The Forebrain Song System Mediates Predictive Call Timing in Female and Male Zebra Finches

Abstract

The dichotomy between vocal learners and non-learners is a fundamental distinction in the study of animal communication. Male zebra finches (Taeniopygia guttata) are vocal learners that acquire a song resembling their tutors', whereas females can only produce innate calls. The acoustic structure of short calls, produced by both males and females, is not learned. However, these calls can be precisely coordinated across individuals. To examine how birds learn to synchronize their calls, we developed a vocal robot that exchanges calls with a partner bird. Because birds answer the robot with stereotyped latencies, we could program it to disrupt each bird's responses by producing calls that are likely to coincide with the bird's. Within minutes, the birds learned to avoid this disruptive masking (jamming) by adjusting the timing of their responses. Notably, females exhibited greater adaptive timing plasticity than males. Further, when challenged with complex rhythms containing jamming elements, birds dynamically adjusted the timing of their calls in anticipation of jamming. Blocking the song system cortical output dramatically reduced the precision of birds' response timing and abolished their ability to avoid jamming. Surprisingly, we observed this effect in both males and females, indicating that the female song system is functional rather than vestigial. We suggest that descending forebrain projections, including the song-production pathway, function as a general-purpose sensorimotor communication system. In the case of calls, it enables plasticity in vocal timing to facilitate social interactions, whereas in the case of songs, plasticity extends to developmental changes in vocal structure.

Figure 1. Call exchanges in a live pair and responses to the vocal robot

(A) Calls exchanged between a male (blue) and a female zebra finch (red) over the first 4 days housed together. Dots represent the pitch vs. amplitude of short calls. Left: Green lines connect female calls to male answers (when answered within 500ms). Middle: Green lines connect male calls to female answers. Inset: Histograms of response latencies to partner calls within 1000ms. Proportion of partner calls answered. Right: representative sonograms by day. Intervals between calls and answers are shaded in green. (B) Left: Schematic of the vocal robot system. Middle: a male answering robot’s 1Hz isochronous calls (ICs). Right: a female answering robot’s ICs (C) The distribution of a bird’s call responses to the robot’s ICs during a 10-minute session are used to compute an optimal jamming window. The robot then produces jamming calls during this window in the following session. (See also Figure S1).

Figure 2. Jamming avoidance

(A) A male’s call responses (blue) to robot ICs (gray). Top: Responses are aligned by IC cycle onset and presented in sequential rows over a 10-minute session. Middle: The male’s responses (red) to robot call cycles (grey) in which the robot produces jamming calls (yellow). Bottom: Distribution of the bird’s responses during the IC session (blue) and during the jamming call session (red). (B) As in (A) for a female. (C) Examples of jamming avoidance strategies in three birds. Birds’ call timing is plotted relative to a normalized jamming window (yellow). (D) Top: as in (C) pooled across 12 birds. (E) As in (D) comparing call responses across 12 birds for ICs (blue) and catch trials during the jamming session (green). (F) Birds predictively reduce the proportion of calling within the jamming window during catch trials compared to ICs (n=12, paired t-test, ***P<0.001; % decrease in 6 males (red) vs. 6 female (blue), t-test, *P<0.05). (G) Across 12 birds, the average proportion of jamming was lower than expected by chance over the course of a session. (H) A female gradually changes response latencies over the course of a session (First third: bright red. Last third: Dark red) to predictively call between the robot’s jamming call pairs (grey and yellow).

Figure 3. Rhythm adaptation

(A) A rhythm pattern of rapidly alternating single calls and jamming call pairs (jamming calls in yellow), produced by a vocal robot (1s cycles, 200ms jamming latency; see also Figure S2). (B) Top: A male’s call responses to the rhythm pattern shown in (A); Green: Responses to the single robot calls. Red: Responses to the jamming call pairs. Middle: Distribution of bird’s responses. Bottom: Cumulative responses within the 200ms following robot call onsets for single calls (green) and the first calls in jamming pairs (red), showing shorter answer latencies prior to jamming. (C) A robot rhythm pattern as in (A) but with slower 2s cycle and 250ms jamming latency. (D) Cumulative responses to the slowly alternating rhythm in 6 birds (green, single calls; red, jamming pairs). (E) Median response latencies for each bird following ICs (blue), single calls in a rhythm (green), and the calls that precede jamming calls in a rhythm (red; Paired t-tests, n=6, *p <0.05).

Figure 4. Effects of RA lesions on precision and jamming avoidance

(A) Left: Control lesion in a male (purple). Control male’s responses to robot’s ICs (grey) prior to (blue) and after bilateral control lesions (purple). Bottom: Overlay of response distributions. Right: RA lesion in a male (orange). Experimental male responses to vocal robot ICs (grey) prior to (blue) and after bilateral RA lesions (orange) Bottom: Overlay of response distributions. (B) As in (A) for a control female and an experimental female. (C) Precision and skewness of responses to ICs in 2 males and 2 females before and after control lesions (n=4, paired t-tests, NS, P>0.65); in 2 males and 3 females after bilateral RA lesions (n=5, paired t-tests, **P<0.01). (D-E) Response distributions for ICs and catch trials pooled over 5 birds (as in Fig. 2E) before (D) and after (E) RA lesions. (F) Percentage of calling within the jamming window for ICs vs. catch trials before (blue, n=5, *P<0.01) and after RA lesions (orange, NS). (G) RA lesions abolish jamming avoidance, measured as the difference in percent of calling within jamming window for ICs and catch trails (n=5, means, s.e.m, **P<0.01). (See also Figures S3 and S4).

Figure 5. HVC activity associated with call production and the effects of HVC_RA projecting axon transections

(A) Partial transections of HVCRA fiber tracts (medial HVCRA tracts intact) allowed retrograde tracer (injected in RA) to reach RA-projecting cells HVC and LMAN. Responses to ICs before (blue) and after (purple) partial transections (n=3) (B) Complete HVCRA transection, verified by absence of tracer within HVC and presence of tracer in LMAN. Responses to ICs before (blue) and after complete bilateral transections (orange, n=3). (C) Complete transections results in decreased precision and skewness of responses to ICs (n=3 birds). Precision is unaffected in transected birds with intact medial projections (n=3). (D) Response distributions for ICs and catch trials (as in Fig. 4F) pooled over three birds before and after (E) complete transections. (F-G) Jamming avoidance is reduced by complete transection of HVC_RA projections.