Mechanisms and time course of vocal learning and consolidation in the adult songbird

Abstract

In songbirds, the basal ganglia outflow nucleus LMAN is a cortical analog that is required for several forms of song plasticity and learning. Moreover, in adults, inactivating LMAN can reverse the initial expression of learning driven via aversive reinforcement. In the present study, we investigated how LMAN contributes to both reinforcement-driven learning and a self-driven recovery process in adult Bengalese finches. We first drove changes in the fundamental frequency of targeted song syllables and compared the effects of inactivating LMAN with the effects of interfering with N-methyl-d-aspartate (NMDA) receptor-dependent transmission from LMAN to one of its principal targets, the song premotor nucleus RA. Inactivating LMAN and blocking NMDA receptors in RA caused indistinguishable reversions in the expression of learning, indicating that LMAN contributes to learning through NMDA receptor-mediated glutamatergic transmission to RA. We next assessed how LMAN's role evolves over time by maintaining learned changes to song while periodically inactivating LMAN. The expression of learning consolidated to become LMAN independent over multiple days, indicating that this form of consolidation is not completed over one night, as previously suggested, and instead may occur gradually during singing. Subsequent cessation of reinforcement was followed by a gradual self-driven recovery of original song structure, indicating that consolidation does not correspond with the lasting retention of changes to song. Finally, for self-driven recovery, as for reinforcement-driven learning, LMAN was required for the expression of initial, but not later, changes to song. Our results indicate that NMDA receptor-dependent transmission from LMAN to RA plays an essential role in the initial expression of two distinct forms of vocal learning and that this role gradually wanes over a multiday process of consolidation. The results support an emerging view that cortical-basal ganglia circuits can direct the initial expression of learning via top-down influences on primary motor circuitry.

Fig. 1.

Neural circuitry contributing to song production and learning. Neural pathways that control song production and learning include a song motor pathway (black nuclei and arrows) that controls much of the moment-by-moment structure of song, as well as a cortical-basal ganglia circuit, the anterior forebrain pathway (AFP; red nuclei and arrows), that plays a crucial role in juvenile song learning and adult vocal plasticity. The song motor pathway includes forebrain nuclei HVC and RA, analogous to vocal motor cortex. RA projects to brain stem structures that control the vocal musculature; changes in RA activity are likely to underlie learned changes to syllable structure (Leonardo and Fee 2005; Sober et al. 2008; Vu et al. 1994; Yu and Margoliash 1996). The lateral magnocellular nucleus of the anterior nidopallium (LMAN), a cortical analog that is part of the AFP, plays a crucial role in juvenile and adult song plasticity. LMAN activity could influence RA during learning via a number of direct and indirect pathways. Neurons in LMAN core make direct glutamatergic projections onto RA neurons and also are thought to release neurotrophins onto RA neurons. Neurons in LMAN core also project to the basal ganglia homolog Area X, which is well positioned to influence activity in neuromodulatory nuclei such as ventral pallidum (VP), a cholinergic nucleus projecting to RA and HVC, as well as the ventral tegmental area (VTA), a dopaminergic nucleus that also projects to RA and HVC (Appeltants et al. 2000, 2002; Gale and Perkel 2010; Gale et al. 2008; Li and Sakaguchi 1997). LMAN activity could also influence RA activity via indirect projections from LMAN shell to the motor pathway. The LMAN shell, part of a distinct motor circuit that plays a functional role in song learning, projects to both the Ad and the medial striatum (MSt), surrounding Area X (Bottjer and Altenau 2010; Iyengar et al. 1999). DLM, medial dorsolateral nucleus of thalamus; DTZ, dorsal thalamic zone; mMAN, medial magnocellular nucleus of anterior nidopallium. [Adapted from Bottjer and Altenau (2010) and Gale et al. (2008).]

Fig. 2.

Example trajectory of changes to syllable structure driven via a reinforcement learning paradigm. A: spectrogram of the syllable targeted for reinforcement learning. An automated system (Tumer and Brainard 2007) reliably detected a specific time point in the target syllable (inverted black triangle). B: on renditions of the target syllable with a fundamental frequency (FF) below a set threshold, no reinforcement was delivered (escape); on renditions of the target syllable with a FF above the threshold, an aversive reinforcement signal, a 60-ms white noise stimulus (WN), was played over the target syllable (hit). C: example trajectory of learning. During the baseline period (days −2 to −1), no reinforcement was delivered. The mean FF at baseline was 2,400 Hz (open black circles and vertical lines indicate daily mean FF ±1SD). During an initial learning period (initial shift, days 1–3), WN was delivered to syllable renditions with FF below a set threshold (dashed green line). In response to this reinforcement, the values of FF of the target syllable (gray data points) gradually increased. After 2 days of upward shift in FF, the threshold was fixed at 2,520 Hz. In response to this stable reinforcement contingency, the FF of the target syllable stabilized at ∼2,560 Hz, a fixed offset of 160 Hz from baseline. This learned shift in FF was maintained for 5 consecutive days (maintained shift, days 4–8). In this and subsequent figures, a random sample (10–15%) of all songs are displayed and were used to measure syllable FF (see materials and methods).

Fig. 3.

Expression of learned changes to syllable structure initially relies on LMAN activity and gradually consolidates to become LMAN independent over multiple days. A: experimental design. Dialysis probes (1 hemisphere shown, sagittal section) were bilaterally implanted into LMAN; retrodialysis solution was switched from a control artificial cerebrospinal fluid (ACSF) solution to muscimol, a GABA_A agonist, or lidocaine, a Na⁺ channel blocker. B: spectrograms illustrate preservation of the overall spectrotemporal structure of song following a switch from dialysis of ACSF to dialysis of muscimol. Also shown is the syllable targeted for learning in this experiment (target). C: effects of muscimol retrodialysis during a trajectory of vocal learning. As described in Fig. 2, WN was used to drive a shift in FF of the target syllable during a period of initial learning (initial shift, days 1–3), and then FF was maintained at a fixed offset from baseline (maintained shift, days 4–8 of reinforcement, corresponding with days 1–5 of the maintained shift). Top: mean values of FF during periods of ACSF dialysis (black circles and vertical lines indicate daily mean FF ±1SD) as well as mean values of FF during periods of several hours of dialysis with 200 μM muscimol (red circles and vertical lines). Bottom: expanded time axis with values of FF for individual syllable renditions on 3 specific days: baseline day −1, reinforcement day 3, and reinforcement day 8 (day 5 of maintained shift). Horizontal bars in bottom panel indicate time periods of drug infusions (ACSF, black; muscimol, red); solid bars indicate the period included for analysis, and open bars indicate periods excluded from analysis due to delayed onset of drug effects. Muscimol infusion on day 3 of learning caused a rapid downward reversion of FF toward the original baseline (downward red arrow, reversion). Subsequent muscimol infusions on days 5, 7, and 8 of learning, while the FF was maintained at a learned offset of ∼180 Hz from baseline, caused gradually decreasing reversions. D: summary across all learning trajectories of the effects of inactivating LMAN during the initial shift period. All data are normalized so that positive values indicate shifts of FF in the direction of learning. At baseline, LMAN inactivation caused no significant change in mean FF compared with ACSF pre- and postinactivation periods (baseline; n = 16 experiments in 7 birds; red bars indicate means ± SE for muscimol/lidocaine experiments; gray bars show means ± SE for pre- and postinactivation periods). In contrast, during the initial shift, inactivation of LMAN caused a significant reversion of mean FF toward baseline (mean reversion 46.9%, P < 0.01, 1-tailed t-test). Inactivations of LMAN reduced rendition-to-rendition variability in FF during both the baseline and initial shift periods (bottom, purple bars). For comparison, in this and subsequent figures, the dashed line shows the reduction in coefficient of variability (CV) of FF (34%) reported following lesions of LMAN in adult Bengalese finches from a previous study (Hampton et al. 2009). E: summary effects of disrupting LMAN activity across 6 learning trajectories (n = 4 birds) in which learning was maintained at a stable offset from baseline for at least 5 days. Over this period, the LMAN-dependent component of learning gradually decreased from 45% to 15% while the LMAN-independent component of learning gradually increased. The reduction in variability caused by LMAN inactivations remained stable across these time periods (bottom, purple bars).

Fig. 4.

Initial expression of learned changes to syllable structure relies on N-methyl-d-aspartate (NMDA) receptor activation in RA. A: experimental design. Dialysis probes were implanted bilaterally into RA (1 hemisphere shown, sagittal section). dl-AP5, an NMDA-receptor antagonist, was dialyzed across the probes. B: example of drug spread (coronal section). Dark biotin stain shows spread of biotinylated muscimol used to infer drug spread, which encompasses the entirety of RA but not Ad. C: effects of dialysis of AP5 at baseline and during learning. Top: daily mean values of FF during periods of ACSF infusion and interleaved periods of AP5 infusion. Bottom: raw values of syllable FF on baseline day −3 and reinforcement day 4. At baseline, retrodialysis of AP5 caused little change to mean FF. In contrast, retrodialysis of AP5 on day 4 of reinforcement caused a rapid and large (119 Hz) reversion of learned changes to FF. D: summary effects of AP5 infusion during the initial shift period (n = 5 experiments in 5 birds). AP5 dialysis at baseline caused no significant change in FF relative to pre- and post-AP5 ACSF periods (P = 0.8, paired t-test). In contrast, during the initial shift period, AP5 dialysis caused a significant reversion of FF toward the original baseline (P < 0.05, 1-tailed paired t-test). During both periods, AP5 significantly reduced the rendition-to-rendition variability in FF (P < 0.05, 1-tailed paired t-test). Conventions are as described in Fig. 3 legend.

Fig. 5.

Syllable structure recovers to the original baseline following cessation of reinforcement even after learning has consolidated. Trajectory of syllable FF for 5 birds is shown over the last day (recovery day −1) of a maintained shift in FF driven via WN and over the following 3 days in which no WN was played (WN off; recovery days 1–3). All data are normalized relative to the magnitude of the shift in FF on the last day of the maintained shift (duration of maintained shifts ranged from 4 to 7 days). Two birds were equipped with retrodialysis probes, allowing confirmation that consolidation had occurred by the last day of the maintained shift, so that the expression of learning no longer depended on LMAN (red triangle indicates muscimol infusion in LMAN; red square indicates AP5 infusion in RA; error bars indicate ±1SD). In all 5 birds, following the termination of reinforcement, syllable FF recovered back toward the original baseline over 3 days (recovery days 1–3; gray triangles and squares show recovery trajectories for birds in which consolidation was confirmed with muscimol and AP5, respectively). By day 3, birds had recovered on average 81% of the difference from the original baseline (green triangle, mean recovery).

Fig. 6.

Contributions of LMAN to recovery of syllable structure. A: example effects of AP5 dialysis into RA during self-driven recovery toward baseline. Top: values of FF for individual syllable renditions. Bottom: FF on an expanded time scale for baseline (day −1), initial shift (day 3 of reinforcement), and initial recovery (day 2 following termination of WN). During the initial shift, infusion of AP5 caused a reversion of FF toward baseline, as previously observed. During initial recovery, infusion of AP5 caused a reversion of FF toward the previously maintained level of learning, away from baseline. B: summary effects of blocking LMAN input to RA, via infusion of muscimol in LMAN (n = 6 experiments) or infusion of AP5 in RA (n = 1 experiment), at 5 stages of learning and recovery. Effects are plotted as % change in FF relative to the magnitude of the maintained shift in FF. Interfering with LMAN's input to RA caused a significant reversion toward baseline during the initial shift period (downward red arrow) and a significant reversion away from baseline during the initial recovery period (upward red arrow), both for experiments in which recovery was self-driven (n = 2 experiments; P < 0.01, permutation test with baseline effects) and those in which recovery was WN driven (n = 5 experiments; P < 0.001, permutation test). Values are means ± SE.