Individual neurophysiological signatures of spontaneous rhythm processing

When sensory input conveys rhythmic regularity, we can form predictions about the timing of upcoming events. Although rhythm processing capacities differ considerably between individuals, these differences are often obscured by participant- and trial-level data averaging procedures in M/EEG research. Here, we systematically assessed the neurophysiological variability displayed by individuals listening to isochronous equitone sequences interspersed with unexpected deviant tones. We first focused on rhythm tracking and tested the anticipatory phase alignment of delta-band activity to expected tone onsets. These analyses confirmed that individuals encode temporal regularities and form temporal predictions, but highlight clear inter- and intra-participant variability. This observation may indicate individual and flexible tracking mechanisms, which show consistency at the single-trial level, but variability over trials. We then modelled single-trial time-locked neural responses in the beta-band to investigate individual tendencies to spontaneously employ binary grouping (“tic-toc effect”). This approach identified binary (strong-weak), ternary (strong-weak-weak), and mixed accentuation patterns, confirming the superimposition of a basic beat pattern. Furthermore, we characterized individual grouping preferences and tendencies to use binary, ternary, or combined patterns over trials. Importantly, the processing of standard and deviant tones was modulated by the employed pattern. The current approach supports individualized neurophysiological profiling as a sensitive strategy to identify dynamically evolving neural signatures of rhythm and beat processing. We further suggest that close examination of neurophysiological variability is critical to improve our understanding of the individual and flexible mechanisms underlying the capacities to rapidly evaluate and adapt to environmental rhythms. Significance statement For decades, music, speech and rhythm research investigated how humans process, predict, and adapt to environmental rhythms. By adopting a single-trial and -participant approach, we avert the common pooling of EEG data in favor of individual time-varying neural signatures of rhythm tracking and beat processing. The results highlight large inter- and intra-individual differences in rhythm tracking, arguing against the typically documented phase-specificity for entrainment. On top of that, we characterize individual variability in beat processing, by showing that binary, ternary and other accentuation patterns are used over time, and ultimately affect the processing of (un-)expected auditory events. The approach aids individual neural profiling and may therefore allow identifying altered neural activity and its consequences in natural listening contexts.


Abstract 23
When sensory input conveys rhythmic regularity, we can form predictions about the timing of 24 upcoming events. Although rhythm processing capacities differ considerably between 25 individuals, these differences are often obscured by participant-and trial-level data averaging 26 procedures in M/EEG research. Here, we systematically assessed the neurophysiological 27 variability displayed by individuals listening to isochronous equitone sequences interspersed 28 with unexpected deviant tones. We first focused on rhythm tracking and tested the anticipatory 29 phase alignment of delta-band activity to expected tone onsets. These analyses confirmed that 30 individuals encode temporal regularities and form temporal predictions, but highlight clear inter- 31 and intra-participant variability. This observation may indicate individual and flexible tracking 32 mechanisms, which show consistency at the single-trial level, but variability over trials. We then 33 modelled single-trial time-locked neural responses in the beta-band to investigate individual 34 tendencies to spontaneously employ binary grouping ("tic-toc effect") . This approach identified 35 binary (strong-weak), ternary (strong-weak-weak), and mixed accentuation patterns, confirming 36 the superimposition of a basic beat pattern. Furthermore, we characterized individual grouping 37 preferences and tendencies to use binary, ternary, or combined patterns over trials. Importantly, 38 the processing of standard and deviant tones was modulated by the employed pattern. 39 The current approach supports individualized neurophysiological profiling as a sensitive strategy 40 to identify dynamically evolving neural signatures of rhythm and beat processing. We further 41 suggest that close examination of neurophysiological variability is critical to improve our 42 understanding of the individual and flexible mechanisms underlying the capacities to rapidly 43 evaluate and adapt to environmental rhythms. 44

45
Significance statement 46 For decades, music, speech and rhythm research investigated how humans process, predict, and 47 adapt to environmental rhythms. By adopting a single-trial and -participant approach, we avert 48 the common pooling of EEG data in favor of individual time-varying neural signatures of rhythm 49 tracking and beat processing. The results highlight large inter-and intra-individual differences in 50 rhythm tracking, arguing against the typically documented phase-specificity for entrainment. On 51 top of that, we characterize individual variability in beat processing, by showing that binary, 52 3 ternary and other accentuation patterns are used over time, and ultimately affect the processing 53 of (un-)expected auditory events. The approach aids individual neural profiling and may 54 therefore allow identifying altered neural activity and its consequences in natural listening 55 contexts. 56 1. Introduction 57 Due to the inherently rhythmic nature of many environmental stimuli, neurocognitive functions 58 such as attention (Lakatos et al., 2008) sensorimotor behavior (Merker et al., 2009), speech 59 (Giraud & Poeppel, 2012), reading (Goswami, 2011), and music (Doelling & Poeppel, 2015) 60 rely on basic timing capacities. To generate a temporally coherent representation of a rhythmic 61 environment, we track stimulus periodicities, use smart grouping, and continuously segment and 62 combine multiple inputs in time (Buzsáki, 2009;Schroeder & Lakatos, 2009; Thut et al., 2012a; 63 Zoefel & VanRullen, 2016). According to the dynamic attending theory (Large & Jones, 1999) 64 these timing processes reflect how internal rhythms synchronize with external rhythms. This and 65 similar theoretical views (Fraisse, 1963; p. 18) suggest that oscillatory brain activity instantiates 66 a realistic model for such "adaptation by anticipation". Accordingly, temporally regular sensory 67 input would make future events predictable and thereby facilitate sensory processing, 68 perception, allocation of attention, and the overall effectiveness of behavior (Friston, 2005;  , and next to regular rhythms, we frequently experience irregular rhythms or sudden 72 changes in the environment. To account for these dynamics, any realistic adaptation mechanism 73 likely tolerates a certain degree of temporal irregularity or unpredictability while trying to 74 achieve synchronization (Barnes & Jones, 2000). Endogenous oscillatory activity must hence not 75 only be precise and stable over time, but also flexible enough to achieve adequate adaptive 76 timing (e.g., by speeding-up or slowing-down accordingly). Indeed, oscillatory brain activity can 77 actively track and process (quasi-)periodic and never strictly isochronous signals such as speech, 78 which includes rhythmic variations at phoneme (25-35Hz) to syllable (4-8Hz) and word (1-3Hz) 79 rates (Giraud & Poeppel, 2012 To address this question, we let participants passively listen to isochronous (1.5Hz) equitone 93 sequences, comprising frequent standard and either one or two amplitude-attenuated deviant 94 tones while their EEG was recorded. Neural signatures of rhythm tracking were assessed by 95 quantifying the trial-level consistency of delta-band phase alignment towards expected tone 96 onsets. We expected to observe a predictive phase-alignment of the high-excitability phase 97 towards tone onsets. 98 Next, we focused on the known human disposition to group two or three adjacent tones when 99 listening to isochronous equitone sequences (Brochard et al., 2003). This results in perceived 100 binary (strong-weak (S-w)) or ternary (S-w-w) accentuation patterns, ultimately resembling beat 101 processing. In other words, auditory sequences are subdivided in regular groups of adjacent 102 events according to a regular superimposed beat structure. Importantly, even with physically 103 identical stimuli, this beat can influence observable behaviour and underlying neural activity  (Brochard et al., 2003). To this end, we modelled single-participant 111 and single-trial time-locked fluctuations in the beta-band, aiming to characterize inter-and 112 within-participant predispositions to superimpose beat-like accentuation patterns and opt for 113 binary, ternary, and combined beat processing over trials. We expected this analysis approach to 114 deliver insights into the intra-and inter-individual neurophysiological variability associated with 115 individual and flexible mechanisms employed to evaluate and adapt to (un)predictable 116 environmental rhythms. The stimuli comprised 192 sequences, consisting of 13-to-16 400Hz, 50ms, 70dB SPL tones 128 (standard; STD), presented in two recording sessions. One or two deviant tones (DEV), 129 attenuated by 4dB relative to the STD tones, were embedded in each sequence replacing STD 130 tones. The first DEV tone could either occur in an odd or even-numbered position (8-11 th ), while 131 the second always fell on the 12 th position ( Fig.1; corresponding to a hypothetical binary Strong-132 weak pattern; S-w). The inter-onset-interval between successive tones was 650ms, resulting in a 133 stimulation frequency of 1.54Hz. Stimuli were thus comparable to those used in previous EEG 134 studies on individual grouping (Brochard et al., 2003;Poudrier, 2020). 135 Participants were seated in a dimly lit soundproof chamber facing a computer screen. A trial 136 started with a fixation cross (500ms), followed by the presentation of the tone sequence. The 137 cross was continuously displayed on the screen to prevent excessive eye movements while tone 138 sequences were played . Immediately after the end of a sequence, a response screen appeared and 139 prompted participants to press a response button to indicate whether they heard one or two softer 140 tones. Button assignments were counterbalanced across participants. The inter-trial interval was 141 2000ms. A session was divided into two blocks of approximately 10 minutes each, with a short 142 pause in between. Therefore, an experimental session lasted for about 25 minutes.   procedure. Lastly, a standard whole-trial rejection procedure based on an amplitude criterion 229 (85uV) was applied. Data for event-related-potential (ERP) analyses ("ERP data") were 230 segmented, including 500ms prior and following each tone onset (1s in total). Data for the time-231 10 frequency representation analyses ("TFR data") were not further segmented at this stage. ERP 232 data were band-pass filtered between 1-30Hz and TFR data low-pass filtered at 40Hz. Data were 233 then downsampled to 250Hz. 234 TFR data underwent time-frequency transformation by means of a wavelet-transform (Cohen,235 2014), with a frequency resolution of .25Hz. The number of fitted cycles ranged from 3 for the 236 low frequencies (<5Hz) to 10 for high frequencies (>5Hz and up to 40Hz). TFR data were then 237 re-segmented to reduce the total length to 2s, symmetric around tone onsets. To test these hypotheses, we first vertically concatenated trials classified as "binary", and then 297 computed a trial-based single-tone pair-wise amplitude difference (corresponding lower-triangle 298 2-D means are provided in Fig. 4D), i.e., amplitude differences between responses to each tone  When individuals listen to these sequences, their neural activity reflects the timing of external 365 events ( Fig.2A-B). Indeed, the Fourier spectrum showed a clear peak at the stimulation 366 frequency (1.5Hz; Fig. 2A) and the time-course of delta-band neural activity showed a tendency 367 to align to tone onsets (Fig. 2B). To quantify the consistency of anticipatory phase alignment to 368 the expected tone onsets, we tested the phase consistency of delta-band (1.5Hz) neural activity in however, we observed intra-and inter-individual differences: participants' delta-band activity did 375 not always synchronize to tone onsets with the same phase relationship (Fig. 2C). Rather, a broad 376 range of possible phase-lags was observed across trials, both at the level of single-participants 377 (Fig. 2C right) as well as when pooling values across participants (Fig. 2D top-left). Hence, the 378 distribution of single-participant phase-angles across trials did not differ from a random 379 distribution (Suppl. Tab. 1). Phase-angles were consistent within a trial (MVL statistics in Fig.   380 2D bottom and Suppl. Tab. 1), but differed across trials. To further explore this variability, we 381 computed a measure of 'relative phase'. This was calculated, at the single-participant, channel-382 and sequence-level as the absolute difference from each phase-angles within the sequence and 383 15 the most common phase. The distribution of relative phase-angles across trials and participants 384 shows a variance which mostly ranges between 0-30 degrees (Fig. 2D, right)

468
A: Beat modelling was performed by means of stepwise regression modelling and using low-beta post-stimulus responses as a 469 dependent variable. The predictors were a binary (1, -1), a triple (1, -.5, -.5) and a constant term (ones). B: grouping 470 preferences, as reported from the beat modelling. In order, we plot the distribution of trials assigned to binary, ternary, 471 combined (binary-ternary) grouping, and 'not classified' (neither binary nor ternary) across participants. At the bottom, we 472 zoom into binary trials and distinguish S-w accents from w-S accents based on trial-level Beta coefficients from the beat 473 modelling. Similarly, on its right side, the distribution of ternary trials showing S-w-w, w-S-w, and w-w-S accentuation patterns.

474
To extract these three accentuation patterns, we performed separate step-wise regression modelling as explained in the method positions. Thus, we tested whether the beat modelling approach delivers a meaningful 489 classification of binary grouping. 490 We isolated the identified 'binary' beat trials and calculated the tone-by-tone pair-wise 491 difference for low-β across 8 positions in the auditory sequence and preceding the DEV tone. For 492 visualization purposes, the resulting matrix was averaged across trials and the upper symmetrical 493 triangle was masked (Fig.4D). The original matrix (all trials) was used to calculate metrics of 494 "Binary similarity" and "Binary dissimilarity" (Fig.4E; see 'Binary group analyses' in the 495 methods). The Binary similarity features the distributions of amplitude differences on odd-and 496 even-numbered positions. For the "Binary dissimilarity" analyses we calculated the amplitude 497 difference for tones on odd-versus even-numbered positions (corresponding to on-beat versus 498 off-beat; thus labeled "Binary difference") and statistically compared it to the Binary similarity 499 (right-side plot in Fig.4E). Statistical testing yielded a significant difference (FDR-adjusted p < 500 .05). 501 In summary, we confirmed that the trials classified as 'binary' in the beat modelling, do indeed 502 show a consistent binary accentuation patter. Hence, the low-β amplitudes in STD tones in S 503 positions significantly differ from those in w positions. To further verify the validity of the beat 504 modelling approach, we tested whether identified 'beat preferences' modulate DEV processing. 505 506 21 4.6. DEV processing based on binary grouping 507 We investigated whether DEV processing is modulated by a binary beat in 'binary' beat trials. 508 Thus, we tested whether ERPs to DEV tones falling on S-w beat positions in the identified 509 "binary" trials would be statistically different. First, we isolated the identified binary trials and 510 discerned 'pure binary' from 'inverse binary' trials based on the beta-coefficient resulting from 511 the regression modelling (see methods). Next, we pooled together odd-numbered (9,11 th 512 positions) and even-numbered (8,10 th ) DEV positions (i.e., corresponding to S-w). Within-513 participant statistical comparison of the respective ERPs yielded significant difference in the 514 time-window between 120-170ms (p FDR adjusted <.05; Fig. 4F). 515 Similarly, we tested whether the same S-w effect would be observed for those trials in which no 516 accentuation pattern could be identified ('non-classified' trials). In these non-classified trials, 517 DEV processing was not modulated by a binary beat (p FDR adjusted >.05; Suppl. Fig. 1). 518 Lastly, we focused on the 'ternary' trials and modelled three possible accentuation patterns: S-w-519 w, w-S-w, w-w-S (Fig. 4b). Given the small percentage of trials belonging to the three 520 accentuation types (~2%), we refrained from performing further analyses due to insufficient 521 statistical power to interpret results. 522 In summary, we here show that DEV processing is modulated by a binary beat, but exclusively onsets. Next, we tested whether neural activity in the low-beta band (12-20Hz) would reflect the 533 emergence of binary-like accentuation patterns, indicating spontaneous engagement in beat 534 processing. 535 When listening to isochronous equitone sequences, participants' neural activity tracked the 536 timing of external events ( Fig.2A), aligning delta-band oscillatory dynamics to expected tone 537 22 onsets (Fig.2) (Buzsáki, 2009;Schroeder & Lakatos, 2009;Thut et al., 2012b;Zoefel & 538 VanRullen, 2016). Hence, sequence-level mean vector lengths of delta-band activity preceding 539 tone onsets displayed anticipatory coupling of brain activity to the timing of environmental 540 stimuli. This finding further confirms theoretical views according to which the brain might 541 generate temporal predictions to achieve successful rhythm tracking to optimize sensory 542 processing, perception, and allocation of attention (Friston, 2005 Next to rhythm tracking, we investigated the neural signatures associated with the human 556 disposition to group two or three adjacent tones when listening to isochronous equitone 557 sequences (Brochard et al., 2003). This spontaneous grouping induces binary (strong-weak (S-558 w)) or ternary (S-w-w) accentuations patterns (Brochard et al., 2003), and might resembles the 559 superimposition of a beat structure. Importantly, while the tones are physically identical, the beat 560 can influence observable behaviour and underlying neural activity (Nozaradan,  The current findings confirm its prevalence in time-locked responses to STD tones (Fig.3B). 572 Furthermore, beta-band activity showed an individuals' spontaneous disposition to superimpose 573 a beat pattern, even when not instructed to do so (Fig.4). Hence, we characterized inter-and 574 within-participant differences in adopting binary, ternary, and combined grouping over time (  588 We acknowledge that classifying trials based on beta fluctuations and then testing beta time-589 locked responses over sequence positions might be circular, but confirmatory. Thus, to test this 590 further, we compared neural responses to unpredicted (deviant) tones falling on S-w positions 591 and found a significant effect in ERP responses to DEV tones (Fig. 4F). Notably, the modulation 592 of DEV processing was absent in those trials which did not adhere to a specific accentuation 593 pattern (non-classified trials; Suppl. In summary, we identify individual rhythm tracking and beat processing differences, and 599 24 associate them with specific delta-band phase-coupling mechanisms and with beta-band 600 dynamics, respectively. The findings showcase the feasibility of using EEG to identify individual 601 neurophysiological signatures in rhythm cognition, suggesting that common trial-and group-602 level averaging approaches might inevitably obscure inter-individual differences and trial-by-603 trial variability. In contrast, the approach adopted here allows the monitoring of

801
The table reports the percentages of individual preferences for (in order) binary, ternary, combined (binary + ternary) beat. The

802
last column includes all trials which could not be classified as the above-mentioned beat strategies. The last column on the right 803 provides the goodness of fit (expressed as Eta-squared, 'R2') of the chosen models, averaged across trials per subject.