Developmentally Regulated Pathways for Motor Skill Learning in Songbirds

Abstract

Vocal learning in songbirds is mediated by cortico-basal ganglia circuits that govern diverse functions during different stages of development. We investigated developmental changes in axonal projections to and from motor cortical regions that underlie learned vocal behavior in juvenile zebra finches (Taeniopygia guttata). Neurons in LMAN-core project to RA, a motor cortical region that drives vocal output; these RA-projecting neurons send a transient collateral projection to AId, a region adjacent to RA, during early vocal development. Both RA and AId project to a region of dorsal thalamus (DLM), which forms a feedback pathway to cortico-basal ganglia circuitry. These projections provide pathways conveying efference copy and a means by which information about vocal motor output could be reintegrated into cortico-basal ganglia circuitry, potentially aiding in the refinement of juvenile vocalizations during learning. We used tract-tracing techniques to label the projections of LMAN-core to AId and of RA to DLM in juvenile songbirds. The volume and density of terminal label in the LMAN-core→AId projection declined substantially during early stages of sensorimotor learning. In contrast, the RA→DLM projection showed no developmental change. The retraction of LMAN-core→AId axon collaterals indicates a loss of efference copy to AId and suggests that projections that are present only during early stages of sensorimotor learning mediate unique, temporally restricted processes of goal-directed learning. Conversely, the persistence of the RA→DLM projection may serve to convey motor information forward to the thalamus to facilitate song production during both learning and maintenance of vocalizations.

Graphical Abstract

We investigated developmental changes in axonal projections to and from motor cortical regions, RA and AId, that underlie learned vocal behavior in juvenile zebra finches (Taeniopygia guttata. We used tract-tracing techniques to label the projections of LMAN-core to AId and of RA to DLM in juvenile songbirds. The volume and density of terminal label in the LMAN-core→AId projection declined substantially during early stages of sensorimotor learning. In contrast, the RA→DLM projection showed no developmental change. The retraction of LMAN-core→AId axon collaterals indicate a loss of efference copy to AId and suggests that projections that are present only during early stages of sensorimotor learning mediate unique, temporally restricted processes of goal-directed learning.

During motor skill learning, cortico-basal ganglia circuits mediate comparisons between self-generated motor output and a goal behavior that the organism is learning to emulate. During early phases of sensorimotor integration animals produce highly variable attempts to mimic the goal behavior, and feedback of their behavioral output is compared to the goal behavior by specific neural circuits. Iterative comparisons between self-generated actions and an internal representation of the goal behavior guide refinement of the organism’s variable behavioral attempts towards an accurate match to the goal behavior. In addition to feedback, efference copy of motor commands contains information about specific motor commands as well as the possibility of predicted sensory feedback; information stemming from efference copy can be relayed back into cortico-basal ganglia loops and integrated with other motor and sensory information.

Vocal learning in zebra finches serves as a powerful model for investigating mechanisms of motor skill learning because, like humans, juvenile songbirds learn a complex motor skill from an adult tutor (normally their father) during a restricted period of development. Juvenile zebra finches memorize the stereotyped song pattern of an adult tutor between 20 and 35 days post-hatch (dph) and the memory of the song provides the goal vocalizations that juveniles learn to imitate (Böhner, 1990; Roper & Zann, 2006). Zebra finches start producing highly variable vocal babbling around 35–37 dph, marking the onset of the sensorimotor phase of learning (Eales, 1985; Immelmann, 1969). Immature vocalizations of juveniles are gradually refined to achieve an imitation of the tutor goal behavior, and ultimately around 100 dph birds produce a highly stereotyped song that is produced throughout adulthood (Immelmann, 1969; Tchernichovski et al., 2000).

Acquisition of learned vocal behavior in juvenile songbirds is mediated by parallel recurrent circuits that traverse core and shell subregions of the cortical nucleus lateral magnocellular nucleus of the anterior nidopallium (LMAN); these subregions give rise to trans-basal ganglia loops as well as trans-cortical loops via the motor cortical regions robust nucleus of the arcopallium (RA) and dorsal intermediate arcopallium (AId) (Fig. 1; Aronov et al., 2008; Bottjer et al., 1984; Bottjer et al., 2000; Luo et al., 2001; Ölveczky et al., 2005; Scharff & Nottebohm, 1991; Vates & Nottebohm, 1995). Core and shell loops through the basal ganglia include connections with different subregions of the medial portion of the dorsolateral anterior thalamic nucleus (DLM): the dorso-lateral subregion of DLM (DLM_DL) projects to LMAN-core, whereas the ventro-medial area of DLM (DLM_VM) projects to LMAN-shell (Carrillo & Doupe, 2004; Iyengar et al., 1999; Johnson et al., 1995). In addition, RA neurons of juvenile (Goldberg & Fee, 2012) and adult zebra finches (Foster et al., 1997; Wild, 1993) project to DLM_DL, and neurons in AId, a cortical region adjacent to RA, project to DLM_VM (Bottjer et al., 2000; Foster et al., 1997). Thus, DLM receives convergent projections from the cortical and basal ganglia circuits that mediate vocal learning and forms a feedback pathway to LMAN (Carrillo & Doupe, 2004; Iyengar et al., 1999; Person et al., 2008).

Figure 1.

A simplified schematic of cortico-basal ganglia circuits that mediate vocal learning and behavior. We made injections of Alexa Fluor dyes into LMAN-core and/or RA in order to examine patterns of anterograde label in RA and AId (see text). Dotted line denotes developmentally regulated projection from LMAN-core to AId that exists in juvenile but not adult birds. We use the term “cortex” in the generic sense as described by Reiner et al. (2004, p. 395) as including that part of the telencephalon that is “pallial in nature and therefore homologous as a field to the brain region of mammals that includes the neocortex, claustrum, and pallial amygdala.” Abbreviations: RA: robust nucleus of the arcopallium; AId: dorsal intermediate arcopallium; LMAN: lateral magnocellular nucleus of the anterior nidopallium; DLM: medial portion of the dorsolateral anterior thalamic nucleus, HVC: letter-based proper name; nomenclature follows the terminology of Reiner et al., (2004).

Vocal production in juvenile birds during early sensorimotor learning is mediated by a projection from LMAN-core to RA, which directly drives vocal motor output (Aronov et al., 2008; Bottjer et al., 1984). Lesions of LMAN-core disrupt production of babbling in juveniles younger than ~50 dph but cause little or no impairment of song production in older juveniles and adult birds (Aronov et al., 2008; Bottjer & Arnold, 1986, Bottjer et al., 1984). Shell circuitry plays an important role in evaluation and refinement of the bird’s own song: lesions of AId in juvenile birds prevent accurate imitation of the goal tutor song but do not produce motor disruption of babbling (Bottjer & Altenau, 2010). The lack of acute performance effects followed by disrupted learning is consistent with the idea that AId receives efference-copy signals (but not motor drive) and participates in evaluative skill learning.

In accord with the idea of efference copy, previous work from our lab based on single-axon reconstructions showed that many RA-projecting neurons in LMAN-core (LMAN-core→RA) send a collateral projection to AId only in juvenile birds (Miller-Sims & Bottjer, 2012). The projection from LMAN-core to AId has been described in birds as old as 35 dph, which corresponds to the earliest age at which juveniles in our breeding colony begin to produce song (Iyengar et al., 1999; Miller-Sims & Bottjer, 2012). However, this collateral projection is completely absent in adults (Bottjer et al., 2000; Iyengar & Bottjer, 2002a; Johnson et al., 1995). Although the projection from LMAN-core to AId is gone by adulthood, the number of LMAN-core→RA neurons does not change during the course of juvenile development (Nordeen et al., 1992). This pattern indicates that the absence of LMAN-core axonal branches in AId is not due to the death of projection neurons, but rather to the retraction of collateral branches of single LMAN-core neurons from AId. In contrast to evidence showing a developmental decrease in the LMAN-core→AId projection, nothing is known about possible developmental changes in the RA→DLM_DL projection in juveniles.

The projections from motor-command neurons in LMAN-core to AId and RA and thence to the thalamus provide pathways by which information about vocal motor output could be reintegrated into cortico-basal ganglia circuitry, potentially aiding in the refinement of juvenile vocalizations as they come to match the goal behavior (tutor song). Thus, examination of the timing of changes in the strength of these projections during vocal motor development in juveniles may shed light on functional changes in neural circuits for vocal learning. To investigate developmental time courses of changes in LMAN-core→AId and RA→DLM_DL axonal connections, we injected LMAN-core and/or RA of juvenile songbirds with fluorescent tracers during early stages of sensorimotor learning (36–49 dph) and analyzed the resultant pattern of terminal label in AId and DLM_DL, respectively.

Materials and Methods

Subjects

All animal procedures were performed in accordance with national regulatory guidelines and followed protocols approved by the Animal Care and Use Committee at the University of Southern California. Surgeries were performed on juvenile male zebra finches between 33 to 46 days post-hatch (dph) (n = 28, Table 1). Birds were perfused three days following surgery, at which point their ages ranged from 36 to 49 dph. Because the fluorescent dye could be transported until the day of perfusion, we used the age at perfusion for all our analyses. All subjects remained in the main colony with their parents in large group aviaries until immediately before surgery. Following surgery, birds younger than 35 dph were returned to their home aviary to be cared for by their parents; birds older than 35 dph were housed in individual holding cages located in the main colony. Thus, all birds remained with their parents until at least 35 dph, and all birds continued to be housed in the main colony up until the day of perfusion. Juvenile birds in our breeding colony begin to produce babbling vocalizations at 35–37 days post-hatch.

Table 1.

Sample Sizes of Experimental Groups

Age	Number of injections into LMAN-core	Number of injections into RA	Number of injections into both LMAN-core and RA	Total number of birds
36–39	6	6	1	11
41–49	9	13	5	17
total	15	19	6	28

Alexa Fluor Dye Injections and Histology

During surgery, each bird was anesthetized with 1.5–1.7% isoflurane (inhalation) and placed in a stereotaxic apparatus. Stereotaxic coordinates were used to approximate the location of RA and LMAN-core; all injections were targeted to the left hemisphere. These locations were confirmed by recording characteristic spontaneous activity of each nucleus using a glass pipette electrode (OD 20 μm) filled with 0.5 M NaCl. Birds received unilateral iontophoretic injections of either one or two dextrans conjugated to different fluorophores: a Grass stimulator (S48) was used to inject either Alexa Fluor 488 (MW 10,000, 10% in sterile-filtered PBS, ThermoFisher Scientific D22910) or Alexa Fluor 594 (MW 10,000, 10% in sterile-filtered PBS, ThermoFisher Scientific D22913) into LMAN-core (n=15) or RA (n=19) by applying pulses of positive current through a silver wire (3–5 μA, 6 sec on/6 sec off) for 20 min through a glass pipette (OD 20–30 μm). Alexa Fluor 488 was used in injections into LMAN-core of 5 birds and RA of 13 birds, and Alexa Fluor 594 was used in injections into LMAN-core of 10 birds and RA of 6 birds. Six birds were injected with fluorescent dye in both RA and LMAN-core (Table 1).

All birds were anesthetized with an injection of 0.05 mL of 5% Euthasol 72 hours following surgery and perfused with 0.7 % saline followed by 10% buffered formalin. The brains were removed and post-fixed in 10% buffered formalin for 72 hours before being cryoprotected in 30% buffered sucrose solution with 0.005% NaN₃ overnight and frozen-sectioned (50 μm thickness, coronal plane). All sections were stored in a covered container to air-dry for 24-hours at room-temperature. Alternate sections were counterstained with thionin for Nissl and coverslipped with Permount (Fisher Scientific). The fluorescent series of sections were coverslipped with Prolong Antifade Diamond Mountant (ThermoFisher Scientific) and patterns of fluorescent label were imaged with confocal microscopy (Zeiss Axioplan 2, LSM 510 Meta or Zeiss LSM 780 Observer). Careful observation revealed that injections of both Alexa Fluor 488 and 594 were transported equally well. However, we present all fluorescent label in green color to avoid the distraction of different colors within multipanel figures. For example, Alexa Fluor 594 was used in the two injections made into LMAN-core shown in Figure 4a–d, while Alexa Fluor 488 was used in the injection shown in Figure 4e–f; using two different colors in this instance would distract from the findings presented in the figure.

Figure 4.

Terminal label in RA and AId following injections into LMAN-core at different ages. Insets depict schematic of injection sites in LMAN-core.

(a) Terminal label in RA and AId following injection into LMAN-core of a 36 dph bird.

(b) Enlarged image of boxed region within AId in (a) showing terminal label that received a rating of 3.

(c) Terminal label in RA and AId following injection into LMAN-core of a 39 dph bird.

(d) Enlarged image of boxed region within AId in (c) showing terminal label that received a rating of 2.

(e) Terminal label in RA and AId following injection into LMAN-core of a 41 dph bird.

(f) Enlarged image of boxed region within AId in (e) showing terminal label that received a rating of 1.

Data Analysis

Injections into LMAN-core that extended beyond the Nissl-defined border of LMAN-core into LMAN-shell (n=5) were not included in our study because LMAN-shell makes a strong projection to AId, thus making it very difficult to determine the origin of terminal label in AId as stemming from core versus shell regions of LMAN (Bottjer et al., 2000; Iyengar et al., 1999; Iyengar & Bottjer, 2002; Johnson et al., 1995). Likewise, some injections into RA extended outside the Nissl-defined border of RA. If the injection site extended into AId (n=3), the data were not analyzed because AId projects to DLM_VM (Bottjer et al., 2000; Foster et al., 1997), and deciphering the origin of axon terminal label would therefore be difficult. Thus, only injections that were completely contained within the borders of each target structure were included for analysis (Table 1).

Terminal label was analyzed qualitatively by an observer unaware of the age of the bird and the size and location of the injection site within LMAN-core or RA. Axon terminals in AId arising from LMAN-core were rated based on terminal density and volume on a scale from 0 to 3; a rating of 0 had no visible terminal label in AId, and a rating of 3 had the highest volume and density of terminal label, normalized to the most intense pattern of label within the LMAN-core→AId projection. We were not able to quantify the volume of terminal label within AId because many labeled LMAN-core axons traverse AId enroute to RA (see Results), making it impossible for imaging programs to discriminate terminal label from axons of passage. Examples of AId terminal label that were judged to have a rating of 3, 2, or 1 are shown in Figure 4 (see Results).

Terminal label in the RA→DLM_DL projection was rated independently also using a scale from 0 to 3; a rating of 0 had no terminal label in DLM_DL and a rating of 3 had the highest volume and/or density of terminal label within this pathway. Thus, rating systems for the two projections were independent of each other since ratings were normalized to the most robust pattern of label within each pathway (i.e., axon terminal label in AId with a rating of 3 is not comparable to terminal label with a rating of 3 in DLM_DL). We utilized separate ratings following injections into LMAN-core versus RA because the resultant terminal label was distributed within post-synaptic targets of substantially different size (AId is larger than DLM_DL) and was differentially distributed within each region (see Results).

We also counted the number of sections in which we observed terminal label in AId as a quantitative estimate of the anterior-posterior (A-P) extent of the terminal field in AId following injections into LMAN-core. Likewise, we counted the number of sections in which we observed terminal label in DLM_DL as a quantitative estimate of the anterior-posterior extent of terminal processes in DLM_DL following injections into RA.

For terminal label in both AId and DLM_DL, we conducted non-parametric Spearman’s rank correlation analyses between age of bird and rating of terminal label, and between age of bird and the A-P extent of terminal label. For correlations we found to be significant (p < 0.05), we conducted K-means clustering utilizing the K-means clustering function on Tableau Desktop 2020.1.

Results

Retrograde label in DLM_DL following injections in LMAN-core

Injections into LMAN-core consistently produced retrogradely labeled neurons restricted to the ipsilateral dorsolateral portion of DLM in birds 36–49 dph (DLM_DL, n = 15 injections, Fig. 2a). This subregion of DLM (DLM_DL) projects to LMAN-core but not to LMAN-shell (Iyengar et al., 1999; Johnson et al., 1995). The region of retrogradely labeled neurons comprised a cluster of brightly labeled neurons, surrounded by numerous lightly labeled cells. We also observed short processes that we identified as putative labeled axons inside the borders of DLM_DL, which likely represent axons of retrogradely labeled somata that project to LMAN-core (Fig. 2a).

Figure 2.

Retrogradely labeled neurons in DLM_DL(a) and LMAN-core (b). Top right insets are schematics of injection sites in LMAN-core (a) and RA (b). Thin grey lines depict the Nissl-stained boundaries of DLM_DL(a) and LMAN-core (b). Medial is left and dorsal is up.

Injections into LMAN-core revealed a clear topographic relationship between the position of somata within DLM_DL and location of the injection site. Injections into lateral LMAN-core produced retrogradely labeled neurons in the ventrolateral portion of DLM_DL (n=3), whereas those into intermediate LMAN-core produced retrogradely labeled neurons along the central region of DLM_DL (n=9), and those into medial LMAN-core produced retrogradely labeled neurons in the dorsomedial portion of DLM_DL (n=3). Previous work from our lab has shown that the projection from DLM_DL to LMAN-core has achieved topographic specificity by 20 dph in juveniles, since the same topography is observed in 20-dph and adult birds (Iyengar et al., 1999). Our results show that topographic organization of the DLM_DL→LMAN-core pathway is preserved during the period of sensorimotor integration, confirming a lack of developmental change. However, our results also revealed a diagonal pattern in which dorsomedial-to-ventrolateral locations within DLM_DL projected to progressively more lateral locations within LMAN-core. These results appear to resolve a minor discrepancy in the descriptions of a dorsal-to-ventral pattern of organization in DLM_DL described by Iyengar et al. (1999) which contrasted with a more medial-to-lateral DLM_DL topography described by Johnson et al. (1995). That is, medial-to-lateral locations within LMAN-core are mapped onto dorsomedial-to-ventrolateral locations in DLM.

Retrograde label following injections into RA

Injections into RA resulted in retrogradely labeled neurons restricted to ipsilateral LMAN-core in birds 36–49 dph (n = 19 injections, Fig. 2b). Retrogradely labeled neurons in LMAN-core had brightly labeled somata, and primary dendrites of heavily labeled neurons were also visible (Fig. 2b). Careful inspection revealed that injections into dorsal RA produced retrogradely labeled neurons in the medial half of LMAN-core (n=3), injections into central RA produced retrogradely labeled neurons within intermediate LMAN-core (n=7), and injections into ventral RA produced retrogradely labeled neurons in the lateral half of LMAN-core (n=9). There was some overlap in the distribution of retrogradely labeled cells between these three subregions within LMAN-core. In addition, while most of the retrogradely labeled neurons in LMAN-core tended to be clustered around these topographically organized positions, sparsely distributed labeled somata were scattered outside these primary topographic locations. For example, Figure 2b shows an instance in which an injection into dorsal RA produced retrogradely labeled neurons mostly clustered within the medial side of LMAN-core as expected; however, we also observed a low density of retrogradely labeled neurons within the lateral portion of LMAN-core.

We also observed retrogradely labeled neurons in ipsilateral HVC following injections into RA (n=19). However, the focus of our study was on developmental changes that occurred within the cortico-thalamic circuits that mediate song learning and production; and thus we did not analyze retrograde label in HVC neurons.

Anterograde label following injections into LMAN-core

LMAN-core to RA Projection

Following injections into LMAN-core (n = 15), we observed numerous labeled axons that entered the dorsal border of AId and then turned sharply within AId to enter RA from its lateral border (Fig. 3, white arrow). As reported by Miller-Sims & Bottjer (2012), LMAN-core→RA axons frequently bifurcated dorsal to AId, giving rise to one branch that traversed the dorsal border of AId and a second branch that extended medially and entered RA via its dorsal border (Fig. 3, gray arrow heads). All injections into LMAN-core consistently produced a high density of terminal label within ipsilateral RA consisting of a dense network of fine branching processes (Fig. 3; Fig. 4a, c, e).

Figure 3.

Anterogradely labeled LMAN-core axons and terminal fields in RA and AId. White arrow points to an example of LMAN-core→RA axon entering the dorsal border of AId before turning sharply medially towards RA. Grey arrow heads point to bifurcation points dorsal to AId of LMAN-core→RA axons.

All 15 injections into LMAN-core produced topographically organized terminal label within RA. Injections into medial LMAN-core resulted in terminal label in dorsal RA (n = 3), injections into intermediate LMAN-core resulted in terminal label in central RA (n = 9), and injections into lateral LMAN-core resulted in terminal label in ventral RA (n = 3). The spatial organization of LMAN-core→RA anterograde label confirmed the topographic pattern established by our retrograde label experiments (see above). These results also confirmed previous findings of broad patterns of topographic organization within the projection from LMAN-core to RA in both 35-dph juvenile and adults (Iyengar & Bottjer, 2002b; Johnson et al., 1995).

LMAN-core to AId Projection

Previous work from our lab found that individual RA-projecting LMAN-core neurons make a collateral projection to AId in juvenile birds between 20 and 35 dph; although this projection shows no diminution during this interval, it is completely absent in adults (Iyengar et al., 1999; Iyengar & Bottjer, 2002b; Johnson et al., 1995; Miller-Sims & Bottjer, 2012). The number of RA-projecting neurons in LMAN-core does not change during vocal development (Nordeen et al., 1992), suggesting that collateral arborizations within AId are retracted at some developmental time point between 35 dph and adulthood.

To characterize developmental changes in the projection from LMAN-core to AId following 35 dph, we rated the pattern of terminal label within AId following injections into LMAN-core. Ratings were made without knowledge of the age of the bird and the location of the injection site (see Materials and Methods). A rating of 3 signified a high density and volume of terminal label in AId as shown in Figure 4a–b; this injection produced terminal label consisting of a network of densely branched processes with a high density of punctae which occupied a large volume within AId. An injection that produced a lower density of branched processes with fewer punctae, and which occupied a smaller volume of AId, received a rating of 2 (Fig. 4c–d). Terminal label within AId which received a rating of 1 occupied a relatively small volume and included very few branched processes with a low density of punctae (Fig. 4e–f).

Ratings of terminal label within AId were highly age dependent. Injections into LMAN-core resulted in robust terminal label in AId of birds aged 36–39 dph: all birds within this age group received a rating of either 2 or 3 (n = 6, Table 2A). However, terminal label in AId showed a rapid diminution with age between 39 and 41 days of age (Table 2B). In contrast to the densely branched labeled terminal processes within AId of a 36 dph bird (Fig. 4a–b), AId of a 39 dph bird contained substantially fewer labeled branches and spanned a smaller area of AId (Fig. 4c–d, rating of 2). Injections into LMAN-core of eight out of nine birds older than 40 dph resulted in substantially diminished terminal label within AId. For example, AId of a 41 dph bird exhibited only a few labeled processes that spanned a considerably smaller volume compared to that of any younger bird (Fig. 4e–f, rating of 1). This pattern suggests that a larger number of birds within this age range might reveal a more progressive decrease.

Table 2.

Age, Location of Injection, and Rating Assigned to AId Terminal Label following LMAN-core Injections

Bird ID	Age ‡	Rating	# of sections with label	Injection site (anterior – posterior)	Injection site (dorsal – ventral)	Injection site (medial – lateral)	LMAN-core volume	Injection site volume
Table 2A: Injections into LMAN-core of birds aged 36–39 dph
Gy72	36	3	4	anterior	ventral	intermediate	0.166	0.010
W107†	37	3	8	anterior	intermediate	medial	0.179	0.008
Gy51	37	2	2	intermediate	ventral	intermediate	0.158	0.008
W145	38	3	4	intermediate	ventral	intermediate	0.098	0.008
Y2	39	3	4	posterior	ventral	intermediate	0.178	0.007
O168B	39	2	5	anterior	dorsal	intermediate	0.173	0.009
Average	37.7	2.67	4.50				0.158	0.008
St Dev	1.2	0.52	1.97				0.031	0.001
Table 2B: Injections into LMAN-core of birds aged 41–49 dph
O58	41	1	3	intermediate	ventral	lateral	0.170	0.010
O66	41	1	3	anterior	dorsal	intermediate	0.241	0.008
Dg14†	42	1	2	posterior	intermediate	intermediate	0.176	0.008
Gy112	45	2	5	intermediate	dorsal	intermediate	0.198	0.006
Dg20†	45	1	2	anterior	ventral	medial	0.207	0.009
W184†	45	1	2	intermediate	dorsal	lateral	0.173	0.007
Db43†	47	0	0	intermediate	dorsal	intermediate	0.145	0.006
Gy117B	48	0	0	anterior	dorsal	lateral	0.147	0.005
Gy117†	49	0	0	posterior	ventral	medial	0.189	0.008
Average	44.8	0.78	1.89				0.183	0.007
St Dev	2.9	0.67	1.69				0.030	0.002

^†birds that received dual injections into LMAN-core and RA; the pattern of label between the two regions never interfered and so all these injections were used for independent analysis

^‡age of birds listed in the table reflects the age of the birds at time of perfusion. All birds were perfused 3 days following surgery to allow for the transportation of fluorescent dye.

We calculated the regression between ratings of terminal label and the age of all birds that received injections in LMAN-core to assess the developmental decrease in the projection from LMAN-core to AId. This analysis revealed a significant negative linear correlation (Fig. 5a, p < 0.001, ρ = -0.873), consistent with the idea that retraction of LMAN-core axon collaterals from AId may occur gradually. However, a K-means clustering algorithm based on ratings of terminal label and age of birds in our sample revealed two distinct age groups, 36–39 dph (n = 6) and 41–49 dph (n = 9) (Fig. 5b, Table 2). In contrast to the robust terminal label of all birds in the younger age group (ratings of either 2 or 3), almost all birds in the older age group received ratings of 0 or 1. This difference in rating of terminal label in the LMAN-core→AId projection was significant between age groups: the younger group received significantly higher ratings than did birds in the older group (Fig. 5b, Mann-Whitney, p < 0.01).

Figure 5.

Scatterplots and box-and-whisker plots of rating and anterior-posterior extent of terminal label in AId by age following injections into LMAN-core.

(a) Scatter plot and correlation analysis of ratings of terminal label in AId by age, Spearman’s correlation.

(b) Box-and-whisker plot of ratings of terminal label in AId by age groups 36–39 dph (n = 6) and 41–49 dph (n = 9). Box-and-whisker plots indicate medians and first and third quartiles, whiskers indicate minimum and maximum values, and circles represent data points from individual birds; all data points were included in statistical analysis, **p<0.01; Mann-Whitney.

(c) Scatter plot and correlation analysis of number of sections of terminal label (anterior-posterior extent) in AId by age; Spearman’s correlation.

(d) Box-and-whisker plots of number of sections of terminal label (anterior-posterior extent) in AId of individual birds by age groups 36–39 dph (n = 6) and 41–49 dph (n = 9), plotted as in 5b; all data points were included in statistical analysis, *p < 0.05; Mann-Whitney.

We also counted the number of sections that contained terminal label in AId following an injection into LMAN-core as a quantitative estimate of the anterior-posterior extent of the terminal field in AId arising from somata in LMAN-core. The linear regression between the number of sections with terminal label within AId and the age of bird revealed a significant negative correlation between the anterior-posterior extent of AId terminal label and age, again suggesting that collateral projections from LMAN-core gradually retract from AId between 36 and 49 dph (Fig. 5c, p = 0.004, r = -0.690). We wondered if the anterior-posterior span of terminal label matched the age groups we found through K-means clustering based on the rating of terminal label and the age of bird. In accord with this idea, we found that the number of sections with terminal label in AId was significantly higher in the younger age group of birds (on average 4.5 sections for birds 36–39 dph, Table 2) than in the older age group (on average 1.89 sections for birds 41–49 dph, Table 2) (Fig. 5d, Mann-Whitney, p=0.0262). On average AId spanned seven sections as defined by Nissl staining (we counted the number of sections that included RA as a measure of the number of sections spanned by AId since AId is always adjoined to the lateral margin of RA and the two nuclei are coplanar on coronal sections).

Both our qualitative and quantitative analyses indicate that LMAN-core→AId collateral axons start to retract between 39 and 41 dph and is gone by 47–49 dph. Interestingly, birds from our breeding colony begin to sing ~35–37 dph, so this projection begins to be retracted soon after the onset of song production, which is thought to correspond to the start of the sensorimotor learning phase.

We also observed anterograde label in the ipsilateral basal ganglia following injections into LMAN-core (n=15). This projection provides another potential source of efference copy (Fee, 2012). However, since the focus of our study was on developmental changes within axonal connections to and from motor cortex, we did not analyze this terminal label. Despite the fact that single LMAN-core neurons send axonal branches to both RA and the basal ganglia, our previous work has shown no developmental changes in the projection to the basal ganglia (Iyengar et al., 1999; Nixdorf-Bergweiler et al., 1995; Vates & Nottebohm, 1995).

Anterograde label following injections into RA

RA to DLM_DL Projection

We wondered if the sparseness of terminal processes within DLM_DL reported previously (Foster et al., 1997; Goldberg & Fee, 2012; Wild, 1993) was due to retraction of axonal processes from RA during early sensorimotor learning. To investigate this possibility, we rated the terminal label observed in DLM_DL of all juvenile birds that received an injection into RA. Our system for rating terminal label in the RA→DLM_DL projection was determined based on the relative incidence of label across all birds that received injections of RA and was independent of the ratings made for label within AId. That is, ratings from 0 to 3 were made based only on comparisons to terminal label within DLM_DL, without reference to AId (see Methods). Figure 6a–b shows an example of terminal label in DLM_DL that received a rating of 3, including multiple branched processes that covered the entire dorsolateral edge of DLM_DL. Careful qualitative examination of terminal label in DLM_DL following injections into RA indicated that ratings were unaffected by age. For example, the pattern of label in two different birds aged 39 versus 44 dph was highly comparable, containing prominently labeled branched processes along the dorsolateral edge of DLM_DL in both cases, and each received a rating of 3 (Fig 6a–d).

Figure 6.

Terminal label in DLM_DL following injections into RA at different ages. Top left insets are schematics of injection sites in RA. Thin grey lines depict Nissl-stained boundaries of DLM_DL.

(a) Terminal label in DLM_DL following injection into DLM_DL of a 39 dph bird.

(b) Enlarged image of boxed region within DLM_DL in (a).

(c) Terminal label in DLM_DL following injection into RA of a 44 dph bird.

(d) Enlarged image of boxed region within DLM_DL in (c).

We calculated the linear regression between ratings of terminal label in DLM_DL and the ages of all birds that received an injection into RA to assess whether the incidence of terminal label remained constant throughout early sensorimotor learning. Our regression analysis revealed no significant effect of age, indicating that RA→DLM_DL terminal label remained constant between 36 and 49 dph (p = 0.453, ρ = -0.183, Fig. 7a, Table 3). We also counted the number of sections with terminal label in DLM_DL after an injection into RA as a quantitative estimate of the anterior-posterior span of terminal label. To determine if the anterior-posterior extent of the RA→DLM_DL projection is developmentally regulated, we conducted a regression analysis between the number of sections with terminal label found in DLM_DL and the age of all the birds that received an injection into RA. This analysis revealed no significant effect of age on the number of sections with terminal label across the anterior-posterior axis of DLM_DL (p = 0.230, ρ = -0.289, Fig. 7b, Table 3). On average the labeled terminal field in DLM_DL: spanned 2.63 sections; DLM_DL spanned a total of 4–5 sections on average as defined by Nissl-staining.

Figure 7.

Scatterplots of rating and anterior-posterior extent of terminal label in DLM_DL by age following injections into RA.

(a) Scatter plot and correlation analysis of ratings of terminal label in DLM_DL by age; Spearman’s correlation.

(b) Scatter plot and correlation analysis of number of sections of terminal label (anterior-posterior extent) in DLM_DL by age; Spearman’s correlation.

Table 3.

Age, Location of Injection, and Rating Assigned to DLM Terminal Label following RA Injections

Bird ID	Age‡	Rating	# of sections with label	Injection site (anterior – posterior)	Injection site (dorsal – ventral)	Injection site (medial – lateral)	RA volume (mm^3)	Injection site volume (mm^3)
O38	36	2	3	intermediate	intermediate	lateral	0.207	0.005
O53	36	1	4	posterior	intermediate	intermediate	0.177	0.006
W107†	37	1	3	posterior	ventral	intermediate	0.149	0.004
Db55	38	0	0	intermediate	ventral	lateral	0.249	0.007
Y171	39	3	7	posterior	intermediate	intermediate	0.297	0.004
W171	39	3	6	anterior	dorsal	medial	0.206	0.003
Dg83	41	1	1	anterior	dorsal	medial	0.241	0.006
Bk105	41	0	0	intermediate	ventral	intermediate	0.246	0.005
Bk90	42	3	6	posterior	intermediate	intermediate	0.216	0.007
W153	42	3	5	intermediate	ventral	lateral	0.202	0.002
Dg14†	42	0	0	anterior	ventral	lateral	0.187	0.001
W104	44	3	9	posterior	dorsal	medial	0.256	0.007
Dg20†	45	0	0	intermediate	intermediate	intermediate	0.216	0.003
W124	45	0	0	intermediate	ventral	intermediate	0.275	0.005
W184†	45	0	0	posterior	intermediate	lateral	0.192	0.004
Db30	46	0	0	intermediate	ventral	intermediate	0.212	0.006
Db43†	47	1	2	intermediate	ventral	intermediate	0.265	0.005
W143	48	1	1	intermediate	ventral	intermediate	0.235	0.006
Gy117†	49	2	3	anterior	intermediate	intermediate	0.256	0.002
Average	42.2	1.26	2.63				0.225	0.004
St Dev	4.0	1.24	2.83				0.037	0.002
Average and St Dev excluding sections without visible terminal label
Average	41.7	2.00	4.17				0.225	0.005
St Dev	4.6	0.95	2.48				0.041	0.002

^†birds that received dual injections into LMAN-core and RA; the pattern of label between the two regions never interfered and so all these injections were used for independent analysis

^‡age of birds listed in the table reflects the age of the birds at time of perfusion. All birds were perfused 3 days following surgery to allow for the transportation of fluorescent dye.

We tried to discern a potential topographic relationship by examining the relationship between the position of the injection site in RA and the location of the resultant terminal label along the anterior-posterior, dorsal-ventral, and medial-lateral axes of DLM_DL. We found that terminal label within DLM_DL was located either in the lateral portion (n = 5) or along the dorsal edge (n=4) or was found dispersed throughout DLM_DL (n = 3). The location of terminal label within DLM_DL appeared to be unrelated to the location of the injection site. This was unexpected, since all projections to and from LMAN are topographic (although the projection from HVC to RA is non-topographic; (Fortune & Margoliash, 1995; Yip et al., 2012). Thus, we were unable to find an obvious topographic relationship between the position of somata in RA and the location of their terminal fields in DLM_DL.

RA Projections to DM, nXIIts, RAm, and PAm

All injections into RA resulted in robust terminal label in several midbrain and hindbrain targets: the dorsomedial nucleus of the intercollicular complex (DM), the tracheosyringeal portion of the hypoglossal nucleus (nXIIts), nucleus retroambigualis (RAm), and nucleus parambigualis (PAm). Consistent with previous findings (Gurney, 1981; Wild, 1993; Wild et al., 2000), labeled axons from RA entered ipsilateral DM along its medial border and produced a dense network of terminal processes (Fig. 8a). Injections into RA also produced labeled terminal fields in ipsilateral nXIIts, consisting of a dense network of terminal branches with multiple brightly labeled punctae (Fig. 8b). In sections of the medulla that included label in anterior nXIIts, a slightly sparser yet prominent terminal field was labeled laterally and ventrolaterally to nXIIts within ipsilateral PAm. In sections of the medulla with label in posterior nXIIts, a brightly labeled terminal field composed of a dense network of branched processes was found laterally and ventrolaterally to nXIIts within ipsilateral RAm (Fig. 8b). Thus, our findings confirmed previous work showing that projections of RA to nXIIts also form adjacent terminal fields in ipsilateral PAm or ipsilateral RAm depending on the relative anterior-posterior location (Wild, 1993; Wild et al., 2000).

Figure 8.

Terminal label in midbrain and hindbrain regions following an injection into RA of the same bird when no label was seen in DLM_DL.

(a) Terminal label in DM following an injection into RA. Inset in bottom right corner depicts schematic of injection site.

(b) Terminal label in nXIIts and RAm stemming from the same injection as shown in (a).

(c) No terminal label was seen in DLM_DL following the same injection as shown in (a). The occipitomesencephalic (OM) fiber bundle was labeled in the top right corner. Thin grey lines depict the Nissl-stained boundaries of DLM_DL.

Injections into RA consistently resulted in labeling of more densely branched terminal processes in DM, nXIIts, RAm, and PAm compared to terminal label in DLM_DL (Fig. 8a–b, cf. Fig 6a–d). Furthermore, terminal label in DLM_DL was not visible in some birds, despite prominent terminal label nXIIts, RAm, PAm, and DM stemming from the same injection site in RA (n=7). For example, a single injection into RA (shown in the inset of Fig. 8a) resulted in robust terminal label in ipsilateral DM, nXIIts, RAm, PAm (Fig. 8a–b; terminal label in PAm not shown for simplicity); however, that same injection into RA failed to produce visible terminal label within DLM_DL (Fig. 8c). We reconducted our regression analyses of DLM_DL terminal label analyzing only birds with visible terminal label in DLM_DL. As in our earlier analyses, we found no significant effect of age on the rating of DLM_DL terminal label (p = 0.851, ρ = 0.061, n = 12) or on the anterior-posterior span of DLM_DL terminal label (p = 0.550, r= -0.192, n = 12). When analyzing only birds with visible terminal label in DLM, we found that injections in RA resulted in terminal label within DLM_DL with an average rating of 2.00 and on average spanned 4.17 sections (Table 3, bottom). In contrast, injections into RA across all birds had an average rating of 1.26 and extended across 2.63 sections of DLM_DL (Table 3, all injections).

There was no mention of this phenomenon in previous literature (Foster et al., 1997; Goldberg & Fee, 2012; Wild, 1993). Thus, it is currently unknown why some of our injections failed to produce terminal label within DLM_DL despite resulting in prominently labeled terminal processes in other downstream targets of RA. One possibility we investigated was that only specific subregions of RA project to DLM_DL. To test whether specific subregions of RA do not project to DLM, we analyzed the pattern of terminal label in DLM_DL following injections into different locations within RA. Interestingly, injections into comparable locations within RA could result in terminal label within DLM_DL in one bird but not in another. For example, we made injections into ventrolateral RA of two separate birds: W153 and Db55 (Table 3). Although we observed robust terminal label within DLM_DL in W153 (rating of 3), we did not observe any terminal label in DLM_DL of Db55. This finding suggests that the absence of terminal label in DLM_DL may not be attributable to the location of the injection site in RA, but rather due to limited transport of fluorescent dyes in RA-to-DLM_DL axons or because of variation in dye uptake by DLM_DL-projecting RA neurons at the injection site.

Discussion

The onset of sensorimotor learning begins around 35 days post hatch when juvenile zebra finches begin to babble and gradually learn to translate the memory of tutor song into corresponding vocal motor output. During early stages of sensorimotor learning, the projection from LMAN-core to RA drives song production, and lesions of LMAN disrupt song production (Aronov et al., 2008; Bottjer et al., 1984). LMAN-core is thought to drive variability in vocal output of juvenile birds, and birds engage in vocal motor exploration as they produce highly variable vocalizations (Achiro et al., 2017; Ölveczky et al., 2005). Motor-related activity in LMAN-core may increase during sensorimotor learning, and the increased informational content could be conveyed in part by the recurrent loops formed by the projections from LMAN-core to RA and AId and in turn to DLM. We report here that the collateral projection from LMAN-core to AId is retracted soon after the onset of babbling behavior, whereas the projection from RA to DLM is fixed and sparse in both juvenile and adult birds. We investigated the time course of developmental changes in these two pathways as a first step in examining whether changes in the strength of axonal connectivity might correlate with a restricted stage of sensorimotor learning.

A previous study from our lab found a prominent projection from LMAN-core to AId in both 20 and 35 dph birds (Miller-Sims & Bottjer, 2012). We found that this collateral projection started to diminish rapidly around 40 dph, soon after the onset of babbling, and disappeared altogether by 47–49 dph. This finding was unexpected since the onset of song vocalizations begins ~35–37 dph in our colony, and birds engage in an extended period of sensorimotor integration as they gradually learn to imitate the tutor song. The transition from babbling (subsong) to plastic song occurs ~35–45 dph (Achiro et al., 2017; Aronov et al., 2008), which may correlate with the gradual disappearance of the LMAN-core→AId projection. In addition, although we investigated structural changes in axonal connectivity after the onset of vocal babbling, sensorimotor learning could potentially occur prior to the onset of babbling. Hatchling zebra finches produce begging calls to elicit parental care until they can care for themselves around 30–35 dph. Begging calls in male chipping sparrows are disrupted by lesions of RA; these calls slowly translate into vocal babbling, suggesting that begging calls may be a first step in vocal learning (Liu et al., 2009; cf. Villain et al., 2015). Thus, LMAN-core→AId collaterals may convey efference copy of motor commands of early vocalizations that facilitate mapping of articulatory-auditory correspondences prior to the onset of babbling. Sensorimotor correspondences are encoded in the firing patterns of RA neurons in adult zebra finches (Dave & Margoliash, 2000), but no studies have examined the time course of the emergence of such mapping. The present results suggest that integration of vocal motor output and auditory feedback may emerge quite early in zebra finches.

Previous research from our lab found that AId does not respond to playback of song stimuli despite receiving major inputs from LMAN-shell, which responds to diverse song stimuli in juvenile birds (Achiro & Bottjer, 2013; Bottjer, 2004; Bottjer et al., 2000; Bottjer & Johnson, 1992; Johnson et al., 1995; Yuan & Bottjer, 2019). Because LMAN-core neurons also respond to playback of diverse song stimuli in juvenile birds (Achiro & Bottjer, 2013; Doupe & Solis, 1997; Doupe, 1997; Solis & Doupe, 1997), it is possible that the transient projection from core to AId could elicit song-evoked activity in AId neurons. If so, such activity would be restricted to juvenile birds younger than 40–42 dph. Thus, the rapid disappearance of the LMAN-core→AId projection we observed could partially explain why Yuan & Bottjer (2019) did not find playback-evoked responses in AId of juveniles since they recorded AId neurons in birds aged 42–48 dph. Investigation of AId responses to song playback in younger juveniles might reveal responsivity to song playback in AId neurons when LMAN-shell projections converge with LMAN-core projections in AId.

Unlike LMAN-core, LMAN-shell has no direct role in driving motor production in juvenile songbirds; rather, LMAN-shell neurons are considered to play a role in song learning (Achiro et al., 2017; Bottjer & Altenau, 2010). LMAN-shell contains separate subpopulations of neurons that are tuned to playback of either the juvenile’s own song or the juvenile’s memorized tutor song (Achiro & Bottjer, 2013). In addition, lesions of AId during sensorimotor integration lead to disruption of song learning; juvenile birds that received AId lesions showed a disruption in stable syllable sequencing and accurate copying of tutor syllables as adults despite producing syllables with normal acoustic morphology (Bottjer & Altenau, 2010). In awake birds that are actively engaged in song production, juvenile utterances evoke a differential response in LMAN-shell neurons depending on how similar (or dissimilar) they are to the tutor song (Achiro et al., 2017). Efference copy regarding ongoing motor production from LMAN-core could converge with information about tutor similarity of self-generated utterances from LMAN-shell in AId. Thus, AId neurons may be well situated to play a role in mediating comparisons of juveniles’ immature sounds to their goal tutor sounds during an early phase of vocal development.

AId projects to several post-synaptic targets, including DLM_VM; the ventral tegmental area (VTA); deep layers of the tectum, which project to the thalamic song-control nucleus Uva; a nucleus in the caudal thalamus (medial spiriform nucleus, SpM), which projects to the cerebellum; a dorsal thalamic zone that includes DLM as well as DMP, which feeds forward to HVC; and a restricted region of the lateral hypothalamus (external cellular stratum of the lateral hypothalamus, SCE), which may also relay information to the DLM/DMP dorsal thalamic zone (Bottjer et al., 2000; Nicholson et al., 2018). The projection from AId to VTA overlaps with the region of VTA neurons that project to the region of the songbird basal ganglia that mediates vocal learning(Area X) (Bottjer et al., 2000). Thus, efferent projections of AId may contribute to integration of reinforcement signals from dopaminergic neurons into cortico-basal ganglia loops and thereby facilitate learning. Combined with inputs conveying efference copy of vocal motor commands via the transient LMAN-core→AId projection and inputs representing song-selective tuning from LMAN-shell, AId neurons may play an important role in one or more aspects of vocal learning during a restricted window of development.

Because RA neurons directly drive vocal motor neurons (nXIIts), the pathway from RA to DLM_DL could also provide efference copy of motor signals in juvenile (Goldberg & Fee, 2012) and adult zebra finches (Foster et al., 1997; Wild, 1993). Multiple sources of efference copy might be useful in juveniles for providing cortico-basal ganglia circuitry with information about self-generated motor output during vocal learning, and we therefore predicted that the projection from RA to DLM_DL would be stronger in juveniles during early sensorimotor learning. However, the RA to DLM_DL projection did not change in volume or density of terminal arborizations in the early stages of sensorimotor learning that we examined. The terminal label in DLM_DL following RA injections tended to occupy small areas of DLM_DL in juvenile birds with a low density of terminal arborizations. This outcome does not rule out the possibility that a more prominent RA→DLM_DL projection could exist in younger juveniles before the onset of sensorimotor learning.

The topographic organization of the projection from LMAN-core to RA is refined between 20 and 35 dph in an experience-dependent manner (Iyengar et al., 1999; Iyengar & Bottjer, 2002a). Thus, despite the lack of apparent morphological changes in the RA→DLM_DL projection, important changes in information transfer from RA to DLM_DL may be occurring during early vocal development. In addition, the topography of DLM_DL to LMAN-core and DLM_VM to LMAN-shell is being refined between 20 and 35 dph (Iyengar & Bottjer, 2002a). In particular, the shell subregion of LMAN and the terminal field of DLM axons within LMAN undergo a striking increase in overall volume during early stages of vocal learning followed by a substantial decrease. The dramatic changes in the overall DLM-to-LMAN circuit are due to dynamic rearrangements in individual DLM axon arbors, which include large-scale retraction suggesting decreased overlap in these axonal projections and an increase in the precision of sensorimotor mapping (Bottjer & Johnson, 1992; Iyengar & Bottjer, 2002a). It is interesting that multiple axonal projections are maturing at such an early age (20–35 dph), presumably at a time when the collateral projection from LMAN-core to AId is strong. The developmental decrease in connectivity between LMAN-core and AId may be, at least in part, a downstream consequence of maturation in these related pathways, which could represent either morphological correlates of song learning or essential prerequisites for acquisition of song. In any case, substantial changes in the neural substrate such as the decreased connectivity we report here between LMAN-core and AId are intertwined with changes in learning: projections that are present only during early stages of sensorimotor learning may mediate unique, temporally restricted processes of goal-directed learning.

Data availability:

all the data for this study consists of stored confocal images, which are not archived in any repository; specific images will be made available upon request.