To study the neural underpinning of sleep deprivation in the human brain, we investigated the selective responses of individual neurons and how such responses change upon cognitive lapses. Twelve individuals with pharmacologically intractable epilepsy, who were undergoing depth electrode monitoring to identify seizure foci for potential neurosurgical treatment, performed a face/nonface categorization variant of the PVT in 31 experimental sessions (Fig. 1a and Supplementary Table 1). Each session included two 12-min blocks in which six images of famous people, familiar landmarks, and animals were presented (24 trials for each image) for 200 ms with long, unpredictable interstimulus intervals (2–8 s) as participants performed the face/nonface categorization task. In four individuals, pairs of PVT sessions were conducted before and after full-night sleep deprivation that was carried out for clinical purposes (time spent awake after sleep deprivation = 24.1 ± 1.6 h; mean ± s.e.m.), thereby providing a unique opportunity to examine the effects of sleep deprivation on behavior and the underlying activity of individual neurons. We were also able to acquire data for two individuals from four PVT sessions conducted before and after normal sleep to address possible circadian and learning effects. Subjects performed the task successfully and accurately (percentage correct = 94.1 ± 1.9%, no response = 2.9 ± 1.1%; mean ± s.e.m. across 31 sessions). We focused only on correct responses in subsequent analyses to maximize the chances that changes in behavior or neuronal activity in some trials were driven by the internal state of the subjects rather than by an impoverished visual stimulus.

Figure 1: Sleep deprivation leads to cognitive lapses in a face/nonface categorization PVT. (a) Schematic illustration of the modified PVT in which images of people, landmarks, and animals were presented infrequently as participants performed a face/nonface categorization task. (b) Distribution of RTs before and after full-night sleep deprivation (SD) in two representative individuals. For each session, an ex-Gaussian fit (Online Methods) defines the right exponential tail of cognitive lapses (orange; highest RTs), and an equal number of trials with the fastest RTs (green) was used for subsequent comparison of neuronal data. (c) Left, mean 1/RT in four session pairs conducted before and after full-night sleep deprivation (black) and two session pairs conducted before and after normal sleep (red). Right, the τ parameter (exponential tail in the ex-Gaussian distribution defining cognitive lapses) before and after sleep deprivation (black; n = 4 paired sessions) and before and after normal sleep (red; n = 2 paired sessions). (d) Scatterplot showing significant correlation (Pearson's r = 0.39, P < 0.03) between τ (ordinate) and TSA (abscissa) across all sessions (n = 31, not only including those conducted before and after sleep deprivation). Full size image

The distribution of behavioral reaction times (RTs) during the PVT experiments was best fit by an ex-Gaussian function, representing a mixture of a normal distribution (standard RTs) and an exponential distribution for slow RTs ('right tail') with large variability between sessions and between participants in the predominance of the exponential component (Fig. 1b and Supplementary Fig. 1), as has been previously observed in healthy individuals21. In each experimental session, 'fast trials' (fastest RTs) and 'slow trials' (slowest RTs) were defined according to the fitted ex-Gaussian distribution (Online Methods), with slow trials comprising 16.9 ± 1.0% of correct responses (mean ± s.e.m., n = 31 sessions). We use the term 'cognitive lapses' throughout to refer to these slow trials characterized by delayed behavioral responses (rather than a complete absence of response), as is customary in the sleep deprivation PVT literature20,21,22,23,24. Other strategies for defining cognitive lapses, such as selecting the slowest 5–10% of trials, yielded similar results (data not shown). Figure 1b illustrates how sleep deprivation altered the distribution of RTs. The reciprocal of behavioral RTs (1/RT) is a sensitive marker of slower performance after sleep loss20 and was decreased after sleep deprivation (–21.3%; Fig. 1c). The parameter τ, denoting the exponential decay component of the ex-Gaussian function, increased from 122.9 ± 41.7 ms (mean ± s.d.) to 306 ± 237 ms (+128 ± 79%; Fig. 1c), whereas the parameter μ, denoting the mean of the normal distribution, was only modestly increased (+14.4 ± 8.1%, mean ± s.e.m.), indicating that sleep deprivation exerted its greatest effect on cognitive lapses, as previously reported23. We also observed a 'time-on-task' effect23, whereby the frequency of cognitive lapses increased with the time spent performing the task (Supplementary Fig. 2).

Given that sleep deprivation had a marked effect on cognitive lapses, we next examined whether the time spent awake (TSA) before each session could predict the measures of slow trial prevalence across the entire data set, such as elevated τ and elevated mean RT in cognitive lapses. ANOVA analysis using TSA and the number of times a participant had performed the task (training effect) as between-session factors revealed that only TSA significantly predicted slow trials (F = 4.3, P = 0.047; mean RTs for cognitive lapses across sessions, training effect and interaction, P > 0.05) and average RTs (F = 4.4, P = 0.046; other, P > 0.05). None of the variables analyzed could significantly predict mean RTs for fast trials across sessions. Time of day (circadian effect) did not correlate with RTs (fast or slow: all, P > 0.5); however, the lack of significant circadian effects here could stem from variability in the precise hours when sessions were conducted (Supplementary Table 1). In contrast, accuracy of performance in the face/nonface categorization task was not affected by TSA (F = 0.03, P = 0.86). Likewise, a relationship between TSA and slow trials was evident when conducting ANOVA on μ, σ, and τ (the parameters of the fitted ex-Gaussian distribution corresponding to the mean, left tail, and right tail, respectively): a significant effect was found for TSA on τ (F = 4.7, P = 0.04; Fig. 1d), but no effect of TSA was found on μ (F = 0.54, P = 0.47) or σ (F = 0.69, P = 0.41), verifying the specific relationship between TSA and cognitive lapses during slow trials. We did not find an effect of repeated sessions (training effect) on performance, in accordance with the literature23. In addition, subjective sleepiness (Online Methods) was significantly correlated with TSA (Spearman's r = 0.42, P < 0.05), but ANOVA did not reveal a significant relationship between subjective sleepiness and slow trials (F = 1.5 and 2.2 for mean RTs and τ of the slow trials, respectively), replicating results from previous studies on the limitation of subjective sleepiness estimates25. Altogether, behavioral data showed that TSA was the dominant factor influencing performance on the task and primarily increased the occurrence of cognitive lapses.

Next, we examined the neuronal activity evoked by the stimuli used in the face/nonface categorization PVT task. An image of the intracranial electrodes used and a schematic of the 104 brain regions monitored in the study are provided in Figure 2a,b. The visual stimuli used in the face/nonface categorization PVT paradigm elicited robust responses in individual neurons (Fig. 2c; 1,481 units recorded in total), especially in the MTL, but occasionally also in cingulate cortex, with variability in the precise intensity, selectivity, and latency of responses across individual neurons. Whenever possible, images were selected that elicited responses in prior screening sessions in order to maximize the likelihood of effectively driving activity in the recorded neurons. Of the 611 recorded neurons in the MTL, 106 (17%) responded significantly to at least one stimulus (Online Methods). When pooling the activity of all responsive neurons (n = 162) irrespective of brain region, an average response profile emerged consisting of increased firing rates at 200–500 ms after stimulus onset (Fig. 2d), with an orderly progression of temporal latencies from the high-order visual cortex to the hippocampus and frontal lobe (Fig. 2e). Robust differences in response latencies across MTL regions were also evident when quantifying the precise timing of the responses detected in each trial separately (Online Methods and Supplementary Figs. 3 and 4), in line with previous findings26,27. Single-unit spiking responses were highly selective and could not be observed when averaging the activity of neighboring neurons not categorized as responsive (Supplementary Fig. 5). Crucially, the robust and highly selective profiles of single-neuron responses allowed for study of the effects of cognitive lapses at the single-neuron level.

Figure 2: Human single-neuron responses during the face/nonface categorization PVT experiment. (a) Macro–micro depth electrodes with eight 40-μm (diameter) platinum/iridium microwires protruding 4–5 mm from the most distal macroelectrode contact. 6–12 of these electrodes were implanted into each subject to simultaneously monitor activity in multiple brain regions. (b) Overview of the 104 brain locations monitored by depth electrodes in 12 individuals, as seen from a medial view. OF, orbitofrontal cortex; AC, anterior cingulate cortex; SM, supplementary motor area; PH, parahippocampal gyrus; HC, hippocampus; E, entorhinal cortex; Am, amygdala; LH, left hemisphere; RH, right hemisphere. Red circles mark more lateral regions, such as superior temporal gyrus. (c) Four representative examples (raster plots and PSTHs) of single-unit spiking responses to different pictures (stimulus 1, 2, and 3) recorded from the anterior fusiform gyrus (top left), anterior hippocampus (bottom left), anterior cingulate cortex (top right), and parahippocampal gyrus (bottom right). Green boxes mark stimuli eliciting significant responses (red bars) above baseline firing (horizontal red lines), and insets show action potential waveforms. (d) Average response (raster plot and PSTH) across all neurons tagged as responsive (n = 162) to pictures that were effective in driving a response. (e) Average response to the same stimuli as in d for each brain region monitored showing an orderly progression of temporal latencies (black arrow; hot to cold colors) from high-order visual cortex to hippocampus and frontal lobe. FG, anterior fusiform gyrus; TPO, temporal–parietal–occipital junction. Full size image

We examined the relationship between cognitive lapses and underlying neuronal activity by testing how the responses of the same neurons to the same physical stimulus might change as a function of behavioral performance (comparing neuronal activity in fast trials versus slow trials in the same session). Importantly, such 'within-session' comparisons minimize confounding by increased epileptogenic activity after sleep deprivation. Given the relationship between cognitive lapses and TSA (Fig. 1d), we compared neuronal responses across all sessions (n = 31): 15.3% of these responses were obtained before and after sleep deprivation, and 71.4% were obtained when subjects were awake for >12 h. In individual neurons, cognitive lapses in slow trials were associated with weaker and delayed neuronal spiking discharges relative to responses in fast trials, with differences particularly evident around 200–300 ms following image onset (Fig. 3a). We proceeded to examine the average normalized response in fast versus slow trials across the entire data set. Slow trials were associated with attenuated, delayed, and prolonged responses to identical stimuli (Fig. 3b–d; Supplementary Fig. 6, data for individual subjects; Supplementary Fig. 7a, non-normalized peristimulus time histograms (PSTHs)). A quantitative paired comparison between the responses of each individual neuron in fast and slow trials (Fig. 3e and Online Methods) revealed that the response magnitude was attenuated by 17% in slow trials (z(376) = −3.05, P = 0.0023, Wilcoxon signed-rank test). Additionally, in slow trials, response latency (detected in individual trials; Online Methods) was delayed by 27 ± 6.9 ms (mean ± s.e.m.; z(376) = 3.5, P = 4.8 × 10−4, Wilcoxon signed-rank test). Analysis that quantified response latency as firing above baseline in PSTHs yielded similar results (data not shown). Response duration in slow trials was increased by 52 ± 19 ms (mean ± s.e.m.; z(376) = 3.2, P = 0.0012, Wilcoxon signed-rank test). Notably, analysis of spiking activity in neighboring nonresponsive neurons during the same trials did not reveal significantly different firing rates between fast and slow trials (P = 0.36, Wilcoxon signed-rank test). Thus, altered neuronal spiking activity during cognitive lapses was specific to responsive neurons and does not reflect a global reduction in activity at those times. Neuronal spiking responses were primarily associated with stimulus onset rather than motor responses (Supplementary Fig. 7). Correlation between the latency of MTL neuronal responses and RTs was also observed across all trials without focusing a priori on comparing fast versus slow trials (Supplementary Fig. 8).

Figure 3: Reduced, delayed, and lengthened single-unit responses during cognitive lapses. (a) Spiking responses (raster plots and PSTHs) in fast trials (lowest RTs; green) versus slow trials (highest RTs; orange) for two representative neurons in the anterior hippocampus and the parahippocampal gyrus (same neurons as those in the bottom row in Fig. 2c). Trials in the raster plots were sorted on the basis of the RT in each trial (slowest on top). Black ticks, action potentials; open red circles, response latency detected automatically; green, gray, and orange circles, behavioral response in fast, other, and slow trials, respectively. Vertical gray bars mark the absence of response in slow trials around 200–300 ms after stimulus onset. (b) Normalized PSTH of all responses (each row represents a response to 1 of 469 stimuli; 162 responsive neurons) during fast trials (left) and slow trials (right). Responses are aligned to the onset of response for each neuron across all trials (x axis), and amplitude (color scale) is normalized to the peak response of each neuron to take into account variability across neurons in response timing and amplitude. Vertical and sloped blue lines mark the average time of response onset and response termination, respectively, for each neuron (sorted by response duration). Green and orange vertical lines mark mean behavioral RTs in fast and slow trials, respectively. (c) Color superposition of PSTH responses (each row represents a response to 1 of 469 stimuli; 162 responsive neurons) in fast and slow trials. Responses are aligned (x axis) and normalized (y axis) as in b. Color intensity (inset legend) corresponds to firing rate magnitude, and hue (green versus orange) indicates stronger responses during fast versus slow trials at that time, respectively (Online Methods). Vertical and sloped white lines mark the average time of response onset and response termination, respectively, for each neuron (sorted by response duration). (d) Grand mean PSTH of all responses (n = 469 responses in 162 neurons) in fast trials (green) and slow trials (orange). Responses are aligned (x axis) and normalized (y axis) as in b. Green and orange arrows mark the mean behavioral RTs in fast and slow trials, respectively. (e) Quantification of response magnitude (left), response latency (middle), and response duration (right) in individual responsive neurons during fast trials versus slow trials (n = 376 pictures in 142 units). Slow trials are associated with statistically significant firing rate reduction (**P < 0.005, Wilcoxon signed-rank test), increased temporal latency (***P < 0.0005, Wilcoxon signed-rank test), and longer response duration (**P < 0.005, Wilcoxon signed-rank test). Gray dots and lines depict 16 individual sessions with at least 5 unit responses each. Error bars denote s.e.m. across responses. Full size image

We also examined responses to images using locally referenced LFPs recorded from the same MTL microwires with which single-unit neuronal activity was observed (Fig. 4). The robust increase in broadband LFP gamma power that occurs following sensory stimulation in multiple modalities is an extensively studied phenomenon. This LFP signal is linked to the neuronal spiking activity of local neuronal populations28 and typically co-occurs with a decrease in low-frequency power, also termed 'desynchronization' (refs. 28, 29, 30, 31). In line with these findings, the 'induced power' LFP response to images (Fig. 4a) consisted of an increase in broadband gamma power (>45 Hz, 50–600 ms after stimulus) and a decrease in slow/theta power (2–10 Hz, 300–700 ms) (Supplementary Fig. 9, examples of the LFP dynamics in single trials). LFP responses were selective: some MTL microwires (n = 270 channels in 31 sessions) showed a robust response (Fig. 4a, MTL responsive channels) whereas other neighboring channels (n = 198 channels in 31 sessions) did not show significant modulations (Fig. 4c, MTL nonresponsive channels), despite the presence of high-quality signals that allowed isolation of neuronal units (Online Methods).

Figure 4: Cognitive lapses are associated with weaker gamma power increase and weaker slow/theta power decrease in MTL LFPs. (a) Time–frequency decomposition of induced power changes in LFPs of MTL responsive channels (n = 270 channels in 31 sessions). Columns show the average power changes for all trials (left), fast trials (lowest RTs; middle), and slow trials (highest RTs; right). Hot and cold colors mark increases and decreases in power, respectively. Black rectangles highlight stimulus-induced increased power in the gamma frequency range (>45 Hz); pink rectangles highlight stimulus-induced decreased power in the slow/theta frequency range (2–10 Hz). (b) Time course of gamma power increase (top) and slow/theta power decrease (bottom) for fast trials versus slow trials. (c) Decomposition as in a for neighboring MTL nonresponsive channels (n = 198 channels in 31 sessions). (d) Time course as in b for neighboring MTL nonresponsive channels. (e) Quantification (median) of gamma power increases (45–100 Hz, 50–600 ms) for responsive (left) and nonresponsive (right) MTL channels. Asterisks indicate significant differences (Wilcoxon signed-rank tests comparing fast trials with slow trials: **P < 0.007). (f) Quantification (median) of slow/theta power decrease (2–10 Hz, 300–700 ms) for responsive (left) and nonresponsive (right) MTL channels. Asterisks indicate significant differences (Wilcoxon signed-rank tests comparing fast trials with slow trials: ***P < 1 × 10−7). In e and f, error bars denote s.e.m. computed across LFP channels (n = 270 and 198 for responsive and nonresponsive channels, respectively), and gray dots and lines denote 22 (responsive channels) and 17 (nonresponsive channels) individual sessions that had at least 5 LFP channels each. Green, fast trials; orange, slow trials. (g) Scatterplot of single-neuron response latency versus strength of gamma power increase showing that during slow trials increased latency in spiking responses is significantly correlated with weaker increase in LFP gamma power (Spearman coefficient r = −0.17, P = 0.007, n = 255 pictures that elicited significant responses across 87 units and 21 sessions; Online Methods). (h) Scatterplot of single-neuron response latency versus strength of slow/theta power decrease showing that during slow trials increased latency in spiking responses is significantly correlated with increased slow/theta LFP power (Spearman coefficient: r = 0.22, P = 4.5 × 10−4, n = 255 pictures that elicited significant responses across 87 units and 21 sessions). Full size image

In responsive LFP channels, cognitive lapses during slow trials were associated with a weaker increase in gamma power in comparison to fast trials (Fig. 4b,e; −19.1%, z(270) = 2.72, P = 0.006, Wilcoxon signed-rank test) and a weaker decrease in slow/theta power in comparison to fast trials (Fig. 4b,f; −76.2%, z(270) = −5.2, P = 2 × 10−7, Wilcoxon signed-rank test). In contrast, no significant effects of cognitive lapses were observed in neighboring nonresponsive MTL channels (Fig. 4d–f; gamma: z(198) = −0.57, P = 0.57; theta: z(198) = −0.98, P = 0.33; Wilcoxon signed-rank tests). In contrast to induced power changes, the power of the evoked (average) LFP at 2–10 Hz was lower during cognitive lapses (Supplementary Fig. 10), suggesting that induced power effects reflect changes in ongoing activity rather than changes in the stimulus-evoked event-related potential. Furthermore, during cognitive lapses, the latency of spiking responses negatively correlated with LFP gamma power (Fig. 4g; r = −0.17, P = 0.006) and positively correlated with LFP slow/theta power (Fig. 4h; r = 0.22, P = 4.5 × 10−4). The significant coupling between the degree of degradation in LFP and neuronal spiking responses suggests that these effects are tightly linked manifestations of neuronal lapses in selective circuits engaged in the task. Whether the cognitive lapses (and underlying neuronal activity) observed after sleep deprivation are qualitatively similar or different from sporadic slow responses occurring throughout wakefulness remains an open question for future studies3,6.

Considering the growing amount of literature on increased theta power (6–10 Hz) as a correlate of sleep pressure16,19,32, we examined theta power during baseline intervals preceding stimulus onset (Online Methods and Supplementary Fig. 11). First, we established that theta power in MTL LFPs was indeed associated with sleep pressure and cognitive lapses. We found that baseline theta power was (i) significantly correlated with TSA (Supplementary Fig. 11a; r = 0.26, P < 4.07 × 10−6), (ii) elevated after full-night sleep deprivation (Supplementary Fig. 11b; P < 2.74 × 10−5, Wilcoxon signed-rank test), and (iii) higher before cognitive lapse trials (Supplementary Fig. 11c; P < 0.0001, Wilcoxon signed-rank test). Baseline theta power also exhibited a modest albeit highly significant correlation with the level of slow/theta power (2–10 Hz) during the response interval (Supplementary Fig. 11d,e; r = 0.05, P < 4 × 10−37), suggesting that baseline theta activity might influence the degraded LFP response during cognitive lapse trials (Fig. 4). Overall, ongoing theta activity is increased with sleep pressure, and its decreased attenuation during cognitive lapses may lead to impoverished neuronal and cognitive responses.

Finally, we ruled out a potential contribution from pathological epileptiform activity. First, we confirmed that all the main findings (degraded neuronal and LFP responses) held when all data collected in regions eventually declared as being within the seizure-onset zone (SOZ; data not shown) were discarded. Second, we detected interictal spikes (IISs) across the entire LFP data set (n = 1,648 LFP channels) to test whether such events might occur more frequently around cognitive lapses (Online Methods and Supplementary Fig. 12a,b). IISs were detected in few trials (5.0 ± 0.23%) and were significantly more frequently detected within the SOZ than in other regions (Supplementary Fig. 12c; 2.7-fold increase, P < 10−48, Mann–Whitney U-test), attesting to successful IIS detection. However, cognitive lapses were not associated with increased frequency of IISs when considering all data (Supplementary Fig. 12d; P = 0.46, Wilcoxon signed-rank test, n = 1,533 channels) or when considering only MTL regions in which selective neuronal effects were observed (Supplementary Fig. 12f; P = 0.48, Wilcoxon signed-rank test, n = 619 channels). In fact, when considering only sessions after complete sleep deprivation, we found a small but significant reduction in the frequency of IISs around cognitive lapses (Supplementary Fig. 12e; P = 0.025, Wilcoxon signed-rank test, n = 186 channels). Thus, we could not find a consistent or robust relationship between IISs and cognitive lapses.

Altogether, these findings show that in sleep-deprived humans engaged in a visual categorization task selective neuronal spiking responses to images are attenuated, delayed, and lengthened before cognitive lapses, and such MTL modulations of spiking activity are associated with a selectively weakened decrease in slow/theta power in responsive LFP channels. Thus, degraded neuronal activity is already evident at the perceptual stage, in which responses of individual neurons in selected trials can predict subsequent cognitive lapses. The extent to which these effects are regionally specific remains unclear, but the current results establish that, within MTL regions, cognitive lapses specifically affect responsive circuits engaged in the task. Progressive delays in neuronal activity may further accumulate in downstream decision-making and/or motor regions during cognitive lapses, ultimately leading to slower behavior. In line with the biased competition model of selective attention33, degraded sensory cortical activity during cognitive lapses may fail to elicit high-quality perceptual representations, and visual information therefore cannot be effectively fed forward to the frontal lobe regions that ultimately determine behavior. It still remains unclear whether degraded MTL activity strictly reflects impaired bottom–up signaling or whether additional top–down attention mechanisms are at play.

Brief periods of silence (OFF periods) accompanied by slow waves in field potentials are hallmarks of non-rapid eye movement (NREM) sleep in both animals34 and humans35, and they are associated with behavioral immobility and unresponsiveness. Following sleep deprivation, awake rats exhibit local sleep-like slow/theta waves and shorter OFF periods that are associated with degraded behavioral performance19. Given that we could only record a few neurons simultaneously in each brain region and that OFF periods in wakefulness are short (∼80 ms), it was not possible to determine reliably whether such brief OFF periods occur in the human brain during sleep deprivation. However, we find that slow/theta activity, previously linked to sleepiness16,17,19,32, is increased before and during cognitive lapses, and these changes were associated with degraded spike responses. Impaired spike responses are observed in individual neurons engaged in a cognitive task without concurrent changes in the firing of neighboring neurons, and these changes predict specific cognitive impairments in sleep-deprived humans. The tight relationship between MTL activity and perception36 suggests that visual recognition itself may slow down as a result of sleep deprivation. PVT lapses are stochastic; i.e., they are unpredictable moment to moment owing to the influence of a random variable. The present findings suggest that degraded neuronal and LFP responses in the MTL do predict such lapses to some extent, which pushes the formal cause of cognitive lapses back another step in the neurobiological chain of events. The mechanisms underlying local neuronal lapses remain to be determined, although it is likely that transient instability in the activity of neuromodulatory systems, including cholinergic and noradrenergic neurons, may play a role37. Indeed, instability in pupil size (tightly linked with central noradrenergic activity38) is correlated with alertness39, and the synaptic release of acetylcholine is transiently diminished during poor behavioral performance40.