GoKawiilInitially, we sought to directly measure the concentration of caffeine (E ox = ≈1.4 V vs. Ag/AgCl25), CGAs (E ox = ≈0.2–0.5 V vs. Ag/AgCl, depending on the isomer23), and other redox-active species in undiluted coffee samples to study their dependence on conventional brew parameters. However, caffeine and CGAs form an aggregate at concentrations typical of brewed filter coffee26, terminally impacting their redox activity—they only become electrochemically well-resolved in acidified dilute solutions with added electrolyte. Literature examples dilute to below 0.02 wt.% TDS27, nearly two orders of magnitude weaker than filter coffee. Other groups have shown that by adulterating coffee with either CGA or caffeine, both boron-doped diamond (BDD) and glassy carbon (GC) electrodes can provide molecule-specific information28,29,30,31,32,33,34,35,36. Yet, these approaches required laborious sample preparations, in opposition of both our and the industry's goal of measuring features of as-consumed coffee. Ideally, ensemble chemical measureables map to sensory experiences and relate the electrochemical features to total beverage strength, rather than measuring the concentration of specific molecules contained within. Thus, we first ascertained the redox landscape in filter-strength coffee by making CV measurements using BDD, GC, and Pt working electrodes across the electrochemical window of the beverage (Fig. 1a).
Fig. 1: Assessing the voltametric features present in coffee extracts. The alternative text for this image may have been generated using AI. Full size image a Cyclic voltammetry performed at 200 mV s−1 using boron-doped diamond, glassy carbon, and platinum working electrodes. Caffeine and chlorogenic acid potential ranges studied by other groups are highlighted. Here, we focus on the protonic features (H UPD and acid redox) region. b These features are suppressed with subsequent CV cycling due to deposition of coffee material on the surface of the electrode, and the total charge can be extracted by subtracting the hydrogen evolution reaction (HER) background. c The current depends on wt.% TDS of the brewed coffee because the protonic and organic concentration scales with wt.% TDS. d Scanning anodically results in mass accumulation on the working electrode, leading to electrode fouling. e The mechanism of mass accumulation is likely proton-assisted, given that appreciable mass does not deposit until the surface has accumulated a critical H-atom concentration. Source data are provided as a Source data file.
As-consumed filter coffee extracts are sufficiently conductive for direct electrochemical analysis without the addition of a supporting electrolyte (ranging from 2-3 mS cm−1, see Supplementary Fig. 1). Additionally, coffee extracts are self-buffered to pH of ≈4.8–5.9 depending on the distribution of compounds in the coffee and brew water composition37,38,39. Even after performing bulk electrolysis at oxygen evolution potentials for two minutes, the pH of a typical filter coffee remains numerically identical, reinforcing the significant buffering capacity of brewed coffee. Yet, despite the plethora of molecules in coffee extracts, the CV response of a Pt working electrode in 1.56 wt.% TDS coffee is consistent with that of dirty acidic water (Fig. 1a)40,41,42,43,44,45.
The cathodic Faradaic features map to the response expected for protonic reactions with the Pt surface (e.g., H UPD ), followed by H 2 evolution at more negative potentials. At positive potentials, OH adsorption and eventual O 2 evolution are also evident. To ascertain whether the anodic feature at –0.6 V is linked with the cathodic features in a reversible redox couple, we probed the scan rate dependence, Supplementary Fig. 2. The linear dependence of peak current on the square root of the scan rate for both features demonstrates the diffusion-controlled nature of the redox events, Supplementary Fig. 3, and the lack of an increase in the peak potential separation with increasing scan rate indicates Nernstian behavior, Supplementary Fig. 4. However, the large peak separation of ≈200 mV at all scan rates suggests that the reversibility of the redox couple is obfuscated by sluggish kinetics.
The same Pt surface sites that adsorb H+ and OH– are also able to adsorb other molecules in solution. In the case of oxidative cycling, some impurities in water compete for the Pt surface, resulting in reduced current with subsequent cycling due to a decrease in the accessible surface area. Given that Pt is known to interact with caffeine and other molecules in coffee46,47, we expected to see a decrease in exchange current density with sequential cycling. When scanning from 0 to –1.0 V, the H UPD and protonic features (E pc = –0.4 and –0.7 V, respectively) smear together and current decreases by ≈34% from CV scan 1 to 2 and ≈18% from scan 2 to 3 (Fig. 1b)48. In pH 7 water purified by reverse osmosis, the same features are not observed (Supplementary Fig. 5), suggesting that the response is due to protonic chemical steps associated with the coffee and not the water.
Further experiments were run to ensure that these features mapped to H UPD /weak acid reduction and its suppression by coffee molecules, rather than fluctuations in dissolved O 2 and other spurious effects, Supplementary Fig. 6. Since the integral of the current density depends on the activity of H+, the H UPD and acid features indirectly provide ensemble insights into the families of molecules in coffee that function as H+ donors and acceptors, the concentrations of which should depend on roast color, brewing parameters, brew water composition, and so forth. Some data in support of this hypothesis is that H UPD /acid reduction current density decreases with decreasing coffee concentration due to the diminished concentration of available H+ (Fig. 1c). Because the feature is concentration dependent, there are also fewer organic molecules competing to bind to the surface of the electrode. As we will show later, the integral of the charge current density of this feature linearly maps to wt.% TDS. Perhaps this is a surprising result, given that a single acidic feature should not necessarily depend on ensemble concentration.
To further support our proposed mechanism that protonic chemistry is being suppressed by adsorbed coffee material, we performed CV measurements using an electrochemical quartz crystal microbalance (QCM) with a Pt working electrode scanning at 50 mV s−1 (Fig. 1d). Scanning cathodically, appreciable mass begins to accumulate on the electrode at potentials more negative than H UPD —the balance is insensitive to surface H-adsorption but can detect larger molecule accumulation. There is a delayed onset, which attribute to a proton-assisted adsorption of Brønsted-basic species like caffeine, following a general mechanism (Fig. 1e). Upon scanning anodically from −1.2 to 0.0 V, mass continues to accumulate until the potential exceeds −0.5 V, when the electrode liberates most of the adsorbed organic material and protons back into solution. The response is reminiscent of a kinetic trap, where the surface assembly forms in kinetically favored conditions49. To ensure that the process was proton assisted, we also scanned from −1.2 to 0.0 V without first scanning cathodically and detected no mass accumulation, indicating that there must be an appreciable concentration of protons on the surface before larger organic species begin to accumulate, Supplementary Fig. 7.
While the total charge passed maps linearly to concentration for any particular coffee, coffees from different origins, processed in dissimilar ways, and roasted to different colors may show major differences in the emergence and suppression of the reductive features. However, before we can probe coffee-related variables we must first ensure that the suppression of the convoluted redox feature at −0.55 V depends on molecules likely found in all coffees. To determine the molecular identities of the adsorbed compounds, we developed the CV cycle shown in (Fig. 2a) to maximize mass accumulation on the Pt working electrode for further characterization. We were able to obtain sufficient coffee material on a Pt-mesh electrode surface by cycling 100 times from −0.55 to −1.2 V (Fig. 2a), followed by solvating the adsorbed material in a sonicated 4 mL bath of 80/20 water/acetonitrile (v/v), and repeating four times. The adsorbates could then be separated and characterized using high performance liquid chromatography coupled with high-resolution mass spectrometry (HPLC-MS, Fig. 2b). We found that caffeine had adsorbed in quantifiable concentration. Our combined samples yielded 7.8 ± 0.1 mg kg−1 caffeine, suggesting that at least one component of the accumulated mass that causes the current to decay over successive CV cycles originates from a molecule common to all coffees, and that our 4 mL solution contained ≈300 µg of caffeine, or ≈0.4% of the total caffeine in an average 180 mL cup of filter coffee50,51,52, (see Supplementary Figs. 8,9). That is, each 100-cycle CV presented in (Fig. 2a) scavenged ≈0.1% of the available caffeine in the cup, as well as other molecules.
Fig. 2: Identification of surface adsorbates at negative applied potentials. The alternative text for this image may have been generated using AI. Full size image a Cyclic voltammetry performed at 200 mV s−1 sampling potentials more negative than –0.5 V, to ensure our accumulated mass is maximized rather than liberated back into solution (Fig. 1d). The first four cycles of the second run are presented. b Combining the extracts from the surface of the Pt electrode sampled 100 times in four separate runs, we can detect the presence of caffeine and quantify it using the calibration curve presented in Supplementary Fig. 8. c The adsorption of caffeine and 5-caffeoylquinic acid is favored on all clear crystal facets of Pt according to density functional theory simulations, suggesting that a collection of organic molecules responsive to roast color and brewing parameters adsorb to the electrode surface. Source data are provided as a Source data file.
Given that a similar electrochemical approach has been used to quantify caffeine content in highly dilute coffee samples through oxidation29, and that bulk Pt also is known to adsorb organic material40, we were somewhat unsurprised to see caffeine in the chromatogram. However, there are of course some coffees that have been decaffeinated, prompting us to resolve whether this mass deposition can be attributed to additional adsorbates beyond caffeine. Because other adsorbates were not in sufficient concentration to quantify with HPLC-MS, we instead turned to density functional theory (DFT) paired with molecular dynamics simulations (MD) to model the Pt surface adsorption energies for caffeine as well as 5-caffeoylquinic acid (5-CQA), an abundant CGA isomer found in all coffees with a concentration that depends strongly on roast parameters50,53,54, (Fig. 2c). The inclusion of caffeine in the DFT study serves as a control to validate the model, and the inclusion of 3 low-index Pt surfaces, (100), (110), and (111), examines the possibility of preferential binding to certain facets of the polycrystalline Pt electrode used for the electrochemical measurements. The model we use does not account for the significant thermodynamic parameter of solvation, as explicit treatments of solvent molecules incur high computational costs and implicit models often fail to reproduce experimental solution-state adsorption energies. Because the penalty of excluding water from the Pt surface upon adsorption is unaccounted for, the calculated adsorption energies likely overestimate the attractive force of the interaction55. Furthermore, the exclusion of an applied electric field in the DFT model precludes both the assertion of equilibrium geometries for the Pt/adsorbate complex at potentials corresponding to electrochemical processes and the calculation of accurate thermochemical parameters for adsorbate formation. Nevertheless, a model of the electrode/adsorbate interface at zero field is instructive for determining relative trends of adsorption between similar molecules that form similar surface dipoles, and for assessing the existence or lack of facet dependence in the complex56.
... continue reading