Post-stimulatory activity in primate auditory cortex evoked by sensory stimulation during passive listening

Under certain circumstances, cortical neurons are capable of elevating their firing for long durations in the absence of a stimulus. Such activity has typically been observed and interpreted in the context of performance of a behavioural task. Here we investigated whether post-stimulatory activity is observed in auditory cortex and the medial geniculate body of the thalamus in the absence of any explicit behavioural task. We recorded spiking activity from single units in the auditory cortex (fields A1, R and RT) and auditory thalamus of awake, passively-listening marmosets. We observed post-stimulatory activity that lasted for hundreds of milliseconds following the termination of the acoustic stimulus. Post-stimulatory activity was observed following both adapting, sustained and suppressed response profiles during the stimulus. These response types were observed across all cortical fields tested, but were largely absent from the auditory thalamus. As well as being of shorter duration, thalamic post-stimulatory activity emerged following a longer latency than in cortex, indicating that post-stimulatory activity may be generated within auditory cortex during passive listening. Given that these responses were observed in the absence of an explicit behavioural task, post-stimulatory activity in sensory cortex may play a functional role in processes such as echoic memory and temporal integration that occur during passive listening.

Scientific RepoRtS | (2020) 10:13885 | https://doi.org/10.1038/s41598-020-70397-0 www.nature.com/scientificreports/ During passive listening, the termination of an auditory stimulus routinely evokes transient offset responses in auditory cortical neurons, typically lasting up to tens of milliseconds [34][35][36][37] . In the cat, the tonal receptive fields of offset responses are often similar to the tuning of responses evoked at the onset of the stimulus although onset and offset receptive fields can also vary their tuning properties 36 . In the mouse however, the tuning of onset and offset responses is largely distinct 35 . Despite this lack of consistency in receptive field dynamics, the ability of auditory cortical networks to routinely generate activity following the termination of an effective auditory stimulus is a consistent feature across species. The maximum duration of activity that can be evoked following stimulus offset, however, is currently unknown.
Here, we investigated whether it is possible to evoke post-stimulatory activity in auditory cortical neurons in the absence of a behavioural task and aimed to characterise the properties of such activity. We analysed activity from single units in the auditory cortex (fields A1, R and RT) and thalamus of awake, passively-listening marmosets that were presented an array of auditory stimuli during experiments not concerned with investigating post-stimulatory activity [38][39][40][41] . We observed post-stimulatory activity lasting for hundreds of milliseconds following the termination of the acoustic stimulus in a sub-population of auditory cortical neurons. This activity followed a variety of response profiles during sensory stimulation, including adapting, sustained and suppressed responses. Post-stimulus activity had a shorter latency and was of longer duration in cortex than in thalamus, indicating that the mechanisms underlying this activity may be primarily cortical.

Results
post-stimulatory activity is observed in auditory cortex units during passive listening. The data analysed in this report were based on a database of 1557 single units recorded from the auditory cortex of 4 passively listening marmosets during the presentation of a variety of auditory stimuli in previous experiments. The stimuli presented were unmodulated pure tones, amplitude-modulated tones, white noise, band-pass noise and click trains. Of these units, 1,188 met our criteria to be included in further analysis (see "Methods"). Units were required to show an elevation in firing > 2 standard deviations over mean baseline firing on at least half of the trials that was longer in duration than for baseline activity observed during the pre-stimulus period. For the initial analysis of post-stimulatory activity units were not separated by the cortical field they were recorded from. Post-stimulatory activity was observed in a subset of these single units in response to particular stimuli (Fig. 1A). Short-duration offset responses were commonly observed, with 77.78% (N = 924/1,188) of units showing significant post-stimulus activity beginning within 50 ms of stimulus offset (Fig. 1B). By searching the stimulus space, it was possible to evoke post-stimulatory activity in 39.31% (N = 467/1,188) of the units recorded (see "Methods") ( Fig. 1C).
Stimulus parameters such as stimulus type and post-stimulus interval were varied across the neurons recorded. In order to quantify the duration of the observed post-stimulus activity, the analysis was initially restricted to stimuli with a post-stimulus interval (PSI) of 300 ms. This subset consisted of responses to tone and noise stimuli lasting 200 ms. For the population of 805 units recorded with these stimulus parameters, the longest duration post-stimulus response observed for each unit exceeded this 300 ms interval in 21.86% of units ( Fig. 2A). When the PSI was extended to 500 ms for tone and noise stimuli 5.23% of the 172 units tested exceeded this 500 ms interval (Fig. 2B). For units presented with tone, noise and click stimuli followed by a PSI of > 500 ms (N = 65), the range of post-stimulus response durations spanned from 12 to 1681 ms (Fig. 2C).
Post-stimulatory activity was observed for all stimulus types tested. For stimuli with a PSI of 300 ms, tone stimuli produced significantly longer duration post-stimulatory activity (N units = 694, mean = 186.93 ms) than noise stimuli (N units = 308, mean = 167.96 ms; Two-Sample t-test, p < 0.001) (Fig. 3A). For stimuli with a PSI of 500 ms, no significant difference was found between the duration of post-stimulus activity in response to tone (N units = 80, mean = 196.3 ms) and noise stimuli (N units = 110, mean = 201.49 ms) (Two-Sample t-test, p = 0.8) (Fig. 3B). For stimuli with a PSI of > 500 ms, neurons were tested with tone (N units = 59, mean = 280.93 ms), noise (N units = 44, mean = 246.05 ms) and click train (N units = 29, mean = 256.21 ms) stimuli. No significant difference was observed in the duration of post-stimulus activity produced by these different stimulus types (ANOVA(2); p = 0.84, Fig. 3C).
post-stimulatory activity is observed following adapting, sustained and suppressed responses. The response profiles observed within auditory stimulation were examined next in order to gain insight into the dynamics that might produce this post-stimulatory activity. For the units that showed post-stimulatory activity (N = 467), within-stimulus evoked activity was quantified by taking the ratio of the within-stimulus firing rate and the baseline firing rate preceding sensory stimulation to produce an evoked ratio (Fig. 4A). An evoked ratio > 1 would indicate that the within-stimulus firing rate was increased with respect to baseline firing while values < 1 would indicate suppression of firing in the within-stimulus period. The dynamics of responses with an evoked ratio of > 1 were first analysed. For each response in this group the median within-stimulus spike time was calculated as a measure of the extent of adaptation that occurred during sensory stimulation. This median spike time was normalised by the duration of the stimulus, producing a normalised median spike time (NMST) for each unit between 1 and 0. This measure of adaptation results in fast adapting responses being associated with values near to zero and ramping responses being associated with values near one. For units with an evoked ratio of > 1 (N units = 332 of 467) the majority of units had NMSTs in the middle of this range, indicating a prevalence of sustained response profiles (Fig. 4B). In order to separate the overlapping adapting and sustained populations, adapting responses were defined as those with an NMST of < 0.35 (N units = 78 of 332) while sustained responses were defined as those with an NMST of > 0.35 (N units = 254 of 332, Fig. 4B). Ramping responses with NMSTs near 1 were rare and were classed as sustained responses here. Post-stimulatory activity was observed following both adapting and sustained responses (Fig. 4C,D). NMSTs were also calculated for Scientific RepoRtS | (2020) 10:13885 | https://doi.org/10.1038/s41598-020-70397-0 www.nature.com/scientificreports/ the mean response to all stimuli for a single unit and showed a qualitatively similar distribution to the response NMSTs that were calculated for the stimulus that produced the longest duration post-stimulatory activity (N units = 332, adapting N = 76, sustained N = 256, Fig. 4E). These two measures also showed a significant positive correlation (r = 0.7, p < 0.001) indicating that within-stimulus response profiles that precede post-stimulatory activity for a single effective stimulus are representative of the average response dynamics of the units (Fig. 4F).
The relationship between within-stimulus response magnitude and the duration of post-stimulus activity was next examined. Within-stimulus tuning curves were calculated by measuring the peak firing rate during the stimulus time period in the mean PSTH for each stimulus (Fig. 5A). Post-stimulus tuning curves were calculated by measuring the duration of significant post-stimulus activity produced by each stimulus. A correlation coefficient (cc) was then calculated for these tuning curve pairs for each unit by correlating the peak firing rates across stimuli during the within period with the duration of post-stimulus activity. Tuning curves showed a significant www.nature.com/scientificreports/ positive correlation across the population of units classified as having adapting responses to their best stimulus (N units = 78, mean = 0.28, t-test; p < 0.001) (Fig. 5B). This was also the case for units that showed sustained responses to their best stimulus (N units = 254, mean = 0.37, t-test; p < 0.001) (Fig. 5C). Tuning curves for units that showed a sustained response (Fig. 5C) showed a significantly greater correlation across the population than units that showed an adapting response to their best stimulus ( Fig. 5B) (Adapting mean cc = 0.28, Sustained mean cc = 0.37, two sample t-test; p = 0.001). Not all units that showed significant post-stimulus activity showed evoked within-stimulus auditory activity in the within-stimulus interval. These units, associated with an evoked ratio of < 1, were examined next (N units 135 of 467). Units in this population typically demonstrated within-stimulus suppression (Fig. 6A,B). Elevated post-stimulus activity can therefore be observed following adapting, sustained and suppressed within-stimulus response profiles. For stimuli with a PSI of 300 ms, significant variation in post-stimulus response duration  post-stimulus activity emerges in cortex. Single units (N = 379) were also recorded from the auditory thalamus of 3 passively listening marmosets (Fig. 7A). The criteria for inclusion in further analysis was met by N = 361 units (see "Methods"). When presented with stimuli lasting 200 ms followed by a 300 ms PSI (N units = 303), thalamic post-stimulus activity was significantly shorter (median = 57 ms) than the post-stimulus activity reported above in cortex (N units = 805, median = 191 ms) (rank-sum test, p < 0.001) (Fig. 7A). Thalamic post-stimulus activity was also significantly shorter (N units = 48, median = 38.5 ms) than cortical activity (N units = 237, median = 196 ms) for units presented with stimuli followed by a PSI of > 300 ms (rank-sum test, p < 0.001) (Fig. 7B). The latency of post-stimulus activity, the time following stimulus termination before firing Comparison of the distributions of post-stimulatory activity duration produced by tone and noise stimuli followed by a 300 ms PSI. Post-stimulus activity in response to tone stimuli was found to be of longer duration than that produced by noise stimuli when tested with this PSI. (B) Comparison of post-stimulatory activity duration produced by tone and noise stimuli followed by a 500 ms PSI. No significant difference in the means of these distributions were observed. (C) Comparison of post-stimulatory activity duration produced by tone, noise and click stimuli followed by a 500 ms PSI. No significant difference in the means of these distributions were observed.
Scientific RepoRtS | (2020) 10:13885 | https://doi.org/10.1038/s41598-020-70397-0 www.nature.com/scientificreports/ became significantly elevated, was significantly shorter in cortex (N units = 467, latency = 4 ms) than in thalamus (N units = 351, latency = 23 ms) (rank-sum test, p < 0.001) (Fig. 7C). We next investigated whether variation in the duration of post-stimulus activity exists between cortical fields. Single units were recorded in areas A1, R and RT in core auditory cortex of two subjects, with units in each field being exposed to all stimulus types (Fig. 8A). In order to compare durations across fields it was necessary to only include units for which the maximum duration of post-stimulus activity could be accurately estimated, this is, where the duration of activity did not exceed the post-stimulus interval. The duration of post-stimulus activity showed significant variation across cortical fields (Kruskall-Wallis test, H(2) = 15.11, p < 0.001) (Fig. 8B). Post-hoc multiple comparisons of mean ranks indicated that post-stimulus activity of significantly longer duration in RT compared to A1 and R (median durations, A1 = 193.

Discussion
Here we report the presence of post-stimulatory activity in a population of auditory cortical neurons in the absence of a behavioural task. We observed that, in the awake marmoset auditory cortex, these responses could last hundreds of milliseconds and in some cases over a second. Post-stimulatory activity has previously been www.nature.com/scientificreports/ observed in subcortical structures in non-primate species. In the dorsal cochlear nucleus of the cat pause-build units have been found to show modulation of spontaneous activity on the order of hundreds of milliseconds following the termination of auditory stimuli [42][43][44] . In the inferior colliculus of the anaesthetised mouse longduration stimuli lasting over 30 s have been found to produce sound-evoked after discharges lasting several minutes 45 . The inheritance of post-stimulus activity from subcortical structures does not appear to account for the activity that we observed in cortex as we found post-stimulus activity to be of shorter duration in auditory thalamus than in cortex. Post-stimulus activity was also found to have a longer latency following stimulus offset in thalamus compared to cortex. This indicates that units displaying such activity in thalamus may simply inherit this activity via cortical feedback projections. We also observed that the duration of this activity increased in RT, a secondary cortical area. These findings indicate that the circuit architecture required to generate these responses may be intrinsic to cortical networks. Post-stimulus activity was observed following several different response profiles during sensory stimulation. As previously reported, units in auditory cortex are capable of showing adapting and sustained responses 46 . The evoked responses of units observed here fell along a continuum from adapting, through sustained to ramping. Post-stimulatory activity was observed following both adapting and sustained response profiles, as well as following suppression. Given the heterogeneity of the stimulus-related activity that preceded the post-stimulus firing, the possibility that a single mechanism is responsible for this activity appears unlikely. In the prefrontal cortex, the mechanistic basis of post-stimulatory activity has not been fully elucidated but three broad classes of mechanism have been proposed to account for this pattern of activity; intrinsic neuronal properties at the cellular level, synaptic dynamics at the network level and the effect of activity from subcortical afferents at the systems level 47 .
The post-stimulus activity observed in sustained units could be accounted for by slow NMDA receptormediated excitatory synaptic currents which can last for hundreds of milliseconds 48 . These currents have been found to have a twofold longer decay time in rat prefrontal cortex compared to primary visual cortex, in keeping with these slow currents playing a role in the generation of prefrontal post-stimulatory activit 49 . It is not known whether NMDA currents in the marmoset auditory cortex are slower than in other cortical areas however. Cortical networks have a highly recurrent architecture 50 and reverberant activity in such a network has long been considered a candidate mechanism for the generation of post-stimulatory activity [51][52][53][54][55][56][57] . A recurrent mechanism of www.nature.com/scientificreports/ this kind would account for the temporal fluctuations in post-stimulus activity that were widely observed in these units. It has been suggested that hybrid mechanisms are most likely responsible for post-stimulatory activity throughout the brain 58 and, in keeping with this, combining slow excitatory currents with recurrent activity has been found to be important for the generation and stabilisation of post-stimulatory activity in computational models of such activity 55,56 . Similarly, asynchronous excitatory activity mediated by AMPA receptors has been found to be important for the maintenance of post-stimulatory reverberant activity in hippocampal cultures 59 . Sustained responses in auditory cortex that persist following stimulus offset and post-stimulatory activity in prefrontal cortex may therefore both be implemented by a hybrid of excitatory mechanisms. In keeping with an excitatory mechanism, the duration of post-stimulus activity following sustained responses was longer than for the other responses types. Furthermore, for sustained responses the peak firing rate in the within stimulus period predicted the duration of post-stimulus activity, indicating a positive relationship between excitatory drive during sensory stimulation and the duration of post-stimulatory activity observed. Future experiments could systematically vary sound duration in order to separate the mechanistic contribution of onset vs offset response dynamics to the generation of these responses. Purely excitatory mechanisms do not seem capable of accounting for the full response profiles of adapting and suppressed response types however. Inhibition is known to play a role in the generation of adapting onset responses 60 and thus may be responsible for the observed alteration in post-stimulus tuning observed for this class of responses. Greater adaptation of inhibitory synaptic input has been observed during sensory stimulation in neurons of the rat somatosensory cortex 61 . Such a pattern of adaptation could lead to excitation dominating during the postsynaptic period, providing a mechanism for post-stimulus firing in these units. Rebound calcium burst activity lasting hundred of milliseconds has been observed following sensory stimulation in the auditory thalamus 62 and may contribute to the emergence of the post-stimulus activity observed in this area. This mechanism may also contribute to the generation of post-stimulus activity in adapting and suppressed cortical units as rebound activity is produced following inhibitory activity 63 . We also observed units that showed dramatic suppression in response to auditory stimuli, followed by the longest duration post-stimulus activity that we observed. www.nature.com/scientificreports/ This dramatic suppression is in keeping with a role for inhibition in generating this post-stimulus activity and this class of responses may share a common mechanism with the adapting response type. An imbalance in balance of excitation and inhibition during the post-stimulus period in favour of excitation may therefore represent a common mechanism for all three response types observed here. The function of post-stimulus activity has been studied in a variety of behavioural contexts but the data presented here indicate that this activity may also play a role during passive listening. Echoic memory is a form of short-term sensory memory that differs from working memory in that it is posited to be active under all behavioral conditions, including passive listening 64 . The content of echoic memory is generally stored on the order of seconds 65-67 , a timescale that fits with the duration of post-stimulus activity observed here. An alternate possibility is that the activity reported here reflects the recruitment of circuits that exist in order to retain information in working memory and provide no function during passive listening. A functional role for this activity appears plausible, however, given the requirement of temporal integration in auditory perception in all species, from insects 68 to humans 69 . Beyond the auditory domain, multi-sensory integration requires combining sensory information from multiple cortical areas with different latencies and temporal dynamics 70 . Such post-stimulatory activity may provide a substrate for integrating disparate sensory signals into a coherent percept of an object. Investigating the function of activity observed in the absence of a behavioural task is particularly challenging however. Experiments addressing the function of this activity could exploit behavioural measures that provide an implicit measure of the processing of sensory information in the absence of a task, such as fear-conditioned freezing 71 . Combining this approach with temporally precise suppression of this activity through optogenetic www.nature.com/scientificreports/ methods, it may be possible to investigate the potential role of this activity in echoic memory and temporal integration. Developing an approach that will enable the investigation of the potential roles of post-stimulatory activity during passive listening will be crucial in elucidating the functional properties of such activity.

Methods
The data analysed in this study were obtained from several previous experiments conducted in the Laboratory of Auditory Neurophysiology at Johns Hopkins University School of Medicine in the laboratory of Prof. Xiaoqin Wang [37][38][39][40][41] . All experimental procedures were approved by the Johns Hopkins University Animal Use and Care Committee and were carried out in accordance with relevant guidelines and regulations. The methods to record single-unit activity in awake marmosets were previously described 61 and are briefly summarised below. Single-unit recordings from auditory cortex and thalamus were conducted in 7 awake, passively listening marmosets sitting on a semi-restraint device with their head immobilized, within a double-walled soundproof chamber (Industrial Acoustics). The inside wall of the chamber was covered by 3-inch acoustic absorption foam (Sonex). For auditory cortical recordings high-impedance tungsten microelectrodes (3-5 M, A-M Systems) were inserted perpendicular to the cortical surface. Electrodes were mounted on a micromanipulator (Narishige) and advanced by a manual hydraulic microdrive (Trent Wells). Action potentials were detected on-line using a template-based spike sorter (Multi-Spike Detector; Alpha Omega Engineering) and continuously monitored by the experimenter while data recording progressed. Typically 5-15 electrode penetrations were made within a miniature recording hole (diameter 1 mm) over the course of several days, after which the hole was sealed with Generation of acoustic stimuli. Acoustic stimuli were generated digitally and delivered by a free-field loudspeaker located one meter directly in front of the animal. All sound stimuli were generated at a 100 kHz sampling rate and low-pass filtered at 50 kHz. The sound level of individual frequency components used in this study was no higher than 80 dB SPL. Frequency tuning curves and rate-level functions were generated using pure-tone stimuli of 200 ms in duration with post-stimulus intervals of 500 ms, and had a minimum of five repetitions each. Stimulus presentation order was fully randomised. Pure-tone stimuli intensity levels were generally 10-20 dB above threshold for neurons with monotonic rate-level functions, or at preferred levels for non-monotonic neurons. Broadband rectangular clicks or narrowband clicks made of brief pulses of white noise or a tone (at an integer multiple of the frequency) were used to generate click trains. Rectangular click trains had a width of 0.1 ms while narrowband clicks had each pulse convolved with a Gaussian envelope with a standard deviation of 0.1-0.4.

Identification of cortical fields.
Single units with significant neuronal discharges to narrowband stimuli, such as tones and band-pass noise, were used to generate cortical characteristic frequency maps. The characteristic frequency of each location on the map was determined by the median characteristic frequency of all electrode tracks within 0.25 mm. Electrode track characteristic frequencies were calculated by computing the median characteristic frequency of units within the track. The anterior-posterior position is reported relative to the boundary between A1 and R. Both the boundary between A1 and R as well as the boundary between R and RT were identified from the frequency reversal between the cochleotopic gradients of the two fields, as their maps are mirror-reversed [72][73][74][75] . Please see 38 for more details.
Quantification of post-stimulus activity duration. Quantification of post-stimulus activity was based on significant difference from baseline firing. Baseline firing rate was calculated from a pre-stimulus interval that was separate from the post-stimulus interval of the preceding stimulus (Fig. 9A). A wide range of pre-stimulus interval, post-stimulus interval, stimulus duration and stimulus type combinations were used resulting in 82 different stimulation conditions. Several subsets with consistent durations were used in our analysis. Stimuli that had a post-stimulus interval of 300 ms had a pre-stimulus interval of 200 ms, creating an inter-tone interval www.nature.com/scientificreports/ of 500 ms. Stimuli that had a post-stimulus interval of 500 ms had a pre-stimulus interval of 500 ms, creating an inter-tone interval of 1,000 ms. For stimuli with a post-stimulus interval of > 500 ms, a range of pre-stimulus intervals from 200 to 8,000 ms were used that varied with the duration of the particular stimulus. Units were required to have a baseline firing rate, calculated across all pre-stimulus periods, of > 1 spike per second and to have been presented with 5 or more repetitions of each stimulus to be included in all further analysis. This criterion was met by 1,188 units of 1,557. The duration of post-stimulus activity was quantified by convolving spike trains for a given stimulus with a bi-directional Gaussian filter (σ = 5 ms, total bandwidth = 20 ms) in order to estimate the mean instantaneous firing rate before, during and after the presentation of the stimulus (Fig. 9B).
The same method was used in order to generate a mean firing rate trace for the pre-stimulus period, combining the data from all pre-stimulus periods for the unit (Fig. 9C). The mean and standard deviations of baseline firing were calculated from this trace (Fig. 9D). Significant post-stimulus activity was required to be > 2 standard deviations over mean baseline firing. Significant post-stimulus activity was defined as periods of firing that crossed the threshold (Fig. 9E). Firing during this period was required to be over the threshold on at least half of the trials, in order to avoid large single trial events being mistaken for reliable post-stimulus activity. The observed post-stimulus activity could be highly variable over time. Therefore, momentary drops below this threshold (< 20 ms) were not considered in our estimate of post-stimulus activity duration. In order to control for chance fluctuations in firing rates, the duration of elevated firing during spontaneous activity during the pre-stimulus period was also calculated. The duration of post-stimulus activity was required to be above the range of values that were observed by chance in the pre-stimulus period for each neuron. ethical approval. All experimental procedures were carried out at Johns Hopkins University in AAALAC approved facilities and were approved by the Johns Hopkins University Animal Use and Care Committee (IACUC). All methods were carried out in accordance with relevant guidelines and regulations.