In a double-blind, randomized, placebo-controlled study, we included 20 clinically healthy male non-smokers, who were between 18 and 35 years of age with a body mass index between 18 and 26 kg/m² and a habitual caffeine intake between 300 and 600 mg/day. The amount of habitual caffeine intake was assessed by a self-report questionnaire adapted from Bühler, Lachenmeier (22), and Snel and Lorist (23).

Each volunteer completed three conditions [orders of condition see (24)]: A placebo condition (150 mg mannitol × 3/ day × 10 days), a daily caffeine condition (150 mg caffeine × 3/ day × 10 days), and an acute caffeine deprivation (caffeine from the first administration up to day 9, then switch to placebo until the end of day 10). Note that a 9-day placebo intake can be considered to be sufficient for a washout of the remaining effects of the participants' prior caffeine intake, at least regarding possible withdrawal symptoms (25). Between each of the conditions, there was at least 1 week of recovery in which caffeine intake was not standardized, as this regular intake pattern was considered to be the status quo. Among the 20 participants (habitual caffeine level mean ± SD: 474.1 ± 107.5 mg), 10 participants habitually consumed more and 10 less than the daily dose at the start of the ambulatory part. An additional period of prior abstinence may have introduced more variance due to individual differences in the course and responses to caffeine withdrawal.

The outcomes were assessed on day 10 for each condition. Figure 1 presents the study protocol, including the timing of the treatments, the outcome measurements, and the corresponding hours of caffeine deprivation in each condition. We scheduled the MRI scans in all three conditions for all participants at 12 h after waking up, to control for the confounding effects of different durations of wakefulness (26) and different levels of sleepiness (27) on cerebral variables.

Through the entire 10 days, participants complied with a regular sleep-wake cycle (8 h ± 30 min night-time sleep, no naps allowed) according to their habitual bedtime, which was monitored and recorded by actimetry (Actiwatch, Cambridge Neurotechnology, Cambridge, United Kingdom) and sleep diaries. Participants abstained from any other caffeine-containing diets, including coffee, tea, energy drink, soda, and chocolate, and the compliance was monitored by measuring caffeine levels in the evenings. The data of one participant's caffeine condition was excluded due to incompliance with the treatment. The laboratory environment from day 9 evening to day 10 was strictly controlled: dim light, half-supine posture (~45°), and regular dietary and lavatory times. Participants were allowed to sleep for 8 h in the night between day 9 and day 10. A digital device without time clues and access to the Internet was permitted. Social interaction was restricted to the group of experimenters.

Twenty volunteers (age: 26.4 ± 4.0 years; BMI: 22.7 ± 1.4 kg/m²; weights: 76.2 ± 8.7 kg; relative dosage of laboratory treatment: 6.0 ± 0.6 mg/kg/day; and self-reported habitual caffeine intake: 6.8 ± 2.3 mg/kg/day) completed the study. We present the 43-h profile of caffeine and paraxanthine in Figure 2, and the exhaustive data of all samples in three conditions are included in the supplement (Supplementary Table 1). The significant main effect of condition [F_{(2, 602)} = 618.0, p < 0.001] on salivary levels of caffeine indicated overall higher levels in the caffeine condition (t = 30.5, p < 0.001) compared to placebo. The overall levels of caffeine in the condition of caffeine deprivation were significantly lower compared to caffeine condition (t = −28.7, p < 0.001) but did not significantly differ from placebo (t = 1.7, p = 0.190). Indicated by the condition x sample interaction [F_{(10, 602)} = 9.3, p < 0.001], the overnight residual level of caffeine already before the first laboratory intake in the caffeine condition was significantly higher than in the placebo condition (see 17 h after the latest intake in Figure 2, t = 3.6, p < 0.001). Caffeine levels in the caffeine condition remained higher than in the placebo condition until 12 h after the latest intake (t = 5.3, p < 0.001).

Regarding the salivary levels of paraxanthine, we also observed a significant main effect of condition [F_{(2, 602)} = 400.4, p < 0.001]. Post-hoc analysis indicated that the paraxanthine levels were significantly higher in the caffeine (t = 25, p < 0.001) than in the placebo condition, while in the condition of caffeine deprivation, the levels were massively reduced (t = −21.7, p < 0.001, compared to caffeine) but still significantly higher than in placebo condition (t = 3.8, p < 0.001). Furthermore, indicated by the condition x sample interaction [F_{(10, 602)} = 2.0, p = 0.007], the overnight residual level of paraxanthine before the first laboratory intake in the caffeine condition was significantly higher than in the placebo condition (t = 5.4, p < 0.001). The paraxanthine levels remained higher than in the placebo condition after 12 h after the last intake (t = 6.9, p < 0.001).

As presented in Table 1, paraxanthine peaked 3.2 h later than caffeine (t = 3.8, p = 0.002) and had a 3.5 h longer half-life than caffeine (t = 2.7, p = 0.15). Furthermore, after correcting for the false discovery (reported in q values), the regression coefficients (Figure 3) indicated that higher habitual intake was significantly associated with shorter half-life of caffeine (β = −0.11, p = 0.002, q = 0.006) and of paraxanthine (β = −0.14, p = 0.04, q = 0.04), as well as at trend associated with the larger disproportionality between paraxanthine and caffeine (AUCc of PX/AUCc of CA: β = 0.05, p = 0.036, q =0.054).

As illustrated by Figure 2, the reduced hippocampal GM during daily caffeine intake reported earlier elsewhere (21) remained significantly lower after 36.5 h of caffeine deprivation compared to the placebo condition (t_W−P = −2.1, p_W−P = 0.039). The magnitude, however, was reduced after 36.5 h deprivation compared to the caffeine condition (t_W−C = 3.2, p_W−C = 0.003). On the other hand, CBF in precuneus, thalamus, and basal ganglia, which was reduced during daily caffeine intake (21), was elevated at 36.5-h caffeine deprivation compared to the placebo condition (t_W−P = 2.1, p_W−P = 0.039). In the supplement (Supplementary Figure 1), we also reported the results from the voxel-based WBA on GM and CBF after 36 h of caffeine deprivation compared to baseline.

Daily repeated intake of caffeine is a very common phenomenon and occurs in adults more frequently as compared to acute irregular consumption of the psychostimulant after a certain phase of abstinence [patterns of regular caffeine intake, see (14)]. Besides the present data, however, the evidence on the course of human caffeine metabolism under conditions of daily intake is scarce.

With a unified dose of 150 mg × 3/day, our study had an average relative dose of 2 mg/kg × 3 times/day in a 4-h interval. The median peak time and half-life of caffeine (<105 min) in this study are very similar to the acute intake reported in earlier studies (7, 29–32). However, compared to the only available study of daily intake (4), our participants exhibited much faster metabolism in caffeine and paraxanthine, which might be due to a shorter dose interval (2 h) in their study compared to ours (4 h). The interval of 4 h was nearly equal to the half-life time of caffeine and thus may have reduced the iteration of the caffeine concentration over repeated intake.

The faster metabolism compared to the earlier study (4) might also reflect genetic traits in our participants, who had an average level of habitual intake level of 474.1 ± 107.5 mg. Several genome-wide association studies (GWAS) have discovered the link between habitual caffeine intake and caffeine metabolism through the polymorphism of CYP1A2, AHR, and CYP2A6 (17–20). Specifically, the genetic variants of CYP1A2 and AHR associated with a lower habitual caffeine intake are also associated with slow caffeine metabolism [i.e., a higher plasma level of caffeine and lower ratios of paraxanthine to caffeine (19)]. This genetic tendency is supported by our salivary data, which showed that a lower habitual caffeine intake was associated with a longer half-life of caffeine, a longer half-life of paraxanthine, as well as a lower ratio of paraxanthine to caffeine. The elevated paraxanthine levels over the entire course might also reflect the tendency to have a fast transformation of caffeine to paraxanthine in high caffeine consumers, yet without a commensurate speed for the metabolism of paraxanthine.

As our study does not include a washout period before each ambulatory part, we cannot exclude that the accumulation of paraxanthine might have been carried over from participants' prior habitual intake. This does, however, not contradict the suggested risk for an accumulated paraxanthine concentration as a consequence of habitual or daily caffeine. A dose-disproportional elevation in paraxanthine is not limited to daily intake but can also be observed when increasing the dose of intake (33).

In summary, we observed an accumulation of paraxanthine present during a pattern of caffeine intake, which imitated a realistic dose and timing of consumption. This accumulation was further intensified in people with a higher dose of habitual caffeine consumption. Paraxanthine is considered to possess lower toxicity and anxiogenesis than caffeine and other methylxanthines (9, 34, 35), while it can still exert similar effects as caffeine in wake-promotion (9), psychostimulation (10), elevated blood pressure, and release of epinephrine (11). Since paraxanthine levels are prone to be mounted over daily intake, understanding the cognitive and physiological outcomes during prolonged exposure to paraxanthine will clarify the long-term benefits and harms of daily caffeine intake.

Earlier GWAS did not support the association between variants of CYP1A2 or AHR and long-term neurocognitive performance (17, 36). Nonetheless, a genetic trait for the susceptibility of the cerebral responses to caffeine intake might instead lie at the ADORA2A gene. Behaviorally, the polymorphism of ADORA2A has been linked to some caffeine-induced neuropsychological responses (37–40) and is also associated with the levels of habitual caffeine intake (37, 41). At the synaptic level, ADORA2A is responsible for not only the expression of adenosine A2A receptors (A2AR) but also the modulation of glutamatergic signaling (42) and dopaminergic receptors (38) as well as the distributions of adenosine A1 receptors (A1R) in the emotion-regulating regions, including the hippocampus (43). Adenosine A1R and A2AR play counteractive roles in the hippocampal synaptic homeostasis (44, 45), where A2AR facilitates presynaptic glutamate release while A1R modulates the postsynaptic NMDA receptors and reduces excitatory signals (44). Hence, a potential modification of A1R and A2AR expressions by a polymorphism of ADORA2A might set a predisposition for the neural responses toward the antagonist, caffeine. For instance, compared to people with C genotypes of ADORA2A, ADORA2A T/T allele carriers had higher average levels of habitual caffeine intake and a higher susceptibility to the anxiogenic effect of caffeine as well as an increased expression of A1R in frontal, hippocampal, and entorhinal cortices (43). As moderate-to-high caffeine consumers, our participants might bear the genetic propensity to have stronger responses toward A1R antagonism by caffeine, which might pose a disposition to hippocampal plasticity or excitotoxicity. Future studies are therefore recommended to include a wider range of habitual caffeine consumers or collect genetic information for better precision.

To date, there is no direct evidence on the link between ADORA2A and caffeine-induced hippocampal plasticity or CBF in healthy humans. It has been found, however, that the variants of ADORA2A were associated with the degeneration of hippocampal gray matter in both schizophrenic patients (46) and aging populations with cognitive decline (42). Adding to this point, the frequently reported association between lifetime caffeine intake and a reduced risk for cognitive decline (47–50) or neurodegeneration (51–53) could be driven by a common genetic factor. Our results ad to this point that at least in a certain population–potentially consumers who naturally accommodate to moderate habitual caffeine consumption–a daily intake of moderate-dose caffeine might instead increase the risk for a reduction in hippocampal gray matter.

Withdrawal responses commonly occur after discontinuing regular caffeine intake. While our reports elsewhere have discussed in detail the implications of cerebral effects of caffeine (21, 54), the current analyses further added that these changes might take longer than the interval of day-to-day intakes to recover. We corroborated the caffeine cessation-induced vasodilation by the elevated CBF (55–60). An increased CBF, which could be attributed to the enhanced adenosine-modulated vasodilation after chronic caffeine exposure (61–63), was also frequently observed during increased sleep pressure, such as in the evening compared to morning (26), after sleep restriction (64), and after sleep deprivation (26). In line with the literature, the cognitive responses in the same volunteers reported elsewhere, including reduced vigilance, increased sleepiness, and enhanced sleep depth (24, 65), confirmed that the participants were experiencing a solid withdrawal state.

Responses toward an acute caffeine deprivation can manifest as a restoration of adapted neural functioning when caffeine intake is ceased. The molecular mechanism of such adaptations includes a mounting concentration of extracellular endogenous adenosine (61) and upregulated adenosine receptors (66–68). Both mechanisms can lead to a surge of cognitive and physiological reactions through the strengthened adenosine binding. These reactions include fatigue, reduced concentration, mood disturbance, headache, as well as increased vasodilation as addressed (25, 69, 70). Intense withdrawal symptoms are usually perceived around 20 and 50 h after the last regular intake and, in extreme cases, can last maximal 9 days (25). The earliest observed neurovascular responses to caffeine deprivation in existing evidence were at 21 h (58).

While combating withdrawal symptoms are often the reason to consolidate the daily repeated consumption of caffeine (71), the typical daily repeated caffeine intake, however, is unlikely to provide enough time for a full “withdrawal-driven restoration.” The first evidence comes from our observation in the overnight residuals of caffeine and paraxanthine levels, which were measured at 17 h after the last intake. Furthermore, during the 24–43 h of caffeine deprivation when the caffeine levels were nearly cleared, the paraxanthine levels remained elevated. In other words, a repetition of intake shorter than this time window is most likely to be insufficient for the full elimination of both caffeine and paraxanthine. The prolonged presence of high levels of such adenosine antagonists may impede the neural homeostasis dependent on the activations of adenosine receptors. In neurogenesis of adult rodents, the presence of a similar dose of caffeine during sleep has been shown to suppress the cell proliferation in the hippocampus [(72); Dose conversion: 10 mg/kg in rodents were estimated to be equivalent to 250 mg/ 70 kg in humans (71)]. The concurrence of a reduced hippocampal GM intensity and an incomplete caffeine elimination, together with a GM recovery after being deprived of caffeine, underscore the importance of a sufficient cessation period. The overcompensation of the CBF response after the caffeine deprivation, again, points to a longer time, perhaps some consecutive days, for full recovery. Future studies should confirm this postulation with a design which allows full elimination caffeine and paraxanthine as well as an observation of the response in cerebral morphology.