Signatures of time interval reproduction in the human electroencephalogram (EEG)

Accurately judging temporal durations is critical for coordinating our actions in everyday tasks such as hitting an oncoming tennis ball, safely crossing the road, or preparing a meal. A significant body of behavioural, computational, and neurophysiological research has sought to identify the fundamental processes through which the brain tracks the passage of time using tasks in which observers are required to reproduce or discriminate between intervals ranging from hundreds of milliseconds to a few seconds, known as interval timing (Buhusi & Meck, 2005; Coull et al., 2011; Finnerty et al., 2015; Grondin, 2010; Hardy & Buonomano, 2016; Mauk & Buonomano, 2004; Merchant et al., 2013; Nobre et al., 2007; Paton & Buonomano, 2018). In non-human primates, behavioural variations in interval timing have been linked to ramping neural activity in cortical areas associated with planning and executing movements (Janssen & Shadlen, 2005; Leon & Shadlen, 2003; Merchant et al., 2011). For example, Jazayeri and Shadlen (2015) found that when macaques reproduced interval durations via eye movement responses, a faster build-up rate of spiking in the lateral intraparietal area (LIP) predicted faster reproduction times. Additionally, LIP activity levels at the start of the reproduction interval were positively correlated with reproduction times. This suggests that in areas involved in preparing the decision-reporting actions, interval timing is achieved via dynamic motor plans that adapt their starting point and build-up rate to reach a fixed action-triggering threshold at interval offset.

The extent to which analogous human brain signals undergo equivalent adjustments has yet to be determined. Perceptual decision-making research has demonstrated that effector-selective mu/beta band (8–30 Hz) signals over premotor sites contralateral to the response hand exhibit many choice-relevant characteristics in common with LIP neurons, including an evidence-dependent ramp to threshold relationship with choice reports and systematic starting level and build-up rate modulations in response to factors such as speed pressure and prior knowledge (de Lange et al., 2013; Donner et al., 2009; Kelly et al., 2021; Murphy et al., 2016; O’Connell & Kelly, 2021; Steinemann et al., 2018). Contralateral mu/beta signals would therefore appear likely candidates for tracing the evolution of interval timing decisions in humans. While several studies have reported positive correlations between fronto-central beta (15–30 hz) amplitudes and interval reproduction times (Kononowicz & Van Rijn, 2015; Kononowicz et al., 2019), we are not aware of any studies that have specifically examined the role of effector-selective mu/beta signals in interval reproduction.

A long-standing account of interval timing proposes that the brain encodes elapsed time by accumulating temporal pulses (Gibbon et al., 1984; Treisman, 1963), but a neural signature of such a process has yet to be identified. A promising candidate for such a signal is the centro-parietal positivity (CPP), another human EEG signature of perceptual decision formation that is functionally distinct from contralateral mu/beta. A feature of the CPP that is apparently unique among decision signals reported in humans and non-humans is that it traces evidence accumulation independent of the specific sensory or motor requirements of the task (Kelly & O’Connell, 2013; O’Connell et al., 2012). Unlike mu/beta or LIP signals, its amplitude at response declines systematically as a function of RT, speed pressure, and prior knowledge; while its starting levels remain unchanged (Kelly et al., 2021; Steinemann et al., 2018). These functional distinctions suggest that the CPP traces a motor-independent representation of cumulative evidence whose amplitude at the time of commitment is determined by strategic influences operating at the motor level (see O’Connell & Kelly, 2021 for full review). Thus far, the CPP has only been examined in the context of tasks that require the evaluation of external sensory evidence. The degree to which the CPP might also accumulate internally generated evidence, such as temporal pulses, has yet to be tested.

Many human EEG studies of interval timing have focused on a fronto-central negativity known as the Contingent Negative Variation (CNV) (Ruchkin et al., 1977; Walter et al., 1964; Weinberg et al., 1974). Early work suggested that the CNV might itself index a temporal pulse accumulator (Bendixen et al., 2005; Casini & Macar, 1999; Macar & Vidal, 2002; Wiener et al., 2012) but subsequent observations that the CNV’s amplitude is larger for shorter rather than longer reproductions (Kononowicz et al., 2015), that its amplitude is modulated by previous trial interval duration (Wiener & Thompson, 2015), and that timing performance is better predicted by post-interval offset ERPs (Bueno & Cravo, 2021; Damsma et al., 2021; Kononowicz & Van Rijn, 2014; Kruijne et al., 2021; Ofir & Landau, 2022), have led to the consensus that the CNV is not specific to timing, but instead indexes a more general anticipatory or preparatory process that reflects the level of readiness to process or respond to upcoming stimuli (Kononowicz & Penney, 2016; Kononowicz et al., 2018; Van Rijn et al., 2011). However, a precise functional account of this process has been lacking. One potential explanation derives from research on perceptual decision making which has shown that, like effector-selective mu/beta, the CNV exhibits larger amplitudes at the outset of decision formation when time pressure is increased and prior to faster response times, and correlates with model-derived estimates of decision boundary elevation (e.g., Boehm et al., 2014). This has led to the hypothesis that the CNV may index the rate of ‘urgency’ (Ditterich, 2006; Drugowitsch et al., 2012; Thura et al., 2012) to respond, effectively adjusting the decision boundaries to take account of the passage of time. Testing this hypothesis in the context of interval timing necessitates not only measurement of the CNV’s starting levels but also its build-up rate during reproduction and its amplitude prior to commitment.

The aim of the present study was to determine whether and how the dynamics of the CPP, premotor mu/beta, and the CNV predict interval timing. In order to bridge the gap between human and non-human primate work, we used an auditory version of Jazayeri and Shadlen’s (2015) time reproduction task as our behavioural paradigm, and developed our hypotheses with close reference to the motor preparation dynamics those authors observed in macaque LIP. We hypothesised that mu/beta band motor preparation activity would build towards a fixed threshold at response, with starting levels shifted closer to this threshold and a steeper build-up rate for faster reproductions. We hypothesised that the CNV would follow a similar pattern to mu/beta, with a steeper pre-response slope and a larger initial amplitude for faster reproductions. We hypothesised that if the CPP indexes temporal accumulation, it would accumulate towards a fixed bound prior to response, with a steeper build-up rate for shorter reproductions.

Participants were 25 young adults (12 females, mean age: 25 years, age range: 18–36 years). All participants had normal or corrected to normal vision, had no personal or family history of epilepsy or unexplained fainting, no sensitivity to flickering light, and no personal history of neurological or psychiatric illness or brain injury. Two participants were left-handed. Participants provided written consent prior to their participation. The experimental procedures were approved by the Research Ethics Committee of the School of Psychology, Trinity College Dublin in accordance with the Declaration of Helsinki. Participants received €20 for their participation. Participants who were undergraduate psychology students at Trinity College had the option to receive research credits as an alternative to the €20 payment. Six participants were removed from the analyses due to excessive artifacts (fewer than 25 trials per condition), leaving 18 participants for the analyses (9 females, mean age: 24.78 years, age range: 19–36 years).

Participants sat in a dark, sound-attenuated room with their head placed in a chin rest located approximately 45 cm away from a computer monitor. Visual stimuli were displayed at a 1024 x 768 screen resolution and a refresh rate of 60 Hz using a 53 cm CRT monitor. The task was programmed and presented using MATLAB (Mathworks, Natick, MA) and the Psychtoolbox extension of MATLAB (Kleiner et al., 2007). The sample time intervals presented to participants were 4800 Hz tones, played at a comfortable volume level. Prior to beginning the experiment, participants completed five training blocks. In training blocks, auditory stimuli were presented using Dell AX210 speakers. In experimental blocks, auditory stimuli were presented using Sennheiser HD 650 headphones.

Participants performed a modified version of the Ready-Set-Go time interval reproduction task (Fig. 1a) originally developed by Jazayeri and Shadlen (2010). Participants self-initiated each trial by pressing the spacebar. Each trial commenced with the appearance of a centrally presented red dot of 2 pixels on which participants were required to maintain fixation throughout the duration of the trial. Participants were then presented with a sample interval, represented by a single, auditory tone of variable duration (600 ms, 650 ms, 700 ms, 750 ms, 800 ms, 850 ms, 900 ms, 950 ms, or 1000 ms with equal probability). Following interval offset, there was a short gap which randomly varied between 350 ms and 550 ms after which participants were presented with a 100 ms cue tone prompting them to reproduce the sample interval by pressing the spacebar such that the time between cue and button press matched their perception of the sample interval. We introduced this gap to create more separation between interval encoding activity and interval reproduction activity due to the potential for signal overlap (e.g., auditory potentials evoked by the offset of the sample tone). Participants performed 16–20 blocks of 45 trials. Each sample interval was presented once every 9 trials in random order.

Participants were given positive feedback (fixation dot turned green at the end of the trial) if the reproduced time was within a certain range around the sample interval. To ensure the probability of positive feedback was comparable across sample intervals, the width of the feedback window was scaled with a constant of proportionality, k. The parameter k was initially set to 0.3 at the beginning of the participant’s first training block and was modified throughout the experiment as a function of the participant’s performance via a one-up one-down procedure which subtracted 0.015 from k for each response within this criterion and added 0.015 to k following responses that exceeded it. For trials with responses exceeding this criterion, the fixation dot remained red.

We measured participants’ reproduction time (RT) from the midpoint of cue onset (50 ms) to the time of response. We removed outlier reproduction times that were less than 150 ms and greater than 1500 ms (less than 1% of reproduction times). We used a Linear Mixed Model (LMM) to assess the relationship between Sample Interval and Reproduction Time with Sample Interval included as a fixed effect. In addition to fitting random intercepts across participants, the slope of the Sample Interval effect was also included as a random effect. We also included Gap duration in the model as a fixed effect to assess whether it impacted Reproduction Time. Due to a recording error, the distribution of gap durations for five of the participants was shifted by around 100 ms (250 ms to 450 ms). Therefore, we z-scored Gap within-participants so that its value was relative to the distribution of gaps a given participant was presented with. To help the model converge, Sample Interval was z-scored so that Sample Interval and Gap were on the same scale. To obtain standardised beta estimates for the fixed effects, Reproduction Time was also z-scored across participants. Previous studies have established that participants’ reproduction times are biased towards the mean of the underlying distribution of sample intervals. Therefore for each participant, we computed linear regressions with Sample Interval and Gap as predictors, and Reproduction Time as the dependent variable, to examine whether some participants were more veridical responders than others.

Continuous EEG data were acquired using an ActiveTwo system (BioSemi) from 128 scalp electrodes, digitised at 512 Hz. Data were analysed in custom MATLAB (Mathworks, Natick, MA) and Python 3 scripts (Kluyver et al., 2016), using functions from the EEGLAB (Delorme & Makeig, 2004) and MNE-python (Gramfort et al., 2013) toolboxes. Blinks were detected using two vertical electro-oculogram (EOG) electrodes that were placed above and below the left eye. EEG data were low-pass filtered offline at 30 Hz. No offline high-pass filter was applied but EEG data for each block were detrended to remove slow linear drifts. Noisy channels were interpolated, and the EEG data were re-referenced offline to the average reference.

EEG data were segmented into three epochs: -200 ms to 1100 ms relative to sample interval onset, -300 ms to 900 ms relative to reproduction cue onset, and -1100 ms to 100 ms relative to the participant’s response time. All three of the epochs were baseline corrected relative to the average activity from -200 ms to 0 ms preceding sample interval onset. Epochs were rejected if absolute scalp electrode activity exceeded 100 μV or bipolar EOG activity (upper minus lower) exceeded 200 μV at any time point in the epoch. Previous work (Kelly & O’Connell, 2013) has demonstrated that volume conduction effects can cause the CPP and CNV to interfere with one another, complicating the interpretation of effects. Therefore, in order to reduce signal overlap, EEG data were transformed into Current Source Density (CSD) following artefact rejection using the CSD toolbox in MATLAB (Kayser & Tenke, 2006).

In order to select electrodes for each signal of interest, we generated a grand-average ERP topography after collapsing across all trial types. Based on the grand-average topographies, we measured the CNV from two fronto-central electrodes (located at standard site FC) centred on the maximum negative grand-average pre-response amplitude. To measure effector-selective Mu/Beta (8–30 Hz) activity, we computed a Short-Time Fourier Transform (STFT) with a window size of 375 ms (3 full cycles of the lowest mu/beta frequency) and a step size of 25 ms. Mu/Beta was measured from four electrodes on the hemisphere contralateral to the hand the participant responded with, centred around the standard sites C3 or C4. As in previous studies examining the CPP, we measured the signal from two central-parietal electrodes, located at the standard site Pz.

We examined the relationship between Sample Interval and the EEG signatures using single-trial LMMs. We focused our analyses of the EEG signatures on three time windows: Sample Interval Offset, Cue Onset, and Reproduction. At Sample Interval Offset, we examined whether the amplitude and build-up rate of the EEG signatures varied with Sample Interval duration. We measured the mean amplitude of the CNV and CPP in a window of -50 ms to + 50 ms from Sample Interval Offset. The build-up rate of each signal was calculated as the slope of a straight line fitted to activity in the same time window. To account for the temporal smearing resulting from the STFT, we measured the mean amplitude and build-up rate of Mu/Beta in a window of -200 ms to -100 ms before Sample Interval Offset. We assessed the effect of Sample Interval on these measures using LMMs with Sample Interval as a fixed effect, including random intercepts and random slopes across participants as random effects. As in the behavioural LMMs, Sample Interval and the dependent variable were standardised across participants to aid model convergence and obtain standardised beta coefficients for evaluating effect sizes.

Around Cue Onset, we examined whether the amplitude of the EEG signatures scaled with Sample Interval. The amplitude of the CNV and CPP at Cue Onset was measured as the average amplitude in a 100 ms window centred at Cue Onset (-50 ms to +50 ms from Cue Onset). To account for the temporal smearing resulting from the STFT, we measured Cue Onset Mu/Beta at -100 ms before Cue Onset. We assessed the effect of Sample Interval on these measures using the same LMM structure as above, with the inclusion of Gap as a fixed effect. Pre-response, we examined whether the build-up rate and amplitude of the signals varied with Sample Interval. The pre-response build-up rate for each of the EEG signatures was quantified as the slope of a straight line fitted to activity from -600 ms to -300 ms prior to response on each trial. The pre-response amplitude of the CNV and CPP was measured as the average activity between -100 ms and -50 ms from response. The pre-response amplitude of Mu/Beta was measured as the average activity from -200 to -150 ms from response. We assessed the effect of Sample Interval on these measures using the same LMM structure as the Cue Onset analysis.

We also examined the link between these signal measurements and variations in Reproduction Time. Since Sample Interval and Reproduction Time were positively correlated, we median-split the data based on RT (‘Faster’ and ‘Slower’ Bins), with the splits performed within participant and Sample Interval. These models included random intercepts only, as the models would not converge if random slopes were included. As this introduces a risk of overestimating the strength of the fixed effect and underestimating the p-value, we also averaged across trials and performed one-way repeated-measures ANOVAs with RT Bin as a within-subjects factor.

As expected, participants adapted their Reproduction Time to the duration of the Sample Interval (Fig. 1b; x²(1) = 31.04, p < 0.001, b = 0.36), while also exhibiting an overall tendency to overestimate across the entire range of intervals, as well as a regression to the mean effect (Fig. 1b). The duration of the variable gap between the end of the measurement phase and the beginning of the reproduction phase had a small but significant effect on Reproduction Time (Fig. 1c: x²(1) = 4.99, p = 0.025, b = -0.02), with Reproduction Time decreasing as Gap increased (Difference in Group Mean RT between Longest and Shortest Bin in Fig. 1c = 7 ms). We also found that Reproduction Time was not influenced by the preceding sample interval (Fig. S1b; x²(1) = 1.34, p = 0.25, b = -6e^-5).

During Sample Interval presentation, we found no effect of Sample Interval or Reproduction Time on either the amplitude or slope of the CNV, Mu/Beta, or CPP measured at the end of the Sample Interval Itself (See left column of Fig. 2a, b and c; all p > 0.05, see S2 for full reporting of statistics). These and all following Mu/Beta signal effects were replicated when we separated the Mu/Beta signal into alpha mu (8–13 Hz) and beta (15–30 Hz) frequency bands (see S6). However, while alpha mu and beta were highly similar at Cue Onset and prior to Reproduction, they differed during Sample Interval Encoding, as alpha mu increased over time while beta decreased over time after an initial increase for 250 ms (S6). Although previous work has shown that the amplitude of offset potentials at fronto-central sites following the end of the sample interval is related to sample interval duration (Bueno & Cravo, 2021; Damsma et al., 2021; Kononowicz & Van Rijn, 2014; Kruijne et al., 2021; Ofir & Landau, 2022), we found no relationship between the amplitude of the P2 and either Sample Interval or RT Bin (S7).

At Cue Onset, CNV amplitude was larger (more negative) and there was greater Mu/Beta desynchronisation for longer Sample Intervals (Fig. 2a and 2b; CNV: x²(1) = 7.58, p = 0.006, b = -0.04; Mu/Beta: x²(1) = 4.21, p = 0.04, b = -0.02). While there was no significant effect of Gap duration on CNV amplitude (x²(1) = 2.8, p = 0.1, b = -0.01), Mu/Beta desynchronisation was greater for longer Gap durations (x²(1) = 17.81, p < 0.001, b = -0.02). Additionally, the amplitude of the CNV was larger and Mu/Beta desynchronisation was greater for Faster responses (Fig. 3a and 3b; CNV: x²(1) = 9.51, p = 0.002, b = -0.03; Mu/Beta: x²(1) = 4.75, p = 0.03, b = -0.01; also see S4). For these and all following RT Bin analyses, the RM ANOVA result supported the LMM result (see S5 for full reporting of the statistics). This suggests that the starting levels of Mu/Beta and the CNV are adjusted to modify Reproduction Time, with an elevated starting level (more negative CNV and greater Mu/Beta desynchronisation) leading to a faster response. Our analyses also indicated that there was no relationship between CPP amplitude at Cue Onset and either Sample Interval (Fig. 2c; x²(1) = 0.12, p = 0.73, b = -0.004) or RT Bin (Fig. 3c; x²(1) = 0.44, p = 0.5, b = 0.006).

Although the pre-response build-up rate of the CNV became shallower as the duration of the Sample Interval increased, this effect did not reach significance (Fig. 2a; x²(1) = 3.1, p = 0.08, b = 0.03). Similarly, the pre-response amplitude of the CNV did not significantly vary with Sample Interval (Fig. 2a; x²(1) = 0.04, p = 0.84, b = 0.002). However, there was a significant effect of RT Bin on CNV build-up rate, with a shallower build-up rate for Slower RTs (Fig. 3a; x²(1) = 43.51, p < 0.001, b = -0.06). Pre-response CNV amplitude was also significantly larger (more negative) for Faster trials (Fig. 3a; x²(1) = 7.67, p = 0.006, b = -0.02). There was no effect of Gap on CNV build-up rate (x²(1) = 2.25, p = 0.13, b = 0.01) or amplitude (x²(1) = 0.001, p = 0.97, b = 0.0001).

The effect of Sample Interval on pre-response Mu/Beta build-up rate did reach significance (Fig. 2b; x²(1) = 9.64, p = 0.002, b = 0.05), with Mu/Beta building more slowly for longer Sample Intervals. As with the CNV, the pre-response build-up rate of Mu/Beta was shallower for Slower reproduction times (Fig. 3b; x²(1) = 42.56, p < 0.001, b = -0.06). Consistent with the idea that Mu/Beta reaches a stereotyped response triggering threshold, pre-response Mu/Beta amplitude did not vary with Sample Interval (Fig. 2b; x²(1) = 1.5, p = 0.23, b = -0.008) or RT Bin (Fig. 3b; x²(1) = 0.03, p = 0.87, b = 0.0001). There was a significant effect of Gap on Mu/Beta build-up (x²(1) = 8.47, p = 0.004, b = 0.02), with a shallower build-up rate for longer gaps. Given that longer gaps are associated with a greater Mu/Beta amplitude at Cue Onset, participants may have slightly attenuated their pre-response build-up to avoid responding too early.

Although there was some positive build-up prior to response at central parietal electrodes (Figs. 2c and 3c), this build-up was attenuated relative to the typically observed levels of pre-response CPP build-up (Kelly & O’Connell, 2013; McGovern et al., 2018). There was no significant relationship between the slope of this pre-response build-up and Sample Interval (Fig. 2c; x²(1) = 0.36, p = 0.55, b = -0.005) or RT Bin (Fig. 3c; x²(1) = 0.11, p = 0.74, b = 0.003). These relationships remained non-significant when using a time window closer to the response of (-200 ms to -100 ms) which is more typically used for measuring CPP build-up rate (Sample Interval: x²(1) = 0.11, p = 0.75, b = 0.003; RT Bin: x²(1) = 1.91, p = 0.17, b = 0.01). Additionally, there was no significant relationship between pre-response CPP amplitude (-100 ms to -50 ms) and Sample Interval (Fig. 2c; x²(1) = 1.92, p = 0.17, b = -0.02) or RT Bin (Fig. 3c; x²(1) = 0.03, p = 0.85, b = 0.002). There was no effect of Gap on CPP build-up rate (x²(1) = 1.07, p = 0.3, b = -0.009) or amplitude (x²(1) = 1.16, p = 0.28, b = -0.009).

As the measurement window for pre-response slope began 600 ms prior to the response, some pre-cue activity was included in the slope measurement for trials with reproduction times shorter than 600 ms. Therefore, we repeated the above slope analyses for each of the EEG signatures excluding trials with reproduction times shorter than 600 ms. Removing these trials from the analyses did not change the above results qualitatively.

The aim of this study was to characterise the human EEG signatures underpinning time reproduction. Specifically, we sought to establish the extent to which the dynamics of the CNV, mu/beta band (8–30 Hz) motor preparation activity, and the CPP evidence accumulation signal predict interval reproduction. By using a task that was highly similar to a reproduction task previously used in an investigation of non-human primates, we were also able to determine the degree to which the observed EEG signal dynamics corresponded to those reported in the spiking activity of effector-selective areas that prepare decision-reporting actions. In terms of behaviour, we found that participants adapted their reproduction times to the presented durations. They also exhibited a central tendency effect that is typical of time reproduction. Our EEG analyses suggest that participants reproduced intervals by dynamically adjusting the starting-level and build-up rate of motor preparation so that it reached a fixed response-triggering threshold at the desired time. Specifically, mu/beta band oscillatory activity had starting levels that were shifted closer to threshold and faster build-up rates prior to shorter reproductions. The CNV exhibited a very similar pattern, with larger amplitudes at reproduction cue onset and faster build-up rates for shorter reproduction times. However, the CNV did not reach a fixed threshold at response, as its amplitude was larger for faster reproductions. Unlike the CNV and mu/beta, we found no relationship between the pre-response slope or pre-cue amplitude of the CPP and either interval duration or reproduction time, suggesting that the evidence accumulation processes indexed by the CPP are not involved in interval reproduction.

There is a striking resemblance between the dynamics of the mu/beta motor preparation activity observed here, and the macaque LIP activity identified by Jazayeri and Shadlen (2015). Both signals built towards a fixed threshold at response regardless of sample interval or reproduction time, indicative of a response triggering threshold, and the pre-response build-up rate of both signals varied with reproduction time, with a steeper slope for faster reproductions. Therefore, as suggested by Jazayeri and Shadlen (2015) for LIP, mu/beta in the lead up to response appears to represent a motor plan for reproducing time intervals. Jazayeri and Shadlen (2015) also found that LIP activity around the time of the reproduction cue was increased for longer sample intervals. Here, we found a similar effect, as pre-cue mu/beta motor preparation increased with sample interval. In the present study, there was a variable gap between sample interval presentation and interval reproduction which was not included in the Jazayeri and Shadlen (2015) version of the task. We found that motor preparation varied with the duration of this gap, with greater motor preparation for longer gaps. Interestingly, we found the opposite effect when we binned by reproduction time within sample intervals, as pre-cue mu/beta build-up was larger for faster reproductions. This suggests that while there is an increase in pre-cue mu/beta that scales with interval duration, faster reproduction of a given interval is, nevertheless, associated with greater motor preparation starting levels. This finding is reminiscent of the elevated pre-stimulus mu/beta motor preparation levels associated with faster responding in perceptual decision-making tasks (Kelly et al., 2021; Murphy et al., 2016; Steinemann et al., 2018).

In line with more recent electrophysiological studies of interval timing (Kononowicz & Van Rijn, 2011; Ng et al., 2011), the present results suggest that the CNV does not represent a veridical temporal accumulator that accumulates pulses at a fixed rate. Rather than reaching a larger (more negative) amplitude for slower reproduction times as would be predicted by temporal accumulator accounts, the CNV reached a larger amplitude for faster reproductions. The CNV exhibited greater similarities with mu/beta, with starting-levels and pre-response build-up rate that varied with reproduction time. Although the effect did not reach significance, CNV build-up rate also had a similar relationship with sample interval as mu/beta, as it became shallower for longer intervals. In line with recent work (Bueno & Cravo, 2021; Damsma et al., 2021; Kononowicz et al., 2015; Kononowicz & Van Rijn, 2014; Kruijne et al., 2021; Ofir & Landau, 2022), these findings suggest that rather than representing a veridical accumulator signal, the CNV more closely resembles a dynamic anticipatory signal like the urgency signals identified in perceptual decision-making research (Hanks et al., 2014; Murphy et al., 2016). Indeed, the pre-cue effect of a larger CNV amplitude for faster reproductions closely resembles the larger pre-evidence CNV amplitude observed under increased speed pressure in perceptual decision making that has been linked to reduced response caution (Boehm et al., 2014). In both cases, reduced time to respond is associated with a larger CNV prior to the response window, suggesting that the process reflected in the CNV plays a role in speeding up or slowing down responding. However, the CNV did not reach a stereotyped threshold at response, suggesting that it is not a build-to-threshold motor planning signal like mu/beta and LIP. Additionally, the CNV is implicated in perceptual timing tasks such as the bisection task, in which participants categorise time intervals without making a motor response (see Kononowicz & Penney, 2016; Van Rijn et al., 2011 for reviews). Therefore, it seems that the CNV is not solely linked to the planning and executing of motor responses, but is more closely aligned with the processes involved in cognitive deliberation.

Rather than building until the reproduction response was made, the CNV plateaued around 250 ms prior to the response and exhibited a return toward baseline around 100 ms prior to the response. Although it did not reverse its trajectory like the CNV, mu/beta also exhibited a deceleration in build-up rate prior to the response. Further work will be required to probe these dynamics and the functional distinctions between the CNV and mu/beta. One speculation that could be examined is that the amplitude of the CNV indexes an urgency process that specifies the rate at which motor preparation should increase.

Given that the CPP has been shown to index the accumulation of sensory evidence in both visual and auditory modalities across a range of perceptual decision-making tasks, we hypothesised that it may also represent a temporal accumulator signal. However, the present results suggest that the CPP does not index the accumulation of temporal pulses, as neither its slope nor amplitude was related to sample interval or reproduction time. One possibility is that the CPP is restricted to accumulating samples of sensory evidence when the decision maker has to choose between multiple alternatives. This possibility could be probed using a perceptual timing task in which participants have two possible responses to choose from (longer vs. shorter, same vs. different). Alternatively, the CPP may only accumulate samples of evidence when an external stimulus is presented, such as an auditory tone. In the absence of external sensory input, as in the reproduction phase of our task, the CPP may not accumulate an internal signal representing elapsed time. Future work could address this question by utilising the same task we used here, but with the addition of an auditory tone played while participants reproduced the interval.

Whereas Jazayeri and Shadlen (2015) found that LIP activity continued to build throughout the sample interval at a decelerating rate, here motor preparation and the CNV appeared to plateau midway through the sample interval. Additionally, we did not find a relationship between interval duration and the build-up rate or amplitude of any of the three signals at sample interval offset. A possible explanation for this apparent discrepancy lies in our adding of a variable gap between sample interval offset and reproduction onset. The fact that observers in Jazayeri and Shadlen (2015) could begin reproduction as soon as the sample interval ended meant that they could continually prepare and tailor their motor plan during the sample interval itself. Here, the variable gap would ensure that such a strategy would not be as beneficial and our data suggest that participants waited until the end of the sample interval to begin preparing their responses. The fact that mu/beta and the CNV did exhibit amplitude modulations immediately prior to cue onset that scaled with sample interval and reproduction time suggests that some anticipatory preparation did occur, but over a narrower interval.

Although we did not find that the amplitude of the P2 evoked by encoding offset was influenced by sample interval or reproduction time, previous studies have found a significant relationship between P2 amplitude and sample interval and perceived duration (Kononowicz & Van Rijn, 2014; Kruijne et al., 2021; Ofir and Landau, 2022), including a recent time reproduction study in which P2 amplitude increased with sample interval and decreased with reproduction time (Damsma et al., 2021). One of the limitations of the present study for examining the P2 is that the gap between encoding offset and cue onset was relatively short, making it more difficult to separate the P2 from CNV activity at cue onset. In the present study, the amplitude of the CNV was more negative for longer sample intervals and shorter reproductions, which is in the opposite direction to the P2 effects found by Damsma et al. (2021). As such, this introduces the possibility that any overlap between these signals would serve to counteract their relationships to sample interval and reproduction time. Therefore, future studies should include a longer gap between encoding and reproduction to facilitate analysis of the P2 and CNV (such as the 1.5 s gap used by Damsma et al., 2021).

In conclusion, our results extend previous work conducted in macaques to human participants, showing that humans reproduce time intervals by flexibly adjusting motor preparation activity to hit a response triggering threshold at the desired time. Furthermore, our findings support a growing body of literature indicating that the CNV does not represent a veridical temporal accumulator. Instead, the CNV bears a striking resemblance to motor preparation, suggesting that it plays a role in speeding up and slowing down responding. Our results also indicate that the evidence accumulation processes that govern perceptual decision making do not play a role in interval timing.

Preprocessed EEG data and analysis code are freely available online:https://osf.io/3vtg9/?view_only=300d5c1e3d7e4891a9ba19e209fe5ccd

H.M., J.S.B., and R.G.O. designed the study, analysed the data, and prepared the manuscript. H.M. collected the data.

This research was funded by a grant from the European Research Council (ERC): European Research Council Consolidator Grant IndDecision—865474.

No conflicts of interest.

Thank you to Pat McKeown and Daria Monakhovych for assisting with data collection. Thank you to all the participants who kindly participated in the study.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00279.