Vocal learning promotes patterned inhibitory connectivity

Abstract

Skill learning is instantiated by changes to functional connectivity within premotor circuits, but whether the specificity of learning depends on structured changes to inhibitory circuitry remains unclear. We used slice electrophysiology to measure connectivity changes associated with song learning in the avian analog of primary motor cortex (robust nucleus of the arcopallium, RA) in Bengalese Finches. Before song learning, fast-spiking interneurons (FSIs) densely innervated glutamatergic projection neurons (PNs) with apparently random connectivity. After learning, there was a profound reduction in the overall strength and number of inhibitory connections, but this was accompanied by a more than two-fold enrichment in reciprocal FSI-PN connections. Moreover, in singing birds, we found that pharmacological manipulations of RA's inhibitory circuitry drove large shifts in learned vocal features, such as pitch and amplitude, without grossly disrupting the song. Our results indicate that skill learning establishes nonrandom inhibitory connectivity, and implicates this patterning in encoding specific features of learned movements.

Fig. 1

Inhibition is a major component of RA circuitry. a Micrograph of acute RA slice preparation at three different magnifications showing RA with four patch electrodes during whole-cell recording (left), recorded and filled neurons visualized in the same slice (middle), and a group of two larger projection neurons (PNs) and two smaller inhibitory fast-spiking interneurons (FSIs, Int) from the middle panel’s dotted region (right). PNs and FSIs are morphologically distinct. b PNs and FSIs express distinct firing patterns. PNs generate spontaneous tonic pacemaking activity at rest (left). After hyperpolarizing them with DC current, PNs produced adapting trains of high-amplitude (>30 mV) action potentials in response to depolarizing current steps, and they expressed voltage sag in response to further hyperpolarization (middle). In contrast, FSIs do not have spontaneous pacemaking activity, and they produce very high-frequency (>150 Hz) non-adapting trains of <30 mV action potentials in response to current injection (right). c Average action potential (AP) waveforms from an FSI (left) and a PN (right). d AHP amplitude and AP width across our sample of FSIs and PNs differentiated these cell types. e Inhibition dominates spontaneous synaptic activity in RA. Example spontaneous inhibitory postsynaptic currents (sIPSCs, top) and excitatory postsynaptic currents (sEPSCs, bottom) recorded from the same PN at equal distance from either E _Cl or E _AMPAR. Both the frequency and amplitude of sIPSCs are much larger than those of sEPSCs despite the pacemaking activity of other PNs in the circuit. f Average excitatory vs inhibitory charge recorded from 46 PNs demonstrates that inhibition outweighs excitation in RA PNs during spontaneous activity

Fig. 2

Learning to copy a tutor song is associated with dramatic changes to RA inhibitory strength and connectivity. a Tutoring Bengalese Finches with an automated tutoring paradigm drives song learning in 10–20 days. Spectrograms from age-matched untutored birds show that they retain unstructured subsong-like vocalizations over this period (left), whereas spectrograms from tutored birds (right) rapidly converge on a copy of the tutor stimulus (bottom panel). The same stimulus was used to tutor all birds so that they ultimately produced similar vocal output at the time of slice recording. b Examples of evoked-IPSCs from paired recordings of PNs and FSIs from untutored (left) and tutored (right) birds. Driving single spikes in FSIs (bottom traces) produced unitary IPSCs in synaptically connected PNs. Gray traces are individual current sweeps and black traces are averages of 15–30 sweeps. IPSCs were large, reliable, and depressing (0.63 ± 0.08 at 50 Hz) in connected FSI→PN pairs. These features are consistent with the high level of spontaneous inhibition that we observed in sIPSC recordings. c After song learning, FSI→PN connections were significantly weaker than that in untutored birds. Blue trace is mean IPSC ± SEM from tutored birds (n = 24 pairs from 11 birds) and red trace is mean IPSC ± S.E.M. from untutored birds (n = 47 pairs from nine birds). Asterisks indicate significance of p < 0.01 determined by t-test. d IPSC amplitude was significantly smaller in tutored birds than in untutored birds (left), and the FSI→PN connection probability was reduced from 0.76 to 0.38 after tutoring. Error bars on the right plot indicate 95% confidence intervals of connection probabilities from a binomial distribution. Asterisks indicate significance of p < 0.01 determined by binomial test

Fig. 3

Enrichment of reciprocal FSI ↔ PN connections during song learning. a Proportion of all tested pairs before (red) and after (blue) tutoring that were unidirectional FSI→PN connections (left) or reciprocal FSI ↔ PN connections (right). Song learning was associated with a 69% reduction in unidirectional FSI→PN connections (binomial test p < 0.001), but did not alter the proportion of reciprocal FSI ↔ PN connections. Error bars indicate 95% confidence intervals. b Frequency of first-order connection patterns that we detected in untutored (left, red) and tutored (right, blue) birds relative to each connection pattern’s expected frequency given unidirectional FSI→PN and PN→FSI connection probabilities in untutored and tutored birds. Each connection pattern was present at chance levels before tutoring, whereas reciprocal FSI ↔ PN connections were 2.39 times more frequent than predicted by random connectivity after song learning (multinomial test p < 0.01, 95% confidence interval = 1.68–3.11-fold, see Methods). Correspondingly, one-way PN→FSI connections were 0.41-fold less frequent than predicted after song learning (multinomial test p < 0.05). Relative to untutored birds, tutoring significantly increased the prevalence of FSI ↔ PN and PN→FSI connections (multinomial test p < 0.005) and reduced the prevalence of FSI→PN connections (multinomial test p < 0.00001). Error bars indicate 95% confidence intervals. Connection pattern sample sizes are parenthetical

Fig. 4

Excitatory PN synapses were not altered during song learning. a Example PN→FSI connection. Evoked PN spikes (bottom) drove monosynaptic EPSCs in a postsynaptic FSI (middle). The black trace is the average of 30 individual sweeps (overlaid gray traces). PN→FSI connections exhibited modest short-term depression (0.63 ± 0.06 at 50 Hz), as observed in other systems. The black trace above the sweeps is the spike-triggered average (STA) of 375 events to enhance sensitivity to small EPSCs. The arrow indicates the presynaptic spike time. PN→FSI strength did not change during song learning (right panel, n = 10 pairs from 3 untutored birds, n = 17 pairs from five tutored birds). b Tutoring did not alter the probability of PN→FSI connections (0.19 untutored, 0.17 tutored). Error bars indicate 95% confidence intervals from a binomial distribution. c Example of quadruple recording from four PNs. Brief current pulses evoked spikes in one PN (bottom), while the other PNs were voltage-clamped at E _Cl to detect evoked EPSCs. Gray current traces are 75 overlaid single sweeps and the black trace is the average. We were unable to detect any PN–PN connections in 262 attempts. d Summary of attempts to record PN–PN connections in untutored (red) and tutored (blue) birds. The upper 95% confidence interval is 0.026 in untutored birds and 0.014 in tutored birds

Fig. 5

Manipulating RA inhibition in singing birds bidirectionally shifts learned song features, while preserving overall song structure. a NASPM reduces spontaneous inhibition in RA slices. Example sIPSCs recorded from a PN in ACSF (gray trace) and after 0.1 mM bath application of NASPM (red trace). b NASPM reduced spontaneous inhibition in all PNs recorded. On average, NASPM reduced inhibitory charge by 42%. c Schematic of in vivo reverse-microdialysis experiment and example spectrograms of songs recorded during reverse-microdialysis of PBS followed by either NASPM (left), which reduces RA inhibition, or midazolam (right), which enhances RA inhibition. Overall song structure was not altered by either manipulation. The gray arrow indicates the syllable that we analyzed for fundamental frequency (FF) and amplitude in d–g. d Fundamental frequency contours of the syllable highlighted by the gray arrow in c during PBS dialysis (gray) followed by either NASPM (red) or midazolam (blue) dialysis. Envelopes indicate SEM. Reducing RA inhibition with NASPM increased the FF without altering syllable structure (left), whereas enhancing RA inhibition with midazolam reduced the FF of the syllable (right). e Changes in syllable FF induced by either NASPM (red) or midazolam (blue) across all experiments. Light circles are fold-changes in FF from PBS in individual experiments and darker circles indicate each condition’s mean. Error bars indicate SEM. f Rectified amplitude waveforms of the syllable marked by gray arrows in c during PBS (gray) followed by either NASPM (red) or midazolam (blue) dialysis. Waveforms are normalized to the mean of syllables produced during PBS dialysis. Envelopes indicate SEM. NASPM significantly increased syllable amplitude, whereas midazolam decreased it. g Change in syllable amplitude relative to PBS induced by either NASPM (red) or midazolam (blue) across all experiments. Light circles indicate individual experiments and darker circles indicate condition means. Error bars indicate SEM