Detection of Individual Differences Encoded in Sequential Variations of Elements in Zebra Finch Songs

Zebra finch songs consist of common elements but have individually unique element sequential arrangements

Zebra finches discriminate individually unique songs (Zann, 1996, 1993a,b; Elie and Theunissen, 2016). Both sexes show a preference for the songs of their own species over those of other species in song-selecting behaviors (Clayton and Pröve, 1989; Clayton, 1990; Braaten and Reynolds, 1999), and male zebra finches preferentially learn from tutors of the same species rather than from heterospecific tutors (Clayton, 1988). Zebra finch songs consist of repetitions of a unique motif, composed of several song elements arranged in stereotypical sequences. To examine the song diversity of zebra finch and how they manage to implement both individual uniqueness and common conspecifics features in their songs, we investigated the elements in song motifs of 192 adult male zebra finches that were hatched and raised in our colony (166 males from 76 breeding pairs, Extended Data Table 1-1) or purchased from commercial sources and raised in our colony (26 males), as well as 38 zebra finch songs from the Millbrook Library [Tchernichovski et al., 2021; in total, 3,706 song elements, 3–19 (8.06 ± 2.83, mean ± SD) elements/song]. Here, we defined the elements as the smallest unit of songs separated by silent gaps and rapid changes in vocal gestures, as previously reported (including combination elements), or as notes (Scharff and Nottebohm, 1991; Zann, 1993a,b; Sturdy et al., 1999a,b; Schneider and Woolley, 2013; Lachlan et al., 2016). We classified song elements into eight types based on a visual inspection of their sound spectrograms, as described in previous publications (Fig. 1A). Previous studies have identified five to fourteen elements by visually inspecting sound spectrographs (Zann, 1993a; Sturdy et al., 1999a,b; Lachlan et al., 2016; James and Sakata, 2017). By applying k-means clustering, an unsupervised clustering algorithm, with different cluster numbers ranging from two to 20, we identified a local peak at k = 8 within the number of element types identified in the previous studies (5–14) in the silhouette score, which is a metric of intracluster relative to intercluster similarity (Fig. 1B); we classified the song elements into eight types. We then identified most of the 3,706 elements (71.8%; 2,662/3,706) as one of the eight types (Fig. 1A,C). While we found all the eight element types in the songs from Millbrook Library, the distribution of element types between songs from our colony and those from the Millbrook Library was significantly different (p = 0.0015; Kolmogorov–Smirnov test; Extended Data Fig. 1-1). To validate our visual classification of song element types, we extracted the acoustic features of the 2,662 identified song elements, applied t-SNE dimension reduction, and visualized them in a two-dimensional space (Fig. 1D). All song elements which were classified as the same type by visual inspection, except for those classified as Type 7, were clustered with a slight overlap with the other types. We also compared the classifications based on visual inspection with those obtained by k-means clustering (k = 8). We discovered that 72% of the visually classifiable elements were grouped into the same clusters by k-means clustering. To further validate our visual classification, we compared the visual classification with the k-means classification by computing multiple clustering performance values, including the silhouette coefficient (Rousseeuw, 1987; visual, 0.119; k-means, 0.143), Davies–Bouldin score (Davies and Bouldin, 1979; visual, 2.27; k-means: 1.82), and Calinski–Harabasz score (Caliński and Harabasz, 1974; visual, 215; k-means, 286). All the values indicate that our visual classification exhibited a clustering performance comparable with that of the k-means algorithm. These results suggest that the songs of the zebra finches in our colony consisted of species-specific song elements with similar acoustic structures as previously reported (Zann, 1993a; Lachlan et al., 2010, 2016).

Next, we investigated the sequential arrangement of song elements by examining the element-to-element transitions between two consecutive song elements. The same sets of song elements can form different song motifs with distinct element sequential arrangements. We found that the percentage of transitions from a given song element type to a specific element type was not different from that of randomized element transitions made with the same elements from the 230 songs (permutation test, p > 0.05) for most element transitions (51/64; 79.7%; Fig. 1E), although there was a higher probability of repeats of the same element type (Types 2, 3, and 6). This suggests that there were no specific sequential element arrangements in the zebra finch songs of our colony, although small trends were observed. We further assessed similarities in the element sequential arrangement among all pairs of the 230 songs by comparing the sequences of the shared song element types. We subsequently found that the majority (73.1%) of the song pairs shared two or more types of song elements (Fig. 1F). In contrast, only 8.46% of those song pairs showed identical sequential arrangements of shared element types (Fig. 1G). The mean element type sequential similarity was significantly higher than the random similarity between song pairs sharing two or three element types (p = 0.003 and 0.036 for song pairs with two and three shared element types, respectively, in the permutation test), whereas between song pairs sharing four to nine element types, sequential similarity was not different from random similarity (p = 0.250, 0.214, 0.598, 0.109, 0.868, and 0.691 for song pairs with four, five, six, seven, eight, and nine shared element types, respectively, in the permutation test). We further calculated the percentage of element transitions and shared element sequential arrangements only within the birds that did not share breeding parents (76 males, one bird from each of the 76 breeding pairs in our colony), to avoid possible effects of song learning on sequence transitions. Notably, some fractions of genetically related individuals exist in natural colonies. Comparable fractions of the 1,164 elements from 76 songs were classified as one of the eight types by visual inspection (69.2%; 805/1,164; Extended Data Fig. 1-1B). We further confirmed that most of the possible song pairs (64.5%) among the 76 songs from unrelated birds shared two or more element types (Extended Data Fig. 1-1C), whereas only 8.67% of the song pairs showed identical sequential arrangements of shared element types (Extended Data Fig. 1-1D). The mean element type sequential similarity was not significantly higher than the random similarity between the song pairs that shared 2–8 (no songs shared nine element types) element types (0.063, 0.267, 0.196, 0.375, 0.999, 0.079, and 0.883 for song pairs with two, three, four, five, six, seven, and eight shared element types, respectively, in the permutation test; Extended Data Fig. 1-1E), as we found in all songs analyzed, including the siblings (Fig. 1D–G). While zebra finch juvenile males tend to learn specific element sequences when they are provided with only a few elements during song playback (James and Sakata, 2017), they cause variations in songs by combining or splitting elements and modifying the sequence of elements (Böhner, 1983), as well as incorporating new elements (Tchernichovski et al., 2021) when learning from tutors under natural conditions. Together with our observation of higher element type sequential similarity in bird pairs that shared only a small number of element types from the bird population, those suggest local sequential rules within a small number of element combinations while random syllable sequences in a song. These findings suggest that zebra finch songs share common song element types with similar acoustic features but are individually distinct in their sequential arrangements.

NCM neurons act collectively as a neuronal ensemble to detect zebra finch song libraries

Lesioning or inactivation of the NCM disrupts song discrimination abilities in adult male zebra finches (Bolhuis and Gahr, 2006; Gobes and Bolhuis, 2007; Canopoli et al., 2014) and decreases song preference in adult females (Tomaszycki and Blaine, 2014), suggesting that the NCM is necessary for individual song-discriminating behavior. Previous electrophysiological studies have shown selective auditory responses to specific songs in the NCM BS neurons (Schneider and Woolley, 2013; Bottjer et al., 2019). We performed single-unit electrophysiological recordings of the NCM in freely-moving adult male zebra finches to determine how NCM neurons respond individually to different zebra finch songs (Fig. 2A). We recorded NCM neurons and probed them with 18 zebra finch songs, termed the song stimulus series, to which the birds had not been previously exposed (see Materials and Methods). The elements of 18 songs were distributed to cover variations in the acoustic features of the song elements found in our previous analysis of 230 songs from our colony and the Millbrook Library (Extended Data Fig. 2-1). Overall, we found that 480 NCM neurons from nine birds [26–107 neurons/bird; 53.3 ± 28.6 (mean ± SD)] responded to at least 1 of the 18 zebra finch songs in the song stimulus series (68.8% of all 698 recorded neurons). Consistent with previous reports (Schneider and Woolley, 2013; Yanagihara and Yazaki-Sugiyama, 2016; Bottjer et al., 2019; Katic et al., 2022), we identified two types of neurons that were distinct in spike shapes and firing rates in the NCM, i.e., BS [n = 388 (80.8%); spike width, 0.43 ± 0.05; spontaneous firing rate, 0.94 ± 1.17 spikes/s; mean ± SD] and narrow-spiking (NS; n = 92 (19.2%); spike width, 0.22 ± 0.03; spontaneous firing rate, 8.51 ± 6.38) neurons (Fig. 2B,C). As reported in previous studies (Schneider and Woolley, 2013; Yanagihara and Yazaki-Sugiyama, 2016; Bottjer et al., 2019; Katic et al., 2022), each BS neuron responded to a small subset of songs in the song stimulus series [4.90 ± 4.15 songs (mean ± SD)], while NS neurons responded to most of the 18 songs [12.4 ± 5.67 (mean ± SD); Fig. 2D]. We found that a group of BS neurons recorded from a single bird as an ensemble (18–92 BS neurons/bird) responded to all 18 songs in the song stimulus series. In contrast, each of the 18 songs elicited responses from different combinations of the NCM BS neurons recorded from a single bird (Fig. 2E,F). This suggests that NCM neurons in birds discriminate individually different songs as an ensemble rather than representing individual songs with a small subset of neurons, allowing the NCM to discriminate an almost unlimited number of songs. When probing birds with a tutor's songs (TS), we found that the proportion of neurons selective for TS was not different from that of neurons selective for other songs (Fig. 2G), in contrast to the proportions of TS-selective neurons found in juvenile birds (Yanagihara and Yazaki-Sugiyama, 2016; Katic et al., 2022). To avoid the effects of experience, we further analyzed the responses to 18 zebra finch songs to which birds had never been exposed previously. To test whether the NCM discriminates individual songs as a neuronal ensemble, we trained the naive Bayes decoder with RS to the song stimulus series of different numbers of BS neurons (1–250) and tested whether the decoder predicted the song stimulus correctly (Robotka et al., 2023; Fig. 3A). The prediction accuracy was measured as the percentage of correct predictions in 100 trials (ensembles). We repeated the training of the decoder and predicted song stimuli using different numbers of neurons (ensemble size). We found that single neurons were insufficient for reliable stimulus prediction, and prediction accuracy improved by increasing the number of neurons, reaching almost 95% with 250 neurons (Fig. 3B,C). The prediction accuracies were better when we trained the decoder with the RS of NS neurons or combinations of NS and BS neurons (4:1, 1:1, or 1:4 ratio), probably owing to the NS neurons’ higher RS (firing rate; Fig. 3D). NS neurons are local inhibitory neurons that modulate the responses of BS neurons, which are excitatory projection neurons (Bottjer et al., 2019; Spool et al., 2021). This suggests that NCM can perform song discrimination as an ensemble of BS neurons (Fig. 3B,C) with local inhibitory inputs from NS neurons.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Individual NCM BS neurons respond to a small subset of zebra finch songs but respond to all as an ensemble. A, The diagram of sagittal section of the zebra finch brain shows the NCM (left) and a Nissl-stained image of a zebra finch brain in which NCM neurons were electrophysiologically recorded (right). The dashed square indicates the electrode track, and the arrowhead indicates the electrical lesion created after the electrophysiological recording. B, Representative sound spectrograms of a song stimulus (top) and time-aligned raster plots of the responses of NS (left) and BS (right) neurons (bottom). C, The spontaneous firing rates against the spike width of each NCM neuron; the spike widths of NS and BS neurons are distinct and separated at 0.3 ms (BS, spike width, 0.43 ± 0.05; spontaneous firing rate, 0.94 ± 1.17 spikes/s; mean ± SD; NS, spike width, 0.22 ± 0.03; spontaneous firing rate, 8.51 ± 6.38). D, The number of NS (left) and BS (right) neurons responding to each song in the song stimulus series. E, Responses of NS and BS neurons recorded in a single bird to each of the 18 songs in the song stimulus series. The filled boxes denote neurons that showed significantly positive responses. Different combinations of BS neurons respond to each song. F, The normalized percentage of BS neurons responsive to each song stimulus in each bird. A unique combination of colors and patterns represents each song stimulus. All 18 songs elicited responses from at least one neuron in all the birds. G, proportion of neurons showing selective responses to each song stimulus in the song stimulus series and each bird tutor's song. The proportion of tutor's song-selective neurons showed no significant difference between the selective and other song stimuli. See Extended Data Figure 2-1 for more details.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Song stimulus prediction from the RS of BS neurons using the Bayesian decoder. A, Process of song stimulus prediction using the Bayesian decoder. B, Prediction performance for song stimulus identity using a naive Bayesian decoder trained with neuronal responses of 1, 100, or 250 neurons. Neurons were randomly selected from the 386 song-responsive BS neurons and averaged across 100 replicates. Each block indicates the percentage of a specific song (actual song ID) within the number of songs predicted as a specific song (predicted song ID), presented as a heatmap. Correct predictions are indicated on a diagonal line in the matrix. C, Prediction accuracies of the naive Bayesian decoder as a function of the number of neurons, ranging from 1 to 250, were used to train the decoder with their responses (average over 100 trials). The dotted line corresponds to the heat map shown in B. D, Prediction accuracies of the naive Bayesian decoder as a function of the number of neurons when the decoder was trained with the response only of BS neurons and the number of BS and NS neurons was 4:1, 1:1, and 1:4 or only with NS neurons. NCM, caudal nidopallium; NS, narrower spike; BS, broader spike; SD, standard deviation.

Element type-selective BS neurons

Previous studies have reported that BS neurons in the NCM respond to the playback of single-song elements (Schneider and Woolley, 2013). We also found that BS neurons respond to specific parts of song stimuli, in contrast to NS neurons, which respond to the entire period of song stimuli (Fig. 2B), as previously reported (Schneider and Woolley, 2013; Yanagihara and Yazaki-Sugiyama, 2016; Bottjer et al., 2019), while response selectivity could differ due to recording conditions (e.g., anesthetized/awake, head restrained/free moving), as suggested in studies on cortical neurons of mammals and birds (Gaese and Ostwald, 2001; Karino et al., 2016). To investigate the correspondence of BS neuronal responses to song elements and songs, we further probed 73 of the 388 song-responsive BS neurons from nine birds with 80 single-song elements, 10 elements of each of the eight types derived from 7 to 10 zebra finch songs (Fig. 4A). We found that 61 of the 73 neurons responded to at least one song element [5.66 ± 5.21 elements (mean ± SD); Fig. 4B]. We further assessed whether these neurons showed selective responses to specific element types by calculating the d′ value between the responses to the two element types using the average RS to 10 song elements of the same type. Slightly more than half (54.1%; 33/61) of the element-responsive neurons were selective to a specific type of song element, as indicated by d′ values higher than 0.5, for a specific element type in all comparisons with other element types (corresponding to RS to a specific type of element being more than twice that of any other element type; Fig. 4C,D). We further found selective neurons for all element types except for element Types 7 and 8 (Fig. 4C). The remaining BS neurons tended to respond, but not significantly, to a smaller number of song elements that were not of the same type (half responded to fewer than four elements) than the element type-selective neurons (Fig. 4E), suggesting that they were selective for specific elements rather than element types. We found that the RS to elements of the same type varied in each element-responsive neuron, especially in nonselective neurons (Fig. 4B). As the RS was the average of 10 trials of the same stimulation, which were presented in a semirandom order, NCM BS neurons are suggested to be sensitive not only to the acoustical features we used to define element types but also to other information, such as temporal patterns or envelopes, rather than response variation occurred due to differences in the birds’ head orientation to the speaker or internal conditions. Spike shapes and spontaneous firing rates of element type-selective and nonselective neurons did not differ (Fig. 4F,G).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Element type-selective neurons. A, Sound spectrographs of a representative element of each type (top) and raster plots of a BS neuron's neuronal response (bottom) to these stimuli. B, Responses to 80 elements of each type-selective (top, n = 33, 54.1%) and nonselective neurons (bottom, n = 28, 45.9%). C, A pie chart showing the percentage of neurons selected for each element type. BS, broader spike. D, Plots of d′ values for each element-type stimulus in element-selective neurons (left) and in element-nonselective neurons (right). E, Box plots (box, 1st and 3rd quartiles; vertical bars, 5th and 95th percentiles; horizontal line, median; square, mean) of the number of song elements to which element type-selective (n = 33) and nonselective (n = 28) neurons responded. Element type-selective neurons tended to respond to a greater number of song elements than nonselective neurons, although the difference was not significant (selective, 6.70 ± 5.50; nonselective, 4.43 ± 4.65; mean ± SD; Mann–Whitney U test; p < 0.05). Each dot denotes data from an individual neuron. F, Box plots (box, 1st and 3rd quartiles; vertical bars, 5th and 95th percentiles; horizontal line, median; square, mean) of the spontaneous firing rates of element type-selective (n = 33) and nonselective (n = 28) neurons, which were not significantly different (selective, 0.81 ± 0.70 spikes/s; nonselective, 0.46 ± 0.27 spikes/s; Mann–Whitney U test; p < 0.01). G, Box plots (box, 1st and 3rd quartiles; vertical bars, 5th and 95th percentiles; horizontal line, median; square, mean) of the spike width of element type-selective (n = 33) and nonselective (n = 28) neurons, which were not significantly different (selective, 0.41 ± 0.06 ms; nonselective, 0.43 ± 0.04 ms; mean ± SD; Mann–Whitney U test; p < 0.05). SD, standard deviation.

Element sequence-sensitive neurons

As we found that BS neurons responded to specific song elements, we investigated whether specific elements caused BS neuronal responses to a song by probing another subset (n = 41) of 388 song-responsive BS neurons with a degressive song stimulus series. The degressive song series was created by deleting song elements one-by-one from the songs to which a neuron showed the strongest response in the song stimulus series and the individual elements of that song (Fig. 5A). We found that most BS neurons (39/41) responded to at least one of the individual song elements and responded most robustly to the individual elements when presented alone. The other two neurons responded only to a sequence of two elements with no gaps, suggesting they responded to a syllable (Extended Data Fig. 5-1A,B). BS neurons responded to a significantly greater number of elements when presented individually than in a song (Fig. 6A). To further examine the effect of a preceding element on neuronal responses to a specific element, we selected neurons that responded to a specific element but not to the preceding element in a song when it was presented individually. We subsequently analyzed the effect of the preceding element (n = 41). Some neurons showed increased (n = 10; Fig. 5B) or decreased (n = 5; Fig. 5C) responses to a specific element when presented without a preceding element. Interestingly, some neurons (n = 5) showed responsiveness to an element when the preceding element was deleted and decreased responses to that element by deleting another preceding element (Fig. 6D; Extended Data Fig. 5-1A). Those suggested that these neurons are sensitive to sequential elemental arrangements. The RS of the remaining 21 neurons did not change, with or without the preceding elements (Fig. 5D). Some neurons exhibited an altered response to a specific song element when the preceding element, which in these cases was a subsyllable, was deleted (Extended Data Fig. 5-1). There were no differences in spike shapes or firing rates between neurons that altered their responses to a specific song element when presented without the preceding element and those that did not (Fig. 6B,C). Next, to test whether the responses of element sequence-sensitive neurons increased the song discrimination ability of the NCM neuronal ensembles, we trained the naive Bayes decoder with the RS of sequence-sensitive or nonsensitive neuronal ensembles with different numbers of neurons ranging from 1 to 15 (100 random neuronal responses per neuron number). We found that the prediction accuracies modeled with the responses of a single neuron between sequence-sensitive and nonsensitive neurons were not significantly different (p = 0.89; Mann–Whitney U test). However, neuronal ensembles with more than two sequence-sensitive neurons showed significantly higher prediction accuracy than ensembles with the same number of nonsensitive neurons (p < 0.01, Mann–Whitney U test). The difference in prediction accuracy between the ensembles of sensitive and nonsensitive neurons gradually increased as the number of neurons in the ensembles increased (Fig. 6E,F). The prediction accuracy reached 60% with 15 sequence-sensitive neurons (Fig. 6F). In contrast, responses from 50 BS neurons (which were thought to include both sequence-selective and nonselective neurons) or >20 NS neurons were needed to train the decoder to reach the same level of accuracy (Fig. 3D), suggesting neuronal response of sequence-sensitive neurons helps to increase the accuracy of song discrimination and decreases the number of neurons in an ensemble required for discriminating songs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Subsets of BS neurons are sensitive to element arrangements. A, Schematic of the degressive song stimulus series. The degressive song stimulus series was created by individually deleting song elements from a song in the song stimulus series, to which each neuron exhibited the strongest response. The individual elements of the song are also presented. B–D, Representative neurons showing an increased (B) or decreased (C) response or no change in response (D) to an element of a song stimulus after deletion of the preceding element. Sound spectrograms (top) of the song stimuli (left), corresponding degressive song stimuli (middle), and individual elements (right) in addition to raster plots of neuronal responses (bottom) to each stimulus. See Extended Data Figure 5-1 for more details.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Properties of element sequence-sensitive neurons and song stimulus prediction with element sequence-sensitive neuronal response. A, The number of elements eliciting responses when presented individually in a song; the NCM neurons responded to a significantly smaller number of elements when they were presented as a song than when they were presented individually (song, 2.42 ± 0.89; individual, 3.42 ± 2.35; mean ± SD; Mann–Whitney U test; p < 0.01). The number of responsive elements in a song was estimated using the number of local maxima of the spike rate functions. The spike rate function was calculated by binning the spike number in a 10 ms time window and smoothing with a 50 ms Gaussian window. Subsequently, the number of local maxima on the spike rate function, which was more than two standard deviations (2σ) above the mean of the prestimulus baseline, was counted. For individual element presentation, significantly higher firing rates in the 250 ms time window from the stimulus onset than in the baseline (the same length of time window before stimulus onset) by the Wilcoxon signed-rank test (p < 0.05) was considered a positive response. B, Box plots (box, 1st and 3rd quartiles; vertical bars, 5th and 95th percentiles; horizontal line, median; square, mean) of the spike widths of neurons that did or did not change their responses to a specific element when the preceding element was deleted. The spike widths did not differ significantly between neurons with and without altered responses. C, Box plots (box, 1st and 3rd quartiles; vertical bars, 5th and 95th percentiles; horizontal line, median; square, mean) of spontaneous firing rates of neurons that did or did not change their responses to a specific element when the preceding element was deleted. NCM, caudal nidopallium; SD, standard deviation. D, The pie chart of the percentage of BS neurons that showed an increase, decrease, both an increase and decrease, or no change in response when a specific element was presented with the preceding element. E, Confusion matrices of prediction accuracy by the naive Bayesian decoder trained with the neuronal responses of one (left) or 15 (right) sequence-selective (top) or nonselective (bottom) neurons. Neurons were randomly sampled from 20 (selective) or 21 (nonselective) neurons, and the results were averaged across 100 trials. Each block indicates the percentage of a specific song (actual song ID) within the number of songs predicted by the song (predicted song ID), as shown in the heatmap (average over 100 trials). Correct predictions are indicated on a diagonal line in the matrix. F, Prediction accuracies of the naive Bayesian decoder as a function of the number of sequence-sensitive or nonsensitive neurons, ranging from 1 to 15, used to train the decoder with their responses (average over 100 trials). The dotted line corresponds to the heat map shown in E. BS, broader spike.

When zebra finch males learn to sing from tutors, they incorporate new elements (Tchernichovski et al., 2021) besides splitting or merging existing song elements and learning chunks of tutor elements (Williams and Staples, 1992; Tchernichovski and Mitra, 2002). To further investigate the effect of sensitivity to sequential element arrangements in NCM neurons, especially if the selectivity was to combinations of specific elements or specific types of elements, we probed another subset of BS neurons (n = 15) with song elements, to which these neurons exhibited the strongest response when the elements were presented in a song, combined with a preceding element (10 elements for each of the eight types) while maintaining the length of the gap between the elements (two-element stimulus series; Fig. 7A). We found that most neurons (13/15) showed increased or decreased responses when presented with at least 1 of the 80 preceding elements [1–39; 14.4 ± 11.8 elements (mean ± SD); Fig. 7B,C]. Eight neurons exhibited decreased responses to the song element when presented with the preceding element. The number of preceding elements that decreased the responses varied between neurons (1–22; 8.13 ± 7.74). The preceding elements that decreased the responses to a specific element were not the same type (Fig. 7C). The other five neurons showed increased or decreased responses to a specific element depending on the preceding element. The preceding elements that caused either an increase or decrease in the response to a specific element were not the same type. Four of these five neurons showed an increase and decrease in response to a specific element based on the preceding elements, whereas they were of the same type (Extended Data Fig. 7-1). The remaining two neurons did not change their responses to a specific element when presented with any of the 80 preceding elements. These findings suggest that responses to element sequences are not depending on the combination of element types but based on particular element combinations. There were no differences in firing rates or spike waveforms between neurons that changed their responses with the preceding elements and those that did not (Fig. 7D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

BS neurons show altered responses to a song element when presented in sequence with another element. A, Schematic diagram of two-element stimuli. B, Sound spectrographs of an element (top) and the corresponding two-element (bottom) stimulus and raster plots of the neuronal responses (bottom) that showed no change (left), an increase (middle), or a decrease (right) in their responses when the element was presented with a preceding element (two-element stimuli). C, Color matrix of changes in neuronal responses when an element is presented with a preceding element. The filled boxes denote neurons with increased (orange) or decreased (blue) responses. See Extended Data Figure 5-1 for more details. BS, broader spike. D, Box plots (box, 1st and 3rd quartiles; vertical bars, 5th and 95th percentiles; horizontal line, median; square, mean) of the spike width (left) and spontaneous firing rate (right) of neurons that did or did not change their responses to a specific element when presented with a preceding element. Each dot denotes the data from an individual neuron. There were no significant differences in either the spike width or the spontaneous firing rate between neurons that altered responses when presented with a preceding element and those that did not (spike width, change, 0.39 ± 0.07 ms; unchanged response, 0.46 ± 0.04 ms; spontaneous firing rate, change, 0.83 ± 0.60 spikes/s; no change, 0.50 ± 0.36 ms; mean ± SD; Mann–Whitney U test; p < 0.05). SD, standard deviation. See Extended Data Figure 7-1 for more details.