Ethical approval

This protocol is a basic experimental study involving humans since it is a systematic study directed toward greater understanding on how coffee intake in healthy individuals influences the microbiome and behaviour with no intention to change the health status of the participants.

The study protocol was approved by the Clinical Research Ethic Committee of the Cork Teaching Hospitals (Study Identification Number APC115) and was registered in the Clinical Trials portal with the ClinicalTrials.gov ID NCT05927038, “Coffee Consumption, the Gut Microbiome, and the Microbiota-Gut-Brain Axis” and NCT05927103 “Differences Between Coffee and Non-coffee Drinkers in the Gut Microbiome and Microbiota-Gut-Brain Axis”. Informed consent was obtained from every participant enrolled in the study.

Study design

Sixty-two healthy adults between 30 and 50 years of age living in Ireland were recruited for the study between September 2021 and January 2023. Thirty-one participants were non-coffee drinkers (NCD, i.e., people that never consume coffee) and thirty-one participants were moderate coffee-drinkers (CD, i.e., people that usually consume between 3 to 5 cups of coffee per day). NCD and CD participants were compared cross-sectionally at baseline. CD participants only proceeded further with the study and were asked to abstain from any kind of coffee for 2 weeks. After the coffee washout period, CDs were block randomized (block of five, stratified by gender) into either caffeinated or decaffeinated coffee consumption group using a double-blinded, parallel design by the first author. Figure S17 details information relative to participants enrolment and allocation.

Study visits and coffee intervention

A scheme of the experimental design is displayed in Fig. 1A. During the screening visit at University College Cork the eligibility of the participants was verified. Demographics, medical and family history information were collected. MINI-international Neuropsychiatric Interview (M.I.N.I, version 7.0.2), Childhood Trauma Questionnaire (CTQ) and Rome IV Diagnostic Questionnaire for Functional Gastrointestinal Disorders in Adults (ROME-IV) were administered to assess the overall mental and gastrointestinal health of the participants^63,64. Habitual caffeine consumption was assessed by a 7-day caffeine consumption diary and a brief measure of verbal IQ was completed using National Adult Reading Test (NART)⁶⁵ (Supplementary Data 1). Both NCD and CD participants were asked to abstain from any other kind of caffeinated drink and dark chocolate during the week before the baseline visit, except their habitual coffee for CDs only.

During the baseline visit, the participants underwent a brief physical examination, and their samples were collected. Subsequently, the participants completed self-reported mood and behaviour questionnaires (see below), cognitive tasks and the Socially Evaluated Cold Pressor Test (SECPT). At the end of the baseline visit, NCDs participant concluded the study and received thirty euros compensation for any travel expenses incurred in taking part in the study. CDs participants were instructed to refrain from any caffeinated drink, any kind of coffee or dark chocolate for 2 weeks. After coffee/caffeine washout, CDs attended the pre-intervention visit in which they underwent a physical examination, samples collection and completed self-reported questionnaires as well as Paced Auditory Serial Additional Test (PASAT) cognitive test. During this visit, CDs were instructed to drink four sachets per day of the provided coffee (Nescafé Classic caffeinated or decaffeinated, 1.8 g instant coffee per sachet) for 3 weeks, beginning on that day.

No other coffee or caffeinated drinks were allowed during the intervention. The coffee provided was consumed with a quantity of hot water, milk, sugar chosen by the participant. CDs completed the study with the post-intervention visit (Visit 4), in which they completed the same measures as for V2 and received sixty euros compensation. Extra stool samples from CDs were collected 2 (day 2) and 4 (day 4) days after the start of coffee withdrawal and 2 (day 16), 4 (day 18) and 14 (day 28) days after the start of coffee intervention. Together with the samples, CDs completed mood, fatigue, and cravings questionnaires.

Exclusion criteria

The participants were instructed not to change dietary and food supplements habits. Participants were excluded from the study in the following situations: if affected by significant acute or chronic coexisting illness, if under any medication (except contraceptive pills and hormonal supplements) and antibiotics/probiotics/prebiotics (minimum 4 weeks of washout to participate), if vegan, if habitual consumer of high quantity of fermented foods (5-6 servings per day), if hypertensive, if pregnant or planning a pregnancy or lactating, if not fluent in English, if dyslexic/dyscalculic, if a current smoker, if involved in other experimental drugs or food trials with other companies or laboratory and if already participated as a volunteer in the Microbiota-Gut-Brain Axis lab (APC Microbiome Ireland) in the past 4 years.

Dietary intake quantification

Dietary intake was monitored and quantified by using a 7-day food diary⁶⁶. The food diary was completed three times, before the baseline visit, before the pre-intervention visits and before the post-intervention visit. Each day, every participant had to note anything they eat or drink throughout the day, specifying the amount/portion size, the method of preparation/brand and comments. The Nutritics tool (https://www.nutritics.com/en/) was used to enter the food diary data and to quantify the macronutrients consumed. The 7-day food diary was a paper-based document that was completed at the timepoints mentioned. It included instructions how to estimate portion size, how to describe the method of preparation and how the food was consumed. The information was then manually entered into Nutritics^TM. Food intake was weighted.

Dietary intake of (poly)phenols

A total of 1424 food items were retrieved from the food diaries. Foods not containing (poly)phenols (n = 391) were excluded from this analysis. All (poly)phenol-containing raw foods were univocally matched with items collected in an in-house database based on Phenol-Explorer 3.6⁶⁷ (www.phenol-explorer.eu). Raw food items that could not be associated to any item in Phenol-Explorer were associated with the most similar food or sourced through multiple literature searches and added to the in-house database. Recipes or complex foods were analysed by considering the individual ingredients and their relative quantity in the food. Specifically, recipes from the food diaries were used and, when not available, an online search was conducted. Changes in weight during cooking or processing were adjusted using yield factors from Bognar’s tables⁶⁸, Phenol explorer⁶⁹, or CREA⁷⁰ (e.g., cooked food, fruit juice, and espresso) wherever relevant.

Microsoft Access was used to match the food composition table with the one related to food intakes, and to calculate (poly)phenol intake by multiplying the amount of food consumed by the (poly)phenol content expressed in mg/100 g. All available compounds belonging to flavonoid, phenolic acid, lignan and other (poly)phenol classes of interest were considered, including total (poly)phenol content assessed through the Folin assay method. Total (poly)phenol content was also calculated as the sum of individual phenolics determined by chromatography without hydrolysis, or by chromatography after hydrolysis if non-hydrolyzed chromatographic values were unavailable, across all considered compound classes except for a few (poly)phenols where the data after hydrolysis is more reliable (as for lignans). Additionally, normal-phase high-performance liquid chromatography (HPLC) data were used for proanthocyanidins, while individual data obtained by reverse-phase HPLC were used when the previous one was not available.

Daily average intakes were analyzed both as total classes (flavonoids, phenolic acids, lignans, and others) and subclasses (anthocyanins, flavan-3-ols, flavanones, flavones, flavonols, isoflavones, hydroxybenzoic acids, hydroxycinnamic acids, hydroxyphenylacetic acids, hydroxyphenylpropanoic acids, chalcones, ellagitannins, prenylflavonoids, avenanthramides, stilbenes and several miscellaneous minor (poly)phenols across study periods.

Lastly, the main dietary sources of (poly)phenolic compounds were also calculated for total (poly)phenol intake and for each phenolic class. Dietary sources were analyzed as food groups as well as subgroups or specific food items (Supplementary Data S,17-S,29).

Caffeine consumption quantification

During the screening visit, both NCD and CD participants completed a 7-day caffeine consumption diary. In the diary, each participant reported all the caffeinated beverages consumed in the previous 7 days. For each day, they reported the kind of caffeinated drinks consumed (e.g., instant coffee, filtered coffee, cappuccino, flat white, black tea, energy drink), the amount consumed (in cups, or cans or bottles in ml and how many e.g., 3 240 ml mugs) and the method of preparation, the brand, or any other known useful information about a specific beverage. Based on the specific caffeinated beverages consumed by the participants, a table with caffeine content of the drinks was created (Supplementary Data 1) and was used to calculate the average of daily caffeine intake, daily caffeine intake from coffee only and daily caffeine amount given up for the study, all expressed in mg of caffeine/day.

Self-reported questionnaires

During visits 2, 3 and 4

Every participant completed the following questionnaires: Cohen’s Perceived Stress Scale (PSS)⁷¹, Emotional and Reactivity Scale (ERS)⁷², Urgency Premeditation Perseverance Sensation Seeking Positive Urgency Impulsive Behaviour Scale (UPPS-P)⁷³, Hopkins Symptom Checklist 90 (SCL 90-R)⁷⁴, Beck Depression Inventory-II (BDI)⁷⁵, State Trait Anxiety Inventory (STAI, during SECPT, State was used, and during baseline, post-withdrawal and post—intervention Trait was used)⁷⁶, Pittsburgh Sleep Quality Index (PSQI)⁷⁷, International Physical Activity Questionnaire (IPAQ)⁷⁸, Gastrointestinal Symptoms Visual Analogue Scales (GI-VAS). These questionnaires aimed to assess the level of stress, emotional reactivity, impulsivity, depression, sleep quality, physical activity gastrointestinal symptoms. Stool types were reported using the Bristol stool chart (BSC)⁶¹.

During washout and intervention

At the timepoints (T2W, T4W, T2I, T14I) during coffee withdrawal and intervention periods as well as during visit 3 and 4, CD participants were asked to complete the following questionnaires to monitor caffeine cravings and fatigue: Caffeine Withdrawal Symptom Questionnaire (CWSQ)⁷⁹, Questionnaire of Caffeine Cravings (QCC)⁸⁰ and Visual Analogue Scale to evaluate Fatigue severity (VAS-F)⁸¹.

Neuropsychological assessment

Verbal list learning and episodic memory

Modified Rey Auditory Verbal Learning Test (ModRey). The ModRey was administered to estimate the participant’s episodic memory. Participants learned two lists of words (A and B) each consisting of twenty words. Participants had to recall the words at three moments: immediately, following a short delay (5 min) and after a long delay (30 min), followed by a recognition task.

Word list A was played aloud via a standardised audio recording, after which the participant recalled as many words as possible (immediate recall). This process was completed two more times. After which, a new list of words (list B) was introduced, and the participant was again asked to immediately recall this word list. 5 min later (short delay), participants were asked to recall words from List A. After 30 min (long delay), participants again recalled words from both List A and B.

A recognition phase was then performed in which participants had to indicate whether each presented word came from list A, B, or neither of those lists⁸². To prevent learning effects, separate word lists were used for baseline and endpoint visits.

Visuospatial memory and new learning

Paired Associates Learning (PAL) from Cambridge Neuropsychological Test Automated Battery (CANTAB) included eight levels with increasing difficulty. White boxes were displayed on the screen and revealed in sequence. Some boxes would be empty, and others would reveal a pattern which the participants had to memorize. Once the content of every box was revealed, all the previously shown patterns would appear in the centre of the screen, one by one. For each pattern appearing, the participants had to touch the screen to select the box in which the pattern had originally appeared. In the case of an error, the sequence was shown again. The main outcome measure was the total number of errors summed across the entire task. This test does not include different versions of forms; thus, it cannot be repeated too frequently to avoid learning bias. Due to this limitation, this test was administered during visit 2 and visit 4 and not during visit 3.

Emotional processing

Emotion Recognition Task (ERT) from CANTAB has been used to assess emotional processing, which is often associated with depression and affective disorders. ERT measures the ability to identify six basic emotions in facial expressions along a continuum of expression magnitude (Happiness, Anger, Surprise, Sadness, Fear, Disgust). Consecutively, one of the six emotions computer-morphed images were displayed for 200 milliseconds on the centre of the screen. The subject had to then select amongst the six options which emotion had been shown. This test lasted between 6 to 10 min. Outcome measures include numbers of correct or incorrect answers and overall response latencies in individual emotions and across all emotions. This test does not include different versions of forms; thus, it cannot be repeated too frequently to avoid learning bias. Due to this limitation, this test was administered during visit 2 and visit 4 and not during visit 3.

Attention and vigilance

Paced Auditory Serial Additional Test (PASAT) included an audio recording where single digit numbers played. Participants were asked to sum the two most recent digits spoken on the tape. PASAT is composed by two trials⁸³. The single digit numbers (60 numbers each trial) were spoken at a rate one every 3 s (3’) and 2 s (2’) respectively for each trial. The PASAT outputs included the percentages of correct answers for each trial, 3’and 2’, as well as the total percentage of correct answers.

PASAT was performed during visit 2, 3 and 4 since it has limited learning bias, and two different forms are available. Form A was administered during visits 2 and 4, and form B was administered during visit 3. Form A and Form B differ in the numbers used on each 3’ and 2’ trial.

Socially evaluated cold pressor test

The Socially Evaluated Cold Pressor Test (SECPT) is a psychological and physical stressor that was used to induce an acute stress response by the activation of the hypothalamus-pituitary-adrenal axis (HPA axis)⁸⁴. The participants were led into a room different from the one used in study visit one, in which a researcher was present, a video camera and a bucket with water containing ice (0 °C). Each participant had to sit facing the camera and the researcher and was told that their facial expression would be recorded for later analysis and that the researcher was recording the non-verbal behaviour. Then, they were asked to submerge their hand, including the wrist, in the ice-cold water for three minutes unless the participant indicated they could no longer continue. During those three minutes, each participant was instructed to stare at the camera. At the end of the procedure, the participant was given a paper towel to dry their hand and allowed to leave the room.

Before and after the SECPT, eight saliva samples were collected as detailed in the methods section and the participants had to complete different stress reported questionnaires as an indication of anxiety, stress, mood, and pain related to the stress procedure only.

Prior to the SECPT, each participant completed the following questionnaires: STAI-State (Ferreira & Murray 1983), Primary Appraisal Secondary Appraisal (PASA)⁸⁵, Positive Negative Affect Schedule PANAS⁸⁶, Bond-Lader Visual Analogue Scales (BL-VAS)⁸⁷ and level of stress felt before the procedure (0% to 100%, VAS stress). After SECPT, each participant completed the following questionnaires: STAI-State, VAS stress after the procedure, VAS stress during the procedure, difficulty to keep the hand in the water (0% to 100%) and how unpleasant was the SECPT (0% to 100%), PANAS, BL-VAS and level of pain felt during the SECPT (1 to 10, VAS pain).

Biological samples

Biological samples were collected on the day of Visit 2 (V2), Visit 3 (V3) and Visit 4 (V4) and analysed as follows.

Saliva

The morning of the visit, participants were asked to collect four saliva samples. The first sample was collected within 3 minutes after awakening, the second 30 min later, and the third and fourth 45 min and 60 min respectively after wakening. These were used to measure the concentration of the cortisol awakening response (CAR) which gives an indication of HPA axis function^46,88. In addition, participants were asked to collect eight saliva samples during the SECPT in the order indicated in Fig. S15.These samples were used to assess the short-term release of cortisol before, during and after the stress procedure.

Saliva samples were collected using Salivette Cortisol tubes (Sarstedt) and kept refrigerated until delivery to the laboratory. The saliva tubes were centrifuged at g for 5 min at 4 °C. The obtained saliva was stored at −80 °C until analysis.

Blood

Blood samples were collected from the antecubital vein during each visit. Whole blood was collected into 4 ml lithium-heparin-containing tubes (Vacuette, Greiner Bio-One, 454029). Blood for plasma samples was collected into 4 ml K2E K2EDTA tubes (Vacuette, Greiner Bio-One, 454023). Blood for serum samples was collected into 3.5-ml CAT SERUM Sep Clot Activator tubes (Vacuette, Greiner Bio-One, 454071). Blood for serum samples was incubated for 20 min at room temperature and blood for plasma samples was stored at 4 °C. Afterwards, blood samples were centrifuged at 1500 × g for 10 min at 4 °C to obtain serum and plasma. Both were aliquoted and stored at −80 °C.

An aliquot of 1 mL of whole blood was diluted 1:10 in medium (DMEM, 10% Foetal Calf Serum (FCS) and 5% Pen strep). A 500 µL aliquot of diluted blood was stimulated with 5 µL of LPS-EK -TLR4 agonist, lipopolysaccharide from E. coli K12 (Human TLR1-9 Agonist Kit, InvivoGen) and incubated for 24 h at 37 °C. Stimulated and unstimulated whole blood was aliquoted and stored at −80 °C.

Urine

The first urine of the day was collected and kept refrigerated in a sterile collection tube by each participant, and 4 ml of urine were aliquoted into 1 ml of 0.25% sodium azide and stored at −80 °C.

Stool

Participants were asked to collect the first freshly voided stool samples of the day in plastic containers containing an AnaeroGen sachet (Oxoid AGS AnaeroGen Compact, Fischer Scientific, Dublin). Stool samples were refrigerated using ice packs until delivery to the laboratory, where they were aliquoted and stored at −80 °C. Stool samples were collected the day of the visits as well as during the extra timepoints during the coffee washout and coffee intervention.

Biological samples analysis

Cortisol analysis

Cortisol concentration within the saliva samples was measured using Cortisol enzyme-linked immunosorbent assay (ELISA) kit (Enzo Life Sciences, #ADI-901-071, Exeter, UK) according to manufacturer’s instructions. Saliva samples were diluted 1:3 in the provided assay buffer (200 μL of Assay Buffer + 100 μL of saliva sample). The absorbance was measured with Synergy (BioTek) plate reader and Gen5 software, with a lower detection limit of 57 pg/mL. Each saliva sample was tested in duplicate, and the average of each concentration was used to determine the final concentration of cortisol. The area under the curve was calculated⁸⁸.

Inflammatory markers analysis

Inflammatory markers were analysed both in whole blood (LPS-stimulated and unstimulated) and in plasma. Concentration of tumour necrosis factor alpha (TNFα), interleukin-1 beta (IL-1β), interleukin-6 (IL-6), interleukin-8 (IL-8), interleukin-10 (IL-10), and interferon gamma (IFNγ) was determined by using V-PLEX Proinflammatory Panel 1 (human) Kit MSD®MULTI-SPOT assay system (Meso Scale Discovery, USA) in either plasma or whole blood or both, according to the manufacturer’s protocol. No samples dilution was required. The concentration of C-reactive protein was determined by using V-PLEX Vascular Injury Panel 2 (human) Kit MSD®MULTI-SPOT assay system (Meso Scale Discovery, USA) in plasma samples diluted 1:1000 with diluent 100 and 101, following the manufacturer’s instructions. Cytokine concentrations were calculated using a 4-parameter logistical curve within the MSD DISCOVERY Methodological Mind Software. Results for each cytokine were expressed in pg/mL. C-reactive protein (CRP) was converted to mg/L. The results are shown as normalised concentration of cytokines in the whole blood. The normalized concentration was calculated by subtracting the unstimulated whole blood concentration from the LPS-stimulated whole blood concentration. This calculation was performed to better account for inter-individual differences.

Faecal gut microbiota analysis

DNA was extracted from 250–500 mg of sample using the QIAamp Power Faecal Pro Kit from Qiagen with homogenisation on a Tissuelyser (Qiagen) at 30 Hz for 10 min. DNA was quantified using the Qubit Broad Range Assay (Invitrogen) and 200–300 ng were used as input for shotgun library preparation. Samples were prepared using Illumina DNA prep (Illumina). Following library prep, samples were pooled at equal concentrations. The average size of the DNA fragments was determined using the Agilent High Sensitivity kit (Agilent) and concentration was determined using the Qubit High Sensitivity Assay. The final pool was then sequenced on a NovaSeq 6000 S2 flow cell according to the manufacturer’s instructions.

Microbiome: taxonomic and functional analysis

We performed quality checks on raw sequences from all faecal samples using the FastQC program. Shotgun metagenomic sequencing data were then cleaned, and host genome sequences were filtered using Bowtie2⁸⁹ via the Kneaddata wrapper program with following parameters: ILLUMINACLIP:/NexteraPE-PE.fa:2:30:10, SLIDINGWINDOW:5:25, MINLEN:60, LEADING:3, TRAILING:3. Reads were aligned to the Web of Life database using Bowtie2 and taxonomic and functional profiling of the microbial community was performed using woltka⁹⁰. Next, the uniref90-based gene abundance matrix was further collapsed by KEGG Orthology (KO) term mapping via the “woltka tools collapse” function provided within woltka. Woltka SOP is available online (https://github.com/qiyunzhu/woltka/blob/master/doc/wolsop.sh). Gut-Brain Modules (GBMs) and Gut-Metabolic Modules (GMMs) were calculated using the R version of the Gomixer tool⁹¹.

Metabolomics

Semi-polar metabolite analysis was carried out by MS-Omics as follows. The analysis was carried out using a Thermo Scientific Vanquish LC coupled to a Orbitrap Exploris 240 MS, Thermo Fisher Scientific. An electrospray ionization interface was used as ionization source. Analysis was performed in positive and negative ionization mode under polarity switching. The UPLC was performed using a slightly modified version of the protocol described by Doneanu et al. (UPLC/MS Monitoring of Water-Soluble Vitamin Bs in Cell Culture Media in Minutes, Water Application note 2011, 720004042en). Peak areas were extracted using Compound Discoverer 3.3 (Thermo Scientific). Identification of compounds were performed at four levels; Level 1: identification by retention times (compared against in-house authentic standards), accurate mass (with an accepted deviation of 3ppm), and MS/MS spectra, Level 2a: identification by retention times (compared against in-house authentic standards), accurate mass (with an accepted deviation of 3ppm). Level 2b: identification by accurate mass (with an accepted deviation of 3ppm), and MS/MS spectra, Level 3: identification by accurate mass alone (with an accepted deviation of 3ppm).

SCFA analysis was carried out by MS-Omics as follows. Samples were acidified using hydrochloride acid, and deuterium labelled internal standards where added. All samples were analyzed in a randomized order. Analysis was performed using a high polarity column (Zebron™ ZB-FFAP, GC Cap. Column 30 m x 0.25 mm × 0.25 μm) installed in a GC (7890B, Agilent) coupled with a quadropole detector (5977B, Agilent). The system was controlled by ChemStation (Agilent). Raw data was converted to netCDF format using Chemstation (Agilent), before the data was imported and processed in Matlab R2021b (Mathworks, Inc.) using the PARADISe software described by Johnsen et. al (DOI: 10.1016/j.chroma.2017.04.052).

Urine sample preparation for targeted metabolomics

For phenolic analysis, urine samples were defrosted, vortexed, diluted in 0.1% formic acid in water (1:5 v/v), centrifuged at 17,968 × g for 5 min, and filtered (0.22 μm nylon filter)²⁶. For caffeine and caffeine metabolites, a subsequent dilution in acidified water was performed (1:50 v/v). For pyridine analysis, 10 µL of urine mixed with pure acetonitrile (1:100 v/v), vortexed for 1 min, and centrifuged at 13 765 × g for 10 min at room²⁷. For creatinine, urine samples were centrifuged at 10,980 g for 5 min and subsequently diluted with water 0.1% formic acid (1:1000 v/v), vortexed, and transferred into vials for analysis²⁸.

UPLC-ESI-QqQ-MS/MS analysis for phenolics, caffeine, and caffeine metabolites:

An Ultra-high performance liquid chromatography (UHPLC) Acquity UHPLC I-Class Plus system (Waters Corporation, Milford, MA, USA) equipped with a binary pump, autosampler, and oven with an Acquity Premier HSS T3 column (1.8 µm, 2.1 × 100 mm i.d., Waters) installed with an Acquity VanGuard Premier HSS T3 1.8 um (2.1 × 5 mm) precolumn were used. The injection volume was 5 µL, the samples were maintained at 10 °C, and the column at 40 °C. The mobile phase consisted of water with 0.01% formic acid (A) and acetonitrile (ACN) with 0.01% formic acid (B). A gradient elution was applied as follows: 0 min, 99% (A); 0.5 min, 99% (A); 3.0 min, 85% (A); 6.0 min, 50% (A), 9.0 min, 5% (A), 10.0 min, 5% (A), 11.0 min, 99% (A); 14.0 min, 99% (A). A flow rate of 400 µL/min was applied. A Xevo XS triple-quadrupole mass spectrometer with an electrospray ionization source (UHPLC-ESI-QqQ-MS/MS, Waters) operating in negative and positive modes was used. The capillary voltage was set at 2.3 kV. The source temperature was set to 150 °C, and the desolvation gas to 600 °C. The gas flows were set to cone gas 150 L/h, desolvation gas 800 L/h and nebulizer at 7.0 L/h. The mass spectrometer operated in multiple reaction monitoring (MRM) mode and cone, and collision voltage were optimized for each compound by infusion experiments with standards. When not available, the energies were optimized in chromatography or used from molecules with similar structures. Equipment control and data acquisition were performed using MassLynx software version 4.2 and data processing by TargetLynx Software (Waters). A total of 125 urine (n = 1) samples were analyzed,

UHPLC-ESI-QqQ-MS/MS analysis for pyridines (trigonelline derivatives):

Samples were analyzed using a UHPLC DIONEX Ultimate 3000 fitted with a TSQ Vantage triple quadrupole mass spectrometer equipped with a heated-ESI source (Thermo Fisher Scientific Inc., San Jose, CA, USA)^25,26. Separations were performed using XBridge® BEH HILIC column (2.1 × 100 mm; 2.5 µm particle size; Waters) installed with a precolumn cartridge (Waters). Volume injection was 5 µL and column oven was set to 35 °C. Flow rate was set at 0.5 mL/min, mobile phase A was water containing 0.1% formic acid, mobile phase B was acetonitrile containing 0.1% formic acid, and mobile phase C was ammonium formate 20 mM containing 0.1% formic acid. For mobile phase C, the proportion of 10% was kept throughout the run. Following 0.5 min of 1% A in 89% of B, the proportion of B was decreased to 80% for a period of 3 min. Solvent B was decreased again to 47% for 1.5 min and increased back to 89% from 6.5 min to the end of the run (total run: 13 min). The HESI-II source interface was set to a capillary temperature of 270 °C and the source heater temperature was 200 °C. The sheath gas pressure was set at 40 (arbitrary units), the auxiliary gas pressure at 5 (a.u.), and the source voltage was 3.5 kV. A total of 3 compounds were monitored through selective reaction monitoring (SRM) mode. Quantification was performed with calibration curves of authentic standards. Data processing was performed using Xcalibur software version 2.1 (Thermo Scientific Inc., Waltham, MA, USA). A total of 125 urine samples (n = 1) were analysed (Supplementary Data 22).

UHPLC-ESI--MS/MS analysis for creatinine:

Analysis was performed with an UHPLC Accela 1250 equipped with a linear ion trap-mass spectrometer (LTQ XL) fitted with a heated-ESI probe (Thermo Fisher Scientific) (Tosi et al., 2023). Separations were performed using XSelect HSS T3 column (50 × 2.1 mm; 2.5 µm particle size; Waters) installed with a precolumn cartridge (Phenomenex, CA, USA). Volume injection was 5 µL and column oven was set to 40 °C. Flow rate was set at 0.2 mL/min, mobile phase A was water containing 0.1% formic acid and mobile phase B was acetonitrile containing 0.1% formic acid. Chromatographic separation was performed in an isocratic mode keeping A and B at 50% for 2 min. The HESI-II interface was set to a capillary temperature of 275 °C and the source heater temperature was 300 °C. The sheath gas (N₂) flow rate was set at 40 (a.u.) and the auxiliary gas (N₂) flow rate at 5 a.u. The source voltage was 4.5 kV, and the capillary and tube lens voltage were +20 and +95 V, respectively. Ultra-high purity helium gas and a CID equal to 55 were used to obtain MS² fragmentation. Creatinine was analyzed with a full-scan MS² analysis in a positive ionization mode, monitoring all the ions with a m/z value between 35 and 200 Da. Chromatograms and spectral data were acquired using XCalibur software 2.1 (Thermo Fisher Scientific). Identification and quantification were performed with calibration curves built with the available standard. A total of 125 urine samples (n = 1) were analysed.

Targeted metabolomics of faecal samples

Faecal samples were thawed, weighed, proportionally diluted in ethyl acetate acidified (1200 µL for every 300 mg of sample) with formic acid (0.1%), vortexed, and sonicated for 10 min at room temperature^29,30. Vortexing and sonication (5 min) were repeated once before centrifuging (14000 rpm, 10 min, 4 °C). The supernatant was recovered, and the pellet was re-extracted following the same procedure. The resulting supernatants were pooled, vortexed, dried in a vacuum concentrator, resuspended in 200 μL of MeOH: H2O (50:50, v/v acidified with formic acid (0.1%)), sonicated, centrifuged, and filtered (0.22 μm nylon) into vials for UHPLC-ESI-QqQ-MS/MS analysis.

Faecal extracts were analyzed using the same instrument described for pyridine analysis. Separation was carried out by means of a Kinetex EVO C18 column (100 × 2.1 mm; 2.6 µm particle size; Phenomenex), installed with a precolumn cartridge (Phenomenex) (Di Pede et al., 2022). Mobile phase, pumped at a flow rate of 0.4 mL/min, consisted of a mixture of acidified water (0.01% v/v formic acid) (solvent A) and acidified acetonitrile (0.01% v/v formic acid) (solvent B). Following 0.5 min of 5% solvent B in A, the proportion of B was increased linearly to 40% over a period of 7 min. Solvent B was increased again to 80% in 1 min, maintained for 2 min and then the start conditions were re-established in 0.5 min and maintained for 3 min to re-equilibrate the column (total run: 14 min). The HESI-II source interface was set to a capillary temperature of 275 °C and the source heater temperature was 250 °C. The sheath gas (N₂) flow rate was set at 40 a.u., the auxiliary gas (N₂) flow rate at 5 a.u. and the sweep gas flow was set at 15 a.u. The source voltage was 3 kV; the capillary voltage was −9 V and tube lens voltage was −53 V. A total of 52 compounds were monitored through SRM mode. Quantification was performed with calibration curves of standards, when available, or using the most structurally similar compound. Data processing was performed using Xcalibur software version 2.1. A total of 123 faecal samples (n = 1) were analysed once. Quality controls were performed using one point of the calibration curve during the batch for every 30 samples (Supplementary Data 23).

Target metabolomics quality assessment

For all targeted metabolomics analyses, metabolites were identified by matching retention times and fragmentation patterns with authentic standards. When standards were not available, characteristic fragments of the corresponding phenolic class were monitored and cross-checked with both the in-house library and literature references. Peak quality was evaluated based on peak shape, and peaks deviating by more than 0.05 minutes from the reference were excluded. Quality control was ensured by injecting a calibration-curve point every 30 samples to monitor potential instrumental drift and sensitivity loss. Calibration curves included at least seven points, with linearity values above r = 0.98. Accuracy for each point ranged between 80% and 120%. The limit of quantification was defined as the first calibration point with a signal-to-noise ratio above 10 and meeting the linearity criteria. Final metabolite concentrations were expressed relative to creatinine for urine samples and per unit weight (mg/g) for faecal samples. The validated dataset was then used for all subsequent analyses. The targeted metabolomics data have been deposited to MetaboLights repository with the study identifier MTBLS13494⁹² (Supplementary Data 22 and 23)

Adenosine A2A receptor (ADORA2A) genotyping

DNA was extracted from 100 µl of whole blood using the DNeasy Blood&Tissue kit (Qiagen). Two Single Nucleotide Polymorphisms (SNPs) related to caffeine sensitivity were selected using the data base https://www.ensembl.org/index.html and previous scientific evidence^{40,93,94,95,96}. The SNPs selected are rs5751876 (Frequency MAF: 0.44. synonymous variant) and rs2298383 (Frequency MAF: 0.40, non-coding transcript exon variant). DNA was diluted to obtain a concentration of 4 ng/µl in 50 µl of elution buffer.

SNPs genotyping analysis of rs2298383 and rs5751876 was performed at Translational Analysis in Molecular Medicine (TAMM) at Karolinska University Hospital (Huddinge, Sweden) using iPLEX® Gold chemistry and the MassARRAY® mass spectrometry system⁹⁷ (Agena Bioscience, San Diego, CA, U.S.A.). A two-plexed assay was designed using MassARRAY® Assay Design Suite v3.0 software (Agena Bioscience), genotyping the two SNPs in one reaction per sample. The protocol for allele-specific base extension was performed according to Agena Bioscience’s recommendation. Analytes were spotted onto a 384-element SpectroCHIP II array (Agena Bioscience) using Nanodispenser RS1000 (Agena Bioscience) and subsequently analysed by MALDI-TOF on a MassARRAY® Analyzer 4 mass spectrometer (Agena Bioscience). Genotype calls were manually checked by two persons individually using MassARRAY® TYPER v5.0 Software (Agena Bioscience).

The genotyping assay was validated using a set of fourteen trio families, in total forty-two individuals, with genotype data available for rs2298383 through the HapMap consortium for39 individuals, and for rs5751876 through the 1000 Genomes Project Phase 3 for 21 individuals, respectively. Concordance analysis with the available data showed 100% concordant genotypes for both SNPs. Parent-offspring-compatibility analysis was performed, and no Mendel error was observed in the genotyped trio families for any of the SNPs. Three individual control samples were genotyped in parallel with both the validation sample run and the run of the study samples. All control samples showed 100% concordant genotypes for both SNPs. Additionally, genotyping the study samples in duplicate resulted in 100% concordant genotypes for both SNPs.

Bioinformatics

Further data-handling was undertaken in R (version 4.2.0) using the Rstudio GUI (version 2022.2.2.485). In all microbiome analysis except for alpha diversity, taxa with a prevalence of <60% of samples at the species level were excluded from analysis as ratios are invariant to subsetting and this study employs compositional data analysis techniques. Principal component analysis was performed on centred log-ratio transformed (clr) values. Zeroes were replaced using the ‘const’ approach described by Lubbe and colleagues. Beta diversity was computed in terms of Aitchison distance, or Euclidean distance between clr-transformed data, and assessed by PERMANOVA using the vegan package. Alpha diversity was computed using the iNEXT library. Differential abundance of microbial features and metabolites was assessed using linear models, or linear mixed effect models in the case of longitudinal (repeated measures) data. Variables were selected based on the Benjamini-Hochberg procedure with a q-value of 0.2 as a threshold. This procedure controls the False Discovery Rate (FDR) using sequential modified Bonferroni correction for multiple testing. Figures were generated in R, using ggplot2 and its extensions patchwork and ggforce.

Statistical analysis

Our primary outcome measure was microbiota composition and function (see NCT05927103) while secondary outcome measures included gut microbial metabolites (including SCFAs) and Coffee-related metabolites. Other secondary outcomes included cognitive performance, response to an acute stressor, peripheral blood inflammatory profile and the cortisol awakening response. g*Power was used for power calculations considering an α-error probability (false positive) of 0.05 and a power of 80%. This calculation indicated that the minimum required samples size per group was 18, thus, 36 participants were required for analysis. Effect sizes of 0.25 will be able to be differentiated using n = 18 per group (ANOVA repeated measures) for microbiota composition and function. Allowing for attrition, we estimated the requirement to recruit a minimum of 60 volunteers were recruited. Final recruitment numbers were 62, following screening of 118 potential participants and enrolment in the study of 92. Statistical analyses were performed using SPSS version 28 (IBM, Armonk, NY, USA) and n R (version 4.2.0) using the Rstudio GUI (version 2022.2.2.485). Outliers have been identified using the Grubb’s test (α = 0.05) and were removed before analysis. Every set of data except demographic data and (poly)phenols intake from food diaries were tested for outliers. Only statistical outliers were removed due to modest sample size used in the study. General Linear Model Univariate was used to perform statistical analysis of all the data of NCD versus CD in visit 2 (baseline), except CAR and SECPT timepoints as well as immediate versus delayed recall of List A in the ModRey. Age, gender, BMI, and amount of caffeine given up were used as covariates. All the other statistical analyses were performed using Mixed Models followed by Bonferroni’s post hoc pairwise comparisons at each timepoint. Statistical analysis for categorical data, such as mode of delivery and genotyping data, were performed using Pearson’s chi-squared test. p < 0.05 was deemed significant in all analysis. Data are expressed as mean + SEM unless otherwise stated.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.