach 2005).

In A1, Tian et al. (2013) proposed that neurons could be divided into two classes—type-S and type-C cells—analogous to simple and complex cells in visual cortex, based on a bimodal distribution of metrics quantifying the overlap of ON/OFF subregions of spectral receptive fields in A1. In our sample, we also observed many neurons that had clearly nonoverlapping and clearly highly overlapping ON and OFF spectral receptive fields (see Figs. 2 and 3). Two metrics that we investigated were bimodally distributed for spectral tuning in both A1 and CL. Both of these metrics took into consideration only the responses at ON and OFF best frequencies rather than the entire response profile. However, in our study, we found that distributions of ON/OFF similarity metrics for spectral tuning of A1 and CL neurons (see materials and methods) were mostly unimodal (see Technical considerations below). Although there were several neurons that displayed different ON and OFF best frequencies, consistent with the type-S cells of Tian et al. (2013), many of these exhibited substantial overlap between the ON and OFF subregions of the receptive fields. Thus the relationship between ON and OFF spectral receptive field subregions ranged from highly nonoverlapping to highly overlapping, with no clear distinction between classes. This difference between V1 and A1 neuronal classes was also observed in a previous study by Ahmed et al. (2006). They examined the other criterion commonly used to classify visual cortex neurons: response linearity. That study evaluated the response linearity of ferret A1 neurons to spectral ripple stimuli (proposed to be the acoustic equivalent of drifting sine-wave gratings) and found that the response modulation ratios were unimodally distributed, unlike V1 neurons. Furthermore, relative proportions and distributions of linear vs. nonlinear neurons in ferret A1 did not parallel those seen in V1. Finally, for ON/ OFF response profiles in the spatial domain, the unimodality of overlap metrics was even more apparent. Although a small fraction of neurons displayed nonoverlapping receptive fields, for the most part, they were highly overlapping. Our data suggest that ON/OFF tuning relationships in A1 form a continuous range by most criteria, likely due to the much broader spectral and spatial tuning seen in auditory cortical neurons compared with the narrow spatial tuning seen in V1 neurons. The broadening of the tuning functions would increase the overlap between tuning functions with different best responses (either spectral or spatial), giving rise to the unimodal distributions that were most common in our present study.

Highly nonoverlapping ON/OFF receptive fields were more common for frequency tuning compared with spatial tuning, as evidenced by nonunimodal distributions for two of the overlap metrics considered in the spectral domain compared with strongly unimodal distributions across all overlap metrics in the spatial domain. This difference could be reflective of dissimilar roles of OFF responses in spectral processing compared with spatial processing and potential differences in the circuits underlying these two functions within the same cortical area.

Comparison of ON/OFF tuning in A1 and CL.

Our study is the first to characterize ON/OFF tuning relationships in a higher-order auditory cortical area of primates. We examined the level of segregation of ON and OFF responses in the CL field, in addition to A1. Overall, ON/OFF-overlap metrics were similarly distributed for A1 and CL neurons; however, there was a slight shift (∼10%) in the distribution of CL neurons toward values indicating high overlap. This increase was small compared with the increases in the proportions of complex cells from V1 to next higher-order visual area, V2. In V1, there is an approximately equal representation of simple and complex cells, whereas in V2, a large majority of neurons are complex cells (Levitt et al. 1994). Further investigation of ON/OFF tuning relationships in other auditory cortical fields in the belt is necessary to gain insight into whether proportions of ON/OFF overlap observed in our study, relative to A1, are a common feature of higher-order cortical areas.

Technical considerations.

Tian et al. (2013) proposed classification of macaque A1 neurons into type-S and type-C classes based on the bimodal distribution of quantitative metrics of ON/OFF subregion overlap. However, in our data, although we examined several of the same quantitative metrics, as well as a few metrics that were modified versions of metrics used by Tian et al. (2013) (see materials and methods), in most cases, we saw highly unimodal distributions for almost all metrics of ON/OFF segregation that we examined. Below, we address some potential technical reasons for the differences between the two studies.

Tian et al. (2013) reported that for type-S cells, OFF responses to tones were highly reduced compared with bandpass noise stimuli. As noted above, we observed a large proportion of OFF responses even with tone stimuli. OFF responses exhibited lower firing rates compared with ON responses for most neurons in our sample, but this was also true for bandpass noise stimuli used by Tian et al. (2013). In our study, we addressed this difference in firing rates by normalizing ON and OFF tuning curves to their respective peaks so that the shape of ON and OFF tuning curves and the extent of spectral or spatial overlap between them could be examined. Given that and since the use of bandpass noise stimuli does not seem to cause major alterations in the shape of the tuning curves (Tian et al. 2013), we do not expect that our conclusions, with regard to ON/OFF spectral tuning, are affected by the use of tone stimuli.

Another factor to consider is the sound level at which ON and OFF tuning response profiles were measured. In our experiments, ON/OFF tuning was measured at moderate sound levels, spectral tuning at 65 dB SPL, and spatial tuning at 55 and 75 dB SPL. The apparent level of segregation between ON and OFF tuning profiles could be higher if tested at lower intensities.

Differences could also be attributed to differences in laminar distribution of recording sites. Unfortunately, neither study is able to report the laminar distribution of the neurons that were studied, given the electrode approach and inherent difficulties of accurately discerning laminar location in cortical structures first encountered after many millimeters of electrode travel. Finally, although recordings were performed in awake animals in both studies, differences in receptive field structure could have resulted due to differences in attentional or arousal state or potentially, even in the ages and previous stimulation histories of the animals.

GRANTS

Funding for this work was provided by the National Institute on Aging (Grant R01AG034137) and National Institute on Deafness and Other Communication Disorders (Grant R01DC015232; both to G. H. Recanzone) and the National Eye Institute (Grant 2T32EY015387-11) and National Science Foundation Graduate Research Fellowship Program (Grant DGE-1148897; both to D. L. Ramamurthy).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

G.H.R. performed experiments; D.L.R. and G.H.R. analyzed data; D.L.R. and G.H.R. interpreted results of experiments; D.L.R. and G.H.R. prepared figures; D.L.R. and G.H.R. drafted manuscript; D.L.R. and G.H.R. edited and revised manuscript; D.L.R. and G.H.R. approved final version of manuscript.