Seasonal changes of perineuronal nets and song learning in adult canaries (Serinus canaria)

Abstract

Songbirds learn their song during a sensitive period of development associated with enhanced neural plasticity. In addition, in open-ended learners such as canaries, a sensitive period for sensorimotor vocal learning reopens each year in the fall and leads to song modifications between successive breeding seasons. The variability observed in song production across seasons in adult canaries correlates with seasonal fluctuations of testosterone concentrations and with morphological changes in nuclei of the song control system (SCS). The sensitive periods for song learning during ontogeny and then again in adulthood could be controlled by the development of perineuronal nets (PNN) around parvalbumin-expressing interneurones (PV) which limits learning-induced neuroplasticity. However, this relationship has never been investigated in the context of adult vocal learning in adult songbirds. Here we explored PNN and PV expression in the SCS of adult male Fife Fancy canaries in relation to the seasonal variations of their singing behaviour. We found a clear pattern of seasonal variation in testosterone concentrations and song production. Furthermore, PNN expression was significantly higher in two specific song control nuclei, the robust nucleus of the arcopallium (RA) and the Area X of the basal ganglia, during the breeding season and during the later stages of sensorimotor song development compared to birds in an earlier stage of sensorimotor development during the fall. These data provide the first evidence that changes in PNN expression could represent a mechanism regulating the closing-reopening of sensitive periods for vocal learning across seasons in adult songbirds.

1. Introduction

Passerine songbirds from the oscine suborder are the largest known animal group in which juveniles acquire vocalisations through social learning [1,2]. Indeed, although birds in this group have a (genetic) predisposition to produce specific vocalisations, interactions with an adult tutor during a sensitive period of development are necessary to develop a normal song as an adult [3,4]. The process of vocal learning in songbirds can be divided into two stages: a sensory learning phase, during which the tutor song is memorized, and sensorimotor phase, during which the bird progressively learns to produce species-specific songs by matching the memorized template. During sensorimotor learning, songbirds start singing “subsongs” that are unstructured vocalisations in which song elements are poorly identifiable. Then follows a period during which young birds produce “plastic songs” that are more structured and contain identifiable elements, but these elements are still variable from one rendition to the next [4-6]. This phase of sensorimotor learning continues until the song is well structured and produced in a stable manner with stereotyped song elements; this is the adult “crystallised song” [7]. As with most behaviours that rely on learning during a sensitive period, vocal learning is associated with a period of enhanced neural plasticity that generally terminates at the end of song development. In songbirds, this learning-associated neuroplasticity especially plasticity related to the sensorimotor phase [8] mainly concerns a set of interconnected brain nuclei that are involved in song learning and production and is called the song control system (SCS). The SCS contains two main pathways. The first pathway directly connects HVC (used as a proper name) to RA (robust nucleus of the arcopallium) and then to the motoneurones innervating the syrinx and is directly controlling song production. The second pathway connects HVC to RA through a more indirect route including, among others, Area X of the basal ganglia and is involved in song learning and the control of song variability [9-12].

In some songbird species, vocal learning is limited to sensitive periods of ontogenesis. Their ability to learn new vocalisations or modify their song during adulthood is very limited (close-ended learners) [2]. Other songbird species have the unique ability to modify their vocalisations and learn new songs as adults (open-ended learners) and they experience each year some sort of re-opening of the sensitive periods for vocal learning [7]. The extent to which seasonal songbirds are able to learn new songs during adulthood nevertheless varies widely between species [2]. Some species, such as European starlings (Sturnus vulgaris), are fully able to memorize new song motives as adults and incorporate them into their repertoire so that this repertoire increases with age [13-15]. Other species, such as canaries (Serinus canaria), modify their repertoire between each breeding season, but it is not known whether this is based on the acquisition of new motives during adulthood or via incorporation each year of a different subset of syllables that were all learned during ontogeny. In both cases, changes in adult singing behaviour are made possible because there are periods during the annual cycle of renewed brain plasticity that recapitulate to some extend the plasticity observed during ontogeny.

As singing behaviour is mainly used by males to attract females and to compete with other males, song is considered to be a reproductive behaviour. Since reproduction is limited to the breeding season in seasonal songbirds, adult song learning and production are regulated by seasonal changes in the environment [7,16]. The main factor driving these adult seasonal changes is day length, which regulates the annual growth and regression of the gonads along with seasonal fluctuations of the plasma testosterone concentrations [17,18], which are associated with seasonal plasticity in the SCS [19,20]. In open-ended learner species such as canaries, birds enter in a new period of sensorimotor learning and sing again a plastic song during the fall and early winter; the song becomes more variable and shorter in duration [20-22]. Later, the progressive photostimulation and resulting increase in plasma testosterone will crystallise the song again [23].

There is consequently a change in the type of syllables used in singing (repertoire composition) between the non-breeding and the breeding season [22,24-26]. These seasonal changes in song are associated in some species with an extensive adult plasticity in the SCS [19,27-29]. In seasonally breeding birds of the temperate zone, the volume of HVC, RA and Area X are larger during the breeding season than outside this season [23] This is due namely to an increased neurogenesis of RA-projecting neurones in HVC [30], to trophic effects of this nucleus on RA and Area X through the axons of projection neurones [31] and an increased neuronal volume and spacing in RA [32].

The development of perineuronal nets (PNN) could be an important mechanism controlling the closing of the sensitive period for learning [33,34]. PNN are aggregations of chondroitin sulfate proteoglycans, tenascin-R and hyaluronic acid that create a scaffold around interneurones often expressing the parvalbumin (PV) protein. It was for example established that, in the mammalian visual system, PNN development is closing the sensitive period for visual learning by limiting synaptogenesis [35]. In addition, PNN have been implicated in the stabilization of memories. For example, they develop in parallel with the acquisition of a fear conditioned response and their degradation following stereotaxic injection of chondroitinase ABC into the amygdala specifically makes subsequently acquired memories susceptible to erasure [36]). Similarly, removal of PNN in the secondary visual cortex by injection of chondroitinase ABC impairs the recall of remote but not recent visual fear memories [37] More recently, it was shown that PNN are found in higher numbers in the SCS of adult male zebra finches compared to juvenile males that are still in the song learning process [38,39]. Male zebra finches also have more PNN in their SCS than females that do not sing [40,41]. Moreover, isolation from tutors during development, which delays song learning, reduces the numbers of PNN present around PV-expressing neurones in adult zebra finches [38], suggesting that the development of PNN around PV-interneurones closely correlates with song learning and crystallisation. We therefore hypothesized that PNN expression might relate to the seasonal changes in song plasticity and crystallisation in adult open-ended learning species.

In contrast, PV cells develop early during ontogeny in both zebra finches [38,39] and canaries [42] and they may mark, as in mammalian systems, the onset of experience-dependent brain plasticity [43,38]. In avian species, PV expression in song control nuclei is up-regulated in vocal learners as compared to non-vocal learners [44]. Both PNN and PV neurones thus seem to be closely associated with brain plasticity related to experience-dependent learning and we therefore considered the question as to whether one or both types of structures might vary during the annual cycle of adult canaries in parallel with the different stages of their singing behaviour.

In the present experiments we investigated the relationships between seasonal changes in PNN, PV and vocal behaviour. We used canaries as the model species for this investigation, because we had demonstrated already that testosterone increases the number of PNN in the SCS of adult canaries in parallel with song crystallisation [45] and that, in juvenile male canaries, PNN develop around PV-interneurons during the winter in parallel with the process of song crystallisation [42]. Together these data suggest that PNN could be an important mechanism that regulates the seasonal closing and re-opening of sensitive periods of sensorimotor vocal learning in adult canaries. If this was the case then the number of PNN in the SCS should decrease in the fall when the song becomes plastic to allow synaptic reorganization and song modifications, and PNN density should be higher at the onset of the breeding season when the song is crystallized. To this end, we performed a longitudinal study, spanning over two years and three consecutive breeding seasons, to explore the existence of seasonal cycles of singing behaviour and of testosterone concentrations in adult male canaries of the Fife Fancy breed. Secondly, we quantified the PNN in three key nuclei of the SCS and the singing behaviour at three critical stages of the annual cycle of adult male canaries.

2. Methods

2.1. Animals

All canaries used in this study were provided by the Behavioural Ecology and Ecophysiology Research group at the University of Antwerp, Belgium. These birds were born and raised in the animal facility of this laboratory and maintained in the animal facility of Antwerp until their transfer to Liege. All subjects belonged to the Fife Fancy breed and were maintained in indoor collective aviaries exposed to natural photoperiods before their arrival in the GIGA Neurosciences laboratory. Three groups of birds were used in three separate experiments; one longitudinal study analysed seasonal changes of singing behaviour and reproductive physiology over two years, and two experiments quantified song and PNN expression in the SCS at critical stages of the annual cycle.

All subjects received food (canary seeds) and water ad libitum during all experiments. Anise scented sand, cuttlebones and perches were always present in the cages as enrichments. Additionally, the birds received egg food once a week. During all experiments, the photoperiod of the aviaries was adjusted monthly to the outdoor photoperiod to maintain the birds as much as possible in natural conditions that match seasonal variations in day length. The room temperature was maintained at the same level between 20 °C and 22 °C all year long. During recording sessions that lasted between 1 and 4 weeks, birds were kept singly in a cage (70 × 30 × 35 cm) placed in a sound-attenuated box to allow individual song identification. Observation of the birds suggested that isolation never induced abnormal behaviour, nor signs of distress. No experimental procedure more invasive than a needle puncture to collect a small amount of blood was performed in these studies. All experimental procedures complied with Belgian laws concerning the Protection and Welfare of Animals and the Protection of Experimental Animals, and experimental protocols were approved by the Ethics Committee for the Use of Animals at the University of Liège (Protocol 1739).

2.2. Experimental design

2.2.1. Experiment 1: Longitudinal study of seasonal variations in song and testosterone concentrations

Five adult male canaries (age > 2 years old) arrived in the laboratory in May 2016 and were used as tutors in another study exploring the ontogenesis of PNN in canaries [42]. As tutors, they were maintained in a collective cage (90 × 40 × 35 cm) placed in front of an indoor aviary containing a group of first year males until April 2017. Afterwards, they remained as a group in the same cage except when signs of direct aggression were detected which sometimes occurred during the breeding season. Birds were then temporarily isolated in single cages (45 × 40 × 35 cm) in the same room.

The songs of these 5 males were recorded for a few days during each season from March 2017 until March 2019 at dates and under photoperiods that are described in Table 1. Over the two years, the shortest photoperiod was set at 8L:16D in December and the longest photoperiod was set at 16.5L:7.5D in June. The changes in photoperiod were exactly the same across the two different years. The day before the beginning of the recording sessions, each bird was caught within the collective cage and a small blood sample was collected from the wing vein. Each bird was then immediately transferred to an individual cage within a sound-attenuated chamber to avoid multiple manipulations and limit the associated stress. At the end of the recording session, each bird was returned to the original collective cage. Due to an oversight, for two of the recording sessions (Summer and Fall 2019), the blood was collected at the end of the recording session instead, but this had no obvious effect on the pattern of results observed.

Table 1.

Photoperiods, days of blood collection and days of song analysis for each season over two years. The relative photoperiod indicates if the photoperiod was increasing or decreasing compared to the previous month’s photoperiod at the time point considered.

Seasons	Photoperiod	Relative photoperiod	Blood sample	Song analysis
spring 2017	12L:12D	increasing	March 21 st	March 26th
summer 2017	16.5:7.5D	increasing	June 21 st	June 26th
fall 2017	9L:15D	decreasing	November 15th	November 19th
winter 2018	9L:15D	increasing	January 10th	January 14th
spring 2018	12L:12D	increasing	February 28th	March 4th
summer 2018	16.5:7.5D	increasing	June 25th	June 24th
fall 2018	9L:15D	decreasing	November 14th	November 13rd
winter 2019	9L:15D	increasing	January 15th	January 19th
spring 2019	12L:12D	increasing	March 13rd	March 17th

The recording sessions lasted 3–4 weeks between spring 2017 and spring 2018, and 1–2 weeks from the summer 2018 until the spring 2019. When birds are transferred to a new cage, some may not sing on the following day(s). Therefore, we allowed the birds to habituate to their new environment for 3 days before sound recordings. Plasma testosterone concentrations were correlated with singing behaviour recorded during the fourth or fifth day of recording after blood collection, or during the day before blood collection when blood was collected at the end of the recording session.

One bird died from natural cause before the end of the experiment and was consequently excluded from the song analyses, so that the full longitudinal study of singing behaviour consisted at the end of the study of 4 adult males.

2.2.2. Experiment 2: Seasonal variation of song and PNN

The next experiment quantified PNN in parallel with singing behaviour in two batches of 12 and 10 adult male canaries that arrived in the laboratory in early October and early November 2017 respectively. Before arrival, all birds were kept indoors under a 12L:12D photoperiod and upon arrival the photoperiod was decreased to 10L:14D. All birds were born in April 2016. They were maintained in 4 collective cages (90 × 40 × 35 cm) at a density of 5–7 birds/cage from arrival until the end of the experiment and kept in the same room as birds from experiment 1 under the photoperiod described in the previous section that mimicked natural changes in day length. All birds were distributed in 3 experimental groups of 7–8 subjects to be used at three different time points (fall, winter and spring) based on their housing cage and their arrival date (October or November), so that each group contained 1–2 birds from each cage and 3–4 birds that had arrived on the same day in our laboratory.

At the specified time point, a blood sample was obtained from each bird before it was transferred in an individual cage placed in a sound-attenuated chamber used for sound recordings. The first group (fall, n = 8) was recorded from November 16^th until December 12^th under a 9L:15D photoperiod that was decreasing. The second group (winter, n = 7) was recorded from January 11^th until February 6^th under a 9L:15D photoperiod that was increasing. Finally, the third group (spring, n = 7) was recorded from March 1^st until March 27^th under a 12L:12D photoperiod that was increasing. On the day following the end of the recording session (on December 13, February 7 and March 28 respectively), a second blood sample was taken from the wing vein of each bird and its brain was collected to study PNN expression in the SCS. One bird died naturally before the recordings of the third group resulting in a total sample of 21 adult male canaries.

2.2.3. Experiment 3: Variation of song and PNN between fall and spring

A replicate of experiment 2 was performed because results suggested that song learning had already progressed towards crystallisation in the fall group of the first experiment so that song no longer reflected the plastic stage we had anticipated. The replica was performed with 20 males that arrived in the laboratory either in September 2018 (n = 10) or in March 2019 (n = 10). One half of the birds from each batch were born in April 2015 and the other half in April 2016. All birds were again maintained in collective cages (90 × 40 × 35 cm) of 5–7 birds/cage from arrival until the end of the experiment.

The first group (fall, n = 10) arrived in the laboratory on September 18^th and had been maintained before in Antwerp in an indoor aviary under a natural photoperiod that was set at 13L:11D. Upon arrival, the photoperiod was set at 12L:12D and song recordings were started two days later. The second group (spring, n = 10) was maintained in Antwerp in a photoperiod of 10L:14D before being transferred to our laboratory on March 12^th, where the photoperiod was set at 12L:12D, and song recordings were again started two days later. Both groups were recorded for 7 days. A blood sample was collected before the recordings and on the day following the recording sessions. Brains were then collected (on September 27 or March 27 respectively) for the analysis of PNN expression in the SCS.

When singing behaviour was analysed at the end of the recording session of the fall group, we realized that 5 birds were only producing plastic songs, whereas the 5 other birds were already engaged in the song crystallisation process. We therefore decided to divide this group into two subgroups: fall-plastic song (PS, n = 5) and fall-crystallized songs (CS, n = 5). Data were then analysed separately comparing the full fall group to the spring group, the fall-PS group to the fall-CS group and the two fall groups to the spring group. These analyses provided similar conclusions for all song and brain measurements. All song characteristics of subjects from the two fall subgroups were clearly different and there was no overlap between data from these two subgroups. We only report here the results of analyses simultaneously considering the three groups of birds.

2.3. Testosterone analysis

2.3.1. Blood collection

Blood (50–100 μl) was collected from the wing vein of all subjects at specific time points as described in the “Experimental design” section. Blood collection was always performed within 3 min after catching the birds in their collective cage during the morning within 1.5 h after lights on. Blood was collected into Na-heparinized micropipettes and any further blood flow was stopped by pressing cotton on the punctured vein after a maximum of 150 μl was collected. Blood was centrifuged at 9000 g for 9 min and the supernatant plasma was collected and stored at −80 °C until assayed.

2.3.2. Testosterone enzyme immunoassay

As previously described [46,47], 10 μl of plasma from each sample was diluted in 150 μl of ultra-pure water. A few additional samples were spiked with 20,000 CPM of tritiated-testosterone (Perkin-Elmer) to estimate the percentage of hormone recovery during the extraction process. All samples were extracted twice with 2 ml of dichloromethane. The organic phase was eluted into clean tubes, dried with nitrogen gas and stored at −20 °C until further use. Average recovery rate was 71.7%.

Extracted samples were re-suspended in 400 μl Enzyme Immunoassay (EIA) buffer by vortexing for 30 s and shaking for 120 min at 1350 rpm. A 50 μl aliquot of the re-suspended samples was used for assay. Samples were assayed in triplicates for testosterone concentrations using a Cayman Chemicals testosterone EIA kit following manufacturer’s instructions within 96-well plates. The 42 samples of the second experiment were assayed in 2 plates on the same day. The intra-assay variation ranged between 4.4 and 5.0% (mean = 4.7%) and the inter-assay variation for all points in the three standard curves ranged from 1.9% to 16.6% (mean = 7.8%). The samples of the third experiment and of the longitudinal study were assayed together in 3 plates using 60 samples on the same day. All birds from the third experiment were assayed in the same plate. The intra-assay variation ranged between 2.3 and 3.1% (mean = 2.7%) and the inter-assay variation for all points in the three standard curves ranged from 1.0% to 9.8% (mean = 5.4%). The minimum and maximum detection limits of the EIA, as determined by the range of concentrations detected within the standard curves, were 0.11-0.17 pg/well and 19.97–24.21 pg/well respectively. Concentrations in a few samples from the third experiment and from the longitudinal study were below the detection limits. These samples were assigned a value extrapolated between the zero point and the first point of the curve since they belonged to birds that were part of the groups displaying the lowest testosterone levels.

2.4. Song analysis

All song analyses were performed as described previously [42]. All songs had been recorded while birds were individually housed in sound-attenuated chambers. Records were collected every day during 2 consecutive hours starting immediately after lights on. Sounds from all chambers were acquired simultaneously via custom-made microphones (microphone from Projects Unlimited/Audio Products Division, amplifier from Maxim Integrated) through an Allen & Heath ICE-16 multi-channel recorder connected to a computer. The sound files were 16-bit acquired at a frequency of 44,100 Hz which translates to a frequency range of 0–22,050 Hz and saved as 1 min. wav files sequences using Raven Pro v1.4 software (Bioacoustics Research Program 2011; Raven Pro: Interactive Sound Analysis Software, Version 1.4, Ithaca, NY: The Cornell Lab of Ornithology).

The sound analyses were performed with the same software. The daily 2 h sound recordings were first reassembled for each channel corresponding to each experimental bird. Spectrogram views of these files were constructed with a direct Fourier transform (DFT) size of 256 samples (172 Hz per sample) and a temporal frame overlap of 50% with a hop size of 128 samples. These parameters were automatically determined by the software to provide an optimized frequency/time resolution for the spectrographic analysis and were identical for all recordings analysed in the two studies.

The first hour of recording obtained two days before brain collection was analysed in detail for each bird in the two experiments on PNN changes. In the longitudinal study, we analysed in detail the first hour of recording obtained four or five days after the blood collection. One hour of recording was sufficient to obtain at least 240 s of songs for each bird. As previously reported [48], this time is necessary and sufficient to identify the complete repertoire of the canary.

Vocalizations were considered as songs if they lasted a minimum of 0.5 s separated by a gap of minimum 0.5 s. Previous studies on canaries used a minimum song duration of 1 s [22,49,50]. However, these criteria cannot be applied to plastic songs produced during early sensorimotor learning since they generally last less than 1 s. The minimum sound duration to be considered as a song was therefore reduced and calls, vocalizations containing a single element, were manually excluded from the analyses. A vocalisation had to contain multiple spectrographic elements (different syllables) to be considered as songs (see examples in Fig. 1). All songs corresponding to these criteria were manually selected throughout the entire one hour-long recording and counted (number of songs per hour or song rate). The duration of each song was provided by the software and was averaged for each bird. The duration of all songs was summed (to obtain the total duration of singing in seconds during one hour; not analysed as such) and divided by 3600 to obtain the percentage of time spent singing (% time singing). Each individual song was also processed through the automated sound analysis of the Raven software and results were averaged for each bird. These additional automated measurements characterized the song “loudness” (average and maximum power (dB), RMS (root mean square) and maximum amplitude (U)), the energy distribution across frequencies (5%, 1^st quartile, centre, 3^rd quartile and 95% frequencies (Hz)), the bandwidth (Hz) of this energy distribution between the 1^st and 3^rd quartile (IQR bandwidth) and between 5% and 95% (90% bandwidth) and the average entropy (bits), a measure of disorder within the energy distribution; a lower song entropy is associated with a higher stereotypy, a characteristic of the adult crystallized song.

Fig. 1.

Representative sonograms illustrating a short (A) and a longer (B) subsong and the fully developed song (C) that were recorded across the annual cycle in male Fife Fancy canaries.

Additionally we attributed a qualitative developmental score ranging from 1 to 5 to each selected song to characterize the level of song development from subsong (1), through advanced subsong/poorly organized plastic song (2), plastic song (3), advanced plastic song (4) to crystallized song (5). Briefly, the score was assigned following spectrogram inspection based on multiple qualitative criteria including the possibility of identifying individual syllables, the presence of a song structure typical of the canary song including different phrases that are repetitions of a same syllable, the sharp representation of syllables in the sonograms indicating the presence of crystallized song syllables and the general accuracy of syllable repetition in terms of frequency and time (see detailed criteria in [42]). For each bird, the score of all songs was averaged to obtain the mean developmental score. A similar measure of song development was previously used to study the effect of testosterone on song development in juvenile song sparrows [51]. Here, we used a similar development scale, but the criteria corresponding to each grade were adapted to the specificity of the canary song. This scale was originally developed for the study of the ontogeny of song in male canaries [42] but it should be noted that vocalizations never reverted to subsong during the annual cycle of adult males; 2 was the lowest score ever observed.

2.5. Immunohistochemistry

The following procedures only concern experiments 2 and 3 and equally apply to both experiments.

2.5.1. Tissue collection

After the recording sessions, subjects were weighed, their cloacal protuberance was measured, a blood sample was taken from the wing vein and then birds were anaesthetized with Nembutal™ (0.04 ml at 0.6 mg/ml of pentobarbital molecule). Once reflexes had stopped, birds were intracardially perfused with phosphate-buffered saline (PBS) to remove blood, immediately followed by 4% paraformaldehyde in PBS (PFA) to fix the brain. After perfusion the brain was immediately extracted from the skull, the 2 testes and the syrinx were dissected and weighed. The mean weight of the 2 testes of each bird is used in all analyses presented here. Brains were post-fixed during 24 h in 15 ml PFA and on the following day, they were transferred to 15 ml of 30% sucrose solution for cryoprotection. Once brains had sunk to the bottom of the vial, they were frozen on dry ice and stored at −80 °C until used. Brains were cut coronally in 30 μm-thick sections on a Leica CM 3050S cryostat into 4 series of 4 wells that were stored in anti-freeze solution at −20 °C.

2.5.2. Immunostaining

Half a series (2 non-adjacent wells; 240 μm between sections) was double-labelled for parvalbumin and chondroitin sulfate, one of the main components of the perineuronal nets, following a previously described protocol [38,39,41,52]. Each experiment was stained separately. Briefly, sections were blocked in 5% Normal Goat Serum (NGS) diluted in Tris-buffered Saline (TBS) with 0.1% Triton-X-100 (TBST) for 30 min. They were incubated overnight at 4 °C in a mixture of 2 primary antibodies diluted in TBST: a mouse monoclonal anti-chondroitin sulfate antibody (CS-56, 1:500; C8035, Sigma Aldrich) specific for the glycosaminoglycan portion of the chondroitin sulfate proteoglycans that are the main components of the PNN and a polyclonal rabbit anti-parvalbumin antibody (1:1000; ab11427, Abcam). The third experiment used a concentration of 1:1000 for the chondroitin sulfate immunostaining because the antibody lot (lot 026M4834-V) used for this batch was different and optimization tests indicated that this lower concentration was optimal. Sections were then incubated at room temperature in a mixture of secondary antibodies diluted in TBST. A goat anti-mouse IgG coupled with Alexa 488 (green, 1:100, Invitrogen) was used to visualize PNN staining and a goat anti-rabbit IgG coupled with Alexa 546 (red, 1:200, Invitrogen) was used to visualize PV cells. Finally, sections were mounted on slides using TBS with gelatine and coverslipped with Vectashield containing DAPI (H-1500, Vector laboratories) that was used to confirm that PNN that were not surrounding PV-positive cells were localized around a cell nucleus, which was almost invariably the case. On a few occasions, fluorescent rings were observed at the borders of small holes in the tissue, but the absence of DAPI in the centre was used as a criterion to discard these structures.

2.6. Nucleus volume quantification

The dense parvalbumin and chondroitin sulfate pattern of stainings was used to quantify the volume of HVC, RA and Area X. Photomicrographs of all stained sections containing these nuclei were acquired at 5X magnification and their volume was quantified as previously described [52]. Briefly, the area of the Regions of Interest (ROIs in mm²) within each section was measured using the Image J software (NIH, https://imagej.nih/ij). The volume of each ROI was then estimated by multiplying the measured surface in each section by the distance between sections (240 μm) and then summing the results for all the sections. Finally, the mean of the volumes in the left and right hemispheres was calculated.

2.7. PNN & PV quantification

The numbers of PV-positive cells (PV), of cells surrounded by PNN (PNN) and of PV-positive cells surrounded by PNN (PV + PNN) were counted in the 3 song control nuclei HVC, RA and Area X. The boundaries of the ROIs were determined based on the bright PV and/or PNN staining. Two photomicrographs were acquired on each brain side in 2 sections equally spaced in the rostro-caudal axis for each ROI. These photomicrographs were obtained with a Leica fluorescence microscope with a 40X objective and fixed settings. Each photomicrograph was entirely contained within the ROIs so that quantifying the entire image always sampled a similar area. The numbers of PV, PNN and PV + PNN were consequently counted in the entire photomicrographs with the Image J software (NIH, https://imagej.nih/ij) as previously described [52].

Briefly, for each ROI, the mean of data collected in the left and right side of each section was calculated and averaged to obtain the number of stained structures per counted surface in a given ROI. These numbers were converted in densities/mm² and also used to compute the percentage of PV surrounded by PNN (%PVwithPNN) and the percentage of PNN surrounding PV (%PNNwithPV). Finally the volume of each nucleus of each bird was used to estimate the total number of counted objects (PV, PNN and PV + PNN) in the entire nuclei using the following formula: (number of counted object)*(nuclei volume/(counted area*section thickness)). All quantifications were made with the experimenter being blind to treatment.

2.8. Statistics

One-way Repeated Measures ANOVAs were used to explore the variation in testosterone and singing behaviour across seasons over two years in the first experiment. Significant effects were further explored with post hoc analyses using Tukey multiple comparison tests to identify the seasons when differences appeared. In the two other experiments, because all the data were not normally distributed, non-parametric rank tests were used for analyses. For all measures, one-way Kruskal-Wallis ANOVAs were used to explore the effect of seasons on physiological, behavioural and brain measurements. When appropriate, Dunn’s multiple comparisons tests were used to further explore the differences between groups. The critical significance level was set at p < 0.05 for all tests. The effect sizes were calculated using partial eta squared for the Repeated Measures (RM) ANOVAs and H eta squared for Kruskal-Wallis ANOVAs (as described in [53]).

3. Results

3.1. Experiment 1: Seasonal variations of song and testosterone levels, a longitudinal study

3.1.1. Plasma testosterone peaks in early summer

Plasma testosterone concentrations changed significantly across seasons over two years (RM ANOVA: F_8,24 = 9,43, p < 0.0001, η²_p = 0.76). Post hoc analyses showed that testosterone concentrations were significantly higher during summer (p < 0.05 for all comparisons except for the summer 2017 vs. spring 2018). The largest decrease was systematically observed between the summer and fall, and the largest increase between spring and summer (see Fig. 2).

Fig. 2.

Plasma testosterone concentrations (Means ± SEM in ng/ml) across seasons over two years in four adult male Fife Fancy canaries. RM ANOVA indicated a main effect of time. Post hoc analysis results are indicated as follows: *=p < 0.05 compared to all other data points except spring 2018.

3.1.2. The song becomes plastic every fall

The song rate was not significantly affected by seasons (Fig. 3A; RM ANOVA: F_8,24 = 1.70, p = 0.1486, η²_p = 0.36, see Fig. 2A), but several song characteristics were: This was the case for song duration (Fig. 3B; RM ANOVA: F_8,24 = 4.16, p = 0.0031, η²_p = 0.58), the percentage of time spent singing, which is indicative of the motivation to sing (Fig. 3C; RM ANOVA: F_8,24 = 3.37, p = 0.0099, η²_p = 0.53), the song RMS amplitude (Fig. 3D; RM ANOVA: F_8,24 = 8.18, p < 0.0001, η²_p = 0.73), the song stereotypy measured by the entropy (Fig. 3E; RM ANOVA: F_8,24 = 5.64, p = 0.0004, η²_p = 0.65) and the song developmental score (Fig. 3F; RM ANOVA: F_8,24 = 13.99, p < 0.0001, η²_p = 0.82). The post hoc analyses revealed that most measures of song quality were higher in the summers than at other seasons in particular during the fall (see detail of statistics in Fig. 3). Conversely, song entropy was higher in the fall 2018 compared to all other seasons, except the winter 2019 (see Fig. 3E). Together these data thus indicate that song becomes more plastic and less stereotyped each year during the fall as compared to the spring seasons

Fig. 3.

Seasonal changes in singing behaviour (Means ± SEM) across two years in four adult male Fife Fancy canaries as measured by the song rate (A), song duration (B), percentage of time spent singing (C), song RMS amplitude (D), song average entropy (E) and song developmental score (F). The effect of time revealed by the repeated measure ANOVA is indicated in the inserts (**=p < 0.01; ***=p < 0.001). Results of the post hoc analyses are indicated as follows: * or # or $=p < 0.05, ** or ## or $$=p < 0.01, *** or ### or $$$=p < 0.001 versus fall 2018, or fall 2017 or winter 2019 respectively.

3.2. Experiment 2: Seasonal variation of song and PNN

3.2.1. Plasma testosterone and morphological measures

Testosterone concentrations before and after recordings were positively and significantly correlated (Spearman rank correlation: r_s = 0.84, N = 21 p < 0.0001). Therefore, only data obtained on the day of brain collection are presented here. As expected, testosterone concentrations increased significantly from fall to spring (Kruskal-Wallis test: H₂ = 16.24, p = 0.0003, η²_H = 0.68) and post hoc tests indicated that it was higher during the winter and spring than in the fall (Fig. 4A-left). A similar increase was observed in the mean testis weight (Kruskal-Wallis test: H₂ = 15.96, p = 0.0003, η²_H = 0.66), but only the spring group had heavier testes than the fall group. Testis weight in the winter group was intermediate between the two groups and did not significantly differ from the weight at the two other time points (Fig. 4B-left). Somewhat surprisingly, the cloacal protuberance area, an indirect measure of testosterone action, was not significantly affected by seasons (Kruskal-Wallis test: H₂ = 3.78, p = 0.1509, η²_H = −0.01, Fig. 4C-left) and this was true also for the syrinx weight (Kruskal-Wallis test: H₂ = 0.83, p = 0.6593, η²_H = −0.18, Fig. 4D-left).

Fig. 4.

Physiological measures on the day of brain collection for experiments 2 (left part of each panel) and 3 (right part of each panel) as represented by plasma testosterone concentrations (A), testis weight (B), cloacal protuberance area (C) and syrinx weight (D). Means ± SEM of all data are represented together with individual values (dots). Significant Kruskal-Wallis ANOVAs (see text for details) were followed by post hoc tests and their results are indicated as follows: *p < 0.05, **p < 0.01, ***p < 0.001 compared to the fall 2017 group (in Experiment. 2) or the fall-plastic song (PS) group (in Experiment 3), # p < 0.05 compared to the fall-crystallized song (CS) group (in Experiment 3).

Despite the large seasonal changes in testosterone concentrations, there was no systematic variation in the volume of HVC (Kruskal-Wallis test: H₂ = 2.14, p = 0.3437, η²_H = −0.10, Fig. 5A-left;) and Area X (Kruskal-Wallis test: H₂ = 0.23, p = 0.8924, η²_H = −0.22, Fig. 4C-left). In contrast, RA volume varied across seasons (Kruskal-Wallis test: H₂ = 7.72, p = 0.0211, η²_H = 0.21) and was significantly larger during the winter than during the fall, but this difference was no longer significant in the spring, possibly because the sample size was smaller in the spring than in the winter. RA volumes were however essentially identical at these two time points (Fig. 5B-left).

Fig. 5.

Volume of the three song control nuclei HVC, RA and Area X (A–C), and numbers of PNN (D–F), of PV (G–I), of PV + PNN (J–L) and percentage of PV with PNN (M–O) in these three nuclei for experiments 2 (left part of each panel) and 3 (right part of each panel). Means ± SEM of all data are represented together with individual values (dots). Abbreviations: PS, plastic song; CS, crystallized song. Significant Kruskal-Wallis ANOVA (see text for details) were followed by post hoc tests and their results are indicated as follows: (*)p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001 compared to fall 2017 (Experiment 2) or fall-PS (Experiment 3).

3.2.2. PV and PNN in the song control nuclei

No effect of season could be detected in HVC on the numbers of PNN (Kruskal-Wallis test: H₂ = 3.84, p = 0.1417, η²_H= −0.01, Fig. 5D-left), PV (Kruskal-Wallis test: H₂ = 5.49, p = 0.0644, η²_H = 0.08, Fig. 5G-left), PV + PNN (Kruskal-Wallis test: H₂ = 3.94, p = 0.1395, η²_H = 0.00, Fig. 5J-left) and percentage of PV with PNN (Kruskal-Wallis test: H₂ = 1.03, p = 0.5986, η²_H = −0.17, Fig. 5M-left). However, the overall density of PV-expressing neurones varied between seasons, even if the analysis failed to identify differences between specific groups (see Table 2-top part for details). Similarly, in Area X there was also no significant seasonal variation in the total numbers of PNN, PV and PV + PNN as well as the percentage of PV with PNN (Kruskal-Wallis test: PNN: H₂ = 4.61, p = 0.0997, η²_H = 0.04, Fig. 5F-left; PV: H₂ = 0.30, p = 0.8604, η²_H = −0.225, Fig. 5I-left; PV + PNN: H₂ = 3.46, p = 0.1775, η²_H = −0.03, Fig. 5L-left; % PV with PNN: H₂ = 4.85, p = 0.0883, η²_H = −0.11, Fig. 5O-left). There was however an effect of season on the percentage of PNN localized around PV-expressing neurones with significantly fewer PNN located around PV cells in the spring compared to fall and winter (see Table 2 for details).

Table 2.

Quantitative analyses of PV and PNN densities (numbers per mm²) and ratios during experiments 2 and 3. The table shows the means ± SEM of PV, PNN and PV + PNN densities, percentages of PV surrounded by PNN and percentages of PNN located around PV in the three SCS nuclei. Abbreviations: PS, plastic song; CS, crystallized song. Statistical results of the Kruskal-Wallis ANOVAs for experiment 2 (top) and experiment 3 (bottom) are also presented. Results of significant post hoc tests by Dunn’s multiple comparison test are indicated by * when significantly different from fall (Exp. 2) or fall-SS (Exp. 3) or ^# when significantly different from winter (Exp. 2). Effects sizes were calculated as follows η²_H = (H-(k+1))/N-k and are indicated for each Kruskal-Wallis test. Significance levels are indicated as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Brain	Fall ‘17	Winter ‘18	Spring ‘18	Kruskal-Wallis test
HVC
PNN density (/mm2)	37.7 ± 8.6	56.4 ± 7.8	57.1 ± 12.8	H 2 = 2.35, p = 0.3086, η2H = −0.09
PV density (/mm2)	444.8 ± 46.7	690.0 ± 64.1	693.7 ± 57.7	H 2 = 7.60, p = 0.0224, η2H = 0.20
PV + PNN density (/mm2)	26.8 ± 7.3	34.8 ± 4.0	48.4 ± 8.0	H 2 = 3.89, p = 0.1437, η2H = −0.01
% PNN with PV	66.5 ± 13.0	65.3 ± 7.7	90.6 ± 5.4	H 2 = 4.85, p = 0.0883, η2H = 0.05
RA
PNN density (/mm2)	38.5 ± 11.2	60.5 ± 9.6	50.3 ± 14.4	H 2 = 3.80, p = 0.1496, η2H = −0.01
PV density (/mm2)	590.7 ± 66.3	420.44 ± 38.6	376.4 ± 31.3 *	H 2 = 7.15, p = 0.0280, η2H = 0.17
PV + PNN density (/mm2)	27.6 ± 8.5	53.9 ± 7.5	43.5 ± 10.1	H 2 = 5.92, p = 0.0461, η2H = 0.11
% PNN with PV	71.9 ± 8.1	91.2 ± 4.6	91.0 ± 4.6	H 2 = 4.22, p = 0.1215, η2H = 0.01
Area X
PNN density (/mm2)	46.4 ± 7.7	63.9 ± 11.0	73.5 ± 7.0	H 2 = 4.32, p = 0.1152, η2H = 0.02
PV density (/mm2)	262.9 ± 21.2	252.1 ± 70.1	263.2 ± 55.7	H 2 = 0.32, p = 0.8514, η2H = −0.22
PV + PNN density (/mm2)	37.3 ± 6.6	54.7 ± 10.1	44.5 ± 3.6	H 2 = 2.03, p = 0.3782, η2H = −0.12
% PNN with PV	83.1 ± 3.9	86.0 ± 3.9	62.6 ± 5.6*#	H 2 = 8.20, p = 0.0166, η2H = 0.25
Brain	Fall-PS ‘18	Fall-CS ‘18	Spring ‘19	Kruskal-Wallis test
HVC
PNN density (/mm2)	67.3 ± 15.1	81.3 ± 13.3	71.4 ± 8.4	H 2 = 0.32, p = 0.8507, η2H = −0.22
PV density (/mm2)	327.4 ± 39.5	379.6 ± 44.1	459.2 ± 34.5	H 2 = 4.89, p = 0.0865, η2H = 0.05
PV + PNN density (/mm2)	67.3 ± 15.1	77.8 ± 15.5	70.8 ± 8.2	H 2 = 0.21, p = 0.9054, η2H = −0.22
% PNN with PV	100 ± 0.0	93.3 ± 6.7	99.4 ± 0.6	H 2 = 1.17, p = 0.5583, η2H = −0.17
RA
PNN density (/mm2)	33.7 ± 7.7	77.8 ± 2.3(*)	79.0 ± 11.3*	H 2 = 8.11, p = 0.0174, η2H = 0.24
PV density (/mm2)	679.2 ± 37.8	506.2 ± 44.5	431.9 ± 27.3**	H 2 = 10.13, p = 0.0063, η2H = 0.36
PV + PNN density (/mm2)	33.7 ± 7.7	77.8 ± 2.3*	73.7 ± 11.7	H 2 = 7.37, p = 0.0177, η2H = 0.20
% PNN with PV	95.0 ± 5.0	100 ± 0.0	92.5 ± 3.1	H 2 = 3.34, p = 0.1884, η2H = −0.04
Area X
PNN density (/mm2)	52.2 ± 15.0	80.1 ± 10.5	101.6 ± 7.9*	H 2 = 7.89, p = 0.0194, η2H = 0.23
PV density (/mm2)	255.4 ± 11.6	220.6 ± 15.7	222.9 ± 8.6	H 2 = 4.77, p = 0.0922, η2H = 0.05
PV + PNN density (/mm2)	49.9 ± 14.8	65.0 ± 6.7	85.9 ± 7.1	H 2 = 6.15, p = 0.0391, η2H = 0.13
% PNN with PV	96.2 ± 2.3	83.7 ± 6.3	84.8 ± 3.0	H 2 = 5.24, p = 0.0728, η2H = 0.07

In contrast, significant changes were detected in RA. There was a significant effect of season on the number of PNN (Kruskal-Wallis test: H₂ = 6.98, p = 0.0305, η²_H = 0.17, Fig. 5E-left), PV + PNN (Kruskal-Wallis test: H₂ = 7.70, p = 0.0212, η²_H = 0.21, Fig. 5K-left) and percentage of PV with PNN (Kruskal-Wallis test: H₂ = 6.98, p = 0.0305, η²_H = 0.17, Fig. 5N-left), although the number of PV-expressing neurones did not change (Kruskal-Wallis test: H₂ = 1.19, p = 0.5517, η²_H = −0.16, Fig. 5H-left). The post hoc analysis revealed that the significant effects of season resulted from higher values in winter compared to fall (Fig. 5E, K and N-left). There was additionally a significant decrease of PV density in RA, that was significantly lower in spring than in fall (see Table 2-top). There were no further significant seasonal effects in the other density measures (Table 2).

3.2.3. Song quantity and quality

Three birds from the fall group and one bird from the winter group did not sing (singing rate and the percentage of time spent singing = 0), so that it was impossible to quantify their song quality. This reduced the sample size for these measures to 5 in fall, 6 in winter and 6 in spring. No seasonal variation was detected in song rate (Kruskal-Wallis test: H₂ = 0.75, p = 0.6873, η²_H = −0.18, Fig. 6A-left), song duration (Kruskal-Wallis test: H₂ = 4.40, p = 0.1093, η²_H = 0.03, Fig. 6B-left), percentage of time spent singing (Kruskal-Wallis test: H₂ = 3.10, p = 0.2125, η²_H = −0.05, Fig. 6C-left) and song RMS amplitude (Kruskal-Wallis test: H₂ = 4.75, p = 0.0894, η²_H = 0.05, Fig. 6D-left). There was however a seasonal effect on the average song entropy (Kruskal-Wallis test: H₂ = 9.52, p = 0.0030, η²_H = 0.39, Fig. 6E-left) that was significantly decreased in the winter and spring compared to fall and on the song developmental score (Kruskal-Wallis test: H₂ = 9.35, p = 0.0035, η²_H = 0.38, Fig. 6F-left) that was significantly higher in spring compared to fall. There was also no seasonal variation in the power distribution across frequencies, bandwidths or other measures of the song loudness (see details in Table 3-top).

Fig. 6.

Singing behaviour as measured by the song rate (A), song duration (B), percentage of time spent singing (C), song RMS amplitude (D), song average entropy (E) and song developmental score (F) during experiments 2 (left part of each panel) and 3 (right part of each panel). Abbreviations: PS, plastic song; CS, crystallized song. Means ± SEM of all data are represented together with individual values (dots). Significant Kruskal-Wallis ANOVA (see text for details) were followed by post hoc tests and their results are indicated as follows: (*)p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001 compared to fall 2017 (Experiment 2) or fall-PS (Experiment 3). #p < 0.05 compared to winter 2017 (Experiment 2).

Table 3.

Quantitative analysis of singing behaviour during experiments 2 and 3. The table presents the means ± SEM of power distribution across frequencies (1st quartile, 3rd quartile, 5%, 95% and center frequency), bandwidths of this distribution (inter-quartile range and 90% range) and additional measures of vocalization loudness (average power, maximum power and maximum amplitude). Abbreviations: PS, subsong; CS, crystallized song. Statistical results of the Kruskal-Wallis ANOVAs for experiment 2 (left) and experiment 3 (right) are also presented. Results of significant post hoc tests by Dunn’s multiple comparisons are indicated by * when significantly different from fall (Exp. 2) or fall-SS (Exp. 3). Effects sizes were calculated as follows η²_H = (H-(k+1))/N-k and are indicated for each Kruskal-Wallis test. Significance levels are indicated as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Singing behaviour	Fall ‘17	Winter ‘18	Spring ‘18	Kruskal-Wallis test
1 st quartile frequency (Hz)	3681 ± 275	4169 ± 111	4157 ± 100	H 2 = 2.84, p = 0.2543, η2H = − 0.08
3rd quartile frequency (Hz)	4651 ± 323	5171 ± 134	5048 ± 119	H 2 = 1.75, p = 0.4336, η2H = −0.16
5% frequency (Hz)	2893 ± 276	3356 ± 52	3413 ± 125	H 2 = 4.90, p = 0.0841, η2H = 0.06
95% frequency (Hz)	5314 ± 346	5867 ± 138	5836 ± 100	H 2 = 2.58, p = 0.2930, η2H = −0.10
Center frequency (Hz)	4186 ± 300	4712 ± 121	4570 ± 103	H 2 = 2.82, p = 0.2583, η2H = −0.08
IQR bandwidth (Hz)	969 ± 73	1002 ± 48	890 ± 43	H 2 = 1.74, p = 0.4402, η2H = −0.16
90% bandwidth (Hz)	2422 ± 103	2511 ± 94	2422 ± 71	H 2 = 0.88, p = 0.6640, η2H = −0.22
Average power (dB)	43.4 ± 4.2	50.4 ± 1.6	51.9 ± 1.5	H 2 = 5.09, p = 0.0722, η2H = 0.08
Maximum power (dB)	75.6 ± 4.7	82.4 ± 1.7	83.4 ± 1.7	H 2 = 3.32, p = 0.1967, η2H = −0.05
Maximum amplitude (U)	1470 ± 397	2176 ± 333	2531 ± 460	H 2 = 2.90, p = 0.2493, η2H = −0.08
Singing behavior	Fall-PS ‘18	Fall-CS ‘18	Spring ‘19	Kruskal-Wallis test
1 st quartile frequency (Hz)	3140 ± 67	4090 ± 58*	4146 ± 126**	H 2 = 10.70, p = 0.0048, η2H = 0.42
3rd quartile frequency (Hz)	3870 ± 121	5050 ± 88**	4861 ± 140*	H 2 = 10.52, p = 0.0013, η2H = 0.41
5% frequency (Hz)	2467 ± 247	3388 ± 68(*)	3527 ± 88**	H 2 = 11.21, p = 0.0037, η2H = 0.45
95% frequency (Hz)	4755 ± 192	5739 ± 118*	5621 ± 153*	H 2 = 8.90, p = 0.0117, η2H = 0.31
Center frequency (Hz)	3533 ± 91	4569 ± 54*	4477 ± 142*	H 2 = 9.38, p = 0.0037, η2H = 0.34
IQR bandwidth (Hz)	730 ± 135	960 ± 53	715 ± 37	H 2 = 6.81, p = 0.0332, η2H = 0.18
90% bandwidth (Hz)	2289 ± 363	2351 ± 68	2094 ± 96	H 2 = 1.82, p = 0.4025, η2H = −0.14
Average power (dB)	37.1 ± 2.2	47.1 ± 1.7	51.9 ± 1.3**	H 2 = 11.64, p = 0.0030, η2H = 0.48
Maximum power (dB)	66.4 ± 1.9	78.6 ± 1.6	82.4 ± 1.4 **	H 2 = 12.29, p = 0.0021, η2H = 0.52
Maximum amplitude (U)	494 ± 126	1539 ± 210	2452 ± 280 **	H 2 = 12.47, p = 0.0020, η2H = 0.53

3.3. Experiment 3: Seasonal variation of song and PNN – Early faR vs. Spring

One of the birds from the spring group never sang and was consequently excluded from the analysis of song quality. This experiment was designed to document the period of plastic song during the fall and compare it to the period of fully crystallized song. One group of birds was thus recorded in early fall during the last week of September. However, even at this early time point, there was a substantial amount of individual variability in the degree of song development: some birds were clearly in the plastic song stage of sensorimotor learning, whereas other birds were in the latest stage of sensorimotor learning and their song had already started crystallizing. Therefore, we decided to divide the fall group, based on a qualitative evaluation of sonograms, in one subgroup only singing plastic songs (fall-PS) and a second group singing partly crystallized songs (fall-CS). All birds from the spring group were singing mostly crystallized songs.

3.3.1. Physiology and morphology

As testosterone concentrations before and after the song recordings were highly correlated in the second experiment and one week only separated the two blood samples in this experiment, we only measured and analysed the second blood sample of each bird. Plasma testosterone concentrations were again different between groups (Kruskal-Wallis test: H₂ = 12.32, p = 0.0021, η²_H = 0.49) and significantly higher in spring compared to the two fall subgroups (males singing plastic songs (PS) and or crystallized songs (CS), even if absolute concentrations were lower than in experiment 2 (Fig. 4A-right). There was also a seasonal effect on the testis size (Kruskal-Wallis test: H₂ = 14.43, p = 0.0007, η²_H = 0.61); they were much larger in the spring than in the fall-PS group (Fig. 4B-right). The cloacal protuberance area was in this case affected by season (Kruskal-Wallis test: H₂ = 9.38, p = 0.0092, η²_H = 0.32) and was significantly larger in the spring group compared to the fall-PS group only (Fig. 4C-right). The syrinx weight also changed significantly (Kruskal-Wallis test: H₂ = 10.79, p = 0.0045, η²_H = 0.40) and was larger in the spring and the fall-CS groups than in the fall-PS group (Fig. 4D-right).

These physiological changes were accompanied by significant alterations in the volume of the three song control nuclei (HVC: H₂ = 11.75, p = 0.0004, η²_H = 0.46; RA: H₂ = 11.03, p = 0.0008, η²_H = 0.41; Area X: H₂ = 9.35, p = 0.0040, η²_H = 0.31; Fig.5A-C right). Post hoc analyses revealed that the volume of HVC, RA and Area X was significantly larger in the spring group compared to the fall-PS group only. The fall-CS group displayed intermediate values and was not significantly different from the two other groups for these measures.

3.3.2. Variability in song development in early fall

The detailed song analyses confirmed the large difference in terms of the stage of song development between the two fall subgroups (Fig. 6). The Kruskal-Wallis ANOVA indeed identified significant differences between the three groups for the song duration (H₂ = 10.62, p = 0.0049, η²_H = 0.41), the song average entropy (H₂ = 11.22, p = 0.0006, η²_H = 0.45) and the song developmental score (H₂ = 10.54, p = 0.0014, η²_H = 0.41). In each of these cases, the post hoc analyses confirmed that the spring group and the fall-CS group had significantly different measures compared to the fall-PS group (Fig. 6B, E and F-right). In contrast, the singing rate also differed between groups (H₂ = 10.50, p = 0.0053, η²_H = 0.38), but in this case it was significantly higher in the fall-CS than in the other two groups (Fig. 6A-right). A similar pattern was consequently observed in the percentage of time spent singing (H₂ = 11.36, p = 0.0034, η²_H = 0.43) that was significantly higher in the fall-CS than in the fall-PS group (see Fig. 6C-right), but the difference with the spring group was not significant. The song RMS amplitude also differed between groups (H₂ = 12.16, p = 0.0002, η²_H = 0.51) and was significantly higher in the spring group compared to the fall-PS group only (Fig. 6D-right).

The other measures of song loudness revealed similar patterns (see Table 2-right for details). All three measures were different between groups with the spring group showing larger values than the fall-PS group. Additionally, there was a significant difference between groups for the power distribution across frequencies that showed a displacement of the power towards higher frequencies. This effect was significant for the comparisons of both the spring and the fall-CS groups with the fall-PS group (see Table 2-right for details). Finally, there was also a significant group effect for the inter-quartile range bandwidth, but the post hoc analysis failed to identify specific differences between groups. The 90% bandwidth did not differ between groups (see Table 3-right for details).

3.3.3. Seasonal changes in PNN and PV

In HVC, as observed in the previous experiment, no differences between fall and spring could be detected in the numbers of PNN (Kruskal-Wallis test: H₂ = 3.22, p = 0.2062, η²_H = −0.05), PV + PNN (Kruskal-Wallis test: H₂ = 3.25, p = 0.2040, η²_H = −0.04) and the percentage of PV with PNN (Kruskal-Wallis test: H₂ = 2.70, p = 0.2702, η²_H = −0.08), even though average values were again higher in the spring than in the fall-PS group (Fig. 5D, J and M-right; see representative microphotographs in Fig. 7). There was in contrast a significant difference between groups in the numbers of PV-expressing neurones (Kruskal-Wallis test: H₂ = 11.53, p = 0.0005, η²_H = 0.44) that were more numerous in the spring group compared to the fall-CS group (see Fig.5G-right). Additionally, no changes in the densities of PNN, PV and PV + PNN could be detected in this nucleus. Although the density of PV cells changed significantly in the previous experiment, there was only a statistical trend in the present experiment (p = 0.0865). Finally, the percentage of PNN surrounding PV cells did not differ across groups (Table 2).

Fig. 7.

Representative photomicrographs of the PV and PNN staining in HVC, RA and Area X in canaries from the fall-PS and the spring groups in experiment 3.

In RA, the changes in PNN expression were reminiscent of the changes in singing behaviour. As observed in experiment 2, there was a seasonal variation in the numbers of PNN (Kruskal-Wallis test: H₂ = 8.93, p = 0.0057, η²_H = 0.29), PV + PNN (Kruskal-Wallis test: H₂ = 8.61, p = 0.0073, η²_H = 0.27) and in the percentage of PV with PNN (Kruskal-Wallis test: H₂ = 8.55, p = 0.0077, η²_H = 0.27), but no changes in the numbers of PV (Kruskal-Wallis test: H₂ = 2.77, p = 0.2638, η²_H = −0.07). The post hoc analyses showed that the three measures of PNN were significantly higher in the spring group than in the fall-PS group (Fig. 5F, L and O-right). Additionally, there was a trend for the fall-CS group to have a higher number of PV + PNN and a higher percentage of PV with PNN than the fall-PS group (Fig. 5K and N-right). There was also a significant decrease in the density of PV cells in RA in the spring compared to the fall-PS group as shown in the previous experiment (Table 2-right).

Similar differences were found in Area X. There was an effect of group on the number of PNN (Kruskal-Wallis test: H₂ = 9.47, p = 0.0037, η²_H = 0.32), PV + PNN (Kruskal-Wallis test: H₂ = 8.14, p = 0.0171, η²_H = 0.24) and the percentage of PV with PNN (Kruskal-Wallis test: H₂ = 6.93, p = 0.0238, η²_H = 0.17) without changes in the numbers of PV-expressing neurons (Kruskal-Wallis test: H₂ = 4.35, p = 0.1108, η²_H = 0.02). Post hoc analyses showed that these effects were due to the spring group that had significantly more PNN and PV + PNN and a higher percentage of PV with PNN than the fall-PS group (Fig. 5F, L and O-right). For both RA and Area X, the pattern of changes in the total numbers of PNN and of PV + PNN per nucleus was similar for the densities of these measures (Table 2-right). The decrease seen during experiment 2 in the percentage of PNN surrounding PV was just a statistical trend in the present experiment (Table 2-right).

4. Discussion

Together, the present studies demonstrate that a) male Fife Fancy canaries display distinct annual cycles in plasma testosterone concentrations and in song production, b) seasonal plasticity in song structures varies in parallel, and c) PNN numbers and densities are lower in the song control nucleus RA during the fall when song is more plastic. Qualitative trends in the same direction were observed in HVC and Area X, but were not statistically significant based on the current data set. These observations thus suggest a role of PNN present in the song control nuclei in the maintenance of song stability, as will be discussed below.

4.1. Photoperiodic control of testosterone and singing in the Fife Fancy canary

It is commonly assumed that one of the key environmental changes regulating seasonal plasticity in the adult song control system is the seasonal change in day length that is responsible in turn for the seasonal fluctuation of testosterone [17]. This androgen is then thought to play a key role in the control of SCS plasticity [23]. This conclusion has been supported by many correlational studies of testosterone SCS plasticity from field caught birds, photoperiodic manipulations in the laboratory as well as concomitant castration and testosterone replacement studies, which mimicked the effects of photoperiodic manipulations [29,54-59]. Because seasonal fluctuations of plasma testosterone correlate with seasonal variations in the volume of the SCS nuclei and in song production in adult seasonal songbirds [20,24,60,61], exogenous testosterone has also been used as a means to study vocal learning-induced neuroplasticity in adult songbirds. However, such manipulations often lead to supraphysiological testosterone concentrations in the blood [62] and are thus not necessarily representative of the seasonal brain plasticity associated with seasonal variations of song production in ecological conditions. Moreover, a few studies have previously shown that aspects of the seasonal changes in singing behaviour and neural plasticity in adult songbirds, such as the song rate and the volume of song control nuclei, can be testis- and androgen-independent [59,63,64]. Additionally, the magnitude of seasonal changes in singing behaviour varies widely between species and the same is also true for different aspects of adult brain plasticity [2]. Here, we confirmed in one breed of domestic canaries that most aspects of song variation that can be induced by treatment with exogenous testosterone also occur in natural conditions during the breeding season.

Canaries have been domesticated for centuries and artificial selection combined with inbreeding has produced multiple strains with substantially different behavioural and physiological traits [65-67]. These different strains display different relationships between singing behaviour, physiology and patterns of neural plasticity. Therefore, before attempting to analyse the relationships between PNN and singing behaviour in a seasonal context, it was important to confirm that the strain of canaries we were using was photoperiodic and exhibited seasonal changes in singing behaviour and testosterone concentrations.

The first studies of adult seasonal plasticity in brain and behaviour were made with the Belgian Waterschlager strain that displays a clear pattern of seasonal variation in terms of song plasticity, testosterone concentrations and SCS nuclei volume [19,20,68]. During the breeding season of Waterschlager canaries, song is more stereotyped, plasma testosterone concentration is high and the volumes of the SCS nuclei are larger than outside the breeding season. These results were partly confirmed using outbred common domesticated canaries [17] and wild canaries [21] Laboratory studies with the Border or the Fife Fancy strain of canaries [46,49,56,69] also confirmed that these strains respond robustly with variation in reproductive physiology and brain plasticity to changes in the photoperiod that places the birds in different reproductive states. However there are apparently no completed studies describing a full annual seasonal cycle in these strains.

In a first experiment, we showed here that the Fife Fancy strain of canaries displays a clear pattern of testosterone fluctuations across two seasons that parallel the seasonal fluctuations in singing behaviour. Testosterone concentrations began to increase in the spring and apparently peaked in late June whereas this peak was previously shown to occur in March in the Belgian Waterschlager strain [20]. The time resolution of the present study does not allow us to be sure this represents a true difference. There are many reasons why there might be a difference in the timing of the testosterone peak. The different breeds of birds could have different photoperiodic thresholds required to stimulate testicular growth, it could be due to the slight difference in photoperiod corresponding to 41 °N (New York) or 51 °N (Antwerp) or to multiple aspects of the housing conditions (presence or not of females, of nest material,…) that could provide additional non-photoperiodic factors related to the control of sexual maturation. In any case the present study, showing similar peaks in two successive years, clearly indicates the periodic, presumably annual, nature of testicular activity in these birds.

Detailed analyses of singing behaviour showed that, as observed in the studies of Nottebohm and colleagues [68], song crystallization began during the winter, the song was fully crystallized in the spring and remained crystallized until early summer at least. Thus, it seems that the song may remain crystallized for a longer period in the Fife Fancy strain than in the Belgian Waterschlager which barely sings during the summer [19]. Again, whether these differences reflect differences between strains of canaries or in the housing conditions cannot specified at this time. The largest decrease of testosterone was observed between the early summer and the fall time point and it is at that time that song became more plastic. It has been shown experimentally in song sparrows and swamp sparrows that testosterone is required for song crystallization to occur and that the removal of testosterone can re-induce a state of plastic song [70]. Additional measures of song structure and quality such as the song duration, RMS amplitude, entropy and developmental score also confirmed that the largest changes occur between the summer and fall when the song becomes more plastic. Similar changes in song quality between fall and spring and/or winter were also observed in the two other experiments focusing on specific time points. An analysis of the testosterone concentrations derived from the second experiment confirmed that the testosterone increase that starts during the winter is already sufficient to initiate the process of song crystallization. The second and third experiment also demonstrated the presence of clear seasonal variations in the testes weight, which is a good indicator of the physiological state of these birds. All birds used in these three experiments remained in similar environmental conditions including constant presence in the room of other males, ad libitum food availability and constant temperature. The only factor that varied systematically across time was the photoperiod that was adjusted every month to match the outdoor photoperiod of Belgium. These data therefore indicate that the Fife Fancy canary is responsive to seasonal variation in photoperiod and that changes in brain and vocal behaviour are the result of the photoperiodic changes inducing fluctuations in plasma testosterone concentrations.

4.2. Seasonal changes in song rate and structure in the Fife Fancy Canary

It was previously shown that canary song becomes more plastic and variable in fall and that this period closely resembles the sensorimotor learning period taking place during developmental song learning [20,68]. After this adult sensorimotor period, the song crystallizes again and then displays a different repertoire composition. It was also shown that in common domesticated canaries, the song duration is increased during the breeding season [21]. This was confirmed here in the Fife Fancy strain. We also confirmed that the song becomes more plastic and variable in fall using both a qualitative measure (the song developmental score) and automated quantitative measures (the song average entropy) in the Fife Fancy strain.

Multiple song parameters also changed seasonally in this strain. Fife Fancy canaries are singing louder songs during the breeding season compared to the fall period and the frequency at which the song power is concentrated is displaced towards higher frequencies. Interestingly, these changes in the song amplitude and entropy were similarly observed in the Border canary and other domesticated strains of canaries after treatment with exogenous testosterone, suggesting that these seasonal changes are androgen-dependent [69,71]. However, in contrast, there was no increase of the song rate during the breeding season in the three experiments described here, whereas testosterone was previously shown to increase this song rate in castrated males [49,69,71]. This apparent discrepancy is probably due to the fact that the experiments studying the effect of testosterone on singing behaviour used birds that were first placed in short day photoperiods for a long time (6–8 weeks) and were even castrated which generally leads a complete cessation of singing. An increase in song rate is thus easier to induce and detect in these conditions.

Some measures of singing behaviour did not significantly change during the second experiment, which was to some extent unexpected. This is likely due to the fact that the birds from the fall group were recorded in late fall and had already started crystallizing their song; three birds already had a high score of song development around 4 (see Fig. 5D). For this reason, we decided to record birds earlier in fall for the third experiment, and to classify these birds according to their song development. This allowed us to identify specific aspects of song that change across the early stages of adult song plasticity between early sensorimotor learning (the fall-PS group) and the first stages of song crystallisation. It should be noted that many aspects of singing were already changed to the spring level in the fall-CS group, which means that song crystallisation takes place relatively early in the fall. Other traits, such as the song RMS amplitude and other measures of song loudness, had started to increase in the fall-CS group, but were still significantly higher in the spring group, suggesting that this aspect of singing is more closely related to the breeding period and the production of a fully crystallized song.

4.3. PNN in the SCS and the regulation of adult seasonal plasticity

In parallel with the seasonal changes in vocal behaviour, there was a significant increase of HVC volume between the fall-PS group and the spring group, but this growth was not accompanied by any changes in the number or density of PNN in HVC. It was previously shown that PNN expression increases in HVC at the end of developmental song learning period of male canaries [42], but we failed to detect similar changes across seasons, suggesting that PNN in HVC do not play a major role. It must however be noted that the total number of PNN and their densities were numerically higher during the spring in both experiments 2 and 3, while song had often started to redevelop in the fall samples that we collected. Therefore, we cannot exclude that our sampling schedule missed the lowest point of PNN expression in HVC during the late summer/early fall. This is also suggested by the very low values observed in the fall group from experiment 2 and in the fall-PS group during experiment 3. There was however in spring an increase of PV-expressing neurones in this nucleus that might relate to the adult song plasticity. This observation raises the possibility that PNN could also increase during the summer after the increase of PV-interneurones is first observed. Indeed it was previously shown that, in female canaries, testosterone increases the number of PV and that this increase is followed by development of PNN around these PV-expressing neurones [45]. More detailed time-course studies would be needed to test this possibility.

In RA, we detected clear variations in the PNN numbers, density and their localization around PV across seasons. There was a higher PNN expression and localization around PV neurones during the winter in the second experiment, and in both the fall-CS and spring group in the third experiment. As these groups of birds had initiated their song crystallisation process, these data suggest that PNN develop around PV interneurones during the song crystallisation process before the onset of the breeding season. The small decrease in PNN and PV + PNN that took place between winter and fall during experiment 2 was not significant in itself and it must be considered that we are dealing here with measures in independent groups of birds. We would therefore argue that all data suggest that PNN and PV + PNN are higher when song is crystalized in males. RA is a premotor nucleus that receives projections from both HVC and LMAN [9]. This nucleus is directly connected to the motorneurones of nXIIts that control the syrinx muscles contractions [72]. It was previously shown that testosterone action in RA controls, among other things, the song duration and the variability of specific song elements [50]. We previously showed that treatment with exogenous testosterone increases the numbers of PNN in RA of castrated Fife Fancy canaries [45]. It is known from experimental studies during development in male sparrows and from the treatment of adult females with testosterone that testosterone crystallizes the song [70,73]. It is thus likely that the development of PNN in RA allows a decrease of specific aspects of song variability to allow production of stereotyped elements that are typical of crystallized song.

Interestingly, there was also an increase of PNN numbers, density and colocalization with PV in Area X. However this increase was statistically significant in the spring group only during the third experiment. The lack of effect in the second experiment was probably due to the smaller differences between groups in terms of singing development related to a sampling carried out too late in the fall. As PNN development around PV interneurons occurred only in the spring when the song was fully crystallized, it is possible that the role of PNN in this nucleus relates to the maintenance of the song stability after the end of the crystallisation process. It is well known that auditory feedback is necessary to maintain a stable crystallized song [74]. It was also shown that the anterior forebrain pathway is regulating the action of the auditory feedback on song learning mainly through the action of LMAN [75]. This nucleus also plays a role in the production of song variability. But LMAN is controlled upstream by DLM which receives inhibitory projections from Area X [76]. Consequently, a higher inhibition provided by Area X decreases LMAN-induced variability. As part of the anterior forebrain pathway, Area X probably plays an important role in song maintenance through the auditory feedback. The increase of PNN in Area X could be a result of the birds hearing their crystallized song and this would in turn inhibit the action of DLM on LMAN activation, ultimately maintaining the song crystallized across the breeding season. This process would probably progressively lead to a ceiling effect in the expression of PNN in Area X when the song is fully crystallized. It is also interesting to note that the song RMS amplitude followed a very similar pattern as the numbers of PNN, of PV + PNN and the percentage of PV with PNN in Area X. It is thus tempting to specifically relate the control of the song amplitude to the development of PNN in Area X, but this should be tested by causal mechanistic manipulations.

5. Conclusions

Overall, we provide here evidence that, in canaries, PNN expression in RA and Area X is lower in fall especially during the early phases of sensorimotor learning. The expression of PNN in some SCS nuclei thus varies seasonally in parallel with changes in song production and structure. These changes in PNN expression are anatomically quite specific and actually more specific than what is observed during ontogeny [39,42]. The seasonal plasticity of the song system thus seems to recapitulate ontogeny only to a limited extent in some specific nuclei. The low expression of PNN during fall might be causally related to the increased vocal plasticity observed at that time. This is possibly an important mechanism controlling vocal learning in open-ended learner songbirds - at least in species with only limited seasonal changes in song. European starlings, which maintain a high degree of vocal plasticity as adults and are able to add new syllables to their repertoire all year long [13,14,77], do not seem to display seasonal variation of PNN expression and they have a very low density of PNN in the SCS compared to zebra finches and canaries [52]. It is thus likely that PNN, which are known to close sensitive periods of learning in mammalian models, also regulate the seasonal opening and closing of sensitive periods for sensorimotor learning in adult songbirds. This hypothesis should now be tested via causal manipulations of PNN numbers in the song control nuclei.