Sensory ecology of prey rustling sounds: acoustical features and their classification by wild Grey Mouse Lemurs



    Corresponding author
    1. Tierphysiologie, Zoologisches Institut, Universität Tübingen, Germany
    Search for more papers by this author

    1. Tierphysiologie, Zoologisches Institut, Universität Tübingen, Germany
    Search for more papers by this author
    • Present address: Max Planck Institute for Ornithology, Sensory Ecology Group, 82319 Seewiesen, Germany.

†Author to whom correspondence should be addressed. Neurobiology, Biocentre, Ludwig-Maximilians-Universität München, Großhaderner Str.2, 82152 Martinsried, Germany. E-mail:


  • 1Predatory mammals and birds from several phylogenetic lineages use prey rustling sounds to detect and locate prey. However, it is not known whether these rustling sounds convey information about the prey, such as its size or profitability, and whether predators use them to classify prey accordingly.
  • 2We recorded rustling sounds of insects in Madagascar walking on natural substrate and show a clear correlation between insect mass and several acoustic parameters.
  • 3In subsequent behavioural experiments in the field, we determined whether nocturnal animals, when foraging for insects, evaluate these parameters to classify their prey. We used field-experienced Grey Mouse Lemurs Microcebus murinus in short-term captivity. Mouse Lemurs are generally regarded as a good model for the most ancestral primate condition. They use multimodal sensorial information to find food (mainly fruit, gum, insect secretions and arthropods) in nightly forest. Acoustic cues play a role in detection of insect prey.
  • 4When presented with two simultaneous playbacks of rustling sounds, lemurs spontaneously chose the one higher above their hearing threshold, i.e. they used the rustling sound's amplitude for classification. We were not able, despite attempts in a reinforced paradigm, to persuade lemurs to use cues other than amplitude, e.g. frequency cues, for prey discrimination.
  • 5Our data suggests that Mouse Lemurs, when foraging for insects, use the mass–amplitude correlation of prey-generated rustling sounds to evaluate the average mass of insects and to guide their foraging decisions.


Foraging theory predicts that a forager will maximize its energetic net rate or efficiency (Schoener 1971; Stephens & Krebs 1986; Hughes 1993; Cuthill & Houston 1997). It can do so by selecting food with high profitability. Profitability increases with energy content and decreases with handling costs (Cuthill & Housten 1997). To assess profitability as a basis for decision making, foragers rely on sensory information (Dusenbery 1992) which is correlated with profitability, e.g. visual, olfactory, acoustic or other sensory cues. As efficient foraging is vital for an animal's fitness (Stephens & Krebs 1986), such sensory access to external information can be assumed to be under strong selection and may even contribute to niche separation in closely related species (Siemers & Schnitzler 2004; Siemers & Swift 2006). Here we consider the case of animals that forage for insect prey at night. As acoustic information is independent of illumination and only slightly affected by vegetation cover, it may be the most useful of the above-mentioned cues for perception of insects over some distance in dense, cluttered habitats such as forests (Marten & Marler 1977; Marten, Quine & Marler 1977; Wehner 1997; Dominy et al. 2001; Dominy, Ross & Smith 2004). At close distance, visual and possibly olfactory cues will likewise play an important part (Nekaris 2005).

It has been shown that predators and parasitoids exploit communication sounds and rustling sounds of moving prey (Payne 1971; Charles-Dominique 1977; Zuk & Kolluru 1998; Arlettaz, Jones & Racey 2001; Page & Ryan 2005). While foragers of such diverse taxa as owls, bats and primates use prey-generated rustling sounds to detect and localize prey (Payne 1971; Charles-Dominique 1977; Arlettaz et al. 2001), it is not yet known whether they also use these sounds to classify their prey. A prerequisite for this would be that prey rustling sounds convey information about prey properties such as size, taxon or profitability.

We used the Grey Mouse Lemur Microcebus murinus as a model species to test whether insect-eating animals can extract and use prey size-correlated cues to guide their foraging decisions. The 60 g Grey Mouse Lemur feeds largely on fruit, gum, insect secretions and arthropods (Martin 1972, 1973; Petter 1978; Sussman 1978; Radespiel et al. 2006). As in this study we were interested to explore the sensory ecology of insect rustling sounds, we concentrated on those albeit Mouse Lemurs occasionally also take small vertebrates. Mouse Lemurs can use visual motion cues and hearing to find insects (Siemers et al. 2007). Given their nocturnal lifestyle and dense habitat (Martin 1972), we assumed that acoustic prey cues will be more helpful than visual cues for prey classification over a distance.

First, we recorded rustling sounds from Madagascan insects and analysed the information content with respect to insect mass. We then conducted a series of behavioural experiments with wild-caught Mouse Lemurs to explore whether and how they make use of this information. In a series of preference experiments, we tested the hypothesis that certain acoustic features mediate prey choice in Mouse Lemurs. Specifically, we tested the predictions that temporal structure, high-frequency content or bandwidth would be used as indicators of large prey. In a reinforced discrimination experiment, we tested the hypothesis that Mouse Lemurs will be able to classify prey mass based on rustling sounds. Specifically, we tested the predictions that Mouse Lemurs can achieve this classification independent of the individual insect, of the insect species and of the walking substrate.


sound recordings

Recordings of 530 rustling sounds from 50 individual Madagascan insects of at least 10 different genera (Coleoptera & Blattodea, electronic Appendix S1*) were made in April 2003 in a large wooden building in the Mandena littoral rainforest [Tolagnaro (Fort-Dauphin), SE-Madagascar]. We used custom-built equipment (ultrasonic condenser microphone: frequency response ± 3 dB between 20 and 160 kHz, ± 4 dB between 7 and 20 kHz; external A/D-converter: 16-bit depth, 480 kHz sampling rate, additional 8 × oversampling, digital antialiasing; Department of Animal Physiology, University of Tübingen) connected to a laptop computer running custom-made recording software. Insects walked on the floor of a plastic bowl (height: 15 cm; diameter at bottom: 26 cm, at top: 33 cm) covered with dry leaves collected below one individual tree (Syzygium emirnense, Myrtaceae). The microphone was vertically suspended 20 cm above the centre of the bowl. For most insects, i.e. 36 individuals, 10 recordings were made (for 14 insects, less or more recordings, i.e. between 8 to 25, were made; for details see electronic Appendix S1). Each recording had a duration of 1 s (for analysis we discarded the first and last 5 ms) and was taken while the insect was walking on the substrate in the middle of the bowl.

sound analyses

The recordings were high-pass filtered at 1 kHz (elliptic fourth order filter, 80 dB attenuation, 0·1 dB peak-to-peak ripple) and analysed with Matlab 6·5 (MathWorks, Inc, Natick, MA, USA).

Amplitude parameters

We calculated three different amplitude parameters from the time signal: (1) Peak [highest amplitude, i.e. the value of one sampling point (smpl)]; (2) MaxRMS (the maximum root mean square (RMS) amplitude of a gliding 10 ms window = 4800 smpl); and (3) TotalRMS (the RMS-amplitude of the whole recording = 475 200 smpl). Both measurements for signal roughness, i.e. the fourth moment (Hartmann & Pumplin 1988; Grunwald, Schörnich & Wiegrebe 2004) and the crest factor (Peak − TotalRMS), showed no significant relationship with the mass or size of the insects (data not shown).

Frequency parameters

We calculated four frequency parameters from the amplitude spectrum (256 smpl FFT, Hann-window, 50% overlap) of 94 recordings, which were well above the background noise level (i.e. only recordings where the best frequency of the amplitude spectrum of the whole recording had an amplitude 14 dB above the mean amplitude of the amplitude spectrum). The four frequency parameters were: (1) OverNoiseBW, which is the bandwidth at 40 dB SPL (i.e. about 10 dB above spectrogram noise level), and (2–4) the bandwidths at thresholds 3, 6 and 10 dB below the peak frequency of the amplitude spectrum. All four frequency parameters were calculated three times for each recording, using three different increasingly sized sections of the whole recording. The sections were selected to correspond to the three different amplitude parameters we calculated (Peak, MaxRMS, TotalRMS) and their respective position within each recording. For every recording, each frequency parameter was thus calculated (1) using 256 smpl around the position of Peak; (2) using 4864 smpl starting with the first smpl of the 10 ms window of MaxRMS; and (3) using all 475 200 smpl of the recording.

mouse lemurs

We conducted playback experiments from October to December 2003 in the Forêt de Kirindy (60 km north of Morondava, West Madagascar) with five male, experimentally naïve, wild Grey Mouse Lemurs Microcebus murinus (J. F. Miller 1777). Subjects 1, 2 and 3 were used in the discrimination experiments and subjects 3, 4 and 5 in the preference experiment. Subject 3 was used in both experiments, with the preference experiment first. No animals were excluded from the experiment or the analysis. The lemurs were kept and tested individually in mesh-covered wooden enclosures (75 × 75 × 70 cm) with unlimited access to water and were released at their capture site after 12–23 experimental days. Each enclosure had two choice platforms (24 × 21 cm) 17 cm above the ground in the left and right front corners, separated by a gap of 15 cm and connected by a horizontal branch. During playbacks, the lemurs were trained to sit on a central branch leading from the choice platforms to the rear part of the enclosure (electronic Appendix S2). Food (banana, mango, insects) was offered only during the experiments and afterwards. All experiments were carried out under license of the Ministière de l’Environnement des Eaux et Forêts, Madagascar.

playback set-up

Two loudspeakers (ribbon-tweeter RT 2H-A, Swans, Montery Park, CA, USA) were fixed outside the enclosure opposite each choice platform and directed toward the centre of the enclosure. Stimuli were broadcast from a laptop computer with CoolEdit 2000 (Syntrillium, Phoenix, AZ, USA) via an external soundboard (96 kHz; Hammerfall DSP Multiface, RME/Synthax, Haimhausen, Germany) and two amplifiers (no. 19 00 91 50, Conrad electronics, Hirschau, Germany). The frequency response of the playback system decreased slightly from 1 to 40 kHz (± 6 dB) and again by about 8 dB between 40 and 45 kHz. The frequency response of the playback and the recording system together was flat ± 3 dB between 0·5 and 40 kHz and ± 5 dB between 0·2 and 45 kHz.

familiarization with the set-up

The experiments were preceded by 2–4 days in which we familiarized the lemurs with the set-up and the task. Lemurs were rewarded for moving to the loudspeaker broadcasting one out of three 1-s long artificial rustling-like sounds. The artificial rustling-like sounds were generated from white noise with Cool Edit 2000 (Syntrillium, Phoenix, AZ, USA) by manually introducing amplitude envelope fluctuations. One sound was additionally low-pass-filtered. All three sounds were checked to sound ‘rustling-like’ by a human listener. At the end of the familiarization phase, the animal's performance showed no further improvement (correct choice in > 85% for subject 2, 100% for the other subjects).

general procedure of playback experiments

We conducted two types of playback experiments to identify (1) the acoustic features preferred by Mouse Lemurs (preference experiments), and (2) their ability to discriminate acoustically between big (i.e. heavy) and small (i.e. lightweight) insects (discrimination experiment). We used a two-alternative-choice paradigm, i.e. we presented the animal with two simultaneous playbacks from two loudspeakers. The animals had to decide for one of them by entering the choice platform in front of the respective loudspeaker (electronic Appendix S2). The experiments were carried out at night. As stimuli we used 1 s long recordings of rustling sounds from beetles and cockroaches. For each playback, the loudspeaker and stimulus were chosen randomly in the familiarization phase and the training phases 1–6. In training phase 7 and all tests, we used a balanced design with sequence of presentation being randomized within blocks (one block containing all given stimuli of the respective phase or test). As a result, there was no repetition of the same stimulus–speaker combination and not more than two presentations of the same stimulus type on different speaker sides. The lemurs had to move to one of the two platforms within 6 s after stimulus onset. As a reward, they were given by hand a piece of banana (0·1 g) on a stick (for rewarding scheme, which differed between the two experiments, see below). The next playback began after the lemur had left the platform and returned to the central branch (example video in electronic Appendix S3). If the lemur did not react at all, the stimulus was repeated up to three times before switching to the next one. Rewarding was necessary because in a previous study Mouse Lemurs showed little response to repeated unrewarded playbacks of rustling sounds (Siemers et al., 2007). As many choice decisions as possible were registered per animal and night, with an average of 126 decisions (range: 0–380, maximum 3 h testing per night). We stopped experiments when the subjects were satiated.

preference experiments: white noise test and frequency test

The preference experiments aimed at identifying the acoustic features of rustling sounds that mediate prey choice in Mouse Lemurs. To assess spontaneous reactions, the test phase commenced immediately after the familiarization phase, without reinforcement of any particular stimulus. The lemurs were confronted with a choice between two sounds that they had not experienced before in our experiments. To obtain uninfluenced choices, we rewarded only 25% of all trials, following a pseudo-random schedule that was independent of the stimulus and the decision. We alternated the blocks of stimuli of both tests.

White noise test

We hypothesized the lemurs would prefer sounds with a rustling-like temporal structure over noise-like sounds because the former are typical for prey. One rustling sound from a large beetle (Fig. 1a) was presented simultaneously with white noise (48 kHz bandwidth). Both were set to the same TotalRMS and then played back at six different amplitudes (30, 39·5, 49, 68, 77·5, 87 dB SPL TotalRMS at a distance of 1 m from the loudspeaker), chosen to encompass the entire range of insect rustling sound amplitudes we had measured in our recordings.

Figure 1.

Example and analysis of rustling sounds. Typical examples from a large (a) and a small beetle (b); note the different scales on the amplitude axis (linear arbitrary units, 1 corresponds to the full scale of the recording system). (c) All three amplitude parameters [(i) Peak; (ii) MaxRMS; and (iii) TotalRMS] increased with the base-10 logarithm of insect mass. Dots: all recordings per specimen. Squares: mean of all recordings per specimen. Lines: linear regression of the amplitude-means and its 95% confidence interval. (d) Lower and upper boundary of the bandwidth at 40 dB SPL (OverNoiseBW). The OverNoiseBW was calculated using three increasingly sized sections of the recording, which corresponded to the calculated amplitude parameters (i.e. Peak, MaxRMS, and TotalRMS). The boundaries of the OverNoiseBW were plotted against the respective amplitudes [i.e. against Peak (i), against MaxRMS (ii) and against TotalRMS (iii)]. Circles: upper and lower boundary of the OverNoiseBW. Lines: linear regression and its 95% confidence interval. Significance of the regression of OverNoiseBW vs. amplitudes: against Peak: r2 = 0·182, P < 0·0001; against MaxRMS: r2 = 0·402, P < 0·0001; against TotalRMS: r2 = 0·626, P < 0·0001.

Frequency test

Based on the above sound analysis, we hypothesized that the lemurs would prefer rustling sounds with either broadband and/or high-frequency content because these indicate large prey. To test this hypothesis, we varied the frequency content of the playbacks by filtering a rustling sound recording with one of five bandpasses (1–12, 1–24, 1–48, 24–48 and 36–48 kHz; stopband > 30 dB attenuation) and set all stimuli to a TotalRMS of 60 dB SPL. We presented all 10 possible pairwise combinations (e.g. 1–12 vs. 1–24, 1–12 vs. 1–48, etc.).

discrimination experiment

The discrimination experiment aimed at investigating whether Mouse Lemurs were able to discriminate acoustically between large (i.e. heavy) and small (i.e. lightweight) insects. Therefore, the familiarization phase was followed by training to reinforce the rustling stimuli of large beetles. We simultaneously presented one of five recordings of large beetles (Kheper subaeneus, 0·77–1·07 g) and one of five recordings of small beetles (Phalos sp., 0·20–0·26 g) in different pairwise combinations (Table 1). Both beetle species are common in the study area and are readily taken and eaten by the Mouse Lemurs. The next training phase commenced after the performance showed no further improvement. Subsequently, we conducted three tests, as detailed below (only lemurs 1 and 3). We always rewarded the selection of the playback of the larger insect, which was not necessarily the louder one.

Table 1.  Discrimination experiment: amplitudes and sequence of presentation of the stimuli (without familiarization phase). Training phases 1–5: presentation of five pairs of recordings. Training phase 6: all five pairs were presented in a random order. Training-phase 7: all 25 possible combinations of the five recordings of large with the five recordings of small beetles were presented pseudo-randomly. Tests: several pairs of large and small beetles and cockroaches were presented pseudo-randomly
PhaseLargeSmallNo. of playbacks per lemur
No. of filesPeak dB SPLTotalRMS dB SPLNo. of filesPeak dB SPLTotalRMS dB SPL
Training 1 19156 1 7051120–280 
Training 2 18562 1 7151 90–150 
Training 3 17955 1 7553120–250
Training 4 18054 1 8655120–250
Training 5 18156 1 8153120–290
Training 6mixture of the five pairs of the training phases 1–5, each pair 32 ×160 
Training 7all combinations, each pair 10 ×250 
Test 14173–9551–6541 67–8651–56  6 per file
Test 22068–9451–6520 70–8651–61  6 per file
Test 31069–8651–5710 68–8551–55 10 per file

Known species test

We tested whether Mouse Lemurs can acoustically classify unfamiliar recordings of the same species. We randomly combined 41 new recordings of K. subaeneus (large) and Phalos sp. (small) having amplitude ratios (large to small) in the range of −6·4 to +23·7 dB (Peak), −5·4 to +21·3 dB (MaxRMS) and −2·8 to +12·8 dB (TotalRMS). Note that an amplitude ratio equates to a difference on the (logarithmic) dB SPL scale.

Unknown species test

To test whether Mouse Lemurs will generalize the classification to discriminate between unknown species, we used 20 recordings of beetle and cockroach species not presented before (amplitude ratios of playbacks: Peak: −12·8 to +22·5 dB; MaxRMS: −8·8 to +18·4 dB; TotalRMS: −5·3 to +14·0 dB).

Substrate-dependency test

Walking substrate influences acoustic parameters of rustling sounds (Siemers & Güttinger 2006). Specifically, rustling sounds of insects walking on fresh, green leaves have lower amplitude than those on dry leaves (Goerlitz & Siemers, unpublished). To determine whether lemurs can classify prey independently of the substrate over which it is moving, we combined and tested 10 recordings of the larger beetle K. subaeneus on fresh, green leaves with 10 recordings of the smaller beetle Phalos sp. on dry leaves (amplitude ratios of playbacks: Peak: −16·7 to + 17·8 dB; MaxRMS: −13·5 to + 11·4 dB; TotalRMS: −4·5 to + 4·4 dB).

handling time measurement

We fed 85 insects to 12 male Grey Mouse Lemurs (individuals 1, 4, 5 of the playback experiments, and nine additional animals) to measure handling times for differently sized insects. Housing for the nine additional Mouse Lemurs was the same as for the subjects in the playback experiments, except for the size of the enclosures (150 × 75 × 70 cm) and the presence of only a single experimental platform (49 × 31 cm). Living insect prey was offered to the Mouse Lemurs in plastic dishes covered with a lid. Handling time, starting with the first touch of the insect until the feeding ended, was measured with a stopwatch. Insect mass was determined beforehand with a balance to the nearest 0·01 g.


For analysis of the playback experiments, we used a two-tailed binomial test (JMP 4.0.4, SAS, Cary, NC, USA) to test the null hypothesis that the distribution of the Mouse Lemurs’ decisions between two stimuli did not differ from an equal distribution. In the discrimination experiment, we correlated the lemur's decisions to the amplitude ratio of the playbacks (large insect to small insect) with a logistic model (Systat 10, SPSS Inc., Chicago, IL, USA):


As logistic constraints limited us to a small number of animals, we used within-subject controls in all behavioural experiments.


rustling sounds

We recorded 530 rustling sounds from 50 individual Madagascan insects of at least 10 different genera (Coleoptera & Blattodea; electronic Appendix S1) walking on a natural substrate, dry leaves. The rustling sounds consisted of a series of broadband clicks with variable amplitudes, bandwidths and click intervals (Fig. 1a,b) produced by the insects’ feet and body when touching the leaves and by the corresponding shifts of the substrate. All three amplitude parameters calculated were widely distributed for each mass and individual, yet significantly correlated with the base-10 logarithm of insect mass (Fig. 1c). For statistical analysis, we discarded all TotalRMS-values for insects with masses < 0·64 g because their TotalRMS did not differ from the noise level (Dunnett test, P > 0·05).

Higher-amplitude clicks contained higher frequencies. This was revealed by the correlation between the clicks’ bandwidth at 40 dB SPL (OverNoiseBW) with amplitude (Fig. 1d). OverNoiseBW ranged from 36 to 146 kHz when calculated from the amplitude spectrum around Peak (using 256 sampling points [smpl]). When calculated from the amplitude spectrum of the whole recording (475 200 smpl), it reached maximally 38 kHz because of the integration of clicks with high and low amplitudes. The lower boundary of OverNoiseBW was between 2 and 13 kHz and decreased only slightly with increasing amplitude, whereas the upper boundary increased strongly as amplitude increased; this was the main reason for the increase of OverNoiseBW.

The main energy (i.e. 10 dB below peak frequency) was between 3 and 40 kHz for each recording as a whole and between 10 and 32 kHz when calculated only around Peak. The −3 dB boundaries were 7 and 18 kHz for the whole recording and 16 and 24 kHz around Peak (see electronic Appendix S4 for details). Similar to OverNoiseBW, variations between recordings in the bandwidths 3, 6 and 10 dB below peak frequency became smaller when greater sections of the recordings were included in the analyses (256 smpl around Peak, 4864 smpl at MaxRMS and 475 200 smpl over the whole file). However, contrary to OverNoiseBW, no systematic correlation was found between amplitudes and the bandwidths 3, 6 and 10 dB below peak frequency, so that the spectral envelope was roughly the same for all rustling sounds.

Because mass was correlated with amplitude (Fig. 1c), and amplitude with OverNoiseBW (Fig. 1d), heavier insects produced higher amplitudes, higher frequencies, and greater bandwidths.

preference experiments

To identify the acoustic features mediating prey choice in Mouse Lemurs, we conducted rewarded playback experiments without systematic reinforcement using a two-alternative-choice paradigm.

White noise test

One rustling sound from a large beetle (Fig. 1a) was presented simultaneously with white noise (48 kHz bandwidth). Both were set to the same TotalRMS and then played back at six different amplitudes. Lemurs chose white noise at high and the rustling sound at low stimulus amplitudes, with two significant cases each (Fig. 2a). At 77·5 dB SPL, subjects 4 and 5 significantly preferred the white noise playback (P < 0·05, two-sided binomial test), whereas at 30 dB SPL, the same subjects significantly decided for the rustling sound playback (P < 0·05, two-sided binomial test).

Figure 2.

Results (a) and tentative interpretation (b) of the preference experiment between rustling sound and white noise (White noise test). (a) No. of decisions for the rustling sound. Sixteen playbacks were presented per animal and amplitude (no presentation at 58·5 dB SPL); 3, 4, and 5 denote the experimental subjects. (b) Hearing curve of Microcebus murinus (thick grey line; measured as evoked potentials in the auditory cortex by Niaussat & Petter 1980), shown together with the amplitude spectra of our simultaneously presented rustling sound and white noise (which are plotted at four of the six presented amplitudes). The sensitivity peak of the hearing curve is positioned arbitrarily at the lowest white noise amplitude, showing that the white noise would not be audible at this TotalRMS, whereas the rustling sound's amplitude would be above hearing threshold. The relative amplitude values for the lowest amplitude spectrum are indicated on the right side.

Frequency test

We used a rustling sound recording filtered with one of five bandpasses and presented all 10 possible pairwise combinations. Lemurs preferred the 1–48 kHz stimulus (Fig. 3a–d), except for the combination with 1–24 (Fig. 3c). Two lemurs preferred the low-frequency stimulus 1–24 to the high-frequency stimulus 24–48 (Fig. 3e). In the combinations of 24 kHz vs. 12 kHz bandwidth, all Mouse Lemurs preferred the low-frequency stimulus 1–24 vs. both 12 kHz stimuli (Fig. 3f,g), but did not prefer the high-frequency stimulus 24–48 vs. both 12 kHz stimuli (Fig. 3h,i). There was also no preference for 36–48 vs. 1–12 (Fig. 3j).

Figure 3.

Frequency test: Preference experiment between the same rustling sound filtered with different bandpasses (1–12, 1–24, 1–48, 24–48 and 36–48 kHz; TotalRMS always 60 dB SPL). Each panel (a–j) presents the results for one combination of two stimuli, which are symbolized in the top row (e.g. 1–12 kHz vs. 1–48 kHz, 36–48 kHz vs. 1–48 kHz, etc.). The results within each panel show the number of decisions for each of the two simultaneously presented stimuli for three Mouse Lemurs. Sixteen playbacks were presented per animal and stimulus combination; 3, 4, and 5 denote the experimental subjects.

discrimination experiment

A reinforced discrimination experiment was used to test whether Mouse Lemurs are able to use non-amplitudinal acoustic information to discriminate between large and small prey (example video in electronic Appendix S3).


The lemurs first went through several training phases to reinforce the acoustic stimuli of large beetles. During the first three training phases, in which large beetles had higher amplitudes (stimuli of training phases 1–3, Table 1), the Mouse Lemurs selected the larger one (two-tailed binomial test, P < 0·05 for 41 of 52 blocks with 20 decisions each). However, this was not the case when playbacks of larger beetles had similar or slightly lower amplitudes than playbacks of smaller beetles (stimuli of training phases 4 and 5, P < 0·05 for only five out of 43 blocks, Table 1). The same pattern was found in training phase 6 (electronic Appendix S5). The Mouse Lemurs apparently learned in the first three training phases to recognize the reinforced playback by amplitude and did not or could not switch to another cue afterwards. Their decisions in response to all possible combinations of training stimuli (phase 7) depended again significantly on amplitude (Fig. 4a).

Figure 4.

Last training phase (a) and all tests (b–d) of the discrimination experiment. The choice behaviour of the Mouse Lemurs 1 and 2 for the playbacks of larger and smaller insects is plotted against the Peak-ratio of the respective playback pairs. The no. of choices is indicated by the size of the circles. The amplitude dependence of the choices was described by a logistic model (thick solid lines, (a): subject 1: P < 0·0001, 2: P = 0·0004; (b): 1: P = 0·0007, 2: P < 0·0001; (c,d): P < 0·0001 for both lemurs). Solid vertical lines: 0·5-probability value of the logistic regression (i.e. the Peak-ratio which will be classified with a probability of 0·5 as being a large insect). Dotted vertical line: 95% confidence interval of the 0·5-probability value.

Tests (known species, unknown species, substrate-dependency)

Although the lemurs did not learn the reinforced task of choosing the playback of the larger insect irrespective of amplitude, we still conducted three tests with subjects 1 and 3 to test their ability to acoustically classify unfamiliar recordings of the same species, recordings of unknown species, and recordings independent of walking-substrate. As during training, both lemurs chose the playbacks with the higher amplitudes in all three tests, independently of mass and reward (Fig. 4b–d).

We analysed the influence of the amplitude on the choice behaviour by fitting the choice behaviour with a logistic regression and calculating the 0·5-probability value of the logistic fit (Fig. 4). The 0·5-probability value indicates the amplitude-ratio at which the larger insect is selected in 50% of all playbacks. At amplitude values higher than the 0·5-probability value the larger insect is selected more often, at lower amplitude values the smaller insect is selected. A shift of the 0·5-probability value to negative values indicates a preference for the larger insect, which is increasingly independent of amplitude as the 0·5-probability value gets more negative. If there was no influence of the amplitude on the choice of large insects at all, the logistic regression would be a straight horizontal line.

In the substrate-dependency test, all six logistic regressions (2 subjects [1 and 2] × 3 amplitude-ratios [ratios of Peak-, MaxRMS- and TotalRMS]) included zero in the 95% confidence-interval of the 0·5-probability value (see Fig. 4d for the logistic regressions to the Peak-ratio; data for MaxRMS- and TotalRMS-ratios not shown). In the unknown species test (Fig. 4c) and the known species test (Fig. 4b), three of the six logistic regressions included zero in the 95% confidence-interval, whereas only one regression did so in the training phase 7 (Fig. 4a). In all other cases, where the 95%-confidence interval did not include zero, the 0·5-probability value was shifted to negative amplitudes (Fig. 4a–c).

handling time measurement

We fed 85 insects (mass: 0·06–5·94 g) to 12 Mouse Lemurs and measured handling time, which ranged from 10 to 1100 s (Fig. 5). A power model with an exponent < 1 best described the relation between total prey mass and handling time (Fig. 5), i.e. the increase of handling time became smaller with increasing mass.

Figure 5.

Handling times of 85 insects eaten by 12 Mouse Lemurs, plotted vs. total mass of the insects. The data were best fitted with a power law model (F = 95·67, P < 0·0001).


rustling sounds

Based on experiments with laboratory recordings and more artificial substrates, Siemers & Güttinger (2006) have suggested a correlation between arthropod length and amplitudes. The present study was the first to identify mass-related amplitude and frequency cues produced by arthropods moving over a natural substrate. It shows that rustling sounds will convey reliable information about insect mass to a potential predator despite the high variation between recordings. This variation was reduced by integrating over larger sections of the recordings. We thus assume that prolonged listening may likewise enable a predator to estimate the average acoustic properties of a rustling sound, and thereby prey mass, more exactly.

Rustling sound amplitudes depend heavily on the substrate (Goerlitz and Siemers, unpublished), as well as on the distance between prey and predator. Their perception can be impaired by background noise from wind, moving leaves and other animals. This background noise approximated 50 dB SPL in Mandena Forest, Madagascar (own data; similar in Kibale Forest, Uganda: Waser & Waser 1977), resulting in a human sensation level for white noise of about 30 dB SPL under Malagasy field conditions (own observation). Sound amplitude is reduced by spherical attenuation (6 dB per doubled distance), atmospheric attenuation (0·11 dB m−1), and vegetation cover (0·4 dB m−1; all values calculated for 12 kHz and a typical night climate in the Forêt de Kirindy: T = 30 °C, moisture = 80%; Marten & Marler 1977; Marten et al. 1977; Crocker 1998). A 12 kHz signal with the lowest MaxRMS observed in this study (50 dB SPL) will therefore fall below sensation level at 2 m distance, whereas the detection radius of a signal with the highest observed MaxRMS (87 dB SPL) will be 8 m larger. These are rough estimates, but in any case, rustling sounds emanating from a centimetre-sized prey in the forest at night will be perceivable to predators over a far greater distance than visual cues.

To judge prey mass on the basis of sound amplitude, the predator must also know the prey's distance and substrate. The predator may estimate the distance to a sound source from the sound's direction and its own position, known by spatial memory. Reverberation too is a robust cue to the absolute distance from a sound source (Bronkhorst & Houtgast 1999; Naguib & Wiley 2001). Another cue may be atmospheric attenuation, which affects higher frequencies more than lower ones (Crocker 1998; Bronkhorst & Houtgast 1999; Naguib et al. 2000). Its potential usefulness as a robust cue is corroborated by our finding that the spectral envelope of all recorded rustling sounds was shifted only along the amplitude axis and did not differ in shape between insects or substrates. Substrate information may be retrieved from the memory of the predator or from the rustling sound itself, as different substrates produce not only different amplitudes, but possibly different spectral envelopes as well, along with characteristic glints and temporal patterns.

Grey Mouse Lemurs have acute hearing at about the frequency range which we measured in rustling sounds (3–40 kHz, −10 dB bandwidth of the whole recordings), with greatest sensitivity at 12 kHz (Fig. 2b; Niaussat & Petter 1980). They also use similar or slightly higher frequencies for alarm (12–17 kHz) and advertisement calls (14–34 kHz), with the dominant frequencies at 19–22 kHz and higher (Zimmermann & Lerch 1993; Zimmermann et al. 2000; Zimmermann & Hafen 2001). Whereas hearing ability and communication frequency are interdependent and can be mutually adapted during evolution, the selective advantages of acoustic prey detection, discrimination and localization are likely to have played a unidirectional and major role in the evolution of hearing ability of predatory species. The Mouse Lemur is thought to resemble the ancestral primate, having retained many of its basal characteristics (Martin 1995; Gebo 2004), such as nocturnality, partly insectivorous diet, and good olfaction. Likewise, the Mouse Lemur's audiogram is not specialized, but represents the generalized mammalian audiogram (Heffner 2004). We therefore assume that this broadband hearing ability already evolved relatively early in mammals, possibly as an adaptation to the pursuit of arthropod prey. Detection of rustling sounds that reveal an approaching predator are likewise a conceivable important selection pressure in this respect.

preference experiments

White noise test

Mouse Lemurs consistently chose the stimulus that was highest above their hearing threshold. Although TotalRMS was set identical, the stimuli were likely not perceived as equally loud by the Mouse Lemurs because their hearing threshold is frequency-dependent (Niaussat & Petter 1980). While the white noise had a flat frequency response (± 1 dB), the amplitude of the rustling sound was higher than that of the white noise between 6 and 24 kHz and had a peak frequency at 10 kHz, with an amplitude of 7 dB above the white noise (Fig. 2b). As the energy of the white noise was distributed equally over all frequencies, the white noise with the lowest amplitude (TotalRMS = 30 dB SPL at a distance of 1 m, almost inaudible to a human in Madagascan field conditions) may have been inaudible to the lemurs because it was not above their hearing threshold (Fig. 2b, lowest plot). The rustling sound with the same amplitude, on the other hand, had more energy around 10 kHz and thus would have excited the auditory system more strongly. At higher amplitudes (Fig. 2b, above plots), the white noise excited lower and higher frequency channels much better than the rustling sound, presumably resulting in a higher perceived overall amplitude. In spite of the familiarization phase with an artificial rustling-like sound, the lemurs selected the white noise if it appeared louder, thus disregarding temporal and spectral envelope information.

Frequency test

Our results showed that the Mouse Lemurs preferred the frequency range from 12 to 24 kHz (Fig. 3), which matches their auditory system's region of greatest sensitivity (Fig. 2b). This largely invariant preference of the 12–24 kHz frequency range contradicts our hypothesis that the preference of rustling sounds would be based on the presence of higher frequencies and/or greater bandwidths as compared with the other stimulus presented. Rather, the 1–24 kHz stimulus was apparently perceived as loudest by the Mouse Lemurs. Here too, the lemurs’ decisions are explicable by perceived amplitude ratios alone. Our data suggest that they did not use frequency or bandwidth information, even though both have reliable informative value in real rustling sounds.

discrimination experiment

The choice behaviour of the Mouse Lemurs depended on the amplitude of the playback and not on the mass of the insect. This was especially obvious from the logistic regression of the substrate-dependency test, where the subjects selected the playback with the higher amplitude (Fig. 4d). The shift of the 0·5-probability value to negative amplitudes ratios, which was observed in some of the other logistic regression (training phase 7, known species test, unknown species test), is presumably caused by the limited number of available playbacks with negative amplitude ratios and not by a preference for the playbacks of larger insects. In the substrate-dependency test, where all six logistic regressions had zero as 0·5-probability value, the amplitude ratios were equally distributed above and below zero, e.g. Peak-ratios ranged from approximately −18 dB to 18 dB (Fig. 4d). In the known species test, the unknown species test and the training phase 7, the rustling sound combinations with negative amplitude ratios became increasingly rare (see Fig. 4a–c for the Peak-ratios). Had recordings of small insects with amplitudes as high as or higher than those of large insects been available, the lemurs would presumably have selected the playback of the smaller, but louder insect. However, we cannot rule out the possibility that subjects became fixated on amplitude only because the first three rewarded training stimuli were higher in amplitude than the three unrewarded ones (besides possessing all other acoustic features of large insects). However, we assume this to be unlikely, as the results of the preference experiments, where the lemurs showed spontaneous reactions, are explicable by amplitude ratios only.

general discussion

The experimental two-alternative-choice situation between two prey items does not directly mimic a natural foraging situation because a mouse lemur will most often hear prey consecutively and not simultaneously in the wild. However, it must decide whether to approach and attack a detected prey or to continue searching. Our experiment aimed at assessing the classification of prey underlying this decision process.

All Grey Mouse Lemurs tested by us relied consistently on amplitude information alone rather than on spectral or temporal information in determining which of two acoustic stimuli to approach. Despite attempts in the reinforced paradigm, we were not able to persuade the lemurs to use other than amplitude cues. We therefore assume that Mouse Lemurs in the wild also rely on amplitude cues. However, we can not fully rule out the possibility that the training stimuli in the discrimination experiment specifically influenced our subjects in this direction.

Our experiments produced a type of behavioural audiogram that mirrors the evoked potential audiogram obtained from the auditory cortex (Niaussat & Petter 1980). On a behavioural level, we found that Grey Mouse Lemurs preferred high amplitude stimuli. Therefore, amplitude is presumably the lemurs’ main auditory means of access to the mass of insects whose rustling sounds they overhear. We are unable to determine whether this choice of high amplitude stimuli reflects the action of a passive sensory mechanism biased towards high amplitudes (Siemers & Güttinger 2006), an active selection of high amplitude stimuli, or both (the two are not mutually exclusive). Whatever the mechanism, however, the evolutionary selected ecological benefits of choosing on average larger prey remains the same.

Selecting larger prey is not beneficial per se, as long as prey size is not correlated with profitability. Energetic profitability is defined as consumed energy per time spent feeding. The relationship between total insect mass and energetic content is linear (Green, Dryden & Dryden 1991; Brodmann & Reyer 1999; Chen, Thompson & Dickmann 2004), with beetles (Coleoptera) containing more energy per gram fresh-weight (7·2–8·7 kJ g−1, Brodmann & Reyer 1999; Chen et al. 2004) than cockroaches (Blattodea, 5·6 kJ g−1, Green et al. 1991). Given this linear relation of energy content to insect mass and the decelerating increase of handling time with mass that we found for Mouse Lemurs (Fig. 5), energetic profitability of insects can be assumed to increase with insect mass for Mouse Lemurs.

This gross estimation lends some support for an at least rough correlation of insect mass and (energetic) profitability. If this holds true, Mouse Lemurs might not only be able to assess mass, but also profitability of potential prey from their rustling sounds. Future refinement of our gross estimation should consider, however, that ingested prey mass rather than total mass would be more suitable for calculating energetic profitability. This will require measuring the mass and energetic content for ingested vs. discarded parts separately. Based on own observations we predict that replacing total by ingested mass would increase the slope of the handling time curve. The energetic content of ingested mass, on the other hand, will presumably be higher than that of total mass, as the discarded parts (e.g. wings, carapax) are usually of low energy. Empirical data will be needed to assess whether these two effects cancel out and will yield an increase of profitability with prey mass (until a certain limit) also for ingested mass. Furthermore, beside energetic aspects, many other factors have to be considered by a predator, like its current own nutritional needs, prey nutrient composition and prey toxicity. It might therefore be worthwhile to investigate whether foragers can tell toxic from palatable insect species from there rustling sounds.

Besides getting larger prey, there is a second conceivable advantage to the preference of high amplitude rustling sounds: namely, getting closer prey. Prey proximity will also result in higher amplitudes (due to reduced geometric attenuation), which will additionally be less affected by atmospheric attenuation than those of prey at a greater distance. Close-by prey will be of higher profitability than distant prey because of reduced approaching costs and reduced prey escape probability.

In summary, our behavioural data indicate that Mouse Lemurs, when foraging for insects, use the mass–amplitude correlation of prey-generated rustling sounds to guide their foraging behaviour. Their broadband hearing sensitivity appears to be an adaptation to rustling sounds of prey (and predators), which evolved earlier in ancestral insectivorous mammals to match the passive peripheral filtering of their auditory system to the spectrum of behaviourally relevant sounds. Subsequent active prey selection processes, e.g. based on amplitude, can solely decide using this filtered information.



We thank the Directions des Eaux et Forêt and the Commission Tripartie for authorization and the National Geographic Society and the Alfred Gottschalk-Stiftung Tübingen for funding our research. All research was conducted under license of the Ministière de l’Environnement des Eaux et Forêts. Excellent logistic support was provided by QIT Madagascar Minerals (special thanks to Jean-Baptiste Ramanamanjato and Manon Vincelette) and the German Primate Center (DPZ) Göttingen. We gratefully acknowledge the help and company of Eric Robsomanitrandrasana and Tiana Andrianjanahary in the field. We thank Olga Ramilijaona for insect identification and Jörg Ganzhorn, Peter Kappeler, Hans-Ulrich Schnitzler, Joachim Ostwald, Jean-Baptiste Ramanamanjato and Manon Vincellete for their friendly help and invaluable support. Rachel Page, Thomas Rice, Lutz Wiegrebe and two anonymous referees provided detailed and very helpful comments on earlier versions of this manuscript.