Adolescent mice exhibit lower performance during self-initiated auditory learning in the ‘Educage’.

A. Schematic model of the ‘Educage’ (left), the trial structure and trial types (FA, false alarm; CR, correct reject). Created with BioRender. B. Experimental timeline. Total training time was 21 days (±2). C. The sounds used for training. Light blue-easy task; dark blue-hard task; Grey-catch-trials. D. Learning curve examples. Adolescent mouse (left, P20-to-P37); adult mouse (right, P60-to-P77). Vertical dashed lines indicate the easy-hard transition. Horizontal line is d’=1. E. Number of trials to reach threshold (d’>=1; adolescents, n =15; adults, n=15; z=-0.1659; p=0.8682, two-sample Wilcoxon rank sum test). F. Discriminability (d’) of the easy task in adolescent mice at P30 (grey; n=15, d’=2.0001±0.1791; n trials=8317±712) and adult mice at P70 (black; n=15, d’=2.1986 ± 0.2441; n trials=13229 ± 514, top: marked by the arrowhead). Dashed lines: mean trials per group (t-stat=-5.6314, df=28, p=4.9566e-06, two-sample independent t-test, vertical line), and mean d’ per age group (z=-0.2074; p=0.8357, two-sample Wilcoxon rank sum test, horizontal line) G. Change in discriminability (Δd’) of the easy task before and after the introduction of the hard task (Top: arrowheads; left: adolescent, signed rank=14, p=0.0067; right: adult, signed rank=14, p=0.1514, one-sample Wilcoxon signed rank test; Δd’ between adult and adolescent mice: z=-2.0739; p=0.0381, two-sample Wilcoxon rank sum test) H. Same as ‘G’ for the first 100 and last 100 trials of the experiment in the easy task (adolescent signed rank=120, p=6.1035e-05; adult signed rank=120, p=6.1035e-05, one-sample Wilcoxon signed rank test; Δd’ between adult and adolescent mice: z=-1.0370; p=0.2998, two-sample Wilcoxon rank sum test). I. Same as ‘F’ for the hard task (adolescent-grey; n=15, d’=1.1895±0.1783; n trials=4163 ± 297; adult-black; n=15, d’=1.8342 ± 0.1743; n trials=4102±475) (mean trials per group: t-stat=0.1306, df=28, p=0.8970, two-sample independent t-test, vertical line; mean d’ per group: z=-2.2398; p=0.0251, two-sample Wilcoxon rank sum test, horizontal line). J. Same as ‘G’ for the hard task. (adolescent signed rank=73, p=0.4887; adult signed rank=114, p=8.5449e-04, one-sample Wilcoxon signed rank test; Δd’ between adult and adolescent mice: z=-1.9495; p=0.0512, two-sample Wilcoxon rank sum test).

Behavioral differences between adolescent and adult mice are age-, but not sex-related.

Fixed effects of age and sex, and the random effects of co-housing in the ‘Educage’ on the discriminability (mean d’ of the last 100 trials of the easy and hard task to avoid pseudo replication) of all mice (Number of observations = 30, Fixed effects coefficients = 3, Random effects coefficients = 7, Covariance parameters = 2). Coefficient estimates, STE, T-statistic, degrees of freedom, p-values (adjusted for multiple comparisons with the Bonferroni method) and the lower and upper Confidence Interval (95%). The model includes random effects coefficients of the Cage ID in each group of co-housed mice (7 cages in total; see methods, equation 7).

Adolescent mice exhibit lower response inhibition and higher variability in performance over the course of learning.

A. Schematic timeline of tone discrimination). B. Average psychometric curve of the lick rate across all stimuli (catch and learned; No-Go stimuli: p = 0. 0085, Go stimuli: p = 0. 0963, Wilcoxon rank sum test, after Bonferroni correction). C. Same as ‘B’, but normalized and fitted to a sigmoid function0(lick rate at 10kHz, p = 0.1163; frequency at 50% lick rate, p = 0.5896, Wilcoxon rank sum-test, after Bonferroni correction). D. Maximal lick bias (i.e., criterion bias) per mouse during the last 2000 trials (c-bias: p = 0.0136, Wilcoxon rank sum test). E. Average psychometric curve, fitted to a sigmoid function across all stimuli (catch and learned) at different periods of the experiment — Left: first tertile-easy task (per mouse; adolescent n = 15, adult n = 15); Middle-left: last tertile – easy task; Middle-right: first tertile-hard task; Right: last tertile-hard task. (Dashed vertical =category boundary). F. Same as E. as the maximal C-bias. From left to right:(z = -0.5104, p = 0.6098, two-sample Wilcoxon rank sum test), (z = -2.6546, p = 0.0079, two-sample Wilcoxon rank sum test), (z = -4.0739, p = 4.6228e-05, two-sample Wilcoxon rank sum test), (z = -3.2768, p = 0.0011, two-sample Wilcoxon rank sum test). G. Change in lick bias (Δ c-bias) between the last 100 trials before and after the introduction of the hard task. (left, adolescent signed rank = 17, p = 0.0125; right, adult signed rank = 83, p = 0.2078, one-sample Wilcoxon signed rank test; Δ c-bias between adolescents and adults, z = -2.9035; p = 0.0037, two-sample Wilcoxon rank sum test). H. Coefficient of variation (CV) of the d’ of the easy task before the introduction of the hard task. (z = -0.1659, p = 0.8682, two-sample Wilcoxon rank sum test) I. Same as ‘H’ after the introduction of the hard task (z = 2.1569, p = 0.0310, two-sample Wilcoxon rank sum test). J. Same as ‘H’ but for d’ of the hard task (z = 2.5302, p = 0.0114, two-sample Wilcoxon rank sum test).

Adolescent mice exhibit lower performance in the head-fixed discrimination task.

A. Experimental timeline of training followed by recordings. Created with BioRender. B. Trial structure during the recording. Solid lines indicate the tone period. Dashed lines show the reward or punishment delay (0.6 sec), and the response window (2 sec). C. Example session. Licks (grey ticks) and trial outcomes (hit = green, false alarm = yellow, miss = red and correct reject = blue) across all trials in one recording session. D. Discriminability during training sessions for the easy task (light blue) and hard task (dark blue). E. Change in d’ after the introduction of the hard task (last 100 trials of the last session of the easy task compared to last 100 trials of the first hard session; rank-sum = 14; p = 0.0381, two-sample Wilcoxon rank sum test). F. Expert d’ of the last 100 trials during the last training session of the easy task (rank-sum = 21; p = 0.1255, two-sample Wilcoxon rank sum test). G. Same as ‘F’, but for the hard task (rank-sum = 17; p = 0.0173, two-sample Wilcoxon rank sum test). H. Behavioral performance (average d’ of the easy and the hard task) per mouse during recording sessions for adolescents (n =13, left) and adults (n= 14, right) (trials per recording: adolescent: 340.5385 ± 45.0650; adult: 431.1429 ± 30.3367; independent t-test, t-statistic = -203.7581, p = 0.1116). I. Same as ‘H’ but only for the first 148 trials. The color bar shows the p-values between the groups. J. Average cumulative licks per trial in adolescents (dashed-line) and adults (solid-line) from - 200ms before tone -onset until the reward or punishment delay, 500ms after tone-offset. K. Lick latency per trial for adolescent (left) and adult (right) groups during electrophysiological recordings (LME statistics are shown in supplemental Table 1). J. Same as ‘K’ for the Lick count.

ACx neurons in adolescents exhibit lower discriminability in stimulus- and choice-related activity

A. Recordings in ACx when the mouse is engaged in the task, using Neuropixels-1 probes. Left: Recordings were performed in AUDd, AUDp, AUDv, and TEa. Right: Fluorescent micrograph of a coronal brain slice showing the probe tracks of three recordings (red = DiI, yellow = DiO). Created with BioRender. B. Top: 3D-Reconstruction of recording sites in adolescents (n = 13; light green) and adults (n = 14; dark green). Bottom: distribution of the spike-depth of all excitatory tone-responsive L5/6 neurons in adolescents (n = 455; light green) and adults (n = 607; dark green). C. Normalized PSTH and lick-rate (LR) from -200ms to +600ms after tone-onset in adolescents (light green) and adults (dark green). D. Spiking activity from one example neuron sorted by trial outcome (hit, miss, false alarm, correct reject). Top: PSTH per trial outcome. Bottom: Heat map of the FR sorted pertrial outcome. E. Discriminability values (AUC) over time (from -200ms to 600ms after tone onset) for one example neuron (same neuron as in ‘D’). AUC values are shown for stimulus related activity (left: easy task, middle: hard task) and choice-related activity (right). Shuffled distribution in all curves is shown in grey. F. Same as ‘E’ for all neurons. The curves are average (+-SEM) neuronal discriminability of adult neurons (solid line) and adolescent neurons (dashed line), for easy (adolescent neurons = 190, mice = 4, recordings = 7; adult n = 358, mice = 4, recordings = 8; left) and hard stimulus-related activity (adolescent n = 429; adult n = 562, mice = 5, recordings = 9; middle), and choice-related activity (adolescent n = 429; adult n = 562, mice = 5, recordings = 9; right). G. 3D plots of the onset-latency of discriminability (ms), duration of discriminability (ms), and maximal discriminability (AUC) of all neurons that showed significant discriminability. Left: easy task (adolescent neurons = 178, mice = 4, recordings = 6; adult n = 346, mice = 4, recordings = 8; left); Center: hard task (adolescent neurons = 368, mice = 5, recordings = 10; adult n = 544, mice = 6, recordings = 12; middle); Right: choice-related activity (adolescent neurons = 368, mice = 5, recordings = 10; adult n = 544, mice = 6, recordings = 12; right).

Neuronal discrimination is later, shorter, and less precise in adolescent neurons.

Linear mixed effect models of the neuronal discriminability in adolescence and adulthood per stimulus-related activity in the easy task (Number of observations = 524, Fixed effects coefficients = 2, Random effects coefficients = 10, Covariance parameters = 3), stimulus related activity in the hard task (Number of observations = 943, Fixed effects coefficients = 2, Random effects coefficients = 14), and choice-related activity (Number of observations = 520, Fixed effects coefficients = 2, Random effects coefficients = 10, Covariance parameters = 3). The table shows the fixed effects of the coefficient estimates, STE, T-statistic, degrees of freedom, p-values (corrected for multiple comparisons with Bonferroni-correction) and the upper and lower CI of the effect of age on the onset latency of discrimination, duration of discrimination and maximal neuronal discrimination (AUC). Each model also included random effect coefficients of each mouse, and recording per mouse. P-values for were adjusted with post-hoc tests using Bonferroni-correction (see methods, equation 9).

Decoding in adult neuronal populations outperforms decoding in adolescents.

A. Decoding accuracy for the first 200ms across all recordings in both adults (dark green) and adolescents (light green) for the easy task (adolescents compared to adults, p=0.5000, Student’s t-test) and the hard task (adolescents compared to adults, p=0.0300, Student’s t-test). Decoding is better in the easy task for both age groups (adults: p=0.0030; adolescents: p=0.01, paired t-test). B. Decoding latency for all recordings in the easy task (p=0.0200, Student’s t-test) and the hard task (p=0.0030, Student’s t-test), as well as compared between age groups (easy task, p=0.05400, pared t-test; hard task: p=0.0100). C. Decoding accuracy over a time window from -0.5s to 10s. D. LDA separation for easy and hard tasks. Lines represent robust linear regression fits without intercept (Huber loss; robust linear regression, p=0.0001) E. Single trial variance for easy and hard tasks in adolescent and adult recordings (adults: p=0.0040; adolescents: p=0.0300, paired t-test; easy task: p=0.4500; hard task: p=0.4100, Student’s t-test). F. Visualization of population representations for the stimuli in easy and hard tasks. Dotted lines indicate decoding dimensions, and ellipses represent the covariance of the representations.

Cortical activity during behavior reflects both age- and learning-induced effects.

A. Training and recording schedule for novice mice, compared to expert mice. Created with BioRender. B. 3D-Reconstruction of recording sites in novice adolescent (n = 6; light green) and novice adult (n = 6; dark green) mice. Bottom: spike-depth of excitatory tone-responsive L5/6 adolescent (n = 107; light green) and adult (n = 177; dark green) neurons. C. Normalized FR and lick-rate (LR) PSTH from -200ms to 600ms after tone-onset in adolescents (light green) and adults (dark green). Average +-sem. D. Single neuron data from novice adolescent mice. Left: Heat map of the FR per trial from one example neuron sorted by trial outcomes. Center: the AUC of the neuron from the left for the easy and hard stimulus pairs (light and dark blue, respectively). Right: Average (+-SEM) AUC of all neurons in the novice group (n = 140 neurons). E-G. Same as ‘D’ for novice adult (n = 186 neurons), expert adolescents (n = 455 neurons; Easy vs hard), and expert adults (n = 604 neurons; Easy vs hard.). H. Linear regression analysis between the average AUC per recording and the behavioral d’ during the recording (the correlation and p values are indicated for each plot). I. Same as ‘I’ for adult mice.

The effect of age, learning and task difficulty on the latency, duration, and ability to discriminate tones in ACx neurons.

Linear mixed effect models of the effect of age, learning and task difficulty on onset-latency of discrimination, duration of discrimination and maximal discriminability (Number of observations = 2590, Fixed effects coefficients = 8, Random effects coefficients = 20, Covariance parameters = 3). The table shows the fixed effects of the coefficient estimates, STE, T-statistic, degrees of freedom, p-values (corrected for multiple comparisons with Bonferroni-correction) and the upper and lower CI. The model also includes random effects coefficients of each mouse (adolescent novice = 3, adult novice = 3, adolescent expert = 5, adult expert = 6) and recording per mouse (n = 3). P-values for were adjusted with post-hoc tests using Bonferroni-correction (see methods, equation 10).

Unlike adults, neuronal tuning properties of adolescents do not change after learning

A. Schematic showing that for the passive listening protocol, we continued our recording following the session of the engaged task (i.e. in satiated mice) by removing the waterspout. Created with BioRender. B. Example raster plot of a neuron from an adolescent mouse (top) and an adult mouse (bottom). C. FRA’s of the neurons shown in ‘B’. D. Distribution of best frequencies in our dataset. Values are normalized firing rates calculated at 62 dB SPL. Matrices are sorted by BF for clarity. Dotted line marks the decision boundary. E. Tuning bandwidth at 62 dB SPL of neurons in adolescents and adults. Side by side comparisons of novice versus experts. (adolescents p = 0.0882, adults p = 0.0001, Kruskal Willis Test after Tukey-Kramer correction for multiple comparisons). F. Same as E. for the Population sparseness (adolescents p = 0.9549, adults p = 0.0013, Kruskal Willis Test after Tukey-Kramer correction for multiple comparisons). G Same as E. for the distance (in octaves) between the best-frequency of each neuron to the easy Go-stimulus (adolescents p = 0.0816, adults p = 0.6391, Kruskal Willis Test after Tukey-Kramer correction for multiple comparisons). H Same as E. for the average neuronal d’ of frequencies in the learned frequency spectrum (adolescents p = 0.1627, adults p = 0.0026, Kruskal Willis Test after Tukey-Kramer correction for multiple comparisons).