Caffeine and Caffeine Metabolites in Relation to Insulin Resistance and Beta Cell Function in U.S. Adults

The relationship between caffeine and insulin resistance (IR) has been assessed only in terms of caffeine intake, and the association between caffeine and beta cell function (BCF) remains unclear. This study examines the association between urinary caffeine and its metabolites, IR, and BCF in nondiabetic, noninstitutionalized US adults in order to account for the inter-individual differences in caffeine metabolism. Data on urinary caffeine and its metabolites, IR and BCF from adults aged 20 years and older who participated in the 2009–2010 and 2011–2012 National Health and Nutrition Examination Surveys were analyzed (n for caffeine = 994). IR and BCF were assessed using homeostatic model assessment (HOMA) and urinary caffeine and its metabolites were measured using high-performance liquid chromatography-electrospray ionization-tandem quadrupole mass spectrometry. After adjusting for all covariates, increases in urinary 1,3-DMU, 1,7-DMU, 1,3,7-TMU, theophylline, paraxanthine, caffeine, and AAMU were significantly associated with increased HOMA-IR and HOMA-β (HOMA of insulin resistance and beta cell function). Compared with individuals in the lowest quartile of urinary 1,3-DMU, 1,7-DMU, 1,3,7-TMU, theophylline, paraxanthine, caffeine, and AAMU, the regression coefficients for HOMA-IR and HOMA-β were significantly higher among those in the highest quartile. After stratification by prediabetes status, HOMA-IR and HOMA-β showed significant positive associations with urinary caffeine and its metabolites among subjects with normal fasting plasma glucose levels. Our cross-sectional study showed that caffeine and its metabolites were positively related to IR and BCF.


Introduction
Diabetes mellitus (DM) is a worldwide public health problem considered one of the four priority noncommunicable diseases targeted by world leaders [1]. Globally, the prevalence of DM has doubled from 1980 from 4.7% to 8.5% and, in 2014, approximately 422 million adults were living with DM [1]. In the United States, 34.2 million adults (13.0%) had DM and 1.5 million new cases were diagnosed in 2018 [2]. Factors such as age, body weight, sex, genetic predisposition, and lifestyle factors such as physical activity and diet are associated with insulin resistance (IR) [3]. In addition to carbohydrates, fibers, and fatty acid composition, an association between caffeine and IR has been reported [3][4][5][6][7].
The association between caffeine consumption and IR has been reported in a diverse population [7][8][9]. Previous studies investigating the relationship between caffeine and IR have reported that acute caffeine intake is associated with an increase in IR [5,6,[8][9][10], while chronic coffee consumption is associated with insulin sensitivity and a protective effect against type 2 DM risk [7,[11][12][13][14]. Conversely, the relationship between coffee and beta cell function (BCF) has been studied to a lesser extent and shows conflicting results; studies in Japanese adults and men from the Uppsala Longitudinal Study of Adult Men reported no association, while other studies from the Insulin Resistance Atherosclerosis Study (IRAS) and Shanghai High-Risk Diabetic Screen (SHiDS) study reported a positive association [5,[15][16][17]. Furthermore, no studies so far have investigated the relationship between caffeine and BCF.
Previous studies on the relationship between caffeine and IR have assessed caffeine exposure only in terms of caffeine intake [6,8,18]. Although food frequency questionnaires or 24-h dietary recalls are the source of most estimates of caffeine intake, they are usually limited to beverages and do not capture the full spectrum of caffeine-containing products [19,20]. Additionally, the interindividual differences in caffeine metabolism, which may result in different amounts of circulating caffeine and caffeine metabolites, have not been considered [21].
To account for the inter-individual differences in caffeine metabolism, and in order to study the associations between caffeine and its metabolites, IR and BCF, we sought to determine the association between IR, BCF, and urinary caffeine and its metabolites in a representative sample of nondiabetic, noninstitutionalized US adults using homeostatic model assessment (HOMA) modelling.

Study Population
Data on urinary caffeine and caffeine metabolites, insulin, and fasting glucose were obtained from the 2009-2010 and 2011-2012 National Health and Nutrition Examination Surveys (NHANES), a representative survey of the civilian population in the US [22]. A total of 20,293 people were included from both cycles. Urinary caffeine was measured in 5,112 people. The analysis was restricted to individuals with urinary caffeine and caffeine metabolite levels at or above the lower limits of detection due to the fact that the regression equations, accounting for the possible differences resulting from an instrument change between the two cycles, did not include values below the lower limits of detection [23,24]. As a result, urinary caffeine data was available for a total of 4782 individuals.
The analysis was restricted to adults aged 20 years and older (n = 3426) without a history of diabetes, diabetic drug use, or insulin use (n = 3009). Among these subjects 1361 were tested for fasting insulin and fasting glucose; however, because the interpretation of HOMA results generated when fasting insulin is less than or equal to 5 uU/mL and fasting glucose is less than 4.5 mmol/L are not valid, the study sample was limited to individuals with fasting insulin levels greater than 5 uU/mL and fasting glucose levels greater than or equal to 4.5 mmol/L (n = 1192) [25]. Among these subjects, 994 individuals had data for all covariate variables.

HOMA Modelling
IR was evaluated by HOMA-IR (fasting serum insulin (µIU/mL) * fasting plasma glucose (mmol/L)/22.5) and BCF was evaluated by HOMA-β (20 * fasting serum insulin (µIU/mL)/(fasting plasma glucose (mmol/L) − 3.5)) [26]. Fasting glucose and insulin blood tests were performed on participants after a nine-hour fast. Specimens were stored at −70 • C until analyzed at Fairview Medical Center Laboratory at the University of Minnesota, Minneapolis Minnesota [27,28]. To account for a change in methods for insulin measurements during the NHANES 2009-2010 cycle from the Mercodia sandwich ELISA assay to a Roche chemiluminescent immunoassay performed on the Elecsys 2010 analyzer in late 2009, a fractional polynomial regression was used to trend insulin from the NHANES 2011-2012 cycle to match the NHANES 2009-2010 cycle [27]. Glucose was measured using an enzymatic method measuring the increase in absorbance at 340 nm resulting from the reduction of NADP + to NADPH during the conversion of glucose-6-phosphate to gluconate-6-P by glucose-6-phosphate dehydrogenase [27].

Statistical Analysis
Concentrations of urinary caffeine and caffeine metabolites were analyzed as both continuous and categorical variables categorized into quartiles. Distributions of urinary caffeine and caffeine metabolites were right-skewed and log transformed. Pearson's correlation coefficients among HOMA-IR, HOMA-β, and urinary caffeine and caffeine metabolites were calculated. Linear regression analysis was performed to determine associations between urinary caffeine and caffeine metabolites, HOMA-IR and HOMA-β. Linear regression analyses provided beta coefficients and standard errors (SE) for HOMA-IR and HOMA-β among individuals with higher levels of urinary caffeine and caffeine metabolites (quartiles 2 through 4) compared to those with the lowest level of urinary compounds (quartile 1) as the reference group. The adjusted linear regression model was adjusted for all potential confounders including age, sex, race, income, education, marital status, smoking history, alcohol consumption, physical activity, BMI, urinary creatinine clearance, and caffeine intake. Linear regression analyses were also performed after stratification for prediabetes status.
According to the NHANES analytic guidelines, weighted estimates of the population parameters were applied to account for the complex sampling design. Analyses were performed with the PROC SURVEY procedures of SAS 9.4 statistical analysis package (SAS Institute, Cary, NC, USA). The statistical significance was set at α = 0.05.

Ethical Approval and Consent to Participate
The present study was exempt from formal ethics review as a secondary analysis of existing NHANES public data under the US Health & Human Services (HHS) regulations at 45 CFR 46.101 (b) [31]. The NHANES surveys were approved by the US National Center for Healthcare Statistics (NCHS) Research Ethics Review Board (ERB), and the NCHS IRB/ERB protocol number for NHANES 2009-2010 was a continuation of #2005-06 and NHANES 2011-2012 was #2011-17 [32]. Participants gave written informed consent before the home interview and health exams [33].
Our results regarding IR are consistent with results from a previous cross-sectional study which reported that high serum caffeine and paraxanthine concentrations were associated with HOMA-IR in fasting subjects enrolled in the Pregnancy Exposures and Preeclampsia Prevention Study (PEPP) [34]. With the exception of this cross-sectional study, this is the first study examining the relationship caffeine and its metabolites and IR. However, previous studies have reported a positive association between acute caffeine intake and IR [6,[8][9][10][35][36][37]. A recent systematic review by Emami et al., (2019) revealed that acute caffeine supplementation increases glucose concentration (weighted mean difference (WMD) = 2.48, 95% CI: 0.75 to 4.20; p = 0.005) and decreases insulin sensitivity indices (WMD = 0.03, 95% CI: −0.84 to 0.89) in both healthy subjects and those with type 2 DM [8]. In contrast, chronic intake has been associated with insulin sensitivity and a reduced type 2 DM risk [5,7,[10][11][12][13][14]. A cross-sectional study from the Uppsala Longitudinal Study of Adult Men (ULSAM) revealed a positive association between coffee and tea consumption and insulin sensitivity, showing that long-term consumption may show a different metabolic effect compared to short-term caffeine consumption [5]. A study in rats also showed that chronic caffeine intake reversed age-induced IR [38]. Additionally, studies have shown that both caffeinated and decaffeinated coffee are associated with a decreased risk of type 2 DM, revealing that constituents of coffee other than caffeine such as chlorogenic acid, polyphenols, and lignin may be responsible for coffee's protective role in type 2 DM [39][40][41].
Some of the most common mechanisms regarding the effects of caffeine on IR include increases in adenosine receptor antagonism and catecholamine release [42]. An animal study testing the effects of caffeine on insulin sensitivity of skeletal muscles revealed that acute caffeine administration decreased insulin sensitivity in a dose-dependent manner mediated by A1 and A2B adenosine receptors [43]. Caffeine is also known to stimulate the release of catecholamines, especially epinephrine, which impairs both hepatic and peripheral insulin sensitivity [6,44]. Metabolites of caffeine including paraxanthine, theobromine, and theophylline are biologically active and may play a role in causing IR [21]. An experimental study in humans showed that caffeine and paraxanthine have similar sympathomimetic actions producing an increase in diastolic blood pressure, plasma epinephrine levels and free fatty acids, and paraxanthine is as potent as caffeine in blocking adenosine receptors [21,45]. In line with these results, 1,3,7-TMX (caffeine), 1,3-DMX (theophylline) and 1,7-DMX (paraxanthine) showed positive associations with HOMA-IR in our study. However, no previous studies have considered 1,3-DMU, 1,7-DMU, 1,3,7-TMU, and AAMU in relation to IR. As a result, it is difficult to compare our results with those of previous studies. The mean fasting plasma glucose level for the study participants was in the range of prediabetes (5.6-6.9 mmol/L). In order to find out whether fasting plasma glucose levels influenced the results, we conducted linear regression analyses after stratifying for fasting plasma glucose. Significant positive associations between HOMA-IR, HOMA-β, and urinary caffeine and its metabolites were consistently observed among individuals with normal fasting glucose levels. In contrast, the regression coefficients were generally less significant among individuals with prediabetes, and further studies should be conducted in order to clarify this relationship.
We also observed a positive association between caffeine and its metabolites and BCF; however, evidence on the association between coffee intake and BCF is limited and inconsistent. A cross-sectional study in Japanese men showed that coffee consumption was significantly, inversely associated with HOMA-IR but showed no association with HOMA-β [17]. A cross-sectional study from the Uppsala Longitudinal Study of Adult Men (ULSAM) showed that coffee consumption was associated with an increase in insulin sensitivity but not with insulin secretion measured as the early insulin response, attributing the proposed antidiabetogenic effect of coffee as the effect of increased insulin sensitivity rather than improved BCF [5]. A study from the Insulin Resistance Atherosclerosis Study (IRAS) reported that decaffeinated coffee, but not caffeinated coffee, was favorably related to measures of BCF, measured by the intact and split proinsulin to C-peptide ratios, and a cross-sectional study in Sweden showed that there was an association between high coffee consumption and beta cell function in those with diabetes and impaired glucose tolerance, but not in subjects with normal glucose tolerance [4,16]. In contrast, a recent study in a large high-risk diabetic Chinese population showed that coffee intake is positively related to BCF measured using HOMA-β [odds ratio (OR) = 2.270, 95% CI: 1.456-3.538] [15]. Evidence on the relationship between coffee intake and BCF is inconsistent and remains limited, and furthermore, the association between caffeine and BCF remains unclear. Studies in mouse islet grafts showed that glucose-induced insulin secretion was potentiated by caffeine and in a randomized controlled trial of nine healthy fasted volunteers, insulin concentrations were higher after caffeinated coffee consumption than after decaffeinated coffee consumption (p < 0.05) [46,47]. Accordingly, caffeine or its metabolites, but not coffee, may be what is actually associated with increased BCF. Our results showed that caffeine may show an antidiabetogenic effect via increased BCF, in contrast to a previous study by Ärnlöv et al., (2004) which proposed that the antidiabetogenic effect of coffee may involve insulin sensitivity rather than improved BCF [5]. Further studies are recommended to clarify the association between coffee, caffeine, and BCF.
We have accounted for the inter-individual differences in caffeine metabolism by investigating the association of caffeine and caffeine metabolites and IR and BCF in a representative sample of nondiabetic, noninstitutionalized US adults. Nevertheless, our study has several limitations. Firstly, due to its cross-sectional nature we cannot confirm any casual or temporal relationships. In addition, due to the nature of the NHANES data, we were unable to account for the differences in acute and chronic consumers. Further studies should address the differences between acute and chronic consumption in relation to HOMA-IR, HOMA-β, and urinary caffeine and its metabolites. Secondly, BCF and IR was assessed using HOMA modelling; however, HOMA modelling has been shown to correlate well with established methods such as hyperglycaemic and euglycaemic clamps [26,48]. Thirdly, the relationship between HOMA-IR, HOMA-β, and caffeine and its metabolites may not necessarily be linear. As a result, the associations found in this study may not be generalized to populations with different ranges of urinary caffeine and caffeine metabolites. Fourthly, although basic sociodemographic confounding variables were considered, not all possible confounding variables may have been controlled for. Some studies have also included waist circumference, total energy intake and a family history of diabetes as confounding variables [5,15,16]. Therefore, we cannot exclude the possibility of residual confounding. Fifthly, serum levels of caffeine and its metabolites have not been measured. Despite the fact that urinary caffeine and its metabolites are associated with caffeine intake, urinary levels may not precisely reflect serum levels due to the fact that caffeine is reabsorbed from the renal tubule to reach equilibrium with unbound plasma caffeine [30,49]. Further studies evaluating the relationship between plasma caffeine and caffeine metabolite levels are recommended.

Conclusions
In conclusion, our cross-sectional study showed that caffeine and its metabolites were positively related to IR and BCF using HOMA modelling, suggesting that caffeine may play a protective role in T2DM via increased BCF. Further studies should be conducted (1) to investigate the effects of individual metabolites on BCF and IR, and (2) to clarify the relationship and mechanisms behind coffee, caffeine, and beta cell function.