LHb receives inputs from VP and AIV and projects to VTA.

A: Injection of viral tracer scAAV9-CBh-GFP into VP, and retrograde tracer cholera toxin subunit B (CTB) 555 into VTA resulted in co-localization of VP axons (green) and VTA projecting neurons (yellow) in LHb. Scale bar: 50 um. B: LHb- and VTA-projecting neurons (green and magenta, respectively) were identified in both VP and AIV following injection of retrograde tracer Dextran 488 into LHb and CTB 647 into VTA. Note LHB- and VTA-projecting neurons spatially intermingled in both VP and AIV but rarely co-labeled. Scale bar: 50 um (center), 75 um (right). C: LHb axons (yellow) innervated the area of Area X projecting VTA neurons (magenta) following injection of CTB 647 into Area X and viral tracer HSV-mCherry into LHb. Scale bar: 75 um. All histological images are sagittal slices with dorsal upwards and anterior to the right. D: A summary schematic of input pathways to songbird and mammalian VTA. The excitatory (+) or inhibitory (-) nature of the functional connections were established in mammals and, more recently, in songbirds (see also Fig. 2 of this paper).

VP activates LHb, which suppresses Area X-projecting VTA DA neurons.

A: Brain schematic of the experimental approach to test how VTA-projecting LHb neurons (LHbVTA) respond to VP stimulation. B: Summary of LHbVTA (left) and LHbother (right) responses. C: antidromic identification of an LHbVTA neuron. Collisions (red) and antidromic (black) spikes are indicated by arrows, and the stimulation artifact by the black triangle. D: example LHbVTA neuron (same as in C). Raster plot (top) and rate histogram (bottom) aligned to a single VP stimulation pulse. Horizontal bars show a significant response (p < 0.05, Z test; Methods). E: data plotted as in D for a second LHbVTA neuron. F and G: data plotted as in D for two LHbother neurons. H: brain schematic of the experimental procedure to record the responses of Area X projecting VTA neurons (VTAX) to LHb stimulation. I: summary of VTAX (left) and VTAother (right) responses to burst LHb stimulation. J: scatterplot showing the Z-scored firing rate change within a 100 ms window from burst LHb stimulation against the average spontaneous spike width duration (VTAX neurons: red triangles; VTAother neurons: black circles). K: data plotted as in C for an antidromically identified VTAX neuron. L: example VTAX neuron (same as in K). Raster plot (top) and rate histogram (bottom) aligned to a triple LHb stimulation (interpulse interval, 3 ms). Horizontal bars show a significant response (p < 0.05, Z test). M and N: data plotted as in L for two additional VTAX neurons. O-Q: data plotted as in L for three VTAother neurons.

LHb lesions in juvenile birds cause atypical vocalizations in adults.

A: experimental timeline. B: schematic showing the location of habenula in the diencephalon. C: lesions were confirmed using dark field (DF, left column), anti-NeuN (middle column), and CTB647 to label VTA-projecting LHb neurons (right column), LHb indicated by white circle. Top row: example of a completely lesioned hemisphere as evident by extensive tissue damage (small, bright dots in all three images), no neurons visible in anti-NeuN, and absence of LHbVTA neurons. Bottom row: example of a sham bird with intact LHb at roughly the same mediolateral position as hemisphere in the top row. Note the cluster of LHbVTA (magenta) absent in the fully lesioned condition. Scale bars: 75 um. D: spectrograms showing tutor song (top row), and adult undirected song of an LHb lesioned juvenile bird (bottom three rows). Spectrograms in D-G have frequency limits of 500-8500 Hz and are 4 seconds in length. Scale bar: 500 ms. E-F: data plotted as in D for two additional LHb lesioned juveniles. Red underlines indicate abnormal vocalizations G: data plotted as in D-F except for a sham bird. H: reconstruction error distributions for all datasets from one of the six replicates of a trained VAE. The red vertical line corresponds to the 99.9th percentile of sham bird (test set) vocalizations, which was used to define a boundary between normal and abnormal vocalizations (Methods). Sham birds (training set: black outline; test set: gray), LHb lesion birds (average: red; individual bird shown in D: pink), and tutor bird (yellow) vocalizations. I: example original and reconstructed spectrograms from the VAE of LHb lesioned birds with reconstruction error values less than the threshold (normal) and greater than the threshold (abnormal). J: average anomaly scores for each individual bird across 6 replicate VAEs (n = 9 sham birds; n = 12 tutor birds; n = 17 lesion birds; mean +/- STD shown in dot/vertical line). Lesion group is significantly different from the other two groups (sham and tutor; Kruskal-Wallis test with Bonferroni method multiple comparison: α = 0.05). K: two-dimensional projection of the PC space for the first three components. Top row corresponds to tutor vocalizations, middle row corresponds to all sham bird vocalizations (training and test set), and bottom row corresponds to LHb lesioned birds vocalizations (blue: normal syllables; red: abnormal syllables as determined by the threshold in H).

LHb-lesioned birds produce abnormal vocalizations during courtship song.

A: spectrograms showing undirected song for a juvenile LHb lesion bird (top) and female-directed song for the same bird (bottom; bird not shown as an example in Fig 3). Scale bar: 500 ms. Red underlines indicate abnormal vocalizations. B-C: same as A for two additional LHb lesion birds, not shown as examples in Fig. 3. D: same as A-C except for a sham bird, not shown as an example in Fig. 3. E: reconstruction error distribution for sham bird (training set: black; test set, gray), and the LHb lesioned bird’s female-directed vocalizations (average: red; individual bird shown in A: pink). The red vertical line corresponds to the threshold set at 99.9th percentile of sham lesioned controls (test set) vocalizations. F: example original and reconstructed spectrograms from the VAE of LHb lesion birds with reconstruction error values less than the threshold (normal) and greater than the threshold (abnormal). G: two-dimensional projection of the PC space for the first three components. Top row corresponds to sham bird female-directed vocalizations (training and test sets), and bottom row corresponds to LHb lesion bird female-directed vocalizations (normal syllables: blue; abnormal syllables: red, as determined by the threshold in E).

LHb lesions in adulthood do not impair vocalizations or the ability to produce female-directed song.

A: spectrograms showing pre-lesion lone song for an adult LHb lesion bird (top), 60 days post-lesion lone song (middle), and post-lesion female-directed song for the same bird (bottom). Scale bar: 500 ms. B-C: same as A except for two additional example adult lesion birds. D: the self-imitation score was high for each bird, and there was no significant difference between any of the three groups (n = 3 birds control; n = 5 birds pre- vs. pre-lesion and pre- vs. post-lesion; p = 0.0625, two-sided, paired Wilcoxon sign rank test comparing pre- vs pre-lesion and pre- vs. post-lesion; p = 0.571, two-sided, unpaired Wilcoxon rank sum test comparing pre- vs. post-shamand pre- vs. pre-lesion; p = 0.0714, two-sided, unpaired Wilcoxon rank sum test comparing pre- vs. post-shamand pre- vs post-lesion). E: reconstruction error distribution for all vocalizations. The red vertical line corresponds to the threshold set at 99.9th percentile of pre-lesioned undirected song (test set), which is used to divide the LHb post-lesion undirected data into normal and abnormal vocalizations. LHb pre-lesion undirected (training set: black outline; test set: gray), LHb post-lesion undirected (average: red; individual bird shown in A: pink), and LHb post-lesion female-directed (green) vocalizations. F: average anomaly score for each individual bird across 6 replicate VAEs (n = 3 control; n = 5 lesion undirected; n = 5 lesion female-directed; mean +/- STD shown in dot/vertical line). Groups are not significantly different from each other (Kruskal-Wallis test, p = 0.9139). The outlier observed in the lesion directed group is due to determining the 99.9th percentile with a small sample size. The outlier (green) can be observed in E and Sup Fig 5_1A and C. G: two-dimensional projection of the PC space for the first three components. Top row corresponds to pre- lesion undirected vocalizations (test and training sets), middle row corresponds to post-lesion undirected vocalizations, and bottom corresponds to LHb post-lesion female-directed vocalizations.

(with Figure 3). Abnormal vocalizations can exhibit long and variable duations, evident as long tails in the syllable duration distribution.

A: example undirected song spectrograms from adult birds that experienced an LHb lesion as a juvenile. Vocalizations were recorded at 92 dph and illustrate the distribution of durations for this single syllable type (bird also shown in Fig. 3D). All spectrograms are 1 second in duration; red dashed line denotes abnormal syllable onset. B: same as in A, except for the LHb lesioned bird shown in Fig. 4C. C: same as A for a bird not previously shown. D: same as A, except each column depicts a different vocalization from bird shown in Fig. 3E. E: same as D, except bird not previously shown. F: the distribution of sham, tutor, and lesion bird syllable durations during adult undirected song (black: sham; yellow: tutor; lesion: red). Pairwise Kolmogorov-Smirnov tests with Bonferroni correction were performed for statistical differences between the distributions (sham - tutor p = 3.48e-18; sham-lesion p = 5.35e-197 and tutor-lesion p = 5.81e-7).

(with Figure 3). Uniform Manifold Approximation and Projection (UMAP) for example LHb lesion and sham birds.

A: example UMAP projection for the example LHb lesion bird shown in Fig 3D. Each cluster is differently colored, and corresponds to a distinct syllable. Example spectrograms from each cluster are populated around the UMAP projection. B: data same as A but for example sham bird seen in Fig 3G.

(with Figure 3). Identification of abnormal syllables in LHb lesioned birds is robust across independent replicate VAEs trained with different subsets of sham birds.

A: reconstruction error distribution for all datasets. The red vertical line corresponds to the threshold set at the 99.9th percentile of sham bird (test set) vocalizations, which is used to divide the lesion dataset into normal and abnormal vocalizations. Sham birds (training set: black outline; test set: gray), LHb lesion birds (average: red; individual bird shown in 3D: pink), and tutor bird (yellow) vocalizations. B: two-dimensional projection of the PC space for the first three components. Top row corresponds to tutor vocalizations, middle row corresponds to all sham bird vocalizations (training and test set), and bottom row corresponds to LHb lesioned birds vocalizations (blue: normal syllables; red: abnormal syllables as determined by the threshold in A). C-D: same as A-B with an additional cross validation example from a VAE independently trained on a distinct subset of sham birds.

(with Figure 4). Abnormal vocalizations are produced in both undirected and female-directed courtship songs.

A: example song and calls from a single bird (left: undirected song; right: female-directed song; same bird as Fig. 3F). Scale bar is 500 ms. Red underlines indicate abnormal vocalizations. B: same as in A, except the bird shown in Fig. 4A. C: same as in A, except the bird was not previously shown. D: same as in A, except the bird previously shown in Fig. 4B.

(with Figure 5). Results from figure 5 are robust across replicate VAEs trained with different subsets of sham birds.

A: reconstruction error distribution for all datasets. The red vertical line corresponds to the threshold set at the 99.9th percentile of the test set (pre-lesion control) vocalizations, which is used to divide the post-lesion dataset into normal and abnormal vocalizations. Pre-lesion data (training set: black outline; test set: gray), post LHb lesion undirected data (average: red; individual bird shown in Fig. 5A: pink), and post LHb lesion female-directed data (green) vocalizations. B: two-dimensional projection of the PC space for the first three components. Top row corresponds to pre-lesion undirected vocalizations, middle row corresponds to post-lesion undirected (blue: normal syllables; red: abnormal syllables as determined by the threshold in A), and bottom row corresponds to post-lesion female directed vocalizations. C-D: same as A-B except another cross validation analysis example.

(with Figs 3 and 4).

Variational Autoencoder (VAE) results for six different runs where the sham juvenile birds were permuted between training and test sets.

(with Fig. 5).

Variational Autoencoder (VAE) results for six different runs where the adult birds were permuted between training and test sets.