Introduction

A universal feature of sensory processing is that the brain's response to a sensory stimulus depends on the recent history of sensory experience. This history dependence can take many forms: depending on the stimulus, brain area, and delay, one stimulus might increase or reduce responses to subsequent, similar stimuli. History-dependent increases and decreases in firing have different implications for how sensory stimuli are processed, and likely relate to differences in functional requirements. For example, for many cells in the auditory system, a repeated pure tone will reduce (or “depress”) responses to subsequent, similar tones, which likely enhances representation of novel sounds (Nelken, 2014). In others, sound stimuli can increase (or “facilitate”) responses to a subsequent sound stimulus, thus enhancing the representation of sound sequences (Brosch and Schreiner, 2000).

In the auditory cortex (AC), in addition to responding (mostly transiently) to the onset of sound, some neurons respond to the ends (or “offsets”) of sounds. Much is known about the history dependence of cortical sound “onset” responses, including the relative tendencies of responses to facilitate or depress (depression predominates), the time constants of facilitation or depression, and the relationship between cell type and facilitation or depression (Brosch and Schreiner, 1997; Ulanovsky et al., 2004; Wehr and Zador, 2005). Yet almost nothing is known about the history dependence of sound offset responses. Sound offset responses are important for many aspects of hearing, which are key to processing continuous sounds, such as speech (Liberman et al., 1954; Ghitza, 2011), including the encoding of sound duration (Malone et al., 2015; Li et al., 2021; Sołyga and Barkat, 2021), the discrimination of up versus down frequency-modulated (FM) sweeps (Sollini et al., 2018), and the detection of silent gaps in between sounds (Anderson and Linden, 2016; Weible et al., 2020). Thus, understanding how offset responses depend on recent stimulus history is likely to aid in our understanding of how the brain processes complex, naturalistic sounds.

Several aspects of AC circuitry support the prediction that onset and offset responses will be differently affected by recent sound history. The input pathways driving onset and offset responses are distinct and largely nonoverlapping (He, 2001; Scholl et al., 2010; Liu et al., 2019), which could allow differences in history dependence to be inherited from differences in the dynamics of their thalamic inputs. In addition, the many types of cortical interneurons may shape history dependence in diverse and specific ways. Prior work has shown that parvalbumin-positive (PV) and somatostatin-positive (SST) interneurons control separate types of history dependence in the cortex (Natan et al., 2015; Phillips et al., 2017), in part because of specializations in their own short-term synaptic dynamics. Because the types of cortical interneurons that tend to show onset versus offset responses are different (Liu et al., 2019; Li et al., 2021), this implies that cortical interneurons may shape onset and offset history dependence in distinct ways.

Here, we compare the prevalence of history-dependent facilitation and depression in offset and onset responses, both across populations of cortical neurons of different excitatory and inhibitory types and within individual neurons. We presented awake mice with a series of repeating noise bursts and silent intervals, recorded through layers of primary AC with a silicon probe, and measured the dynamics of onset and offset responses over sound repetitions. We separately tracked responses in opto-tagged populations of PV and SST interneurons and in untagged broad-spiking (putative pyramidal) and narrow-spiking (nonpyramidal) cells. We show that offset response magnitudes do depend on recent stimulus history and that they are less likely to depress than onset responses. Different cell types have different proportions of onset- and offset-responsive cells, and their onset responses, but not their offset responses, show diverse forms of history dependence. Last, we show that, while early onset firing tended to depress, later or more sustained firing was more facilitatory; in contrast, nearly all offset responses were transient, and there was almost no relationship between offset response duration, cell type, and history dependence.

Results

We recorded action potentials from 64 channel linear silicon probes implanted in the primary AC of n = 18 awake mice (male and female). We presented 120 trials of a sound stimulus consisting of four noise bursts, each 150 ms in duration and separated by 150 ms of silence. Each stimulus was followed by 8 s of silence (intertrial interval). This stimulus allowed us to measure changes in sound-evoked (“onset”) and sound-offset (“offset”) response magnitudes across repetitions of the same stimulus. Firing from an example unit is shown in Figure 1A.

Figure 1.

Identifying onset- and offset-responsive cells. A, Spike raster (top) and PSTH (bottom) showing firing from an example unit. Red shading represents onset response windows. Blue shading represents offset response windows. This unit responded both to the sound (onset) and the termination of the sound (offset). B, Histogram represents the spike-time reliability scores for all single units. Of 3118 single units, 2127 (green) showed a reliability score that was ≥0.3 and statistically significant (p < 0.01) (other units were not included in further analyses). C, Criteria for identifying unambiguous onset and offset responses. The PSTHs show example onset (left) and offset (right) responses. The response maxima (circles) exceed the 0.025 spikes/bin/trial above baseline threshold (orange dashed line), the 3× z-score threshold (dark yellow dashed line), and the 3× prior FR threshold (dark brown dashed line). D, Left, Count of units showing ≥1 onset response (red), ≥1 offset response (blue), both onset and offset responses (purple), or neither response type (gray). Right, Count of onset- and/or offset-responsive units as a function of cortical depth. E, Normalized PSTHs, grouped according to whether they showed onset responses only, onset and offset responses, offset responses only, or neither onset nor offset responses. F, Mean normalized PSTHs for all units. Groups are the same as in D and E.

Onset and offset firing in AC

Among the 3118 stable, well-isolated single units recorded from primary AC, 2127 (∼68%) showed significant stimulus-locked firing to this protocol (Fig. 1B, reliability ≥0.3 and significant at p < 0.01; see Materials and Methods). Only these units were analyzed further. We constructed PSTHs using 10 ms bins, defined onset and offset response windows for each of the four noise bursts (Fig. 1A), and defined a window as containing an unambiguous response if the maximal FR within the window exceeded 0.025 spikes/bin/trial, had a z score >3, and was at least 3× the mean FR in the 50 ms before the window (Fig. 1C). The 0.025 spikes/bin/trial and 3× z-score thresholds ensured that stimulus-driven firing exceeded spontaneous firing and was large enough to not be swamped by noise. The 3× prior FR threshold ensured that stimulus-driven firing unambiguously exceeded any lingering firing from a prior response. A unit that had onset responses to at least one of the four noise bursts was considered onset-responsive, and a unit that had offset responses to at least one of the four noise bursts was considered offset-responsive. Among the 2127 units with significant stimulus-locked firing (Fig. 1B), 891 (∼42%) showed onset responses only, 245 (∼12%) showed offset responses only, 849 (∼40%) showed both onset and offset responses, and 142 (∼7%) showed neither onset nor offset responses (these neurons were typically inhibited by the test stimulus) (Fig. 1D–F). These results differed depending on cortical depth (Fig. 1D, right). Cells in the shallower layers were more likely to show offset responses, whereas cells in the deeper layers were more likely to show onset responses only. These results are consistent with previous studies, which show that a substantial proportion of onset-responsive units also respond to sound offsets (>25% in previous studies; roughly half in this study), and onset firing is more common than offset firing (Sołyga and Barkat, 2019, 2021; Chong et al., 2020; Li et al., 2021).

Quantifying history dependence

We first aimed to quantify the prevalence of response facilitation or depression among onset and offset responses. For this, we had to make two analysis choices: first, how to measure response magnitude (e.g., as maximal response height, area under the curve [AUC], etc). Second, because in this dataset many cells showed responses that were so sustained as to bleed into the next response window (e.g., the first onset response in Fig. 2A, left), we had to choose whether to measure responses with respect to the pretrial baseline or with respect to the FR at the end of the prior response. We first analyzed facilitation and depression using the difference between the maximal response and the prior FR (Fig. 2A). In order to compare results across cells with very different FRs, we normalized changes in response magnitude to produce an FDI: for each onset response, we divided the change in response magnitude with respect to the first onset response by the size of the largest onset response, and we calculated offset response FDIs analogously but using only offset response magnitudes. If a response is facilitated with respect to the first response, then its FDI will be >0; and if a response is depressed with respect to the first response, its FDI will be <0.

Figure 2.

Offset responses are less depressive than onset responses. A, Before calculating the maximal FR in a response window (A, right, circles), the FR at the end of the prior response window (A, right, horizontal bars) was subtracted. B, Left, Onset FDIs for all units showing at least one onset response. PSTHs (Bi,Bii) show firing from example units with facilitatory (Bi) or depressive (Bii) responses. C, Same as in B, but for offset responses. D, Scatter-violin plot showing the FDI from the fourth onset (red) and offset (blue) response. Offset FDIs were significantly more positive than onset FDIs (p < 0.001). E, Onset and offset history dependence of responses to a stimulus consisting of either four 100 ms noise bursts separated by silent intervals of 100 ms (Ei–Eiv) or four 200 ms noise bursts separated by silent intervals of 200 ms (Ev–Eviii). Ei, Ev, Firing from example units in response to these stimuli. Eii, Evi, Count of units showing onset and/or offset firing in response to these stimuli. Eiii, Evii, Mean normalized PSTHs for the first (black) and fourth (red) onset response (left), or the first (black) and fourth (blue) offset response (right). All units showed either ≥1 onset response (left) or ≥1 offset response (right). Shaded error bars indicate SEM. Eiv, Eviii, Onset and offset FDIs for all units depicted in Eiii (Eiv) or Evii (Eviii). Offset FDIs were significantly more positive than onset FDIs for both stimuli (Eiv,Eviii, p < 0.001). ***p < 0.001.

Offset responses tend to be less depressive than onset responses

Figure 2B, C shows onset and offset FDIs from all units exhibiting at least one onset (Fig. 2B) or offset (Fig. 2C) response. Onset responses were more likely to have negative FDIs (i.e., more likely to depress than facilitate), whereas offset responses were roughly equally likely to have positive or negative FDIs (i.e., to be facilitating or to be depressing). Because the direction of response change was typically uniform across Responses 2-4, and depression or facilitation was typically strongest for Response 4, we performed all subsequent analyses on FDIs from Response 4 only. Indeed, the FDIs for offset Response 4 were significantly more positive than the FDIs for onset Response 4 (Fig. 2D, −0.35 ± 0.50 vs −0.07 ± 0.45, Mann–Whitney U test: p < 0.001), implying that for these stimuli, offset responses were less depressive than onset responses. Because previous studies have shown that the strength of depression and facilitation varies depending on interstimulus intervals (Brosch and Schreiner, 1997; Wehr and Zador, 2005), we wondered whether this result was specific to these particular stimulus timings. To test this, we recorded 15 penetrations from an additional 5 male and female mice, in which we presented randomly interleaved trials of repeating noise burst stimuli with different temporal characteristics (either 100 ms noise bursts with 100 ms internoise intervals, or 200 ms noise bursts with 200 ms internoise intervals; 240 trials in total) (Figs. 2Ei,Ev). Changing stimulus timing did not substantially alter the proportion of cells showing onset and/or offset responses (Figs. 2Eii,Evi, for 100 ms noise bursts and 100 ms internoise intervals, for 192 cells in total: onset only: ∼44%, onset and offset: ∼39%, offset only: ∼8%, neither: ∼8%; for 200 ms noise bursts and 200 ms internoise intervals, for 182 cells in total: onset only: ∼54%, onset and offset: ∼32%, offset only: ∼9%, neither: ∼5%), and offsets consistently remained less depressive than onsets (Fig. 2Eiii,Eiv and Fig. 2Evii,Eviii, for 100 ms noise bursts and 100 ms internoise intervals: onset FDI: −0.51 ± 0.45, offset FDI: −0.01 ± 0.56, Mann–Whitney U test, p < 0.001; for 200 ms noise bursts and 200 ms internoise intervals: onset FDI: −0.40 ± 0.40, offset FDI: −0.04 ± 0.43, Mann–Whitney U test, p < 0.001). These results suggest that offset responses are more facilitative than onset responses, regardless of stimulus timing.

Offset responses are still less depressive than onset responses when different analysis choices are made

We wondered whether our results were confounded by lingering firing driven by the prior stimulus in the noise burst (“prior FRs”), as this firing frequently bled across response windows (e.g., Fig. 2A, left). To test for this, we compared onset and offset FDIs calculated in three different ways. First, we calculated FDIs using prior FR-subtracted responses, as in the previous analysis (Fig. 3Ai, left; data identical to Fig. 2). Second, we calculated FDIs for the same cells, but using baseline-subtracted responses (Fig. 3Ai, middle). Third, we calculated FDIs using baseline-subtracted responses, but only from cells in which all four prior FRs returned to within 1 SD of baseline (Fig. 3Ai, right). We used a two-way ANOVA to analyze the effect of the calculation method on onset and offset FDIs (Fig. 3Aii). This test revealed a nonsignificant interaction between response type (onset vs offset) and calculation method (F(2, 6602) = 0.05, p = 0.95), and simple main effects analyses revealed that the FDI differed significantly between onsets and offsets (F(1, 6603) = 341.4, p < 0.001), but not between the three calculation methods (F(2, 6602) = 0.49, p = 0.95). Post hoc Tukey tests showed that offsets were significantly less depressive than onsets for all three calculation methods (Method 1: onset FDI: −0.36 ± 0.50, offset FDI: −0.07 ± 0.45, p < 0.001; Method 2: onset FDI: −0.34 ± 0.49, offset FDI: −0.06 ± 0.47, p < 0.001; Method 3: onset FDI: −0.34 ± 0.51, offset FDI: −0.05 ± 0.51, p < 0.001). These results show that offsets are more facilitatory than onsets, regardless of how prior FRs are handled.

Figure 3.

Offsets remain less depressive than onsets, even when different analysis choices are made. Ai, To assess the effect of prior FR subtraction on measured FDIs, we compared FDIs calculated in three different ways: using response maxima from which the immediately preceding FRs were subtracted (left, Method 1: -Pr), using response maxima from which the baseline firing was subtracted (middle, Method 2: -Base), and using baseline subtraction, but only including cells for which all four prior FRs returned to within 1 SD of baseline (right, Method 3: Pr0). Aii, Onset and offset FDIs for the three prior FR conditions (-Pr, -Base, Pr0). A two-way ANOVA revealed a significant effect of response type (onset and offset, p < 0.001), but not calculation method (p = 0.95). Bi, The FDI was plotted against the response maximum (in spikes/trial) for Response 1, for onset (left) and offset (right) responses. Marginal histograms (top) represent the distributions of first response maxima for cells with negative (cyan, depressive) or positive (orange, facilitative) FDIs. For both onset and offset responses, cells with negative/depressing FDIs on average had stronger first responses than cells with positive/facilitating FDIs (onset: p < 0.001; offset: p < 0.001). Bii, Across all cells, onset responses typically showed higher maximal FRs than offset responses (left, p < 0.001). Among cells matched for maximal FRs (middle), the offset FDIs were still significantly more positive than the onset FDIs (right, p < 0.001). Bi, Bii, 0.1 was added to all spikes/trial values to allow plotting these data on a log scale (for strongly facilitating or depressing cells, the maximal baseline-subtracted response within a given window occasionally fell between −0.1 and 0). C, PSTH (left) from an example unit showing the onset (red shading) and offset (blue shading) AUCs. AUCs are calculated with respect to the mean FR just before the response window (i.e., prior FRs). Right, Offset AUC FDIs were significantly more positive than onset AUC FDIs (p < 0.001). D, Onset (left) and offset (right) FDI as a function of cortical depth. Horizontal error bars indicate SEM. ***p < 0.001.

We additionally wondered whether our results related to differences in the magnitude of firing evoked by sound onsets and sound offsets. One mechanistic explanation for response depression is that the underlying synapses may be undergoing short-term synaptic depression through vesicle depletion, which is more prominent at stronger synapses with higher release probabilities (Tsodyks and Markram, 1997; Schneggenburger et al., 2002). Because of this, cells with stronger initial responses may show stronger depression. Indeed, in our dataset, cells with larger initial responses (i.e., in response to the first noise burst) were generally more depressive than cells with smaller initial responses, and this was true for both onset responses (Fig. 3Bi, left, 0.42 ± 0.46 spikes/trial vs 0.16 ± 0.25 spikes/trial, Mann–Whitney U test, p < 0.001) and offset responses (Fig. 3Bi, right, 0.19 ± 0.24 spikes/trial vs 0.12 ± 0.20 spikes/trial, Mann–Whitney U test: p < 0.001). Moreover, onset responses were stronger than offset responses (Fig. 3Bii, left, 0.37 ± 0.44 spikes/trial vs 0.32 ± 0.46 spikes/trial, Mann–Whitney U test, p < 0.001), which suggests that they may be more liable to depress. However, even when comparing populations of cells with equivalent initial (first) response sizes (Fig. 3Bii, middle), offsets were still significantly less depressive than onsets (Fig. 3Bii, right, mean FDI: −0.32 ± 0.58 vs −0.08 ± 0.45, Mann–Whitney U test: p < 0.001). These results indicate that the difference in onset and offset FDI is not because of differences in response strength. Next, we tested whether the tendency for offsets to be less depressive than onsets was still found when response magnitude was measured as an AUC instead of a maximum FR, and again found that offset responses were less depressive than onset responses (Fig. 3C, FDI: −0.29 ± 0.50 vs −0.11 ± 0.47, Mann–Whitney U test: p < 0.001). Finally, we checked to see whether our results were dependent on cortical depth. Previous research has shown that history dependence is different across cortical depth, with deeper layers showing more onset depression than shallower layers (Christianson et al., 2011). We wondered whether offsets were also more depressive in the deeper layers. To determine whether cortical depth was significantly related to FDI, and whether this relationship differed for onset and offset FDIs, we binned the fractional depth into seven linearly spaced bins and performed a two-way ANOVA (depth bin vs onset/offset) (Fig. 3D). This test revealed a significant interaction between response type and cortical depth (F(6, 2827) = 4.52, p < 0.001). Post hoc multiple comparisons tests confirmed that offsets were significantly less depressive than onsets for most of the cortical depth bins (5 of 7). This difference was not significant for two of the shallowest depth bins (the first and third shallowest depth bins), which is consistent with previous results (Christianson et al., 2011), that onset responses in superficial layers are less depressive than in deep layers.

In summary, these results show that, regardless of how they are measured or normalized, onset responses mainly depress in response to repeated sound stimulation, consistent with prior reports (Phillips et al., 2017; Seay et al., 2020). In contrast, and in addition to previous reports, offset responses are less likely to depress and more likely to either remain stable or facilitate.

Within cells, the history dependence of onset and offset firing does not correlate

We found that ∼40% of cells in our sample responded both to sounds and to sound offsets. We wondered whether, in these cells, there was any correlation between the history dependence of the onset response and the offset response; this would be important for understanding how neurons downstream might decode the information they carry, and also for establishing whether the mechanisms underlying this history dependence are mainly presynaptic versus postsynaptic. Figure 4A shows the normalized PSTHs for all units showing both onset and offset responses, showing that, among units with onset and offset responses, onset responses were on average depressive and offset responses were on average stable, but that there was considerable variability in the short-term dynamics for both onset and offset responses. To quantify this difference and determine whether the cells with the most depressing onset responses also had the most depressing offset responses, we compared their FDIs. Within units, the offset FDI was significantly more positive than the onset FDI (Fig. 4B, right, −0.37 ± 0.54 vs −0.09 ± 0.46, Wilcoxon signed-rank test: p < 0.001). However, there was no tendency for units with larger onset FDIs to also have larger offset FDIs (Spearman's ρ: −0.05, p = 0.18, n = 849), and the offset FDI was not significantly different for units with very negative (<−0.2), intermediate (≥−0.2 and ≤0.2), or very positive (>0.2) onset FDIs (Fig. 4C, Kruskal–Wallis test: χ2 = 0.4, p = 0.81). The lack of a clear relationship between onset and offset FDIs within units can further be seen in Figure 4D, which shows an alluvial plot illustrating the relationship between cells showing very negative (depressive), intermediate (stable), or very positive (facilitatory) responses. Within cells, all three onset FDI profiles (depressive, stable, facilitatory) coexist with all three offset FDI profiles, and no dominant relationships are apparent. These results suggest that, over the time intervals used here, history dependence is mainly implemented presynaptically (independently for different ascending stimulus pathways), consistent with previous research, which shows that onset and offset responses are primarily driven by nonoverlapping sets of synapses (Scholl et al., 2010; Liu et al., 2019).

Figure 4.

Within cells, onset history dependence does not correlate with offset history dependence. A, Normalized PSTHs (top) and the mean of these normalized PSTHs (bottom) from all units showing both onset and offset responses (n = 849). Rows are organized by the maximum offset response amplitude for each unit. B, The offset FDI plotted against the onset FDI, for all units showing both onset and offset responses (left). Right, The difference between offset and onset FDIs shows that offset FDIs tend to be significantly more positive (more facilitating) than onset FDIs in the same units (p < 0.001). C, The offset FDI for units that had both onset and offset responses, split according to whether the onset FDI was very negative (<−0.2, left), around zero (≥−0.2 and ≤0.2, middle), or very positive (>0.2, right). No significant difference was found between groups. D, The alluvial plot represents the diverse relationships between onset and offset temporal response profiles within units. Cyan represents depression, FDI <−0.2. Gray represents stable, FDI ≥−0.2 and ≤0.2. Orange represents facilitatory, FDI >0.2. Numbers on bars indicate number of cells. ***p < 0.001.

Onset history dependence varies between cell types, but offset history dependence does not

In vitro, thalamocortical inputs to AC show cell type-specific short-term synaptic plasticity. For example, inputs driving PV cells typically depress, whereas inputs driving SST cells typically facilitate (Tan et al., 2008; Takesian et al., 2013). We therefore wondered whether the short-term dynamics of onset and offset responses in vivo would vary between cell types. We used opto-tagging to identify PV and SST cells in Pv-Cre;Ai32 and Sst-Cre;Ai32 mice (Fig. 5A–E). We additionally split the nontagged cells into BS and NS groups because prior work has shown that most BS cells are excitatory and nearly all NS cells are inhibitory (Barthó et al., 2004) (Fig. 5E).

Figure 5.

Opto-tagging PV and SST neurons. A, Histology showing eYFP fluorescence (green) from a representative PV-Cre;Ai32 (left) and SST-Cre;Ai32 (right) mouse using coronal slices (50 μm thickness) containing the AC. B, Identifying opto-tagged cells. The photo-tagging stimulus consisted of 75 trials of a 10 ms blue light pulse, each interspersed by a 1.5 s intertrial interval (left). The waveforms on the right show the mean spontaneous (black) and light-evoked (blue) waveform for the same cell depicted on the left (cross-correlation between the two waveforms = 0.99). Orange text (labeled i-iv) represents the criteria required for a cell to be classified as successfully opto-tagged. C, Opto-tagged PV cells. From the 511 units recorded from all PV-Cre;Ai32 mice used in this study, 68 showed significantly elevated light-evoked firing, and 64 of these units passed all remaining opto-tagging criteria (left). Normalized PSTHs on the right (right, top) and the mean of these normalized PSTHs (right, bottom) show the light pulse responses of these 64 successfully opto-tagged PV cells. Each row in the top panel is a normalized PSTH from a single unit (amplitude expressed using color scale). D, Opto-tagged SST cells. Format the same as in C. Of the 2027 units recorded from all SST-Cre;Ai32 mice used in this study, 115 cells showed significantly elevated light-evoked firing, and 83 of these units passed all remaining opto-tagging criteria. E, Spontaneous FR plotted against the waveform trough-to-peak delay for all units recorded from PV-Cre;Ai32 (green) and SST-Cre;Ai32 (magenta) mice. The marginal histograms are normalized to have a maximum of 1. Nontagged (“Other”) cells were typically broad-spiking (trough-to-peak delay ≥600 μs, red dashed line). PV and SST cells were typically narrower spiking and showed higher spontaneous FRs.

We first determined the proportion of cells of each type showing onset and/or offset responses. Most PV cells responded both to sounds and to sound-offsets, and rarely responded to one but not the other (Fig. 6Ai). In contrast, SST cells typically responded only to sounds but not to sound-offsets (Fig. 6Aii), a result consistent with previous reports (Liu et al., 2019; Li et al., 2021). Some nontagged NS cells fired similarly to PV cells (i.e., firing to both sounds and sound-offsets), and others fired similarly to SST cells (i.e., firing to sounds but not to sound-offsets) (Fig. 6Aiii), consistent with the hypothesis that the NS cell population is a mixture of the PV and SST populations (for review, see Tremblay et al., 2016). Most cells in our recordings were broad spiking. These cells could show any combination of onset and offset responses, including “offset-only” and “neither” (i.e., sound-suppressed) firing (Fig. 6Aiv), patterns that were not observed among tagged or putative interneurons.