Auditory processing in the zebra finch midbrain: single unit responses and effect of rearing experience

Animals and ethical considerations

The experimental work was carried out in compliance with and approval from the University of Auckland Animal Ethics Committee (R425) in accordance with the University of Auckland Code of Ethical Conduct for the Use of Animals for Teaching and Research, the Animal Welfare Act 1999 (New Zealand), and The National Animal Ethics Advisory Committee (NAEAC) Good Practice Guide for the Use of Animals in Research, Testing and Teaching. Birds were provided with ad-lib water and commercial dry finch-mix seeds (Animates™), spray millet and additional weekly treats (fresh grasses, corn boiled egg, and rice). Birds were also supplied with calcium supplements and grit.

Normal reared zebra finches

Six adult zebra finches (2 males and 4 females >100 days old) that were raised having been exposed only to their conspecifics were used in this study. Birds were bred in an aviary at the Department of Psychology, University of Auckland (Campbell & Hauber, 2010). Light was provided via a compact Arcadia fluorescent bird lamp (2.4% UVB and 12% UVA) which imitated outdoor conditions. The mean temperature in the aviary was kept at 21 ± 3 °C, daily humidity between 35% and 65% and care was taken to keep a constant airflow. Pairs of zebra finches (Taeniopygia guttata) were housed together in indoor cages and once the chicks reached adulthood (>100 days), they were transferred to a holding indoor aviary (2 m × 1 m × 2 m) in the Faculty of Medical and Health Sciences. At all times, birds were held at a constant light/dark cycle of 14 h/10 h in the aviary and birds were provided with food and water ad lib. Thus, the ZF-ZF group had been only exposed to conspecific song.

Cross fostering

Six male zebra finches (>100 days old) that were raised by Bengalese finches (ZF-CF) were used. These zebra finches had previously been used in behavioural experiments and a full description of the rearing conditions is available in Campbell & Hauber (2009). Briefly, zebra finch chicks (3–5 days old) were transferred into the nest of a Bengalese finch pair in one of two outdoor aviaries in Auckland (2 m × 1 m × 2 m), and exposed to natural photoperiod and weather conditions. The chicks were therefore reared hearing male and female Bengalese finches and tutored by a Bengalese finch male, but were also exposed to wild birds species found commonly in suburban Auckland (Blackbird Turdus merula; House Sparrow Passer domesticus; Song Thrush Turdus philomelos; Starling Sturnus vulgaris; and Tui Prosthermadera novaeseelandiae). After reaching adulthood (>100 days), the cross-fostered birds were transferred into single-sex aviaries where they were housed with fostered conspecifics. After the behavioural experiments were completed, the birds were finally transferred to an indoor aviary (2 m × 1 m × 2 m) at the Faculty of Medical and Health Sciences, co-housed with the group of normal reared zebra finches. At all times, birds were held at a constant light/dark cycle of 14 h/10 h in the aviary and birds were provided with food and water ad lib. These birds did not hear normal conspecific vocalisations before behavioural testing, at which time their songs were crystallised.

Surgery

Birds were deeply anesthetised with an intramuscular injection of a mixture consisting of 55 mg/kg of ketamine (Parnell Laboratories, Auckland, New Zealand) and 11 mg/kg Xylazine (Rompun, Bayer) delivered in a volume of 0.04–0.05 ml. Additional small doses of anaesthetic were given as required during the course of the experiment. The head was fixed in a custom-made stereotaxic apparatus (Herb Adams, Los Angeles, CA, USA) with ear and beak bars and the head was tilted down 28° from the vertical axis. The head skin was reflected from the midline, the skull surface cleaned and a small metal plate was fixed to the skull using dental cement (Land Dental Manufacturing Co., Inc., Wheeling, IL, USA). This allowed the head to be held at a fixed angle during experimentation after removing the ear bars to present auditory stimulation. The coordinates for electrode insertion were measured with respect to the bifurcation of the mid-sagittal sinus (Y sinus). A small opening in the skull and dura mater was made over the chosen coordinates to allow electrode penetration into the midbrain.

Stimulus generation and presentation

Auditory stimuli consisted of pure tones, white noise, conspecific and heterospecific songs. Both white noise and tones were generated using Adobe Audition 3.0 software taking into account described hearing ranges for zebra finches (see Okanoya & Dooling, 1987). Pure tone stimuli (1, 2, 3, 4, 5 and 6 kHz ) were of 1 s duration with built-in rise and fall ramps of 150 ms at onset and offset to avoid stimulus clicking through the speaker. The white noise (WN) stimulus was also 1 s in duration and contained frequencies ranging from 0 to 10 kHz. Noise and tone stimuli were presented at 75 dB SPL, as measured with an SPL metre (RadioShack) 30 cm from a free-field speaker.

The songs used were conspecific (zebra finch) and heterospecific (Bengalese finch) songs. The conspecific songs were chosen from birds raised in the same aviary as the birds used in this experiment and were therefore familiar to both ZF-ZF and ZF-CF birds. They were obtained by recording vocalizations of each individual bird using Sound Analysis Pro software. They were then bandpass filtered (300 Hz–12 kHz) using Raven 1.3 software. Heterospecific songs were kindly provided by Dr. Sarah Woolley from Columbia University, New York. Song lengths ranged between 1.816 s and 2.691 s for the conspecific songs and 1.940 s to 3.091 s for the Bengalese finch songs. All songs were calibrated to be presented with an average power of 75 dB (as calibrated using a ½-inch free-field microphone, Brüel and Kjaer and Raven 1.3 software). Songs were played in the forward and backward directions.

Each stimulus (white noise, tones and songs) was constructed so as to contain 500 ms of silence preceding and following the actual stimulus. Each stimulus was presented to the bird 15 consecutive times, with different stimuli presented in a randomised order. The interval between the 15 consecutive presentations of a single stimulus type was kept constant within a sequence, but these intervals were randomised for each separate sequence (between 0.5 and 1.5 s, 0.1 s steps). Each unit therefore received 19 stimuli: white noise, 6 pure tones and 12 songs (3 conspecific songs in their forward and reverse directions and 3 heterospecific songs in their forward and reverse directions). At the end of the presentation of each set of stimuli, 30 s of baseline activity for each unit was also recorded at least 2 min after the end of the last stimulus presentation, and this activity was used to calculate the spontaneous firing rate.

Stimuli were presented to the bird through a multifunction processor (TDT System 3; Tucker-Davis Technology, Alachua, FL, USA), routed through an attenuator (TDT) that allowed equal intensity presentation of all stimuli at 75 dB SPL average power. The signal was then processed through a TDT Stereo Amplifier connected to a free-field magnetic loudspeaker (TDT) placed 30 cm in front of the bird. Before an experiment, the output of the loudspeaker was checked using a RadioShack SPL metre so to ensure constancy of the average power at which songs were delivered.

Electrophysiological recordings

All recordings were made in a sound attenuation chamber (Microbooth, All-DUCT Fabrication, PTY. Ltd, Melbourne, VIC, Australia), using tungsten or platinum backed electrodes (FHC Inc., Bowdoin, ME, USA; impedances ~10 MΩ at 1 kHz). The location of MLd was determined using a variety of search stimuli (white noise, hand claps, clicks, vocalizations), so as to avoid any bias that could result from missing cells responsive to a narrow range of stimuli. Tones were not used as search stimuli, so the data set may contain an underrepresentation of units responsive only to specific tones. The recorded signal was filtered (300 Hz low/5 kHz high pass) and amplified (gain 100×) using an AC amplifier (A-M systems Model 1800) and digitised using the TDT multifunction processor RX6. Threshold and spike discrimination were achieved in real time using the OpenController interface of the TDT OpenEx Suite. Raster plots and peristimulus-time histograms (PSTH) were visualised in real-time using the OpenScope interface, for a rapid qualitative assessment of the auditory selectivity of isolated units prior to the stimulus presentation protocol. The sequence of song presentations and unit recordings were made using a custom-made programme designed using the TDT RPvdsEx control (P. Logerot). During the experiment, individual stimuli were chosen from a prebuilt collection of stimuli through the OpenController interface, which also allowed us to set the number of consecutive presentations of the stimulus (15 for this study), the inter-stimulus interval and the attenuation. The order in which stimuli were presented was determined randomly for each series of presentations. The epoch store tracked the onset of each stimulus repeat within a stimulation sequence, and this was then used to align sequential presentations to construct the peristimulus-time histograms and raster plots.

Electrolytic lesions (40 μA, DC for 10 s) were made to identify and reconstruct recording site locations but, unfortunately, these proved frequently too large to assess the sites with precision. They could, however, be confidently used to confirm the recording site within MLd.

Analysis

Units were analysed off-line using TDT’s OpenSorter and OpenExplorer packages of the OpenEx suite. OpenSorter was used to manually eliminate outlier spikes and to confirm that recordings represented single units. When in doubt, the data were not included in the final analysis. The spontaneous firing rate of the unit (spikes/s) was calculated using the first 20 s of the 30 s baseline activity recorded at the end of the session.

Units were classified as ‘auditory’ when they reached criterion for at least one of the stimuli (whether WN, a tone or a song). Post-stimulus histograms (PSTHs) of 20 ms bins were constructed using TDT Openexplorer. Neurons were considered to reach the criterion if the responses within at least one 20 ms bin were above threshold, with threshold set at the mean +5 S.D. of the baseline firing rate (defined as the firing rate during the 500 ms of silence preceding the sound stimulus) (Prather et al., 2009). This choice of threshold biases the population used for analysis, since units that show an inhibition as a result of the auditory stimulus would not be included in the analysis, unless they reached criterion when presented with another stimulus. Once a unit met the criterion for at least one stimulus, the responses to all of the stimuli were considered for the analysis.

Generalized linear mixed model

We used a Generalized Linear Mixed Model (GLMM) in SAS 9.3 and SPSS 26 for the analysis of the data. This analysis took into account the lack of independence of the individual data points from multiple neurones recorded from different animals, which otherwise may have led to pseudoreplication (for reviews on these issues, see Lazic, 2010; Nakagawa & Hauber, 2011). The analysis took into account the following: (1) That each isolated single unit in this experiment was presented with 19 different stimulations each repeated 15 times, that is the recording session presented 285 stimuli; and (2) that several units were recorded from a single animal.

The data were first tested for the normality of the residuals using the Shapiro–Wilk test and then natural log transformed in an attempt to rectify their non-normal distribution. Homogeneity of variance for the residuals was tested using the Levene test and it was confirmed that this assumption was satisfied. The Levene test was 0.13 before transformation and 0.42 after a natural log transformation; therefore, the null hypothesis of homogeneity of variance was retained.

In order to investigate factors influencing the response of units within birds, a GLMM was fitted with bird and unit within bird as random effects to allow for the clustering of units within birds. Bird category was a fixed, between group factor, and song ID, conspecific/heterospecific (CON/HET) and song direction (i.e. song played either forwards or backwards; FOR/REV) were designated as repeated measures using an unstructured covariance matrix model. The GLMM was conducted using a restricted maximal likelihood estimation (REML) procedure and the Kenwood-Rogers approximation for the degrees of freedom (Gurka & Edwards, 2007; McCulloch, Searle & Neuhaus, 2008). The best covariance matrix model (i.e. unstructured) was chosen on the basis of the smallest Akaike Information Criterion (Gurka & Edwards, 2007; McCulloch, Searle & Neuhaus, 2008; Field, 2011). The interactions of bird category, song type, CON/HET and song direction, were initially included as explanatory variables.

The general equation (in matrix notation) for the GLMM analysis used was: $$y=X\beta +Zu+\epsilon$$in which y = an N × 1 column vector of outcome variables; X = an N × p matrix of p predictor variables (X included Bird Category, Song type (CON/HET), direction (FOR/REV), Song ID and Bird Category × CON/HET); β = a p × 1 column vector of fixed effects regression coefficients; Z = an N × q matrix of the q random effects related to a fixed β (Z included factors Bird ID and unit within bird); u is a q × 1 vector of random effects; and ε = an N × 1 column vector of residuals (Gurka & Edwards, 2007; McCulloch, Searle & Neuhaus, 2008). Initially, the full factorial model was fitted but then non-significant interactions were systematically omitted in search of the best model for the data, as indicated by the lowest AIC (Rouder et al., 2016). This was a model using unstructured covariance, excluding all interactions except for Bird Category × CON/HET. As with regression, it is appropriate to remove non-significant terms in search of the best model for the data, providing that some independent criterion such as the AIC is used as a guide (Gurka & Edwards, 2007; McCulloch, Searle & Neuhaus, 2008; Rouder et al., 2016).

Post-hoc comparisons were conducted using the least square means estimates from the GLMM (Field, 2011). Mann–Whitney U tests were used to compare the distributions and medians of neuronal responses (Field, 2011).

In normal reared zebra finches we found that for both sexes there was evidence of a greater response to CON than HET songs (males P = 0.001, females P = 0.0003) and that the size of this difference was greater in males than females. Therefore, sex was not introduced in the analysis, since it is unlikely that the uneven distribution of sex between the two groups would influence our results. This is consistent with behavioural data where sex was not found to be a factor in similar rearing experiments (Campbell & Hauber, 2009).

Spontaneous activity and responses to tones and white noise

Most units in the MLd of ZF-ZF and ZF-CF showed no spontaneous activity (<1 spike/sec) or showed low spontaneous activity (<5 spikes/s) (66% in ZF-ZF, 75% in ZF-CF) (Fig. 2). Units with higher spontaneous rates were found in ZF-ZF units, but not in ZF-CF midbrain units. As high spontaneous rate units were infrequent in the ZF-ZF birds, our failure to find units with strong discharge rates in cross fostered zebra finches may be related to the smaller number of units recorded.

Unit responses to white noise (WN) and tones are summarised in Table 1. Most ZF-ZF units and all ZF-CF units responded to at least one of the simple stimuli. Unit responses showed mostly excitation, although inhibition was also seen (6 units in ZF-ZF and 3 units in ZF-CF). Most units showed responses to WN and at least one of the tone stimuli, and most responded to more than one frequency, although 4 units in the ZF-ZF MLd failed to respond to any of the tones.

We derived the neurone’s best frequency (BF) from the response curves of each unit when excited by the presentation of tones at 75 dB SPL. Most units were mainly tuned to lower frequencies, 2 kHz being the most represented best frequency, followed by 1 and 3 kHz (Table 2).

The temporal response patterns of each unit are based on a qualitative assessment of the shape of the PSTH obtained in response to BF and to WN and are summarised in Fig. 3E. Units were classified as onset, sustained, primary-like or primary-like with notch (Figs. 3A–3D). Not all units could be confidently classified into any of these 4 categories based on stimulation with BF or WN. Units showing a sustained response to BF and WN were the most frequently found in both ZF-ZF and ZF-CF.

Tuning properties and response patterns to tones and WN of MLd units were comparable in the midbrains of ZF-ZF and ZF-CF birds. Thus, our qualitative analysis failed to detect an effect of rearing on the basic tuning properties of MLd neurones to simple stimuli beyond that which may be accounted for by differences in sample size.

Effect of learning in responses to song

Behavioural data have shown that normal-reared zebra finches prefer to spatially associate with conspecific song over Bengalese finch song, whereas zebra finches that were cross fostered by Bengalese finches associated equally with both types of song. Using birds from the same population used for these behavioural experiments, we sought to examine the extent to which responses of auditory units in the auditory midbrain mirrored these behavioural differences. We challenged single units in the MLd of normal reared (ZF-ZF) and cross fostered (ZF-CF) zebra finches with 12 song stimuli: 3 conspecific (CON) songs and 3 heterospecific, Bengalese finch (HET) songs. All song stimuli were presented in the forward (FOR) and reverse (REV) direction. Of all the units that were classified as auditory in the present study most (ZF-ZF: 79%; ZF-CF: 75%) responded to all types of stimuli (WN, tones, and songs). A minority of neurones responded only to song stimuli (Table 3). Those units where WN failed to elicit a response did respond to both tones and song stimuli. Most units were excited by the stimuli, although inhibition was also seen. One unit in a ZF-CF bird was found to be inhibited by WN, 1 kHz, 2 kHz, and 3 kHz stimuli, but excited by 5 kHz and a few song stimuli.

Both conspecific (CON) and heterospecific (HET) song elicited robust responses from MLd units in ZF-ZF and ZF-CF zebra finches. Thirty-four of the 58 units of ZF-ZF (59%) and 15 of the 20 units of ZF-CF (75%) responded to all 12 song stimuli. The remaining 24 units (41%) did not respond to at least one of the song stimuli (Fig. 4). Only 10% of the ZF-ZF and 15% of the ZF-CF units responded to 8 songs or less. There was no indication of any bias towards conspecific song or forward version of the song at this level of analysis.

Response patterns of MLd units of ZF-ZF and ZF-CF birds, were heterogenous, with many units showing strong responses throughout the song stimulus presentation. Qualitatively, responses appeared to be mainly isomorphic, following the AM envelope of the conspecific or heterospecific stimulus (Fig. 5A) with fewer units responding in a more sustained manner throughout the entire song stimulus (Fig. 5B). Other unit responses appeared to show some degree of selectivity in their responses, either by responding more selectively to some elements within songs or by responding much more vigorously to some songs in particular (Figs. 5C and 5D).

The type of response to one stimulus was not, however, a good predictor of the unit’s response to another one. For example Fig. 6 shows the responses of a ZF-CF MLd unit which is inhibited by white noise (Fig. 6A), and conspecific song 2 (CON-2) but increases its firing rate in response to heterospecific song 2 (HET-2) (See also Supplemental Table). Note that the increased response at the onset of CON-2 corresponds to a long silent period within the stimulus. Some other units would only respond to a subset of songs.

Determining what aspects of the song elicit these different types of responses is beyond the scope of this study. These results underscore the complexity of the neuronal responses encountered in this study, and show that the population of units in this study was not homogeneous with respect to response characteristics.

d′ and SI

To compare the responses of units to pairs of stimuli we calculated the selectivity index (SI) and d′. In the present study, a d′_A→B > 2 indicates a neurone’s stronger response to stimulus A over stimulus B while a d′_A→B < −2 indicates a neurone’s stronger response to stimulus B over stimulus A. A value of d′ falling between [−2, 2] (non-selectivity zone) indicates no preference of one stimulus over the other (see “Material and Methods”). We plotted the cumulative distribution of the d′ values for a given song against its reverse version and the other 5 forward songs for all instances in normal reared birds (58 units × 5 −1 comparisons) and for all instances in cross-fostered zebra finches (20 units × 5 comparisons).

Figure 7 shows these distributions of d′ for all instances when conspecific song 2 (CON-2, in ZF-ZF MLd, Fig. 7A and ZF-CF MLd, Fig. 7D) and heterospecific song 1 (HET-1, in ZF-ZF MLd, Fig. 7C and in ZF-CF MLd, Fig. 7B) were compared against their respective reverse version and all other forward versions of songs. Most comparisons fall within the non-selectivity interval, but many comparisons are found outside of this interval. In the case of CON-2, for those instances showing preference, most prefer CON-2 over its reverse or other forward songs (d′ values > 2). As in the ZF-ZF, in cross fostered zebra finches the cumulative distributions of d′ values of each song against all others for all instances suggest that, at the population level, there is no preference for an individual song stimulus with most comparisons falling within the non-selectivity zone (Fig. 7B).

At the unit level, all but 2 units from ZF-ZF and 1 in ZF-CF MLds showed discrimination to at least one pair of stimuli (d′ outside the non-selectivity interval, Fig. 8). A unit’s discrimination between a pair of stimuli could not predict discrimination response to a second pair.

Taken together, these results show that the majority of units in the MLd of both ZF-ZF and ZF-CF are able to discriminate at least one feature within the stimulus set.

Response strength

The response strength (RS) measures the amount of activity within the total firing rate that can be directly attributed to the presentation of the stimulus. The RS to each song stimulus of the MLd units recorded in this study ranged from −3.58 to 41.37 spikes/s in ZF-ZF and −0.28 to 17.03 spikes/s in ZF-CF zebra finches, and there was no tail towards high values of RS in ZF-CF zebra finches (Fig. 9). In the ZF-ZF auditory midbrain, other than the peak at 0, the majority of instances were confined between 0 and 5 spikes/s whereas in ZF-CF higher RS values appear to fall less sharply (Fig. 9).

Thus, MLd units from cross-fostered zebra finches appeared to have lower RS values and a narrower range of RS values than those in normal reared birds, although formal statistical analysis indicated that the mean difference was not statistically significant (F_1,15 = 0.33, P = 0.58). The RS values obtained for the forward and reverse presentation of a given song stimulus in a single unit are plotted against each other in Fig. 10. In 59% of the instances (here 58 units × 6 −1 FOR/REV song comparisons), slightly greater responses of the units toward the reverse version of songs were observed, which was confirmed by GLMM analysis (F_1,467 = 16.87, P = 0.0001).

In order to investigate factors influencing the response strength of units within birds, a GLMM was fitted with bird and unit within bird as random effects to allow for the clustering of units within birds, and song and direction (i.e. song played either forwards or backwards) as repeated measures with unstructured covariance. Bird Category (reared with conspecifics or cross-tutored), song type (conspecific or heterospecific, CON/HET), song within type (conspecific song 1, 2 or 3; heterospecific song 1, 2 or 3), direction of song (forward or reverse, FOR/REV) as well as the interactions of category, song type and song direction and the interaction of song with direction were initially included as explanatory variables and then interactions that were not significant were removed from the analysis (Rouder et al., 2016), as explained and justified in the Analysis section of the “Methods”.

Initially, a full factorial model was fitted to the data. Using this model, no significant difference was found between ZF-ZF and ZF-CF birds as a main effect (F_1,15 = 0.43, P = 0.51) but there were significant main effects of individual songs (Song ID) (F_4,384) = 4.4, P = 0.003) and the direction of songs (FOR/REV: F_1,467 = 13.26, P = 0.0001). A reduced model was investigated and compared to the initial model using the AIC values, where the smaller value is better. When the reduced model was used, on the basis of the smaller AIC value (Rouder et al., 2016), the results were, in general, similar: Bird Category as a main effect was non-significant (F_1,15 = 0.33, P = 0.58), individual song (Song ID) as a main effect was significant (F_4,384 = 4.95, P = 0.0007), as was FOR/REV (F_1,467 = 16.87, P = 0.0001). In addition, there was a significant difference between ZF-ZF and ZF-CF birds in the effect of CON or HET, with ZF-ZF birds having larger reductions in response to HET than to CON compared to ZF-CF birds (i.e. a significant interaction: (F_1,384 = 4.12, P = 0.043; see Tables 4 and 5; Figs. 9B and 10). These appear to result from a reduction in response strength to conspecific song and an increase in response strength to the song of the foster species (mean RS: ZF-ZF = 5.76 (CON), 4.49 (HET); ZF-CF = 5.20 (CON), 5.12 (HET); both P = 0.0001; see Table 5). CON/HET was also significant as a main effect but because it was significant in interaction with Bird Category, it was not examined further as an independent effect (Nelder, 1977; Martinez, 2015). Despite this strong evidence that direction influences RS depending on rearing, the size of the effect was not large. This is demonstrated when the RS values obtained for the forward and reverse presentations of a given song stimulus in a single unit are plotted against each other where most data points fall on or near the equality line (Fig. 10).

Table 4:

Significant main effects and interactions as determined by the GLMM analysis.

Effect	Num DF	Den DF	F Value	Pr > F
Type 3 tests of fixed effects
Individual song	4	384	4.95	0.0007
Song direction	1	467	16.87	<0.0001
Rearing × CON/HET	1	384	4.12	0.0429

DOI: 10.7717/peerj.9363/table-4

Table 5:

Least squares means.

Effect	Category	Song ID	Song type	Direction	Estimate	Standard error	DF	T value	Pr > |t|
Least squares means
Category × song type	ZF-CF		CON/HET		0.9706	0.05651	22.7	17.18	<0.0001
Category × song type	ZF-ZF		CON/HET		0.9479	0.04053	8.3	23.38	<0.0001
Direction				FOR/REV	0.9407	0.03458	15	27.21	<0.0001
Song (song type)		CON-1	CON		0.9380	0.03551	16.8	26.41	<0.0001
Song (song type)		CON-2	CON		0.9679	0.03551	16.8	27.25	<0.0001
Song (song type)		CON-3	CON		0.9719	0.03551	16.8	27.37	<0.0001
Song (song type)		HET-1	HET		0.9506	0.03551	16.8	26.77	<0.0001
Song (song type)		HET-2	HET		0.9466	0.03551	16.8	26.65	<0.0001
Song (song type)		HET-3	HET		0.9128	0.03551	16.8	25.70	<0.0001

DOI: 10.7717/peerj.9363/table-5

These results show that the MLd units of all birds did not respond equally to individual songs within one category (CON or HET) and indicate that early auditory experience shapes responses in the auditory midbrain in the zebra finch (Table 4).

General description

Effect of rearing in responses to song

We sought to answer two questions: Do units in MLd show a response bias towards conspecific songs, and if this exists, can this be modified by developmental experience? To this end, we challenged MLd units of ZF-ZF and ZF-CF zebra finches with stimuli consisting of 3 familiar zebra finch (CON) and 3 unfamiliar Bengalese finch (HET) songs, each played forwards and backwards. We found that in both groups of birds, responses to song stimuli were heterogeneous, that units in MLd were able to discriminate between different stimuli, and that a preference towards conspecific song was developmentally modified.

The response parameter of choice for the GLMM analysis was RS because it better reflects the responses of the units to different song stimuli. While the SI is good at detecting selectivity to certain features, its shortcoming lies in its failure to detect the strength of that response. The RS, although it provides little information regarding the temporal pattern of the evoked response, is a more direct measure of the ‘total’ response of the cell. The distribution of RS values in ZF-CF differs from that seen with ZF-ZF units (Mann–Whitney U test for equality of distributions, P = 0.001), although the mean differences were non-significant. Woolley, Hauber & Theunissen (2010) showed that cross-fostering resulted in a lower firing rate in MLd neurones. In contrast we find that the median RS for cross fostered birds is higher than for normal-reared birds (ZF-ZF: 2.4, ZF-CF: 4.4) (Mann–Whitney U test for medians, P = 0.0001). Schumacher, Schneider & Woolley (2011) found that units in MLd of male zebra finches under urethane anaesthesia had lower spontaneous and song-evoked firing rates and response strength (compared to unanaesthetised birds). In guinea pig IC, Astl et al. (1996) showed that frequency tuning was similar under urethane versus Ketamine/Xylazine anaesthesia, although the number of spontaneously active neurones was higher under Ketamine/Xylazine. While the anaesthetic regime may account for differences in overall response levels between our and Woolley’s results, it is unlikely to account for the direction of the changes observed as a result of cross-fostering.

Three main findings resulted from the GLMM analysis of RS values. First, the reverse versions of songs evoked stronger responses than the forward versions of songs in both ZF-ZF and ZF-CF zebra finches. Second, there was an effect of individual songs within CON and HET categories. Third, ZF-ZF MLd units responded more strongly to CON than to HET songs than did their ZF-CF counterparts. Each of these findings will be considered separately below.

Ascending and descending inputs into MLd

We did not establish whether the responses we observed are influenced by ascending or descending inputs to MLd (Pannese, Grandjean & Frühholz, 2015). In mammals, subcortical structures have been shown to contribute to the processing of vocalizations (Bauer, Klug & Pollak, 2002; Pannese, Grandjean & Frühholz, 2015; Roberts & Portfors, 2015), but recordings from the lower brainstem in songbirds have been limited to studies of tonotopy (Konishi, 1969; Konishi, 1970; Sachs & Sinnott, 1978). To our knowledge, no attempts have been made to examine the responses of hindbrain auditory nuclei in the processing of complex sounds to infer what influence they may have on MLd responses. It should be noted, however, that the terminal fields of ascending inputs to MLd originating in the hindbrain do not conform to the typical avian pattern where the nucleus laminaris (NL) terminal field defines an MLd core (Conlee & Parks, 1986; Takahashi & Konishi, 1988). Instead, the terminal fields appear to overlap more broadly within MLd (Krützfeldt et al., 2010). The organisation of the NL of songbirds also shows morphological modifications, the significance of which remains unknown (Kubke & Carr, 2006). Whether these anatomical differences reflect an adaptation for specific aspects of sound processing has not been established.

Descending inputs to the auditory midbrain have been shown to influence neuronal responses in a number of vertebrates (Endepols & Walkowiak, 1999; Gao & Suga, 2000; Suga et al., 2000; Endepols & Walkowiak, 2001; Ma & Suga, 2001; Suga et al., 2002; Popelář et al., 2016). It is thus possible that some of the changes observed in MLd may be influenced by forebrain auditory nuclei (Gentner & Margoliash, 2003; Cousillas et al., 2004; Jeanne et al., 2011; George & Cousillas, 2013; Araki, Bandi & Yazaki-Sugiyama, 2016). The auditory forebrain has been implicated in the categorisation of vocalizations (Chew, Vicario & Nottebohm, 1996; Grace et al., 2003; Elie & Theunissen, 2015). Strong responses to conspecific song have been recorded in Field L, and the development of neuronal selectivity for conspecific song appears to parallel the emergence of behavioural song preference (Amin, Doupe & Theunissen, 2007). Changes in forebrain responses that are influenced by development have also been reported in the ventral caudal hyperstriatum (cHV) (Grace et al., 2003) and caudal medial nidopallium (Stripling, Kruse & Clayton, 2001; Menardy et al., 2012). However, descending inputs to MLd in songbirds are not substantial. The only descending input that has been described is from the robust nucleus of the arcopallium (RA) cup (Mello et al., 1998), but these inputs are largely confined to a region external to MLd, and the adjacent intercollicular complex. Similarly with respect to the RA cup’s projection to the vicinity of lower auditory brainstem nuclei, such as the dorsal and ventral nuclei of the lateral lemniscus (LLD and LLV), both of which project strongly to MLd (Wild, Krützfeldt & Kubke, 2010). But in neither the case of MLd nor in that of LLD and LLV have injections of neural tracers been found to retrogradely label adjacent neurones in the region of the RA cup’s descending projections.