key: cord-0319722-15zsg17t
authors: Snipes, Sophia; Krugliakova, Elena; Meier, Elias; Huber, Reto
title: The theta paradox: 4-8 Hz EEG oscillations reflect both local sleep and cognitive control
date: 2022-04-07
journal: bioRxiv
DOI: 10.1101/2022.04.04.487061
sha: 729359cbfef42f1b800125d33bbbb3e2ff5b5518
doc_id: 319722
cord_uid: 15zsg17t

Human brain activity generates electroencephalographic (EEG) oscillations that characterize specific behavioral and vigilance states. The frequency of these oscillations is typically sufficient to distinguish a given state, however theta oscillations (4-8 Hz) have instead been found in near-opposite conditions of drowsiness during sleep deprivation and alert cognitive control. While the latter has been extensively studied and is often referred to as “frontal midline theta”, the former has been investigated far less but is considered to be a marker for local sleep during wake. In this study we investigated to what extent theta oscillations differed during cognitive tasks and sleep deprivation. We measured high-density EEG in 18 young healthy adults performing 6 tasks under 3 levels of sleep deprivation. We found both cognitive load and sleep deprivation increased theta power in medial prefrontal cortical areas, however sleep deprivation caused additional increases in theta in many other, predominantly frontal, areas. The sources of sleep deprivation theta were task-dependent, with a visual-spatial task and short-term memory task showing the most widespread effects. Notably, theta was highest in supplementary motor areas during passive music listening, and highest in the inferior temporal cortex during a spatial game. This suggests that theta caused by sleep deprivation may preferentially occur in cortical areas not involved in ongoing behavior. While our results find differences in topography from frontal midline theta, they raise the possibility that a common mechanism may underly both theta oscillations during cognition and during sleep deprivation.

Research of the healthy human brain is limited to what can be measured non-invasively. 23 Electroencephalography (EEG) offers the possibility of measuring synchronized 24 neural activity, i.e. oscillations. Since the discovery of EEG, specific oscillations have been 25 associated with behavioral states such as alertness, drowsiness, and different stages of 26 sleep. This allows oscillations to be used as objective markers for vigilance. While the exact 27 function of a given oscillation is not always understood, its predominant occurrence is limited 28 to a specific vigilance state. The exception are theta oscillations (4) (5) (6) (7) (8) , which have been 29 separately identified as an indicator of drowsiness and intense cognition.

Theta oscillations are associated with drowsiness in both animals (Vyazovskiy & Tobler, 2005) 1 and humans (Finelli et al., 2000) , most easily observed during sleep deprivation. More 2 specifically, theta activity reflects sleep pressure, i.e. the interaction between circadian Not only does theta track sleep pressure in time, but also in space. Sleep pressure appears 8 driven by daytime neural plasticity; brain areas that undergo high plasticity during the day will 9 have higher sleep need at the beginning of the night, reflected as a local increase in slow wave 10 activity (1-4 Hz oscillations) during deep sleep (Tononi & Cirelli, 2014) . This increased sleep 11 pressure is especially true for frontal association areas, where both theta and slow waves are 12 highest (Finelli et al., 2000) . Similarly, daytime activity can induce local differences in sleep 13 pressure in areas involved in learning, leading to corresponding local changes in slow wave 14 activity (Huber et al., 2004 (Huber et al., , 2006 . The tight relationship between daytime activity and local 15 changes in theta was convincingly illustrated in a study by Hung et al. (2013) , in which 16 participants spent a 24 h sleep deprivation period either playing a driving simulator or 17 listening to audiobooks. Theta during waking rest between tasks increased more over visual 18 (occipital/parietal) areas after playing the driving simulator, and more over auditory/speech 19 (left temporal) areas after listening to audiobooks. 20 Given the presence of theta oscillations when and where sleep pressure is highest, they have 25 found these off periods to also occur during sleep deprived awake rats, over localized cortical 26 areas. These off periods in wake corresponded to theta oscillations in the local field 27 potentials. While evidence for sleep slow wave off periods in humans has been found with 28 intracortical recordings in epileptic patients , the same has not been possible 29 for theta in wake due to technical limitations in single-cell recordings in humans (Nir et al., 2017) . It is therefore still uncertain if theta in humans corresponds to off periods in the cortex 1 and could thus be considered localized slow waves. 2 Equally robust research has separately linked theta activity to cognition and wake. Theta has 3 been associated with an unparalleled variety of functions (for a review see Buzsáki, 2005) , 4 most notably hippocampal theta during spatial navigation in rats (Buzsáki, 1996 ; O'Keefe & 5 Recce, 1993) and frontal-midline theta during cognitive tasks in humans. Frontal-midline 6 theta (fmTheta) is an oscillation that is visually identifiable in approximately 10-40% of the 7 population (Inanaga, 1998) , occurs in bursts around 1-10 seconds, with amplitudes around 8 30-60 μV (Mitchell et al., 2008) . It has been associated with arithmetic (Ishihara & Yoshii, 9 1967 ; Ishii et al., 2014) , working memory (Gevins et al., 1998; Jensen & Tesche, 2002) , and 10 even meditation (Banquet, 1973 ; D. J. Lee et al., 2018) . It has been shown to increase with 11 short-term / working memory load. fmTheta has been source localized to the anterior , where it has in turn been anti-correlated to fMRI BOLD (functional magnetic 14 resonance imaging, blood-oxygen level dependent) activity in these areas (Scheeringa et al., 15 2008, 2009). 16 The exact function of fmTheta oscillations in cognition is still unresolved although various 2001; Sauseng et al., 2010) . One of the most well-elaborated hypotheses is that theta is 20 responsible for synchronizing neuronal firing to specific phases of distant cortical regions 21 (Lisman & Jensen, 2013) . Such a role for theta has been supported by intracortical recordings independent task-related changes in theta (Brzezicka et al., 2018) . Nevertheless, given the 25 strong association between theta and tasks, it is generally considered to be functionally 26 relevant for cognitive processing. 27 Currently, research in theta oscillations caused by sleep deprivation (hereafter referred to as 28 sdTheta) has remained largely independent from research in cognition and fmTheta. sdTheta 29 has been studied almost exclusively during quiet resting states, whereas fmTheta is studied 30 during cognitive tasks under low sleep pressure conditions (i.e. normal wake). It is therefore 31 unknown if these represent either two distinct oscillations in the theta range or one and the 1 same, as has been suggested by Takahashi et al. (1997) and Mitchell et al. (2008) . If sdTheta 2 and fmTheta are qualitatively distinct, this would resolve the apparent paradox of an 3 oscillation reflecting both drowsiness and cognition. If sdTheta is instead a manifestation of 4 fmTheta, then its interpretation as local sleep should be reconsidered. Instead, fmTheta 5 during sleep deprivation would speak of some form of compensation mechanism 6 counteracting the detrimental effects of prolonged wakefulness. 7 We conducted a sleep deprivation study in young healthy adults to disentangle the changes 8 in theta related to both drowsiness and cognition using high-density EEG. Participants The tasks included: 14 • A short-term memory task (STM) to reproduce classic fmTheta memory load effects 15 • The psychomotor vigilance task (PVT) which involved responding to the start of a 16 countdown every 2-10 s with the push of a button 17 • The lateralized attention task (LAT), an adaptation of the PVT in which faint circles 18 would appear in half of the screen to which participants had to push a button in 19 response 20 • A speech fluency task (Speech) in which participants had to read out loud English 21 tongue twisters 22 • The Arkanoid-based game BBTAN (Game) which involved aiming a ball to bounce off 23 bricks to make them disappear before they covered the screen 24 • Passive music listening (Music) 25 To determine whether sdTheta and fmTheta could be considered the same oscillation, we 26 first looked at their topography within the STM task, and with source localization identified 27 their neural substrates. We then investigated how they interact by reproducing the classic 28 fmTheta memory load effect in the STM task at each session. Then, to determine more 29 generally how sdTheta is affected by behavioral state, we compared its topography and 30 source localization in all 6 tasks. In a final attempt to distinguish fmTheta from sdTheta, we 1 inspected the spectrograms for each task to determine if they could be differentiated by peak 2 frequency. Altogether, these results provide the first look of theta simultaneously under 3 different behavioral and vigilance states. 4 

Changes in sleep architecture and subjective sleepiness confirm the effectiveness of 7 the sleep deprivation protocol 8 To determine whether the sleep deprivation protocol was successful in increasing sleep 9 pressure, we evaluated changes in sleep architecture between the baseline night and 10 recovery night following sleep deprivation (Table 1) . We found shorter sleep onset latencies 11 and more deep sleep (NREM3), key indicators of increased sleep pressure. 12 All sleep stages except REM sleep showed a significant change between baseline and 13 recovery, with NREM3 increasing 30% at the expense of wake (-30%), NREM1 (-47%), and 14 NREM2 (-16%). Sleep onset latency (SOL) significantly decreased from 16.8 minutes to 5.6 15 minutes. Overall sleep duration was shorter during the recovery night, although this was not 16 statistically significant (p-value = .108), and sleep efficiency increased from 92% to 96%. 17 Together, these results indicate that sleep pressure, specifically for slow wave sleep, 18 increased over the 24h wake period. To determine the degree of sleep deprivation experienced by the participants, a brief 6 questionnaire was administered after each task at every session (all questionnaire results are 7 provided in Supplementary Figure 1) . Participants indicated on a continuous Karolinska 8 Sleepiness Scale (Åkerstedt & Gillberg, 1990 ) how alert or sleepy they felt ( Figure 1A) . A two- 9 way repeated measures analysis of variance (rmANOVA) was conducted with factors session 10 and task, as well as their interaction. There was a highly significant and extremely large effect Frontal-midline theta is more localized than sleep deprivation theta 26 For fmTheta and sdTheta to be considered the same oscillation, they should originate from 27 the same brain areas. To determine if this was the case, we analyzed changes in theta from 28 the short-term memory task (STM). Briefly, participants were presented with 1, 3, or 6 symbols for 2 s, which they then had to hold in memory during a 4 s retention period. Finally, 1 they were presented a probe symbol and had to indicate whether it was part of the original 2 set or not. Each session consisted of 120 randomized trials, 40 for each memory load level 3 (L1, L3, L6). 4 Theta activity was measured as power spectral density (PSD) from the first 2 s of the retention 5 period. To account for large interindividual differences (Suppl. Figure 2 ) as well as the 1/f 6 power amplitude distribution across frequencies, PSD values were z-scored for each 7 frequency, then the PSD for frequencies between 4 and 8 Hz was averaged. Power spectra for 8 the retention period across load levels and sessions is provided in Suppl. untransformed theta power to the source space, trials were averaged by load level for each 12 session and the data was z-scored across voxels, load, and sessions. Separate pipelines (Suppl. Figure   21 6). The right medial cortex also showed increases in theta however these areas did not survive 22 correction for multiple comparisons. 23 sdTheta was calculated comparing L1 trials from BL to L1 trials from SD ( Figure 2B ). sdTheta 24 was more widespread across the cortex than fmTheta, showing cluster-corrected increases in 25 38% of gray matter voxels relative to 21%, respectively. All areas showing load-effects of 26 fmTheta were also significant for sdTheta ( Figure 2C ), and the areas showing highest sdTheta 27 were not among those significantly increasing in fmTheta. Specifically, the peak location of 28 sdTheta was different in both the channel space (ch5) and source space: right middle frontal 29 gyrus (t = 6.95) and superior frontal gyrus (t = 5.94; 2B III). sdTheta extended along the medial cortex up to the cuneus (tmax 1 = 5.15) and was additionally present around the left insula (tmax 1 = 4.58), and the temporal poles (tmax = 3.67). Therefore, sdTheta and fmTheta have different 2 primary sources, and different spread throughout the cortex. (4) (5) (6) (7) (8) was only significant for fmTheta. Exact t-values can be seen in Figure 6 . Acronyms: BL (baseline), SD (sleep 18 deprivation), FDR (false discovery rate), STM (short term memory task), AAL (Automated Anatomical Labelling). 19 20 1 When tmax is specified, then the highest t-value from neighbouring or group of channels/areas is used Frontal-midline theta fades with increasing sleep deprivation 1 If sdTheta and fmTheta are independent oscillations, they should both be present during 2 sleep deprivation when performing the STM task. fmTheta was therefore calculated at every 3 session, for both L3 vs L1 and L6 vs L1 ( Figure 3A) . Surprisingly, fmTheta decreased in 4 amplitude with increasing sleep deprivation, until no channel showed statistically significant 5 differences with memory load during SD. 6 A two-way rmANOVA was conducted with factors session, load, and their interaction, The interaction between load and session was driven by a larger increase in theta for low 4 memory load trials during sleep deprivation ( Figure 3B ). To better understand this, we 5 compared sdTheta topographies (BL vs SD) for each memory load level ( Figure 3C ). L1 showed 6 the largest and most widespread increase in theta (tmax = 7.28, p < .001, g = 1.57), L3 the Sources of sdTheta are task dependent 12 The results from Figure 2 show distinct topographies for fmTheta and sdTheta. The literature 13 has identified fmTheta to consistently originate from the same medial region, however similar 14 source localization has never been done for sdTheta. To determine whether the location of 15 sdTheta is consistent or task-dependent we compared theta changes from BL in 6 different 16 tasks, using the first 4 min of EEG data from each task. Figure 4 depicts the sdTheta changes 17 for both SR and SD relative to BL in the channel space, and Figure 5 provides the source 18 localization for SD relative to BL. Figure 6 provides the t-values for all regions found to be 19 significant in at least one comparison of SD relative to BL. Mean theta values for all tasks in 20 regions of interest are provided in the supplementary material (Suppl. Figure 4) , as well as 21 statistical comparisons between tasks (Suppl. Table 2 ). 22 All tasks showed increases in theta between BL and SR, however no channel was significant 23 for the Speech and Music conditions after FDR correction. The highest overall increase was 24 seen for the LAT over ch109 (tmax = 5.74, p = 0.002, g = 1.33), accompanied by widespread 25 increases. Due to the otherwise medium-low effect sizes, the comparison between BL and SR 26 was not further investigated with source localization. For BL to SD, already from the channel space it is evident that the location and spread of 1 sdTheta is task dependent. The LAT, STM, and PVT showed the most widespread increases, 2 as well as the highest amplitude (PVT: tmax = 7.52, p < .001, g = 1.85; LAT: tmax = 7.10, p < .001, 3 g = 1.24; STM: tmax = 6.31, p = .001, g = 1.80). The Speech task showed the lowest and most 4 local increase in theta (tmax = 5.50, p = .005, g = 1.51). 5 The source space allowed further anatomical localization of the origin of theta activity. All 6 tasks showed a predominantly right, frontal increase in theta, although no anatomical area 7 survived FDR correction for the Speech task. One of the primary sources of sdTheta across all 8 tasks was the right superior frontal gyrus. All tasks (except Speech) also had significant theta 9 originating from the right hippocampus, parahippocampus, anterior and middle cingulate 10 cortex. The STM and LAT had further extensive increases across both dorsal and medial frontal 11 areas, with the STM showing high theta activity along the left lateral sulcus (Rolandic 12 operculum, insula), and the LAT in the right lateral sulcus (Heschl's gyrus, Rolandic operculum, 13 insula). Unfortunately, source localization along this sulcus is challenging due to how gray 14 matter is folded and would require subject-specific MRI structural scans for accurate results. 15 The overall strongest source of sdTheta was the left supplementary motor area during the 16 Music task (t = 6.54), extending contralaterally as well as into the middle cingulate cortex. 17 Bilateral supplementary motor areas were also the main sources of theta for the PVT (tmax = Finally, the most atypical distribution of sdTheta came from the Game, which showed minimal 22 increases in frontal cortices and primary sdTheta originating from the right inferior temporal 23 cortex (inferior temporal gyrus, mid temporal gyrus, fusiform gyrus; tmax = 5.65). The only 24 other task to show significant sdTheta in these regions, to a lesser extent, was the LAT (tmax = 25 2.99). 26 Overall, the majority of sdTheta occurred in medial and superior frontal cortices, with a right 27 lateralization. LAT and STM were the most widespread in the source space (39% and 35% of 28 significant voxels, respectively), the Game, Music, and PVT intermediate (28%, 27%, 25%), and 29 Speech the least (9%). While most sdTheta sources were frontal, there were substantial differences between tasks. The high theta from the supplementary motor area in the Music 1 task and in the inferior temporal cortex in the Game suggests a preference of sdTheta for 2 cortical areas not critical for the ongoing behavioral task. sdTheta occurs in addition to task-related fmTheta found at BL. In order to determine whether 8 sdTheta could be further distinguished from this baseline fmTheta, we inspected the 9 spectrograms of the different tasks for all participants. In particular, we were interested in 10 whether tasks with high frontal BL theta showed an additional distinct peak in the theta range 11 following sleep deprivation. This would support the hypothesis of theta during sleep 12 deprivation as a separate oscillation from task-related, baseline fmTheta. 13 While statistics at each frequency confirmed that the effect of sleep deprivation was specific 14 to the theta range (Suppl. Figure 5) , sdTheta often did not occupy a single consistent peak 15 within or across individuals (Figure 7) . Instead, individuals' peaks were spread over the entire 16 theta range, often with multiple smaller peaks within the same participant. Furthermore, the 17 peak frequency for a given participant was not consistent across tasks (Suppl. Figure 6C-D) . 18 The exception was the Game, which showed the overall highest amplitude frontal theta as 19 well as the most clearly defined peak both during BL and SD, with prominence values 2 (MEAN 20 ± STD) of 1.72 ± 1.14 and 2.94 ± 1.88 respectively. By contrast, the STM task had a prominence 21 of 0.33 ± 0.27 at BL, and 0.71 ± 0.94 at SD (Suppl. Figure 6A-B) . In general, the STM had low 22 BL frontal theta, along with Speech and Music (Suppl. Figure 4D ). 23 Due to the clear presence of fmTheta at BL in the Game, we considered this task to be the 24 most likely to show both an fmTheta peak and an sdTheta peak during SD. The BL peak 25 frequency was significantly different from the SD peak frequency, increasing from 5.7 ± 1.0 26 Hz to 6.4 ± 0.5 Hz (t = 2.62, p = .018, g = 0.84). For reference, the STM peak was 6.0 ± 1.4 Hz 27 at BL, and 6.4 ± 0.7 Hz at SD, but the increase was not statistically significant (t = 0.87, p = 28 2 Difference in z-scores between the maximum theta amplitude and the closest trough in the spectrum .397, g = 0.30). However, as can be seen in the individual Game spectrums in Figure 7 , only a 1 single peak is present for most participants, with the baseline theta peak merely shifted in 2 frequency and increased in amplitude during SD. Multiple peaks were instead found in all 3 other tasks during SD, which may indicate a multitude of different theta oscillations not found 4 in the Game. Visual inspection of the EEG data provided further insight into task-related theta differences. 12 At BL, fmTheta bursts as described by Mitchell et al. (2008) were visible primarily in the Game 13 task ( Figure 8A ) in 11 individuals. These were frontal-midline bursts that lasted 1-5 s with 14 amplitudes around 15-20 μV. No other prominent theta oscillations were detectable by visual 15 inspection in any task (best example, Figure 8C ). During SD, fmTheta became even more 16 prominent in the Game EEG ( Figure 8B) , with higher amplitudes and longer bursts, appearing 17 for 13 participants and increasing in other tasks as well. In addition to fmTheta, widespread 18 bursts often with frontal peaks appeared during sleep deprivation especially in the LAT and STM ( Figure 8D ). These had a much shorter duration (2-3 oscillations), but with a higher peak 1 amplitude (> 40 μV). As can be seen from the spectrums (Figure 8 II) , game theta bursts 2 yielded narrow-band theta, whereas the LAT bursts had more widespread spectrums. These 3 examples support the interpretation of at least 2 types of oscillations in the theta range that 4 increase with sleep deprivation. be a compensation mechanism to counteract the detrimental effect of sleep deprivation, 24 similar to the increase in fmTheta with memory load and cognitive demand. These results 25 could mean that sleep deprivation in humans induces two types of changes in theta: an 26 increase in fmTheta when already present at baseline, and the appearance of local sleep. 27 An alternative, simpler explanation is that theta may reflect the same mechanism during both 28 cognition and sleep deprivation, regardless of waveform. Simultaneous EEG-fMRI studies 29 previously found that fmTheta originating from the medial prefrontal cortex corresponds to BOLD deactivations in these areas, both during passive rest (Scheeringa et al., 2008 ) and 1 increasing short term memory load (Scheeringa et al., 2009 ). Our source localization of 2 sdTheta across the different tasks also suggests that these oscillations may be a marker for 3 cortical areas not in use. 4 First, we found high sdTheta activity in the bilateral (but especially left) supplementary motor 5 area in the Music listening condition. It is compelling that the one task not requiring 6 movement showed such strong theta activity in brain areas involved in complex motor 7 planning (Goldberg, 1985) . Notably, the PVT also showed strong activity in bilateral 8 supplementary motor areas, which may seem contradictory. However, the PVT required 9 simply pushing a button after a very obvious stimulus appeared; this made for an almost 10 reflexive response, with little need for deliberative action. By contrast the LAT, which had 11 identical motor requirements but difficult to detect stimuli, despite otherwise widespread 12 high amplitude theta, showed less activity in the supplementary motor areas than the PVT 13 ( Figure 6 ). Supporting this distinction between reflexive and deliberative action, mean 14 reaction times of the LAT were ~20% slower than during the PVT (Suppl. Figure 7B -C), despite 15 identical task requirements (respond within 0.5 s). Vice versa, supplementary motor area 16 activity did not significantly increase in the Speech or Game, two tasks characterized by 17 deliberative action. 18 Second, high sdTheta was found in the right inferior temporal cortex in the Game, extending 19 all the way to the fusiform gyrus. These areas collectively form the ventral visual pathway 20 responsible for object recognition (Ishai et al., 1999) . This is in opposition to the dorsal visual 21 pathway running from the occipital cortex to dorsal parietal areas such as the supramarginal 22 gyrus and parietal sulcus, where object location is processed (Freud et al., 2016) . The Game 23 was almost exclusively a spatial task, requiring participants to map out a target path for a 24 bouncing ball. The only other task to show significant activity in the inferior temporal cortex 25 was the LAT, a spatial attention task. Instead the STM, in essence an object recognition task, 26 showed no significant increase in these areas. 27 One possible interpretation for theta in cortical areas not in use is that it has a role in active 28 inhibition. Such a hypothesis has already been proposed for theta during cognition. Buzsáki 29 in 1996 suggested that theta oscillations in the hippocampus could act as a low-energy 30 solution to selective inhibition (Buzsáki, 1996; Thompson & Best, 1989) , such that only neurons synchronized to fire at the correct phase of an ongoing oscillation would successfully 1 transmit action potentials. The role of theta phases in inhibition was supported by phase-2 targeted closed loop stimulation in mice (Siegle & Wilson, 2014) . It may therefore be the case 3 that fmTheta and sdTheta in humans also reflect a low-energy active inhibitory state that 4 conflicting brain networks enter to compensate for cognitive load and sleep deprivation, 5 respectively. 6 Alternatively, theta could reflect passive cortical disengagement. In this scenario, an entire 7 network or brain area ceases to receive inputs, and essentially goes in "standby". This is 8 comparable to alpha oscillations in visual areas during eyes closed (Kirschfeld, 2005) , and may 9 even be related to the bistable default state in which cortical networks enter when physically 10 disconnected from the rest of the brain or during anesthesia (Sanchez-Vives et al., 2017). An 11 interpretation of theta as disengagement, more so than inhibition, would also explain theta 12 activity occasionally found in NREM1 (Santamaria & Chiappa, 1987) , at the transition between 13 wake and sleep. In essence, theta as inhibition would be a compensation mechanism for sleep 14 deprivation, whereas theta as disengagement would be a consequence of sleep deprivation, 15 bringing the brain closer to true sleep. 16 In our results, theta in the supplementary motor areas during Music listening and the inferior 17 temporal cortex in the Game could be interpreted as either active inhibition or passive 18 disengagement. However, the widespread theta increases in the low memory load trials of 19 the STM task relative to the local theta increases in the medium memory load trials ( Figure   20 3C) is more compatible with disengagement during less demanding conditions. Likewise, 21 theta was more widespread in the LAT and STM, subjectively considered more boring tasks 22 (Suppl. Figure 1C) , compared to the Speech and Game tasks. However, different experimental 23 methodologies (especially intracortical data) are needed to confirm this link with local 24 neuronal inactivity, and to determine the distinction between active inhibition or passive 25 disengagement. 26 If fmTheta and sdTheta reflect the same neural process, even cortical disengagement, it 27 becomes more difficult to consider theta as a marker for local "sleep". The fact that theta is 28 reliably present during rested conditions, especially during complex tasks, speaks against this. 29 The fact that the amplitude of theta power is dependent on local sleep pressure and neural of memory load effects of fmTheta, likely due to the higher increase in sdTheta in easier trials. 8 Vice versa, the condition with the overall highest frontal theta was the Game during sleep 9 deprivation, which is also when participants felt, out of all tasks, most awake ( Figure 1) . As 10 such, theta cannot be used in absolute terms as an objective marker for drowsiness. 11 Moving forward, research in the EEG of cognition and sleep pressure should not continue to 12 operate blind to each other. Experiments investigating cognitive changes in theta should aim 13 for constant, low levels of sleep pressure. An experiment conducted in the morning may yield 14 different results from an experiment conducted in the evening, or patients who have 15 comorbid sleep disturbances may not show the same theta effects as healthy controls, even 16 with intact cognition. Vice versa, experiments trying to quantify sleep deprivation need to 17 avoid participants arbitrarily increasing their theta by engaging in meditation-like activity, or 18 other forms of focused attention. In general for such studies, it may be more appropriate to 19 record sdTheta during a task like the PVT, which is often used in sleep deprivation studies and 20 simultaneously provides behavioral data. Given the already high variability in theta power 21 across individuals (Suppl. Figure 2) , it would be best to reduce variability in mental state. 22 Similarly, we would recommend the use of an "addictive" game to study fmTheta when 23 possible, as this is both enjoyable for the participants and produces remarkably reliable and 24 robust oscillations, more so than the standard short-term memory task. 25 While our study offers unique insight into theta under different conditions within the same 26 participants, it also suffers limitations. First and foremost, because the participants are from 27 such a narrow age group, these results cannot be generalized to other populations. Theta 28 both during wake and sleep is known to undergo substantial changes across ages, with some 29 theta oscillations disappearing in adolescence and others re-appearing in old age (Ebersole & 30 Pedley, 2003). Additionally, while we mainly interpret these sdTheta results as originating from time spent awake based on previous literature (Cajochen et al., 2002; Finelli et al., 2000) , 1 it is likely that they are also affected by circadian fluctuations. Figure 4 illustrates how the 2 frontal spot in particular is only prominent in the SD condition and not SR; this may be due to 3 either the "extreme" sleep pressure or additional circadian components. Furthermore, given 4 the lack of structural MRIs and digitization of electrode positions, the source localization 5 results need to be taken with some caution. It is possible that deeper brain sources were 6 behind these diffuse theta increases, or some other bias we did not account for. Simultaneous 7 EEG-fMRI experiments could possibly resolve these issues. Furthermore, by looking at theta 8 power averaged over several minutes, we do not know if the oscillations appeared 9 synchronously in this network of areas, or in isolation. It is therefore imperative to verify and 10 expand these results with other recording methods, analyses, and more targeted tasks.

In conclusion, we do not provide a definitive resolution to the theta paradox but suggest three 12 possible explanations for our results: 1) fmTheta and sdTheta are separate oscillations, but 13 both can occur during sleep deprivation, one as a compensation mechanism, the other as 14 local sleep; 2) sdTheta is merely a more widespread form of fmTheta, and both reflect active 15 cortical inhibition of task-irrelevant networks; 3) both reflect passive cortical disengagement. 16 EEG offers a limited view of the brain but makes up for it by being easy to use and readily 17 applied in clinical populations and children of any age. As such, it becomes essential to 18 understand as much as possible of the EEG signals we can actually measure. Since its 19 inception, clinicians have been able to visually recognize peculiar oscillations, but such an 20 approach does not lend itself to quantification. Vice versa, analyses such as ours in the 21 spectral domain allow rapid quantification, but at the cost of oversimplification. The results 22 of this study are a reminder that there is still much more to be learned from even basic 23 oscillations like theta. Detailed screening criteria are provided in the supplementary material. Briefly, participants 6 had to be between 18-25 years old, completely healthy, neurotypical, good sleepers, and at 7 least somewhat vulnerable to sleep deprivation. Due to scheduling restraints caused by the 8 COVID-19 pandemic, some leniency was allowed for edge cases (e.g. one participant was 26 9 at the time of recording). 10 75 applicants conducted the screening questionnaire. One was recruited for technical pilots 11 (data not included), and an additional 31 passed but did not initiate contact or were unable 12 to meet the scheduling requirements. 19 participants were recruited, however one dropped 13 out midway and so was not included in further analyses. Of the 18 participants who 14 completed the experiment, 9 were female and 3 were left-handed. Mean age (± standard 15 deviation) was 23 ± 1 years old. All participants self-reported above-average English fluency 16 (68% ± 13% on a scale from terrible to native speaker), with 1 participant a native English 17 speaker. All had corrected-to-normal vision and self-reported no hearing impairments. 18 Data collection and interaction with participants was conducted according to Swiss law 19 (Ordinance on Human Research with the Exception of Clinical Trials) and the principles of the 20 Declaration of Helsinki, with Zurich cantonal ethics approval BASEC-Nr. 2019-01193. All 21 participants signed informed consent prior to participation and were made aware that they 22 could terminate the experiment at any time. 23 24 Experiment design 25 Participants came to the laboratory twice, first for the baseline then the sleep deprivation 26 recordings, separated by at least 4 days. During the week prior to each session, participants 27 were asked to maintain a regular sleep wake cycle, going to bed and waking up within 1h of 28 a pre-determined sleep and wakeup time based on their personal preference. These 29 individualized sleep and wake times were then used during the experiment. During the control week, participants wore a wrist accelerometer (GENEActiv, Activinsights Ltd.) and 1 filled out regular sleep reports to ensure compliance. Participants were further asked to 2 abstain from alcohol in the 3 days prior to the measurement, and limit caffeine consumption 3 to no more than the equivalent of 2 cups of coffee, and never after 16:00. They were asked 4 to avoid time-zone travel and any activities they knew could affect their sleep (e.g. parties, 5 skiing, sauna). 6 Data was collected in Zurich, Switzerland, between February and December 2020, wake onset and within 2.6 ± 10.5 min of the prior night's bedtime. After 23.6 ± 0.5 h of wake, 29 participants went to bed and slept for as long as they wished. During all wake recordings, 30 participants were monitored by an experimenter to ensure they did not fall asleep. From the evening before the first night to the day after the recovery night, participants remained in the 1 sleep laboratory and did not have access to clocks or external time cues. Two participants 2 reported nausea with increasing sleep deprivation and were therefore provided a break 3 outside just prior to the SD block (in complete nocturnal darkness). Each task block lasted approximately 2 hours. The order of tasks was randomized and 14 counterbalanced across participants. For each participant, tasks were conducted in the same 15 order for all three blocks. Each task began and ended with a 1 min rest period allowing 16 participants to adjust and get comfortable. After each task, participants answered a task 17 battery questionnaire asking how they experienced the task (Suppl. Figure 1 ). Participants were presented with a red fixation rectangle on a gray background ( Figure 10B ). 13 Every 2-10 s, the rectangle was replaced with a millisecond countdown and participants had 14 to press a button as fast as possible to stop it. The response time would then freeze for 1 s 15 and be colored in yellow if less than 0.1 s (false alarm), green if between 0.1 and 0.5 s (correct 16 response), and red if later than 0.5 s (lapse). If participants did not respond within 5 seconds, 17 an alarm would sound to wake them up. periods, 50 ms pink noise tones were presented every 1.5-5 s at ~50dB. Participants were 27 instructed to ignore these tones. 28 Speech Fluency Task: Participants performed a tongue-twister reading task in English for 10 min. This consisted of 20 trials, one for each sentence. Each trial began with the sentence 30 written on the screen ( Figure 10D ). Participants were instructed to read it in their head once or twice to get familiar with it, but not practice speaking. When they were ready, they could 1 press a button, and a green bar would appear below, steadily shrinking to count down a 10 s 2 reading window. In this time, participants had to read out loud the sentence as many times 3 as possible, as clearly as possible, and as correctly as possible. Arkanoid by Taito) for 10 minutes ( Figure 10E ). They started each session from level 1. The 6 game involved a robot with a ball at the bottom of the screen, and a row of 1-6 bricks at the 7 top. By tapping and dragging on the screen, participants could orient an arrow from the robot, 8 and the ball would be launched from the robot in the indicated direction. The goal was to 9 bounce the ball against the walls and hit as many bricks as possible, such that every time the 10 ball hit a brick, the brick lost a point, and when the brick had no more points, it disappeared. 11 At each round, after the ball was launched, hit the bricks, and bounced back to the bottom, 12 the remaining set of bricks descended by 1 row, and a new row of bricks appeared at the top. 13 When the bottom-most row of bricks reached the robot, the player lost the game. There were 14 additional game features to help remove bricks faster. This was a "simple but addictive" 15 game, requiring a minimum amount of spatial strategy to win, without any time pressure. High-density EEG was recorded using HydroCel Geodesic Sensor Nets™ with 128 channels, 5 connected to DC BrainAmp Amplifiers and recording software Brainvision Recorder (Vers. 6 1.23.0003, Brain Products GmbH, Gilching, Germany). Data was recorded with a sampling rate 7 of 1000 Hz with Cz reference. 8 All data preprocessing, analysis, and statistics was done with custom scripts in MATLAB 1 (R2019b) based on the EEGLAB toolbox v2019.1 (Delorme & Makeig, 2004) . All further 2 analyses involving source localization were performed with the FieldTrip toolbox v20210606 3 (Oostenveld et al., 2011) . Detailed analysis pipelines are provided in the supplementary 4 material, and code is available on GitHub (https://github.com/snipeso/Theta-SD-vs-WM).

Preprocessing: EEG data was filtered between 0.5-40 Hz and downsampled to 250 Hz. Visual 6 detection of major artifacts and bad channels was conducted by author SS, blind to 7 participant, task, and session. ICA was then used to remove physiological artifacts, mainly eye 8 movements, heartbeat, and muscle activity (Dimigen, 2020) . Bad channels were interpolated, 9 and only the first 4 minutes of clean data were used. The full pipeline is depicted in Suppl. Channel space power calculation: 120 channels were used. Power source density (PSD) was 12 calculated using MATLAB's pwelch function, with 8 s windows, Hanning tapered, and 75% 13 overlap. Data for each frequency was z-scored. For theta topographies (e.g. Figure 4 ), z-scored 14 PSD values between 4-8 Hz were averaged. For power spectrums (e.g. Figure 7 ), z-scored PSD 15 values were averaged into 3 pre-selected regions of interest (ROIs). Exact channels are 16 available in the Supplementary Material. For mean theta values (e.g. Figure 3B ), these ROI 17 spectrum averages were further averaged between 4-8 Hz. The channel space analysis 18 pipeline is depicted in Suppl. Figure 11 . standard MRI template brain. A 3D grid with 10 mm resolution (3294 voxels) was used as a 23 source model. After being projected into the source space, power was z-scored for each 24 frequency. For visualization, t-tests were conducted for all gray-matter voxels, cluster 25 corrected for multiple comparisons, and significant clusters projected onto the inflated brain. 26 To determine the main anatomical sources, z-scored data was parcellated based on the of all voxels within each area was then averaged across frequencies. For both pipelines, only 29 cortical areas were included, as there is currently little evidence that activity from deep brain 30 structures reaches the scalp. Further details can be found in Supplementary material: Methods: Source localization, with the source space analysis pipeline depicted in Suppl. Figure   1 12.

Trial analysis: Power calculations in both channel and source space for the 2 s retention 3 windows of the STM task were done as described above, except using artifact-free non-4 overlapping trials from the entire 25-minute recording. Trials were first averaged within each 5 session by load level, and then z-scored pooling levels, sessions, and channels for each 6 participant and each frequency. The minimum number of trials for each level for each session 7 was 25. 

Subjective and objective sleepiness in the active individual

Active Forgetting: Adaptation of Memory by Prefrontal Control

Spectral analysis of the EEG in meditation

Maximizing sensitivity of the Psychomotor Vigilance Test (PVT) to sleep loss

Event-induced theta responses as a window on the dynamics of memory

Controlling the False Discovery Rate: A Practical and Powerful Approach to 16 Multiple Testing

Local Patterns of Sleep and Wakefulness

A two process model of sleep regulation

Working memory 22 load-related theta power decreases in dorsolateral prefrontal cortex predict individual differences in 23 performance

The hippocampo-neocortical dialogue

Theta rhythm of navigation: Link between path integration and landmark navigation, 28 episodic and semantic memory

Separation of circadian and wake duration-30 dependent modulation of EEG activation during wakefulness

EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics 33 including independent component analysis

Optimizing the ICA-based removal of ocular EEG artifacts from free viewing experiments

Current practice of clinical electroencephalography

Dual electroencephalogram markers of human 40 sleep homeostasis: Correlation between theta activity in waking and slow-wave activity in sleep

What' Is Happening in the Dorsal Visual Pathway

Monitoring working 1 memory load during computer-based tasks with EEG pattern recognition methods

Supplementary motor area structure and function: Review and hypotheses

EEG spectral 6 power and sleepiness during 24 h of sustained wakefulness in patients with obstructive sleep apnea 7 syndrome

Dynamic imaging of 9 coherent sources: Studying neural interactions in the human brain

An event-related 12 fMRI study of the neurobehavioral impact of sleep deprivation on performance of a delayed-match-to-13 sample task

Computation of measures of effect size for neuroscience data sets

Frontal midline theta oscillations during working memory maintenance and 17 episodic encoding and retrieval

Arm 20 immobilization causes cortical plastic changes and locally decreases sleep slow wave activity

Local sleep and learning

Local experience-25 dependent changes in the wake EEG after prolonged wakefulness

Frontal midline theta rhythm and mental activity

Distributed representation of objects 30 in the human ventral visual pathway

The interaction between paroxysmal EEG activities and continuous addition work 33 of Uchida-Kraepelin psychodiagnostic test

Frontal midline theta rhythm and gamma power changes during 36 focused attention on mental calculation: An MEG beamformer analysis

Frontal theta activity in humans increases with memory load in a working 39 memory task

The physical basis of alpha waves in the electroencephalogram and the origin of the "berger 42 effect

The functional significance of theta and upper alpha oscillations

Review of the neural oscillations underlying meditation

Phase locking of single neuron activity to theta 2 oscillations during working memory in monkey extrastriate visual cortex

Theta coupling between V4 and prefrontal cortex 5 predicts visual short-term memory performance

The Theta-Gamma Neural Code

Frontal Midline Theta Reflects 10 Individual Task Performance in a Working Memory Task

Simultaneous 13 EEG-fMRI during a working memory task: Modulations in low and high frequency bands

Frontal-midline theta from the perspective of 16 hippocampal "theta

Selective neuronal 19 lapses precede human cognitive lapses following sleep deprivation

Regional Slow Waves 22 and Spindles in Human Sleep

Phase relationship between hippocampal place units and the EEG theta 24 rhythm

Frontal midline EEG dynamics during working memory

FieldTrip: Open source software for advanced 28 analysis of MEG, EEG, and invasive electrophysiological data. Computational Intelligence and 29 Neuroscience

Gating of human theta oscillations by a working memory task

Shaping the Default Activity Pattern of the Cortical 34 Network

The EEG of drowsiness in normal adults

Control mechanisms in working memory: A 38 possible function of EEG theta oscillations

Frontal 41 theta EEG activity correlates negatively with the default mode network in resting state

Local aspects of sleep and wakefulness

Enhancement of encoding and retrieval functions through theta phase-4 specific manipulation of hippocampus

Natural waking and sleep states: A view from inside neocortical 6 neurons

Subjective sleepiness correlates 8 negatively with global alpha (8-12 Hz) and positively with central frontal theta (4-8 Hz) frequencies in the 9 human resting awake electroencephalogram

Frontal Midline Theta Rhythm in Young Healthy 12 Adults

Place cells and silent cells in the hippocampus of freely-behaving rats

Sleep and the Price of Plasticity: From Synaptic and Cellular Homeostasis to 16 Memory Consolidation and Integration

Automated anatomical labeling of activations in SPM using a macroscopic anatomical 20 parcellation of the MNI MRI single-subject brain

The FieldTrip-SimBio pipeline for 23 EEG forward solutions

Local sleep in awake rats

Theta activity in the waking EEG is a marker of sleep propensity in the rat

A 30 unified view on beamformers for M/EEG source reconstruction