Evaluating potential interactions when medications are ingested by individuals increasing their physical activity is becoming more important given the rise in drug prescriptions (OECD, 2019) and exercise recommendations (Task Force on Community Preventive, 2002). Previous drug-exercise investigations have focussed on exercise-induced hemodynamic changes on drug pharmacokinetics (PK) (Sweeney, 1981), yet drug binding in blood remains unstudied. This is puzzling since exercise is associated with physicochemical and biochemical changes in the blood, with temperature and acidity being major factors that both affect drug binding and, thus, its efficacy. Therefore, a patient given prescriptions for both exercise and medications may experience unintended drug effects. Examining exercise’s effect on drug-plasma protein binding will add to our repository of drug interactions for personalized medicine and provide insight into exercise as a modality to elicit specific physiological changes that can predictably change a medication’s effect.

Drugs travel in the blood in two forms—that which is bound to plasma proteins, known as the bound fraction (f _b ), and that which is not bound to plasma proteins, known as the free/unbound fraction (f _u ). The degree and relative affinity of drug binding to plasma proteins (e.g., albumin, α1-acid glycoprotein [AGP], and lipoprotein) is based on the drug’s physicochemical properties, with basic drugs binding well to AGP, and acidic and neutral drugs preferentially binding to albumin (Brown et al., 2010). Since the f _b is bound to proteins, it is not available to bind to its receptor and is, thus, rendered inactive. In contrast, the f _u is free in the circulation and can: 1. bind to its receptor to elicit a pharmacological response; 2. move to extravascular tissues, increasing the drug’s volume of distribution (V_d); and 3. interact with enzymes and transporters, increasing the rate of drug metabolism and excretion (Roberts et al., 2013). For example, a 9.7% increase in the %f _u of the antibiotic, ertapenem, resulted in a 112% increase in its clearance (CL) and 160% increase in its V_d (Burkhardt et al., 2007). Thus, drug binding is highly relevant to understanding both drug PK as well as pharmacodynamics (PD).

All drugs have an inherent affinity for plasma proteins which, in exception to pregnancy and aging, remains relatively constant for healthy individuals under normal physiological conditions (Kremer et al., 1988). However, disruption of the drug-protein complex through competitive binding with exogenous substances and conformational changes from physiological perturbations can alter the degree of drug-protein binding. Competitive binding is commonly observed with drug-drug interactions, whereby drug A will competitively bind to a similar binding spot occupied by drug B, effectively disrupting its interaction and increasing drug B’s f _u . For example, valproic acid displaces phenytoin from a shared albumin binding spot because of its higher association constant (McElnay and D’Arcy, 1983). Conformational changes to proteins from endogenously-generated perturbations can affect ligand-binding sites, thus, affecting drug-protein affinity. For example, in vitro studies give evidence that changes in blood pH and temperature can affect a drug’s f _u . In pooled human serum spiked with propranolol (a beta blocker with %f _u < 10%), a decrease in pH from ∼7.6 to ∼6.6 was associated with an increase in %f _u from 6% to 10.5% (Paxton and Calder, 1983). The same study showed that an increase in temperature from 25°C to 30°C was associated with a ∼1.5% increase in propranolol’s %f _u (Paxton and Calder, 1983). This effect is rather common. A review of in vitro studies investigating >20 compounds revealed that all of the basic drugs investigated experience an increase (e.g., ranging from 3.5% [practolol] to 136% [fentanyl]) in f _u when pH is decreased, while acidic drugs exhibit unpredictable changes in f _u (e.g., ranging from −15% [etopside] to 95% [tenoxicam]) (Hinderling and Hartmann, 2005). The therapeutic significance of alterations in fu of these magnitude are maximized in drugs that: 1. are highly protein bound, 2. have a high intrinsic clearance (CL_int); or 3. are not titrated to the pharmacological response (Roberts et al., 2013). The first point is exemplified with (R)-warfarin, a commonly prescribed oral anticoagulant that has a plasma protein binding of 99%, which, if decreased to 97%, for example, would result in a 2-fold increase in f _u , which would affect V_d and CL.

Mechanistically, increasing the temperature results in increased enzyme activity, known as the Q₁₀ effect, up until a threshold where the enzyme’s tertiary (i.e., globular) and secondary structures are denatured (Sinha et al., 2008). Changes in pH will alter the degree of ionization and the resulting charge on amino acid side chains (Baler et al., 2014). In the case of albumin, deviating the pH from 7.4 decreases α-helical and increases β-sheet structures, with alkalinic conditions (e.g., pH = 9) reducing overall charge and promoting the N-B (normal-to-base) transition and acidic conditions resulting in the N-F (normal-to-fast) transition at pH = 4 (Peters and Peters, 1995). These conformational changes are reversible up to the extremes of pH < ∼2 and > ∼9, beyond which disulfide bonds are cleaved and, in basic conditions, tyrosyl hydroxyl groups are deprotonated (Peters and Peters, 1995). These changes can be further accelerated by fluctuations in calcium content, urea, and temperature (Peters and Peters, 1995), which are all increased in some capacity during exercise (Foran et al., 2003), and, cumulatively, may promote the conformational changes to albumin within the physiological pH range. For example, the N-B transition at higher pH values is augmented with increasing calcium concentration, which, in the case of warfarin, led to higher binding with albumin (Wilting et al., 1980).

The normal blood pH range is 7.38–7.44, yet exercise above the anaerobic threshold results in the formation and immediate dissociation of lactate to its ion (La⁻) and H⁺, which can decrease blood pH to 6.8 during exhaustive exercise (McArdle et al., 2007). Exercise also increases metabolic heat production, elevating core body temperature (T_c), with values of >41°C being reported in athletes during competition (Byrne et al., 2006). Given that in vitro experiments have characterized changes in drug f _u resulting from changes in the pH and temperature of a drug’s medium, and exercise is associated with changes in blood pH and increased temperature, it is reasonable to speculate that exercise has the potential to affect a drug’s f _u . The current investigation is the first to assess in humans the effects of acute exercise on a drug’s f _u . Because of the potential risk for aberrant drug effects, we chose to study these interactions using caffeine, the world’s most consumed drug that is toxic at doses ∼30-fold higher than the average daily intake (Nawrot et al., 2003). Moreover, any change in its %fu is likely to be reflected in its efficacy, which will be of interest to the ∼75% of professional athletes who ingest caffeine as an ergogenic aid (Aguilar-Navarro et al., 2019). An increase in the %fu may cause a stronger ergogenic effect, which may engender recommendations for a lower dose while a decreased %fu may engender the opposite.

Caffeine is a weak base that has been shown to bind to albumin (Wu et al., 2009); yet, information about its binding to AGP is scant. After being absorbed within 45 min, caffeine distributes throughout well-hydrated tissues with a V_d of 0.5–0.8 L/kg and is primarily (90%) metabolized by the enzyme, CYP1A2, to paraxanthine (84% of CYP1A2 metabolites), which makes caffeine a commonly used probe drug to phenotype 1A2 activity (Nehlig, 2018). Caffeine experiences relatively low protein binding (%f _u = 70–90%) (USIM, 2001) and few studies have evaluated changes in its binding capacity. Comparison of caffeine f _u between young (25–30 years) and elderly (66–78 years) individuals using CF-25 ultrafiltration ‘cones’ revealed that age is associated with a decrease in plasma albumin concentration (45.9–41.13 g/L) but is not associated with a change in %f _u (Blanchard, 1982a). Another study using fluorescence spectroscopy to evaluate albumin-caffeine interactions found a decrease in the apparent binding constant (K _b ) and number of binding sites with increasing temperatures, giving evidence that the albumin-caffeine complex may dissociate at higher temperatures and increase the f _u of caffeine (Wu et al., 2009). Therefore, the purpose of the current study was to clarify if the f _u of caffeine is influenced by exercise-mediated changes in body temperature and blood pH changes in healthy humans. The study was designed to test the hypothesis that increases in body temperature due to passive heating would increase f _u , and that high intensity exercise—but not low intensity exercise—would further increase f _u via decreases in blood pH.

The f _u of a drug determines many PK and PD properties that can dictate its efficacy and potential for side effects. This is the first study, to our knowledge, to investigate in vivo changes in blood temperatures and pH on the f _u of a drug using ultrafiltration. An experimental protocol was designed with the intent of eliciting changes in T_c and blood pH, which are normal physiological responses to acute exercise. The results demonstrate that the intervention protocol was successful in raising T_c and decreasing blood pH compared to a non-exercise, control trial. However, the hypothesis that the intensity and duration of exercise used in the current study would lead to changes in T_c and blood pH that would affect f _u was not supported.

We used a combination of passive heating and various intensities of cycling to induce physiological changes in T_c and blood pH. T_c showed a gradual increase over the course of the interventions in the exercise trial with the greatest relative increase occurring from passive heating and the highest absolute temperature peaking at 38.37°C during HIIT. Even with this profound 1.37°C increase in T_c relative to BASE, and the swiftness of our research personnel to record blood temperature after collection in <1 min, blood temperature changed in the opposite direction to T_c but was still statistically slightly higher than time-matched results during the Rest trial after SS (+0.82°C) and REC (+0.90°C). Steady-state cycling was successful in maintaining a consistent blood pH while increasing T_c, whereas the HIIT protocol effectively decreased blood pH while further increasing T_c compared to the Rest trial. Despite these pronounced changes, the plasma concentrations of caffeine and paraxanthine showed no difference between trials when corrected for PV fluctuations. Accordingly, the MR of CYP1A2 did not change during exercise, indicating that over the measurement period heat stress and exercise did not change the enzymatic rate of CYP1A2. Similarly, C_u did not differ between trials, resulting in no difference in the %f _u of caffeine or paraxanthine. Our resting data shows a %f _u of 89% (SD = 0.04%), which is at the upper end of the reported %f _u (70%–90%) (USIM, 2001), greater than the 64.55% reported using another ultrafiltration device (Blanchard, 1982b), and comparable to a study using the same ultrafiltration tubes (Lammers et al., 2018). Thus, our mean values appear acceptable; however, the same cannot be said for the variability, which might be masking potential differences. Of similar concern is that some of the %f _u values in our study were >100%, which is not physiologically possible. This, however, is consistent with other studies have also reported %f _u values above 100%. A major analysis of 25 epileptic medications showed mean unbound fractions of 102.7% for both gabapentin and pregabalin (Patsalos et al., 2017), which is slightly above their reported %f _u of 99% and 97%, respectively. This is obviously a smaller margin of error than the current study, yet still highlights the variability that is possible when deriving a value (e.g., %f _u ) using multiple mass spectrometry data. Nonetheless, there could be other factors that should be considered. For example, the use of protein precipitation in the sample preparation could be interfering with the caffeine concentration in C_t. Although we did not validate the currently used assay under the full measures of the FDA guidelines (FDA, 2018), a protein precipitation method very similar to the one used in the current investigation has been validated with great % recoveries of caffeine and its metabolites recently (Chen et al., 2017). It is possible that the slight difference in reagents used (acetonitrile vs. formic acid (Chen et al., 2017)) resulted in differences in drug precipitating with protein, leading to the aberrantly low C _t values and resulting in %f _u >100%. Additionally, a long spin time might also artificially inflate the C _u as previously reported with valproic acid (Liu et al., 1992). This could result from perpetual flux from C _b to C _u according to the law of mass action as C _u is continually being spun through the filter and a true equilibrium is never attained. Findings by Kratzer et al. (2014) contradict this by showing no effect of centrifugation time on vancomycin, which they attribute to the constancy of K _b , and that free protein and drug-bound protein are both concentrated in the sample reservoir at the same rate. Regardless, since all of the samples were processed similarly, a systemic effect on all the samples would have been apparent, which is not the case. This study highlights the need for standardized ultrafiltration methodology.

The cumulative thermal load (+1.37°C over BASE) during EXS in the current study is higher than the +0.9°C increase (36.9 → 37.8°C) reported previously by Kelly and colleagues (Kelly et al., 2016), who employed a similar protocol that had a 3 min warm-up at 60% V ˙ O 2 ⁡ max , followed by fifteen high-intensity intervals (work:rest) of 30:30 s sprints at 90%:30% V ˙ O 2 ⁡ max , respectively. Comparing the increases from exercise alone, however, reveals that their protocol elicited a slightly more elevated response in T_c (+0.9 vs. +0.8°C), which is likely a combination of their longer duration (7.5 vs. 3.3 min) of high intensity (i.e., 90% V ˙ O 2 ⁡ max ) cycling and the higher T_c starting point of our participants who had just undergone passive heating (HEAT). Further comparisons show that our values are more in line with rectal temperatures reported by Reeve and colleagues (Reeve et al., 2019) (38.17°C), who performed a similar protocol of 10 min of passive heating, followed by a 6 min warm-up at 50% of work max (W_max; watts), and then 12 × 1-min intervals at 100% W_max with rests of 1 min. Blood temperature may not mirror this increase because of differences in the measurement site. In our study, T_c was measured by a capsule situated in the GI tract. The temperature gradient has been reported to reach as high as 0.4°C between muscle tissues generating metabolic heat during exercise and the GI tract (Parkin et al., 1999). Therefore, blood flowing through arterial and venous conduits during exercise experiences temperatures ranging from ∼39°C at the muscle (Parkin et al., 1999) to ∼32°C in a superficial vein, based on our values. This reduction in temperature in mixed venous blood sampled from an arm vein is the result of combined efforts of radiative, convective, and evaporative heat loss, and is a display of our profound capacity to dissipate heat. This was elegantly displayed in exercising horses where temperatures of the gluteal muscle, pulmonary artery (representing T_c), and superficial thoracic vein were 43.9, 42.8, and 39.5°C, respectively, after 10 min of running at 90% V ˙ O 2 ⁡ max (Hodgson et al., 1985). This showcases the cascade of heat transfer and the rate at which blood is cooled in the vasculature during exercise. The greater difference between T_c and blood temperature in our study (6.6°C vs. 3.3°C) is likely a result of more efficient cooling in humans through a greater surface area to body weight ratio (Hodgson et al., 1985), better convection through our well-exposed dermis, the delay in measurement time after collecting the blood, and an evaporative response that begun in the sauna and was effective before the start of any exercise in our study.

The maintenance of blood pH over SS cycling can be explained by the increase in mean blood lactate ([La⁻]_b) to 4.42 (range = 0.97–7.79) mmol/L, indicating that participants were very close to their lactate threshold, which is typically ∼4 mmol/L, yet varies inversely with training status (Stegmann et al., 1981). This increase in [La⁻]_b is greater than we anticipated since we had participants cycle at 55% of their V ˙ O 2 peak with the intent to keep intensity below anaerobic threshold during SS cycling, which was previously reported to be 61.6% V ˙ O 2 ⁡ max in untrained participants (Simon et al., 1985). Therefore, although lactate threshold was not determined in the present study, it is likely that SS cycling was performed near the individual lactate anaerobic threshold in the majority of participants. To efficiently lower blood pH, we chose a HIIT protocol with a work:rest ratio derived from work by Nicolo et al. (2014), which showed that “iso-effort” cycling at 20:40 s provided the largest increase in [La⁻]_b compared to 30:30 s, 40:20 s, and continuous cycling. In comparison, our participants engaged in 1/3rd of the HIIT volume in sets, yet displayed considerably higher [La⁻]_b levels (12.1 vs. 9.5 mmol/L), which is likely from a generally lower level of fitness (i.e., V ˙ O 2 peak ; 40 vs. 67 mL⋅kg⁻¹⋅min⁻¹).

The calculated changes in PV were similar until initiation of cycling (i.e., SS, HIIT, and REC) in the EXS trial. Indeed, the %PV change after HEAT was comparable to the time-equivalent at rest (EXS: 1.63 ± 4.97% vs. Rest: 1.68 ± 1.87%); however, the variability was much greater in response to passive heating (Figure 2). This suggests that 10 min of passive heating may not be long enough to facilitate the same ∼5% hemodilution expected after 30 min of heating (Fortney and Miescher, 1994), which results from absorption of interstitial fluid into the vasculature in response to peripheral vasodilation and as an early compensatory mechanism for the ensuing sweating response (Bass and Henschel, 1956). This response shifts to hemoconcentration with the addition of exercise as the greater metabolic demands, coupled with an increased systolic arterial pressure, leads to greater vasodilation and an increased hydrostatic pressure of more capillary beds causing filtration to the extravascular compartments (Harrison, 1986). Additionally, since participants were instructed to abstain from drinking fluids after 15 min from the trial initiation, fluids lost from sweating during HEAT were not replaced and likely magnified the hemoconcentrated state during EXS since hydration status is a determinant of the direction of PV changes during exercise (Sawka et al., 1984). The 3.52% decrease in PV after 20 min of SS cycling at 55% V ˙ O 2 peak is almost half of the previously described 6.8% decrease after 15 min of cycling at 67% V ˙ O 2 ⁡ max (Novosadova, 1977). The 8.05% decrease at the end of HIIT is also shy of the reported ∼19% reduction after various HIIT protocols employed by Bloomer and Farney (Bloomer and Farney, 2013). After our participants stopped cycling, there was a classic PV rebound where they netted a +3.98% PV increase 10 min into REC. This is expected as exercise-induced decreases in PV tend to rebound to pre-exercise levels after a brief recovery period of ≥30 min (Novosadova, 1977).

Applying PV corrections avoids the inaccurate interpretation that total-body solute concentrations increase during hemoconcentration, and typically decreases the magnitude of change elicited by exercise (Berthoin et al., 2002). This is reflected in the current study where the Time × Trial interaction for paraxanthine was no longer significant when PV corrections were applied since variability from PV changes during EXS were accounted for. The results suggesting that caffeine is not sequestered in the intravascular space as fluid is expelled to the extravascular compartments during exercise is supported by previous research showing that [caffeine]_pl. was no different after 60 min of walking at 30% V ˙ O 2 ⁡ max (Collomp et al., 1991) and appeared slightly lower in lean individuals walking for 90 min at 40% V ˙ O 2 ⁡ max (Kamimori et al., 1987). In proportion to the studies evaluating caffeine and exercise, relatively few researchers have compared the MR of caffeine and its metabolites during rest and exercise. One investigation in horses (Aramaki et al., 1995) revealed increases in the enzymatic conversion of caffeine to theophylline—the major metabolite in horses and converted by CYP1A2 and 2E1—in blood samples taken 3 to 71 h after 4,000 m of various intensity exercise. Theobromine, another metabolite, showed elevated conversion after 22 h of exercise, but paraxanthine, which is the only metabolite solely metabolized by CYP1A2, showed no change after exercise. Regardless, the combination of these processes resulted in caffeine’s half-life time (t_1/2) to be hastened during exercise. This is congruent with post-exercise data in humans that showed a similarly faster t_1/2 after 60 min of walking at 30% V ˙ O 2 ⁡ max ; yet, MR changes cannot be substantiated since paraxanthine was not measured (Collomp et al., 1991). Based on these studies, it is possible that a longer sampling period in the current study would have revealed a change in the MR in the hours following the trial completion, or, simply, that the activity of CYP2E1, and not CYP1A2, increases with heat stress.

Given the complexity and time-dependence of measurements in the current study, there were some limitations that, if replicated, should be accounted for. The sample size met the a priori sample size calculation described previously in the Methods, but was relatively small, increasing the risks associated with extrapolation of our results to a larger sample size. At the time of our sample size calculations, very limited data were available to enable an a priori determination of the statistical power associated with that sample size; we thus determined power post factum. The results of our investigation can hopefully be used by those who will plan future related research, including an a priori determination of statistical power. The temperature of the sauna was not controlled by our study personnel since we lacked access to the controls. Controlling the temperature would allow for consistent heating among participants. Dermal and muscle temperatures were not measured. These values could describe the range of temperatures experienced by various drug-protein complexes and provide insight into thermal unloading of blood. Storing the vacutainers at mean venous blood temperature (e.g., 32°C), instead of room temperature, and insulating it upon collection would have minimized heat loss and provided greater confidence in the temperature of the blood sampled. To keep sample processing time to a minimum, the volume of fractions before (C_t) and after separation (C_b and C_u) were not measured, which precludes the possibility of calculating % recovery of C_u through the ultrafiltration method. Lastly, the results are only applicable insofar as they relate to the exercise employed and cannot preclude the possibility of changes in fu from greater perturbations in T_c and pH from other forms of exercise.

This is the first report of an in vivo study of healthy humans in which the effects on a drug’s f _u were assessed after a challenge involving passive heat stress and acute exercise. Passive heating followed by steady state and high-intensity cycling on a cycle ergometer were responsible for increasing T_c and decreasing blood pH 45–105 min after ingestion of caffeine. Despite previous reports on the necessity to control temperature and pH when analyzing drugs in vitro, we find no reason to be concerned about the effects of acute whole-body heat stress and exercise on the %f _u of caffeine. The validity and reproducibility of the methodology used to derive the f _u should be further investigated.