Central contributions to acoustic variation in birdsong

Abstract

Birdsong is a learned behavior remarkable for its high degree of stereotypy. Nevertheless, adult birds display substantial rendition-by-rendition variation in the structure of individual song elements or "syllables." Previous work suggests that some of this variation is actively generated by the avian basal ganglia circuitry for purposes of motor exploration. However, it is unknown whether and how natural variations in premotor activity drive variations in syllable structure. Here, we recorded from the premotor nucleus robust nucleus of the arcopallium (RA) in Bengalese finches and measured whether neural activity covaried with syllable structure across multiple renditions of individual syllables. We found that variations in premotor activity were significantly correlated with variations in the acoustic features (pitch, amplitude, and spectral entropy) of syllables in approximately a quarter of all cases. In these cases, individual neural recordings predicted 8.5 +/- 0.3% (mean +/- SE) of the behavioral variation, and in some cases accounted for 25% or more of trial-by-trial variations in acoustic output. The prevalence and strength of neuron-behavior correlations indicate that each acoustic feature is controlled by a large ensemble of neurons that vary their activity in a coordinated manner. Additionally, we found that correlations with pitch (but not other features) were predominantly positive in sign, supporting a model of pitch production based on the anatomy and physiology of the vocal motor apparatus. Collectively, our results indicate that trial-by-trial variations in spectral structure are indeed under central neural control at the level of RA, consistent with the idea that such variation reflects motor exploration.

Figure 1.

Acoustic variation in Bengalese finch song. a, Top, Raw sound amplitude waveform of five syllables (labeled A–E) from the song of an adult Bengalese finch. Middle, Smoothed rectified sound amplitude waveform. Bottom, Spectrograms of three different iterations of this motif. Spectrograms show the power at each frequency (color scale) as a function of time. The topmost spectrogram corresponds to the example sound waveform. Acoustic features such as pitch and amplitude were measured at a fixed time (red dashed line for syllable B) relative to syllable onset (green dashed line). b, Distributions of the pitch (top) and amplitude (bottom) of syllable B across 1919 renditions of the syllable. c, The song system is composed of a direct motor pathway consisting of nuclei HVC (used as a proper name) and RA and an AFP containing area X, the medial portion of the dorsolateral thalamus (DLM) and lMAN. RA sends projections to motor neurons in the tracheosyringeal portion of the 12th motor nucleus (nXIIts), which innervates the muscles of the syrinx, and to motor nuclei retroambigualis (RAm) and paraambigualis (PAm), which innervate the respiratory musculature (Vicario and Nottebohm, 1988; Wild, 1993; Reinke and Wild, 1998).

Figure 2.

Three models of how RA might drive trial-by-trial variation in song. Circles represent populations of RA projection neurons, and wavy arrows represent their contribution to variations in syllable structure (pitch in this case). a, A small number of independently varying RA neurons each drive a substantial amount of pitch variation (heavy arrows). In this case, a small proportion of RA neurons would exhibit correlations with pitch, and these correlations would be quite strong (indicated by filled black circles), reflecting the substantial influence of each neuron. b, Large numbers of independently varying RA neurons each make small contributions to pitch variation (thin arrows), with pitch variations resulting from the sum of independent modulations in firing. In this case, the activity of many neurons would be correlated with pitch, but these correlations would be weak (indicated by the lightly shaded circles), because each neuron is responsible for only a small fraction of behavioral variation. c, Large numbers of RA neurons generate correlated activity (curved arrows), driving pitch modulations with coordinated changes in activity. In this case, many neurons might exhibit correlations with each acoustic feature, and the strength of these correlations could be quite high (black circles), because variations in the firing of any one cell are correlated with variations in the entire population. In Discussion, we consider the implications of our results in discriminating between these models.

Figure 3.

Neural activity in Bengalese finch RA. a, Extracellular recording of an RA single unit in a nonsinging, awake bird. b, Firing of the same unit during singing. Neural recordings are shown for five repetitions of the song motif “ABCDE,” shown in the spectrogram at top. The five neural traces are aligned to the onset of syllable C (white dots). Onsets of the other syllables are shown as well (red dots). Data in a and b are plotted with the same time scale. c, Left, Instantaneous firing rates (1/ISI) during song for all single units. A 50 Hz rate threshold (dashed red line) was used to segregate bursting from nonbursting epochs to compute the distribution of burst durations during singing (right).

Figure 4.

a, Population activity and unit isolation estimates of 25 units recorded from bird 1. Each tick mark represents one spike; each row represents the activity during one iteration of the song motif “ABCDE,” and 20 iterations from each unit are shown. Colors differentiate recordings from the 25 units. Mean syllable durations are shown as gray boxes. The unit shown in Figure 3, a and b, is unit 2 (asterisk at left), and the unit shown in Figure 5a is unit 8 (arrow at left). The plot at the far right shows the estimated isolation error (see Materials and Methods). Units with isolation errors of <0.01 (red line) were classified as single units. b, Fraction of units active (mean rate, ≥25 Hz) as a function of time.

Figure 5.

Correlation between neural and behavioral (pitch) variation. a, Spectrogram of a song motif from bird 1. Below the spectrogram are five neural traces showing the activity of unit 8 from this bird (Fig. 4a, arrow). Premotor neural activity for this unit was quantified in a 40 ms window (red box) preceding the time of pitch measurement (dashed red line). The white arrow indicates the measured pitch of syllable E. Neural traces are aligned on the onset of syllable E. b, Distribution of pitches measured from syllable E while recording from the unit shown in a. c, Distribution of the number of spikes within the premotor window preceding syllable E. d, Plot showing the relationship between pitch and the number of spikes in the premotor window for each rendition of syllable E (n = 371), along with the correlation coefficient (r) and p value of the correlation.

Figure 6.

Significant correlations of premotor neural activity with acoustic structure across a population of RA units. a, Correlations with pitch for all units recorded in bird 1. Each row represents data from one single-unit or multiunit recording (see Fig. 4), and each column represents firing in the premotor window before one of the six syllables produced by the bird (as indicated at the bottom: A–F). A dot indicates that the unit in question was active (mean rate ≥25 Hz in the premotor window preceding a syllable). Dot color indicates whether neural activity was positively correlated (green), negatively correlated (red), or uncorrelated (white) with pitch. Black, gray, and white arrows indicate units 8, 9, and 19, respectively, as referred to in text. b, c, Same conventions as a, but dot color signifies correlations between premotor neural activity and amplitude (b) or spectral entropy (c).

Figure 7.

Prevalence and explanatory power of neuron–behavior correlations. a, Bar plot of the proportion of cases in which neural activity was significantly (p < 0.05) correlated with each of the three acoustic parameters measured. Note that one case corresponds to a unit's being active in the premotor window before a single syllable, and that one unit can therefore contribute several cases to each acoustic feature. Black lines and asterisk above bars indicate that the proportions for pitch and amplitude were both significantly greater than the proportion for entropy (p < 0.05, Z test for proportions). The dashed line shows the significance threshold for the proportion of cases correlated exceeding chance at p < 0.05, as determined by a permutation test (supplemental text, available at www.jneurosci.org as supplemental material). All three acoustic parameters were correlated with premotor activity significantly more often than chance. b, c, Probability density functions (b) and cumulative distributions (c) of r² values for the three acoustic parameters, corresponding to the fraction of behavioral variability in each case that can be accounted for by neural activity. Color conventions are the same as in a.

Figure 8.

Correlation signs. Bars show the fractions of significant correlations with positive (green) and negative (red) slopes for single-unit (filled) and multiunit (empty) recordings. Asterisks indicate a ratio of positive to negative correlations significantly different from equality (p < 0.05, binomial test). a, Results from the primary analysis. b, Results from the partial correlation analysis (see Materials and Methods).

Figure 9.

Acoustic control in the descending motor pathway. In our schematic, neurons are represented as white circles, synaptic connections are represented as lines connecting brain regions and muscles, and causal influences on acoustic structure are represented as arrows. RA projection neurons excite brainstem motor neurons in the tracheosyringeal portion of the 12th motor nucleus (nXIIts), nucleus retroambigualis (RAm), and nucleus paraambigualis (PAm), which in turn activate the muscles controlling the syrinx and respiratory apparatus (Sturdy et al., 2003). Recordings from syringeal muscles reveal strong positive correlations between muscle activity and pitch, and one muscle in particular, the musculus syringealis ventralis (vS), has been shown in the brown thrasher to have an exceptionally strong association with pitch production (Goller and Suthers, 1996). Models of syringeal function suggest that increased muscular tension raises pitch by putting more tension on the vibrating structures of the syrinx, thereby increasing their vibrational frequency when air is blown through the syrinx (Gardner et al., 2001; Laje et al., 2002; Suthers and Zollinger, 2004). This string of excitatory relationships, between RA and nXIIts, nXIIts and muscle contraction in the syrinx, and between muscle contraction and pitch, likely results in a net excitatory relationship between RA activity and the pitch of song. The excess of positive correlations with pitch in our data might therefore be attributable to a subpopulation of RA projection neurons (far left, dashed box) that activate the motor neurons innervating the vS muscle of the syrinx (middle, dashed box) or other muscles for which activation drives increases in pitch. The approximately equal mixture of positive and negative correlations with other acoustic features might be attributable to the mix of positive (+) and negative (−) influences of other syringeal and respiratory muscles. Relative neuron numbers in the four nuclei are not shown to scale.