Temporal processing properties of auditory DUM neurons in a bush-cricket

Abstract

Insects with ears process sounds and respond to conspecific signals or predator cues. Axons of auditory sensory cells terminate in mechanosensory neuropils from which auditory interneurons project into (brain-) areas to prepare response behaviors. In the prothoracic ganglion of a bush-cricket, a cluster of local DUM (dorsal unpaired median) neurons has recently been described and constitutes a filter bank for carrier frequency. Here, we demonstrate that these neurons also constitute a filter bank for temporal patterns. The majority of DUM neurons showed pronounced phasic-tonic responses. The transitions from phasic to tonic activation had different time constants in different DUM neurons. Time constants of the membrane potential were shorter in most DUM neurons than in auditory sensory neurons. Patterned stimuli with known behavioral relevance evoked a broad range of responses in DUM neurons: low-pass, band-pass, and high-pass characteristics were encountered. Temporal and carrier frequency processing were not correlated. Those DUM neurons producing action potentials showed divergent processing of temporal patterns when the graded potential or the spiking was analyzed separately. The extent of membrane potential fluctuations mimicking the patterned stimuli was different between otherwise similarly responding neurons. Different kinds of inhibition were apparent and their relevance for temporal processing is discussed.

Introduction

Insect species frequently serve as model systems for the neuronal basis of sensory processing, since their nervous system is considered simple in comparison to vertebrates (e.g., Comer and Robertson 2001; Hedwig 2014). However, more detailed investigations may show an unexpected complexity of neuronal networks. As a representative example of Ensifera, auditory processing in crickets has focussed on three neurons with prothoracic origin: one local (Omega neuron 1 = ON1) and two with ascending axon to the brain (Ascending neurons 1 and 2 = AN1, AN2; e.g., Wohlers and Huber 1978). The prime candidate for song recognition, connecting the thorax with the brain (AN1, reviewed in Hedwig 2016), shows rather simple temporal copying properties for cricket songs. Only three additional local auditory neurons have been described for crickets (Wohlers and Huber 1978; Stiedl et al. 1997). In contrast, in a bush-cricket, a cluster of at least 15 local auditory neurons (dorsal unpaired median neurons = DUM neurons; Lefebvre et al. 2018) have been identified in addition to ON1. DUM neurons constitute a filter bank for carrier frequency (Lefebvre et al. 2018). Song carrier frequencies differ widely between bush-cricket species, and these differences likely have some role in song recognition (Schul et al. 1998; Schul and Patterson 2003; Montealegre-Z 2009; Heller et al. 2015). However, temporal properties of songs differ between bush-cricket species as well, and at least some are relevant for song recognition (Zimmermann et al. 1989; Dobler et al. 1994b; Schul 1998; Korsunovskaya 2009; Schul et al. 2014). As such, knowledge about the processing of temporal patterns at the first central stages in Orthoptera is currently limited. In grasshoppers, at least seven local neurons with differing temporal response properties have been described (Marquart 1985; Stumpner and Ronacher 1991; Stumpner et al. 1991). In bush-crickets, clear temporal response differences have been described only for neurons with axons projecting from the prothoracic ganglion into the brain (Rheinlaender 1975; Stumpner and Molina 2006; Triblehorn and Schul 2009). However, it is unknown how these response properties arise in the thoracic neuronal network and whether local neurons in the bush-cricket’s prothoracic ganglion differ in their temporal properties.

Previous studies have shown that DUM neurons of the bush-cricket Ancistrura nigrovittata differ considerably in their responses to carrier frequency (Lefebvre et al. 2018). In the present study, we investigated this population of DUM neurons and found strong differences with respect to their temporal processing. This finding may also be of interest for other systems, since auditory DUM neurons are not specific to bush-crickets. DUM neurons were also described in grasshoppers (e.g., Stumpner and Ronacher 1991) and recently in flies (Drosophila melanogaster: IVLP–IVLP–PN1 or AMMC–AMMC PN1 or aLN(m); Lai et al. 2012; Vaughan et al. 2014; Matsuo et al. 2016). DUM neurons are organized in clusters. In Drosophila, Clemens et al. (2015) described these neurons (named AV4 in that study) as members of a group of central neurons that process behaviorally relevant information of the male song. The group exhibits linear temporal filters of various filter widths and constitutes a filter bank for temporal features.

Ancistrura nigrovittata possesses a sexually dimorphic song that provides the basis for mate finding. Males produce a specific pulse pattern at 16 kHz that is answered by females with a 28 kHz click. Thus, their song differs between sexes in carrier frequency (Dobler et al. 1994b) and in temporal patterns. The male produces a group of pulses (typically eight pulses of 7 ms duration with 22 ms pauses) and after an approximately 350 ms delay produces a single pulse that serves as a trigger for the female reply (Heller and von Helversen 1986; Dobler et al. 1994a, b). Females reply with a short click response to the trigger pulse if the previous pulse group is close to the species-typical carrier frequency and temporal parameters (Dobler et al. 1994a). Related species consistently differ in the temporal parameters of their songs (e.g., Stumpner and Meyer 2001; Korsunovskaya 2008, 2009; Heller et al. 2015). Here, we report that the population of DUM neurons in the bush-cricket A. nigrovittata constitutes a filter bank for temporal parameters, a feature that may also be relevant in many other bush-cricket species.

Materials and methods

Animals

Both sexes of the bush-cricket A. nigrovittata (Brunner von Wattenwyl) were studied and were either wild-captured from Northern Greece or laboratory-reared from eggs collected from the laboratory culture each summer (the majority of our population). Our experiments complied with the laws of Germany, Directive 2017/63/EU, and the “Principles of Animal Care”, publication No. 86-23 revised 1985 from the National Institutes of Health. The results presented here are from the same set of experiments as in Lefebvre et al. (2018). From 277 animals, we had 682 recordings from putative (cells not stained but recorded in the cluster of DUM somata; see Lefebvre et al. 2018 for arguments) or identified DUM neurons. Of those, 472 recordings were in males and 210 in females. We recorded up to 11 neurons in one individual. Data used for the final analyses came from 183 at least partly stained neurons and 102 unstained neurons. In addition, 22 sensory neurons from 7 females and 6 males were evaluated.

Recording and staining techniques

The electrophysiological methods used have been described in detail by Stumpner (1997) and Lefebvre et al. (2018). In short, an anesthetized animal was fixed using a wax–resin mixture in a normal (inverse) standing position. The prothoracic ganglion was exposed and stabilized, and collagenase (Sigma) was briefly placed on the ganglion to facilitate penetration of the glass capillary. Fielden’s saline (Fielden 1960) was used to bathe the ganglion during recordings. Thick-walled borosilicate glass capillaries were filled with 5% Lucifer yellow CH (Sigma or Molecular Probes), 5–10 mM Alexa Hydrazide 488, 555, 633 (Life Technologies, Darmstadt, D), 5 mM CF555, CF633 (Biotium, Hayward, CA, USA), or in few cases with 1 M potassium acetate (SIGMA). Recordings were amplified (NPI BA-1S, NPI, Tamm, Germany, or custom-made model) and recorded at a 20 kHz sampling rate (CED power 1401, Cambridge Electronic Devices, Cambridge, UK) using Spike2 (CED) software. Following physiological characterization, dye was ionophoretically injected (0.5–15 min) and ganglia were fixed in paraformaldehyde (4% in buffer, pH 7.4, > 30 min). Ganglia were dehydrated and cleared in methylsalicylate for visualization with a confocal microscope (Leica SP2 AOBS or Leica SP8 AOBS). Projections of confocal scans were calculated using the Leica program.

Physiological data evaluation

Recordings were analyzed using custom-made scripts in Spike2. Several parameters were evaluated (Fig. 1). Graded potential (GP) in mV × ms was calculated as the integral of the potential (action potentials clipped) between 10 ms after stimulus onset (due to the longer response latency) and 30 ms after stimulus offset relative to the respective reference-null value (membrane potential during 100 ms preceding a stimulus; for details see Lefebvre et al. (2018). Mean membrane potential (MP) in mV was calculated as the average value of the membrane potential (relative to reference-null) 50 ms after stimulus onset (excluding the phasic onset) to stimulus offset. With 300-ms block stimuli, the MP was calculated in the phasic (10–60 ms after stimulus onset) or tonic (110–310 ms after stimulus onset) response part. Standard deviation (SD) of the mean potential in mV was calculated between 50 ms after stimulus onset and stimulus offset. SD was used as a proxy for the ability of a cell to represent a pulsed pattern with its membrane potential, where the greater the effect of the pulse pattern on the response, the greater the SD. The number of action potentials (AP) was only counted in recordings with unequivocal AP production. Fits for the averaged membrane potential (DUM neurons) or instantaneous AP frequency (sensory neurons) in response to constant 300-ms block stimuli were calculated in a window from 40 to 290 ms after stimulus onset using Origin Pro 8.5G (Northampton, MA, USA). Only single or double exponential fits were calculated using 100 iterations. Origin Pro was used for statistical testing.

Fig. 1

Sketch indicating the types of measurements made in the present study. Area (integral) of the graded potential (GP) relative to the reference-null value; Mean potential (MP) = averaged value of the individual data points of the membrane potential; SD = standard deviation of the mean potential. Below the sketch, a symbolized stimulus and time frames used to evaluate the different parameters are shown

Acoustic stimulation

Our setup was identical to the one described in Lefebvre et al. (2018) where two speakers were placed to the left and right of the insect. Acoustic stimuli were generated with a computer-controlled interface (Lang et al. 1993). Carrier frequency tuning was tested using 50 ms stimuli (with 1.5 ms rising and falling ramps) presented at 250 ms intervals; each stimulus was repeated five times. Carrier frequencies ranged from 3 to 46 kHz at 50 or 70 dB SPL (sound pressure level re 2 × 10⁻⁵ Pa). In most experiments (Fig. 2), a rough test for frequency tuning was presented using 11 stimuli (50 ms, 250 ms interval) with increasing frequency between 3 and 50 kHz at 70 dB SPL. Aspects of temporal processing were studied with three programs. In the first, 300-ms white noise stimuli (2 ms ramps) were presented at increasing sound amplitudes from 30 to 84 dB SPL (7 steps) with a 997 ms interval and 5 repetitions. In the second, an approximately 210 ms-long pulse group from white noise (as used in Dobler et al. 1994a to test behavioral responses of this species) was repeated five times and separated by pauses of 300 ms duration. Pulse durations varied from 2 to 25 ms with rise/fall times of 0.5 ms. Pauses between pulses were constant, so that the number of pulses varied to fill 210 ms. Pulse durations were presented in pseudorandomized order. One stimulus (8 pulses of 7 ms duration) was roughly comparable to the natural calling song of the male. In the third, the program kept pulse duration constant at 7 ms (ramps 0.5 ms), while pauses between pulses varied from 2 to 61 ms (pseudorandomized order). Thus, the number of pulses varied accordingly within the 210 ms pulse group. The acoustic system had an accuracy of ± 2 dB and was calibrated with a continuous sound using an amplifier (type 2610, Bruel & Kjær, Nærum, DK) and Bruel & Kjær condenser microphones (1/2” or 1/4”; calibrated with a pistonphone 4230) directed toward the speaker from the position of the insect.

Fig. 2

Three DUM neurons recorded and stained in one male. a Morphology of the three neurons stained with Lucifer yellow (a1, likely type “loops”), Alexa 555 (a2, type “loops”), and Alexa 633 (a3, type “narrow”; see Lefebvre et al. 2018). Somata not shown. Bluish turquoise structures in a and b are from autofluorescence of tracheae. b Responses of the three neurons to sequential 50 ms stimuli of increasing carrier frequency at 70 dB SPL. Colors correspond with the stainings in a b1a, b2a, b3a are complete responses and b1b, b2b, b3b are extended views of the response to 24 kHz (= stimulus no. 7). c Responses of the three DUM neurons (c1–c3) to three temporal patterns with different pause durations (e.g., 7/2 ms: 7 ms pulse, 2 ms pause) at white noise, 70 dB SPL. 7/22 ms corresponds to the conspecific male song. d Responses of the three DUM neurons (d1–d3) to two temporal patterns with different pulse durations of white noise

Results

Neuronal diversity

Three DUM neurons from one male (Fig. 2a) varied in their frequency tuning (Fig. 2b) and, at the same time, responded differently to variations in pause and pulse durations of calling song models (Fig. 2c, d). Even responses to identical 50 ms block stimuli (24 kHz, 70 dB SPL; Fig. 2b1b, b2b, b3b) differed fundamentally. Although all neurons had a strong initial phasic response, only neuron 1 (Fig. 2b1b) had a distinct and ongoing tonic component. Neuron 2 (Fig. 2b2b) had a slowly diminishing tonic part and an inhibitory postsynaptic potential (IPSP) in the rising flank (barely visible at this magnification, but see Fig. 9). Neuron 3 (Fig. 2b3b) expressed a prominent hyperpolarization after the phasic onset. Accordingly, the responses to varied temporal patterns also differed. The most consistent response and best representation of the pulsed stimulus pattern with the longest pause duration was seen in neuron 1 (Fig. 2c1; each pulse elicited an action potential). In contrast, neuron 3 had the weakest and least reliable response to subsequent pulses (Fig. 2c3). For the shortest pulse durations, neuron 3 responded strongest (Fig. 2d3) and neuron 2 responded the weakest (Fig. 2d2). With the longest pulse durations, neuron 3 showed relatively short excitation due to its postexcitatory inhibition. This inhibition also suppressed the response to all subsequent pulses. The other neurons responded much stronger to the longest pulse duration than to the shorter pulses (Fig. 2d). In total, neuron 1 showed strong adaptation and responded best to sufficiently separated pulses. Its responses depended least on pulse duration. Neuron 2 adapted less to the stimulus than neuron 1 and produced APs only during the initial phasic onset response, which was lost with short pulse durations. Neuron 3 responded best to very short, clearly separated pulses. It, otherwise, showed a strong delayed inhibition, which may have suppressed ongoing responses (following the short onset excitation) to sustained stimuli or higher pulse rates. Overall, high neuronal diversity, as shown by multiple DUM recordings (up to 11 cells) within a specimen, was found in both sexes of A. nigrovittata. Based on our large data set, there were no indications of sex-specific differences in DUM neurons or their processing of acoustic signals, which was found for all other auditory neurons in A. nigrovittata in previous studies (Stumpner and Molina 2006; Ostrowski and Stumpner 2010).

Temporal processing: variation of pause duration

The behavioral preference of A. nigrovittata females to varied temporal parameters of the male song exhibited band-pass characteristics for stimuli with varied pause durations (gray areas in Fig. 3a, b; Dobler et al. 1994a). DUM neurons may show band-pass characteristics in their graded potential responses as well (Fig. 3a). In a group of 285 analyzed cells, 39 cells showed band-pass properties, i.e., a clear peak at one of the tested pause durations between 5 and 35 ms (and at least 20% response reduction at shorter and longer pause durations). About a quarter (11) of these 39 cells were tuned to 22 ms, which is behaviorally most attractive. Another 98 cells showed low-pass or high-pass characteristics (Fig. 3b). These cells had above 90% responses at the shortest or longest pause durations tested and consistently lower responses at all other pause durations tested. DUM neurons with strongest responses at 2 or 5 ms pause (therefore constituting a high pass for pulse rate) were by far the most frequent cell types (79 of 98 cells). High-pass neurons for pause duration were comparatively rare (8 of 98 cells). Rare types of neurons might appear to be an exception. However, recordings of multiple neurons in one specimen, as shown in Fig. 3c, exhibited the whole spectrum of low-pass, band-pass, and high-pass neurons. For the 39 cells with well-defined band-pass properties (Fig. 3a), we found no correlation between temporal and frequency processing (y [best pause in ms] = 0.26 × [best frequency in kHz] + 22.54, R² = 0.076, n = 39).

Fig. 3

Responses (area of the graded potential) of DUM neurons to stimuli with varied pause durations, while pulse duration (7 ms) and duration of the pulse group (about 210 ms) remained constant (see Fig. 2c). Gray areas in a and b show the average behavioral response probability of five females (from Dobler et al. 1994a). a Response functions showing band-pass characteristics, i.e., a clear peak in their response (< 80% of the peak response at higher and lower pause durations). Means, SE, and numbers of individual cells indicated. b Response functions showing low-pass or high-pass characteristics, i.e., more (low pass) or less (high pass) than 90% response up to a certain pause duration. Means, SE, and numbers of individual cells indicated. c Eight DUM neurons from one male showing low-pass, band-pass, and high-pass characteristics

The majority of DUM neurons produced APs that can be seen in recordings from the soma (Lefebvre et al. 2018). Since the two types of signals (GPs and APs) in a local neuron may transfer different aspects about the stimulus to postsynaptic cells, the temporal processing properties of both response types were compared (Fig. 4). Figure 4a shows the spiking responses of six DUM neurons with a graded response peaking at 5 ms pause duration. However, the spiking responses (with maximum AP numbers per stimulus between 5.4 and 14.2) peaked at all tested pause durations (except for 2 ms). Figure 4b shows responses of 40 DUM neurons grouped according to their peak spiking response. Even though some response functions were relatively rare, all tested pause durations were represented in a band-pass characteristic. Among these neurons, GPs peaked mostly at short pause durations (i.e., with high pulse numbers), and AP numbers (maximum between 4.2 and 23.6) peaked at more distributed values, although they peaked predominantly at shorter pause durations, as well (Fig. ESM 1). We found no correlation between GPs and AP numbers.

Fig. 4

Comparison of spiking responses (at least four APs at the strongest response) and graded responses of DUM neurons to variations of pause duration. a Spiking responses of six DUM neurons from six specimens peaking at 5 ms pause duration in their graded responses (arrow). Peak spiking of these neurons was observed at six different pause durations. b Spiking responses of 40 neurons grouped according to their peak activity. Means, SE, and numbers of cells indicated. c Number of APs per pulse in dependence of pause duration from 40 DUM neurons. The broken line represents one AP per pulse. Black lines depict 30 cells responding with one AP per pulse (< 50% variation at the longest pause) at a certain pause duration (and longer). Gray lines depict cells that produced at least 1.5 APs or much less than 1 AP per pulse at the maximum pause duration

In general, there was a negative correlation between maximum AP number and the pause duration at which this response occurred (R² = 0.54, data not shown). This result means that stimuli with fewer pulses evoked fewer APs. The majority of neurons (30 of 40) produced a maximum of one AP per pulse (from some pause duration on; Fig. 4c). Thus, these cells entrained to the pulse rhythm as soon as the pauses (which varied between 5 and 61 ms) were sufficiently long. At very short pauses, only a minority of cells produced about one AP per pulse (Fig. 4c; few gray color-coded cells). These cells reached a maximum of two APs per pulse with longer pause durations.

The responses to block stimuli also indicated different temporal coding modes in clusters of DUM cells (Fig. 5a), varying from tonic (cell 1) to phasic-tonic (cells 2, 4, 6) to mainly phasic responses (cells 3, 5; all six DUM neurons recorded in one male). Cells 3 and 6 additionally showed signs of inhibition, where cell 6 had a short inhibition directly after the onset (see Lefebvre et al. 2018 and Fig. 9) and cell 3 had a long-lasting hyperpolarization after the phasic response. Figure 5b presents a quantitative assessment of the different responses by measuring the mean potential in the phasic (first 50 ms) and tonic part (last 200 ms; indicated at the stimulus in Fig. 5a) with increasing sound amplitudes. The phasic responses showed saturating or optimum-type dependencies. The amplitude of most tonic responses was below phasic responses (except for two neurons). Cell 3 showed a negative potential in the second part of the response at all amplitudes that were above threshold. In line with the divergent responses to block stimuli, responses to pulsed stimuli also showed striking differences (Fig. 5c; this stimulus mimics a male song). Responses may not represent the temporal structure of the song (cell 3) or give a near perfect representation of the pulse-pause pattern with one AP per pulse (cell 5). Such response differences between cells 3 and 5 existed, even though their response to the block stimulus was similar (Fig. 5a). In addition to GPs (total area of the response), Fig. 5d shows the SD (measured in the last 150 ms of the 200-ms stimulus, i.e., after the stereotypic phasic response) of the membrane potential to quantify how well a neuronal response is mimicking a given pulse pattern. The bigger the SD of a cell at a given pause, the better the representation of the pulsed stimulus (e.g., cell 5 in Fig. 5c was best at 22 ms pause). The GP in three of the six cells dropped with increasing pause duration as does stimulus energy (represented by a solid gray line in Fig. 5d). In two cells, GP dropped at slightly longer pauses; and in one cell (cell 3 from Fig. 5a), a reverse relationship was seen. Conversely, the SD of cells 3–6 increased with increasing pause duration, but the pause duration with the fastest rise in SD varied between cells. The tonically responding cell 1 produced a highly variable GP (even to block stimuli; Fig. 5a), and correspondingly, its SD was high with short pauses. Cell 2 showed a more complex SD and responded strongest at shortest and longest pause durations.

Fig. 5

Response characteristics of six DUM neurons from one male to block stimuli (75 dB SPL, white noise) and stimuli comprising pulses (white noise, 70 dB SPL). a Single responses to block stimuli. Numbers and colors label the individual cells. The dashed arrows under the stimuli show the evaluation windows for (b). b Mean potential of responses to block stimuli. Solid lines show mean potential during first 50 ms (starting 10 ms after stimulus onset), and broken lines show the mean potential during the last 200 ms (again with 10 ms latency). Data were normalized to the maximum positive values for both measures. Negative values represent hyperpolarization. Colors and numbers next to the broken lines as in (a). c Single responses to a pause pattern representing the male song (white noise, 70 dB SPL; four cells that represent the variation of responses). d Standard deviation (SD) of the mean potential between 50 and 200 ms after stimulus onset (solid lines) and graded potential (GP, broken lines). Colors and numbers next to the broken lines as in a). The solid gray line represents the stimulus energy

In a bigger group (136 cells), the majority of cells had their largest GP (total area) with short pause durations, but the opposite was true for SD (Fig. 6a). However, when grouping cells with an SD peaking at 61 ms (Fig. 6b), not all cells produced the largest GP with the shortest pause. The same was true when cells were grouped according to other SD or GP peaks (Figs. ESM 2a and b). Consistent with these data, there was no correlation between the peak pause retrieved from GP and peak pause retrieved from SD of the same cells (Fig. ESM 2c; R² = 0.037 for linear regression, nonlinear regressions increase the coefficient only marginally, if at all). This lack of correlation was not because of the different analysis windows used (GP calculated for the whole response vs. SD calculated without the phasic response; see Fig. 1). The mean MP calculated for the identical time window as that used for SD was highly correlated with the GP calculated for the complete response (R² = 0.97, n = 74; Fig. 6c). These data also demonstrated that the phasic onset peak (included in GP but not in MP) was indiscriminate for responses to different pause durations and the decisive processing that occurs thereafter. Finally, there was no correlation between the pause duration at the largest SD (more exactly, 75% of the largest SD to reduce noise in the data) and the best frequency of the same neurons (Fig. ESM 2d).

Fig. 6

Responses (graded potential of the whole response and standard deviation of the membrane potential from 50 to 200 ms after stimulus onset) of DUM neurons in response to stimuli with varied pause duration. a Number of cells with maximal responses at a given pause duration with their graded response (GP) or standard deviation (SD) of the response. b Responses from 62 cells that show largest SD (black lines) at 61 ms (longest pause tested). Gray lines depict the GPs of the same cells. c Bubble diagram relating to the pause duration eliciting the strongest GP (integral of the potential between 10 and 230 ms after stimulus onset) and the largest mean membrane potential (MP, measured between 50 and 200 ms after stimulus onset). Dashed line represents the linear regression. Bubble area defines sample size (between 1 and 32)

Temporal processing: adaptation to block stimuli

Responses of DUM neurons to block stimuli (see spike traces in Fig. 5a) were averaged (five repetitions) and a second-order exponential was fitted to the decaying MP (Fig. 7). The four examples in Fig. 7a come from DUM neurons recorded in one male in response to a stimulus of 75 dB SPL. In all four responses, the exponential fit described the decaying portion of the phasic response quite well (R² all above 0.8). However, the fits and their first time constant differed considerably between cells. Variations of the second time constant were even larger (of 123 fits, c1 was 31 times larger than c2, median difference 0.3 ms, and c2 was 92 times larger than c1, median difference 25.9 ms). Figure 7b illustrates the phasic character resulting from the decaying potential in response to a 500 ms white noise block stimulus at 70 dB SPL in two DUM neurons from one male. The lower (gray) cell showed a constant tonic potential that was about 40% of the phasic onset amplitude. APs were produced during the phasic part only. The black cell had a slowly declining tonic potential, starting at about 55% of the phasic part. It produced APs throughout the whole response with a decreasing frequency that corresponded to the declining MP.

Fig. 7

Responses of DUM neurons to block stimuli. a Averaged responses (n = 5) of four cells (from one male) to a 300 ms block stimulus (white noise, 75 dB SPL). A double exponential fit is overlaid between 40 and 290 ms after stimulus onset (c1, c2 = first and second time constant). b Responses of two DUM neurons from one male to a 500 ms block stimulus at 70 dB SPL, white noise. The dotted lines help visualize the constancy (gray) or slow decline (black) of the potential in the tonic response phase. c First time constants from fits (as shown in a) of 114 DUM cells from 79 specimens plotted over the R² of the fit with stimulus white noise, 75 dB SPL. In addition, first time constants from single or, where possible, double exponential fits from 15 receptor cells (fits of the AP frequency) are shown. d Logarithm of the ratio of the membrane potential (MP, averages of five repetitions) in the phasic (10–60 ms after stimulus onset) and tonic part (110–310 ms) from three females and two males with multiple DUM recordings (between 5 and 11 per specimen, together 39 cells) at the given amplitudes. Negative ratios not included. The boxes show the median and 25% quartiles, the whiskers extend to 5% and 95%, and the asterisks show the minimum and maximum values. The median values are given and the dashed lines indicate the de-log ratios

A decaying MP may depend on adaptation of the input into DUM neurons, namely auditory receptors, and on other intrinsic mechanisms of DUM neurons. Graded responses of auditory receptors in A. nigrovittata are not accessible in the preparation used for studying DUM neurons. Therefore, APs were recorded in the prothoracic ganglion. In Fig. 7c, first time constants from fits to AP frequency adaptation in 16 sensory neurons (15 displayed in the diagram, one time constant > 200 ms) were compared with fits of 120 DUM membrane potentials at 75 dB SPL (114 are included in Fig. 7c; three had time constants larger than 200 ms, three had negative time constants). AP frequencies of sensory cells typically varied more than MPs in DUM neurons. Therefore, no restriction for R² was made. Instead, time constant one was plotted over the R² value of the fit. It was immediately obvious that DUM neurons adapted much faster than sensory neurons (Mann–Whitney test, p < 0.001) with an accumulation of values around 5–15 ms. This result indicated that DUM neurons had additional mechanisms for adaptation and did not merely represent the adaptation of the presynaptic neuron.

The ratio of the averaged potential in the first 50 ms and last 200 ms of a response to 300-ms block stimuli directly indicated the phasic characteristics of the response. Data from five specimens for which multiple DUM neurons were recorded at different sound amplitudes (Fig. 7d) demonstrated a large spread of values. Therefore, log-transformed data are reported in a box-plot (ranges of linear values are indicated). A perfectly tonic response would have a log value of zero (a normal value of 1), which corresponds with no adaptation. Larger values indicate a phasic response. All medians, and even all 25% quartiles, were larger than zero at all sound amplitudes tested. Only a few single values were below zero, indicating a greater response at the end of a stimulus. At the same time, phasic responses occurred that were 5–10 times stronger than the tonic part. These data support the notion that DUM cells, in general, exhibit mainly phasic responses with the phasic part being on average 1.5–1.6 times stronger than the tonic part.

Temporal processing: variation of pulse duration

In response to varied pulse duration, GPs differed between DUM neurons from one specimen (Fig. ESM 3a). In a larger sample, the responses were categorized in seven groups (Fig. 8a). The neurons showed either a weak band-pass characteristic for certain pulse durations or strongest responses with longest pulse durations. Fourteen of seventy-four cells had best responses for pulse durations similar to the male song (7 ms), which is also preferred in female behavior (Dobler et al. 1994a). Nine cells had no preference for any pulse duration. A weak correlation (R² = 0.25) existed between pulse duration tuning and pause duration tuning of the GPs (Fig. 8b). Neurons responding best to long pulse durations also responded well to short pause durations, indicating that a high stimulation energy (or duty cycle) was most effective. Neurons that responded best to long pause durations tended to prefer short pulses. These neurons had a strong phasic onset followed by a more or less obvious inhibition with high pulse rates or long pulses. The SD of the mean potential rose with longer pulses, indicating a better representation of longer pulses in the MP. There was a weak positive correlation between the SD of pauses and pulses (R² = 0.11; Fig. ESM 3b).

Fig. 8

Responses of DUM neurons to stimuli with varied pulse duration, while pause duration (22 ms) and duration of the pulse group (ca. 210 ms) remained constant (see Fig. 2d). a Graded response functions (means and SE) of 74 DUM neurons sorted according to their maximal response at certain pulse durations (i.e., band pass at one pulse duration or low or high pass) or completely unspecific response (i.e., more than 80% at all tested values). b Correlation of the pulse duration eliciting the strongest graded response with the pause duration eliciting the strongest graded response of 93 DUM neurons. Bubble area defines number of cells (between 1 and 27). Function and equation of an exponential regression are shown

Temporal processing: the role of inhibition

About one-third of the recorded DUM neurons had indications for inhibition in their responses (Lefebvre et al. 2018). Two-thirds of those neurons with inhibition had an IPSP in the rising flank of their response (Fig. 9a). This delayed inhibition was seen in each onset response with sufficiently loud stimuli (Lefebvre et al. 2018). It was repeatedly triggered with pulsed stimuli (Fig. 9b). In a test series with decreasing pause duration (at 7 ms pulses), the IPSP at pulse onset became increasingly smaller and then slowed down the rise (less steep) of the excitatory postsynaptic potential before it was lost with the shortest pauses tested (2 and 5 ms). A comparison of cells with (32 cells) and without (59 cells) inhibition indicated that the presence of inhibition had no strong effect on peak pause durations (Fig. 9c).

Fig. 9

Inhibition and temporal processing. a Individual responses of three DUM cells from one female to the onset of a 500 ms stimulus (white noise, 70 dB SPL). The gray arrows mark the hyperpolarizing deflections in the rising flank of the excitatory response. b Averaged responses from one of the cells shown in a) to the onset of stimuli comprising 7 ms pulses and variable pause length (see Fig. 2c; pause duration indicated on the right). Averaged responses either to the onset of all 35 stimuli (0 ms pause) or to individual pulses after the initial phasic excitation (therefore, the number of averaged responses varies between 15 and 76). The gray arrow indicates the occurrence of the IPSP at longer pause durations. c Distribution of most effective pause durations as judged from the graded response in cells with inhibition (n = 32, gray) and without inhibition (n = 59, black). d Responses of two neurons to five different temporal patterns showing phasic excitation followed by strong inhibition. e Excitatory (solid lines, nine cells) and inhibitory (dashed line, mean of nine cells plus standard error) graded potentials of DUM neurons with strong inhibition after the phasic onset excitation for stimuli with varied pulse durations (pause constant at 22 ms). f Standard deviation of the mean graded potential between 50 and 200 ms after stimulus onset (excluding the phasic onset response; nine cells with the most consistent inhibition evaluated). Each line represents one neuron

A relatively small number of recorded DUM neurons (< 3%) had a strong hyperpolarization after phasic excitation at the onset of the response for longer stimuli (Fig. 9d; see also Fig. 2c3 or Fig. 5a, neuron 3). In multiple DUM recordings in one specimen, this response type was never encountered twice, indicating that only one DUM neuron with these properties might exist per specimen. In general, this inhibition got stronger with longer pulse durations and with shorter pauses between pulses (i.e., high stimulus energy). Such filtering resulted in a predominant inhibition of these neurons by the species-specific song pattern, while very short pulses elicited relatively greater excitation (Figs. 2c, d, and 9d). All ten tested neurons showed the strongest excitation with the shortest pulse duration and the strongest inhibition with the longest pulse duration (Fig. 9e). At the same time, all neurons showed the largest potential fluctuations (SD) with the longest pause duration tested (phasic onset excluded; Fig. 9f), indicating that pulses were able to trigger new excitatory responses when pauses were longer. Nevertheless, the amount of excitation (GP) varied between specimens, showing band-pass or low-pass characteristics for stimuli with varied pause durations (Fig. ESM4). Only one of these cells had a best frequency (70 dB SPL) at 16 kHz, whereas all others peaked in the ultrasound.

Discussion

Most DUM neurons produced a strong phasic onset response that was followed by a diminished tonic component. Despite this common property, their responses to temporal patterns differed widely and were at least as divergent as their carrier frequency processing, which gave them the attribute of a filter bank for carrier frequencies (Lefebvre et al. 2018). Sex differences were not encountered.

Diversity of temporal filters

Female A. nigrovittata evaluate the temporal pattern of artificial songs and respond strongest to stimuli comprising the species-specific durations of pulses, pauses, and pulse groups (Dobler et al. 1994b). When pause duration was varied with otherwise species-typical song parameters (7 ms pulses, about 210 ms pulse group), DUM neurons exhibited a large diversity of responses (Figs. 1, 2, 3, 4, 5). Low-pass, high-pass, and band-pass properties were found with peaks at all tested values. This diversity also existed in DUM neurons from one specimen, whether male or female. Therefore, DUM neurons represented a temporal filter bank for pause durations (or repetition rates; see Figs. 3, 4). Among these, low-pass and high-pass neurons formed the largest fraction (when the integral of the GP was evaluated). Band-pass filters for pause duration that were similar to behavioral filters (Dobler et al. 1994a) were underrepresented.

Temporal and carrier frequency processing was not correlated. In particular, neurons that responded best to the species-typical pause duration of 22 ms did not preferably respond best to the species-typical male song frequency of 16 kHz. In fact, this combination was very rare. As a general conclusion, there was no indication that populations of DUM neurons specifically filter the conspecific song. This finding is not surprising considering the likely inhibitory nature of these neurons for further processing (Lefebvre et al. 2018).

When AP numbers were evaluated instead of GPs, different filter properties emerged in the same neurons, but a correlation between the two filters did not exist. This finding was true, even though DUM neurons that produced APs only during the phasic onset response were not included in the comparison. As before, the filters based on APs included low-pass, high-pass, and band-pass characteristics. Neurons with temporal filtering properties for the spiking response similar to those of DUM neurons also existed in the cricket brain (Gryllus bimaculatus; Schildberger 1984). In the fly brain (Drosophila melanogaster), band-pass characteristics of membrane responses for interpulse intervals of the male song have been identified in the antennal mechanosensory and motor center (AMMC, neuron aPN1). The filter properties, when used to construct encoder models (Clemens et al. 2015), were able to predict female behavior for song recognition. Furthermore, in the brain of anurans, which produce and process signals similarly to many acoustic insects, various low-pass, high-pass, and band-pass neurons have been described (Rose and Capranica 1983; Hall 1994; Gerhardt and Huber 2002).

Spiking responses of auditory interneurons with axons to the brain varied little (about 20–30%) with variations of pulse duration. Three of five previously described neurons responded better to short pulse durations (Stumpner and Molina 2006). The same was true for many DUM neurons, even though peak responses were found for all pulse durations tested. Neurons that responded best to long pulses were the most frequent and, again, neurons that responded best to the modal value of the male song (7 ms) were underrepresented (Fig. 8). Ostrowski and Stumpner (2014) argued that combined activity of neurons ascending from the thorax might give rise to the behavioral filter for pulse duration.

Proximate mechanisms for filter diversity

Adaptation in DUM neurons and presynaptic neurons

Auditory sensory neurons, in principal, copy the temporal pattern of a signal (e.g., Mason and Faure 2004), but also exhibit adaptation of the MP and AP frequency during longer stimuli (e.g., Gollisch and Herz 2004). Therefore, their response should be classified as a phasic-tonic response. This phenomenon also exists in vertebrates (e.g., Galambos and Davis 1943). Consequently, neurons postsynaptic to afferents are expected to also show phasic-tonic responses, which, at least in part, should arise from their phasic-tonic input (e.g., Benda and Hennig 2007).

The progression of the MP in response to a longer block stimulus was well described by a double exponential fit (for most neurons) when ignoring the initial response peak. AP frequency adaptation in sensory neurons, in general, is sufficiently well described by a single exponential fit (Benda and Herz 2003; Clemens et al. 2018; but see Smith and Zwislocki 1975 for guinea pig afferents). This is true for Ancistrura receptor cells, as well. Adaptation in DUM cells of Ancistrura at a given intensity was much faster than in receptor cells (Fig. 7c). First time constants (typically called τ) below 20 ms were found in more than 80% of DUM neurons but in only one of 16 receptor cells (this one had a very low regression coefficient). Therefore, in addition to adaptation by the input (which is mostly driven by the AP generator; Benda and Herz 2003), other mechanisms must be effective in DUM neurons (e.g., Hildebrandt et al. 2009). The possibility of two independent mechanisms is also in line with the finding that the time course of the DUM potential is better represented by two (instead of one) time constants of the exponential fit. Regression coefficients of fits in sensory cells were generally much lower than those in DUM neurons. The likely reason for this difference is that adaptation in sensory neurons was calculated from instantaneous AP frequency instead of GPs. However, AP frequency in sensory cells is generally thought to have a nearly linear correlation with the underlying GP (e.g., Terzuolo and Washizu 1962). Therefore, a direct comparison of the fits is justified.

Adaptation time constants in DUM neurons varied widely between (and within) specimens. Similar differences have been described for frog auditory nerve fibers (Feng et al. 1991). The generally fast and strong adaptation was visible as a phasic-tonic response of DUM neurons (Figs. 2, 5, 7). The ratio between the phasic and tonic portion of a representative cell group (Fig. 7d) was similar to ratios in ferret auditory afferents (Sumner and Palmer 2012). The speed of adaptation will likely determine the number of filtering properties seen with patterned stimuli (see Benda and Herz 2003). Recently, fast sound intensity adaptation of Drosophila was described as enhancing the encoding of conspecific sound signals (Clemens et al. 2018). The positive impact of different adaptations on pattern processing may arise at the expense of sound localization (Hildebrandt et al. 2015). DUM neurons, however, are non-directional and cannot have a function in sound localization (Lefebvre et al. 2018). In the bush-cricket Neoconocephalus triops, dendritic mechanisms have been identified that account for stimulus-specific adaptation (Triblehorn and Schul 2013), and dendritic compartmentalization is an important aspect in the tonotopically organized neuropil (Prešern et al. 2015). Such mechanisms were not studied in Ancistrura, but DUM neuron anatomy and organization of the neuropil would allow similar mechanisms to act in DUM processing.

Coding mechanisms in DUM neurons

Various analysis methods were used in the present study to extract information about how DUM neuron responses depended on temporal patterns (see “Materials and methods”). Only the combination of these methods revealed the variety of responses of DUM neurons. Since nothing is known about the influence of DUM cells on postsynaptic neurons and networks, each of the different analyses may have its own justification.

The integral of the GP may provide some indication of the information transfer of a neuron’s response in the auditory network (e.g., Juusola et al. 1996). For temporally structured stimuli, however, the structuring of the GP may also play a role. A stronger response in a subsequent neuron may be caused by a generally increased MP, but it could also be caused by a larger number of onset responses during a stimulus. These onset responses were detected in the present study by measuring the SD of the MP. The largest SDs (resulting from the largest membrane fluctuations elicited by pulses and pauses) were found in most neurons at the longest pause duration tested (e.g., Fig. 6b). However, the integral of the GP (or the average MP, since the two are closely correlated) of the same cells may also vary widely. Cells that peaked at the same pause duration with both measures were an exception; nevertheless, the broad range of SDs for pause duration across different cells (Figs. 5, 6) demonstrated that clear dependencies existed and differed between cells. Again, these dependencies were not correlated with carrier frequency processing.

Many DUM neurons produced no or small APs (Lefebvre et al. 2018). Many neurons that reliably produced APs did so only at the phasic onset of the response. In such neurons, numbers of APs varied little with changing stimulus parameters. Many neurons that continued eliciting spikes after the initial phasic response had highly divergent filtering for temporal patterns in their graded and spiking response (see above). If both GP and spiking carry relevant information for auditory processing in the network, their impact must be quite different. Real measurements in crabs (Wildman and Cannone 1990) and model calculations (Kretzberg et al. 2001) have shown that the information transfer in graded and spiking responses may differ considerably depending on parameters such as the speed of potential changes or the evaluation window. The often abortive appearance of APs and the scarcity (< 10%) of DUM neurons reliably producing more than one or two APs at the phasic onset, however, calls into question the importance of spiking in DUM neurons in general. Unfortunately, the current data set cannot adequately address this question.

Relevance of inhibition for temporal processing

DUM neurons showed various kinds of inhibition in the present study. A delayed inhibition, typically seen in the rising phase of the onset excitation, has been previously described and occurs in about 20% of DUM neurons (Lefebvre et al. 2018) and in the auditory DUM neurons of other insects (Stumpner and Ronacher 1991). As for carrier frequency processing, occurrence of this inhibition was not correlated with specific filter properties. Furthermore, such inhibition was gradually reduced with shorter pause durations (higher pulse rates). Since DUM neurons receive excitation from both ears, it is unlikely that onset inhibition is important for the precise timing of the first AP, which is known from directional processing (Kelly and Phillips 1991; Carr 1993; Mason et al. 2001). An inhibition kicking in shortly after the initial excitation is similar to the pause/build pattern described in the dorsal cochlear nucleus of the guinea pig (Stabler et al. 1996). However, the temporal scale is much shorter, and the strength is much smaller (except perhaps for high intensities; see Lefebvre et al. 2018). Therefore, a functional role of this inhibition in the temporal processing of Ancistrura is not apparent. This is in contrast to the important role of the inhibitory timing in directionally sensitive interneurons in grasshoppers (Römer et al. 1981; Römer and Dronse 1982).

A strong hyperpolarization after the phasic onset excitation was observed in a subset of cells (e.g., Figs. 2c3, 5a, 9d). This inhibition reduced the excitatory response in stimuli with longer pulse durations. A species-typical song pattern elicited only weak excitation in such neurons. Very short pulses of 2 ms only faintly triggered the inhibition and consequently elicited the strongest excitation. Therefore, the potential role of such neurons may lie either in the detection of female response clicks (Stumpner 1999) or outside the processing of conspecific signals (e.g., bat echolocation sounds; see Faure and Hoy 2000a, b). In concordance with these ideas, nearly all of these neurons responded best to ultrasound at medium-to-high intensities.

Function of DUM neurons

The function of local auditory neurons must be local processing or neuromodulation of a local network (as certain amacrine cells do in the vertebrate retina; e.g., Gustincich et al. 1997). Most auditory DUM neurons branch in large areas of the auditory neuropile and therefore overlap at least partly with all other described auditory interneurons (Stumpner 1999; Molina and Stumpner 2005; Stumpner and Molina 2006; Ostrowski and Stumpner 2014; Lefebvre et al. 2018; see also Römer et al. 1988). Therefore, the diversity of DUM neurons renders them prime candidates for exciting or inhibiting various other local and intersegmental neurons with auditory input in the prothoracic ganglion. An inhibitory role for DUM neurons in frequency processing has been previously considered (Lefebvre et al. 2018). DUM neurons might also assist in shaping species-specific temporal filters in auditory neurons. Therefore, related bush-crickets with songs differing in temporal and frequency aspects (e.g., Stumpner and Meyer 2001; Korsunovskaya 2008, 2009) may have a very similar set of DUM neurons. Actually, the large number and physiological diversity of DUM neurons may have played a role in the diversification of acoustic signaling in bush-crickets (e.g., Korsunovskaya 2008, 2009; Heller et al. 2015), which only becomes evolutionarily relevant with a parallel diversification of song recognition. Concordant with this idea is the finding that species closely related to A. nigrovittata with different song frequencies and temporal patterns also differ in the tuning of auditory neurons. These differences are purely caused by inhibition, not excitation (Stumpner 2002). Such inhibition may be the result of DUM activity.

In A. nigrovittata, one target of DUM neurons may be neurons with ascending axon to the brain. Among these neurons, however, a temporal selectivity for the male song, similar to that of female behavior, has only been found in TN1 (Stumpner and Molina 2006). A striking variation among TN1 from different specimens was attributed to interindividual variation, since TN1 is a single identified neuron (Stumpner and Molina 2006). The mechanisms relevant for temporal processing seen in neurons such as TN1, AN2, and AN5-AG7 are unknown (Stumpner 1999; Stumpner and Molina 2006), but may include inhibition. Unpublished data (Andreas Stumpner) show clear effects on temporal processing of AN2 when inhibition is blocked. DUM neurons are the only known local candidate neurons for such inhibition. Therefore, some members of the DUM cluster may be involved in frequency processing and others in temporal processing. Even an excitatory function is conceivable, since polysynaptic excitation of the local neuron ON1 at the song carrier frequency has been described in some crickets (Pollack 1994). A homologous neuron to ON1 also exists in A. nigrovittata (Molina and Stumpner 2005).

Surprisingly, local auditory DUM neurons with properties similar to those of bush-crickets also exist in grasshoppers (Stumpner and Ronacher 1991), even though both groups of Orthoptera diverged about 300 million years ago (Song et al. 2015). Furthermore, their auditory systems must have evolved independently, since the ears are situated in different segments of the body. In crickets, some of the known prothoracic DUM neurons have weak auditory input, but none of them is a local neuron. They all have axons into the periphery, and all of them likely have octopamine as a transmitter (Gras et al. 1990). However, local auditory neurons other than the two Omega neurons (ON1 and ON2) are rarely studied (Stiedl et al. 1997), even though frequency-dependent inhibition has been described for several neurons (Schildberger 1984; Stumpner et al. 1995). From immunohistochemical stainings, the existence of γ-aminobutyric acid (GABA)-positive DUM somata in the cricket’s prothoracic ganglion seems likely, but they are less abundant than in A. nigrovittata (Silvia Gubert, Andreas Stumpner, unpublished data) and are potentially vibration sensitive. In non-hearing ancestors, the likely role for DUM neurons was processing of vibration (Stritih and Stumpner 2009), and vibratory responses (with various sensitivities) have been seen in some DUM neurons of A. nigrovittata (Paule Lefebvre, Andreas Stumpner, unpublished data).

DUM neurons: open questions

The population of local DUM neurons in the prothoracic ganglion of a bush-cricket comprises 15–50 individual cells. As an ensemble, they provide a filter bank for carrier frequency (Lefebvre et al. 2018) and temporal patterns (pulse rates and pause durations described in this paper). These two properties, however, are not clearly related, which makes it difficult to judge their effects in the auditory network. Therefore, it would be desirable to corroborate preliminary results of GABA as the transmitter of local DUM neurons using immunohistochemistry combined with stained characterized neurons. This analysis might also show whether auditory DUM neurons not containing GABA as transmitter exist (large soma DUM neurons have octopamine but are not local neurons; see Bräunig and Pflüger 2001). Applying blockers of GABA channels to the ganglia could demonstrate the function of inhibition that DUM neurons receive (most likely originating in other DUM neurons). Blockers like picrotoxin had clear and repeatable effects on auditory interneurons in Ancistrura (Stumpner 1998; Molina and Stumpner 2005) and other Orthoptera (Römer and Seikowski 1985; Stumpner 2002). The most direct evidence for DUM effects on other neurons would be through either double recordings or cell killing of a DUM neuron while recording from a different auditory neuron (see Molina and Stumpner 2005; Stumpner and Molina 2006). Both are not easy to achieve.