I. Introduction

Top Next References

Caffeine is the most widely consumed behaviorally active substance in the world. Almost all caffeine comes from dietary sources (beverages and food), most of it from coffee and tea. Acute and, especially, chronic caffeine intake appear to have only minor negative consequences on health. For this reason and because few caffeine users report loss of control over their caffeine intake, governmental regulatory agencies impose no restrictions on the use of caffeine. Ordinary caffeine use has generally not been considered to be a case of drug abuse, and is indeed not so classified in DSM-IV (Diagnostic and Statistical Manual of Mental Disorder).³ However, some years ago it was pointed out that caffeine may be a potential drug of abuse (see Gilliland and Bullock, 1984), and more recently caffeine has been described as "a model drug of abuse" (Holtzman, 1990) and the possibility that caffeine abuse, dependence, and withdrawal should be added to diagnostic manuals has been seriously considered (Hughes et al., 1992b; Strain et al., 1994; Pickworth, 1995; Hughes et al., 1998)

In the present review we discuss the evidence regarding caffeine and dependence in light of increasing knowledge regarding the actions of caffeine on specific neuronal brain substrates. Because the use of caffeine is probably related to its diverse effects on several brain functions, these are also briefly presented. Even though we have attempted to cover many of the aspects that are relevant to this complex issue, we are aware of several omissions and we also realize that the complexoften somewhat contradictoryliterature lends itself to more than one interpretation.

	II. Consumption and Metabolism of Caffeine

Top Previous Next References

Although coffee and other caffeine-containing beverages were introduced in Europe only a few hundred years ago, consumption of these beverages now occupies a significant place in our national cultures. The same can be said for most nations of the world (see Table 1). The national consumption of caffeine summarized in this table relies heavily on official statistics, which are notoriously unreliable. It is, for example, possible that the rather low figures for caffeine consumption in countries that produce the relevant plants may partly be due to the fact that not all the production has entered into the official statistics. In addition, Table 1 does not include soft drinks, although they are a major source of caffeine for example for children in Western society.

Caffeine is present in a number of dietary sources consumed worldwide, i.e., tea, coffee, cocoa beverages, chocolate bars, and soft drinks. The content of caffeine of these various food items ranges from 40 to 180 mg/150 ml for coffee to 24 to 50 mg/150 ml for tea, 15 to 29 mg/180 ml for cola, 2 to 7 mg/150 ml for cocoa, and 1 to 36 mg/28 g for chocolate (Barone and Roberts, 1996; Debry 1994; see also Table 2). Difficulties in taking all the sources into account may partly explain the considerable differences, such as in the estimates of caffeine consumption in the United Sates [from 196 to 423 mg/ 24 h; Weidner and Istvan (1985)] or in the UK [from 359 to 621 mg/24 h; Bruce and Lader (1986)].

Caffeine consumption from all sources can be estimated to around 70 to 76 mg/person/day worldwide (Gilbert, 1981, 1984) but reaches 210 to 238 mg/day in the US and Canada and more than 400 mg/person/day in Sweden and Finland, where 80 to 100% of the caffeine intake comes from coffee alone (Debry, 1994; Barone and Roberts, 1996; Viani, 1996). In the UK, the consumption is as high as in Sweden and Finland, but 55% comes from tea, 43% from coffee, and 2% from colas (Barone and Roberts, 1996). According to the recent survey of Barone and Roberts (1996), the daily intake of caffeine from all sources in the US is estimated at 3 mg/kg/person, two-thirds of it coming from coffee in subjects more than 10 years old. If only consumers are taken into account, the daily caffeine consumption reaches a value of 2.4 to 4.0 mg/kg (170-300 mg) in a 60- to 70-kg individual. In 7- to 10-year-old children, the daily consumption of caffeine ranges from 0.5 to 1.8 mg/kg. The soft drinks represent 26 to 55%, chocolate foods and beverages 17 to 40%, tea 6 to 34%, and coffee 0 to 22% of the total caffeine intake (Morgan et al., 1982; Arbeit et al., 1988; Ellison et al., 1995). It is also clear from the data given below that the amounts of caffeine ingested via these sources are biologically active. This emphasizes that caffeine is indeed the most widely used of all psychoactive drugs.

Caffeine absorption from the gastrointestinal tract is rapid and reaches 99% in humans in about 45 min after ingestion (Marks and Kelly, 1973; Bonati et al., 1982; Blanchard and Sawers, 1983a,b; Arnaud, 1993). Caffeine absorption is also complete in animals (Arnaud, 1976, 1985). Pharmacokinetics are comparable after oral or i.v. administration of caffeine in humans and animals, leading to superimposable plasma curves (Arnaud, 1993). Absorption is, however, not complete when the substance is taken as coffee (Morgan et al., 1982). It is also known that when very large doses of caffeine are accidentally ingested, toxic effects appear, with an LD₅₀ of about 200 mg/kg in rats (see Eichler, 1976). In patients who have been admitted to hospital due to acute caffeine poisoning, levels of a few hundred micromoles per liter have been recorded.

The hydrophobic properties of caffeine allow its passage through all biological membranes. There is no blood-brain barrier to caffeine in the adult or the fetal animal (Lachance et al., 1983; Tanaka et al., 1984), and the blood-to-plasma ratio is close to unity (McCall et al., 1982), indicating limited plasma protein binding and free passage into blood cells. In newborn infants, caffeine concentration is similar in plasma and cerebrospinal fluid (Turmen et al., 1979; Somani et al., 1980). There is no placental barrier to caffeine (Ikeda et al., 1982; Kimmel et al., 1984) and unusually high levels of caffeine have been reported in premature infants born to women who are heavy caffeine consumers (Khanna and Somani, 1984). Finally, saliva concentrations of caffeine, which are considered to be a reliable index of plasma caffeine levels, reach 65 to 85% of plasma concentrations (Cook et al., 1976; Khanna et al., 1980).

Peak plasma caffeine concentration is reached between 15 and 120 min after oral ingestion in humans and equals 8 to 10 mg/l for doses of 5 to 8 mg/kg (Arnaud and Welsch, 1982; Bonati et al., 1982). Ingestion of a single cup of coffee provides a dose of 0.4 to 2.5 mg/kg. It can therefore be estimated that this gives a peak concentration of 0.25 to 2 mg/l or approximately 1 to 10 µM.

For doses lower than 10 mg/kg, caffeine half-lives range from 0.7 to 1.2 h in rat and mouse, 3 to 5 h in monkey (Bonati et al., 1984-1985) and 2.5 to 4.5 h in humans (Arnaud, 1987). There are no differences in caffeine half-life in young and elderly humans (Blanchard and Sawers, 1983b). Conversely, caffeine half-life is increased during the neonatal period due to lower activity of cytochrome P-450 (Aranda et al., 1979) and to the relative immaturity of some demethylation and acetylation pathways (Aranda et al., 1974; Carrier et al., 1988). The half-life of caffeine is about 80 ± 23 h for the full-term newborn infant (Aranda et al., 1977; Le Guennec and Billon, 1987) and can be over 100 h in premature infants (Parsons and Neims, 1981). Thereafter, the half-life of caffeine decreases exponentially with postnatal age to 14.4 and 2.6 h in 3- to 5- and 5- to 6-month-old infants, respectively (Aldridge et al., 1979; Parsons and Neims, 1981; Paire et al., 1988; Pearlman et al., 1989). The clearance of caffeine is low in 1-month-old infants (31 ml/kg/h), increases to a maximal value of 331 ml/kg/h at 5 to 6 months, and is 155 ml/kg/h in adult humans (Aranda et al., 1979). In adult males, caffeine half-life is reduced by 30 to 50% in smokers compared with nonsmokers (Hart et al., 1976; Joeres et al., 1988; Murphy et al., 1988), whereas it is approximately doubled in women taking oral contraceptives (Patwardhan et al., 1980) and greatly prolonged (up to 15 h) during the last trimester of pregnancy (Aldridge et al., 1981; Knutti et al., 1981; Brazier et al., 1983).

Caffeine is metabolized by the liver to form dimethyl- and monomethylxanthines, dimethyl and monomethyl uric acids, trimethyl- and dimethylallantoin, and uracil derivatives (Arnaud, 1987, 1993). The demethylation, C-8 oxidation, and uracil formation occur mostly in liver microsomes. The major metabolic difference between rodents and humans is that, in the rat, 40% of the caffeine metabolites are trimethyl derivatives as compared with less than 6% in humans (Arnaud, 1985, 1993). Metabolism in humans is characterized by the quantitative importance of the 3-methyl demethylation leading to the formation of paraxanthine. This first metabolic step represents up to 72 to 80% of caffeine metabolism (Arnaud and Welsch, 1982; Arnaud, 1993). Many of the metabolic steps may be saturable in humans as the elimination half-time for not only caffeine, but also some of its metabolites, is dose-dependent (Kaplan et al., 1997).

Some metabolites of caffeine also have marked pharmacological activity. Thus, 1,3-dimethylxanthine (theophylline) and 1,7-dimethylxanthine (paraxanthine) must be taken into account when considering the biological actions of caffeine-containing beverages. In rodents, paraxanthine is the major metabolite in plasma, but levels of theophylline are also high. The metabolism of caffeine to paraxanthine can be used to phenotype individuals with regard to one subform of cytochrome P-450, CYP1A2 (Fuhr et al., 1996; Miners and Birkett, 1996). By contrast, the formation of theophylline from caffeine does not correlate with any specific subform.

It has recently been shown that, after long-term caffeine ingestion, the levels of theophylline in the mouse brain may be higher than those of caffeine during a substantial part of the day and almost always higher than the levels of paraxanthine (Johansson et al., 1996a). This could mean that caffeine in the brain is metabolized partly via specific, local enzymatic pathways and that caffeine administration leads to high central nervous system (CNS) concentrations of theophylline, whereas peripheral theophylline levels are kept low. It is possibly relevant that demethylation of caffeine to paraxanthine in rats appears to be predominantly catalyzed by cytochrome P-450, whereas demethylation to theophylline and theobromine may also take place via flavin-containing monooxygenases (Chung and Cha, 1997). Future studies will have to be performed to determine if the situation is similar in humans. It is, however, clear that the contention that most of the effects of caffeine in the CNS are direct or indirect consequences of adenosine receptor blockade (see Section III below) increases in strength if local CNS concentrations of theophylline and/or paraxanthine are high after caffeine ingestion. Theophylline is some three to five times more potent than caffeine as an inhibitor of both adenosine A₁ and A_2A receptors, and paraxanthine is also at least as potent as caffeine. Indeed it has been shown that, in humans, some tested effects of caffeine are readily mimicked by paraxanthine (Benowitz et al., 1995).

Because so much of the background information is derived from animal experiments, we must try to extrapolate the data to humans. However, it is not a trivial task to compare doses of caffeine in animals and humans. For example, it must be kept in mind that in most experiments on rodents, one single high dose of caffeine is administered, whereas human consumption of coffee is divided up during the day. Gilbert (1976) suggested the use of a metabolic body weight correction factor when comparing the effect of a given dose of caffeine in animals and humans. However, not everyone agrees that such a correction based on the metabolic body weight should be applied. Indeed the LD₅₀ of caffeine is fairly consistent across species, including Homo sapiens (Dews, 1982). The plasma level resulting from 1.1 mg/kg caffeine (a single cup of coffee containing 80 mg of caffeine ingested by a 70-kg human) ranges from 0.5 to 1.5 mg/l. A similar dose-concentration relationship is found in many species, including rodents and primates (Hirsh, 1984). However, because the metabolism of caffeine differs between rodents and humans and the half-life of the methylxanthine is much shorter in rats (0.7-1.2 h) than in humans (2.5-4.5 h) (Morgan et al., 1982), it seems reasonable to correct for the metabolic body weight when comparing animal and human doses. Thus, it is generally assumed that 10 mg/kg in a rat represents about 250 mg of caffeine in a human weighing 70 kg (3.5 mg/kg), and that this would correspond to about 2 to 3 cups of coffee.

	III. Molecular and Cellular Action of Caffeine in the Brain

Top Previous Next References

The biochemical mechanism that underlies the actions of caffeine at doses achieved in normal human consumption must be activated at concentrations between the extremes (between barely effective doses and doses that produce toxic effects; see Fig. 1). This tends to rule out the direct release of intracellular calcium [probably via an action on ryanodine receptors (McPherson et al., 1991)], which occurs only at millimolar concentrations. Also the inhibition of cyclic nucleotide phosphodiesterases (Smellie et al., 1979; see Fredholm, 1980; Nehlig and Debry, 1994) occurs at rather higher concentrations than those attained during human caffeine consumption. Xanthines can influence 5'-nucleotidase and alkaline phosphatase, but these actions are also exerted only at millimolar concentrations (Fredholm et al., 1978; Fredholm and Lindgren, 1983). In fact, the only known mechanism that is significantly affected by the relevant doses of caffeine is binding to adenosine receptors and antagonism of the actions of agonists at these receptors (see Fredholm, 1980, 1995). Thus, in the remainder of this section, adenosine receptor antagonism is taken to be the mechanism of action of caffeine even though there are data, especially from behavioral experiments, that could be interpreted as evidence for some other, as yet unidentified mechanism of action (see, e.g., Garrett and Holtzman, 1995).

View larger version (29K): [in this window] [in a new window]   Fig. 1.   Effect of caffeine on different biochemical targets in relation to its levels in humans. Note that caffeine is able to significantly block adenosine effects on A2A (most potent) and A1 receptors already at the low concentrations achieved after a single cup of coffee. To inhibit cyclic nucleotide breakdown via inhibition of phosphodiesterase, 20 times higher concentrations are required; to block GABAA receptors, 40 times higher concentrations; and to mobilize intracellular calcium depots, concentrations of 100 times higher are needed. These latter levels are unlikely to be reached in humans by any form of normal use of caffeine-containing beverages (modified from Fredholm, 1980).

There are two enzymes that constitute the major pathways of adenosine removal: adenosine kinase and adenosine deaminase. The latter enzyme is present mostly intracellularly but is also found in some extracellular compartments. The preferred substrate of the enzyme is not adenosine but 2-deoxyadenosine (Fredholm and Lerner, 1982). The K_m for adenosine is well above 5 µM and adenosine deaminase is therefore of particular importance when adenosine levels are high (Arch and Newsholme, 1978). Adenosine kinase, by contrast, has a K_m level in the range of physiological intracellular adenosine concentrations. Indeed, blockade of adenosine kinase has a much larger effect on the rate of adenosine release than does blockade of adenosine deaminase (Lloyd and Fredholm, 1995). Another enzyme of importance is S-adenosylhomocysteine hydrolase. This enzyme sets the equilibrium between S-adenosylhomocysteine and adenosine + L-homocysteine. When the level of the amino acid is low, this enzyme serves to generate adenosine. On the other hand, when the level of L-homocysteine is raised, it can trap adenosine formed via AMP breakdown as S-adenosylhomocysteine inside the cell. This reaction has been used to demonstrate that the bulk of the adenosine formed by energy deprivation or electrical field stimulation in hippocampal slices is formed intra- rather than extracellularly (Lloyd et al., 1993).

Extracellular ATP is very rapidly hydrolyzed to adenosine and other metabolites. Thus, if ATP is released from neuronal (or glial) cells, e.g., as a transmitter or an intercellular signal, it will provide a source of extracellular adenosine. It seems likely that this may be significant in some circumstances and in some locations. However, extracellular ATP is not the major source of adenosine released from brain slices during field stimulation (at least when relatively low frequency stimulation is used) or following hypoxia/hypoglycemia. This is shown by the fact that agents that block extracellular AMP hydrolysis fail to affect the rate of adenosine release significantly (Lloyd et al., 1993). Thus, intracellular adenosine formation is quantitatively most important.

Intra- and extracellular adenosine concentrations are kept in equilibrium by means of equilibrative transporters. These transporters are blocked by several agents such as nitrobenzylthioinosine, propentofylline, dipyridamole, and dilazep. In addition there are sodium-dependent, concentrating transporters that move extracellular adenosine into cells. These latter transporters are not blocked by the above agents, and their precise role in the CNS is unknown. When inhibitors of equilibrative transport are given, the levels of adenosine rise in the CNS despite a decrease in the release of adenosine metabolites such as inosine and hypoxanthine (Andiné et al., 1990; Fredholm et al., 1994b). The reason for this has been discussed elsewhere (see Fredholm et al., 1994b). Adenosine, once released, can secondarily be taken up by cells and metabolized to inosine and hypoxanthine. It should, however, be pointed out that transport inhibitors block the overall release of adenine nucleotide breakdown products (Jonzon and Fredholm, 1985), as expected for a model where equilibrative transporters are critically important. Furthermore, the addition of L-homocysteine in the presence of transport inhibitors leads to a very substantial reduction in the efflux of adenosine. The reason is that an excess of L-homocysteine forces the S-adenosylhomocysteine hydrolase reaction to occur in reverse and intracellular adenosine levels are very much reduced. When intracellular levels are decreased the extracellular levels also go down.

From these facts it can be deduced that adenosine levels in the extracellular fluid should be raised whenever there is a discrepancy between the rate of ATP consumption and ATP synthesis. In addition, it is expected that drugs that interfere with the key enzymes and with the transporters should affect adenosine levels. Extracellular adenosine levels have been measured using microdialysis. In the first paper using this method it was shown that the level of adenosine, while initially high, stabilized at about 1 µM within a few hours of implantation of the dialysis probe (Zetterström et al., 1982). It was also shown that the level of adenosine was raised about 3-fold following a mild hypoxia. The level of adenosine can increase dramatically to 10 µM or more following ischemia (Andiné et al., 1990; Dux et al., 1990). However, a later study showed that it took much longer to reach the true equilibrium level and that consequently the disturbance produced by the microdialysis probe lasted for perhaps 24 h (Ballarin et al., 1991). Our current best estimate of the basal level of adenosine in the brain of awake, unrestrained rats is between 30 and 300 nM. Interestingly, these levels are close to the estimated levels of adenosine in plasma (Reid et al., 1991). There is one report to the effect that caffeine, particularly prolonged administration of caffeine, increases the levels of adenosine in plasma dramatically (Conlay et al., 1997) at least in rats, and that this effect is receptor-mediated. This finding clearly needs to be reproducedespecially in humans.

We turn now to the question of whether there are adenosine receptors that are activated not only by the high adenosine levels seen in ischemia, but also by the low (high nanomolar concentrations) physiological levels.

C. Adenosine Acts on Several Types of G-Protein-Coupled Receptors

1. Receptor Subtypes. At present four distinct adenosine receptors, A₁, A_2A, A_2B, and A₃, have been cloned and characterized in several species (Fredholm et al., 1994a; Table 3). Of these subtypes, the rat A₃ receptor was originally shown to be but little affected by many methylxanthines, including caffeine. In humans, the A₃ receptor is blocked by caffeine with a K_D of close to 80 µM. Therefore, this receptor is not the best target for caffeine actions in humans. The A_2B receptor has been shown to require higher concentrations of adenosine for activation than those found in resting animal tissues. Thus, inhibition of adenosine actions at this receptor is similarly unlikely to provide an explanation for the actions of caffeine under physiological conditions. Under pathophysiological conditions, however, A_2B receptors are likely to be activated by endogenous adenosine and caffeine may then very well act also on these receptors.

2. Receptor Distribution. A₁ and A_2A receptors in the brain can be localized by receptor autoradiography with radioactive ligands. In addition, the sites of receptor synthesis can be determined using in situ hybridization. Adenosine A₁ receptors are present in almost all brain areas, with the highest levels in hippocampus, cerebral and cerebellar cortex, and certain thalamic nuclei (Goodman and Snyder, 1982; Fastbom et al., 1987). Only moderate levels are found in caudate-putamen and nucleus accumbens. The corresponding mRNA shows a somewhat different distribution (Mahan et al., 1991; Reppert et al., 1991), indicating that some of these receptors are located on nerve terminals rather than cell bodies (Johansson et al., 1993a). Indeed, the presence of presynaptic adenosine A₁ receptors mediating inhibition of transmitter release has been demonstrated on virtually all types of neurons [for review see Fredholm and Dunwiddie (1988)]. In the caudate-putamen, adenosine A₁ receptor mRNA was found to be present, albeit in low abundance, on all the major types of neurons (Ferré et al., 1996).

View larger version (58K): [in this window] [in a new window]   Fig. 2.   The similarity in the distribution of adenosine A2A, dopamine D2, and dopamine D1 receptors in rats. These film autoradiograms (negatives) show the distribution of mRNA (in situ hybridization) or protein (receptor autoradiography with antagonist radioligand) for the three types of receptors in coronal sections of rat brain. Note the excellent colocalization.

View larger version (32K): [in this window] [in a new window]   Fig. 3.   A2A receptor mRNA is specifically associated with D2, but not with D1 receptors, with enkephalin, but not with Substance P, and is essentially absent from large aspiny cholinergic neurons. In situ hybridization experiments were carried out using two separate probes in each experiment. A radioactive probe for the adenosine A2A receptor was combined with a nonradioactive probe for preproenkephalin, preprotachykinin, dopamine D1 and D2 receptors, or choline acetyl transferase. Using an image analysis system, the number of cells expressing mRNA for the A2A receptor was calculated as a percentage of the total number of cells expressing mRNA for the other probes. Cells were counted in several areas of the brain, including dorsolateral and dorsomedial caudate putamen (CP), nucleus accumbens (NC) core and shell, and in the olfactory tubercle. Data are from Svenningsson et al. (1997b).

The inhibitory effect of adenosine on transmitter release was first noted in the peripheral nervous system, but similar effects in the CNS were soon demonstrated (see Fredholm and Hedqvist, 1980; Fredholm and Dunwiddie, 1988). There is some evidence that the release of excitatory transmitters is more strongly inhibited by adenosine than that of inhibitory neurotransmitters (Fredholm and Dunwiddie, 1988). This would be in keeping with a proposed role of adenosine as a homeostatic regulatory factor that serves to match the rate of energy consumption to the rate of substrate supply. The receptors involved are similar to adenosine A₁ receptors.

As discussed previously (Fredholm and Dunwiddie, 1988), adenosine appears to use several mechanisms in order to produce inhibition of transmitter release. The electrically evoked release, but not the spontaneous release of neurotransmitter, is strongly dependent on the concentration of calcium in the extracellular environment (see Fredholm and Hu, 1993). On the other hand, the electrically evoked release is poorly affected by buffers of intracellular calcium, whereas the spontaneous transmitter release is strongly affected. This supports the contention that calcium entering via voltage-dependent calcium channels and acting on docked vesicles in the neighborhood of the channel is important. It is interesting to note that adenosine acting on A₁ receptors has been shown to decrease calcium entry via N-type channels in hippocampal CA1 and CA3 neurons (Scholz and Miller, 1992; Mogul et al., 1993). However, in some types of neurons the effect of adenosine is only moderately or not at all affected by -conotoxin, a selective inhibitor of N-type channels (Fredholm, 1993). This could mean that adenosine acts either on some other type of calcium channel (Takahashi and Momiyama, 1993), perhaps the putative Q-type channel (Sather et al., 1993; Takahashi and Momiyama, 1993). Another possibility is that adenosine affects, directly, a calcium-sensitive member of the release machinery, a contention for which there is some support (Silinsky, 1984; Scholz and Miller, 1992; Thompson et al., 1992). However, when it has been possible to actually measure Ca²⁺ influx, the reduction has been found to adequately account for the decrease in transmitter release (Yawo and Chuhma, 1993; Wu and Saggau, 1994). There is also some evidence that increases in cyclic AMP in nerve endings are associated with an increase in transmitter release (Chavez-Noriega and Stevens, 1994). Because activation of adenosine A₁ receptors is known to cause a decrease in cAMP formation, it is conceivable that this may also be a mechanism of decreased transmitter releaseat least under some circumstances.

There is also considerable evidence that adenosine acts to decrease the rate of firing of central neurons (Phillis and Edstrom, 1976). This effect appears to be quite general and is due to an activation of potassium channels via adenosine A₁ receptors (Dunwiddie, 1985). When the effect of endogenous adenosine at these receptors on glutamatergic neurons is blocked by caffeine, it leads to epileptiform activity in vitro (Dunwiddie, 1980; Dunwiddie et al., 1981), and this could be the mechanism by which methylxanthines produce seizures in vivo.

It is also known that caffeine increases the turnover of several monoamine neurotransmitters, including 5hydroxytryptamine (5-HT) dopamine, and noradrenaline (Fernström and Fernström, 1984; Bickford et al., 1985; Fredholm and Jonzon, 1988; Hadfield and Milio, 1989). There is evidence that methylxanthines increase the rate of firing of noradrenergic neurons in the locus ceruleus (Grant and Redmond, 1982). The increase in noradrenaline turnover is probably the explanation for the fact that methylxanthines also reduce the number of -adrenoceptors in rat brain (Fredholm et al., 1984; Shi et al., 1993a). It has also been shown that the mesocortical cholinergic neurons are tonically inhibited by adenosine and that caffeine consequently increases their firing rate (Rainnie et al., 1994). It was postulated that this effect is of importance in the electroencephalogram (EEG) arousal following caffeine ingestion. Because dopamine and noradrenaline neurons also are involved in arousal, there is ample neuropharmacological basis for assuming that central stimulatory effect of caffeine could be related to inhibition of adenosine A₁ receptors. Also there are increases in 5-hydroxytryptamine receptors, muscarinic receptors, and -opioid receptors following higher doses of caffeine (Shi et al., 1993a, 1994). The functional relevance, if any, of these changes remains to be elucidated.

There is considerable evidence for a link between adenosine A₁ receptors and dopamine D₁ receptors (see Ferré et al., 1997). Thus, blockade of adenosine A₁ receptors enhances motor effects of D₁ receptor agonists. Infusion of an adenosine A₁ receptor agonist into the caudate-putamen does not per se modify the levels of GABA in the entopeduncular nucleus, the output structure of the dopamine D₁ receptor-expressing, medium-sized GABAergic neurons, but blocks the stimulatory effect of a D₁ receptor agonist (Ferré et al., 1996). There are several possible mechanisms that could underlie these behavioral and neurochemical effects. It has been shown that activation of adenosine A₁ receptors influence the binding of dopamine D₁ agonists (Ferré et al., 1994, 1996, 1998). There are also more indirect interactions, and an involvement of N-methyl-D-aspartate (NMDA) receptors has been implicated. It is interesting to note that a recent study (Harvey and Lacey, 1997) presented strong evidence that combined dopamine D₁ and NMDA receptor stimulation increases the release of adenosine, which then acts at adenosine A₁ receptors to decrease the release of the excitatory neurotransmitter. Some of these interactions between A₁, D₁, and NMDA receptors are schematically represented in Fig. 4. In addition, D₁ receptors in the ventral tegmental area (VTA) interact with adenosine A₁ receptor effects (Bonci and Williams, 1996).

View larger version (30K): [in this window] [in a new window]   Fig. 4.   Schematic illustration of the effect of caffeine on striatopallidal and striatonigral neurons. A, potential interactions between A2A and D2 receptors in the GABAergic neurons that comprise the so-called indirect pathway and project to the ventral pallidum. B, a simplified wiring diagram of the nucleus accumbens and some of its input and output structures. Synapses are shown as stimulatory () or inhibitory (). In this part of the figure are also indicated areas where adenosine and dopamine receptor subtypes are enriched. C, the interactions between A1, D1, and glutamate receptors in neurons that comprise the so-called direct pathway. In particular, it should be noted that activation of dopamine D1 receptors can enhance the actions mediated via NMDA receptors. This causes release of adenosine, which activates A1 receptors located on the terminals of the excitatory input. Hereby the release of glutamate is reduced.

As noted above A_2A receptors are located preferentially in the subpopulation of the medium sized spiny GABAergic neurons that project to globus pallidus, a subpopulation in which they are colocalized with dopamine D₂ receptor mRNA. These colocalized receptors have been shown to interact functionally. Thus, activation of A_2A receptors has been shown to decrease the affinity of dopamine binding to D₂ receptors (Ferré et al., 1991), but not antagonist binding affinity. This type of change mimics that observed following the addition of sodium ions. However, the effect of the A_2A receptor agonist was at least as large in the presence as in the absence of sodium ions. The interaction could not be observed when high levels of dopamine D₂ and adenosine A_2A receptors were transiently expressed in Cos-7 cells (Snaprud et al., 1994). In these cells the receptors were not functionally coupled to an effector response. This suggests that the interaction may not occur between the receptor molecules only but that some interactions with other membrane components are also necessary. Conversely, in fibroblasts stably transfected with both A_2A and D₂ receptors there was a clear-cut interaction at the binding level despite the fact that the A_2A receptors in these cells were very poorly coupled to adenylyl cyclase (Dasgupta et al., 1996). This indicates that the interaction at the level of binding does not require the full effector response. The latter contention is also supported by the fact that the binding studies were conducted on broken cell preparations and under conditions when adenylyl cyclase activity and protein phosphorylation can be expected to proceed at negligible rates. More recently this binding interaction was observed in Chinese hamster ovary cells cotransfected with A_2A and D₂ receptors, and in this cell type there were also very clear-cut interactions at the second messenger level and beyond.

There is evidence that these interactions between adenosine A_2A and dopamine D₂ receptors observed in vitro have functional correlates in intact striatum. Thus, it is interesting that dopamine administered in the striatum has been shown to block the release of GABA in the globus pallidus (Ferré et al., 1993) and that this effect is reduced by endogenous adenosine. Furthermore, activation of adenosine A_2A receptors increases GABA release from pallidal slices (Mayfield et al., 1993). The effects of adenosine on striatal GABA release are much more complex, possibly indicating a complex interaction between different neuronal populations. In slices of striatum, adenosine A_2A receptor agonists do not directly influence the release of dopamine or acetylcholine (Jin and Fredholm, 1997), but adenosine A_2A receptor stimulation has been shown to block the inhibitory effect of a dopamine D₂ receptor agonist on acetylcholine release from striatal slices (Jin et al., 1993). This could indicate that part of the dopamine D₂ receptor-mediated control of acetylcholine release from striatal slices is indirect and mediated via actions exerted at GABAergic neurons.

An increased neuronal activity is often accompanied by an expression of so-called immediate early gene (IEGs) such as c-fos, c-jun, junB, junD, NGFI-A (also called zif/268), and NGFI-B. Thus it is possible to determine which neuronal pathways are activated by caffeine by examining the effect of caffeine on immediate early gene expression. Caffeine causes a concentration-dependent increase in c-fos expression, which is confined to the striatum (Johansson et al., 1994). However, the increase does not become apparent until caffeine doses exceed 50 mg/kg, i.e., doses clearly higher than those required to elicit behavioral stimulation (see below). This could mean that the caffeine-induced increase in immediate early genes is related to the second phase of caffeine action, which involves a behavioral depression. Alternatively, the dose-response relationship could indicate that substantially higher concentrations are required to observe a generalized c-fos increase than are needed to activate a sufficient number of neurons to produce a behavioral stimulation. Some support for the latter contention is provided by the finding that other central stimulants, including amphetamine and cocaine, have to be given in very much higher concentrations to induce c-fos than to cause behavioral stimulation.

Because amphetamine and cocaine are known to act by releasing dopamine and because caffeine is presumed to act in part by increasing dopaminergic transmission, it is of interest to compare the effects of these three agents. A recent study revealed a gross morphological difference in the pattern of c-fos induction: cocaine and amphetamine increase the c-fos mRNA expression throughout the striatum, not least in the nucleus accumbens; caffeine, on the other hand, increases c-fos mRNA expression primarily in dorsolateral straitum (Johansson et al., 1994). Furthermore there is a marked difference at the cellular level. Amphetamine and cocaine primarily increase c-fos in the cells that express dopamine D₁ receptors and Substance P, but not in those that express D₂ receptors and met-enkephalin. By contrast, caffeine increases c-fos expression in both types of cells (Johansson et al., 1994). These data point to differences between the two types of agents, differences that could have some bearing on the question of whether caffeine is an addictive drug much like cocaine and amphetamine (Griffiths and Woodson, 1988a-c), but, as noted, the doses used to elicit IEG expression are high and not necessarily relevant in discussing behavioral stimulation.

The effect of an increase in the release of dopamine on IEG expression in the nucleus accumbens was directly studied by electrical activation of the medial forebrain bundle (Chergui et al., 1996). Burst activation of these neurons causes a marked increase in the evoked release of dopamine in the nucleus accumbens as assessed by voltametry. It is also associated with a marked increase in the expression of several IEGs, including c-fos, jun-B, NGFI-A, and NGFI-B. This increase can be blocked by dopamine D₁ receptor antagonists and is confined to the striatonigral neurons that express D₁ receptors. This shows that increased dopamine releasewhether brought about by nerve activation or by pharmacological meanscauses a D₁ receptor-mediated increase in IEG expression. Because high doses of caffeine increase c-fos both in D₁- and D₂-expressing neurons, the mechanism underlying its actions cannot be explained solely by an increase in dopamine.

As noted above, there are adenosine A₁ receptors on virtually all types of neurons that have the ability to decrease transmitter release. They are certainly present on the dopaminergic neurons (see Jin et al., 1993; Jin and Fredholm, 1997). However, they are also present on glutamatergic neurons. The striatum receives a strong glutamatergic input from both cortex and thalamus (see Gerfen, 1992), and part of the caffeine-induced increase in c-fos could be due to elevated release of glutamate. Indeed, at least part of the elevation in c-fos could be blocked by NMDA receptor antagonists (Svenningsson et al., 1996). Because the blockade was not complete, it is clear that additional mechanisms are also operative, but it may be relevant that the largest effect of NMDA receptor antagonism was observed in nucleus accumbens.

Caffeine injections lead to an increased expression not only of c-fos but also of other members of the same family of IEGs, notably c-jun and jun-B. Furthermore, there is an increased expression of the AP-1 transcription factor (Svenningsson et al., 1995b). Moreover, there are later changes in the expression of neuropeptides that are known to have AP-1-sensitive regulatory elements, notably preproenkephalin. These results suggest that even a single, albeit high, dose of caffeine can induce changes in gene expression that could lead to adaptive changes in the brain. The mRNA for four different neuropeptides, dynorphin, enkephalin, neurotensin/neuromedin, and Substance P, is elevated in the striatum by high doses of caffeine (Svenningsson et al., 1997a). Two of these, neurotensin/neuromedin and Substance P, are dependent on a rise in c-Fos as evidenced by the effect of a specific antisense oligonucleotide. By contrast, mRNA for dynorphin and enkephalin, were unaffected by blocking c-Fos increases, suggesting that other transcription factors are more important. Cyclic AMP response element-binding protein (CREB) is a likely candidate.

The high doses of caffeine used in these previous studies lead to a behavioral depression in experimental animals (see Daly, 1993). Therefore the induction of IEG expression might reflect behavioral depression rather than the behavioral stimulation that is the basis for the widespread human use of caffeine. It is known that the basal expression of mRNA for NGFI-A (Milbrandt, 1987) and NGFI-B (Milbrandt, 1988) is relatively high in striatum (Watson and Milbrandt, 1990; Schlingensiepen et al., 1991; Worley et al., 1991; Bhat et al., 1992). Several studies have shown that the striatal levels of NGFI-A mRNA can be regulated via dopaminergic transmission and an increase in the expression of the gene is seen following treatment with D₁ agonists, D₂ antagonists, and indirect dopamine agonists like cocaine and amphetamine (Cole et al., 1992; Moratalla et al., 1992; Nguyen et al., 1992). In addition, it has been reported that a significant reduction of NGFI-A mRNA occurs following chronic treatment with cocaine (Bhat et al., 1992).

Two recent studies examined the expression of mRNA for NGFI-A and NGFI-B in an attempt to reveal effects of low, behaviorally relevant doses of caffeine (Svenningsson et al., 1995a, 1997c). They showed that lower doses of caffeine (7.5-25 mg/kg) decrease the expression of mRNA for NGFI-A and NGFI-B in striatum (see Fig. 5). Indeed, the effect seen at the lowest dose was almost 75% of that maximally observed, suggesting that the threshold effect may be on the order of a few milligrams per kilogram. This may be the first evidence for direct neurochemical changes induced by such low, clearly stimulant doses of caffeine. As noted above, the only known biochemical action of caffeine, in the concentrations reached following administration of doses similar to those attained during normal human caffeine consumption, is blockade of adenosine receptors. Because the effect was most clear-cut in the striatum, where A_2A receptors are abundant, the data suggest that antagonism at adenosine A_2A receptors plays an important role in mediating the effects of caffeine. This is further supported by the finding that the caffeine-induced changes are located specifically to the striatopallidal neurons, which express A_2A receptors in high abundance. It has also been shown that the effect of a low dose of caffeine can be mimicked by the selective adenosine A_2A receptor antagonist SCH 58261, but not by the selective adenosine A₁ receptor antagonist 1,3-dipropyl-8-cyclopentylxanthine (DPCPX; Fig. 5) (Svenningsson et al., 1997c).

View larger version (41K): [in this window] [in a new window]   Fig. 5.   The effect of low doses of caffeine, of SCH 58261, and of DPCPX on NGFI-A expression in striatum and cortex. Left, the effect of increasing doses of caffeine (7.5, 15, or 30 mg/kg i.p.) on locomotion and rearing. Middle, the effect of caffeine on the expression of NGFI-A (measured in arbitrary optical density units; O.D.) in the same animals in the dorsal caudate putamen (upper panel-A), ventral caudate putamen (upper middle panel-B), nucleus accumbens (lower middle panel-C), and in several areas of cortex (lower panel-D). Right, the NGFI-A expression (in optical density units; O.D.) in the animals given the adenosine A2A antagonist SCH 58261 or the A1 antagonist DPCPX in the same brain regions. *p < .05; **p < .01; ***p < .001. Reprinted with permission from Elsevier Science [Svenningsson P, Nomikos GG, Ongini E and Fredholm BB (1997c) Antagonism of adenosine A2A receptors underlies the behavioural activating effect of caffeine and is associated with reduced expression of messenger RNA for NGFI-A and NGFI-B in caudate-putamen and nucleus accumbens. Neuroscience 79:753-764].

There is a parallelism between caffeine (or SCH 58261)-induced increase in locomotion and a decrease in the expression of the mRNA for some IEGs in the striatum (Svenningsson et al., 1995a, 1997c). The parallelism does not necessarily imply a direct causal relationship. Clearly the fall in mRNA cannot be the cause of the altered motor behavior since the latter occurred very rapidly. Conversely, the alteration in locomotor behavior is unlikely to cause the change in mRNA expression, because other drugs such as amphetamine cause an increase in locomotor behavior and an increase in the expression of mRNA for NGFI-A (Svenningsson et al., 1995a). The possibility exists, however, that the parallelism may be due to the fact that a single mechanism is the cause of a change both in mRNA and in locomotion after caffeine.

There is reason to believe that a reduction of intracellular levels of cyclic AMP is important for the observed decrease in the expression of the IEGs, because both NGFI-A and NGFI-B have CRE-like binding sites in their 5' flanking sequence (Watson and Milbrandt, 1989; Sheng and Greenberg, 1990). Adenosine A_2A receptors are coupled to G-proteins that activate adenylyl cyclase. By antagonizing the actions of adenosine at these receptors, caffeine would decrease intracellular cyclic AMP levels. Dopamine D₂ receptors are coupled to G_i-proteins, and decrease the levels of cyclic AMP. It should be pointed out that a D₂ agonist will be able to depress cyclic AMP formation only if there is a high basal rate of cyclic AMP generation. Adenosine acting on A_2A receptors is a probable mediator of such basal cyclic AMP generation (see Fig. 4).

These considerations focus the attention on the cyclic AMP system in the basal ganglia. Indeed, there is evidence that cAMP-dependent protein kinase is very important in the acquisition of cocaine self-administration and also in relapse into cocaine-seeking behavior (Self et al., 1998). As illustrated in Fig. 4, cAMP formation is stimulated via D₁ receptors in the GABAergic neurons of the direct pathway, whereas adenosine A_2A receptors mediate the important cAMP-raising signal in the neurons of the indirect pathway. Additional support for this scheme has recently been provided by the observation that D₁ agonists and A_2A agonists cause additive effects on striatal cAMP and on cAMP-dependent phosphorylation of DARPP-32 (Svenningsson et al., 1998a).

Bidirectional changes in gene expression following low and high doses of caffeine were also found for jun-B. The basal expression of jun-B is known to be relatively high in striatum (Mellstrom et al., 1991), and it has been reported that the expression of this IEG increases markedly following administration of a high dose of caffeine (Svenningsson et al., 1995b). Interestingly, it has been reported that cyclic AMP can regulate also the expression of jun-B (de Groot et al., 1991).

A working hypothesis is illustrated in Fig. 4. It is assumed that the level of cyclic AMP is important to determine the expression of mRNA for NGFI-A, NGFI-B, and Jun B in striatopallidal neurons. It is further assumed that the rate of cyclic AMP production is importantly controlled by adenosine, acting on A_2A receptors to stimulate adenylyl cyclase, and by dopamine, acting on D₂ receptors to inhibit the enzyme. In agreement with this basic hypothesis, the D₂ receptor agonist quinpirole was found to induce a marked reduction of the expression of mRNA for NGFI-A and NGFI-B. Quinpirole does not alter the expression of c-fos (Paul et al., 1992) unless c-fos expression is enhanced, e.g., by reserpine treatment (Cole and Di Figlia, 1994). Caffeine (7.5-25 mg/kg) had an effect of equal magnitude, and its effect was not clearly additive to that of quinpirole. Because the effect of caffeine was confined to the striatopallidal neurons, the data suggest that these neurons are the target also for quinpirole.

It is known that neuroleptic drugs with D₂ antagonistic properties cause a rapid and transient increase in IEG expression; this effect has been attributed to a removal of an inhibitory D₂ receptor tone (Robertson et al., 1992; Merchant and Dorsa, 1993). In one study (Svenningsson et al., 1998b), haloperidol was given with or without caffeine and the animals were sacrificed 30 min later. Under these circumstances the expected increase in IEGs was observed after the D₂ antagonist. The effect of the D₂ antagonist was reduced by caffeine in the dorsomedial striatum and nucleus accumbens, but it was increased in the caudal part of striatum (Svenningsson et al., 1998b). This suggests that caffeine not only acts on the basal ganglia neurons but also affects the striatal inputs. In another study (Svenningsson et al., 1995a) the D₂ receptor antagonist raclopride was given 4 h before sacrifice, and then there was no significant effect of the antagonist per se, probably because IEG levels had returned to control by this time. Furthermore, raclopride did not inhibit the depressant effect of caffeine given simultaneously with raclopride (Svenningsson et al., 1995a). These findings can be explained if adenosine and dopamine are both tonically active at their respective receptors. Thus, when dopamine receptors are blocked with raclopride, the stimulatory effect of adenosine is unhampered, leading to a transient increase in the IEG expression; when the adenosine receptors are also blocked, gene expression is brought down to essentially normal levels. The finding is less easy to explain if it is assumed that the major effect of adenosine A_2A receptor stimulation is to regulate signaling via the D₂ receptors. Thus, we have to assume that adenosine plays an important role in regulating gene expression in striatopallidal neurons that is independent of its established ability to influence the affinity of dopamine as an agonist at D₂ receptors (Ferré et al., 1992). The scheme in Fig. 4 also indicates that GABA release in the pallidum may be regulated by adenosine and dopamine in opposite directions, and as noted above, studies of GABA release support this proposal.

Given that adenosine acting on A_2A receptors is expected to increase the release of GABA in globus pallidus, caffeine is expected to decrease it. As a consequence of the decreased release of the inhibitory transmitter, caffeine is then also expected to increase activity in this brain area. This contention has been borne out in studies examining the expression of IEGs in globus pallidus following caffeine or selective adenosine A_2A receptor antagonists (Le Moine et al., 1997; Svenningsson and Fredholm, 1997). Furthermore, there are important synergistic effects of adenosine A_2A receptor antagonism and stimulation of dopamine D₁ receptors (Pinna et al., 1996; Le Moine et al., 1997). It is also of potential relevance that the human adenosine A_2A receptor gene has been linked to a potential schizophrenia locus on chromosome 22 (Deckert et al., 1997). If this tentative identification holds up, the link between adenosine and dopamine-related functions would be strengthened.

The link between adenosine A_2A receptors and dopamine-related effects in the striatum is further supported by the finding that a selective adenosine A_2A receptor agonist, 2-[(2-aminoethylamino)carbonylethylphenylethylamino]-5'-N- ethylcarboxamido adenosine (APEC), can antagonize the motor stimulant effects of amphetamine. The A_2A agonist also reduced the effects of amphetamine on c-Fos in nucleus accumbens core and shell (Turgeon et al., 1996). These authors also demonstrated effects of an A₁ receptor agonist and thus supported several previous reports that not only A_2A but also A₁ receptors are important in the regulation of striatal function (Ferré et al., 1997). Indeed, blockade of adenosine A₁ receptors has been shown to potentiate the motor stimulation afforded by a dopamine D₁ receptor agonist (Popoli et al., 1996b). Conversely, stimulation of A₁ receptors blocks the EEG arousal afforded by D₁ receptor stimulation (Popoli et al., 1996a).

	IV. Actions of Caffeine on Brain Functions and Behavior

Top Previous Next References

Having discussed the molecular and neuronal actions of caffeine, especially as they relate to a primary effect on adenosine receptors, it is important to consider some actions at a more integrated level. Even though the primary action of caffeine may be to block adenosine receptors this leads to very important secondary effects on many classes of neurotransmitters, including noradrenaline, dopamine, serotonin, acetylcholine, glutamate, and GABA (Daly, 1993). This in turn will influence a large number of different physiological functions. It would clearly be outside the scope of this review to cover all aspects of caffeine action in the CNS. Nonetheless, some specific aspects need to be brought forward as they relate directly or indirectly to the issue at hand. Below we will briefly consider a set of such responses and attempt to relate them to the primary actions of caffeine. Finally, we will briefly comment upon the similarities and dissimilarities between caffeine and known addictive drugs such as cocaine, morphine, and nicotine.

The interaction between adenosine A_2A and dopamine D₂ receptors highlighted above could provide a mechanism for several actions of caffeine and some of its metabolites on dopaminergic activity. Thus, an inhibition of A_2A receptors by caffeine would be expected to increase transmission via dopamine at D₂ receptors (Ferré et al., 1992). There is indeed ample evidence that caffeine (and other adenosine receptor antagonists) can increase behaviors related to dopamine. The first demonstration of an adenosinedopamine interaction on behavior was the finding that several adenosine receptor antagonists, including caffeine, theophylline, and isobutyl-methylxanthine, could increase dopamine receptor-activated rotation behavior (Fredholm et al., 1976). This finding was preceded by the observation that theophylline could enhance such rotation behavior (Fuxe and Ungerstedt, 1974), but in that study the authors proposed that the mechanism was phosphodiesterase inhibition. In the later study (Fredholm et al., 1976) this possibility was discounted. This type of finding has since been repeatedly confirmed and elaborated (see Daly, 1993; Ferré et al., 1992; Ongini and Fredholm, 1996). Indeed, dopamine receptor antagonists can inhibit the stimulatory effects of caffeine on motor behavior (Fredholm et al., 1983; Herrera-Marschitz et al., 1988; Garrett and Holtzman, 1994b), and long-term treatment of rats with caffeine reduces the effects of both caffeine and dopamine receptor agonists (Garrett and Holtzman, 1994a).

Besides the direct effects on striatopallidal neurons mediated via an antagonism of A_2A receptors, caffeineat least at high doseshas been reported to influence the turnover of dopamine [for review see Nehlig and Debry (1994)]. Adenosine A₁ receptors (in contrast to adenosine A_2A receptors) have been shown to influence dopamine release in slices of the striatum (Jin et al., 1993; Jin and Fredholm, 1997). Caffeine has been reported to cause a dose-dependent (30-75 mg/kg) increase in dopamine in the striatum (Morgan and Vestal, 1989). In that study electrochemistry was used, which presents a potential problem since caffeine itself appears to influence the response of the recording electrode (F. Gonon, personal communication). In a recent study, microdialysis techniques were used to study this question (Okada e