Abstract
An organism's ability to detect and discriminate sensory inputs depends on the recent stimulus history. For example, perceptual detection thresholds for a brief tone can be elevated by as much as 50 dB when following a masking stimulus. Previous work suggests that such forward masking is not a direct result of peripheral neural adaptation; the central pathway apparently modifies the representation in a way that further attenuates the input's response to short probe signals. Here, we show that much of this transformation is complete by the level of the inferior colliculus (IC). Single-neuron extracellular responses were recorded in the central nucleus of the awake marmoset IC. The threshold for a 20 ms probe tone presented at best frequency was determined for various masker-probe delays, over a range of masker sound pressure levels (SPLs) and frequencies. The most striking aspect of the data was the increased potency of forward maskers as their SPL was increased, despite the fact that the excitatory response to the masker was often saturating or nonmonotonic over the same range of levels. This led to probe thresholds at high masker levels that were almost always higher than those observed in the auditory nerve. Probe threshold shifts were not usually caused by a persistent excitatory response to the masker; instead we propose a wide-dynamic-range inhibitory mechanism locked to sound offset as an explanation for several key aspects of the data. These findings further delineate the role of subcortical auditory processing in the generation of a context-dependent representation of ongoing acoustic scenes.
Introduction
Sensory systems are exposed to a continuous barrage of complex stimuli, and our perception of specific components within such an environment is dependent on the temporal relationships between individual elements in the signal. A simple and robust example of temporal context dependence in hearing is forward masking, where the detectability of a short probe signal is degraded when a masking stimulus is presented before the probe. Forward masking contributes to changes in the neural representation of behaviorally relevant sounds such as consecutive speech phonemes, even in the auditory nerve (AN) (Delgutte, 1980).
Neural correlates of behavioral forward masking have been largely sought in AN responses (Relkin and Turner, 1988). However, the responses of single AN fibers (ANFs) do not account for two important aspects of psychophysical forward masking (see Fig. 1C). First, the masked threshold shift is proportional to the discharge rate during the masker in ANFs, so that neural masking has a limited dynamic range (often <20 dB). In contrast, psychophysical masking grows continuously over all suprathreshold levels (Lüscher and Zwislocki, 1949; Jesteadt et al., 1982; Plack and Oxenham, 1998). Second, maximum threshold shifts in AN responses (<21 dB) (Relkin and Turner, 1988) are much less than their psychophysical counterparts (>40–50 dB). As a result, the behavioral threshold is substantially higher than the peripheral single-neuron threshold at moderate and high sound pressure levels (SPLs). Higher processing centers seem to discard information present in the AN, possibly in an effort to maintain a robust representation of new or dynamic environmental signals.
Several questions surface when the psychophysics and the available physiology are considered together. Clearly, central mechanisms must shape the neural representation of these stimuli, but details of this transformation remain unclear. Can the psychophysical behavior be explained with responses of single neurons? If so, at what level of the pathway do the perceptually relevant response features emerge? How are forward-masked threshold functions related to simple tone-driven responses? And what is the mechanism of central forward masking? Is it better described as a persistence of the response to the masker (as assumed in many psychophysical models used to account for forward masking) (Moore et al., 1988), or as a decrease in excitability (i.e., adaptation or inhibition) following masker offset?
We measured forward-masked responses in single neurons in the central nucleus of the inferior colliculus (IC) of the awake marmoset. We found that responses of IC neurons could account for the behavior of growth of masking (GOM), in terms of wide dynamic ranges, maximum amounts of masking, and recovery from masking that was roughly linear in log time. The underlying mechanism of masking was almost universally adaptation or offset inhibition; <10% of the population showed characteristics consistent with persistence of the masker response when best frequency (BF) maskers were used. Our data provide the first examination of forward-masked thresholds in the central auditory system, and they suggest that neural mechanisms underlying the behavioral phenomena are largely established at the level of the IC.
Materials and Methods
Animal preparation.
Animal care and recording procedures are similar to those used previously for studies in the awake marmoset auditory cortex and thalamus (Lu et al., 2001; Bartlett and Wang, 2007). During a one-month period preceding surgery, the animal was adapted to sitting quietly in the restraint device for the 2–4 h period used for recording. A sterile, isoflurane-anesthetized surgery was performed to mount posts for subsequent head fixation during recordings and to expose a small portion of the skull to provide access to the IC. The head posts were surrounded by dental acrylic, which bonded to the skull and also to a set of stainless steel screws that were mounted in the skull.
After a 2- to 3-week recovery period including daily antibiotic treatment and wound cleaning and maintenance, a small (∼1-mm-diameter) craniotomy was made in the skull. The location of the hole was based on stereotaxic coordinate maps as well as superficial landmarks on the skull that were marked during the sterile surgery (e.g., the lateral sulcus). The exposed recording chamber surrounding the craniotomy was covered with polysiloxane impression material (GC America, Alsip, IL) between recording sessions, and after many penetrations (usually >20), the hole was filled with a layer of bone wax and dental acrylic before making another craniotomy to provide access to other regions of the same IC. Multiple craniotomies were performed in both hemispheres before the animal was killed and perfused for histological evaluation. All procedures were approved by the Johns Hopkins University Institutional Animal Care and Use Committee and conformed to National Institutes of Health standards.
Recording methods.
Once the craniotomy was made, an epoxy-coated tungsten microelectrode (A-M Systems) was slowly advanced during daily recording sessions using a hydraulic microdrive (Kopf Instruments). During recording, the animal was awake or drowsing with its body loosely restrained in a comfortable seated position with the head held in a natural orientation. Depending on the angle of approach (which varied across animals and sessions), the electrode traversed 0.8–1 cm of brain tissue before reaching the IC. Usually the electrode was in an approximately frontal plane at an angle of ∼45° lateral to medial. Stimulus presentation, animal monitoring via video camera, and electrode advancement were controlled from outside a double-walled soundproof booth. Spike times of Schmidt-triggered single-neuron responses were recorded with respect to stimulus onset. Only well isolated single neurons were studied. Sessions were terminated after 3–4 h, or earlier if the animal showed signs of discomfort. Using these techniques, it was possible to collect data from several hundred electrode penetrations over the course of more than a year per animal.
Often, the external nucleus (equivalently, lateral cortex) (Oliver, 2004) or dorsal cortex of the IC was encountered using our lateral-access approach. We assumed that neurons were located in the CNIC based on three criteria: (1) short-latency (minimum first-spike latencies less than ∼ 20 ms), (2) reliable and qualitatively nonhabituating responses, and (3) a well defined location within the tonotopically organized frequency lamina (Aitkin et al., 1975). Neurons meeting these criteria are called “CNIC-like”; data from neurons not meeting these criteria are not included in the results presented here.
Acoustic stimuli.
Sound stimuli were digitally generated in Matlab (Mathworks), converted to analog signals (RP2.1, Tucker-Davis Technologies) and presented in the free field through a speaker (FT-28D, Fostex) positioned ∼1 m directly in front of the animal. With few exceptions, previous studies have tended to use monaural earphone stimulation in nuclei below the IC (Relkin and Turner, 1988; Shore, 1995), binaural headphone stimulation in IC (Sanes et al., 1998; Finlayson, 1999), and free-field stimuli at higher levels of the pathway (Bartlett and Wang, 2005; Scholl et al., 2008). No systematic differences in forward masking related to the stimulation mode are apparent across physiological or psychophysical studies. The free-field stimuli contained reverberation of the stimuli produced by the speaker, despite the foam padding on the chamber walls. The reverberation lasted ∼10 ms but was at a low level (<20 dB) relative to the stimuli. The strongest potential effect of such reverberation would be to produce a persistence of the response to the masker stimulus. As we show, such persistence was seen very rarely in the responses of neurons, suggesting that the impact of the weak acoustic echoes on the data was small or negligible.
Wideband noise bursts (50 or 200 ms in duration, 48.8 kHz bandwidth) and tone bursts of variable frequency and level were used to characterize neural activity as sound-driven and to estimate BF and threshold. BF was defined as the frequency at which sound-driven responses were observed at the lowest SPL. Each neuron was characterized by its response map (RM) (Ramachandran et al., 1999) and rate-level function (RLF). To obtain a RM, 200 ms tones (10 ms rise/fall) were presented (1/s) over a 2–4-octave range centered on the previously estimated BF, using 25 log-spaced frequencies per octave. This procedure was repeated for at least three sound levels, usually 20, 40, and 60 dB above the audio-visually determined estimate of BF threshold. RLFs were measured with 200 ms stimuli over a range from approximately −10–85 dB SPL (with SPL defined as the sound level at the position of the animal's head in the absence of the animal) in 1 dB steps, using both BF tones and broadband noise bursts. Spontaneous rate was estimated using the 400 ms periods immediately preceding sound onset, and was used as another part of the basic characterization of the neurons. All stimuli were presented once; when plotted, rate functions were smoothed using a triangular window spanning three consecutive SPLs.
With the more precise values of BF and threshold obtained from the RM and BF RLF, a family of forward-masked RLFs was measured. First, a control RLF was obtained using a 20 ms BF probe alone (10 ms rise/fall, no steady state, repetition rate of 1/s), over a range of 50–70 dB. The lowest tested probe SPL was at least 10 dB below the threshold estimated from the 200 ms tone RLF. Following this baseline condition, forward-masked RLFs were measured, consisting of a 200 ms masking tone followed by the probe tone (as in Fig. 1A). The masker had fixed properties and the probe level was varied as above. The default parameter values for the masker were a frequency at BF, an SPL ∼40 dB above the neuron's threshold, and a delay between the masker offset and probe onset of 10 ms. The forward-masked paradigm was repeated with various combinations of two of these parameters fixed and the third varied, until contact with the neuron was lost. Masker SPL was varied in 10–20-dB steps from near threshold levels to 60–80 dB SPL above threshold. Masker-offset/probe-onset delay was varied from 0 ms up to 310 ms (0, 10, 30, 70, 150, and 310 ms). Finally, the frequency of the masker tone was varied in some neurons. The frequencies chosen were based on features of the pure-tone responses, often within inhibitory regions of the RM.
Response analysis.
For direct comparison with psychophysical data, we focused on describing the changes in probe detection thresholds given different parameter variations in the masker. Threshold was defined as the lowest probe level that elicited a reliable increase in response rate in a 40 ms window surrounding the probe. The beginning of the window was delayed by 10 ms following the physical probe onset to allow for acoustical and neural delays. Probe threshold was determined based on a comparison response which was defined as the average rate observed for the 10 stimulus conditions at the lowest probe levels (all set to SPLs below the threshold for 200 ms tones). For each probe SPL above the lowest 10 tested, a “hit” was recorded if the number of spikes in the test window was greater than the comparison spike count. Otherwise, the trial was labeled a “miss.” Probe threshold was empirically defined as the lowest level in a string of 10 or more probe SPLs that resulted in a hit, with no more than a single miss allowed between otherwise consecutive hits.
This threshold criterion was adopted because of our decision to use high resolution when varying probe SPL (1 dB steps), as opposed to collecting multiple repetitions of each stimulus condition. This is in contrast to the method used by Relkin and colleagues (Relkin and Pelli, 1987; Relkin and Turner, 1988), who collected many repetitions of a more sparsely sampled probe SPL space. Our definition was almost always consistent with the threshold determined by directly examining dot rasters, and was actually trivial in many conditions where the comparison spike rate was zero. If anything, our rate-based rule would tend to underestimate sensitivity in probe-alone conditions because of the higher background comparison rate (as in ANF responses). The reliability of our method was supported by the reliability of threshold estimates in simulations of Poisson spike-trains with rates typical of our data and smoothing to produce comparable variability in the spike counts.
Excluded neurons.
Type O neurons (Ramachandran et al., 1999) were not common in this population, but four that were encountered were excluded from the analysis because they gave weak and nonmonotonic probe responses that were usually completely suppressed by the presence of a preceding masker (i.e., masked thresholds were undefined). The responses of a single type V neuron were not included for similar reasons: the probe response did not reach the criterion for threshold over the entire range of levels tested. Neurons were included in the population analyzed for different probe delays only if thresholds were measured for at least three masker-probe delays with a masker presented at an SPL ∼40 dB above threshold. One additional neuron was excluded because thresholds did not recover toward probe-alone conditions, even after a 320 ms masker-probe delay (the probe threshold shift remained larger than 40 dB for all tested delays); the interstimulus delay was apparently not long enough in this case to avoid cumulative adaptation.
Results
IC responses to the default masker
The responses to the forward-masking paradigm were recorded from 102 CNIC-like neurons from two animals; recordings were made in the IC on both sides. When the masker and probe tone frequencies were both set to the neuron's BF, the effect of the masker on the probe response was almost exclusively suppressive over the entire range of probe levels tested (usually ≥50 dB).
The responses of two example neurons to a 20 ms BF probe tone over a range of probe SPLs are illustrated in Figure 2, in the form of spike-time raster plots (Fig. 2A,B,D, and E) and rate-level functions (Fig. 2C,F). Probe-alone (unmasked) responses are shown in Figure 2, A and D, and responses to the sequential presentation of the default masker [40 dB sensation level (SL) BF tone] and the subsequent probe are shown in Figure, 2, B and E. Downward triangles indicate the neural detection threshold for the probe; shaded areas mark the timing of the masker (when present) and the 40 ms probe analysis window. For each of these neurons (and for 101 of 102 of the population), the preceding masker caused an increase in the neural detection threshold for the probe. This can be seen as a rightward shift in the triangles in the second versus the first row of panels. The two examples include a neuron with sustained responses to the masker (Fig. 2A–C), as well as a transient (onset) responder to both the masker and the probe (Fig. 2D–F). Both neurons exhibited a forward-masked threshold shift of 24 dB.
Across the population, the average amount of masking was 18.8 ± 9.6 (mean and SD) dB for the default masker. This value is well within the range of shifts observed psychophysically across studies and subjects for comparable stimulus conditions (∼ 15–25 dB) (Jesteadt et al., 1982; Plack and Oxenham, 1998). The amount of masking was not strongly dependent on BF, although the widest range of threshold shifts was found in the mid-frequency (4–16 kHz) region, and we did not observe any shifts larger than 20 dB in neurons with BFs of <1.5 kHz (Fig. 3A). These low-frequency neurons tended to have V-shaped frequency response maps, with a pronounced lack of sideband inhibition above or below BF.
In contrast to the near independence of threshold shift and BF, there was a weak but significant dependence of the amount of masking on spontaneous rate (SR): low-SR neurons tended to exhibit more masking than high-SR neurons (Fig. 3B) [amount of masking vs log(SR) correlation, r = −0.35; p < 0.001]. This is also true of ANFs (Relkin and Doucet, 1991), but the magnitude of the threshold shift in the IC population was much greater than that observed in ANFs. The gray hatched areas in Figure 3B represent the range of ANF threshold shifts measured under nearly identical stimulus conditions [redrawn from the work of Relkin and Doucet (1991), their Fig. 3]. The largest threshold shifts were observed in onset neurons (Table 1) (2 × 2 contingency test, p < 0.06 for all combinations of onset and other unit types) with no significant difference between the other neuron types (2 × 2 contingency test p > 0.17 for the remaining combinations).
Response dependence on masker level
As discussed above, in ANFs forward masking varies with the discharge rate in response to the masker, typically leading to saturating GOM curves (Fig. 1C) (Relkin and Turner, 1988). IC neurons show a variety of RLFs, many showing nonmonotonic behaviors. However, GOM in IC neurons with the forward masking paradigm usually did not exhibit similar nonmonotonicities. We found that most (51 of 74) IC neural GOM functions increased monotonically over the 40–80 dB range of masker levels tested, despite saturation or turn-over in the excitatory masker-driven RLFs (49 of 74 had saturating or nonmonotonic RLFs over the same SPL range).
Spike-time rasters for three neurons' responses to variations in masker SPL (across different panels) and probe SPL (within panels) are shown in Figure 4. When presented in isolation, the short probes reliably evoked a spike or burst of spikes at a relatively low probe SPLs; detection thresholds are shown by the downward triangles as before. These examples illustrate the difference between the behavior of the probe threshold and the rate in response to the masker (given by the numbers to the right of the dot displays). In the three examples shown, detection threshold shifted monotonically with masker level in the face of a nonmonotonic masker response (left column), essentially no sustained response to the masker (middle column), or a saturating response to the masker (right column).
Typical probe rate functions are shown in the panels of Figure 5A for four other example neurons. The parameter in each panel is masker SPL, and different colors are used for different SPLs of the preceding masker (masker responses not shown). These examples show a suppression of the probe response magnitude over the entire range of probe SPLs tested, with the strongest suppression and largest threshold shift caused by the loudest masker. Thresholds derived from these probe responses were used to construct individual-neuron GOM functions (Fig. 5B).
A straight line was fit to each set of points in Figure 5B to determine the GOM slope. The slopes in these example neurons were 0.55, 0.42, 0.28, and 0.15 (Fig. 5Aa-d, respectively). Across the population, the average GOM slope was 0.36. A comparison of GOM slopes between IC neurons (Fig. 5C, shaded histogram) and ANFs [Fig. 5C, red line redrawn from Relkin and Turner (1988)] reveals a tendency for IC GOM slopes to be greater than those measured in the AN (Wilcoxon rank-sum test p < 0.02). This is true although the AN slopes were computed only for masker levels on the dynamic portion of the ANF's RLF (typically 20–30 dB SL) and used no gap between the masker and probe; the IC slopes were derived from probe responses over the entire range of masker levels tested (i.e., including levels that potentially caused a saturated or decreased high-level response to the masker) and a 10 ms masker-probe gap.
On average, the amount of masking in IC neurons grew gradually as a function of masker level (Fig. 6A), with no sign of hard saturation at the highest masker levels tested. The dashed line in Figure 6A has a slope of 0.5 dB/dB and assumes no threshold shift for a masker presented at 0 dB sensation level.
On-BF masking is not consistent with persistence
Inspection of the dot displays in Figures 2 and 4 shows that there is a clear gap between the response to the masker and the response to the probe. This behavior shows that the persistence model often used in psychophysical studies (Moore et al., 1988; Oxenham, 2001) does not apply to the IC; in that model, masking occurs because of persistence of response to the masker, which overlaps with the response to the probe. To show that this behavior was typical, we used a persistence/adaptation ratio (PA ratio: the average rate in the probe analysis window in the masker-alone condition divided by the average spontaneous rate). Values less than one indicate an adaptation or inhibition-driven mechanism; values greater than one suggest a persistent or offset excitatory response. Individual points contributing to the averaged GOM function are shown in Figure 6B, with the symbol size related to the value of the PA ratio. Clearly, a persistent response was not required to generate a threshold shift comparable with those observed psychophysically (small points correspond to small PA ratios).
The relatively few persistent (or offset) responses that we observed did not emerge until the masker level exceeded ∼30 dB SL. The vast majority of masker conditions resulted in PA ratios of <1 (259 of 312, 83%), and more than half (170 of 312) elicited no spikes in the masker-alone observation window (PA ratio = 0) (Fig. 6D). Consistent with the independence of individual-neuron GOM and masker RLF slopes (above), there was not a strong relationship between threshold shift and the masker-driven rate when the entire population was considered (Fig. 6C).
Response dependence on masker-probe delay
Psychophysical forward masking (in dB) is proportional to the log of signal delay (Lüscher and Zwislocki, 1949; Widin and Viemeister, 1979; Kidd and Feth, 1982; Jesteadt et al., 1982; Moore and Glasberg, 1983). In other words, probe threshold shifts exhibit a linear dependence on signal delay when delay is plotted on a logarithmic scale. Here we show a similar behavior for IC neurons.
In a subset of neurons (n = 41), we examined the effect of changing the delay between masker offset and probe onset, over a range from 0 to 310 ms. Responses of two example neurons are shown in Figure 7A. Spike-time raster plots are overlaid in different colors for the different delay conditions, although the conditions were presented in separate independent blocks. The 40-dB SL, 200 ms masker elicited a sustained response in the first neuron, and a transient onset response in the other cell. Responses to the probe emerged at higher SPLs for probe signals presented with shorter delays. For these examples, the probe threshold decreased with delay and recovered toward baseline (unmasked) conditions. Responses to the probe-alone case (Fig. 7A, right column) emerged at lower probe sound levels than the masked responses, and were also larger in magnitude at suprathreshold levels.
Probe threshold recovery functions for individual neurons are shown in Figure 7B (thin lines), along with the across-neuron average (thick line with circles, ± SD). It is clear that most of the probe threshold-shift functions, along with the average curve, do recover approximately linearly in log time (even ignoring the point at 0 delay which is arbitrarily positioned on the log abscissa). There is a small amount (∼5 dB) of residual masking that does not change between the 150 and 310 ms delay conditions. This was observed despite the fact that the silent time between the probe offset and masker onset was always >500 ms, an off-period that is typical in psychophysical measurements. The magnitude of this residual masking was likely amplified by a bias in the data collection: neurons that showed little or no masking for shorter delays were often not tested at the longest delay.
Response dependence on masker frequency
Generally, maskers had their largest effect when placed at BF. As has been shown above, the type of masker-driven response pattern (i.e., sustained or onset) was secondary. To illustrate this behavior, the responses of three neurons are shown in Figure 8 for stimuli with various masker frequencies. The neuron in Figure 8A was studied with maskers at three frequencies, all within the excitatory region of the response map. Each masker produced a qualitatively similar sustained spiking pattern throughout the masker (driven rates during the masker were 30, 80, and 57 sp/s, for the below-BF, on-BF, and above-BF maskers, respectively). The on-BF masker had the largest forward masking effect (blue circle), producing a 17 dB shift in probe threshold and a large decrement in the rate response to the probe (Fig. 8Ab). The other two masker frequencies produced much smaller threshold shifts (filled square and diamond).
Usually, maskers positioned within an inhibitory region of the frequency response map did not cause significant elevations in probe threshold (Fig. 8Bb, orange square and red diamond). In fact, maskers in the inhibitory regions of the response map could produce a modest improvement in probe thresholds over the unmasked condition (11 of 39 tested). This “negative masking” had a magnitude between 5 and 12 dB, with one unusually high spontaneous-rate neuron exhibiting negative masking >20 dB for all masker conditions tested (this neuron was the outlier omitted from the population plots in Fig. 2).
In some cases, the response to the probe was affected by offset discharges in response to off-BF maskers. The example in Figure 8C shows the most extreme example of such a case: the above-BF masker (red diamonds) caused such a strong offset response (compare red and black RLFs at low probe SPLs) that the addition of the probe did not cause a further increase in firing rate over the entire range of probe levels tested. This neuron is identified in Figure 9 (see below) by symbols surrounded by parentheses to indicate its status as an outlier in this sense.
The data in Figure 8 are typical in that the largest masking was observed within the excitatory response area of the neuron; that is, the masking was often tuned at least as narrowly as the excitatory response of the neuron. Figure 9 shows population data on this point. On-BF maskers (blue symbols) generally produced more masking (ordinate) than off-BF maskers (red symbols). Again, the amount of masking was not predicted by the discharge rate during the masker (Fig. 9A). However, some inhibitory maskers (negative masker-driven rates) produced as much masking as on-BF maskers. Those were mostly cases with PA ratios of >1 (Fig. 9A, open symbols), in which the masking was produced by an offset response as in Figure 8C. The decline of forward masking for this group of neurons as the masker frequency is moved away from BF is shown in Figure 9B.
Discussion
The nature of physiological forward masking in IC
To our knowledge, these data represent the first description of neural forward-masked detection thresholds at any level of the central auditory system. A large number of papers have described forward masking at various levels of the auditory system [AN: Smith (1977), Harris and Dallos (1979); CN: Starr (1965), Boettcher et al. (1990), Palombi et al. (1994), Shore (1995), Backoff et al. (1997), Bleeck et al. (2006); SOC: Finlayson and Adam (1997); IC: Finlayson (1999), Faure et al. (2003), Furukawa et al. (2005); A1: Calford and Semple (1995), Brosch and Schreiner (1997), Brosch et al. (1999), Bartlett and Wang (2005), Wehr and Zador (2005), Nakamoto et al. (2006), Scholl et al. (2008)]. However, these have all considered the effect of the masker on the responses to a fixed probe stimulus. Because of nonlinear relationships between neural responses and detection threshold, it is not possible to estimate detection thresholds from these data and therefore not possible to compare them directly to behavioral data. Thus we considered it important to explicitly estimate probe thresholds.
The principal result of this study is that GOM in the awake marmoset IC resembles the monotonic 0.5 dB/dB growth seen in human psychophysical forward masking (Plack and Oxenham, 1998) and not the saturating growth seen in anesthetized chinchilla ANFs (Relkin and Turner, 1988). Probe masking in the IC also shows a linear decay with log time (Fig. 7) and shows stronger masking for on-BF maskers (Figs. 8, 9), both features seen in psychophysical forward masking, although the frequency dependence cannot be quantitatively compared with the data available. Although there are no forward masking psychophysical data available for marmosets, we would expect that at least the gross trends in the fundamental perceptual features of the phenomenon would be conserved across primate species. An additional implicit assumption underlying our interpretation is that AN responses in the chinchilla are similar to those in primates.
The results also clearly show that masking does not depend on the net masker-driven excitatory activity in the cell (Figs. 6C, 9). Other studies examining the temporal context dependence of responses in the mammalian IC have reached conflicting conclusions as to whether the history of the excitatory spiking response is sufficient to account for the observed “adaptation” effects (McAlpine et al., 2000; Ingham et al., 2001) or not (Malone and Semple, 2001). The current data support the idea that the history of excitation is insufficient to predict the magnitude of forward masking in the marmoset IC. Another way of stating this conclusion is that IC forward masking is not strictly an adaptation process defined as a change in a neuron's properties due to its history of activation.
Potential inhibitory mechanisms
Inspection of the probe RLFs (Figs. 2, 4, 5) show a universal decrease in overall probe response rates relative to the unmasked condition, including a suppression of spontaneous activity. This behavior suggests that inhibitory mechanisms are involved in forward masking in the IC, either within the IC or at lower levels of the auditory system. There is considerable evidence for poststimulus inhibitory activity in IC. Sound-evoked IPSPs and currents are present in IC neurons following the cessation of the acoustic stimulus (Nelson and Erulkar, 1963; Kuwada et al., 1980, 1997; Covey et al., 1996; Pedemonte et al., 1997; Tan and Borst, 2007; Xie et al., 2008). Also, local blockade of inhibitory neurotransmitters (GABAA, and to a lesser extent, glycine) releases extracellular responses from the persistent (i.e., continuing beyond stimulus offset) inhibition that is often present in control conditions (Faingold et al., 1991; Pollak and Park, 1993; Bauer et al., 2000; LeBeau et al., 2001).
A parsimonious explanation for several aspects of the data relies on inhibitory inputs from the superior paraolivary nucleus (SPON) to generate the strong and wide-dynamic-range offset suppression of the probe. Many SPON neurons make a direct inhibitory (GABAergic) projection to the ipsilateral IC (Kelly et al., 1998; Kulesza and Berrebi, 2000), and their physiological responses are often characterized by no response during the stimulus and a burst of spikes locked to the offset of a tone burst that can last as long as 50 ms (Kuwada and Batra, 1999; Kulesza et al., 2003). These offset responses (1) tend to have higher thresholds than nearby “primary-like” SOC neurons, (2) are sharply frequency tuned near threshold and become more broadly tuned at higher SPLs, and (3) can exhibit rate responses with remarkably wide dynamic ranges (Kuwada and Batra, 1999; Kulesza et al., 2003). Such input and response patterns could at least qualitatively account for enhanced suppressive strength in the IC at higher masker SPLs (Figs. 2⇑⇑–5) and sharp tuning of this suppression (Figs. 8, 9). A schematic illustration of the frequency-response profile of a hypothetical IC neuron receiving such an input is shown in Figure 10. At low SPLs, the response to the masker is sharply tuned and purely excitatory, and any forward suppression is presumably inherited from lower levels. At higher SPLs, broadly tuned inhibitory inputs (possibly from the DNLL) are activated and interact with the excitation (Ramachandran et al., 1999). This inhibition could be sluggish, working to augment the forward suppression found in the inputs by lasting beyond the offset of stimulation. Finally, the SPON-IC pathway is active at still higher SPLs, potentially providing the substrate for monotonically increasing GOM functions in IC neurons.
Other possible sources of forward suppression include intrinsic membrane properties in IC (e.g., GABA receptors with slow kinetics or a voltage-sensitive current that changes its properties following masker stimulation), local recurrent inhibition within the IC, and descending influences from auditory cortex. Interestingly, forward suppression in auditory cortex is not related to the strength of poststimulus hyperpolarization (Scholl et al., 2008), and patterns of probe suppression are broadly similar to those we recorded extracellularly in IC neurons. This, along with the tight match between behavior and IC responses we report, suggests that much of the processing underlying behavioral forward masking is set up before the thalamocortical pathway.
Functional implications
The so-called temporal-window model has influenced much of the recent interpretation of nonsimultaneous perceptual masking (Moore et al., 1988; Plack and Oxenham, 1998; Oxenham and Plack, 2000). Forward masking in this model results because the masker response is assumed to last beyond the stimulus offset in the form of persistent excitation, and responses to the subsequent probe tone are swamped by the extended masker-induced firing. While we did occasionally find persistent excitation in a small group of neurons at high masker levels (Fig. 6B,D), poststimulus inhibition was much more common. Importantly, neural threshold shifts consistent with psychophysical masking were regularly observed in the IC over the entire range of tested masker SPLs even if the few persistent neurons were excluded (Fig. 6B). We conclude that mechanisms like inhibition or adaptation (discussed above) are more likely to account for behavioral forward masking, despite circumstantial psychophysical evidence suggesting otherwise (Oxenham, 2001) and the relatively weak offset adaptation that is present in ANFs (Relkin and Turner, 1988).
Differences in the susceptibility of different neuron types to conditioner-induced suppression (Table 1, Fig. 3) may provide a hint as to their functional importance in hearing. Some IC neurons show small threshold shifts (similar to those measured in ANFs); this information is apparently unavailable or not used by listeners in a forward masking task. In contrast, other IC neurons, especially onset neurons, respond strongly and reliably to the short probe stimuli and show threshold shifts that are comparable with behavioral ones. This finding is consistent with previous studies of responses to sequential tones in the IC (Finlayson, 1999), and it suggests that the onset neurons may be more heavily weighted by the auditory system for tasks like teasing apart a string of rapidly changing stimuli.
The recovery of response strength with increasing delay between masker and probe (Fig. 7) follows a time course that might be useful for maintaining a robust representation of species-specific social communication calls in the marmoset midbrain. One of the most common vocalizations made in this species is the “twitter” call, which is made up of a series of brief frequency-modulated phrases separated by silent intervals ranging from ∼100–160 ms (DiMattina and Wang, 2006). In our data, probe thresholds were recovered in most neurons for silent gaps of this duration. These slow fluctuations would be more likely to be present at the output of IC neurons than faster modulations that might be attenuated by the same mechanisms underlying the forward suppression described here.
Summary
A forward-masking paradigm modeled after those used in psychoacoustic studies was applied in the awake marmoset IC, revealing level-dependent, frequency tuned offset suppression with characteristics strikingly similar to those of perceptual forward masking. These response profiles were largely emergent between ANFs and IC neurons; indirect evidence points to the de novo generation of many of the key features at the level of the IC, including the excitation-independent, wide dynamic range of masker-evoked shifts in probe detection threshold.
Footnotes
-
This work was supported by National Institutes of Health Grants DC00115, DC00023, and DC009164. We are grateful to Sean Slee, Bradford May, Sharba Bandyopadhyay, and two reviewers for comments on earlier versions of this work, and to Xiaoqin Wang, Yi Zhou, Jenny Estes, and William Tam for help with animal preparation and care.
- Correspondence should be addressed to Paul C. Nelson, 505 Traylor Research Building, 720 Rutland Avenue, Baltimore, MD 21205. pcnelson{at}jhu.edu