All the animal care, housing, experimental procedures were approved by the National University of Mexico Institutional Animal Care and Use Committee and conformed to the principles outlined in the Guide for Care and Use of Laboratory Animals (NIH, publication number 85-23, revised 1985). Both monkeys were monitored daily by the researchers and the animal care staff, and every second day from the veterinarian, to check the conditions of health and welfare. To ameliorate their condition of life we routinely introduced in the home cage (1.3 m³) environment toys (often containing items of food that they liked) to promote their exploratory behavior. The researcher that tested the animals spent half an hour interacting with the monkeys directly, giving for example new objects to manipulate. We think that this interaction with humans, in addition to the interaction that was part of the task performed, can help to reduce potential stress related to the experiment. Food and water where given ad libitum.

Two rhesus monkeys participated in the ERP measurements. Aji, a 2 year old male (referred to as monkey A) and Yko, a 5 year old male (referred to as monkey Y). Both monkeys have normal hearing. They were awake (i.e. not sedated) during the measurements, sitting in a quiet room [3 (l)×2 (d)×2.5 (h) m] with dimmed lighting and two loudspeakers in front of them. The ERP measurements were performed after a morning session of unrelated behavioral experiments. The animals were seated comfortably in a monkey chair where they could freely move their hands and feet. No head fixation was used and the EEG electrodes were attached to the monkey’s scalp using tape. To ease the fixation of the electrodes, the monkey’s hair on the scalp and reference ear was shaved.

In Experiment 1 pure sine-wave tones were used for the two-stimulus oddball paradigm. Their frequencies were 500 Hz and 1500 Hz, with a duration of 50 ms, and a rise and fall of 5 ms. The frequencies of these tones were within the audible range of both monkeys.

In Experiment 2 a sine-wave with a frequency 1000 Hz was used, with a duration of 50 ms and a rise and fall of 5 ms.

In Experiment 3 sound sequences based on a typical 2-measure rock drum accompaniment pattern (S₁) were used, composed of snare, bass and hi-hat, spanning equally spaced positions (see Figure 1). Four further variants of the S₁ pattern (S₂–S₄ and D) were created by omitting sounds in different positions. Within the patterns the onset-to-onset interval between successive sounds was 150 ms with 75 ms onset-to-offset interval (75 ms sound duration). Patterns in the sequence were delivered as a continuous sound stream. Loudness of the sounds was normalized so that all stimuli had the same loudness.

Sound stimuli were presented through 2 loudspeakers placed 1.1 meters away from the subject (and 1 meter apart from each other). The sound intensity measured at the subject position was approximately 60 dB SPL.

In all three experiments an auditory oddball paradigm was used. The experimental paradigm was adapted from previous studies (Experiment 1: [16]; Experiment 2: [21]; Experiment 3: [20]).

In Experiment 1 and 2 sound inter-onset-intervals were 600 ms and 150 ms, respectively. In both experiments standards (0.9 probability) were randomly replaced (0.1 probability) with deviants and deviant omissions (i.e. silence), respectively. In Experiment 1 for half of the blocks one frequency was used as deviant and the other as standard (i.e. S₅₀₀, D₁₅₀₀), switching roles for the other half of the blocks (i.e. S₁₅₀₀, D₅₀₀). In Experiment 2 the inter-onset-interval was 150 ms (an interval motivated by human studies [21], and that is within the ‘preferred tempo’ range of rhesus monkeys [22]).

In Experiment 3 the 4 strictly metrical sound patterns (S₁–S₄; standards) made up the majority of the patterns in the sequences (0.225 probability, respectively). In the standard patterns regular omissions occurred in metrically weak positions, leaving these patterns metrically intact. Occasionally, the D pattern was delivered (0.1 probability) in which the downbeat was omitted, which interrupted the metricality of the pattern. The order of the five patterns was pseudo-randomized, enforcing at least three standard patterns between successive D patterns and no D after S₄ to avoid two consecutive omissions. A control sequence (deviant-control) repeating the D pattern 100% of the time was also delivered (see [20] for more details).

The ERP measurements were conducted in a repeated session, containing all three experiments in random order. The monkeys participated in one recording session per day, to a total of 11 sessions for monkey A and 23 sessions for monkey Y (monkey Y moved considerably more than Monkey A). All measurement was completed in about one month per monkey. Each experiment consisted of 10 blocks with 306 repetitions for each block.

The EEG was recorded from electrodes (Grass EEG electrodes; #FS-E5GH-60) attached to five scalp positions (Fz, Cz, Pz, F3, F4) according to the 10–20 system (see Figure 2).

The electrodes were connected to a Tucker-Davis Technologies (TDT) headstage (#RA16LI) for low impedance electrodes. This headstage was connected to a TDT RA16PA preamplifier, which in turn was connected to a TDT RZ2 processor. RZ2 was programmed to acquire the EEG signals with a sampling rate of 498.25 Hz and the bandpass filters were set at 0.01–100 Hz.

All electrodes were attached using Ten20 Conductive EEG Paste and medical tape, and were referenced to the right ear (fleshy part of the pinna). In the offline analysis, a 0.1–30 Hz band-pass FIR filter (Kaiser-window) was applied. With zero latency set to the onset of the stimuli, epochs of −100–500 ms (Experiment 1), 0–450 ms (Experiment 2), and 0–600 ms (Experiment 3) were extracted. All epochs were baseline corrected to zero using a 100 ms pre-stimulus interval in Experiment 1 and the whole epoch in Experiments 2 and 3. Epochs that exceeded +/−150 µV amplitude were excluded from the statistical analysis. EMG recordings were obtained from the temporalis muscles. No event-locked activity was found in these recordings. The number of epochs accepted for analysis for the three experiments are given in the Tables 1–4.

Statistical analysis was performed on the mean amplitudes in a 50 ms wide time window centered on the absolute maximum peak of difference waveforms (i.e. the difference between the standard and deviant wave). The resulting windows are stated underneath the Tables 1 to 4 and marked with gray-shaded rectangles in Figures 3, 4, 5. In all three experiments channel Cz was used for the latency measurements.

The resulting values were fed into an analysis of variance (ANOVA), where Electrode sites were treated as a within subject variable and all other variables as grouping variables. For Experiment 1 factors Stimulus (500 Hz vs. 1500 Hz) × Type (Deviant vs. Standard) × Electrode (Fz vs. Cz vs. Pz vs. F3 vs. F4) were used, for Experiment 2 Type (Omission vs. Sound) × Electrode (Fz vs. Cz vs. Pz vs. F3 vs. F4), and for Experiment 3 Type (Deviant vs. Deviant control vs. S_1–4) × Electrode (Fz vs. Cz vs. Pz vs. F3 vs. F4). Greenhouse-Geisser correction was used where necessary (corrected p, df and epsilon values reported).

In Experiment 1 we presented two rhesus monkeys with a sequence of sounds using a two-tone oddball paradigm (see Methods) to see whether an MMN-like response can be elicited.

Figure 3 shows that the electrical brain responses elicited by the standard and deviant stimulus are different for both monkeys, with a morphology comparable to a human MMN, though with a shorter latency (peaks around 90 ms, instead of 150 ms) and slightly larger amplitude as compared to humans (around 10 µV, instead of 5 µV) [23]. These differences in latency and amplitude can be attributed to the anatomical differences between human and monkey brains (e.g., skull size, thickness, and the distribution of musculature [24]).

For monkey A the ANOVA with factors Stimulus (500 Hz vs. 1500 Hz) × Type (Deviant vs. Standard) × Electrode (Fz vs. Cz vs. Pz vs. F3 vs. F4) revealed significant main effects in Type (F (1, 47395) = 104.555, P<0.001, η² = 0.002), Stimulus (F (1, 47395) = 12.045, P<0.001, η²<0.001) and Electrode (F (3.202, 151750.6) = 151.684, P<0.0001, η² = 0.002, ε = 0.800) as well al a Stimulus × Type interaction (F (1, 47395) = 31.476, P<0.001, η² = 0.003). All interactions involving the Electrode factor were significant, namely Electrode × Stimulus (F (3.202, 151750.6) = 2.723, P<0.05, η² = 0.002, ε = 0.800), Electrode × Type (F (3.202, 151750.6) = 51.294, P<0.0001, η²<0.001, ε = 0.800) and Electrode × Stimulus × Type (F (3.202, 151750.6) = 3.113, P<0.01, η²<0.05, ε = 0.800). Tukey unequal-N HSD post-hoc tests revealed no significant difference between F3 and F4 and and Type having no effect on Pz. Additionally the effect of Type was only marginally significant (df = 47395, P = 0.066) on 1500 Hz stimuli.

For monkey Y the mean negative amplitude for deviant stimuli was significantly greater than that for standard stimuli. An ANOVA with the same factors revealed significant main effects in Type (F (1, 78644) = 206.474, P<0.001, η² = 0.003) and Electrode (F (2.336, 163892.2) = 181.928, P<0.0001, η² = 0.002, ε = 0.584). All interactions involving the Electrode factor were significant, namely Electrode × Stimulus (F (2.336, 163892.2) = 3.543, P<0.05, η² = 0.002, ε = 0.584), Electrode × Type (F (2.336, 163892.2) = 35.920, P<0.0001, η²<0.001, ε = 0.584) and Electrode × Stimulus × Type (F (2.336, 163892.2) = 6.034, P<0.0001, η²<0.001, ε = 0.584). Tukey unequal-N HSD post-hoc tests revealed no difference between F3 and F4 and Type having no effect on Pz.

An MMN-like response was found for the deviant responses as compared to physically identical standards in a time-window centered on the absolute maximum of the difference waves (D₅₀₀–S₅₀₀, D₁₅₀₀–S₁₅₀₀; See Table 1 and gray-shaded windows in Figure 3).

The results show that physically identical deviant and standard stimuli elicited different responses. The average amplitude of the responses for both monkeys tended to be large in the frontal and central areas, similar to a human MMN [23]. Table 1 shows the mean amplitudes for monkey A and monkey Y, for each condition, stimulus type and electrode position. There was no indication of hemispheric differences.

These results are in line with another study showing an MMN-like response in a single chimpanzee (Pan troglodyte) [16] using the same two-tone odd-ball paradigm with scalp-recorded EEG. Together with the current experiment these studies provide evidence that ERP and MMN can be measured in both monkeys and apes.

To study whether an MMN can be elicited in response to omissions as well, the same rhesus monkeys were presented with a tone sequence in which tones were omitted (i.e. replaced by silence, see Methods).

Figure 4 shows the electrical brain responses elicited by the standard (S) and the deviant (D; an omission). (Note that Figure 4 shows a time window with three repetitions of the standard tone, marked by rectangles at either side of the time line.) This allows for a comparison of the responses to the first and second tone after the omission. To test the effects of the omission we concentrate on the time range closest to the occurrence of the omission (see Methods; Table 2). In both monkeys the standard stimuli elicit a steady-state response with increased amplitude, phase-aligned to the stimuli. The amplitude of the response for the first tone after the omission (see Figure 4), most notably in monkey Y, neural activity increased after the short period of silence, but returns near to previous levels by the second tone. This could also be interpreted as a response marking the beginning of a rhythmic group [25].

Mean amplitudes of responses elicited by standard and deviant stimuli were measured within a time window centered on the absolute maximum of the D minus S difference waves (see Table 2 and gray-shaded windows in Figure 4).

For monkey A an ANOVA with factors Type (Omission vs. Tone) × Electrode (Fz vs. Cz vs. Pz vs. F3 vs. F4) revealed significant main effects in Type (F (1, 15708) = 32.906, P<0.0001, η² = 0.002) and Electrode (F (2.894, 45465.48) = 32.049, P<0.001, η² = 0.002, ε = 0.724). Tukey unequal-N HSD post-hoc tests revealed no significant difference between F3 and F4.

For monkey Y an ANOVA with the same factors revealed significant main effects in Type (F (1, 27307) = 10.648, P<0.005, η²<0.001) and Electrode (F (2.255, 61581.99) = 7,477, P<0.001, η²<0.001, ε = 0.564). Tukey unequal-N HSD post-hoc tests revealed no significant difference between F3 and F4.

Again for both monkeys the average amplitude tended to be large in the frontal and central areas, without any laterality effects.

The ERP responses to the omission (red lines in Figure 4) have a morphology comparable to human MMN (i.e. negative in early latencies). However, the polarity of the responses, probably due to inter-individual differences, were different in the two monkeys. Nevertheless, there is a small, but significant amplitude difference between the standard tone and the omission in a time range comparable to human MMN [21], [26] suggesting that the omission was indeed detected.

In Experiment 3 we presented the same two rhesus monkeys with complex stimuli consisting of sound sequences based on a typical rock drum accompaniment pattern (see Figure 1).

The standard stimuli are four randomly presented and strictly metrical sound patterns (S₁–S₄), with a deviant pattern (D) presented which the ‘downbeat’ omitted. Humans adults perceive the D pattern within the context of standards as if the rhythm was broken, stumbled, or became strongly syncopated for a moment [20]. We refer to the omission at the start of D as a ‘loud rest’ and the omissions in S₂–S₄ as ‘silent rests’; Music theory suggests the former to sound ‘syncopated’ (a violation of a metric expectation) and the latter not [3].

A sequence repeating the D pattern 100% of the time was also presented (‘deviant-control’ or D_control) to allow controlling for acoustic effects on the ERP.

On the basis of the dissociation hypothesis, and the observation that monkeys apparently can not synchronize to a beat [7] but are sensitive to auditory timing [12], one might expect that monkeys are sensitive to rhythmic structure (interval-based timing) but not to metric structure (beat-based timing). This hypothesis predicts that omissions that play a role in rhythmic grouping [27] can be detected, as they mark the structure of a rhythmic pattern (as is the case in D_control), consequently not eliciting an MMN as they are part of the regularity. In contrast, the omissions that do not affect the rhythmic grouping will not be detected as part of a regularity, since they occur irregularly (as is the case in S₂–S₄ and D) and hence may elicit an MMN.

In humans these differences in salience appear to be related to the coding of an internal representation of the rhythmic structure of a sound pattern [27], with the first sound after a relatively long inter-onset interval determining the rhythmic group structure [25]. If this is the case we expect the first sound of a repeated rhythmic pattern (D_control) – but not a randomly inserted pattern (D) – to elicit a response marking the beginning of a rhythmic group [25].

An alternative hypothesis is based on the observations made in human adults and newborns using the same stimuli and experimental paradigm [17], [18], [19], [20]. This hypothesis predicts that primates are not only able to sense rhythmic grouping, but are also able to detect the regular beat that is induced by a varying rhythmic stimulus. The perception of a ‘loud rest’ – a violation of a temporal expectation reflected by an MMN-like signal– can serve as evidence for the presence of a strong metric expectation [3]. This hypothesis predicts an large and early MMN for the omission in the deviant (D, containing a ‘loud rest’), but no or considerably smaller MMN for the omissions in the standard (S₂–S_4, containing ‘silent rests’). And since the omission in the deviant-control (D_control) is expected – the pattern is presented repeatedly –, there as well no MMN is predicted. If these three aspects are observed (as they were found in human adults and newborns [20]), they suggest that a regular beat is extracted from the auditory stimulus. This could be interpreted as evidence against the vocal learning hypothesis.

Figure 5 shows that the electrical brain responses elicited by omissions in the standard (S₂–S₄) and deviant-control (D_control) are relatively flat, and different from the deviant (D), with the latter eliciting a more pronounced negative peak, most notably in monkey Y. This suggest a similar result as was found human adults and newborns. However, the ERP response to S₁ (dotted black line in Figure 5) is not different from that in response to D (solid red line in Figure 5), while D contains an omission and S₁ does not. This seriously weakens the interpretation that the monkeys are able to extract the beat from the stimulus.

Mean amplitudes of responses elicited by standard and deviant stimuli were measured within a time window centered on the absolute maximum of the D minus S_2–4 difference waves (see Table 3 and the early gray-shaded windows in Figure 5).

For monkey A an ANOVA with factors Type (S₁ vs. S₂ vs. S₃ vs. S₄ vs. D_control vs. D) × Electrode (Fz vs. Cz vs. Pz vs. F3 vs. F4) in the early window (105–155 ms) showed significant main effects in Type (F (5, 37269) = 89.318, P<0.0001, η² = 0.012) and Electrode (F (3.006, 112063.6) = 11.221, P<0.0001, η²<0.001, ε = 0.752), as well as a significant Electrode × Type (F (15.034, 112063.6) = 7.475, P<0.0001, η² = 0.001, ε = 0.752) interaction. Tukey unequal-N HSD post-hoc tests were performed. All channels differed from each other (df = 149076, P<0.05) except for Cz, F3 and F4 not differing from each other. All Types differed from each other (df = 49071, P<0.01), except D, D_control and S₁ from each other and S₃ from S₄.

For monkey Y an ANOVA with factors Type (S₁ vs. S₂ vs. S₃ vs. S₄ vs. D_control vs. D) × Electrode (Fz vs. Cz vs. Pz vs. F3 vs. F4) in the early window (73–123 ms) showed significant main effects in Type (F (5, 49071) = 74.323, P<0.0001, η² = 0.008) and Electrode (F (2.412, 118344.7) = 48.423, P<0.0001, η² = 0.001, ε = 0.603), as well as a significant Electrode × Type (F (12.059, 118344.7) = 9.479, P<0.0001, η² = 0.001, ε = 0.603) interaction. Tukey unequal-N HSD post-hoc tests were performed. All channels differed from each other (df = 196284, P<0.05) except for F3 and F4 and Fz not differing from F3. All Types differed from each other (df = 49071, P<0.001), except D from S₁; S₂ from S₃ and S₄ from D_control also the difference between S₃ and S₄ was less significant (P<0.05) than other differences.

So in short, while there is a difference between D (containing a ‘loud rest’) and S₂–S₄ (containing ‘silent rests’) and as such evidence in support of beat perception, there is no difference between D and S₁: a pattern with and without an omission. This makes the interpretation that the monkeys are detecting the beat (by distinguishing ‘loud rests’ from ‘silent rests’) less likely and leads to the alternative hypothesis that the monkeys are solely detecting rhythmic groups [21]–[22]: the first note of a rhythmic group (separated by an omission) eliciting an MMN-like response in D_control (but not in D).

Mean amplitudes were measured in a late time window just after the first tone (after 200 ms), centered on the absolute maximum of the D minus D_control difference waves (see Table 4 and the late gray-shaded windows in Figure 5).

For monkey A the ANOVA with the same factors on the late window (214–264 ms) showed significant main effects in Type (F (5, 49071) = 71.134, P<0.0001, η² = 0.009) and Electrode (F (2.975, 110879.9) = 35.850, P<0.0001, η²<0.001, ε = 0.744), as well as a significant Electrode × Type (F (14.876, 110879.9) = 19.880, P<0.0001, η² = 0.003, ε = 0.744) interaction. Tukey unequal-N HSD post-hoc tests were performed showing that D was significantly different from D_control (df = 37269, P<0.001) while not differing from S₁. All channels differed from each other (df = 149076, P<0.001) except for Cz and F4.

For monkey Y the ANOVA with the same factors on the late window (220–270 ms) showed significant main effects in Type (F (5, 49071) = 195.816, P<0.0001, η² = 0.020) and Electrode (F (2.412, 118344.7) = 283.270, P<0.0001, η² = 0.006, ε = 0.604), as well as a significant Electrode × Type (F (12.059, 118344.7) = 47.789, P<0.0001, η² = 0.005, ε = 0.604) interaction. Tukey unequal-N HSD post-hoc tests were performed showing that D was significantly different from D_control (df = 49071, P<0.001) while not differing from S₁. All channels differed from each other (df = 196284, P<0.001) except for Pz and F4.

These results suggests that the monkeys are actually sensing surface-level rhythmic grouping (i.e. detecting the start of a repeating rhythmic group) instead of the induced beat (i.e. detecting a regular pulse in a varying rhythmic pattern). As such, we have to conclude that rhesus monkeys, contrary to what has been shown for human adults and newborns, show no sign of representing the beat in music, but apparently do represent rhythmic groups.