Participants

Forty‐five adult volunteers (25 women, aged 18–50 years) participated in this study between 2005 and 2007. The Institutional Review Board of Wake Forest University School of Medicine approved this study, and all participants gave informed consent to the study procedures. Potential volunteers were recruited from the Winston‐Salem, North Carolina area with flyers, newspaper advertisements, and by word of mouth. Respondents who met the following inclusion criteria were invited to participate: (1) no history of migraines, diabetes, stroke, hypertension, or any neurologic or vascular disease, (2) no current symptoms of depression or an anxiety disorder, (3) no previous brain surgery or serious CNS trauma, (4) no use of vasoactive medications, (5) no current abuse of alcohol or illicit drugs, (6) visual acuity able to be corrected to 20/40, and (7) no MR incompatible implanted devices. Participants’ average daily caffeine intake was estimated from their responses to a 7‐day caffeine consumption diary (7‐day CCD; Addicott et al.,2008) modified from the Caffeine Consumption Questionnaire [Landrum,1992] and from data on the caffeine content of common beverages (Center for Science in the Public Interest web site, http://www.cspinet.org, published July 1997). Participants were then identified by their daily caffeine use as “low users” (n = 18; < 200 mg/day), “moderate users” (n = 9; 200–600 mg/day), and “high users” (n = 18; > 600 mg/day). Although there is no currently accepted cut offs for categorizing caffeine use levels, group divisions used here are consistent with previous work [Schuh and Griffiths,1997; Tinley et al.,2003]. In addition, we present analyses that correlate caffeine use with CBF to supplement findings based on these groupings. The average daily caffeine use (mg/day) recorded in the participants’ 7‐day CCD was converted into body‐weight adjusted caffeine use (mg/kg/day) using participants’ weight taken at the first visit.

Participant characteristics appear in Table ​I. There were expected differences between groups in daily caffeine use (mg/day) (one‐way analysis of variance, ANOVA: F(2,42) = 146.90, p < 0.001; Bonferroni t‐tests: low < moderate < high, p < 0.001) and in body‐weight adjusted caffeine use (mg/kg/day) (F(2,42) = 54.17, p < 0.001; t‐tests: low < moderate, p < 0.05; low and moderate < high, p < 0.001). Differences between groups in weight (1‐way ANOVA), gender, race, cigarette use, and oral birth control use (χ² tests) were not significant. Groups differed slightly in age (F(2,42) = 6.56, p < 0.005) with low users being younger than high users (p< 0.005). Since age correlated with some measures of CBF, it was used as a covariate in CBF analyses.

Study Design

Each participant underwent quantitative perfusion imaging on four separate occasions in a randomized, double‐blind study design. Test sessions were at least 1 week apart to allow participants to resume normal caffeine use between scans. Two of these scans followed participants’ normal caffeine use (native state) and two followed abstention from caffeine (abstained state). For the native state, participants were instructed to continue their normal daily caffeine usage until 15 min before their scheduled visit. For the abstained state, participants were instructed to avoid consuming any foods, beverages, or drugs that contain caffeine for 30 hours before their scheduled visit. This duration of abstinence was selected so that caffeine concentrations would most likely be below detection threshold (0.2 μg/ml) at the time of testing. Caffeine use or abstention from use during the 30 hours before a scan visit was verified with a 3‐day CCD. In addition, a saliva sample was obtained upon arrival to the scan visit to verify compliance with study requirements. Saliva samples were collected using Salivettes® (Sarstedt, Inc. Newton, NC), and caffeine concentrations were assayed by HPLC (Global Lifescience Solutions, LLC, Ann Arbor, MI).

Following the first saliva sample, participants ingested a capsule containing placebo or caffeine (250 mg, anhydrous). A second saliva sample was obtained 1 hour after drug administration to measure the varied levels of caffeine in each participant. Respiration rate, end‐tidal CO₂, oxygen saturation (Capnocheck® II, SIMS BCI Inc., Waukesha, WI), heart rate, and blood pressure (IVAC vital check model 4410, ALARIS medical systems, Dublin, OH) were also recorded when the saliva samples were collected. Immediately following collection of the second saliva sample, participants were placed in the MRI scanner. Structural, functional (results to be reported elsewhere), and perfusion imaging were completed. The perfusion scan was collected approximately 1.5–2 hours after drug administration. During the perfusion sequence, participants were instructed to maintain their gaze on a fixation cross (E‐Prime, Psychology Software Tools, Pittsburgh, PA) delivered through MR‐compatible virtual reality goggles (Resonance Technology Inc., Los Angeles, CA). Participants’ wakefulness throughout the scan was monitored via an infrared camera embedded in the goggles. Participants were scanned within the 3 hours of their peak caffeine use as determined by the 7‐day CCD.

Image Acquisition

Imaging experiments were performed on a 1.5 T GE scanner (Twin‐speed; GE Healthcare, Milwaukee, WI) with a 4‐channel neurovascular MR coil array (MEDRAD, Inc. USA, Warrendale, PA). The imaging protocol consisted of a high‐resolution anatomical image for tissue classification and a perfusion‐weighted image to measure CBF.

A T₁‐weighted structural scan was acquired using an Inversion Prepared 3D Spoiled Gradient Echo; IR‐3DSPGR [Mugler and Brookeman,1990]. The parameters for this scan were as follows: TE = 2.60 msec, TR = 9.00 msec, TI = 600 msec, bandwidth = 16.93 kHz, flip angle = 20 degrees, field of view = 240 mm (frequency) × 240 mm (phase), and an acquisition matrix of 256 (frequency) × 256 (phase), 124 slices, 1.5 mm thickness, 0 mm slice gap, and an in‐plane resolution of 0.94 mm.

CBF was measured with QUantitative Imaging of Perfusion using Single Subtraction with Thin Slice TI₁ Periodic Saturation: QUIPSS II TIPS, also known as Q2TIPS [Luh et al.,1999] with Flow‐sensitive Alternative Inversion Recovery (FAIR) encoding [Kim and Tsekos,1997]. Images were acquired with a single shot gradient echo Echo Planar Imaging (EPI) sequence [Mansfield,1977]. To improve perfusion sensitivity by minimizing slice imperfections [Frank et al.,1997; Yongbi et al.,1999], a C‐shaped frequency offset corrected inversion pulse (β = 1361, μ = 6) [Ordidge et al.,1996] was used for tagging the blood. Very selective saturation (VSS) pulses were used [Tran et al.,2000] and applied every 25 msec between 800 ms (TI₁) [Wang et al.,2002] and 1,200 msec (TI_1s) to minimize the uncertainty of the labeled blood’s transit time to the imaging slice [Wong et al.,1998a]. Three VSS pulses were also applied immediately before and after the inversion pulses to suppress tissue signal in the imaging plane. The Q2TIPS‐FAIR sequence consisted of 80 alternating slice‐selective and non‐selective radiofrequency inversion pulses (label/control pairs) to improve the signal to noise ratio (imaging time 8 min 36 seconds). These label/control pairs are pair‐wise subtracted then averaged to generate a perfusion‐weighted image. The first 36 seconds (10 volumes with an initial 6 second quiescent delay) were used to establish steady state. After the initial 6 seconds, a single‐shot EPI proton density (M
₀) image was acquired. The mean white matter signal from the M₀ image was used to scale perfusion‐weighted images to accurately calculate absolute quantitative CBF maps according to the general kinetic model [Buxton et al.,1998]. Immediately after the EPI 90 degree excitation RF pulse, a bipolar diffusion gradient with an equivalent b value of 5.25 mm²/sec was used to suppress intraarterial spins [Yang et al.,1998]. The 10 oblique slices were parallel to the AC/PC line (2 slices below and 8 slices above the AC/PC line, 8 mm thickness, 0 mm slice gap) and were acquired inferior to superior. Additional parameters of interest for the Q2TIPS‐FAIR‐EPI sequence used in this experiment are as follows: TE = 30.4msec, TR = 3,000 sec, TI = 2,000 msec, bandwidth = 62.5 kHz, flip angle = 90 degrees, field of view = 240 mm (frequency) × 240 mm (phase), and an acquisition matrix of 64 (frequency) × 40 (phase).

Data Analysis

Perfusion data were preprocessed using SPM99 software (Wellcome Department of Imaging Neuroscience, Institute of Neurology, London). The reconstructed control and label images were motion corrected with a six‐parameter rigid body transformation using SPM99. After motion correction, control/label images were pair wise subtracted; the different images were averaged; and quantitative perfusion maps were calculated from the equation:

where CBF is the cerebral blood flow, ΔM(TI₂) is the mean difference in the signal intensity between the label and control images, M
_0,blood is the equilibrium magnetization of blood, α is the tagging efficiency, TI₁ is the time duration of the tagging bolus, TI₂ is the inversion time of each slice, and T
_1,blood is the longitudinal relaxation time of blood, q
_p is a correction factor that accounts for the difference between the T
₁ of blood and the T
₁ of brain tissue [Wong et al.,1998a]. For this study, the correction factor q
_p was assumed to be 1_, which is a reasonable approximation when the T
₁ of blood and the T
₁ of brain tissue are similar. Based on the literature, the T
₁ of blood was assumed to be 1,200 ms at 1.5T [Simonetti et al.,1996]. The inversion efficiency was measured to be 0.95 from another experiment (data not shown). The M
_0,blood is approximated from the M
_{0,white matter}, which is measured directly from the M
₀ image acquired with the perfusion weighted images [Wong et al.,1998a].

Tissue probability maps were created using an SPM99 automated segmentation of the T
₁‐weighted structural scan. The T
₁‐weighted image was coregistered to the M
₀ image and the same transformation was applied to the tissue probability maps. The coregistered tissue probability maps were used to create a white matter mask (81% probability) and a gray matter mask (51% probability). Then the average signal intensity of white matter was calculated. The signal intensity of white matter and global scaling factors were applied to convert the perfusion‐weighted image to a quantitative CBF map that represents the magnitude of perfusion (ml/100 g tissue/min) for each voxel. Mean gray matter CBF values were calculated by averaging the perfusion values from all the voxels contained in the gray matter mask. Images were normalized within SPM5 by warping the M
₀ to the SPM EPI template. The transformation was then applied to the perfusion maps. Perfusion maps were visually inspected, and no errors were detected. The normalized images were used to perform voxel‐wise comparisons, which showed no regionally‐specific changes after correcting for global blood flow changes. These results are not depicted, but the mean normalized images are shown in the figures to demonstrate perfusion changes.

Primary analyses were conducted using univariate analyses of variance (ANOVA). The pre‐ and post‐drug salivary caffeine concentrations were analyzed separately with repeated‐measures 2 (state) × 2 (drug) × 3 (caffeine use) ANOVAs. The same analyses were performed on pre‐ and post‐drug measures for respiration rate, end‐tidal CO₂, O₂ saturation, heart rate, and blood pressure. Potential interactions between gender and gray matter CBF were investigated with a 2 (state) × 2 (drug) × 2 (sex) ANOVA. There were no significant interactions between gender and any of the main effects or interactions; therefore, the genders were combined in further analyses. The signal to noise ratio of the M
₀ images were analyzed with a 2 (state) × 2 (drug) × 3 (caffeine use) repeated‐measures ANOVA, and no significant effects were found. Gray matter CBF was analyzed with a repeated‐measures 2 (state) × 2 (drug) × 3 (caffeine use) analysis of covariance (ANCOVA) using age as the covariate. The difference from the mean age was used instead of the raw value, to avoid interfering with the main effects [Delaney and Maxwell,1981]. Difference measures in CBF between caffeine states and drug conditions were analyzed with a 2 (state) × 3 (caffeine use) and a 2 (drug) × 3 (caffeine use) ANCOVA, respectively. Follow‐up analyses included two‐tailed independent‐samples t‐tests to measure between‐caffeine use group effects, and paired‐samples t‐tests to measure within‐ caffeine use group effects. Correlations were performed using two‐tailed Pearson’s product‐moment correlation. Analyses were conducted with SPSS version 15.0 (SPSS Inc., Chicago, IL) with alpha set to 0.05. CBF images were created with MRIcro v1.40 [Rorden and Brett,2000].

Salivary Caffeine Concentrations

Pre‐drug salivary caffeine concentrations showed participant compliance with the study requirements (see Table ​II). Caffeine concentrations were near zero in the abstained state and were not different between the caffeine use groups. In the native state there were differences in caffeine concentrations between caffeine use groups (state x caffeine use interaction effect [F(2,42) = 18.26, p < 0.001], Between‐subjects' Caffeine Use effect [F(2,42) = 18.53, p < 0.001]). Since there was no effect of drug condition, caffeine concentrations were averaged from the two native state measures and significant differences were found between caffeine use groups (low < moderate [t(25) = 3.71, p < 0.05]; low < high [t(34) = 5.92, p < 0.001]; moderate < high [t(25) = 2.35, p < 0.05].

As expected, post‐drug salivary caffeine concentrations increased from pre‐drug levels in the caffeine conditions, but not in the placebo conditions (drug effect [F(1,42) = 189.08, p < 0.001]). Post‐drug caffeine concentrations were greater in the native state than in the abstained state and depended on caffeine use group (state x caffeine use interaction effect [F(2,42) = 11.86, p < 0.001]). Differences between caffeine use groups were observed in the NP condition (low < moderate [t(25) = 4.30, p < 0.001], low < high [t(34) = 4.58, p < 0.001]). The state x drug interaction did not reach significance, but there were greater caffeine concentrations in the NC than the AC condition for moderate (t(8) = 3.21, p < 0.05) and high users (t(17) = 4.01, p < 0.005).

There were no main effects or interactions in the pre‐ and post‐drug measures of respiration rate, end‐tidal CO₂, or O₂ saturation, nor did any of these post‐drug measures correlate with CBF in each of the four state × drug conditions. The average pre‐drug levels were: respiration rate: 16.1 ± 3.5 breaths/min, end‐tidal CO₂: 36.6% ± 4.5%, and oxygen saturation: 97.1% ± 1.1%. The average post‐drug levels were: respiration rate: 15.6 ± 3.4 breaths/min, end‐tidal CO₂: 35.9% ± 5.2%, and oxygen saturation: 97.3% ± 1.2%. There were expected changes in heart rate and blood pressure. Following caffeine administration, heart rate decreased by an average of 8 beats/min from a resting rate of 74 beats/min, and blood pressure increased by an average of 3.0/2.4 mmHg from a mean resting value of 121/70 mmHg. These changes are similar to previously published reports on the effects of caffeine on blood pressure [Nurminen et al.,1999].

Cerebral Perfusion

Caffeine reduced gray matter CBF by 27% to approximately 60 ml/100 g of tissue/min across all caffeine users, although the decrease was larger in the abstained state (33%) than in the native state (20%; see Fig. ​1). The effects of caffeine on CBF depended on the caffeine state (F(1,41) = 21.81, p < 0.001), drug condition (F(1,41) = 190.73, p < 0.001), and the level of caffeine use (state x drug x caffeine use interaction effect [F(2,41) = 8.26, p < 0.005]). Caffeine reduced CBF within each caffeine use group in both the abstained state (low [t(17) = 7.85, p < 0.001], moderate [t(8) = 5.78, p < 0.001], and high [t(17) = 8.88, p < 0.001]) and the native state (low [t(17) = 6.77, p < 0.001], moderate [t(8) = 4.36, p < 0.005], and high [t(17) = 3.13, p < 0.05]). Differences in CBF between caffeine users in the AP and NP conditions appear to be driving the interaction effect. The low users exhibited virtually no difference between these conditions, whereas there was a considerable increase in CBF in the AP condition relative to the NP condition among the moderate users (t(8) = 4.78, p < 0.005) and high users (t(17) = 5.90, p < 0.001). Compared with the low users, high users had greater CBF in the AP condition (t(34) = 2.46, p < 0.05) and had a trend for greater CBF in the AC condition (t(34) = 1.97, p = 0.057); high users also had a trend for less CBF in the NP condition (t(34) = 1.67, p = 0.104). The difference between moderate and low users in the AP condition approached significance as well (t(25) = 1.91, p = 0.068). Potential contributions of other between‐group differences to the 3‐way interaction effect were investigated with further analyses.

To investigate how the caffeine states contributed to the state x drug x caffeine use interaction effect, differences in CBF between drug conditions in each state (abstained placebo − abstained caffeine, AP‐AC; native placebo − native caffeine, NP‐NC) were calculated (see Fig. ​2). The effects of caffeine on CBF differed between caffeine users in the native state but not in the abstained state (state x caffeine use [F(2,41) = 8.26, p < 0.005]). In the native state, high users exhibited a smaller caffeine‐induced reduction in CBF than low users (t(34) = 3.53, p < 0.005) and there was a trend for high users to have a smaller reduction than moderate users (t(25) = 2.06, p = 0.050). There was also a greater reduction in CBF following caffeine in the abstained state than in the native state for moderate (t(8) = 5.90, p < 0.001) and high users (t(17) = 5.36, p < 0.001). Figure ​3 illustrates the average whole brain CBF in the NP and NC conditions. The reduced CBF following caffeine administration is most evident in the low user group.

Since the quantification of caffeine use groups was arbitrary, correlational analyses were conducted between changes in CBF and body‐weight adjusted caffeine use (mg/kg/day). Caffeine use was normalized by taking the square root of the usage in mg/kg/day to produce a more uniform distribution along the abscissa. There was a negative correlation between caffeine use and the difference between placebo and caffeine conditions in the native state (NP‐NC; r = −0.48, p < 0.005; see Fig. ​4). Correlations between the percent change and caffeine use were similar (r = −0.48, p < 0.005).

To investigate how the drug conditions contributed to the state x drug x caffeine use interaction effect, difference in CBF between states for each drug condition (abstained placebo − native placebo, AP‐NP; abstained caffeine − native caffeine, AC‐NC) were calculated (see Fig. ​5). Differences in CBF varied within the placebo condition, but not within the caffeine condition (drug x caffeine use [F(2,41) = 8.26, p < 0.005]), and the effects of caffeine on CBF were dependent on the caffeine use group (between‐group effect [F(2,41) = 5.72, p < 0.05]). There was a greater difference in CBF between placebo conditions in the high users compared to the low users (t(34) = 4.23, p < 0.001) and in the moderate users compared to the low users (t(25) = 2.66, p < 0.05). Additionally, there was a greater difference in CBF in the AP‐NP condition than in the AC‐NC condition among moderate (t(8) = 5.90, p < 0.001) and high users (t(17) = 5.36, p < 0.001). Figure ​6 illustrates the average whole brain CBF for the three groups in the AP and NP conditions. The increased CBF in the high users following caffeine abstention compared to the low and high users’ native state is evident. There was a positive correlation between caffeine use and the difference between abstained state and native state during the placebo condition (AP‐NP; r = 0.62, p < 0.001; see Fig. ​7). Correlations between the percent change and caffeine use were similar (r = 0.60, p < 0.001).

Correlations between caffeine use and CBF in each of the four state by drug conditions were also performed. Similar to the previous study by Field et al. [2003], there was a significant relationship between CBF and caffeine use in the AP condition (r = 0.42, p < 0.005). A correlation was also found in the AC condition (r = 0.31, p < 0.05). This relationship did not reach significance in the native state. There were no correlations between CBF and salivary caffeine concentrations in any of the state by drug conditions.