|Home | About | Journals | Submit | Contact Us | Français|
Acoustic communication in Drosophilid flies is based on the production and perception of courtship songs, which facilitate mating. Despite decades of research on courtship songs and behavior in Drosophila, central auditory responses have remained uncharacterized. In this study, we report on intracellular recordings from central neurons that innervate the Drosophila AMMC (antennal mechanosensory and motor center), the first relay for auditory information in the fly brain. These neurons produce graded-potential (non-spiking) responses to sound; we compare recordings from AMMC neurons to extracellular recordings of the receptor neuron population (Johnston’s Organ neurons or JONs). We discover that while steady-state response profiles for tonal and broadband stimuli are significantly transformed between the JON population in the antenna and AMMC neurons in the brain, transient responses to pulses present in natural stimuli (courtship song) are not. For pulse stimuli in particular, AMMC neurons simply low-pass filter the receptor population response, thus preserving low-frequency temporal features (such as the spacing of song pulses) for analysis by postsynaptic neurons. We also compare responses in two closely related Drosophila species, D. melanogaster and D. simulans, and find that pulse song responses are largely similar, despite differences in the spectral content of their songs. Our recordings inform how downstream circuits may read out behaviorally-relevant information from central neurons in the AMMC.
In order to select appropriate mates, animals must be able to tell conspecific from heterospecific and to judge fitness among conspecific suitors. Animals take advantage of several sensory channels to make these distinctions, but the use of acoustic cues is quite prevalent across the animal kingdom (Suga, 1989, Bass and McKibben, 2003, Kelley, 2004, Mason and Faure, 2004, Theunissen and Shaevitz, 2006). Processing behaviorally-relevant acoustic signals typically involves neural tuning for conspecific sounds, and such tuning has been reported in animals as divergent as insects, birds and primates (Bentley and Hoy, 1972, Braaten and Reynolds, 1999, Machens et al., 2005, Petkov et al., 2008). However, it is not known how auditory systems achieve their exquisite tuning for conspecific signals, nor how precisely auditory responses are transformed from one layer of the nervous system to the next to accomplish such tuning.
To characterize auditory systems, synthetic stimuli (such as pure tones and broadband noise) have been extremely useful – these stimuli are simple to parameterize and control, and therefore responses straightforward to quantify. However, recent studies using natural sounds have found that auditory neurons encode natural stimuli both differently from synthetic stimuli (Bar-Yosef et al., 2002, Woolley et al., 2006) and more efficiently (Rieke et al., 1995, Hsu et al., 2004, Smith and Lewicki, 2006). As a consequence, responses to synthetic stimuli are generally poor predictors of responses to natural stimuli (Rotman et al., 2001, David et al., 2004, Woolley et al., 2006). It is thought that the temporal structure present in natural stimuli forms the basis for this difference in tuning (Woolley et al., 2006), but the precise mechanisms remain unknown. We investigate this issue here using Drosophila, whose songs are, like those of other animals, highly temporally patterned. Studying auditory processing mechanisms in Drosophila provides a unique opportunity to discover general principles underlying how species-specific sounds are represented within the brain, given the simplicity of the fly nervous system and the wealth of genetic tools available to identify and study individual neurons.
In Drosophila, sound is first detected by the fly’s antenna via vibrations of its feathery arista and associated 3rd segment (Gopfert and Robert, 2001, 2002). Analogous to the mechanical coupling between the basilar membrane and auditory hair cells in vertebrates, the fly’s auditory organ transduces mechanical deflections of the arista into receptor potentials within the ~500 Johnston’s Organ neurons (JONs) housed within the antenna (Todi et al., 2004, Albert et al., 2007, Nadrowski et al., 2011). JONs project along the antennal nerve into the brain, where they arborize in the antennal mechanosensory and motor center (AMMC) (Kamikouchi et al., 2006). Dependent on where their somata are located in the antenna, JONs project to one of roughly five zones in the AMMC – those that project to zones A and B are mostly sensitive to vibratory stimuli, such as courtship song, whereas those that innervate zones C and E respond more robustly to static stimuli, such as wind or gravity (Kamikouchi et al., 2009, Yorozu et al., 2009). In this study, we investigate auditory tuning among central neurons that innervate AMMC zones A and B, and compare responses to those of the JON population.
Courtship song in Drosophilid flies typically comprises two modes, pulse and sine. Pulse song, which is characterized by brief sound impulses separated by species-typical inter-pulse intervals, has been shown through behavioral experiments to be more salient for mate selection (Bennet-Clark and Ewing, 1969, von Schilcher, 1976). However, the majority of tuning curves formed for the Drosophila auditory receiver and coupled receptor neurons come from analyzing responses to synthetic stimuli, such as sinusoids and white noise (Gopfert and Robert, 2002, Gopfert et al., 2006, Effertz et al., 2011). These tuning curves describe responses at steady-state, whereas responses to song pulses are likely to be transient. Here we test the hypothesis that tuning for song pulses (natural sounds) in the central auditory system of Drosophila differs from steady-state tuning for synthetic stimuli. In addition, by making comparisons between the songs and auditory systems of two closely related and hybridizable Drosophila species, we offer insights into the neural basis for species-specific song recognition.
Song recordings (from pairings of single virgin males with virgin females) and JON recordings (from virgin females) were collected from OrR (D. melanogaster) and sim194 (D. simulans collected from Winters, CA (Dermitzakis et al., 2000)) strains. Whole-cell patch clamp recordings were performed on virgin females of the following genotypes: AMMC-A1: (UAS-eGFP2x(II); c767-GAL4(III)); GF: (c17-GAL4, UAS-CD8-GFP(III)); AMMC-B1: (UAS-eGFP2x(II); 8-245-GAL4(III)); AMMC-B2: (JO2(NP1046); UAS-eGFP2x(II)). D. mel/D. sim hybrid females were generated by crossing UAS-eGFP2x; c767-GAL4 females to sim194 males. All flies were reared on standard cornmeal agar medium at constant temperature (25C) and constant relative humidity (65%).
Our sound stimuli consisted of pure tones (17 4s tones (100-900Hz), each delivered at 2 intensities), white noise (2 different 4s variants of white noise, each delivered at 3 intensities), and fly song (10 4s song bouts from D. melanogaster (6 stimuli) and D. simulans (4 stimuli)). For a few experiments, we used synthetic pulse stimuli, which were generated by cutting a single pulse from a D. mel song recording and creating pulse trains consisting of 10-20 pulses separated by IPIs from 20-90ms. Each 4s auditory stimulus was spaced 20s apart from the next stimulus. The intensity of each stimulus was computed using a 20ms sliding FFT window (Hamming) applied to the signal recorded on a calibrated NR23158 (Knowles electronics) pressure gradient microphone connected to a custom-built non-integrating differential amplifier (1.6 Hz high-pass filter with −3dB roll-off; see below for more information on the frequency response of the microphone). The NR23158 microphone was calibrated in an anechoic chamber (Siapas lab, Caltech) at a distance of 2.3m from the sound source (Fostex FX120), and according to the protocol provided in (Gopfert et al., 2006). The intensities of tonal and white noise stimuli were then adjusted to be within the range from 0.1 to 3 mm/s; fly songs were played back at the intensity they were recorded, which fell roughly within this range.
Our sound delivery system consisted of: i) M-audio Audiophile2496 sound card, ii) Samson S-phone 4-channel headphone amplifier, iii) KOSS Sparkplug Headphones (16 Ohm impedance; Sensitivity 112dB SPL/1mW), and iv) 50cm Nalgene 489 rigid Polyethylene tubing (ID: 3/16″ OD: 5/16″) coupled to 5cm of thinner tubing (ID: 1/16″ OD: 3/16″). We estimated the transfer function of the sound delivery system by playing 50 trials of a white noise stimulus (band-pass filtered from 80-1000Hz and delivered at range of intensities) and recording the output on a Bruel and Kjaer ½″ free-field condenser microphone (4133) in a custom-built sound-proof box. We chose this microphone because its frequency response characteristics are flat (±2dB) within the low frequency range we tested. The transfer function between input (x) and output (y) was estimated using the quotient of the cross power spectral density (Pyx) of x and y and the power spectral density (Pxx) of x (see the tfestimate function in Matlab). We then compensated for distortions introduced by the headphone speaker and coupling tube by convolving our stimuli with the inverse transfer function filter before playing the sound through the system (our method for sound calibration is conceptually similar to that used in (Gopfert and Robert, 2003) and (Wightman and Kistler, 1989)). Without compensation, the output of the system amplifies low frequencies (100-300 Hz) and attenuates higher frequencies (300-1000 Hz). After compensation, the output of the system is a near-perfect match with the input (Fig. 1). The sound tube was placed roughly 2mm from the fly’s right antenna, and at a 45 degree angle to the substrate (to mimic male singing (Bennet-Clark, 1971)). The tube creates a planar sound wave, and we empirically determined that the response of the pressure gradient microphone, placed 2mm from the end of the tube, increased linearly with frequency (that is, we observed no flattening of the microphone’s frequency response at short distances from the end of the tube, consistent with (Werner, 2008)). We compensated for the frequency response of microphone post-hoc by dividing FFT voltages by their respective frequencies.
Recordings were made using a Brownlee Precision 440 differential amplifier (gain = 10,000) and by inserting a recording tungsten electrode (18 MΩ impedance) into the gap between antennal segments a1 and a2 of the fly’s right antenna and a reference tungsten electrode (7MΩ impedance) into the eye (similar to (Eberl et al., 2000)). Responses were band-pass filtered between 75Hz and 10KHz, digitized using a National Instruments A-D board, and acquired at 10KHz in IGORPro (Wavemetrics, Inc.).
Whole-cell patch-clamp recordings were, in general, performed as described previously (Murthy et al., 2008). Modifications of the dissection were required to access the different AMMC neurons. For the A1 and GF neurons, we accessed the brain from the posterior side of the head (similar to the protocol described in (Murthy and Turner, 2010)). For B2 neurons, we accessed the brain through the proboscis, as these cell bodies lie ventral to the AMMC. For B1 neurons, which are positioned just above (dorsal to) the AMMC, we accessed the brain from the anterior side of the head. For all preparations, we ensured that all 3 segments of the antennae remained underneath the recording platform (the area above the recording platform is immersed in saline, while the area underneath is kept dry using a wax barrier to separate the two sides (see (Murthy and Turner, 2010) for more information on mounting flies for in vivo recordings). Following dissection and before recording, we checked for self-sustained vibrations of the antennae in response to gentle air flow and for spontaneous leg movements (to indicate the health of the preparation). For AMMC-A1 recordings (these neurons bilaterally innervate right and left AMMCs (Kamikouchi et al., 2009) – our recordings came from neurons in either the right or left hemisphere), we used patch-clamp electrodes pulled from capillary glass (OD = 1.5, ID = 1.1) with resistances of ~5-6 MΩ. The perineural sheath and overlaying neurons were typically removed on the electrophysiology rig with a cleaning pipet, and under visual control. Only neurons with input resistances >100 MΩ were included for analysis. All cells were held between − 55 mV and −65 mV, in current-clamp mode, using either an Axoclamp2B or a Multiclamp700B amplifier, and voltage signals were acquired in IGORPro at 10 kHz via a National Instruments A-D board. Recordings typically lasted for 30-60 minutes, and permitted the measurement of responses to (on average) 80% of our stimulus set.
We recorded D. mel and D. sim songs in a 1cm diameter chamber with a nylon mesh floor that rested 2mm above the surface of a pressure sensitive microphone (Bruel + Kjaer 4179 1″ free field microphone) – the sides of the chamber were coated with fluon, which limited the movement of flies to the floor of the chamber. Songs were recorded in a semi-anechoic box from Acoustic systems (ETS-Lindgren), digitized via a National Instrument A-D board, and acquired at 15KHz in LabVIEW.
Analysis was carried out in Matlab (Mathworks, Inc.).
For steady-state responses: i)JO: response amplitudes were calculated as the sum of the absolute value of the voltage signal (following baseline subtraction) in a 20ms time window, and averaged across all windows for the length of the steady-state response period (0.5-3.5s after stimulus onset). ii)AMMC: response amplitudes were calculated as the change in voltage from baseline, during the steady-state region of the response (DC component). For onset responses (Fig. 4F only): i)JO: response amplitudes were calculated as the sum of the absolute value of the voltage signal (following baseline subtraction), in a 10ms time window that followed stimulus onset by 2.5ms. ii)AMMC: response amplitudes were calculated as the sum of the absolute value of the voltage signal (following baseline subtraction), in an 80ms time window that followed stimulus onset by 2.5ms. For transient responses to pulse stimuli: i)JO: response amplitudes were calculated as the sum of the absolute value of the voltage signal in a 20ms time window (following baseline subtraction) that followed the pulse stimulus by 2.5ms (computed lag between stimulus and response). ii)AMMC: response amplitudes were calculated as max-min in the 20ms time window following the pulse stimulus by 2.5ms. Both JO and AMMC-A1 responses were normalized as follows: i)for pure tone stimuli (Fig. 3), we found the maximum response amplitude across all frequencies (100- 900Hz), for a given animal and stimulus intensity (red or blue), and divided all response amplitudes (for that animal and that intensity) by this maximum. ii)for white noise stimuli (Fig. 4), we found the maximum response amplitude across all stimulus intensities (green, blue, and red), for a given animal, and divided all response amplitudes (for that animal) by this maximum. iii)for fly song stimuli (Figs. 5 and and6),6), we found the maximum response amplitude across all 612 pulse stimuli tested (from 10 song clips), for a given animal, and divided all response amplitudes (for that animal) by this maximum. After normalization, response amplitudes were averaged across animals to form frequency and intensity tuning or response curves.
For each song pulse, response sensitivity was computed as the normalized response amplitude (V/Vmax) divided by the stimulus intensity.
For steady-state responses (to pure tones and white noise), fast fourier transforms (FFT) were computed using sliding 20ms response windows (Hamming), and then absolute values of the resulting coefficients were averaged. For pulse stimuli, FFTs were computed on the 20ms window that included each pulse stimulus; for pulse responses, FFTs were computed on the 20ms response window that followed the stimulus by 2.5ms. Normalization was performed as described above.
All correlation coefficients reported in this study come from taking the maximum computed coefficient from sliding cross-correlations (with mean subtraction of signals and with typical lags of 100ms; see the xcov function in Matlab). To compute the error energy (EE) between signals x and y, we first scaled both x and y between 0 and 1, and then used the following formula:
Pulse responses (lasting >22 ms) were extracted from recordings and only responses with response heights > 3× background noise were selected. We fit the decay of each pulse response (the 20ms period that begins 2 ms after the peak) to a single exponential function. The fits were of the form: a(e−bx) + c, where c is a constant term to account for the DC offset in the response. Decay time then equals 1000/b (in ms). To simulate AMMC responses, male IPI distributions were approximated from (Cowling DE, 1981) using kernel density estimation. For each iteration of the model, 19 IPIs were drawn from one of these distributions (D. mel, D. sim, or the hybrid) and used to construct an artificial train of 20 AMMC pulse responses. Both rise times (taken as the mean rise time of actual pulse responses (Fig. 5G)) and maximum response amplitudes (Vmax) were fixed. That is, the maximum absolute amplitude of any pulse response was set equal to the amplitude of the first pulse response in the train, and rise times were scaled such that the maximum response amplitude was not exceeded (in agreement with our recordings (Fig. 5G)). Response decays were modeled as single exponentials, and no noise was added to the simulated response. For each decay time tested (from 2 to 24 ms), we ran 100 iterations of the model to obtain the mean and standard deviation of the DC offset.
All p values in this study come from either two-tailed Student’s t-tests, assuming unequal variances or a balanced one-way ANOVA (only Fig. 4).
The adjusted r2 statistic is a measure of the correlation between the data values and the predicted data values (based on the fit); perfect correlation = 1. It was calculated as the ratio of the sum of squares of the regression and the total sum of squares, and adjusted for residual degrees of freedom (sample size minus the number of fitted coefficients).
Tuning curves for JO neurons have been previously characterized using electrostatic actuation to vibrate the arista (Albert et al., 2007). This method is not, at present, compatible with patch clamp recordings - we therefore used an alternative (a small earphone speaker coupled to a length of tubing) for delivering both calibrated and localized sounds to flies during electrophysiological recordings (see Methods and Fig. 1). Our sound delivery system was designed to mimic the singing of the male during natural courtship (Bennet-Clark, 1971), and all stimuli were concentrated in the frequency and intensity ranges of recorded D. melanogaster (D. mel) and D. simulans (D. sim) songs (for both species, roughly 100-900 Hz and 0.1 to 10 mm/s (Figs. 1A and and5A),5A), which is within the dynamic range of the D. mel arista (Gopfert et al., 2006)).
The auditory pathway of Drosophila begins in the antenna, where ~500 JONs send projections into the brain, to the AMMC (Kamikouchi et al., 2006). The AMMC is divided into 5 zones, and each JON typically innervates only one. A previous study identified 4 types of central neurons (based on morphology) that innervate AMMC zones A and B – these neurons were proposed to be postsynaptic to zone A and B JONs, and therefore likely responsive to courtship song (Kamikouchi et al., 2009). We consequently focused our analysis on these neurons (Figs 2A and 2B); they include two neuron types that innervate zone A, GF (giant fiber; 1 cell/hemisphere) and AMMC-A1 (2 cells/hemisphere), and two neuron types that innervate zone B, AMMC-B1 (10.5 ± 0.8 cells/hemisphere (n=7 brains)) and AMMC-B2 (2 cells/hemisphere). Antibody staining suggests that AMMC-B1 neurons are cholinergic and AMMC-B2 neurons are GABAergic (Fig. 2B, inset); AMMC-A1 and GF neurons may utilize other neurotransmitters, because neither neuron type was found to be ChAT (choline acetyltransferase) or GABA immunoreactive (data not shown). Zone A AMMC neurons are part of the giant fiber pathway, which senses both visual and mechanosensory stimuli to trigger an escape response; these two types of neurons are also gap-junction coupled to one another ((Phelan et al., 2008) and our dye fills). While the specific inputs to the giant fiber pathway are not known, in wild type flies, the GF neuron is not activated by either visual or mechanosensory stimuli alone (Engel and Wu, 1996, Fotowat et al., 2009).
Using well-established methods for in vivo whole-cell patch clamp in flies (Wilson et al., 2004, Murthy and Turner, 2010), we recorded from at least two animals for each of these AMMC neuron types, and found in all cases graded responses to sound pulses; these potentials are tightly locked to the sound stimulus, and we observed no differences in temporal delay between sound stimulus and graded response across the different AMMC neurons (Fig. 2C). Depolarizing current injection (steps from −60mV to 0mV) at the soma of AMMC-A1 or -B2 neurons (during current clamp recordings) failed to elicit action potentials in these cells (data not shown). Because neuronal morphologies of these cells are different (Fig. 2B), it is unlikely that our inability to control the membrane potential at sites distant from the soma affected our ability to generate and observe spikes via current injection. In contrast, current injection at the GF neuron soma produced a large spike (Fig. 2D; resembling a typical Ca++ spike, found commonly in insect “non-spiking” neurons (Laurent, 1991)), again indicating that the small amplitude graded responses to sound, observed uniformly across these 4 cell types, are likely not an artifact of membrane filtering of spiking events occurring distal to the soma. To test this further, we asked whether or not sound responses could summate with subthreshold somatic current injection to drive spiking in the GF neuron. Playing sound simultaneously with current injection that, on its own, was not able to elicit a spike in the GF neuron, produced a full-blown spike (Fig. 2D). As sound responses were qualitatively similar across these 4 types of neurons (Fig. 2C), we focused the remainder of our analyses on the AMMC-A1 neurons, which proved more accessible for patch clamp recordings.
We then measured response reliability of JO and AMMC-A1 neurons by analyzing responses to ten different clips of courtship song (Fig. 2E). JON population responses were similar whether recorded through the antenna or with a suction electrode to access the antennal nerve directly as it enters the brain (data not shown); that is, LFP (local field potential) recordings did not simply reflect antennal muscle activity. AMMC-A1 neuronal responses are highly correlated across animals: correlation coefficients for comparisons across animals and for each song stimulus are significantly larger than correlation coefficients for comparisons between different song clips (Fig. 2F). As we were unable to record multiple trials for every song stimulus during AMMC-A1 patch clamp recordings (due to the size of our auditory stimulus set and often short duration of patch clamp recordings), we compared across-animal correlation coefficients to those from JON LFP recordings, and found a similar degree of correlation in fly song responses across animals and across trials within an animal (Fig. 2F). This reproducibility in neural responses (both at the level of the JON population and the AMMC-A1 neurons) enabled us to form tuning curves across experiments for the three different types of stimuli we examined: pure tones, white noise, and fly song.
To examine the frequency tuning of the JON population and AMMC-A1 neurons, we recorded responses to tonal stimuli in the frequency range of D. mel and D. sim courtship songs (100- 900Hz), and at two particle velocity levels (Fig. 3A). Because a given animal may have received tonal stimuli of only one of the two stimulus intensities (red or blue), we normalized tuning curves (V/Vmax) for each stimulus intensity and each animal separately (thus, first we focused on only frequency tuning). JON LFP response amplitudes during the steady-state period (0.5-3.5s following stimulus onset; Fig. 3B) were maximal for tones of ~200Hz (Fig. 3D), consistent with tuning curves produced for the arista by laser Doppler vibrometry (Gopfert and Robert, 2002). JON responses are oscillatory; we observed that the dominant frequency of the JON response is twice the input frequency, for stimulus frequencies up to ~450Hz (Fig. 3E), consistent with the known mechanical structure of the JO (two populations of JONs each responding to one phase of the tonal stimulus (Nadrowski et al., 2008)). The AMMC-A1 neurons innervate a sub-region of AMMC zone A, which contains synapses from ~50-100 JO-A neurons (Kamikouchi et al., 2006). AMMC-A1 neurons produce a sustained depolarization during the pure tone stimulus (Fig. 3C) – the amplitude of this depolarization decreases with increasing stimulus frequency (particularly past 600Hz), and, in contrast to the JON population, tuning curves do not show a peak for ~200Hz stimuli (Fig. 3D).
Broadband stimuli such as white noise can be useful to characterize a system’s response to a large range of frequencies simultaneously. We therefore used white noise (band-pass filtered between 80-1000Hz) to examine both frequency response and intensity tuning curves for the JO population and AMMC-A1 neurons. We stepped the intensity of white noise from roughly 0.1 mm/s to 3 mm/s in 3 steps (Fig. 4A), in order to span a range of intensities flies typically encounter during courtship (Bennet-Clark, 1971). We first examined FFT (fast fourier transform) amplitudes for JON population and AMMC-A1 neuron responses during the steady-state response period. FFT amplitudes for the JON population reveal a peak at ~300Hz (Fig. 4D), largely consistent with tuning curves formed with tonal stimuli (tuning curves were peaked for 200Hz stimuli, which produce a ~400Hz response in the JONs; see Fig. 3). However, we observed that FFT amplitudes decreased with increasing stimulus intensity. This same trend could be observed when we instead calculated response amplitudes across stimulus intensities (Fig. 4F) or response variance (data not shown).
To examine the responses of the AMMC-A1 neurons, we first subtracted the DC component of the response and computed the FFT of the voltage fluctuations riding on top of the DC response to white noise. In contrast to the JON population, FFT amplitudes for AMMC-A1 neurons are identical across stimulus intensities (Fig. 4E). That is, while DC-subtracted activity is boosted above baseline during the steady-state response period, it does not vary across stimulus intensities. We also examined the DC component of the AMMC-A1 response to white noise, and found, in contrast to the JON population, no significant trend with increasing stimulus intensity (Fig. 4F). These data collectively reveal a transformation in both frequency and intensity tuning between the JON population and AMMC-A1 neurons during the steady-state response period. Given that a previous study demonstrated that the JO population response to the frequencies and intensities we examined is dominated by the activity of the JO-A neurons, as opposed to JO-B or JO-CE neurons (Kamikouchi et al., 2009), it seems likely that the transformation we observe reflects a difference between AMMC-A1 neurons and JO neurons that also innervate zone A (see Discussion).
Above, we characterized auditory tuning during the steady-state response period. We noticed, however, that responses were often maximal at stimulus onset (Figs. 3B-3C and Figs. 4B-4C). Stimulus onsets have obvious ethological relevance, and nearly all auditory neurons studied respond briskly to onsets (Heil, 1997a, b). All of our white noise stimuli contained a 10ms 3KHz pulse at onset (followed immediately by white noise); analysis of onset responses across the 3 intensities we tested revealed a surprising match between the JON population and AMMC-A1 neurons. At onset, both responses increase linearly with intensity (Fig. 4F). We reasoned that transient responses to courtship song pulses (characteristically 5-20ms pulses spaced with species-typical inter-pulse intervals) might therefore also be matched between the JON population and AMMC-A1 neurons.
Because D. melanogaster (D. mel) females will copulate not only with D. mel males, but also with D. simulans (D. sim) males (Sturtevant, 1920), we chose to make comparisons between these two closely related species (separated by only 2-3 million years of evolution (Clark et al., 2007)). Our song stimuli therefore consisted of 612 pulses presented within ten clips of either D. mel or D. sim song (Fig. 5A); songs from these two species are distinct: D. sim songs contain higher average pulse carrier frequencies (480 Hz in D. sim compared to 280Hz in D. mel) and larger average inter-pulse intervals (IPI; 55ms in D. sim compared to 30ms in D. mel) (Cowling DE, 1981). We first used these stimuli to compare pulse song responses between the JON population and AMMC-A1 neurons in D. mel.
In contrast to responses at steady-state, tuning for fly song pulses was similar between the JON population and AMMC-A1 neurons. With increasing pulse intensity, we observed an increase in response amplitude and a decrease in sensitivity (Fig. 5B). Fitting the data with a straight line (goodness of fit (adjusted r2) for JONs = 0.9, and for AMMC-A1 neurons = 0.8) yielded nearly identical slopes for JONs (−0.9) and AMMC-A1 neurons (−0.8). The relatively flat relationship between intensity and response amplitude reveals that the softest pulses in our stimulus set evoked almost as robust a response as the loudest pulses (see also Fig. 5D). These results are consistent with JO LFP (a.k.a. CAP) onset responses measured using electrostatic forcing to displace the arista and over a similar range of stimulus intensities (Albert et al., 2007). We also compared frequency tuning for pulse stimuli. JON population tuning was flat within the range from 100-900Hz (in contrast to at steady-state), and, again, was matched to AMMC-A1 neuron tuning (Fig. 5C; response sensitivities are not dependent on frequency, and therefore form the mirror image (inverse) of the frequency and intensity profiles of the stimuli themselves, plotted in Fig. 5A).
The similarity between JON population and AMMC-A1 neuron responses to fly song pulses is perhaps unexpected, given the differences observed in steady-state responses to synthetic stimuli across a similar intensity range. The nature of this similarity suggested that the graded response of the AMMC-A1 neurons may be the result of a low-pass filter applied to the JON population response to fly song. We found that a first-order filter with a time constant (tau) of 10ms applied to the averaged JON fly song responses (termed JO response envelopes) matched the AMMC-A1 neuron data extremely well (Fig. 5D). To quantify the match, we computed cross-correlations between each AMMC-A1 neuron response to each of the 10 song stimuli and the corresponding JO response envelope. Correlation coefficients were high (mean = 0.77) between AMMC-A1 responses and JO response envelopes, and significantly larger than between AMMC-A1 responses and stimulus envelopes (Fig 5F). That is, the intermediate filter of the JON population response is required to model the AMMC-A1 neuron response. In further support of this, the error energy, a measure of the difference between two signals after scaling (see Methods), is significantly lower for comparisons between AMMC-A1 responses and JO response envelopes than for comparisons between AMMC-A1 responses and stimulus envelopes (Fig. 5F). The time constant of the low-pass filter (with cutoff at ~50 Hz) is within the biologically plausible range of membrane time constants, and suggests that filtering is, for the most part, passive. Similar to responses to fly song (Fig. 2F), JON white noise responses are reliable across animals (data not shown); we therefore took the average JON response to white noise and passed it through the same low-pass filter to compare AMMC-A1 white noise responses with JO response envelopes; we found that the same filter applied to JON white noise responses at steady-state was a less reasonable fit to the AMMC-A1 data (Figs. 5E-5F). Thus, for transient responses specifically, AMMC-A1 responses are well-matched to the population-level activity of the JO.
The graded responses of the AMMC-A1 neurons enable summation between song pulses that arrive close together, producing a DC shift (sustained membrane depolarization) in the AMMC-A1 response. To determine whether such a shift was correlated with the D. mel IPI range (Fig. 6E), we measured AMMC-A1 neuron responses (n=3) to synthetic pulses spaced 20-90ms apart. Graded response peaks to individual pulses are distinguishable down to an IPI of 20ms (the AMMC-A1 neurons therefore preserve pulse timings in their responses), but a DC shift in the response occurs consistently below an IPI of 50ms (Fig. 5G). If this DC shift is sufficient to support neurotransmitter release, it may serve as one mechanism for signaling the preferred IPI (see Discussion).
We next explored if the early auditory systems of D. mel and D. sim could be used to differentiate between their species-specific pulse songs. While broadband stimuli produced a slight rightward shift (to higher frequencies) in steady-state response curves in D. sim females (data not shown, but consistent with (Riabinina et al., 2011)), tuning for pulses was indistinguishable between the two species (Figs. 6A-6C). That is, normalized JO pulse response amplitudes, relative to either the intensities (Fig. 6B) or frequencies (Fig. 6C) of pulse song stimuli, are nearly identical in D. mel and D. sim. This implies that these sibling species, at the level of the auditory receptor population, are equally sensitive to both D. mel and D. sim pulse songs. What happens at the next layer? While identification of AMMC-A1 neurons in D. sim is not feasible without appropriate genetic tools in that species, matings between D. mel females and D. sim males produce hybrid female progeny that carry the GAL4/UAS transgenes (for neural labeling) from their D. mel parent. We recorded from AMMC-A1 neurons in D. mel/D. sim hybrid females, and observed similar song responses to those from D. mel flies (Fig. 6D).
We observed a consistent DC shift in AMMC-A1 responses to pulse trains containing short IPIs (20 or 30 ms, Fig. 5G); we wanted to next explore how neuronal decay times and stimulus IPIs interacted to generate DC shifts. To do this, we generated a model, and then compared this model to measured decay times from hybrid and D. mel AMMC-A1 neurons. We first established that AMMC-A1 decays could be fit with single exponentials (r2 = 0.947 ± 0.001 (SEM)), by examining D. mel responses to pulses (200-400Hz) separated by IPIs > 50ms (these responses decayed back to baseline). We then simulated AMMC responses to pulse trains by drawing from the pulse song IPI distributions of D. mel, D. sim, or their inter-specific hybrid (Cowling DE, 1981) and varying single exponential decay times (Fig. 6E; see Methods for details on the model). Pulse trains containing IPIs pulled from the D. mel distribution produced DC shifts >5% of the maximum response amplitude with decay times (time to reach 37% of the decay) of at least 6 ms; this tau falls squarely within the range of decay times measured from D. mel AMMC-A1 fly song responses (Fig. 6F). The time constant (10ms) of the low-pass filter used to convolve JO responses (Fig. 5) also falls within this range. Thus, the decay time of D. mel AMMC-A1 neurons is well-suited for responding (with a DC shift) selectively to D. mel pulse song. For D. sim AMMC-A1 neurons to produce a comparable DC shift in response to D. sim pulse trains would require changing the decay time to greater than 13ms (Fig. 6E); however, such a lengthening of decay time would also serve to increase the response to D. mel song, which may be undesired. Analysis of 1428 pulse responses from hybrid AMMC-A1recordings did not reveal such a shift in decay times (Fig. 6F). While the D. mel phenotype may be dominant in the hybrid, our results suggest that sustained depolarization of AMMC-A1 neurons (and likely other AMMC neurons that also produce graded responses to courtship song (Fig 2C)) occurs in both species for IPIs in the D. mel range. We discuss below how such similarity in the early auditory systems of D. mel and D. sim may be exploited for song detection in these sibling species.
In this study, we report on the first recordings from central auditory neurons in the Drosophila brain. We find that AMMC neurons produce graded potential (non-spiking) responses to sound. While most neurons thus far characterized in the Drosophila brain (primarily in the olfactory and visual pathways, but also using whole-cell patch clamp methods) can produce action potentials in response to sensory stimulation (e.g., (Wilson et al., 2004, Joesch et al., 2008)), graded transmission is common in insect nervous systems, where neurite lengths are often short relative to neurite length constants (Burrows and Siegler, 1978, Buschges, 1995). Graded transmission is also used at some vertebrate central synapses (Alle and Geiger, 2006, Shu et al., 2006). For very dynamic signals, graded responses have been shown to encode more information than spiking responses (Kretzberg et al., 2001), and can be more reliable, as they don’t suffer from problems associated with action potential failures. However, it remains to be determined whether and how as of yet unidentified downstream neurons in the Drosophila auditory circuit read out graded signals from the AMMC neurons.
In response to broadband stimuli, we found that JON population responses decrease in magnitude with increasing stimulus intensity. The JONs are mechanically coupled to the arista, which vibrates in response to sound stimulation. A previous study found that arista responses, assayed with laser Doppler vibrometry, shift their peak response frequency with increasing intensity of white noise stimulation, across a similar intensity range (Gopfert and Robert, 2002). While we do not know how JON frequency responses might shift for different white noise intensities and angles of the incident sound wave than those we examined (see Methods), we hypothesize that intrinsic cellular properties of the JONs may transform response curves of the arista, in order to maximize responses at certain distances from the sound source (and therefore particular intensities) and diminish responses to higher intensities (when the male fly comes too close to the female’s receiver, for example), across a broad frequency range. Indeed, other studies that have assayed the JON population with LFP recordings have found that for high intensity tonal stimuli (>1 mm/s), responses decrease with increasing intensity (Kamikouchi et al., 2009, Effertz et al., 2011). As a future direction it would be useful to characterize for our delivery system, designed to mimic courtship conditions, sound responses using simultaneous laser Doppler vibrometry and electrophysiology over a larger range of stimulus intensities and angles than we have examined. Finally, identifying the array of ion channels expressed in the JONs should shed light on biophysical mechanisms that may shift tuning curves between the arista and the JO.
Our study reveals a complex transformation in response profiles between the summed output of the JON population and AMMC-A1 neurons in the brain, at steady-state. While we do not know which of the ~50-100 JO-A neurons are presynaptic to AMMC-A1 neurons, and therefore cannot compare tuning directly between pre- and postsynaptic partners, we do not observe a similarly complex transformation for responses to pulse stimuli. One possibility to explain this difference between steady-state and transient tuning is that the JON LFP represents the activity of JO-A neurons for transients, but not at steady-state (that is, other JO neurons that innervate different zones of the AMMC dominate the response at steady-state). We think this possibility is unlikely given previous calcium imaging experiments, which demonstrated that JO-A+B neurons produce the strongest steady-state responses to sinusoidal vibration (compared to JO-CE neurons), and of the A+B neurons, JO-A neurons respond preferentially to frequencies between 100-900Hz (Kamikouchi et al., 2009). This same study reported that killing either the JO-B or JO-CE neurons did not affect JO steady-state intensity tuning profiles (in response to sinusoidal stimuli), but that killing JO-A neurons specifically shifted tuning curves to higher intensities. While the strength of enhancer lines used for neuronal killing may have varied across these experiments, these data collectively suggest that JO-A neurons dominate the JO response to vibratory stimuli in the frequency and intensity range we examined. We therefore posit that the transformation at steady-state arises rather from an active mechanism within the AMMC neurons themselves. Such a mechanism may involve synaptic adaptation during responses to longer stimuli, which are characterized by sustained depolarization of the membrane potential. Indeed, for responses to pulse trains comprised of short IPIs (20ms), we observed larger DC shifts in the membrane potential than for longer IPIs, but the same peak response amplitudes (Fig. 5G). This is consistent with data from the avian auditory system, where a mixture of short-term facilitation and depression has been shown to similarly maintain subthreshold response amplitudes at even high stimulus rates (MacLeod et al., 2007). An alternative mechanism may involve a change in brain state during listening to natural sounds specifically (Engel et al., 2001) – that is, a feedback system whose activity is based on the social importance of the sound might shape AMMC responses to match JO tuning curves for pulse songs specifically. Regardless of mechanism, differences in steady-state versus transient responses may have evolved based on the different functions of the mechanosensory system in flies. In response to sustained JON activity at steady-state (during constant low amplitude air flow, for example), the tuning curves of the AMMC-A1 neurons allow the system to respond preferentially to low stimulus frequencies, while remaining more invariant to changes in intensity. However, in response to onsets or song pulses, both ethologically relevant stimuli, AMMC-A1 neurons follow the tuning of the JON population, in order to reliably pass on information about stimulus timings and intensities, to downstream circuits for analysis. Such differences in steady-state versus transient responses have been observed in other systems (Mechler et al., 1998), and may be important for encoding natural stimuli, which, regardless of sensory modality, largely consist of transients (Rieke et al., 1997).
We have characterized detailed tuning curves for one subset of neurons within the AMMC. We do not know, however, how these tuning curves might differ for other AMMC neurons, nor which AMMC neurons are involved in the female’s song response. We can therefore only correlate the responses we record with potential behavioral significance. In light of this, do our recordings offer any insights into mechanisms for species-specific song-recognition? Pulse song responses are similar at the level of the JO in D. mel and D. sim, and are likely also to be similar at the level of the AMMC-A1 neurons in these two species. These results may hint at the fact that the detection of conspecific song happens downstream from the AMMC. On the other hand, measurable DC shifts (sustained depolarization) in the AMMC-A1 membrane potential occur in response to pulse songs with IPIs ≤ 35 ms (estimated using the same simulation described in Fig. 6, but instead drawing from the measured AMMC-A1 decay times while systematically varying IPI). If this elevation of the membrane potential is sufficient to support neurotransmitter release (e.g., by causing sustained Ca++ entry into the presynaptic terminal of the AMMC neurons), it could signal the preferred IPI. However, according to this hypothesis, the D. mel IPI range (30 ms ± 5 ms (Cowling DE, 1981)) would also produce a DC shift in D. sim AMMC-A1 neurons, which likely have similar decay constants (Fig. 6). This might not pose a problem to D. sim females, as, due to large differences in pheromone profiles, D. mel males rarely, if ever, sing to them (Moulin et al., 2004). That is, decay times may not have diverged in D. sim, due to a lack of evolutionary pressure to do so. The broad range of D. sim IPIs overlaps that of D. mel; therefore D. sim males, which court D. mel females, occasionally produce IPIs that can drive DC shifts in D. mel AMMC neurons. D. mel females could then use this DC shift (driven more reliably by D. mel song) to signal the difference between the two species. Our model could be validated by comparisons in other species pairs that hybridize in the wild, but with different pulse songs, such as D. yakuba and D. santomea or D. pseudoobscura and D. persimilis (Barbash, 2010), and the development of genetic tools in other Drosophila species should permit such experiments (Holtzman et al., 2010). The mechanism we propose suggests that early auditory neurons (first and second order neurons) already contain sufficient information for making distinctions between species – there is precedent for this in both the grasshopper (Machens et al., 2003) and cricket (Nabatiyan et al., 2003) auditory systems.
In summary, we find that responses to fly song are well-matched between the JON population and AMMC-A1 neurons, and that passing the JO population response through a simple linear filter recapitulates the AMMC-A1 response to song. AMMC-A1 neurons thus filter out information about pulse frequency that is encoded by the JONs (in the frequency doubling the JONs produce at the population level). Put another way, the JONs set the range of frequencies and intensities that the auditory system can respond to, and AMMC neurons, one synapse downstream from the JONs, filter the JON input to primarily preserve information about pulse timings, which are known to have behavioral relevance during courtship (Bennet-Clark and Ewing, 1969). The fly auditory system may thus turn out to be an excellent model for studying the neural encoding of temporal patterns and sequences. By characterizing tuning in fine detail for one neuron type within the AMMC, our study sheds a first light on how sound is processed along the early stages Drosophila auditory pathway. Future studies should be directed at determining i)if all AMMC neurons are similarly tuned and if so, what synaptic properties between JONs and AMMC neurons establish these tuning curves, and ii)the identities and response profiles of downstream neurons in the Drosophila auditory pathway that read out these auditory codes from the AMMC.
We are extremely grateful to Gilles Laurent for critical advice on and funding for this project, to members of the Laurent lab for valuable feedback, to Mark Konishi for donating equipment for fly song recordings, and to Hidehiko Inagaki for help with anatomical characterization of the GAL4 lines used in this study. We also thank Azusa Kamikouchi and Joerg Albert for comments on this manuscript. This work was supported by NIH NIDCD RO1-DC7652 (to Gilles Laurent), funding from the Max Planck Society and a NSERC PGS-M award (to ST), and a Helen Hay Whitney postdoctoral fellowship and startup funds from Princeton University (to MM).