Characteristics of song, brain-anatomy and blood androgen levels in spontaneously singing female canaries

Hormones and Behavior 117 (2020) 104614

Contents lists available at ScienceDirect

Hormones and Behavior journal homepage: www.elsevier.com/locate/yhbeh

Characteristics of song, brain-anatomy and blood androgen levels in spontaneously singing female canaries

T

Meng-Ching Koa, , Vincent Van Meira, Michiel Vellemab, Manfred Gahra ⁎

a b

Max Planck Institute for Ornithology, Dept. of Behavioural Neurobiology, Eberhard-Gwinner str. 6a, 82319 Seewiesen, Germany Utrecht University, Dept. of Experimental Psychology, Yalelaan 2, 3584, CM, Utrecht, the Netherlands

ARTICLE INFO

ABSTRACT

Keywords: Songbird Canary Androgens Birdsong Female canary song Song analysis HVC RA Sex differences Plasticity

Females of many northern temperate songbird species sing sporadically. However, detailed descriptions of female song are rare. Here we report a detailed analysis of song in a small number of spontaneously-singing female domesticated canaries (Serinus canaria) under non-breeding, laboratory conditions in a large population of domesticated birds. In-depth analysis showed that these females sang rarely, and the spontaneous songs varied between and within birds over time. Furthermore, spontaneous female songs were distinct from songs of testosterone-induced singing female canaries and from songs of male canaries in both temporal and spectral features. Singing females had significantly elevated plasma androgen levels and a larger size of the major song controlling brain nuclei HVC (used as a proper name) and the robust nucleus of the arcopallium (RA) than nonsinging females housed under similar conditions. The sporadically observed production of song and accompanying differences in brain anatomy in female canaries may thus depend on minute intraspecific differences in androgen levels.

1. Introduction Songbirds (oscines) constitute approximately half of all extant bird species (Gill and Donsker, 2015) and live in diverse habitats. Male songbirds use songs in various contexts such as mate attraction and territory defense (Catchpole and Slater, 1995; Kroodsma and Byers, 1991; Searcy and Andersson, 1986; Wickler and Seibt, 1980). A recent study suggested that among 1314 surveyed songbird species, females produce songs of different quality in at least 656 of them (Webb et al., 2016). Singing in females might be ancestral and widespread (Odom et al., 2014), but among species of the northern temperate zone, female songs are uncommon in general and, if present, are often restricted to a specific context or to a limited group of individuals (reviewed in (Gahr, 2014)). The patterns of female song have only been analyzed quantitatively in a few northern temperate species: the European robin (Erithacus rubecula) (Hoelzel, 1986), the European starling (Sturnus vulgaris) (Pavlova et al., 2005), the red-winged blackbird (Agelaius phoeniceus) (Beletsky, 1983; Yasukawa et al., 1987), the Northern Cardinal (Cardinalis cardinalis) (Yamaguchi, 1998, 2001), the white-crowned sparrow (Zonotrichia leucophrys) (Baptista et al., 1993) and the dark-eyed junco (Junco hyemalis) (Reichard et al., 2018). Additionally, occasional or individual female song was reported in a larger number of northern species (Langmore, 1998; Ritchison, 1983; Webb et al., 2016). Some of ⁎

these reports are convincing but miss a quantitative analysis of female song structure, for example, the canary (Serinus canaria) (Pesch and Güttinger, 1985), the house wren (Troglodytes troglodytes) (Johnson and Kermott, 1990) and the song sparrow (Melospiza melodia) (Arcese et al., 1988). Further studies report female song only anecdotally and lack clear methods to identify the sex of the singing individuals as in skylarks (Alauda arvensis) (Delius, 1965), chaffinches (Fringilla coelebs) (Halliday, 1948), Eurasian reed warblers (Acrocephalus scirpaceus) (Kuschert and Ekelöf, 1981) and American dippers (Cinclus americanus) (Bakus, 1959). The domesticated canary (Serinus canaria) is a northern temperate songbird originating from wild populations of the North Atlantic islands (Leitner et al., 2001). They breed seasonally, and optimal breeding conditions are tightly associated with an increase in day length, which initiates gonadal development as well as various types of breeding activity including singing. Thus, the canary is an important model species for studying plasticity of birdsong and its underlying neural circuitry and hormonal regulation: for review see (Chen et al., 2013; Gahr, 2004). A major component of the so-called song control system is the HVC, an anatomically well-defined nucleus, critical for many functions such as integration of auditory input and motor output as well as song pattern generation (Hahnloser et al., 2002; Nottebohm et al., 1976; Wild, 2004). Neurons in the HVC project to the robust nucleus of the

Corresponding author. E-mail address: [email protected] (M.-C. Ko).

https://doi.org/10.1016/j.yhbeh.2019.104614 Received 30 December 2018; Received in revised form 8 October 2019; Accepted 8 October 2019 Available online 01 November 2019 0018-506X/ © 2019 Elsevier Inc. All rights reserved.

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

arcopallium (RA), which regulates the motor output of songs by projecting to brain nuclei that control the avian sound producing organ (the syrinx) and the respiratory organ (Schmidt and Martin, 2014). Studies of male canaries showed that testosterone regulates various aspects of singing behavior in multiple brain nuclei within and outside of the song control system (Alward et al., 2018; Alward et al., 2016a, 2016b). The HVC and RA of male canaries undergo anatomical (Gahr, 1990; Nottebohm, 1981) as well as transcriptional (Frankl-Vilches et al., 2015) seasonal plasticity, both processes being testosteronesensitive. The song control system is present in both sexes; singing as well as a growth of the HVC can be easily induced by testosterone treatment in females. Since the neural changes underlying testosteroneinduced female singing are thought to be comparable to changes that naturally occur in males due to seasonality (Balthazart et al., 2010; Brenowitz and Lent, 2002; Devoogd et al., 1985; Nottebohm, 1980), female canaries are used to mimic hormone-dependent plasticity of birdsong and its neuronal correlates (Fusani et al., 2003; Madison et al., 2015; Nottebohm, 1980; Shoemaker, 1939; Vellema et al., 2019). To be able to reliably use female canaries as a model for testosterone-induced singing in songbirds, it is important to evaluate spontaneous female singing in this species. Although sporadic observations of song in female canaries without hormone treatment have been reported (Hartley et al., 1997; Herrick and Harris, 1957; Pesch and Güttinger, 1985; Shoemaker, 1939; Vallet et al., 1996), the structure of spontaneously uttered song has not previously been quantitatively described, and data on the underlying physiological mechanisms are missing. By screening continuous sound recordings of a large population of 112 female canaries, we were able to document spontaneous songs in six individuals. We show that females are capable of producing songs, albeit infrequently and variable between and within individual. In addition, we show that HVC volume, RA volume, blood plasma androgen level, and oviduct mass were greater in spontaneously-singing females than in non-singing ones. By comparing song features of spontaneouslysinging females to testosterone-induced singing females and males in the breeding season, we show that spontaneous female songs were distinct from songs of testosterone-induced singing females and males.

monitoring. Body weights were measured and the brains, syrinxes, oviducts, and follicles were dissected, weighed, snap frozen on dry ice and stored at −80 °C for further analysis. Songs recorded during the breeding seasons, between May and August, of nine male canaries kept under a natural light regime in Germany were included in this study. 2.2. Song monitoring For monitoring singing activity, canaries were housed individually in sound-attenuated boxes (70 × 50 × 50 cm) and recorded continuously for up to four weeks. A microphone (TC20, Earthworks) in each box was connected to a PR8E amplifier (SM Pro Audio), feeding into an Edirol USB audio capture device (Edirol UA 1000, Roland) connected to a computer. Songs were recorded at a sampling rate of 44.1 kHz and 16-bit resolution using the software Sound Analysis Pro 2011 (Tchernichovski et al., 2000). The amplitude filtering (> 24 dB) functionality was used. Recordings provided by Dr. Hans Rudolf Güttinger were recorded using an Uher 4400 report stereo IC tape recorder at a speed of 19 in. per second (ips). The tapes were digitized at a sampling rate of 44.1 kHz and 16-bit resolution using the software Sound Analysis Pro 2011. Female canaries with the potential to develop song were identified by visual inspection of their sound spectrograms (sonograms). Birds that frequently produced sound sequences composed of more than five consecutive syllables were selected for further monitoring. When the recording period exceeded four weeks, they were switched back to same-sex aviaries in non-breeding condition for at least four more weeks before another song monitoring period started. Song monitoring was repeated three to five times from February 2012 to August 2015 to monitor song activity and song development of six spontaneouslysinging female canaries. All non-singing females were monitored at least for four weeks to make sure they did not sing. Female canaries implanted with testosterone were monitored for at least four weeks to make sure they did not sing before testosterone implantation. After implantation, testosterone-treated female canaries were recorded for two weeks. Before song monitoring, male canaries were kept in singlesex aviaries in a local natural light cycle according to South Germany. Between May and August, male canaries were individually housed in sound-attenuated boxes and monitored for their singing activity for four weeks as described above for female canaries.

2. Material and methods 2.1. Animals Adult canaries (Serinus canaria, at least one-year-old) used in all experiments were either purchased from local breeders during the breeding season (June to August) or bred at the animal facility of the Max Planck Institute for Ornithology in Seewiesen, Germany. Sex was confirmed by PCR using P2 and P8 primers for CHD genes (Griffiths et al., 1998) and visual inspection of the reproduction system after sacrifice. These birds were housed in single-sexed aviaries (1.95 × 1.0 × 1.8 m) before song monitoring. Food and water were provided ad libitum at all times. Animal housing and welfare followed the European directive for the protection of animals used for scientific purposes (2010/63/EU). Protocols were approved by the Government of Upper Bavaria (AZ 55.2-1-54-2532-181-12). A total of 112 female canaries were housed in non-breeding conditions 9/15 (light/dark) and were screened for singing activity. Six long-term monitored and one short-term monitored spontaneouslysinging females were included in this study. Dr. Hans Rudolf Güttinger (Pesch and Güttinger, 1985) provided song recordings of one additional spontaneously-singing female canary (short-term). Six non-singing females were implanted with testosterone for two weeks to induce singing activity. Before testosterone implantation, females were held in nonbreeding condition for at least 8 weeks after switching from the breeding condition. The six long-term monitored spontaneously-singing females were sacrificed after at least 6 months of song monitoring and within one hour after the last singing had been observed; nine nonsinging females were sacrificed after at least four weeks of song

2.3. Song analysis Song analysis was done using Multi_Channel_Analyser (MCA), a custom program with graphical-user-interface written in MATLAB (version R2016b, Mathworks). MCA loads song files and generates sound spectrograms using a fast Fourier transformation with sliding time windows of 294 samples and 128 samples of overlap between adjoining sections. These settings lead to a spectrogram with a resolution of 150 Hz on the frequency axis and a resolution of 3.76 milliseconds on the time axis. For long-term monitored spontaneouslysinging females, we analyzed whole day recordings (nine hours) at least every four days for each bird (Fig. 1A and Supplementary Table 1). For testosterone-induced singing females, we analyzed all recordings. For male canaries, we analyzed an average of 150 songs per bird. Due to undesired background noise, syllable detection and song definition were done in three steps. First, audio segments (which might contain several songs) without substantial background noise were manually selected by visual inspection. Second, syllables were automatically detected within the selected segments using MCA. Third, consecutive syllables with inter-syllable pauses shorter than 0.5 s were assigned to one song. When the inter-syllable pause was longer than 0.5 s, the latter syllable sequence was assigned to a new song. We considered songs unstable if song length (the interval between the first and the last syllables) was shorter than 3 s (Fig. 1A). Since canary syllables can be very short and usually occur as phrases 2

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

A

B

2013-08-16 10:32:00

Bird2 (kHz)

8 6 4 2 0

2

4

C

6

(sec)

2014-09-30 10:27:06

Bird3 (kHz)

8 6 4 2 0

2

4

6

8

(sec)

Fig. 1. Long-term monitoring of singing activity reveals that female canaries can sing spontaneously. A, A summary of long-term data collection of six female canaries, which were recorded repeatedly in sound-attenuated boxes (gray dots). The occurrences of songs (blue) and unstable songs (light blue) were marked. Song recordings were analyzed in two to three blocks separately and indicated by shaded boxes. B, An example song spectrogram of bird 2. C, An example song spectrogram of bird 3.

with a high syllable repetition rate, we segmented syllables as follows. First, the audio signal was filtered with a high-pass filter at 2.0 kHz. Next, we detected at which time the values of the audio samples crossed a certain amplitude threshold, which was called the Threshold_wave_amplitude in the MCA. The Threshold_wave_amplitude is relative to the maximum intensity of the recorded audio signal, which is scaled between −1 and 1. For this study, the Threshold_wave_amplitude was set between 0.001 and 0.005. Subsequently, we determined the interval between all subsequent crossing time points and calculated the frequency for each time the threshold was crossed. Finally, we set a threshold for this time series of frequencies, called Threshold_syllable_freq in the MCA, and determined at which time points this threshold was crossed in upward and downward direction, representing the syllable start and its end respectively. We set the Threshold_syllable_freq between 50 and 500 Hz, depending on the syllable repetition rate of each bird. This means that the syllables will be correctly segmented even if the gaps between subsequent syllables in a trill are as low as 20 to 2 milliseconds. Syllables shorter than 5 ms were removed. Subsequently, digital samples of detected syllables were exported and further used to calculate sound features. We measured four songlevel and four syllable-level parameters. Song-level parameters included three temporal features: song length, the number of syllables in a song, and syllable repetition rate, and one spectral feature: slope coefficient (α). Slope coefficient (α) is the downward slope of the loglog transformed spectrum of the amplitude modulation of the song (see below for definition). Syllable-level parameters include two temporal features: syllable length and inter-syllable interval, and two spectral features: peak frequency and Wiener entropy. Song length was calculated by subtraction of the timestamp at the end of the last syllable and the timestamp at the start of the first syllable of a song. Syllable repetition rate was defined as the number of syllables in the song divided by the song length in seconds. Syllable length was calculated by subtraction of the timestamp at the end of the syllable and at the start of that syllable. Inter-syllable interval was calculated by subtraction of the timestamp at the onset of the subsequent syllable and at the end of the

syllable. In order to compare syllable-level spectral features, we used a Fourier transformation to extract the amplitude of each frequency component in the original audio signals, i.e., the syllables, and calculated the peak frequency and Wiener entropy. We first high-pass filtered all syllables at 500 Hz to remove noise and normalized the amplitude by dividing their root mean square (RMS) value. Given that every syllable was variable in length but all were shorter than 1 s and were recorded at a sampling rate of 44.1 kHz, we standardized the length of all syllables by adding 4,096 zeros (approximately equivalent to 93 ms, this amount was chosen arbitrarily) in front of the original signal and adding zeros behind to a total number of 44,100 samples. In addition, because Fourier transformation assumes that the signal oscillates infinitely, adding zeros before and after the signal makes a signal that starts and ends with the same amplitude and would render a more accurate Fourier transformation output. From these normalized syllables, we calculated a spectrum, peak frequency, and Wiener entropy. Peak frequency was the frequency at which the power was at a maximum in the spectrogram of a syllable. Wiener entropy is a unitless measure of signal noise that provides a measure of the width and uniformity of the power spectrum. To have a wider range we used the logtransformed Wiener entropy. In this case, a Wiener entropy of 0 would represent white noise and infinity would represent a pure tone (Tchernichovski et al., 2000). To get a general impression of the song structure, we calculated a song-level spectral feature, the slope coefficient (α), by analyzing the low frequency components (0.5 and 2 Hz) of the amplitude modulation of the song. Specifically, we calculated the power of each sample by means of a Hilbert transformation, applied a low pass filter at 441 Hz and down-sampled them to a rate of 441 samples per second. We then calculated the spectrum and analyzed the frequency range between 0.5 and 2 Hz. This range represents the amplitude modulation below the slowest syllable repetition rate (0.5 s) and above the shortest song length (2 s). Different syllable repetition rates are represented in the spectrum as peaks in the frequency range between 2 and 30 Hz. The spectrum below 2 Hz represents the song organization at the level of 3

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

phrases. In this range, the spectrum is not organized in peaks but as a downward slope. After taking the log transformation of the frequency and the amplitude axes of the power spectrum, the downward slope can be calculated by taking the linear regression. The slope coefficient (α) represents the level of predictability or repeatability in sound structures by reflecting the change in amplitude or loudness of the subsequent phrases (Gardner, 1978; Gisiger, 2001; Voss and Clarke, 1975). The slope coefficient represents the roughness of the transitions in amplitude. Between different phrases there can be a sudden transition or a smooth transition, while within a phrase there are mainly smooth transitions. In unstable songs the change in amplitude between subsequent syllables can be very jumpy, less predictable, even completely erratic, thus the transitions in amplitude are rough. The presence of only rough transitions results in a slope coefficient of 0. Noise in this category (white shaded area in figures) is a random superposition of waves over a wide range of frequencies (Gisiger, 2001). The presence of only smooth transitions results in a slope coefficient of −2 (α ϵ [−2.5, −1.5]). This could result from smooth transitions in amplitude within a phrase, but also between phrases. Noise in this category can be reproduced by adding a random offset to each sample to obtain the next one, thus it is strongly correlated with time (Gilden et al., 1995; Gisiger, 2001; Voss and Clarke, 1975). In figures, we indicated noise in this category in brown shading. A combination of smooth and rough transitions results is a slope coefficient between 0 and −2. The slope coefficient of −1 (α ϵ [−1.5, −0.5]) is believed to represent an optimal balance between predictability and surprise. (Singh and Theunissen, 2003; Yang et al., 2015) The slope coefficient is a way to describe temporal song features ranging from complete chaotic utterances to orderly arranged syllables in a sonogram. In figures, we indicated noise in this category in pink shading. Noise with slope coefficient ranging from 0.5 to 1.5 has a power spectrum with amplitude proportional to its frequency, implying highly repetitive units with short repetitive intervals. In figures, we indicated noise in this category in blue shading. The analysis of the slope coefficient has been used in birdsong analysis (Singh and Theunissen, 2003) and more recently in studies describing natural soundscapes (Yang et al., 2015).

Sciences, Tarzana, USA) as previously described (Goymann et al., 2002). Standard curves and sample concentrations were calculated with Immunofit 3.0 (Beckman Inc. Fullerton, CA) using a four-parameter logistic curve fit and corrected for individual recoveries. For the female singers, the blood was sampled no longer than one hour after their last song before sacrifice. The range between the onset of the first song and sacrifice was 40 min to 3 h. Blood samples of nonsinging females were taken before sacrifice. All blood samples were taken between 8 and 11 am and were taken within 3 min to avoid interference from the effect of handling (Wingfield et al., 1982). Androgen concentrations were assayed in duplicates, in three separate assays. The mean extraction efficiency for plasma testosterone was 86.8 (4.12) % (mean (SD), N = 15). The extraction recovery was measured and corrected for individual samples. All samples were above the lower detection limit of the testosterone (0.35, 0.36 and 0.38 pg per tube). The intra-assay coefficients of variation of a chicken plasma pool were 3.4%, 12.8%, and 1.9%. The inter-assay coefficient of variation as determined by the variation of the chicken plasma pool between all assays was 10.6%. Because the testosterone antibody showed significant crossreactions with 5α-dihydrotestosterone (44%), our measurement may include a fraction of 5α-DHT, thus we called this measurement “plasma androgens” in this manuscript. 2.6. HVC and RA volume measurement Brains were sagittally sectioned in 50 μm × 4 + 14 μm × 3 sections with a cryostat (Jung CM3000 Leica). The thin sections (14 μm) were mounted on RNase-free Superfrost slides for Nissl staining. All sections were stored at −80 °C until further processing. One set of thin serial sections (14 μm) was sequentially hydrated (100%, 90%, 70%, 20% ethanol, and distilled water, each solution for 50–60 s), stained with 0.1% thionin solution for 5–8 s, and serially dehydrated (distilled water, 20%, 70%, 90% and 100% ethanol, each for 30 s). Finally, the slides were immersed in xylene and cover-slipped with Roti-Histokitt II mounting medium (Roth). The HVC and RA areas were measured using ImageJ2 (Rueden et al., 2017) (Fiji distribution (Schindelin et al., 2012)) based on microphotographs taken with a Leica DM6000 B microscope. All brains were coded so that the observers were blind towards any additional information about the sections they measured during the delineations. Volumes were derived from the summed area measurements multiplied by the section thickness and the inter-section distance.

2.4. Testosterone implantation Silastic™ tube (Dow Corning; 1.47 mm inner diameter, 1.96 mm outer diameter, 0.23 mm thickness) was cut to 7 mm in length and was loaded with crystalline testosterone (86,500, Fluka) as dense as possible. The two ends of the Silastic™ tube were sealed with silicone elastomer (3140, Dow Corning). After closure, the implants were cleaned with 100% ethanol to remove testosterone particles from the outside of the implant and then were immersed in ethanol overnight in the hood to ensure no leakage at both ends. Implants with apparent damp were discarded. One day before the implantation the implants were incubated in 0.01 M phosphate buffered saline (PBS) overnight enabling the immediate release of testosterone upon implantation (Rasika et al., 1994). We started implantation right after the light was turned on in the morning, approximately at 8:30 am with 20 min interval between each bird, accounting for the time required for sacrifice. A small incision was made on the back of the bird over the pectoral musculature, and one testosterone implant was placed subcutaneously. The skin was closed by application of tissue glue. We implanted six female canaries for two weeks.

2.7. Experimental design and statistical analysis Fifteen adult female canaries (six long-term monitored spontaneously-singing and nine non-singing) were used for song monitoring, HVC and RA volume measurement, and weight measurements (body, brain, syrinxes, oviduct, and follicles). Six adult female canaries were implanted with testosterone for two weeks and were monitored for singing activity. Songs of nine male canaries from breeding seasons were analyzed. All statistical analyses were performed in R (RCoreTeam, 2015). For statistical comparisons of song features, we used linear mixed-effect models to analyze the effect of different individuals, recording periods, or groups on the eight measured song features using the packages ‘lme4’ (Bates et al., 2014) and ‘arm’ (Gelman and Su, 2018) in a Bayesian framework with non-informative priors. We assumed a Gaussian error distribution and confirmed that model assumptions were met by visual inspection of model residuals. Song rate, song length, syllable repetition rate, syllable length, inter-syllable interval, and peak frequency were log-transformed, while the number of syllables in a song, slope coefficient (α) and Wiener entropy were used un-transformed. Subsequently, we simulated 10,000 times the estimates from the joint posterior distribution of the model parameters using the ‘sim’ function and extracted the 95% credible intervals (CrI) of the mean of the simulated values,

2.5. Radioimmunoassay of plasma testosterone Blood (< 150 μl) was taken from the wing vein and collected in heparinized micropipettes for plasma testosterone analysis. Blood samples were centrifuged (2500 rpm, 10 min) to separate the plasma from blood cells, and stored at −80 °C until further analysis. Testosterone metabolites were measured with a radioimmunoassay using a commercial antiserum against testosterone (T3-125, Endocrine 4

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

monitored for at least 40 days each and we analyzed at least 20% of the recordings of each bird (Supplementary Table 1). We defined a sound segment as an individual song if the segment was longer than three seconds and inter-syllable gap were shorter than 500 milliseconds (ms). All 112 recorded females produced calls.

which represent the uncertainty of the estimates (Gelman and Hill, 2006). We considered an effect to be statistically meaningful when the estimated CrI did not include zero or when the posterior probability of the mean of the difference between compared estimates was higher than 0.95 (for more details on statistical inference see (KornerNievergelt et al., 2015)). For visualization, we plotted the predicted estimate from the models, the 2.5% and 97.5% CrI, as well as the raw data. Statistically meaningful differences between compared groups can be inferred if the CrI of one group does not overlap with the mean estimate of the other. Specifically, for comparing song features between individual spontaneously-singing female canaries, we calculated the daily average for each song feature measured for each bird. Subsequently, we used the daily average of a song feature as dependent variable, individual birds as fixed effect and recording periods as a random effect. For comparing period differences, we used the daily average value of a song feature as the dependent variable, recording periods as fixed effect, bird ID and day of recording were use as random effects (Fig. 2C). Considering that recordings categorized to the same period were not made at the same time across birds, we also performed a more conservative test: each bird was separately tested for effects of recording periods (Supplementary Figs. 2 and 3). In this case, we used recording period as fixed effect and day of recording as random effect. For group comparisons, we used the daily average of a song feature as dependent variable, group as fixed effect and individual bird as a random effect. We used two-sided Mann-Whitney Test (the “wilcox.test” function in R) to test whether HVC volume, RA volume, plasma androgen levels, and organ weights (brain, syringes, oviducts, and follicles) differ between singers (six long-term monitored females) and non-singers (nine non-singing females). To evaluate whether song features correlate with physiological measurements, we calculated the average of a song feature for each bird and calculated Pearson's correlation coefficients using the “cor” function in R. P values were calculated by the “cor.test” function, and further corrected for multiple comparisons using the “p.adjust” function (Bonferroni) in R.

3.2. The occurrence of spontaneous female singing is infrequent and variable between and within individuals We quantified song occurrence, defined as the percentage of days for which we observed female songs among all analyzed days (Supplementary Table 1). We found on average a 34% chance to observe singing females, while the individual differences and the within individual variability were large. For example, bird 4 sang regularly in two periods while bird 3 sang rarely throughout; in a total of 100 recorded days and 38 analyzed days, bird 3 sang only during one day (song occurrence 2.6%). Moreover, even on a “singing-day”, female canaries used < 1% of the 9-hour recording time for singing (Fig. 2A, Supplementary Tables 2, and 3). In contrast, the average song occurrence of breeding male canaries was 99% ± 2.7% (standard deviation (SD)). 3.3. Temporal and spectral characterization of spontaneous female canary songs Next, we quantified temporal and spectral parameters of spontaneous female canary songs. On the song-level, we analyzed three temporal features: the song length, the number of syllables in a song, the syllable repetition rate, and analyzed one spectral feature: the slope coefficient (α). On the syllable-level, we analyzed two temporal features: the syllable length, the inter-syllable silent pauses (inter-syllable intervals), and analyzed two spectral features: the peak frequency and the Wiener entropy (see Methods). All of the above-mentioned features, including syllable repetition rates (Fig. 2B), were statistically different between individuals (Supplementary Fig. 1, Supplementary Tables 2 and 3). Furthermore, the spontaneous female songs of each individual varied over time (Fig. 2C, Supplementary Figs. 2 and 3). For example, differences of syllable repetition rates between different recording periods were statistically meaningful (Fig. 2C) even when we estimated the effects using a more conservative approach by building a linear mixed effect model for each bird: recording periods had effects in four out of five birds with multiple recording periods (Supplementary Fig. 2). Thus, temporal features such

3. Results 3.1. A minority of female canaries sing spontaneously By monitoring the long-term song activity of a total of 112 female canaries housed in non-breeding condition, we observed that six (5%) birds spontaneously started to sing (Fig. 1). The singing females were

A

B

C

Fig. 2. Female canaries sing infrequently and singing activity and song features varies between and within individuals. A, Daily song rate of six spontaneously-singing female canaries showing that on average all birds spent less than 1% of the time singing. Each bird was color-coded. Each light grey dot indicated song rate of a given recorded day. B, Song characteristics differed between individuals of singing female canaries. Syllable repetition rate is shown; other song characteristics are shown in Supplementary Figure 1. Each light grey dot indicates the syllable repetition rate of a given recording day. C, Syllable repetition rate showed within individual differences over time. Other song characteristics are shown in Supplementary Figure 2 and Supplementary Figure 3. Daily average syllable repetition rates were color-coded for each female. For all panels, predicted estimates of the models were indicated with black dots and the 95% credible intervals (CrI) were visualized as vertical grey bars for all panels. If the predicted estimate of one group does not overlap with the 95% CrI of the other group, the comparison is statistically meaningful. See Supplementary Table 2 and Supplementary Table 3 for results of linear mixed effect models and posterior probability of mean differences, respectively. 5

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

1 4

● ●

3

●

0.15

●

5 2

● ●

●

●

0.10

●

0.20

0.15 61

●●

0.10

●

35

● ●

●● ●●

●

● ● ●

2

●

0.05

● ● ●

●

0.05

Singer Non−singer

0.2 0.1 0.05

●

5 14 2 6

● ●● ● ●

● ●● ●● ● ●●

3

720 +

690

+ +

2

++

1

64

5

Plasma androgens (ng/ml)

+

750

6 8 10 Syllable repetition rate (syl/s)

12

3

●

16

● ●

200

●

100

5

●

3

●

2

Non−singer

Singer

●

100

1 6 4

● ●

52

50

● ●

● ● ● ●● ● ●

● ● ●● ● ●

Non−singer

Singer

Non−singer

H 3000

y = 0.025 x - 0.59 r = 0.98 p = 4.78 × 10-4

2000

1000 5 1 ++ 2 + 4

+

y = -0.69 x + 156 r = -0.97 p = 9.35 × 10-4

150 +

0 4

4

0 Singer

**

300

G y = 11 x + 630 r = 0.99 p = 7.01 × 10-5

E

***

3

1

Singer Non−singer

F Brain weight (mg)

●

Follicle weight (mg)

4

●

D

** 3

3

Follicle weight (mg)

0.20

6

RA volume (mm3)

HVC volume (mm3)

●

C

*

Oviduct weight (mg)

B

**

Plasma androgens (ng/ml)

A

+

3

120 +

1 +

90

4

+

6

60

++ 2 5

6

50 100 Number of syllables per song

40

80 120 Inter-syllable interval (ms)

160

Fig. 3. HVC volume and plasma androgen levels of spontaneously-singing female canaries. A, HVC volume of singing and non-singing female canaries. The HVC volume was statistically different between the two groups (Mann-Whitney Test, U = 51, P value = 0.0028). B, RA volume of singing and non-singing female canaries. The RA volume was statistically different between the two groups (Mann-Whitney Test, U = 6, P value = 0.019). C, Plasma androgen levels on the day of sacrifice were statistically different between singing and non-singing females (Mann-Whitney Test, U = 50, P value = 0.0048). Bird 3 had a high androgen level, but removing bird 3 from the analysis did not change the results. D, Oviduct weight was statistically different between singing and non-singing females (Mann-Whitney Test, U = 54, P value = 4.0 × 10-4). E, Follicle weight was different between singing and non-singing females (Mann-Whitney Test, U = 36, P value = 0.0028). A-E: The boxes indicate the 25th/50th/75th percentiles (bottom/middle/top bar), the extent of the whiskers indicate the most extreme values that are within 1.5 times the IQR (inter-quartile range) of the hinge. * P value < 0.05; ** P value < 0.01; *** P value < 0.001. For singers, bird ids were indicated. F, The brain weights were positively correlated with the mean syllable repetition rate. Pearson's r and Bonferroni-adjusted P value are indicated. G, The plasma androgen levels were positively correlated with the mean number of syllables in a song. Pearson's r and Bonferroni-adjusted P value are indicated. However, the significance did not hold after removing bird 3 from the analysis. H, The follicle weights were negatively correlated with the mean inter-syllable interval. Pearson's r and Bonferroni-adjusted P value are indicated. See Supplementary Table 4 for all correlation comparison.

as song length and syllable repetition rate and the overall organization of female song changed over time.

weight and follicle weight were statistically different between singers and non-singers even after excluding bird 3. In contrast, brain weight (Supplementary Fig. 4A, singers 704 mg, non-singers 660 mg, Mann-Whitney U = 43, P value = 0.066, n = 15), body weight (Supplementary Fig. 4B, singers 20.5 g, non-singers 18.5 g, Mann-Whitney U = 42, P value = 0.088, n = 15) and syrinx weight (Supplementary Fig. 4C, singers 21.8 mg, non-singers 19.6 mg, MannWhitney U = 30, P value = 0.77, n = 15) did not differ significantly between singing and non-singing individuals.

3.4. HVC and RA volumes, plasma androgen levels and organ size of singing and non-singing females Artificially administered testosterone strongly affects song development, HVC volume, and RA volume in adult female canaries (Fusani et al., 2003; Madison et al., 2015; Nottebohm, 1980). We asked whether HVC volume, RA volume, the plasma androgen level and other physiological measurements of the spontaneously-singing females differed from those of non-singing female canaries kept under same conditions. HVC volume (Fig. 3A, singers 0.176 mm3 (mean), non-singers 0.0974 mm3, Mann-Whitney U = 51, P value = 2.8 × 10−3, n = 15), RA volume (Fig. 3B, singers 0.119 mm3, non-singers 0.0613 mm3, Mann-Whitney U = 6, P value = 0.019, n = 14), plasma androgen levels (Fig. 3C, singers 512 ng/ml, non-singers 36.7 ng/ml, MannWhitney U = 50, P value = 4.8 × 10−3, n = 15), oviduct weight (Fig. 3D, singers 194 mg, non-singers 18.5 mg, Mann-Whitney U = 54, P value = 4.0 × 10−4, n = 15), and follicle weight (Fig. 3E, singers 89.35 mg, non-singers 11.87 mg, Mann-Whitney U = 36, P value = 2.2 × 10−3, n = 15) of singing females were significantly different from those of non-singing birds. Bird 3 had an extremely high androgen level, but the androgen levels between singing and nonsinging birds remained statistically different when we excluded bird 3 from the analysis (singers 66.6 ng/ml, Mann-Whitney U = 41, P value = 0.012, n = 14). Similarly, HVC and RA volumes, the oviduct

3.5. Plasma androgen levels correlated with temporal song features We further investigated whether song features would be correlated with physiological features, such as brain weight, HVC volume, RA volume, and plasma androgen levels. We calculated the Pearson's correlation coefficients between song features measured in the last recording period and the HVC volume and plasma androgen levels on the day of sacrifice. We did not find a correlation between any of the song parameters and HVC volume or RA volume. Normalization of HVC and RA volume against body weight or brain weight did not influence these results. In contrast, the brain weight was positively correlated with the mean syllable repetition rate (Pearson's correlation coefficient r = 0.99, Bonferroni-adjusted P value = 0.0015, Fig. 3F, and Supplementary Table 4). The follicle weights were negatively correlated with the average inter-syllable interval (Pearson's correlation coefficient r = −0.97, Bonferroni-adjusted P value = 0.021, Fig. 3H, and Supplementary Table 4). The plasma androgen levels were positively 6

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

A

F

Bird7 (kHz)

8 6 4 2 0

2

4

6

B

(sec)

8

F+T

(kHz)

8 6 4 2 0

2

4

6

8

10

12

14

16

C

18

20

22

24

26

28

(sec)

M

(kHz)

8 6 4 2 8

10

●

●

● ●

15

● ●

5

● ● ● ● ●● ●●● ● ●●● ● ●●●● ● ● ● ● ●●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ●

●

F

● ● ●

●

● ●● ● ● ● ● ●

● ● ●●

●

10

●●

●●● ● ● ●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●

●

F+T

● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ●● ●●● ● ● ●● ●● ●● ● ● ● ● ● ● ●● ●

●

●●

M

Number of syllables per song

Song length (s)

300 20

● ● ●

● ●● ●

200

●

●

●● ●

●

● ● ● ●

● ● ● ●●

100

● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●●

●

0

F

● ● ● ● ●● ●● ●● ● ● ● ●● ● ●● ●● ● ●●● ●● ● ●

●

F+T

14

16

F

* * *

●

12

●● ● ● ● ● ● ● ● ● ●●● ●● ●● ● ● ●● ●● ● ● ●● ● ●● ● ● ●● ●● ● ● ● ● ●●● ● ● ●● ● ● ● ● ● ● ● ● ●● ●

●

M

18

20

22

G

* *

20

● ●

●

● ● ● ●● ● ● ●● ● ● ● ● ●●● ●● ● ● ● ● ● ●● ●●●● ●● ●● ●● ● ●●● ● ● ●● ●● ● ●● ●● ●●

●

15

● ●●● ● ●

● ● ● ● ●●● ● ● ● ● ● ●●

10

●●●● ● ● ●

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ●● ● ●● ● ●● ● ●●● ● ● ●●● ● ● ●●● ●

●

5

● ● ●

● ● ●● ●●● ● ● ●

● ● ●● ● ● ●

●

●

● ● ●

●

●●● ● ●● ●●

● ● ● ●● ● ●● ● ●● ● ● ● ● ●● ● ● ●

● ●

H ●

400

* *

F+T

M

* * ●

●

300

●● ●●●

200

● ●● ● ● ●

100

● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ●●● ● ● ● ●● ●● ●● ●●● ● ● ● ● ● ●● ● ●●● ● ● ●● ●● ●● ● ●

●

● ● ● ● ●● ●

●

● ●● ● ●● ●

● ●●● ● ●● ● ●● ● ● ● ● ●● ●● ● ● ●● ●● ● ● ● ● ●●●●

●

0

F

(sec)

24

F

F+T

● ●● ● ● ● ●● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●●

Syllable length (ms)

6

E

* *

25

4

Inter−syllable interval (ms)

D

2

Syllable Repetition rate (syl/s)

0

200 ●

● ● ●

150

● ●

100

● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●●● ● ●● ● ●● ● ● ● ● ●●●● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●

●

50

●

● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ●●● ●● ● ● ● ● ●

●

● ●● ● ● ● ● ● ● ●● ● ●

●

M

●

●

● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ●●● ● ● ●

●

●

F

F+T

M

Fig. 4. Spontaneous female songs were distinct from songs of testosterone-induced females and of breeding males. A, An example song spectrogram of a spontaneously-singing female canary, bird 7. B, An example song spectrogram of a testosterone-induced singing female canary. C, An example song spectrogram of a breeding male canary. Linear mixed effect models estimating group effects on the song length (D), the number of syllables per song (E), syllable repetition rate (F), inter-syllable interval (G), and syllable length (H). In D to H, light grey dots indicate the daily average of song features, black dots indicate predicted estimates of the models and vertical grey bars indicate the 95% credible intervals (CrI) of the predicted estimates. Comparisons are statistically meaningful if the predicted estimate of one group does not overlap with the 95% CrI of the other group. Comparisons are marked by * if the posterior probability is higher than 0.95. See Supplementary Table 2 and Supplementary Table 3 for results of linear mixed effect models and posterior probability of mean differences, respectively.

correlated with the average number of syllables per song (Pearson's correlation coefficient r = 0.98, Bonferroni-adjusted P value = 0.011, Fig. 3G, and Supplementary Table 4). However, after removing bird 3, whose plasma androgen concentration was extremely high, these correlations were no longer significant.

The song length of spontaneously-singing females (F) was shortest compared with testosterone-treated females (F + T) and males (M), and song length of testosterone-treated females was comparable to that of males (Fig. 4D: 5.2 (2.8) s (mean (SD), and here after), F + T: 8.2 (5.9) s, M: 8.0 (6.5) s). The number of syllables per song in all three groups were significantly different (Fig. 4E: 31 (18) syl, F + T: 70 (50) syl, M: 113 (53) syl). Syllable repetition rate, inter-syllable interval, syllable length (Fig. 4F–H), and slope coefficient α (Supplementary Fig. 5D) of the two female groups were not different to each other but both were different from those of male birds (syllable repetition rate: F: 7.0 (2.7) syl/s, F + T: 8.8 (4.1) syl/s, M: 14.6 (3.6) syl/s; inter-syllable interval: F: 113 (81) ms, F + T: 80 (41) ms, M: 32 (12) ms; syllable length: F: 75 (28) ms, F + T: 78 (39) ms, M: 26 (9) ms). Conversely, the two syllablelevel spectral features (peak frequency and Wiener entropy) of the three groups were similar. Overall, spontaneous female songs were different from testosterone-induced female songs and different from male songs. In addition, songs of testosterone-treated females were different from songs of male canaries.

3.6. Spontaneous female songs are distinct from testosterone-induced female songs and male songs Despite the low occurrence of spontaneous songs in female canaries, singing can be easily induced by testosterone treatment (Fusani et al., 2003; Madison et al., 2015; Nottebohm, 1980; Shoemaker, 1939; Vellema et al., 2019). To compare spontaneous female songs (Fig. 4A) with female songs induced by testosterone, we implanted six nonsinging females with testosterone for two weeks. Testosterone-treated females produced unstable songs four days after implantation, and their songs became male-like after 14 days (Fig. 4B). In addition to the six long-term monitored spontaneously-singing females, we obtained recordings of two additional spontaneously-singing females and of nine male canaries from the breeding seasons (Fig. 4C). They were included in this comparison. 7

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

4. Discussion

short transitioning phrases (Guttinger, 1985; Leitner et al., 2001; Madison et al., 2015; Markowitz et al., 2013). Such general organization of phrases was observed in songs of both spontaneously-singing females and testosterone-induced singing females (Figs. 1 and 4). The two syllable-level spectral features were comparable between all three canary groups (Supplementary Fig. 5G–H). However, sex differences exist in all temporal features we quantified and the song-level spectral features (Supplementary Fig. 5A–F). The average slope coefficients (α) of the female groups were close to −0.5, indicating that the sequencing of syllables (i.e., the intervals between syllables) was close to random (−0.5 < α < 0.5). In contrast, male syllable organization was more predictable (mean: −0.79), indicating that male songs were more stereotyped than female songs. Testosterone implantation might extend the song length of female songs but has no effect on the syllable repetition rate (Fig. 4D and F). The maximum syllable repetition rate observed in a spontaneouslysinging female canary was 14.6 syl/s, lower than observed in both breeding and non-breeding male canaries. The so-called sexy syllable of a male canary requires a minimum of 17 syl/s to attract a female (Amy et al., 2015; Vallet and Kreutzer, 1995). Due to the sporadic singing activity and the housing conditions we cannot speculate about the function of singing in female canaries. However, female attraction is unlikely. We did not report the number of different syllable types (i.e., syllable repertoire size) in this study. The spontaneous female songs range from structures with clear distinction between different syllable types to strings of undefined sounds. Precise categorization of syllable types in highly variable female songs is challenging by either manually scoring or computer-based clustering algorithms. This results in high amounts of syllable types which do not reflect the richness of the syllable repertoire. Instead of clustering syllable types, we chose to assess timing or sequencing aspects of the song using the slope coefficient (α). Since the calculation is based on the amplitude modulation of the song, we are not hindered by problems that can arise by attempting to categorize syllable types.

By screening the singing behavior of a large number of female canaries we showed that approximately 5% of the population in our breeding facilities sang spontaneously. Song occurrence and daily song rate were generally low (Supplementary Table 1 and Fig. 2A). Songs were very variable between individuals as well as within individuals over time, differing in terms of the syllable repetition rate, the number of syllables per song and the song length (Fig. 2, Supplementary Figs. 1 and 2). Plasma androgen levels, HVC volumes, and RA volumes of singing females were significantly different from that of non-singing females (Fig. 3). The song speed (syllable repetition rate, the number of syllables in a song, and the inter-syllable intervals) were positively correlated with the brain weight and plasma androgen levels, and negatively correlated with the follicle weight (Fig. 3). Spontaneous female songs were discernible from both songs produced by testosterone-induced singing females and by males in breeding condition. 4.1. A minority of female canaries sings spontaneously Approximately 5% of our female canaries produced song spontaneously. Few spontaneously-singing females have been observed in previous studies (Hartley et al., 1997; Herrick and Harris, 1957; Pesch and Güttinger, 1985; Shoemaker, 1939), but neither their singing activity nor their song structures have been quantified before. The low abundance of singing females within a population is comparable to other species of the northern hemisphere such as the song sparrow (Arcese et al., 1988), the indigo bunting and the rufous-sided towhee (Nolan, 1958). Another similarity between female canaries and song producing females of other species of the northern temperate zone is the rarity of song production. In our study, birds often stayed quiet for the whole day and if singing occurred, the daily song rates were lower than 1%, equivalent to approximately 5 min per day (Fig. 2). Such low song occurrence was already reported for females of the northern temperate zone, such as the Northern cardinals (Ritchison, 1986) and the whitecrowned sparrow (Baker et al., 1984; Baptista et al., 1993). However, in the tropics and the southern temperate zone, female song is much more prevalent (Odom et al., 2014; Slater and Mann, 2004; Voigt et al., 2006; Wickler and Seibt, 1980). The observation of song in Australo-Asian species, as well as ancestral state reconstructions demonstrates that female singing is widespread (Odom et al., 2014). This study suggests that females of the common ancestor of all modern songbirds could sing and provides evidence for the presence of female singing in the ancestral state. Thus, the low occurrence of female singing in the northern temperate zone is likely due to evolutionary loss of singing in multiple genera and species. The example of New World blackbirds illustrates that a composite of life-history traits, including monogamous mating systems, dispersed nesting, and sedentary behavior, was correlated with the loss of female song in this species (Odom et al., 2015; Price, 2009; Price et al., 2009). Although female singing is not lost in canaries, the low prevalence and occurrence imply factors that suppress female singing.

4.3. Mechanisms of female spontaneous singing What are the factors enabling some females to sing but not others? To answer this question, we compared eight physiological measurements between spontaneously-singing and non-singing females kept in social isolation during song monitoring. Plasma androgen concentrations, oviduct weights, and follicle weights were statistically different between the singing and non-singing groups (Fig. 3). Song production of female canaries was observed under non-breeding photoperiodic conditions (this study and (Pesch and Güttinger, 1985)). However, three of our six singing females showed enlarged reproductive tracts. In female birds, the ovarian tissue is the primary source of circulating testosterone (Johnson, 1986). Thus, the occurrences of female singing might depend on plasma testosterone levels and might be a by-product of active ovarian tissue. Because behaviorally active estrogens are likely derived from testosterone in the forebrain of the canary (Schlinger and Arnold, 1991) and because oviduct weight, follicle weight, and plasma androgen concentration were higher in these females, we would expect an increase in estrogen level. Unfortunately, we did not measure blood estrogen concentrations. Our spontaneously-singing females were kept in single-sex aviaries when the birds were not being monitored for song; they did not hear or see a male canary during the experiment. High levels of female-female competition for mates have been associated with female song occurrences (Langmore, 1998). It is possible that the lack of male contact and the fact that they were constantly housed under non-breeding condition (short-day length) might drive these female birds to be “ready for mating” all the time, thus explaining an increase in oviduct weight and plasma androgen level. Since we did not observe song in egg-laying females (i.e. those with large follicles and oviducts) in mixed-sex colonies (Gahr, unpublished data), we suggest

4.2. Song structure comparisons between spontaneously-singing females, testosterone-treated females and male canaries Because of its rare occurrence, the natural song of female canaries has not previously been analyzed in depth. The long-term observations in our study allowed comparisons of natural female songs not only to songs of testosterone-treated females but also to songs of male canaries in the breeding seasons. The song length of the analyzed males and testosterone-induced singing females was comparable to the data reported before (males: (Voigt and Leitner, 2008), testosterone-induced singing females: (Fusani et al., 2003)). The songs of domesticated male canaries are organized in long repetitive sequences of identical syllables called phrases or tours and 8

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

that under natural conditions female singing is facilitated by androgen production resulting from increased ovarian activity, but inhibited by social-sexual cues. To what extent social isolation contributes to singing in female canaries is currently unknown. However, social isolation per se might not be the major cause for singing, because the majority of females did not sing. Social isolation might promote female singing in the short- but not in the long-term because spontaneously-singing female canaries started singing approximately 4 days after social isolation and generally stopped singing after 10 to 14 days in social isolation (data not shown). In fact, social isolation may even suppress spontaneous singing in female canaries. The functions of female songs in other songbird species have primarily been implicated in direct social interactions (Krieg and Getty, 2016; Langmore, 1998), and thus our study may underestimate the natural occurrence of spontaneous singing in female canaries. Future studies employing backpack microphones (Gill et al., 2016), which allows birds to be free-moving and socially interact with each other would provide answers to why and when female canaries sing. For example, experiments designed testing various sex ratios in grouphousing aviaries under a natural light schedule would help to answer whether the occurrence of female singing in canaries is associated with increasing female-female competition or with disinhibition of male suppression. Plasma androgen levels were higher in female singers than in nonsingers, the plasma androgen level of one bird (bird 3) was even as high as breeding male canaries (Voigt and Leitner, 2008). Since that individual's syllable repetition rate was also the highest and was close to that of non-breeding males (Voigt and Leitner, 2008), the high level of plasma androgen of this bird might be biologically relevant. The statistical significance between singers and non-singers in terms of HVC and RA volumes, plasma androgen levels, as well as oviduct and follicle weights remained even after excluding this bird. Plasma androgen concentrations were positively correlated with the mean number of syllables in a song, however, the correlations collapsed after excluding bird 3 (Fig. 3). Thus, a threshold for androgen levels might exist for the singing phenotype but whether song performance positively correlates to androgen levels in female canaries needs to be investigated further. Testosterone generally increases song rate in male and female songbirds (for review (Alward et al., 2018; Gahr, 2014)), however, we did not find a correlation between plasma androgen levels and song rate of spontaneously-singing female canaries (Supplementary Table 4). Although such correlations were lacking in males of red-winged blackbirds (Harding et al., 1988; Johnsen, 1998), quantification of songs of more spontaneously-singing female canaries in the future would increase the power of the correlational analysis. Elevated testosterone levels have been shown to increase the volume of the song control nuclei HVC and RA in adult canaries (Madison et al., 2015; Nottebohm, 1980) and the syrinx weight in zebra finches and canaries (Bleisch et al., 1984; Luine et al., 1980; Wade and Buhlman, 2000). We observed statistical differences of HVC and RA volumes between female singers and non-singers (Fig. 3), but did not find differences in syrinx weights (Supplementary Fig. 4). However, testosterone-induced changes might occur on the molecular, neurochemical and ultrastructural level in the syrinx (Bleisch et al., 1984; Bleisch and Harrelson, 1989; Luine et al., 1980) as well as in the adult vocal control system (e.g. (Devoogd et al., 1985; DeVoogd et al., 1991)). Systemic testosterone implantation in female canaries rapidly (within 2 days) increased the volume of the medial preoptic nucleus (POM), a brain region outside of the song control system (Shevchouk et al., 2019). Moreover, local administration of testosterone or estradiol in POM triggered female singing in canaries, although the songs were short and the quality was poor (Vandries et al., 2019). Future studies need to verify if POM volume of spontaneously-singing females is increased as compared to non-singing females. Our data suggest that relatively small increases of androgen are sufficient to trigger singing and a change in HVC and RA morphology. A

more substantial increase of androgen production might lead to elevated syllable repetition rate, such as in bird 3 (Fusani et al., 2003; Leitner and Catchpole, 2007; Vellema et al., 2019). In our study, the large difference in androgen level did not appear to affect HVC and RA morphology, since the volumes of these areas in bird 3 were well within range of the other singing females. Thus, female singing might be a result of individual-specific genetic predisposition, such as the androgen sensitivity of HVC circuits, the production of ovarian testosterone, or the number of HVC neurons that can be differentiated into a functional circuit by testosterone and its estrogenic metabolites. Higher natural levels of testosterone likely lead to neuronal differentiation that differs from lower levels of testosterone. More observations of singing females will be necessary to get a more detailed insight in the mechanisms underlying song development in adult female canaries. 5. Conclusion Here we provide an in-depth description of spontaneous female canary song. While the overall organization of spontaneous female song was similar to the general organization of song observed in male canaries, the quality was highly variable and differed from that of males and testosterone-induced singing females. We showed that female singers had elevated androgen levels and a larger HVC and RA volume compared to female non-singers. Thus, the presence of spontaneous song in female canaries suggests a natural range of behavioural and neural plasticity in a northern temperate songbird species. Supplementary data to this article can be found online at https:// doi.org/10.1016/j.yhbeh.2019.104614. Author contributions M-CK and MV screened and recorded canaries. M-CK collected female canary blood samples, measured weights of organs and body, sectioned the brain, measured the HVC volume. VVM developed the Multi_Channel_Analyser (MCA) program. M-CK and VVM analyzed the data. M-CK visualized the data. M-CK, VVM, MV, and MG conceptualized the study. M-CK, VVM, MV, and MG wrote the manuscript. Funding This work was supported by Max-Planck-Gesellschaft. Availability of data and materials The raw data (audio wav files) have not been uploaded and made publicly available because they are extremely large files but are available from the corresponding author on reasonable request. The datasets generated and analyzed, and example audio files of the Figs. 1B, C, and 4A–C are available in the figshare repository (project ID 56282). The custom MATLAB sound analysis program, Multi_Channel_Analyser (MCA) is available in GitLab (https://doi.org/10.5281/zenodo. 1489098). Declaration of competing interest The authors declare that they have no competing interests. Acknowledgments We thank Monika Trappschuh and Wolfgang Goymann for hormone analyses, Stefan Leitner and Roswitha Brighton for maintaining the birds, Hans Rudolf Güttinger for providing song recordings of a spontaneously singing female canary and Willi Jensen for digitizing the tape recordings and for maintaining sound-boxes and the equipment. Furthermore, we express our appreciation to Nicolas M. Adreani, Glenn Cockburn, and Susanne Seltmann for the comments on previous 9

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al.

versions of this manuscript and to the International Max Planck Research School for Organismal Biology for training and support.

Behaviour 94, 254–278. Hahnloser, R.H.R., Kozhevnikov, A.A., Fee, M.S., 2002. An ultra-sparse code underliesthe generation of neural sequences in a songbird. Nature 419, 65–70. Halliday, H., 1948. Song of female chaffinch. Brit. Birds 41, 343–344. Harding, C.F., Walters, M.J., Collado, D., Sheridan, K., 1988. Hormonal specificity and activation of social behavior in male red-winged blackbirds. Horm. Behav. 22, 402–418. Hartley, R.S., Chinn, M.S., Ullrich, N.F.E., 1997. Left syringeal dominance in testosteronetreated female canaries. Neurobiol. Learn. Mem. 67, 248–253. Herrick, E.H., Harris, J.O., 1957. Singing female canaries. Science 125, 1299–1300. Hoelzel, A.R., 1986. Song characteristics and response to playback of male and female robins Erithacus-Rubecula. Ibis 128, 115–127. Johnsen, T.S., 1998. Behavioural correlates of testosterone and seasonal changes of steroids in red-winged blackbirds. Anim. Behav. 55, 957–965. Johnson, A.L., 1986. Reproduction in the female. In: Sturkie, P.D. (Ed.), Avian Physiology. Springer New York, New York, NY, pp. 403–431. Johnson, L.S., Kermott, L.H., 1990. Structure and context of female song in a northtemperate population of house wrens. Journal of Field Ornithology 61, 273–284. Korner-Nievergelt, F., Roth, T., Von Felten, S., Guélat, J., Almasi, B., Korner-Nievergelt, P., 2015. Bayesian Data Analysis in Ecology Using Linear Models with R, BUGS, and Stan. Academic Press. Krieg, C.A., Getty, T., 2016. Not just for males: females use song against male and female rivals in a temperate zone songbird. Anim. Behav. 113, 39–47. Kroodsma, D.E., Byers, B.E., 1991. The function(s) of bird song. Am. Zool. 31, 318–328. Kuschert, H., Ekelöf, O., 1981. Singende Weibchen des Teichrohrsängers (Acrocephalus scirpaceus). Vogelwelt 102, 29–30. Langmore, N.E., 1998. Functions of duet and solo songs of female birds. Trends Ecol. Evol. 13, 136–140. Leitner, S., Catchpole, C.K., 2007. Song and brain development in canaries raised under different conditions of acoustic and social isolation over two years. Developmental Neurobiology 67, 1478–1487. Leitner, S., Voigt, C., Gahr, M., 2001. Seasonal changes in the song pattern of the nondomesticated island canary (Serinus canaria) a field study. Behaviour 138, 885–904. Luine, V., Nottebohm, F., Harding, C., McEwen, B.S., 1980. Androgen affects cholinergic enzymes in syringeal motor neurons and muscle. Brain Res. 192, 89–107. Madison, F.N., Rouse Jr., M.L., Balthazart, J., Ball, G.F., 2015. Reversing song behavior phenotype: testosterone driven induction of singing and measures of song quality in adult male and female canaries (Serinus canaria). Gen. Comp. Endocrinol. 215, 61–75. Markowitz, J.E., Ivie, E., Kligler, L., Gardner, T.J., 2013. Long-range order in canary song. PLoS Comput. Biol. 9, e1003052. Nolan, V., 1958. Singing by female indigo bunting and rufous-sided towhee. The Wilson Bulletin 70, 287–288. Nottebohm, F., 1980. Testosterone triggers growth of brain vocal control nuclei in adult female canaries. Brain Res. 189, 429–436. Nottebohm, F., 1981. A brain for all seasons: cyclical anatomical changes in song control nuclei of the canary brain. Science 214, 1368. Nottebohm, F., Stokes, T.M., Leonard, C.M., 1976. Central control of song in the canary, Serinus canarius. J. Comp. Neurol. 165, 457–486. Odom, K.J., Hall, M.L., Riebel, K., Omland, K.E., Langmore, N.E., 2014. Female song is widespread and ancestral in songbirds. Nat. Commun. 5, 3379. Odom, K.J., Omland, K.E., Price, J.J., 2015. Differentiating the evolution of female song and male-female duets in the New World blackbirds: can tropical natural history traits explain duet evolution? Evolution 69, 839–847. Pavlova, D., Pinxten, R., Eens, M., 2005. Female song in European starlings: sex differences, complexity, and composition. Condor 107, 559–569. Pesch, A., Güttinger, H.-R., 1985. Der Gesang des weiblichen Kanarienvogels. J. Ornithol. 126, 108–110. Price, J.J., 2009. Evolution and life-history correlates of female song in the New World blackbirds. Behav. Ecol. 20, 967–977. Price, J.J., Lanyon, S.M., Omland, K.E., 2009. Losses of female song with changes from tropical to temperate breeding in the New World blackbirds. Proc. R. Soc. B Biol. Sci. 276, 1971–1980. Rasika, S., Nottebohm, F., Alvarez-Buylla, A., 1994. Testosterone increases the recruitment and/or survival of new high vocal center neurons in adult female canaries. Proc. Natl. Acad. Sci. 91, 7854–7858. RCoreTeam, 2015. R: a language and environment for statistical computing. http://www. R-project.org. Reichard, D.G., Brothers, D.E., George, S.E., Atwell, J.W., Ketterson, E.D., 2018. Female dark-eyed juncos Junco hyemalis thurberi produce male-like song in a territorial context during the early breeding season. J. Avian Biol. 49, jav-01566. Ritchison, G., 1983. The function of singing in female black-headed grosbeaks (Pheucticus melanocephalus): family-group maintenance. Auk 100, 105–116. Ritchison, G., 1986. The singing behavior of female northern cardinals. Condor 156–159. Rueden, C.T., Schindelin, J., Hiner, M.C., DeZonia, B.E., Walter, A.E., Arena, E.T., Eliceiri, K.W., 2017. ImageJ2: ImageJ for the next generation of scientific image data. BMC Bioinformatics 18, 529. Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., Preibisch, S., Rueden, C., Saalfeld, S., Schmid, B., Tinevez, J.-Y., White, D.J., Hartenstein, V., Eliceiri, K., Tomancak, P., Cardona, A., 2012. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676. Schlinger, B.A., Arnold, A.P., 1991. Brain is the major site of estrogen synthesis in a male songbird. Proc. Natl. Acad. Sci. U. S. A. 88, 4191–4194. Schmidt, M.F., Martin, W.J., 2014. Chapter 15 - the respiratory-vocal system of songbirds: Anatomy, physiology, and neural control. In: Holstege, G., Beers, C.M., Subramanian, H.H. (Eds.), Progress in Brain Research. Elsevier, pp. 297–335. Searcy, W.A., Andersson, M., 1986. Sexual selection and the evolution of song. Annu. Rev.

References Alward, B.A., Madison, F.N., Gravley, W.T., Ball, G.F., 2016a. Antagonism of syringeal androgen receptors reduces the quality of female-preferred male song in canaries. Anim. Behav. 119, 201–212. Alward, B.A., Madison, F.N., Parker, S.E., Balthazart, J., Ball, G.F., 2016b. Pleiotropic control by testosterone of a learned vocal behavior and its underlying neuroplasticity. eneuro 3, 1–17. Alward, B.A., Cornil, C.A., Balthazart, J., Ball, G.F., 2018. The regulation of birdsong by testosterone: multiple time-scales and multiple sites of action. Horm. Behav. 104, 32–40. Amy, M., Salvin, P., Naguib, M., Leboucher, G., 2015. Female signalling to male song in the domestic canary, Serinus canaria. R. Soc. Open Sci. 2. Arcese, P., Stoddard, P.K., Hiebert, S.M., 1988. The form and function of song in female song sparrows. Condor 90, 44–50. Baker, M.C., Bottjer, S.W., Arnold, A.P., 1984. Sexual dimorphism and lack of seasonal changes in vocal control regions of the white-crowned sparrow brain. Brain Res. 295, 85–89. Bakus, G.J., 1959. Observations on the life history of the dipper in Montana. Auk 76, 190–207. Balthazart, J., Charlier, T.D., Barker, J.M., Yamamura, T., Ball, G.F., 2010. Sex steroidinduced neuroplasticity and behavioral activation in birds. Eur. J. Neurosci. 32, 2116–2132. Baptista, L.F., Trail, P.W., DeWolfe, B.B., Morton, M.L., 1993. Singing and its functions in female white-crowned sparrows. Anim. Behav. 46, 511–524. Bates, D., Mächler, M., Bolker, B., Walker, S., 2014. Fitting Linear Mixed-Effects Models Using lme4. arXiv preprint arXiv:1406.5823. Beletsky, L.D., 1983. Aggressive and pair bond maintenance songs of female red-winged blackbirds (Agelaius, Phoeniceus). Zeitschrift Fur Tierpsychologie-Journal of Comparative Ethology 62, 47–54. Bleisch, W.V., Harrelson, A., 1989. Androgens modulate endplate size and ACh receptor density at synapses in rat levator ani muscle. J. Neurobiol. 20, 189–202. Bleisch, W., Luine, V.N., Nottebohm, F., 1984. Modification of synapses in androgensensitive muscle. I. Hormonal regulation of acetylcholine receptor number in the songbird syrinx. J. Neurosci. 4, 786–792. Brenowitz, E.A., Lent, K., 2002. Act locally and think globally: intracerebral testosterone implants induce seasonal-like growth of adult avian song control circuits. Proc. Natl. Acad. Sci. 99, 12421–12426. Catchpole, C.J.B., Slater, P., 1995. Bird Song: Biological Themes and Variations. Chen, Z., Ye, R., Goldman, S.A., 2013. Testosterone modulation of angiogenesis and neurogenesis in the adult songbird brain. Neuroscience 239, 139–148. Delius, J.D., 1965. A population study of skylarks Alauda Arvensis. Ibis 107, 466–492. Devoogd, T.J., Nixdorf, B., Nottebohm, F., 1985. Synaptogenesis and changes in synaptic morphology related to acquisition of a new behavior. Brain Res. 329, 304–308. DeVoogd, T.J., Pyskaty, D.J., Nottebohm, F., 1991. Lateral asymmetries and testosteroneinduced changes in the gross morphology of the hypoglossal nucleus in adult canaries. J. Comp. Neurol. 307, 65–76. Frankl-Vilches, C., Kuhl, H., Werber, M., Klages, S., Kerick, M., Bakker, A., de Oliveira, E.H., Reusch, C., Capuano, F., Vowinckel, J., Leitner, S., Ralser, M., Timmermann, B., Gahr, M., 2015. Using the canary genome to decipher the evolution of hormonesensitive gene regulation in seasonal singing birds. Genome Biol. 16, 19. Fusani, L., Metzdorf, R., Hutchison, J., Gahr, M., 2003. Aromatase inhibition affects testosterone-induced masculinization of song and the neural song system in female canaries. J. Neurobiol. 54, 370–379. Gahr, M., 1990. Delineation of a brain nucleus: comparisons of cytochemical, hodological, and cytoarchitectural views of the song control nucleus HVc of the adult canary. J. Comp. Neurol. 294, 30–36. Gahr, M., 2004. Hormone-dependent neural plasticity in the juvenile and adult song system: what makes a successful male? Ann. N. Y. Acad. Sci. 1016, 684–703. Gahr, M., 2014. How hormone-sensitive are bird songs and what are the underlying mechanisms? Acta Acustica United with Acustica 100, 705–718. Gardner, M., 1978. White and brown music, fractal curves and one-over-f fluctuations. Sci. Am. 238, 16–33. Gelman, A., Hill, J., 2006. Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge university press. Gelman, A., Su, Y.-S., 2018. Arm: Data Analysis Using Regression and Multilevel/ Hierarchical Models, 110–1 ed. (pp. R package version 1.10-11). Gilden, D.L., Thornton, T., Mallon, M.W., 1995. 1/f noise in human cognition. Science 267, 1837–1839. Gill, F., Donsker, D., 2015. IOC World Bird List (v5.4). Gill, L.F., D’Amelio, P.B., Adreani, N.M., Sagunsky, H., Gahr, M.C., Maat, A., 2016. A minimum-impact, flexible tool to study vocal communication of small animals with precise individual-level resolution. Methods Ecol. Evol. 7, 1349–1358. Gisiger, T., 2001. Scale invariance in biology: coincidence or footprint of a universal mechanism? Biol. Rev. 76, 161–209. Goymann, W., Möstl, E., Gwinner, E., 2002. Non-invasive methods to measure androgen metabolites in excrements of European stonechats, Saxicola torquata rubicola. Gen. Comp. Endocrinol. 129, 80–87. Griffiths, R., Double, M.C., Orr, K., Dawson, R.J.G., 1998. A DNA test to sex most birds. Mol. Ecol. 7, 1071–1075. Guttinger, H., 1985. Consequences of domestication on the song structures in the canary.

10

Hormones and Behavior 117 (2020) 104614

M.-C. Ko, et al. Ecol. Syst. 17, 507–533. Shevchouk, O.T., Ball, G.F., Cornil, C.A., Balthazart, J., 2019. Rapid testosterone-induced growth of the medial preoptic nucleus in male canaries. Physiol. Behav. 204, 20–26. Shoemaker, H.H., 1939. Effect of testosterone propionate on behavior of the female canary. Proc. Soc. Exp. Biol. Med. 41, 299–302. Singh, N.C., Theunissen, F.E., 2003. Modulation spectra of natural sounds and ethological theories of auditory processing. J Acoust Soc Am 114, 3394–3411. Slater, P.J.B., Mann, N.I., 2004. Why do the females of many bird species sing in the tropics? J. Avian Biol. 35, 289–294. Tchernichovski, O., Nottebohm, F., Ho, C.E., Pesaran, B., Mitra, P.P., 2000. A procedure for an automated measurement of song similarity. Anim. Behav. 59, 1167–1176. Vallet, E., Kreutzer, M., 1995. Female canaries are sexually responsive to special song phrases. Anim. Behav. 49, 1603–1610. Vallet, E., Kreutzer, M., Gahr, M., 1996. Testosterone induces sexual release quality in the song of female canaries. Ethology 102, 617–628. Vandries, L.M., Ghorbanpoor, S., Cornez, G., Shevchouk, O.T., Ball, G.F., Cornil, C.A., Balthazart, J., 2019. Testosterone or estradiol when implanted in the medial preoptic nucleus trigger short low-amplitude songs in female canaries. eNeuro 6. Vellema, M., Rocha, M., Bascones, S., Zsebők, S., Dreier, J., Leitner, S., Van der Linden, A., Brewer, J., Gahr, M., 2019. Accelerated redevelopment of vocal skills is preceded by lasting reorganization of the song motor circuitry. eLife 8, e43194. Voigt, C., Leitner, S., 2008. Seasonality in song behaviour revisited: seasonal and annual variants and invariants in the song of the domesticated canary (Serinus canaria). Horm. Behav. 54, 373–378.

Voigt, C., Leitner, S., Gahr, M., 2006. Repertoire and structure of duet and solo songs in cooperatively breeding white-browed sparrow weavers. Behaviour 143, 159–182. Voss, R.F., Clarke, J., 1975. ‘1/f noise’ in music and speech. Nature 258, 317–318. Wade, J., Buhlman, L., 2000. Lateralization and effects of adult androgen in a sexually dimorphic neuromuscular system controlling song in zebra finches. J. Comp. Neurol. 426, 154–164. Webb, W.H., Brunton, D.H., Aguirre, J.D., Thomas, D.B., Valcu, M., Dale, J., 2016. Female song occurs in songbirds with more elaborate female coloration and reduced sexual dichromatism. Front. Ecol. Evol. 4. Wickler, W., Seibt, U., 1980. Vocal dueting and the pair bond II. Unisono dueting of the African forest weaver, symplectes bicolor. Z. Tierpsychol. 52, 217–226. Wild, J.M., 2004. Functional neuroanatomy of the sensorimotor control of singing. Ann. N. Y. Acad. Sci. 1016, 438–462. Wingfield, J.C., Smith, J.P., Farner, D.S., 1982. Endocrine responses of white-crowned sparrows to environmental stress. Condor 84, 399–409. Yamaguchi, A., 1998. A sexually dimorphic learned birdsong in the northern cardinal. Condor 100, 504–511. Yamaguchi, A., 2001. Sex differences in vocal learning in birds. Nature 411, 257–258. Yang, M., De Coensel, B., Kang, J., 2015. Presence of 1/f noise in the temporal structure of psychoacoustic parameters of natural and urban sounds. J Acoust Soc Am 138, 916–927. Yasukawa, K., Boley, R.A., Simon, S.E., 1987. Seasonal change in the vocal behavior of female red-winged blackbirds, Agelaius-Phoeniceus. Anim. Behav. 35, 1416–1423.

11