Behavioral vocal learning assay
To assess the effect of adult age on vocal learning ability, we formed novel flocks of adult male budgerigars (Melopsittacus undulatus), belonging to the same age class, either “young adult” (6 mo – 1 year) or “older adult” (≥ 3 year) who were close to or exceeding the mean life expectancy of 4.57 years for this species in captivity, [24] and then collected audio-recordings from all individuals over a 20-day learning period to measure changes in contact call repertoires over time and assess age-related differences in call learning (Fig. 1). Full details of this behavioral vocal learning assay are described in Moussaoui et al., [23] as a subset of birds from that main experiment were used in this study focused on the neural underpinnings of age-related differences in vocal learning ability. In brief, 24 birds of each adult age class were acquired from a commercial breeder (McDonald Bird Farms, Kerrville, TX) and from our own research colony at the New Mexico State University Animal Care Facility, which was derived of birds from the same breeder. The commercial breeder provided young birds and old birds housed in four separately built aviaries and three separately built aviaries, respectively, from a total breeding population that exceeds 10,000 individuals and thus is unlikely to be inbred. These birds, in combination with a set of old males from our colony, generated four independent source populations for each age class such that birds originating from different populations were socially and acoustically unfamiliar and could thus be combined to form flocks of novel membership. Six replicate flocks were formed for each adult age class with each flock being composed of four individuals, such that each individual in a replicate flock originally belonged to a separate source population and thus was unfamiliar to its flockmates. Flocks were housed in 79 × 52 × 102 cm cages with matching layouts of perches, food dishes, and water. Prior to novel flock formation, baseline contact call repertoires were collected for each individual during a 4-day audio-recording block (block 1 in Fig. 1). Upon being placed with novel flockmates, birds were audio-recorded daily across four 4-day blocks (blocks 2–5 in Fig. 1). At the end of this vocal recording period, neural tissue was collected for a randomly selected subset of individuals to measure expression of a key vocal learning related gene. All procedures conducted in this study were approved by the New Mexico State University Animal Care and Use Committee (protocol number 2020-030).
Experimental timeline. Experimental timeline outlining the formation of novel flocks, audio-recording of vocalizations, and neural tissue collection
Acoustic analysis to measure vocal learning
Contact calls were isolated from these audio-recordings using a semi-automated signal detection procedure in the R package ohun (version 1.0.0) [25] in R version 4.0.5 [26]. This involved applying optimized amplitude, frequency, and duration thresholds, the use of supervised random forests to classify detections as “signal” (contact calls) or “noise” (other vocalization types, feather ruffling, cage rattling, background flockmates), and lastly a manual quality control step in which we visually confirmed detections classified as contact calls. Seventeen standard acoustic features were then measured from contact call spectrograms, including various features related to the distribution of power in the time and frequency domains, using the R package warbleR (version 1.1.27) [27]. The dimensionality of these multiple extracted acoustic measures was reduced using t-Distributed Stochastic Neighbor Embedding (t-SNE) [28] implemented in the R package Rtsne (version 0.15) [29]. We mapped contact calls in an acoustic trait space, hereafter “acoustic space”, generated by projecting the first two dimensions such that acoustically similar calls appear closer together in space. We then quantified the kernel density area of each individual’s acoustic space subset by recording block using the R package PhenotypeSpace (version 0.1.0) [30] to assess changes in contact call repertoires over time. All acoustic spaces were generated using the same number of contact calls (180) using a rarefaction subsampling procedure to get the mean area from 30 equal size randomly selected subsets.
From these acoustic space areas, three vocal learning measures were computed for each bird. Firstly, we defined vocal diversity as the change in acoustic space area of an individual’s contact call repertoire from the beginning of the vocal learning assay (audio-recording block 1) to the end (audio-recording block 5). Secondly, we defined vocal plasticity as 1 minus the intersection over union of an individual’s beginning and ending acoustic space areas, where higher values indicate less acoustic similarity between initial and final contact calls, and thus greater vocal plasticity. Thirdly, we defined vocal convergence as the intersection over union of an individual’s acoustic space area and the combined acoustic space area of its flockmates at the end of the vocal learning assay, where higher values indicate greater matching of ones’ contact call repertoire to that of its social group.
Neural tissue collection and preparation
Following the vocal learning assay (Fig. 1), two birds from each flock of 4 birds (N = 12 individuals per age class) were randomly selected for sacrifice to collect whole brains for neural analysis of the FoxP2 gene. As previous work has shown that adult budgerigars exhibit persistent downregulation of FoxP2 regardless of vocal state, we did not record vocal output or time spent vocalizing immediately prior to euthanasia [13]. On day 28 of the experimental timeline, selected birds were euthanized via an overdose inhalation of isoflurane and whole brains were extracted and flash frozen within 5 min using liquid nitrogen and then stored at -80 °C until later use. All collected brains were cryosectioned coronally using a Leica CM1850 cryostat (Leica Microsystems) at -20 °C. The left or right hemisphere of each brain was randomly selected for extraction of 1 mm deep punches with an 18 gauge Luer stub from both MMSt and VSP for future RNA isolation work. The non-punched hemisphere (young adults: N = 6 RH, 3 LH; older adults: N = 5 RH, N = 4 LH) was used in this study for immunohistochemical staining for FoxP2 protein expression. Sections of 20 μm thickness were thaw-mounted onto positively charged slides (Fisher Scientific) in 7 replicate series and stored at -80 °C. One series was Nissl stained for visualization of cytoarchitectural boundaries to enable identification of the key brain regions of interest, MMSt and its adjacent striatum. With reference to the budgerigar brain atlas, [31] adjacent slides were selected for immunohistochemical staining.
Immunohistochemical staining for FoxP2
Brain sections were first fixed with 4% paraformaldehyde (titrated with NaOH and HCl to achieve a pH of 7) for 5 min, dip-rinsed twice with 1X phosphate buffered saline (PBS), then rinsed three times with 1X PBS with 0.4% Triton X-100 (PBST) for 5 min each. Slides were then blocked with 5% sheep serum (Sigma-Aldrich) in PBST for 1 h at room temperature to prevent nonspecific binding followed by overnight incubation at 4 °C in the FoxP2 primary antibody (Mouse, 1:500, Thermo Fisher Scientific) solution. Slides were then rinsed in 1X PBST three times at 5 min each prior to incubation in the Alexa Fluor 594 secondary antibody (Goat anti-mouse, 1:200, Thermo Fisher Scientific) for 2 h at room temperature. Sections were then rinsed four times at 5 min each in 1X PBS, once in ddH20, and finally coverslipped using Vectashield with 405 nm excitable DAPI (Vector Laboratories). Negative controls were performed identically as above except for the omission of primary antibody.
Quantification of FoxP2 expression
Following immunohistochemistry, tissue slides were imaged using a TCS SP5 II Confocal microscope (Leica Microsystems) to capture fluorescent images for quantification of FoxP2 protein expression. Images were taken within the MMSt and VSP regions from each of two sections per bird at 40X magnification. Images of each region were taken sequentially between frames for each channel (405 nm for DAPI, 594 nm for FoxP2, and their overlay) and saved as TIFF image files (Fig. 2). Images were imported into ImageJ 1.53e (NIH), [32] converted to an 8-bit grayscale and auto-thresholded. DAPI and FoxP2 labeled cells were then manually counted using the multi-point tool while referencing cell morphology in original images. To avoid counting noise arising from secondary antibody background staining, FoxP2 labeled cells were only counted if they overlaid atop a DAPI counted cell. For each image, FoxP2 cell counts were divided by the total number of cells (DAPI counted cells) yielding a percentage of neural cells that were expressing FoxP2 in the MMSt and the VSP, which was then used to calculate a MMSt/VSP FoxP2 expression ratio per section. This MMSt/VSP ratio was averaged for the two imaged sections per bird. Cells were counted by two trained observers and inter-observer reliability was assessed at “good” to “excellent” (ICC = 0.965; 95% CI = [0.769, 0.995]) by employing a single-measurement, absolute-agreement two-way mixed-effects model using the package irr (version 0.84.1) [33].
FoxP2 differential expression. Immunohistochemical staining of FoxP2 protein in the magnocellular nucleus of the medial striatum (MMSt) and the adjacent striatal non-vocal learning region (VSP). Images were taken at 40X magnification using a confocal microscope. These example images are from a single older adult male budgerigar. Blue signal (a and d) indicates DAPI stained cell nuclei, representing the total number of cells present in the area imaged. Red signal (b and e) indicates FoxP2 labeled cell nuclei. DAPI and FoxP2 labeled images are overlaid (c and f) to identify the percentage of total cells that express FoxP2
Quantification and statistical analysis
All statistical analyses were carried out in R version 4.2.1 [26] using the stats (version 4.2.1), car (version 3.1-1), and brms (2.18.0) packages. Statistical significance was determined with an alpha level of 0.05 for frequentist analyses, or a 95% credible interval that did not include zero for Bayesian analyses. To test whether FoxP2 expression differs by adult age, we conducted independent samples t-tests for FoxP2 expression in MMSt, VSP, and the MMSt/VSP expression ratio. To determine whether neural FoxP2 expression predicts vocal learning, and whether this relationship differs between young and older adults, we fit Bayesian generalized linear models, with each of the three vocal learning measures (vocal diversity, vocal plasticity, and vocal convergence) as a response variable in three separate models and adult age class and mean centered FoxP2 MMSt/VSP expression as explanatory variables. Regressions were run in Stan [34] through the R package brms [35]. Response variables expressed as proportions (vocal plasticity and vocal convergence) were modeled with a Beta distribution while vocal diversity was modeled with a normal distribution. Effect sizes are presented as median posterior estimates and 95% credibility intervals as the highest posterior density interval. Minimally informative priors were used for population-level effects. Models were run on four chains for 10,000 iterations, following a warm-up of 10,000 iterations. The effective sample size was kept above 3000 for all parameters. Performance was checked visually by plotting the trace and distribution of posterior estimates for all chains. We also plotted the autocorrelation of successive sampled values to evaluate the independence of posterior samples, kept the potential scale reduction factor for model convergence below 1.01 for all parameter estimates and generated plots from posterior predictive samples to assess the adequacy of the models in describing the observed data.
Although, vocal learning measures were computed for each recording block, here we only examine vocal learning measures computed during the last audio-recording block, as this block was closest in time to neural tissue collection, and FoxP2 expression levels more accurately reflect more recent vocal behavior [11]. We extracted this vocal data for each of the 12 young adult and 12 older adult birds for which we had measured FoxP2 MMSt/VSP expression, the key measure that has been linked to persistent vocal learning ability [13, 14]. Four of the older adults, however, had produced fewer than 6 contact calls during the last audio-recording block (2 birds produced 4 calls, and 2 birds did not call), failing to meet our minimum threshold for accurate measurements and comparisons of acoustic space and thus could not be included in this analysis, leaving a sample size of 8 older adults for analyses of call learning.
Add Comment