Earliest Experience of a Relatively Rare Sound But Not a Frequent Sound Causes Long-Term Changes in the Adult Auditory Cortex

Mouse ACX and inferior colliculus (IC) have auditory responses before ECO

We performed wide-field fluorescence imaging of the ACX in Thy1-GCaMP6f (Jax Labs) at early ages (Fig. 1A, left) before (N = 10 mice; P7-P8 = 3, P09 = 4, P10 = 4 mice) and after ECO (P13, Fig. 1A, right), to investigate the presence of sound-driven Ca²⁺ responses in the ACX. After ECO, by P13, mice show strong Ca²⁺ responses and defined rostro-caudal gradient of high to low frequencies (Stiebler et al., 1997), marking out the primary ACX (A1) and other auditory responsive regions (Fig. 1A, right). Since the auditory pathway is immature yet functional before ECO, we next investigate the presence of auditory responses to tones in the ACX of mice before ECO (De Villers-Sidani et al., 2007; Polley et al., 2013) without surgically opening the ear canals. We found that sound-evoked Ca²⁺ responses are present in the mouse ACX at high (90 dB SPL) to moderate intensities (70 dB SPL, in awake state) at ages before ECO (Fig. 1B and Fig. 1D, respectively). At very young ages, the percentage of the responding area is <5%. However, only a positive change in response is considered; hence, the significance is being tested at 2.5%. With increasing age, before ECO, a progressively higher percentage of areas of the cortical surface showed significant responses to multiple tone frequencies (Fig. 1B; Table 1), from 3% in P7 to 48.2% in P10 and then 84.2% in P13 after ECO (Pearson correlation, r = 0.96, p = 0.007, Fig. 1C). The percentage of frequencies of the tone stimuli tested that showed significant wide-field imaging responses also progressively increased (Table 1) from 20% in P7 to 71% in P13. Both of the above reflect the gradual maturation of the ACX, starting from the onset of sound-responsive activity at P7 to achieving tonotopy in A1 by P13. For further confirmation of sound-evoked responses in the auditory pathway before ECO wide-field fluorescence imaging in the IC of Thy1-GCaMP6f (Jax Labs) mice at P9-10 was performed. As in the ACX, we found significant sound-evoked responses in wide-field fluorescence on the IC surface (Fig. 1E, N = 2 mice, P9: 31.32%, P10: 38.06%).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Wide-field fluorescence imaging-based responses to tones in the ACX and IC of Gcamp6f mouse before ECO. A, Left, Experimental timeline. A, Right, Wide-field fluorescence imaging performed at P13 Gcamp6f mouse shows defined rostro-caudal gradient of high to low frequencies. Smoothed df/f traces with error bars indicating SEMs of selected regions (square) responding to selected frequencies (color) are shown to the right of each image. B, Wide-field fluorescence imaging-based significant response regions (white contours) in the ACX to 90 dB SPL tone in 4 different ages (P07-P10) before ECO. Black horizontal bar represents the stimulus duration. C, Percentage responding area at respective ages (Table 1). D, Same as in B, wide-field fluorescence imaging in the ACX based significant response to 70 and 80 dB SPL tone at P09 and P10 awake mouse. Respective df/f traces with SEMs of the selected region are shown at the bottom of each figure. E, Same as in A, wide-field fluorescence responses before ECO in the IC at P9 and P10 to 90 dB SPL tone. Purple “.” represents the first significant window. Sound-evoked Ca²⁺ responses are present in the mouse ACX at high (90 dB SPL) to moderate intensities (70 dB SPL) at ages before ECO (one-sided paired t test): *p < 0.05; **p < 0.01; ***p < 0.001; black, unless the p-value is specified).

L2/3 and L4 ACX single neurons have responses before ECO

The presence of significant auditory responses in Ca²⁺-dependent fluorescence in wide-field imaging (Fig. 1) could potentially indicate either the activity of input axons or spiking responses of neurons in the ACX, or both. To rule out only input axon activity in the ACX wide-field imaging, we performed 2-photon Ca²⁺ imaging of single neurons in L2/3 and L4 of the ACX of mice (N = 14 mice) before ECO and after ECO (N = 2 mice; P12-P13) (Fig. 2A) with high to moderate intensities of tones. Each 50-ms-long tone stimulus was repeated 40 times and consisted of 3-7 frequencies from 6 to 48 kHz, half octave apart, presented at 60-90 dB SPL, with a 5-7 s gap between repetitions. Single neurons in L2/3 in partially anesthetized mice showed Ca²⁺ transients in response to tones at ages before ECO (P7-P9, Fig. 2B, top). The population mean df/f in response to different frequency tones (Fig. 2B, bottom) was investigated, considering all the neurons that responded with at least 25% reliability (fraction of repetitions, showing a significant response; see Materials and Methods). We found an increase in the mean population response strength before ECO (N = 14 mice; P07-P08 = 3, P09 = 5, P10-P11 = 6, ANOVA with Tukey post hoc analysis, F_(2,576) = 33.5, p <10⁻⁴, Fig. 2C). The above indicated a gradual maturation of responses up to ECO (P12-P13) in L2/3 ACX neurons. Along with an increase in tone response strength over age, response reliability increased significantly with age (Fig. 2C, bottom, N = 14 mice; P07-P08 = 3, P09 = 5, P10-P11 = 6, ANOVA with Tukey post hoc analysis, F_(2,6955) = 375.84, p < 0.0001) before ECO, also showing gradual response development. The percentage of overall cases (imaged neuron–tone stimulus pairs) that showed significant responses also showed a gradual increase (Pearson correlation, r = 0.97, n = 6, p < 10⁻³) with age from 2.32% at P7 to 20.2% at P12/13 (Table 1). Furthermore, neuron–tone stimuli pairs also showed higher significant responses to the presented different frequencies (Table 2) as age progressed. All the above observations indicate the presence of sound-driven Ca²⁺ activity before ECO in single neurons and that such activity varies with age in a manner expected from increasing maturity of the neural pathway.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Sound-evoked activity of ACX L2/3 and L4 neurons in the Gcamp6f mouse before ECO. A, Experimental timeline. B, Top, Example traces of df/f of single L2/3 neurons in response to 0 dB attenuation tones, respectively, at three different ages (P08, P09, and P13). Black vertical line on the bottom of each trace indicates stimulus onset. Corresponding green vertical shading bar in each example trace represents a significant moving window (3 frames), which shows the earliest significant p-value among all the frames (corresponding p-value shown above, one-sided paired t test, *p < 0.05). Bottom, Population mean df/f of the age groups (P07 = 35, P08 = 51, P09 = 299, P10 = 110, P11 = 84, P12-P13 = 707, neuron–tone stimulus pairs; Table 2) of L2/3 neurons shown above. C, Top, Box plot showing response peak strength of all L2/3 neurons imaged at different ages (14 mice; P07-P08 = 3, P09 = 5, P10-P11 = 6, ANOVA with Tukey post hoc analysis, corresponding significance value shown to the right). C, Bottom, Box plot showing response reliability of all L2/3 neurons imaged at different ages, as shown above (ANOVA with Tukey post hoc analysis, ***p < 0.001, corresponding significance value shown to the right). D, Similarly, as shown in B, example traces of df/f of single L2/3 neurons of ACX in response to 60 dB-90 dB SPL tones at age group P9-P10 of an awake mouse, with corresponding significance values of the earliest significant window (3 frames) obtained. Bottom, Population mean df/f of L2/3 neurons across different sound intensity level (60-90 dB SPL, 4 mice; 60 dB: 107, 70 dB: 556, 80 dB: 348, 90 dB: 359 neuron–tone stimulus pairs; Table 3). E, Extended 2-photon Ca²⁺ imaging data, showing population mean df/f (below) at two ages (P09, 2 mice, 666 neuron–tone pairs and P10, 138 neuron–tone pairs) before ECO, indicating long (>2 s) spontaneous window. F, Population mean df/f of the same age groups in response to sound (left) and the same neurons without sound stimulation (right), with 40 repetitions in each case (P09-P10, 3mice, 461 neuron–tone pairs). G, Left, Example traces of df/f of 3 different L4 neurons (cyan) in response to 0 dB attenuation tones at two different frequencies (top row, 10 Khz and bottom row, 17 kHz). Green patch represents the earliest significant moving window obtained with respective significance values. Right, Population mean df/f of L4 neurons which showed significant responses (P9-10, 2 mice, 64 neuron- tone pairs). L2/3 and L4 ACX single neurons respond to sound before ECO (one-sided paired t test, *p < 0.05; **p < 0.01; ***p < 0.001, black asterisk, unless the p-value is specified).

In a subset of mice with stable recordings from neurons at all the different frequencies tested (6-48 kHz, 0.5 octave steps, in N = 7 mice, P7-P08 = 2, P09-P10 = 3, P12-P13 = 2, of 14 mice reported above), we investigated the tone peak response strength and response reliability to the BF of the neuron. The above allows ruling out any bias of the effect of possible same few neurons responding to different frequencies. We found a significant increase in strength (ANOVA with Tukey post hoc analysis, F₍₇₂₀₎ = 82.02, p < 10⁻³) and reliability (F₍₇₂₀₎ = 215.98, p < 10⁻³) of the response of neurons to tones at their BF as a function of age. Furthermore, the percentage of responding neurons also increases with age from P7-P8 (18.75%) to P12-P13 (78.40%).

The wide-field Ca²⁺ responses at moderate intensities (Fig. 1D) were also ruled out to reflect only input axon activity as single neurons of L2/3 showed clear Ca²⁺ transients in response to tones down to 60 dB SPL (Fig. 2D, top) in awake young animals. The population mean df/f traces of all neurons (N = 4 mice; 90 dB: 359, 80 dB: 348, 70 dB: 556, 60 dB: n = 144 neuron–tone stimulus pairs) that responded significantly (with 25% reliability; see Materials and Methods) to tones of different frequencies (8.5 kHz: 28.52%, 12 kHz: 52.15%, 17 kHz: 62.03%) at the four intensities used, showed an increase in response strength with increasing intensity, except 80 dB SPL (Fig. 2D, bottom). The percentage of cases of neuron–tone stimulus pairs that showed significant responses also showed a similar trend, increasing from 14% at 60 dB SPL to 69.57% at 90 dB SPL (Table 2).

The previous set of data were collected with usually a little over 1 s of baseline preceding the stimulus. In order to rule out a contribution from any coincident spontaneous oscillating activity (Blankenship and Feller, 2010) to the response, we obtained extended 2-photon imaging data before ECO (N = 3 mice, P9-P10, 6 ROIs) with >2 s of baseline (Fig. 2E) and also with and without sound stimulation (Fig. 2F). The population mean df/f traces of all neurons in L2/3, showing significant activity above baseline in response to tones at P9 (n = 666 neuron–tone pairs) and P10 (n = 138 neuron–tone stimulus pairs), portray the same nature of responses as in Figure 2B. The long duration mean baseline activity from 40 repetitions fluctuated at ∼0, as expected. We confirmed the presence of sound-evoked responses before ECO, considering any coincident ongoing spontaneous oscillations by comparing mean df/f traces of all neurons imaged in with and without sound cases (N = 3 mice, n = 461 neurons-tone pairs, P9-P10, Fig. 2F). No significant activity was observed under the no-sound condition.

As in L2/3, putative L4 single neurons based on the depth of 2-photon imaging (300-350 μm from pia) also showed Ca²⁺ responses. Mean df/f-based responses of three representative L4 neurons to 2 frequencies are shown in Figure 2G (left). As in L2/3, the population mean of all significantly responding cases in L4 (N = 2 mice, n = 64 neuron–tone stimulus pairs) showed responses to sound of different frequencies (8.5 kHz: 10.34%, 12 kHz: 11%, 17 kHz: 10.94%, 24 kHz: 12.80%, 34 kHz: 7.14%) with >2 s of baseline activity showing no spontaneous oscillatory activity (Fig. 2G, right). Thus, based on 2-photon Ca²⁺ imaging of single neurons, we find both L2/3 and L4 to be tone sound-responsive before ECO, showing responses as early as P7 and down to moderate intensities of 60 dB SPL. The above data indicate that wide-field fluorescence imaging-based sound responses in ACX are not because of only input axon-related activity.

SPNs in mice are driven by sounds before ECO

Two-photon Ca²⁺ imaging of single neurons, simultaneously from L4 and SPNs, was beyond our scope (Yildirim et al., 2019). Although SPNs are known to be the first to respond to sounds in ferrets, equivalent properties of SPNs in the mouse at the corresponding age of P6 (Wess et al., 2017) are unknown. Further, our Ca²⁺ responses do not show the nature of spiking activity in L4, thus requiring extracellular electrophysiology in mice before ECO. To reliably target SPNs with electrodes at different developmental ages, we first characterized the depth of SPNs based on SPN-specific markers (Hoerder-Suabedissen et al., 2009; Osheroff and Hatten, 2009) CTGF and Cplx3 (Fig. 3A, B, green and red, respectively) using IHC. Furthermore, cell morphology from Nissl stains (Viswanathan et al., 2012) (Fig. 3D) was also used to characterize the SPNs depth. Electrode tracks were also routinely confirmed to be in the SPN layer (P12, P13, and P18, Fig. 3C, top). In a subset of experiments in SCNN1-td-tomato mice (Jax 9613, P14, Fig. 3C, bottom) with staggered electrodes, confirmation of electrodes being simultaneously in L4 (L4 neurons expressing td-tomato) and SPN (from IHC, Fig. 3C) was done (N = 3 mice, P12, P14, and P17). The thickness of the SPN layer reduced with age, as known (Wang et al., 2010; Hoerder-Suabedissen and Molnár, 2013). At multiple ages, in subsets of experiments, confirmation of electrode location in ACX was done with DiI crystal insertion in the recording site and observing labeling of MGBv (Fig. 3E). Also, in higher age groups, A1 recording was confirmed with the presence of general tonotopy (Fig. 3F) based on the spatial location of BFs of units from multiple electrode penetrations.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Histologic confirmation of recording depth for L4 and SP. A, IHC showing labeling of SP neurons with antibodies of SPN markers, Cplx3 (red) and CTGF (green). The location of SPN is marked by an arrow (magenta). B, SP depth determined with IHC by labeling SPNs with SPN markers Cplx3 (red) and CTGF (green). C, Slices were cut from brains (fixed) in which single-unit recordings were performed in the ACX (representative examples at 3 ages, P12-P14, top), and IHC was performed. Electrode tracks (dotted white lines) show recordings performed in SPN and, in the case of dual recordings in the SCNN1-td-tomato mouse (P14), show electrodes in L4 and SPN simultaneously. The location of SP (magenta arrow) and L4 (cyan) neurons is marked with respective colored arrows. D, Depth and thickness of SP as a percentage of cortical thickness over age determined from cell morphology with Nissl stains (top). Colored arrows on Nissl images indicate the respective location of L4 (cyan) and SP (magenta) layers. E, Representative confirmation of recording in A1 in lower age groups P12, P13, P18 with DiI injections in the recording site with labeling in MGBv (inset). F, Sample locations of recordings of single units mapped onto the cortical surface showing broad clearer tonotopy in P29 than in P15. Black arrow indicates the location of a branch of the rhinal vein, representing a landmark for ACX.

We next performed extracellular recordings in the mouse before ECO (P7-P12) from the ACX SPNs and L4 simultaneously with custom-made staggered electrodes arrays (4 × 4 arrays x-y spacing 125 μm, 8 in SPN and 8 in L4, 400 or 600 μm apart in-depth, Microprobes for Life Sciences). Raw LFPs have been used extensively in the literature to support responses (Liu et al., 2015; De Assis et al., 2016), especially to refer to local synaptic activity in multiple species (Kafaligonul et al., 2018; Sołyga and Barkat, 2019) and multiple brain areas (Galindo-Leon and Liu, 2010; Kafaligonul et al., 2018). Sound-evoked LFPs (see Materials and Methods) simultaneously obtained from SP (≥800 μm; P8-P9, N = 3 mice, responsive/total sites: 28 of 44 (63.63%) and P10-P11, N = 4 mice, responsive/total sites: 44 of 72 (61.11%) and L4 (300-350 μm, P8-P9, N = 3 mice: responsive/total sites: 27 of 44 (61.36%) and P10-P11, N = 4 mice: responsive/total sites: 52 of 72 (72.22%) locations) before ECO showed significant sound-driven activity (Fig. 4A,B). Longer-lasting oscillatory LFPs (mean slope difference; see Materials and Methods, L4: F_(2,390) = 62.07, p < 0.001; SP: F_(2,743) = 121.97, p < 0.001) were observed at ages before ECO (Fig. 4A,B) than after ECO (P12-P13, Fig. 4C). The above results are in accordance with the pattern of LFP activity observed in ferrets at ages before ECO (Wess et al., 2017). The population mean LFPs (population data, Fig. 4A,B) were quantified with the RMS values in 50 ms bins normalized by the mean RMS values of the baseline (200 ms preceding the stimulus onset). Significant responses in both L4 (P8-P9; one-sided paired t test, significant responding site-tone pairs (n), n = 65, t₍₆₄₎ = 8.18, p < 0.0001, P10-P11; n = 140, t₍₁₃₉₎ = 11.42 p < 0.0001) and SP (P8-P9; n = 72, t₍₇₁₎ = 9.377, p < 0.0001, P10-P11; n = 138, t₍₁₃₇₎ = 10.43 p < 0.0001) were observed before ECO that last for >1 s (population data: Fig. 4A,B, top and bottom) unlike after ECO (population data: Fig. 4C, top and bottom, L4 and SP, respectively). Moreover, the peak RMS change (within 500 ms after stimulus onset), a reflection of response strength, increases with age (L4: F_(2,390) = 80.87, p < 0.001; SP: F_(2,743) = 138.36, p < 0.001).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

L4 and SPNs in the mouse ACX respond to tones before ECO. A, B, Example traces LFP in a single channel in L4 (cyan, top) and SP (magenta, bottom), in response to tone at age, P09 and P10. Green patch represents the window across which the significance was calculated (L4 and SP ACX single neurons respond to sound before ECO; one-sided paired t test: *p < 0.05; **p < 0.01; ***p < 0.001; see Materials and Methods). Horizontal red bar represents the stimulus onset. A, B, Population data: Population average of all the significant recording sites and SEM of RMS values of LFP with corresponding p-values (one-sided paired t test: ***p < 0.001). Red vertical line indicates stimulus onset. C, Similar plots of single-site and mean population LFP activity at ages after ECO (P12-P13), as shown in A and B, respectively. D, E, Example traces: Dot raster and z score of an L4 (cyan, top two plots) and SP (magenta, bottom two plots) neuron, in response to tone at age, P9 and P10. Red vertical line indicates stimulus onset. Right, Inset, Spike shape. D, E, Population data: Population mean PSTHs of all the recorded SP (P8-P9, 4 mice; P10-P11, 7mice) and L4 (P8-P9, 4 mice; P10-P11, 9 mice) neuron–tone pairs across different age groups which had a z score > 2 in at least 1 time bin within 500 ms after stimulus onset, with respective significance values (one-sided paired t test). Red vertical line indicates the stimulus onset. F, Similar plots of single-unit and mean population spike activity at ages after ECO (P12-P13) as shown in D and E, respectively. G, Latencies of L4 and SP neurons recorded simultaneously in P8-P11 (before ECO), P12-P13 and P14-P15 age groups. H, The population mean latencies revealed lowering of latencies with age and SPNs having lower latency than L4 neurons at ages before ECO and P12-P13 (one-sided unpaired t test: *p < 0.05; **p < 0.01).

Single units in L4 (P8-P9, N = 4 mice; P10-P11, N = 9 mice, cyan) and SPN (P8-P9, N = 4 mice; P10-P11, N = 7 mice, magenta) layer showed the presence of spiking responses (z > 2) to tones before ECO (Fig. 4D,E, top and bottom rows, respectively) and after ECO (Fig. 4F). Sample dot raster (top) and z scores (bottom) of spiking responses in 50 ms time bins are shown in Figure 4D, E, indicating the presence of significant sound-evoked spiking before ECO. Population PSTHs of all significantly responding neuron–tone pairs (n) in SP (P8-P9; one-sided paired t test, n = 30, 30 of 82 (36.59%), t₍₂₉₎ = 5.17, p < 0.001; P10-P11, n = 107, 107 of 487 (21.97%), t₍₁₀₆₎ = 11.48, p < 0.0001) and L4 (P8-P9; n = 30, 30 of 112 (26.79%), t₍₂₉₎ = 6.46, p < 0.001; P10-P11, n = 261, 261 of 678 (38.5%), t₍₂₆₀₎ = 15.90, p < 0.001) show significant sound-driven activity before ECO, which get stronger with age (Fig. 4D–F).

The response peak latencies (see Materials and Methods) of L4 and SP single units were obtained at ages before (P09 example trace, Fig. 4G) and after ECO (P12 example trace, Fig. 4G). Before ECO (neuron–tone pairs; n_SP = 107 and n_L4 = 193) and at P12-P13 (n_SP = 108 and n_L4 = 146), the mean peak latencies of SPNs were found to be significantly lower than L4 neurons (one-sided t test, before ECO, t₍₂₉₈₎ = −1.86, p = 0.031; P12-P13, t₍₂₅₂₎ = −2.935, p = 0.002, Fig. 4H). The above results suggest that SPNs respond at shorter latency relative to L4 neurons before ECO, and the response properties mature with age. Furthermore, the L4 (ANOVA with Tukey post hoc analysis, F₍₃₉₄₎ = 166.63, p < 0.001) and SP (F₍₂₃₃₎ = 62.65, p < 0.001) responding latencies decreased with age. Thus, based on the data from multiple techniques, we conclude that neurons in the different cortical layers and SP layer respond to sounds before ECO without surgically opening the ear canal. Auditory responses are present as early as P7 in mice and respond down to 60 dB SPL tones.

SPNs, unlike L4 neurons, show oddball selectivity before ECO

Having established the presence of auditory responses before ECO, we investigate the response properties of SPNs and L4 neurons in the context of stimulus-specific adaptation (SSA) (Ulanovsky et al., 2003; Malmierca et al., 2014; Pérez-González and Malmierca, 2014). We use oddball stimuli with a stream of standard tokens (S) with an embedded deviant token (D) to mimic the natural situation of a continuous auditory environment with important sounds occurring with low probability (Ulanovsky et al., 2003; Pérez-González and Malmierca, 2014) (Fig. 5A). The stimuli consisted of 50-ms-long sound tokens presented at 3.3 or 4 Hz of the pattern SSS…SDS…S (D at eighth or seventh position of 15 or 10 total tokens) for the oddball response characterization. One of the S and D tokens was a tone (f, range 6-34 kHz) and the other a broadband noise (N, bandwidth 6-48 kHz with equivalent sound level). Responses were collected in pairs with the S and D swapped to normalize for a neuron's inherent selectivity for a particular S or D (Fig. 5A). Each SD set of a pair was presented with 10-40 repetitions and 5-7 s silence duration between each iteration. A common selectivity index (CSI) was used to quantify a neuron's deviant detection strength as follows: CSI=RD(f) + RD(N)−RS(f)−RS(N)RD(f) + RD(N) + RS(f) + RS(N) (Ulanovsky et al., 2003) where R_S(.) and R_D(.) denote response to S and D, respectively. Two cases for R_S(.) were considered: mean response to the first S or onset and mean response for all the S's; the deviant detection strength values are referred to as CSI(F) and CSI, respectively. To define oddball selectivity, if a neuron evoked higher responses to D than S in an oddball sequence with corresponding CSI value positive, it was considered to exhibit oddball selectivity. We refer to spike rate-based oddball selectivity as deviant detection. Both the metrics helped us to quantify the deviant preference. The response to the first S token is devoid of any underlying context or adaptation. Thus, CSI(F) measures the relative preference of a deviant stimulus with respect to a stimulus presented by itself. Taking the mean response to all the standard tokens, the CSI metric finds this deviant preference with respect to an adapted standard stimulus and thus measures SSA. We use both the metrics CSI and CSI(F) to quantify SSA and deviant selectivity. Positive CSI(F) indicates a higher preference for deviants, and that it is not because of adaptation to the standard but because of the context of the deviant stimulus presented after a stream of familiar standard stimuli.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

SPNs, unlike L4 neurons, show higher oddball selectivity before ECO. A, Schematic represents the tone-noise (f-N) oddball stimulation protocol with the swap on the right. B, PSTHs of a single SP neuron, showing higher response at deviant tone (f, 8.5 kHz) token (swap, 250 ms response window gray patch, right). C, Dot raster and PSTH of the same SP neuron showing response to the same tone (8.5 kHz), when presented as tone-pip stimuli. D, First two columns, Example PSTH and z score of 2 L4 neurons (top rows) and 2 SP neurons (bottom rows), showing response to tone (f, standard in green and deviant in red, top) and noise (N, standard in green and deviant in red, bottom). Light gray patch represents the mean rate window (250 ms) across which CSI(F) was computed. E, Mean population spike rate and mean z score of all L4 and SP neurons in response to tone (f, standard in green and deviant in red, top) and noise (standard in green and deviant in red, bottom) presented at 3.33 Hz. E, Last 2 columns, Mean population spike rate and z score of L4 and SP neurons in responses to tone and noise, delivered at 4 Hz. Gray vertical line indicates a z score of 2. F, Population (3.3 and 4 Hz combined) cumulative distribution of CSI(F) and CSI for SP and L4 (magenta and cyan trace, respectively) shows significantly higher oddball selectivity in SP (one-sided unpaired t test, ***p < 0.001). Similar cumulative distributions with mean CSI calculated for each of 1000 random pair of SD response windows selections for S response and D response either before deviant (light dashed lines, F, left) or after deviant (light dashed lines, F, right). No significant difference between L4 and SP in >90% of the random window sets. Triangles: Magenta and cyan represent mean CSI(F) or CSI for SP and L4 neurons. G, Top and middle rows represent the mean CSI values of the population of L4 and SP neurons, respectively, for different choices of response latencies and response window sizes. *Response latency and response window size combination used in F. The two columns are for the two different CSIs used CSI and CSI(F). Bottom row represents which combinations of response attributes have significantly higher CSI in SP than in L4 (brown, blue indicates NS).

ACX L4 (300-350 μm, 11 mice, 84 neurons) and SP (>800 μm, 11 mice, 34 neurons) single-unit responses were obtained before ECO in response to the presentation of the SD stimulus for f-N and its swap. Sample PSTHs of SPN before ECO show the oddball's preference or to the deviant tone (f) token (Fig. 5B). The tone of a particular frequency was chosen based on the frequency that evoked a significant response (z > 2, Fig. 5C). The SD stimulus–response generally showed peaks following the D-token response, mainly in SPN and the first S-token or onset, primarily in L4 (Fig. 5D, L4 top row, SP bottom row). The earliest response peaks of L4 and SPN at ages before ECO (P9-P12) with the SD stimuli occur in a 250 ms window with a latency of 50-100 ms from the first token and in a similar window following the deviant (Fig. 5D). Comparing the mean response to the S-token and that of the D-token (Fig. 5D; S: green, D: red) shows the above. In the overall population, the mean response before ECO (P9-P12) for all the L4 neurons and SPNs recorded indicates the presence of deviant detection in SP and not L4 (Fig. 5E). Further, based on the spontaneous activity (100 ms preceding the stimulus), z scores of the example neurons (Fig. 5D) and population of neurons (Fig. 5E) demonstrate a significant response to D and S and that in the population, SP has significant activity at the deviant and not L4. To quantify the oddball's selectivity, we obtained spike rates in 250 ms windows because of the width of the peaks and calculated CSI(F) and CSI for all SPNs and L4 neurons. We observed higher deviant detection in SP (n = 34) than in L4 (n = 84) for both CSI(F) (one-sided unpaired t test, t₍₁₁₆₎ = 3.954, p < 0.001, K-S test, p = 0.007, Fig. 5F, left). and CSI (t₍₁₁₆₎ = 3.666, p < 0.001, K-S test, p = 0.002, Fig. 5F, right).

To rule out spurious deviant detection because of coincident spontaneous oscillating activity during the stimulus, we also considered CSIs expected by chance. Two nonoverlapping windows 250 ms long were taken randomly in the standard region before the deviant or after the deviant, not considering the deviant response. The mean CSIs were not significantly different between L4 and SP in >90% of the random window sets. More importantly, the distributions of mean random CSIs for SPN and L4 showed that the means were not significantly different from 0 (Fig. 5F, light dotted lines). Further, to ensure that our results were not sensitive to the choice of latency and window size, we also varied the response window from 50 to 300 ms (in steps of 50 ms) and latency from 0 to 350 ms (in steps of 50 ms). We obtained the CSIs for all neurons, and we found for what combinations SPNs had significantly higher oddball selectivity than L4 neurons (Fig. 5G). The same results were obtained for a substantial range of values around the chosen latency and window size, indicating deviant detection in SPN and not in L4 before ECO.

Unlike before ECO, L4 shows higher oddball selectivity than SP following ECO

Oddball selectivity in adults is a common phenomenon in all cortical layers of ACX (Szymanski et al., 2009). Given the observed absence of oddball selectivity in populations of L4 neurons before ECO, we hypothesize that such selectivity in L4 starts after ECO. Thus, we next investigated how the oddball selectivity changes in L4 and SPNs after ECO. Single-unit responses to the same SD paradigm with f-N sound tokens (schematic: Fig. 6A) were collected simultaneously from L4 (P12-P13, N = 10 mice, n = 45 units, 57 cases; P14-P15, N = 8 mice, n = 66 units, 115 cases; P16-P21, N = 4 mice, n = 20 units, 31 cases; P22-P28, N = 6 mice, n = 38 units, 97 cases; >P28, N = 7 mice, n = 44 units, 63 cases) and SPNs after ECO from P12 to the juvenile/adult at >P28 (P12-P13, N = 10 mice, n = 54 SPN units, 101 SPN cases; P14-P15, N = 8 mice, n = 59 SPN units, 67 SPN cases). We again quantify the selectivity based on CSI(F) and CSI. Comparison of population PSTH segments in response to tone (Fig. 6A, top two rows) or the noise (Fig. 6A, bottom two rows) as the first S token (green), all S tokens (blue), and the deviant (red) show how selectivity to the oddball changes in L4 and SP over age after ECO starting from P12 to P13. Since the inherent selectivity to the f or N confounds direct comparison, we compare CSIs over age (Fig. 6B), including the CSIs before ECO. L4 neurons start showing oddball selectivity immediately after ECO, which gradually declines to stable values with age until adulthood (CSI(F): ANOVA with Tukey post hoc analysis, F_(4,358) = 37.68, p < 0.0001; CSI: F_(4,358) = 12.74, p < 0.0001). On the other hand, SPNs with a significantly lower number at P12-P13 compared with peak at P8 (Wang et al., 2010) show significantly lower CSIs than L4 after ECO (CSI(F): one-sided unpaired t test, P12-13_L4 vs P12-13_SP, t₍₁₅₆₎ = −1.944, p = 0.027; P14-P15_L4 vs P14-P15_SP, t₍₁₈₀₎ = −1.10, p = 0.13; CSI: P12-13_L4 vs P12-13_SP, t₍₁₅₆₎ = −1.6213, p = 0.050; P14-P15_L4 vs P14-P15_SP, t₍₁₈₀₎ = −1.41, p = 0.07). We considered both CSI(F) and CSI since at early ages because of high adaptation (decrease in time constant with age: F_L4(4,455) = 11.92, p < 10^−4f) in L4 neurons with robust responses at the first S token (responding tokens increases: F_L4(4,455) = 6.34, p < 0.0001) and not any others. The same trend was observed in both CSIs, showing that before ECO, SPNs are stronger deviant detectors than L4, while the relative deviant detection strength switches after ECO.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Sound-evoked activity of ACX and MGB neurons after ECO. A, Schematic showing the tone-noise (f-N) oddball stimulation protocol with the swap. Mean population PSTH of all significant responding neurons (top, L4 neurons; bottom, SP neurons), indicating the deviant response (red), the first standard response (green), and the mean of responses at all the standards (blue) to tone (top 2 rows) and noise (bottom 2 rows) for different age groups after ECO. B, Overall mean oddball selectivity for noise and tone tokens measured by CSI(F) (top) and CSI (bottom), respectively, as a function of age. C, Similar mean population PSTH for tone-tone (f1-f2) oddball stimulation protocol. D, Same as in B, overall mean oddball selectivity for tone stimuli measured by CSI(F) and CSI, respectively, as a function of age. E, Left, Example of electrode tracks (white arrowheads) shown in Nissl stains of brain slices showing recording locations in MGB of the thalamus. E, Right, Magnified view of the relevant part of a coronal section with MGB, in which single-unit recordings from MGBv were performed (P14). F, Mean and SEs of CSI(F) (top) and CSI (bottom) for the thalamic neurons, by age groups. *p < 0.05. **p < 0.01. ***p < 0.001.

Above, we used N-f SD stimuli to mimic a repeating broadband auditory environment with the tone as a low probability stimulus (like a pup or adult mouse vocalizations in a normal auditory environment) (Holy and Guo, 2005). Since deviant selectivity has been studied primarily with two tones (Ulanovsky et al., 2003; Pérez-González and Malmierca, 2014) (f₁ and f₂), we also used an f₁-f₂ SD stimulus set to investigate whether our results also hold in such a case. Further, using tones as both tokens allows more direct comparisons between responses. f₁ and f₂ were spaced 1/4, 1/2, or 1 octave apart. Data were obtained from multiple L4 (P12-P13, N = 3 mice, n = 28 units, 47 cases; P14-P15, N = 3 mice, n = 41 units, 63 cases; P16-P21, N = 3 mice, n = 11 units, 15 cases; P22-P28, N = 4 mice, n = 24 units, 53 cases; >P28, N = 3 mice, n = 26 units, 74 cases) and SP (P12-P13, N = 3 mice, n = 9 SPN units, 14 SPN cases; P14-P15, N = 3 mice, n = 34 SPN units, 51 SPN cases) single units across different developmental windows. Data were collected from single units simultaneously, and hence f₁ and f₂ were at different distances from the BF of each of the units. One of the frequencies (f₁ or f₂) was chosen to be at or near the BF of the majority of units. The SD paradigm responses were used to compare mean PSTH segments in response to the first S token, all other S tokens, and the D token (Fig. 6C) as previously. A similar trend of decrease in CSIs in L4 over age was observed, as with N-f stimuli (CSI(F): ANOVA with Tukey post hoc analysis, F_(4,247) = 30.42, p < 0.0001; CSI: F_(4,247) = 13.83, p < 0.0001) (Fig. 6D). From CSI(F) and CSI, we found that L4 neurons are more selective (CSI(F): one-sided unpaired t test, P12-13_L4 vs P12-13_SP, t₍₅₉₎ = 1.8731, p = 0.033; P14-P15_L4 vs P14-P15_SP, t₍₁₁₂₎ = 2.866, p = 0.002; CSI: P12-13_L4 vs P12-13_SP, t₍₅₉₎ = 3.343, p < 0.001; P14-P15_L4 vs P14-P15_SP, t₍₁₁₂₎ = 2.163, p = 0.016) to oddballs than SPNs after ECO.

The observed high oddball selectivity in SPN or L4 at different developmental ages could reflect subcortical deviant selectivity (Antunes et al., 2010; Parras et al., 2017). The MGBv is the main thalamic nucleus from which A1, L4, and SPN receive inputs (Kanold and Luhmann, 2010; Holzhauer et al., 2017) (Fig. 6E). Thus, we quantified CSIs from MGBv single-unit responses to the N-f SD stimuli (Fig. 6F). In both age groups considered (P14-P15 and P24-P26), we find that CSI(F) and CSI were lower than that of L4 (CSI(F): P14-P15_L4 vs P14-P15_MGBv, t₍₁₈₀₎ = −4.373, p < 0.0001; P22-P28_L4 vs P24-P26_MGBv, t₍₁₆₃₎ = −0.031, p = 0.488; CSI: P14-P15_L4 vs P14- P15_MGBv, t₍₁₈₀₎ = −7.370, p < 0.0001; P22-P28_L4 vs P24-P26_MGBv, t₍₁₆₃₎ = −6.209, p < 0.0001) and SPN (CSI(F): P14-P15_SP vs P14-P15_MGBv: t₍₁₃₂₎ = −2.881, p = 0.002; CSI: P14-P15_SP vs P14-P15_MGBv: t₍₁₃₂₎ = −4.006, p < 0.0001) in developing mice and also lower than L4 in the adult. Thus, even at the early ages, the observed oddball selectivity is emergent in the cortex and is, expectedly, a network phenomenon (Yarden and Nelken, 2017).

Exposure in a SD acoustic environment before and after ECO shows differential developmental plasticity for the standard and the deviant frequencies

Having found higher deviant detection in ACX SP than in L4 before ECO (P7-P11) and the opposite after ECO (P12-P15) during the known critical period, we next investigated the functional implications of our observations. Previous studies (De Villers-Sidani et al., 2007; Barkat et al., 2011) have shown no change in the ACX organization in adult rodents following exposure to a continuous stream of a single frequency tone before ECO. Our results indicate that while L4 and SPNs are responsive before ECO, the above observation is expected given the strong SSA observed in SPNs before ECO (Fig. 5D), responding to the D stimulus and not the S. We hypothesize that, because of higher deviant detection by SPNs before ECO (Fig. 5D–F), and because SPNs drive L4 neurons (Holzhauer et al., 2017) at early ages, to an SD auditory environment before ECO could influence the activity-driven plasticity of the ACX.

Adult ACX frequency representation of normally reared mice (EX0, control, N = 6 animals, n = 400 units/815 tuning curves) was compared with mice exposed to an oddball sound sequence consisting of a continuous stream of sound tokens (pure tones), with each token either being a standard (S) or a deviant (D) (S occurring with probability 0.9 and D with probability 0.1) at four different developmental stages between P6 and P21 (Fig. 6A, EX1-EX4). We exposed each of the groups to an f1-f2 combination of S = 12 kHz and D = 17 kHz. The before ECO group (EX2, N = 5, n = 256/504) and the after ECO group (EX3, N = 5, n = 367/700) were the most relevant age groups for our study; hence, we performed additional experiments with another f1-f2 combination (EX2: S = 20 kHz and D = 14 kHz, N = 3, n = 197/525; EX3: S = 14 kHz and D = 10 kHz, N = 3, n = 233/553). In all the above cases, stimuli at 90 dB SPL were used to ensure sound-driven activity in ACX before ECO. Since we found that with ear closed responses in the ACX existed for as low as 60 dB SPL, for our exposure study to be functionally relevant, we did additional experiments by exposing the P6-P11 group at 70 dB SPL also (EX5, N = 5, n = 158/104). Exposure at P0-5 (EX1, N = 4, n = 183/370) and P16-P21 (EX4, N = 4, n = 130/255) was performed as age controls. Finally, as stimulus control, we performed exposures of P6-P11 (EX6, N = 3, n = 194/487) and P11-P16 (EX7, N = 3, n = 327/642) groups with only standard SS paradigm (S occurring with probability 1) in which S = 12 kHz was used (EX6 and EX7, respectively).

Single-unit recordings in L4 (depth 350-450 m) in response to pure tones (13 frequencies, 6-48 kHz, 1/eighth octaves apart, 50-90 dB SPL, 10 dB steps) from A1 and AAF were performed in the exposed (EX1-EX7) and normally reared (EX0) mice at P28-P42 (Fig. 7A). With our limited resolution BF mapping, we did not find any apparent over-representation of any particular frequency in BF maps showing broad tonotopy (A1 and AAF) on the cortical surface in any of the exposure cases (before and after ECO), unlike other studies (De Villers-Sidani et al., 2007) with the equivalent of the EX3 group (Fig. 7B). Other than the lower spatial resolution of single-unit sampling, the above may also be because of exposure parameters (sound frequency, duration, and repetition rate). We considered only mice where recordings based on BF gradients were in A1/AAF. All significant response spike rates from the tuning curves (cartoon tuning curves, Fig. 7C) for each of EX0 to EX7 were obtained. EX1, EX4, and control (EX0) had the same median spike rate (rank-sum test, 12 sp/s for every case, p_EX1 = 0.8475, p_EX4 = 0.1875). All other groups, EX2-3 and EX5-7, show increased median spike rates (rank-sum test, 16-24 sp/s, p_EX2,3,5,6,7, <10⁻⁴) compared with EX0. Thus, the overall median spike rates changed with auditory environment manipulations in the most relevant groups, P6-P11 and P11-P16 (the known critical period).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Exposure to an SD stimulus protocol before ECO induces long-term plastic changes. A, Different exposure paradigms by age and type of exposure stimulus with details (Table 4). For the SD case, green represents standard (S) and red represents deviant (D). For the SS case, blue represents standard (S). B, Bar plot represents the % change in an area of each of the preferred frequencies in A1 and AAF between control (EX0 group of mice) and the only standard P10-P15 group (EX6). C, Schematic explaining the construction of RF plots. Tuning curves (shown with different colors) with the octave difference between the BF (black arrow) and the RF (pink arrow) written on top (♦ represents rate at RF on different tuning curves, with respective color). The mean RF plot (solid black line, below) showing the mean rate at RF (represented by ♦ with respective color). D, Bar plots of all the exposure groups showing RF rates at BF normalized by the RF rates at all the other frequencies 0.5 octaves away on either side (see Materials and Methods). Striped bars in the P6-P11 group represent the SD exposure protocol at 70 dB SPL (EX5). Gray bar represents the control. *p < 0.05. *p < 0.01. Mean of the 13 control cases shown in gray bar in the center (see Materials and Methods). E, Bar plot summarizes population mean absolute response to RF for neurons with BF at RF of all 6 EX cases (EX2-7). Mean of the 13 control cases shown in gray bar in the center. Gray dashed line indicates the maximum boundary of control (see Materials and Methods). F, Bar plot showing population mean absolute response to RF for neurons with BF at RF when presented with loud sound levels (90-70 dB SPL, left) and soft sound levels (50-60 dB SPL, right).

The absence of over-representation of any particular frequency and increased spike rates in both the groups exposed during the known critical period as well as groups exposed during the period immediately preceding the critical period suggested plasticity in the strength of responses. Thus, we next investigated whether the effect of exposure was on spike rate responses to tones and whether changes were frequency-specific. From the tuning curves, the population mean response rates to an RF were plotted as a function of the difference between BF and RF in octaves (Fig. 7C; see Materials and Methods). We termed these plots as RF plots. The RF could be the frequency of S or D or that of SS. For the SD exposure protocol (EX1-5), the RF was taken as the standard frequency (S, Fig. 7D–F, green) and the deviant frequency (D, Fig. 7D–F, red); and for the only standard exposure protocol (SS), the RF was taken as the standard frequency (SS, Fig. 7D,E, blue). We obtained two such plots for SD exposures (EX1-5), with S as RF and D as RF; for SS exposures (EX6-7), S was the RF. For the control group C (gray), since there was no actual RF, each of the 13 frequencies presented was considered as RFs separately, and the RF plot was constructed (EX0, Fig. 7D–F, gray; for details, see Materials and Methods).

Although we did not observe any over-representation, we found a higher median rate for the relevant exposed groups. We investigated the effect of the exposure paradigm on the spike rates of all the neurons with respect to their BFs. We took the RF rates at the center of the RF plot (BF = RF, Fig. 7C, see Materials and Methods) and normalized them with the mean of the RF rates where the absolute difference between BF and RF was >0.5 octaves (|BF-RF|>0.5 octaves). A value >1 will signify an increase in the spike rate at the BF for the neurons with the BF as RF, whereas a value <1 will indicate an increase in spike rate at BF for neurons with BF away from the RF (Fig. 7D, normalized spike rate at BF; see Materials and Methods). We found that, in the SD exposure, before ECO, the normalized spike rate at BF for the neurons where BF and RF is the D frequency is significantly higher than the neurons with BF and RF as S (Fig. 7D, P6-P11; green bar vs red bar, one-sided unpaired t test, t₍₁₂₀₎ = 2.15, p = 0.016; p_KS-test = 0.007 at 90 dB [filled bar] and t₍₅₇₎ = 1.76, p = 0.041; p_KS-test = 0.006 at 70 dB SPL [striped bar]). The opposite is true after ECO (Fig. 7D, P11-P15; green bar vs red bar, one-sided unpaired t test, t₍₁₀₇₎ = 2.71, p = 0.004; p_KS-test < 10⁻³). The normalized spike rate of neurons with BF and RF as S before ECO was significantly less than 1 (Fig. 7D, P6-P11, green bar, one-sample t test, t₍₇₈₎ = −2.75, p = 0.0078), but does not show any effect after ECO (Fig. 7D, P11-P15, green bar, one-sample t test, t₍₇₅₎ = −0.83 p = 0.39). In the SS exposure, the normalized spike rate at BF for neurons with BF and RF as S was not significantly different from 1 both before ECO (Fig. 7D, P6-P11, blue bar, one-sample t test, t₍₆₎ = −0.37 p = 0.72) and after ECO (t₍₁₇₎ = −1.23 p = 0.23). Comparisons (significant or not) with a population of 13 control rate profiles are marked on respective bars (gray '*' or NS). We also looked into the frequency pairs (12/17 kHz and 20/14 kHz) individually and got similar results for both before ECO (Fig. 7D, P6-P11, one-sided unpaired t test; 20/14 kHz: t₍₉₀₎ = 2.16, p = 0.016 and 12/17 kHz: t₍₂₈₎ = 0.95, p = 0.17. 12/17 kHz had comparatively lesser number of tuning curves, both 90 dB SPL [filled bar] and t₍₅₇₎ = 1.71, p = 0.041 12/17, 70 dB SPL [striped bar]) and after ECO (Fig. 7D, P11-P15, 14/10 kHz: t₍₅₁₎ = 1.61 p = 0.056 and 12/17 kHz: t₍₅₄₎ = 2.19 p = 0.016). Thus, there is a differential long-term plasticity of responses to the low probability sound frequency depending on the period of development during which animals were exposed to the SD environment. It coincides with the differential oddball selectivity before and after ECO observed in the ACX L4 and SPNs involved early in development in setting up the ACX circuitry.

The population mean absolute spike rate at BF as a function of the difference between BF and RF in octaves for EX2&6, EX3&7, and EX5 (Fig. 7A) showed differential effects on response strength at BF to the S and D for SD exposure before ECO and after ECO. Exposure after ECO in the known critical period of P11-P15, the response to S frequency in neurons with BF at S is strengthened compared with the response to D frequency in neurons with BF at D (Fig. 7E, P11-P15; green bar vs red bar, one-sided unpaired t test, t₍₁₀₇₎ = 2.65, p = 0.005; p_KS-test = 0.008). However, before ECO, the opposite effect is observed (Fig. 7E, P6-P11; green bar vs red bar, one-sided unpaired t test, t₍₁₂₀₎ = 3.08, p = 0.0013; p_KS-test = 0.0013). No such changes are observed in exposure before P6 (Fig. 7E, P0-6; green bar vs red bar, one-sided unpaired t test, t₍₈₁₎ = 0.15, p = 0.55; p_KS-test = 0.81) and after P15 (Fig. 7E, P16-P21; green bar vs red bar, one-sided unpaired t test, t₍₅₈₎ = 0.88, p = 0.81; p_KS-test = 0.33), similarly at 70 dB SPL exposure (Fig. 7E, P6-P11; striped green bar vs red bar, t₍₅₇₎ = 1.507, p = 0.06; p_KS-test = 0.011). In the SS exposure in P11-P15, there is an increase in response to S frequency in neurons with S as BF beyond the boundary of control (Fig. 7E, P11-P15, blue bar and gray dashed line; for details, see Materials and Methods; one-sample t test, t₍₁₇₎ = 2.68, p = 0.015). A higher response than the boundary of control (for details, see Materials and Methods) for D frequency in neurons with BF as D was found before ECO in SD exposure at both 90 (Fig. 7E, P6-P11, red bar and gray dashed line, one-sample t test, t₍₄₂₎ = 2.80, p = 0.007) and 70 dB SPL (Fig. 7E, P6-P11, striped red bar and gray dashed line, one-sample t test, t₍₃₈₎ = 5.06, p < 10⁻⁴). As observed in normalized spike rate, when considering each set of exposure frequencies, for the merged groups of SD exposure with different sets of S and D frequencies, the response to D in neurons with BF as D was greater than that of S in neurons with BF as S before ECO (one-sided unpaired t test, 20/14 kHz: t₍₉₀₎ = 3.07, p = 0.0014 and 12/17 kHz: t₍₂₈₎ = 0.95, p = 0.17, 12/17 kHz had comparatively lesser number of curves, both 90 dB SPL and t₍₅₇₎ = 1.5, p = 0.06 12/17, 70 dB SPL). After ECO, the response for D in neurons with BF as D was found to be less than that of S in neurons with BF as S (14/10 kHz: t₍₅₁₎ = 1.53 p = 0.06 and 12/17 kHz: t₍₅₄₎ = 2.11 p = 0.019). Tuning curves collected after the presentation of pure tones at louder (70-90 dB SPL) and softer (50-60 dB SPL) sound levels were also analyzed separately to probe the effect of intensities (Fig. 7F). We found the effect to be more consistent at softer sound levels (EX2: t₍₂₀₎ = 2.03, p = 0.023, p_KS-test = 0.01; EX3: t₍₄₂₎ = 2.71, p = 0.004, p_KS-test = 0.016; EX5: t₍₅₎ = 5.47, p = 0.001, p_KS-test = 0.01) compared with the louder sound level (EX2: t₍₅₈₎ = 2.64, p = 0.005, p_KS-test = 0.008, EX3: t₍₆₃₎ = 0.95, p = 0.173, p_KS-test = 0.055, EX5: t₍₅₀₎ = 0.66, p = 0.254, p_KS-test = 0.107).

Model binary network with observed oddball selectivity of SPNs and L4 shows a stronger representation of deviant frequency before ECO and standard frequency after ECO

Having found long-term plastic changes specific to the deviant in the ACX that had exposure to an SD environment before ECO (P6-P11), we next explain our observations based on known circuitry of SPN, L4, and thalamic inputs (Kanold and Luhmann, 2010; Holzhauer et al., 2017). SPNs receive stronger thalamic inputs than L4 early on, and SPNs project to L4 neurons (Holzhauer et al., 2017) (Fig. 8A). In our simultaneous SP-L4 recordings with staggered electrodes before ECO, 42 pairs of neurons (between L4 and SP) were found to be connected (Fig. 8B,C; see Materials and Methods) of 136 simultaneously recorded L4-SP unit pairs. After ECO (P12-P15), we found 176 of 244 L4-SP unit pairs to be connected. The direction of connections was obtained based on significant cross-correlations at positive or negative time lags and confirmed via the GC test (SP->L4, magenta or L4->SP, cyan, corresponding examples in Fig. 8B and population data in Fig. 8C). Assuming columnar sizes of ∼100 µm, we grouped the pairs as columnar (≤125 µm) or not (>125 µm), based on the lateral distance between the electrodes from which connected pairs were obtained (Fig. 8C). Observed directionality of the connected pairs before ECO suggests the existence of connections from SPN->L4 (12 of 16 pairs showed connections from SPN->L4 within a spatial distance of ≤125 µm, while 13 of 26 pairs were found to be connected at a distance of >125 µm, Fig. 8C, first column, top). However, the L4->SP connection probability within ≤125 µm (4 of 16 connected pairs) was found to be significantly lower than SPN->L4 (one-tailed z test, z = −2.828, p = 0.002, Fig. 8C, first column, top). Moreover, L4->SP connections were found to be significantly higher than SP->L4 at ages after ECO (one-tailed z test, z = 4.746, p < 10⁻⁴, Fig. 8C, first column, bottom). The summary of connectivity between L4 neurons and SPNs (Fig. 8C) shows that SP->L4 connections are significantly reduced (one-tailed z test, z = −2.600, p = 0.004) from before ECO to after ECO. GC analysis (Granger, 1969; Francis et al., 2018) also revealed similar connectivity profiles between SP and L4 at ages before and after ECO, as observed using cross-correlation (Fig. 8C, second column, top and bottom). It is also evident from the spike rates of L4 neurons that the thalamic drive increases after ECO, as also observed in slice recordings (Holzhauer et al., 2017). We also found that recurrent connections, obtained with cross-correlation as well as GC, within SP (SP->SP) and within L4 (L4->L4) were differentially altered from before ECO to after ECO (Fig. 8D). Significantly more recurrent connections were present within SP than within L4 before ECO (z = 5.437, p < 10⁻⁴, Fig. 8D, top) and which switched after ECO (z = 9.655, p < 10⁻⁴, Fig. 8D, bottom). SP to SP recurrent connections decreased with ECO while those in L4 increased, very much like the observed CSI values in SPN and L4 before and after ECO, expected from the role played by recurrent connections in deviant detection (Yarden and Nelken, 2017). Our observations on deviant detection strength measured by CSIs are thus further corroborated by the lateral connection changes observed in the two stages in the two layers.

• Download figure
• Open in new tab
• Download powerpoint

Figure 8.

The binary network model of a developing ACX with SP and TC inputs. A binary model is shown for both before ECO and after ECO, with the size of the blobs representing synaptic weights, which changes across time. A, A network schematic showing standard (S, as green in SD and as blue in SS paradigm) and deviant (D, as red in SD paradigm) TC inputs to both SP and L4 and an SP input to L4 (top), SD and only standard (SS) stimulation protocol (bottom). B, Example pair of correlograms for SP to L4 (left) and L4 to SP (right) connections. Gray plot represents the bootstrapped shuffled data (control). *Significant time windows (95% CI) for both before ECO and after ECO. An example EPSP profile showing synaptic depression. C, Population bar plots showing connection probability between SP and L4 for different distances according to cross-correlation (left) and GC (right) before and after ECO. D, Connection probability for SP-SP and L4-L4 connections. E, Top, Example EPSP profile showing synaptic depression. Bottom, The asymmetric spike time-dependent plasticity learning followed by all the plastic synapses, showing a larger LTD window. The synaptic weight modifications occurred over −80 to 40 ms of post-pre spike times. F, Example PSTHs showing higher oddball selectivity in SP (magenta) than L4 (cyan) before ECO and the opposite after ECO. G, Temporal evolution of the relevant synaptic strengths (S) in the two exposure periods (before ECO, SD: EX2; SS: EX6; and after ECO, SD: EX3; SS: EX7) for simulations of exposure with SD or SS protocol. Inset, Schematic represents the final point of the simulation with the blob sizes representing relative synaptic weights. H, Plots represent the mean spike time difference (post-pre) across time. I, Mean relative change in synaptic weight (Δw/w). *p < 0.05. **p < 0.01. ***p < 0.001. Theoretical derivations explaining results are provided in Extended Data Figure 8-1.

Thus, at the earliest stages, stimuli through thalamic inputs that drive SPNs would drive the L4 neurons. Such activation of L4 neurons with coincident activity in the thalamus to L4 synapses would strengthen the synapses following Hebbian plasticity. Thus, the observed deviant detection of the SPNs responding to D and not the S would strengthen the synapses on L4 conveying the D information. On the other hand, after ECO, the thalamus to L4 synapses are already stronger than before ECO, and the L4 neuron responds to both the S and D stimulus, preferring S more. The SP to L4 synapse is weaker than before ECO and is not as strong in evoking spikes in L4 by itself. Thus, on exposure to SD after ECO, synapses conveying D information from the thalamus weaken while conveying S information strengthens.

We test the above explanation for the observed exposure-dependent developmental plasticity based on L4 and SPN deviant detection through a binary network model (Fig. 8A) of early developing TC circuitry of the ACX (Holzhauer et al., 2017) similar to that considered in the visual cortex (Kanold and Shatz, 2006). To have a minimal model that could potentially explain the results, we did not consider lateral connections, although the presence of recurrence will, in principle, only improve the results. Thalamic inputs selective for either of two inputs, S or D (green and red, Fig. 8A), are considered to make synapses on an SPN, sending collateral inputs to L4 neuron (Luhmann et al., 2018). The SP and L4 neurons are modeled as integrate-and-fire neurons with absolute and relative refraction (Kanold and Shatz, 2006). The SPN makes a synapse onto the L4 neuron, as known (Luhmann et al., 2018). All synapses were considered depressing in nature (Mill et al., 2011; Yarden and Nelken, 2017) such that L4 and SPN responses show similar adaptation observed early in development (Fig. 8E, top). Also, all synapses, except the thalamic inputs to SPN, are plastic following an asymmetric Hebbian spike time-dependent plasticity (Kanold and Shatz, 2006) rule (Fig. 8E, bottom). The abscissa in Figure 8E, below represents the difference in presynaptic and postsynaptic spike times (δτ, post – pre), with positive δτ leading to LTP and negative δτ leading to LTD. The ordinate represents the percentage relative change in synaptic weight (Δw/w) for each presynaptic and postsynaptic spike pair.

We consider the model with two different starting points: one representing the period before ECO (Fig. 8A, left) and another after ECO (Fig. 8A, right). In the earlier developmental period (before ECO), the SP CSI was 0.41, and that of L4 was −0.02. In the later developmental period, SP CSI was −0.17, and that of L4 was 0.4 (Fig. 8F), as in our observations (Fig. 6B). The differences in the initial point of the network (see Materials and Methods) in the 2 cases (Fig. 8A, left and right) are supported by other data (Holzhauer et al., 2017) as well as our observations of larger time constants later in development, increased thalamic synaptic weights on L4 because of normal development, lower SPN to L4 connectivity (Fig. 8B,C) and reduced SPN rates through higher thresholds. We use the same exposure protocols, SD and SS, for the two periods to show the nature of how the thalamic inputs to L4 develop (Fig. 8G). Before ECO with higher oddball selectivity in SP than in L4, with an SD exposure, the D thalamic input to L4 strengthens faster than the S thalamic input (Fig. 8G, before ECO).

On the other hand, after ECO, with the oddball selectivity reversed in SP and L4, D to L4 synapse weakens, and S to L4 strengthens, as L4 spikes more because of thalamic input spikes in that of S and not for that of D (Fig. 8G, after ECO). The above is evident when considering the relative presynaptic and postsynaptic spike times (δτ, post – pre) driving LTP or LTD of the S->L4 and D->L4 synapses in the 2 cases, before and after ECO (Fig. 8H; see Materials and Methods). The mean strength of weight change occurring in the synapses based on all Δw/w for each presynaptic and postsynaptic spike pair separated by δτ also shows the same effect on synaptic strength change (Fig. 8I). Mean δτ(D->L4) is positive before ECO and negative after ECO, with a significant difference. Thus, the effect of SD exposure before ECO is overall long-term potentiation of the D->L4 synapse. After ECO, the same synapse weakens as observed in the model results (Fig. 8G). The opposite is true for the S->L4 synapse. Before ECO with an SS exposure, the S thalamic input does not show strengthening at a fast rate (as D, above), leading to the S input's normal representation (Fig. 7D, bar plot P6-P11) equivalent to S in the SD case. After ECO, with the only S stimulus, the S inputs expectedly strengthen, leading to the normal response of S, as observed (Fig. 7D, bar plot).

Sparse coding and information maximization principles imply a strengthening of response to low probability stimuli

While a network model mechanistically addresses the observations, we theoretically address the observations of early exposure to a low probability stimulus leading to the strengthening of its responses and its weakening in later exposure case. Our observations and model results are contrary to the general ideas of long-term plasticity; a repeated presentation of a stimulus in the critical period leads to the strengthening of the representation of that particular stimulus in the long-term (Zhang et al., 2001; De Villers-Sidani et al., 2007; Barkat et al., 2011; Schreiner and Polley, 2014) as theoretically expected from standard Hebbian plasticity. Thus, it is important to reconcile our observations with previous observations from a theoretical perspective as well. Auditory cortical spiking activity is known to be sparse, and sparse coding principles theoretically explain several known receptive field types in the auditory pathway (Smith and Lewicki, 2006; Hromádka et al., 2008). We assumed that a minimal activity-based coding in single-neurons underlying sparse representation in populations and that such representation is one of the developing auditory system's goals while maximizing information about the stimulus in the responses. We use an objective function (Eq. 1) as the desired optimization performed by activity-driven plasticity with these principles (Extended Data Fig. 8-1). The objective function has a mutual information term between the response (R) and the stimuli token (T) that will increase as successive stimuli tokens come. A sparseness constraint term containing the mean rate limits the increase in mutual information. We call this metric as the Constrained Information (CI). This function was optimized in two separate regimens corresponding to the two separate regimens (i.e., before ECO and after ECO) as follows: Constrainedinformation(CI)=Mutualinformation(MI)−ρ(meanrate)(1)

To keep the solution mathematically tangible and straightforward to understand, we optimized a single neuron receiving tokens of stimuli that are either standard (S, occurring with a higher probability) or deviant (D, occurring with a lower probability). We also took the liberty of considering the probability distribution of the responses for standard and deviant as Gaussian with a constant variance but variable mean. A detailed mathematical optimization is provided in Extended Data Figure 8-1. However, the basis is as follows. We focus on the distribution of a single neuron's response to standard and deviant for both before ECO and after ECO regime. For the before ECO case, we have kept the mean response for deviant higher than that of standard and the opposite for the after ECO case, as we got in our experiments (Fig. 5). We see that the mutual information would increase as the separation between the mean response of the two distributions would increase (a will increase; Extended Data Fig. 8-1). However, since the sparse activity requires a minimal response by single neurons to code the two stimuli, α must be positive and small to minimize the summed activity. As such, the response to the deviant needs to be larger than that of the standard. However, the relative weightage of MI and the summed activity (controlled by ρ) can allow a to be increasing or decreasing and still cause CI to be increasing.

We perform this optimization separately for the two regimens and analytically find out the upper bound and lower bound for the Lagrange multiplier ρ and hence the relative weightage of the sparseness, which will keep increasing the mutual information with successive tokens (Extended Data Fig. 8-1). Based on these results, we show that sparseness in later stages (after ECO), when spike rates are higher, needs to be weighted more than earlier stages (before ECO) to explain the differential effect observed before and after ECO in response to the low probability stimulus.