University of Groningen Risk variables for the development of obesity and type 2 diabetes van der Meer, Tom

Risk variables for the development of obesity and type 2 diabetes

van der Meer, Tom

DOI:

10.33612/diss.170143787

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from

it. Please check the document version below.

Document Version

Publisher's PDF, also known as Version of record

Publication date:

2021

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):

van der Meer, T. (2021). Risk variables for the development of obesity and type 2 diabetes. University of

Groningen. https://doi.org/10.33612/diss.170143787

Copyright

Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.

Chapter nine

Summary, general discussion and

future perspectives

9

Summary

The studies contained in this thesis provide insights in the risk factors for the development of obesity and type 2 diabetes, with a focus on non-persistent Endocrine Disrupting Chemicals (EDCs). We provide a general introduction in Chapter 1. Our primary aim was to determine whether parabens, bisphenols and phthalates are hazardous substances in the development of obesity and type 2 diabetes. Second, we aimed to assess the associations and clinical impact of a broader range of potential risk variables in the development of type 2 diabetes. We found that exposure to EDCs is ubiquitous in the Netherlands population and has been gradually declining. Higher concentrations of EDCs are associated with adiposity-related traits, but not with type 2 diabetes. Although many other risk variables show robust associations with the development of type 2 diabetes, only few contribute to risk prediction. Here, we summarized our main findings. Next, we discussed our findings and provided directions for future research.

In Part I of this thesis, we focused on the exposure to parabens, bisphenols and phthalates, and their association with obesity and type 2 diabetes. In Chapter 2, we described the quantification of seven parabens and nine phenols in two brain regions of normal-weight and obese individuals. We detected a larger number of EDCs above the limit of detection in hypothalamus compared to white-matter brain tissue. Yet, concentrations did not differ between tissues. Further, we found higher concentrations of methyl paraben in the hypothalamus of obese compared to normal-weight individuals.

In chapter 3, we present the setup and inter-laboratory validation of two methods which are able to measure EDCs in human urine, using ultra-performance liquid chromatography tandem mass spectrometry (LC-MS/MS). The first method was able to determine five parabens and three bisphenols, whereas the second method was able to determine thirteen different metabolites from eight phthalates. We added ammonium fluoride in the mobile phase of the phenol method, and used acetonitrile as eluent in the phthalate method, instead of the more commonly used methanol. This way, we were able to reduce runtimes without compromising chromatographic resolution, making our methods suitable for high-throughput analysis. We thoroughly validated our methods using reference material from the National Institute of Standards & Technology (NIST). Further, the same urine samples of 40 individuals from the Lifelines cohort study were compared with measurements performed by established LC–MS/MS methods at the Department of Growth and Reproduction, Rigshospitalet, Copenhagen University Hospital, Denmark (1–3). We found a high level of agreement with both the external standard reference materials and the interlaboratory comparison.

In Chapter 4 we present our study in which we assessed EDC exposure in a general Dutch population. We detected bisphenol A, four parabens and eight phthalate metabolites in 84-100% of the samples, indicating ubiquitous exposure. After adjusting for a wide range of covariates including physical activity and caloric intake, we found that high concentrations of bisphenol A and the phthalate metabolites MiBP, MECPP and MBzP were associated

with increased adiposity-related traits. Of these, associations for MiBP and MBzP were robust for Benjamini and Hochberg multiple testing. We found the phthalate metabolites MEP and MECPP to be inversely associated with HDL-cholesterol and triglycerides, respectively. However, these associations did not survive multiple testing. We did not find significant associations between EDCs and glycaemic traits or blood pressure. Some signs of non-linearity were found for blood pressure, and therefore results should be interpreted with care.

In Chapter 5, we investigated EDCs in a dynamic setting of diet-induced weight loss. We found that a reduction in caloric intake led to a decrease in exposure to phthalates with a high molecular weight. Parabens, bisphenol A and most low-molecular-weight phthalates remained similar over the course of the intervention. Concentrations of the low-molecular-weight phthalate MEP increased 1.8-fold during the intervention. Second, we found that a decrease in adipose tissue was not associated with an increase in excreted EDCs. Third, we found that higher baseline concentrations of methyl- and propyl paraben and the phthalate MBzP were associated with impaired weight loss after the intervention. In Chapter 6, we assessed associations between exposure to EDCs and the development of type 2 diabetes in a population with impaired fasting glucose, using repeated measurements. We found that high EDC urinary concentrations were not significantly associated with the development of type 2 diabetes. Contrary, high concentrations of phthalate metabolites were significantly associated with a reduced risk for type 2 diabetes in middle-aged women, a group in which direct associations have been previously reported (4).

In Chapter 7, we investigated the temporal stability of EDCs and their consistency using repeated measurements over the course of multiple years. We found that concentrations of most parabens, bisphenols and phthalates decreased up to 96% (propyl paraben) between 2009 and 2016 in the Netherlands. Second, we found the consistency of EDC concentrations between baseline and follow-up to be low. Intraclass correlation coefficients (ICCs) were highest for parabens (0.34 to 0.40) and poorest for bisphenols (0.15 to 0.23). For phthalates, we found consistency for short time intervals to be fair for some metabolites (e.g. ICC <48 months, MiBP: 0.48; MBzP: 0.44), but to reduce over time (e.g. >48 months, MiBP: 0.21; MBzP: 0.07). Further, we found that categorizing continuous EDC concentrations did not improve consistency, with a large proportion of individuals not remaining in their respective quartile.

In Part II, we shifted our focus to a broader set of risk variables for the development of type 2 diabetes and their applicability in risk prediction. In Chapter 8, we used a data-driven methodology to assess and contextualize 134 potential risk variables for the development of type 2 diabetes. We identified 63 risk variables of which none were solely driven by impaired fasting glucose at baseline. Even though a number of risk variables have been described in literature (5), we found novel associations for quality-of-life indicators and medications (e.g. proton-pump inhibitors, anti-asthmatics). Second, we found that the hazard caused by the increase of one standard deviation of HbA1c

9

as BMI, HDL-cholesterol and uric acid were required to change up to 2 standard deviations. For the majority of the variables, for example lifestyle changes such as dietary fat intake or work-related activity, it is impossible to achieve equivalence if focusing on these variables alone.

Next, we used a machine learning approach to assess the applicability of previously identified risk variables in risk prediction. We found that correlations between risk variables in general are modest. For variables which showed a stronger correlation, we did observe a clustering pattern for groups of risk variables with similar physiological origins. In contrast, we found that only a handful of risk variables were able to predict disease risk independent of similar risk variables, making most more or less interchangeable. Moreover, we found that the inclusion of up to three variables satiated prediction models. Also, we found that invasive measurements other than glycaemic traits (i.e. glucose, HbA1c) did only marginally improve the prediction model. Further, we found that hazard ratios of established risk variables were dependent on other variables in the model.

General discussion

The technical methodology behind measuring endocrine disrupting

chemicals

As parabens, bisphenols and phthalates have been receiving a growing amount of attention over the decades, the need to quantify these EDCs has resulted in the development of different analytical methods. One of these methods is enzyme-linked immunosorbent assay (ELISA). Although ELISA is convenient and useful for screening a large number of samples, it is unable to reliably measure low concentrations and is easily affected by irregular noise (6). Moreover, it is only able to measure one EDC at a time, making the assessment of multiple EDCs time consuming. Gas chromatography combined with mass spectrometry (GC–MS) does enable the quantification of multiple EDCs at low concentrations, but often requires a time intensive derivatization step for sample volatility making it less ideal for high throughput analysis. In contrast, liquid chromatography tandem mass spectrometry (LC-MS/MS) does not require derivatization while this technique is able to measure multiple analytes simultaneously. Therefore, we developed two ultra-performance LC-MS/MS methods for the measurement in urine of 21 different EDCs. By using ammonium fluoride in the mobile phase of the phenols method, and using acetonitrile as eluent in the phthalate method we were able to reach lower runtimes compared to methods described in literature (7–12). These low runtimes, together with the assessment of a large set of EDCs per method (i.e. 5 parabens and 3 bisphenols; 13 phthalate metabolites) makes our methods well suited for fast high throughput analysis. As the number of analytical methods rises, the need for harmonization between methods increases. Yet, interlaboratory comparisons between biomonitoring programs remain rare. Previously, a European project has compared bisphenol A and several phthalates between

different programs (13), in which they showed that even the reference laboratories showed considerable interlaboratory imprecision (relative standard deviations ranging from 6 to 39% for phthalates, and 11 to 20% for BPA). Compared to these findings, our newly developed methods show high levels of agreement and are therefore able to reliably measure parabens, bisphenols and phthalates in urine samples. As the quantification of a component forms the foundation on which research often builds, we highly encourage future comparisons between novel as well as established analytical methods to improve transparency and quality assurance.

Human exposure to parabens, bisphenols and phthalates is usually quantified in serum or urine samples. Although serum samples quantify direct exposure (i.e. EDCs that are in the human body at that specific moment), the quick metabolization and excretion of EDCs make serum concentrations present a short and potentially volatile snapshot of exposure. In contrast, urine can be collected over a longer time period, and therefore represents a more chronic time frame of exposure. Moreover, the collection of urine does not require invasive methods, making it suitable for large population studies. As the EDCs of interest are largely excreted within 24 hours after administration (14–16), we chose to use 24 hour urine samples for the determination of recent exposure to EDCs. Although the collection of 24 hour urine is more strenuous compared to spot- or morning-urine and requires motivated participants, it has shown less variation in EDC concentrations and fewer samples showing high exposure outliers (17). Moreover, urine dilution can be easily corrected for by the calculation of absolute excreted concentrations per 24 hour and does not require additional adjustments such as creatinine (18,19) which are prone to introduce inconsistencies (20).

Monitoring exposure to parabens, bisphenols and phthalates in the Dutch

population

Exposure to parabens, bisphenols and phthalates have been shown to be ubiquitous across the globe. Several countries use surveillance programs such as the National Health and Nutrition Examination Survey (NHANES; US), the Canadian Health Measures Survey (CHMS), the Korean National Environmental Health Survey (KoNEHS) and the German Environmental Specimen Bank (ESB) to monitor exposure (21,22). Other countries, such as Denmark, have been monitoring EDC exposure through active research programs (23,24). However, there are very few studies which have examined EDC exposure in the Netherlands, of which most focussed on pregnant women from the city of Rotterdam (25–27). We found EDC concentrations in a general Dutch population to be in line with the Rotterdam studies and concentrations reported in other countries (21–23,25,26). Moreover, we found that concentrations of most EDCs within individuals decreased over time. When we grouped concentrations per year of urine collection, we found a similar decrease in concentration between 2009 and 2016, an effect also observed in populations from the US, Denmark and Germany (21,24). A number of paraben and bisphenol concentrations showed strong decreases from one year to another, which coincides with

9

(28,29). This implies that legislation can have a limiting, thus positive effect on the use of EDCs in consumer products and subsequent exposure in the population. On the other hand, we observed concentrations of analogues such as bisphenol F to remain stable over time, a trend also reported in literature (24,30). As some legislation has recently (i.e. January and July 2020) become active (31,32) and the deadline of others is due next year (33), exposure is expected to further change in the future. Therefore, we recommend the continuation of monitoring potentially hazardous EDCs. Further, the lack of decrease of analogues such as bisphenol F exemplifies the shift to less known chemicals. As chemicals such as bisphenol F and bisphenol S have been proved to possess similar hormonal activity and endocrine disrupting effects as bisphenol A (34), they should be included in future biomonitoring studies.

Assessing the routes of exposure using a diet-induced weight loss program

Main routes of exposure to EDCs include ingestion, inhalation and transdermal exposure, and vary based on the foodstuff or consumer products in which the respective EDC is contained. We found the exposure to parabens to be independent of food products, which is in line with their main use in personal care and cosmetic products, and the literature pointing towards the main route of exposure to be transdermal (35,36). On the other hand, bisphenol A is used in a wide variety of food products (37), and concentrations have been shown to decline during a short period of fasting (38). However, we found that exposure to bisphenol A concentrations remained similar over a period of dietary weight loss (Chapter 5). This discrepancy may be caused by the run-in period prior to the intervention, in which individuals were consulted to meet an energy restriction to 100% of their energy requirements. As the median bisphenol A concentration at baseline was nearly half of that reported in a General Dutch population (Chapter 4), a change in type of food products rather than a quantitative restriction could have led to a decrease in exposure. This is in line with a study which showed that exposure to bisphenol A can be substantially reduced by avoiding specific nutritional products (39). The decrease which we observed in exposure to high molecular weight phthalates as a result of the dietary intervention is in line with food products being the main source of exposure (40,41). Most low molecular weight phthalates remained stable during the dietary intervention, further confirming their non-dietary routes of exposure (36,42). Yet, MEP concentrations showed a strong increase, suggesting that specific food-products play an important role in its exposure. A similar effect has been reported in a weight loss study, in which the absolute concentration of MEP decreased, but its proportion compared to other phthalates increased (43). Two studies have shown associations between high MEP concentrations and vegetable consumption (44,45), which may explain the source of increased exposure. Given its purported obesogenic properties (46), this increase is alarming and future research should further elicit its source of exposure in relation to weight reduction.

Long-term reproducibility of exposure to endocrine disrupting chemicals

We previously described our decision to use 24 hour urine samples to assess chronic EDC exposure as accurate as possible. However, we found a mostly poor consistency of EDCs over the course of multiple years, which could only be partly ascribed to the decreasing temporal trend. Even though the intraindividual variability has been modest to poor in literature (47), we found that this variability further declined after a follow-up beyond the scope of previous studies (48,49). These results have a large impact on studies which show associations between EDCs and disease development. The majority of these studies base exposure to EDCs on a single baseline measurement (4,50–52), thereby incorrectly assuming that this single measurement is representative for the exposure during the development of the disease. This is exemplified by our study focussing on associations between EDCs and the development of type 2 diabetes, in which we found that the inclusion of a repeated measurement at follow-up impacted the associations. Therefore, the results of this thesis stress the importance of repeated measurements in EDC-focussed research.

The literature describes many studies in which continuous EDC concentrations are divided in categories, after which the group with the lowest exposure is compared with the one with the highest exposure (51–53). Our findings showed that the categorization of EDCs did not improve the reproducibility compared to continuous variables. Moreover, roughly half of the individuals which were categorized in the highest or lowest category

Figure 1. Potential routes of exposure to endocrine disrupting chemicals determined by its response to caloric restriction. Potential routes of exposure to parabens, bisphenols and phthalates include absorption via the gastro-intestinal tract, inhalation and transdermal application. When individuals adhered to a 3-month dietary intervention limiting caloric intake to 33% of their daily requirements, we observed concentrations of parabens and low molecular weight phthalates to remain stable suggesting exposure through inhalation and/or transdermal application. Concentrations of high molecular weight phthalates decreased over the intervention suggesting that exposure largely occurs through the gastro-intestinal tract. Bisphenol concentrations were lower compared to a sample from the general population potentially due to the two-week run-in period prior to the baseline measurement, and remained similar over the course of the dietary intervention suggesting exposure through food-products, inhalation and/or transdermal application.

Parabens Bisphenols Phthalates

Low mol. weight High mol. weight

9

Therefore, categorization severely reduces data dimensionality (i.e. from continuous to few groups) while not increasing the consistency of the measurements and should be avoided.

Associations with adverse metabolic outcomes

High exposure to EDCs has been associated with adverse metabolic outcomes such as obesity and type 2 diabetes. In this thesis, we described a study investigating cross-sectional associations between EDCs and obesity-related traits, and a study focussing on obesogenic properties of EDCs in a dynamic setting of weight loss. Of the three EDC groups of interest, the obesogenic evidence of parabens in humans is weakest. First of all, few studies have investigated associations between parabens and obesity-related traits, of which all have a cross-sectional design. Even though some studies show significant associations between high paraben concentrations and increased adiposity-related traits (54,55), a large study conducted in NHANES observed the opposite effect (56). In our cross-sectional analysis, we found no significant associations between parabens and obesity-related traits. Yet, we did find an association between high baseline paraben concentrations and impaired weight loss after a dietary intervention in the, to our knowledge, first prospective study focussing on parabens so far. In contrast to parabens, bisphenol A is one of the most thoroughly researched EDCs in literature, and high concentrations have been associated with obesity-related traits in the US population (4,57,58). Yet, associations found in a Lebanese study did not hold up after adjusting for confounding factors (59). In the studies described in the current thesis, we did not find significant associations between bisphenol A and adiposity-related traits. As the significant associations were all reported in studies from the US, discrepancies may be population specific. A number of cross-sectional studies performed in NHANES showed that high concentrations of phthalates were associated with higher body weight (60–62). Moreover, prospective studies link high baseline exposure to weight gain over time (4,46). Our results add to this growing body of evidence. Further, we found that high phthalate concentrations are associated with impaired weight reduction, suggesting obesogenic properties in a dynamic setting of weight loss.

Second, we investigated associations between EDC concentrations and glycaemic traits (i.e. glucose, HbA1c), as well as investigated associations between EDC concentrations and the 5-year development of type 2 diabetes. Several prospective studies have tested the association for parabens, bisphenols and phthalates with the development of type 2 diabetes (50–53,63), with conflicting results. Here, we found that exposure to high EDC concentrations were not significantly associated with higher glucose and HbA1c levels in a cross-sectional setting. Moreover, higher EDCs were not significantly associated with the development of type 2 diabetes over time. When stratifying for age and sex, we did find significant inverse associations in middle-aged women. Direct associations between EDCs and type 2 diabetes have been previously described specifically in middle-aged

and not older women (50), suggesting that this subpopulation is especially susceptible for the endocrine disrupting effects of these chemicals. This may be explained by the fact that concentrations of natural hormones are very different between males and females, but also between premenopausal and postmenopausal women. As EDCs interact with the same system, co-exposure and susceptibility from natural hormones may lead to different effects.

Pathological processes underlying the obesogenic properties of endocrine

disrupting chemicals

We found a higher quantity of EDCs in human hypothalamus compared to white matter brain tissue. This may be explained by the central role the hypothalamus plays in hormonal-signalling pathways, making it more susceptible to hormones due to a more permeable blood-brain-barrier and relatively high vascularity (64). Further, we found higher methyl paraben concentrations in the hypothalamus of obese compared to normal-weight individuals. As the hypothalamus plays a central role in energy homeostasis, methyl paraben may induce adiposity via a local disrupting effect in the hypothalamus.

For persistent EDCs, weight loss has been shown to lead to mobilisation of EDCs from adipose tissue (65–67), increasing circulating concentrations and therewith potential hazardous effects. In case of obesogenic EDCs, this release in turn could lead to increased weight gain, creating a so-called yo-yo effect (65). Although non-persistent EDCs have been widely detected in human adipose tissue (68–71), we found that increased weight loss was not significantly associated with an increase in urinary EDC concentrations. Although a number of factors (e.g. fixed EDC intake, blood serum and adipose tissue measurements, more frequent urinary EDC measurements) should be controlled for to confirm our results, this yo-yo effect does not seem to be the consequence of non-persistent EDCs.

Regular risk variables for the development of type 2 diabetes

A plethora of different risk variables have been associated with the development of type 2 diabetes. Through our novel data-driven risk variable wide association study methodology (RV-WAS), we were able to independently discover and replicate a wide variety of risk variables. Our results were largely in line with conventional meta-analysis, and we discovered a similar proportion of risk variables as a recent umbrella-review (5). In contrast to conventional methods, our methodology enabled us to put the hazard ratios of the discovered risk variables into perspective. This led to the insight that although we identified a large number of risk variables, their impact was modest to weak. When we compared risk variables to the variable with the highest hazard ratio (i.e. HbA1c), only a handful of adiposity-related and biochemical variables were able to affect disease risk to a comparable extent. For all others, it was physiological not possible to attain a similar disease hazard and adjusted hazard ratios approached one. This was further exemplified by our novel machine learning approach in which we assessed the applicability of variables

9

risk variables suggested that each variable would be able to be of some value in risk prediction. Nevertheless, we found that most of these variables predicted the same part of disease risk and thus only a limited number of variables were able to contribute to risk prediction. Thus, even though the majority of the identified risk variables may attribute to the discovering pathophysiological pathways, they are not of use in risk prediction. Therefore, we recommend that the search for novel risk variables in for example the ‘omics field (e.g. genomics, proteomics and metabolomics) should primarily focus on the aetiology of the disease rather than attempting to outperform existing models.

When in turn excluding key variables, we found that slightly larger models were able to predict risk as well. This feat is reflected by the existence of a large body of slightly different risk prediction models with a similar performance (72,73). Our results imply that existing risk prediction models could be simplified without losing performance and should be reevaluated. Further, our findings show that measurements in blood or serum, with the exception of glucose and HbA1c, do not further contribute to the prediction of type 2 diabetes. As invasive measurements are costly, require venipuncture and analysis in specialized centers, they should be considered to be left out. Glucose and HbA1c measurements are becoming increasingly accessible at outpatient clinics and general practitioners, can be measured with a simple finger stick, and can therefore be easily implemented in screening methods for the development of type 2 diabetes.

Even though only a few key variables were able to maximize the prediction of type 2 diabetes, we found that the inclusion of other risk variables, while not improving the model discrimination, were able to attenuate hazard ratios of risk variables already included in the model. Therefore, risk variables should only be compared between study populations when similar covariates are included in the models.

Future perspectives

Narrowing down sources and reducing exposure to endocrine disrupting

chemicals

We showed that exposure to parabens, bisphenols and phthalates has been declining over recent years, but remains widespread with large differences between individuals. We further investigated the routes of exposure of different EDCs by distinguishing whether exposure was affected by a restriction of food-related products. For example, our findings suggest that exposure to high concentrations of the phthalate metabolite MEP originate from food products with a low caloric value such as vegetables, potentially due to a combination of packaging and its water solubility compared to other phthalates (74). These results provide the opportunity for future research to pinpoint the source of EDC exposure to a specific set of products. When considering large population-based cohort studies which include EDC exposure data and food-frequency questionnaires, such as Lifelines in the Netherlands, and NHANES in the US, one could use a methodology similar to the RV-WAS described

in chapter 8 to assess a the associations between EDC exposure and the consumption of a large number of food products to elicit high-exposure products. However, this approach will be solely useful for EDCs for which the route of exposure includes food-products. Further, food-frequency questionnaires do not give information on how the products are packaged. For example, even though the number of carrots consumed is known, it remains unclear whether these were plastic-wrapped and bought in a supermarket or were paper-packed and purchased freshly from a farm. Further discrimination between food products and packaging requires study designs specific for the task. By closely assessing the use of products including dietary intake, personal care products, product packaging, and cooking utensils (e.g. detailed diaries), one could narrow the potential sources of exposure down. Successively, the suspected sources of exposure could be omitted from a person’s lifestyle in an experimental setting. The potential of such a study has been shown previously, in which a study was able to reduce exposure to bisphenol A and DEHP metabolites with 53 to 66% by using fresh food products and avoiding them to come into

Figure 2. Risk variables fit for the prediction of type 2 diabetes development. The model performance for predicting the development of type 2 diabetes (determined by its discrimination, i.e. c-index) satiated after the inclusion of sex, HbA1c and glucose when considering all risk variables, and after the inclusion of age, body-mass index and waist-to-hip ratio when solely considering non-invasive variables. Although risk variables from most other domains were found to be associated with the development of type 2 diabetes, they did not contribute to risk prediction models next to the variables mentioned above.

9

addressed in this thesis can be particularly well monitored in such experiments due to the short timeframe between exposure and excretion (14–16). These projects would create insights in how to actively and accurately avoid exposure to EDCs on an individual level in parallel with the broader restrictions from government entities such as the European Union and the US Food and Drug Administration (FDA).

Determining hazardous effects of endocrine disrupting chemicals in

humans

We found that both a general decrease of exposure over time as well as a poor intra-person reproducibility resulted in a mostly poor consistency of EDCs over the course of multiple years. This shows that a single baseline measurement does not reflect chronic exposure during the timeframe in which a disease develops. Further, the inclusion of one baseline, one follow-up or both measurements affected the effect sizes and significance of associations, stressing the need for repeated measurements in order to carefully assess the effect of EDCs. While still rare, few studies have been including a second EDC measurement (53,63). We recommend that future studies include repeated measurements to monitor chronic exposure as closely as possible.

As described previously, we found that the associations between EDCs and the development of type 2 diabetes were largely driven by the subset of middle-aged women in line with a previous study (50). Future research should elaborate on these sex- and age-specific effects by at the least performing secondary analysis on these subsets. Otherwise, subgroup specific significant associations might be diluted and missed. Moreover, future studies should focus on the underlying reason behind these differences. Besides the menopause as described earlier, middle-aged women in the Netherlands often use hormone-based contraceptives. Through that route they are exposed to high concentrations of estrogens and/or progestogens. Adverse effects of hormone-based contraceptives include elevated blood pressure and weight gain. Therefore, it would be interesting to investigate how the use of contraceptives influences the associations between EDCs and metabolic diseases.

We primarily focussed on linear dose-response associations between EDC exposure and clinical outcomes, in which we expected both to constantly change in respect to each other. Although the majority of associations described in the literature are linear, there is a growing body of evidence which points towards non monotonic dose-response relationships of EDCs (75,76). For example, exposure to low and high concentrations can be strongly associated with a trait, while exposure to medium concentrations are not. Such a curvilinear relationship can be described as a so-called “U-shaped” curve (figure 3). For our association analysis in chapters 4 and 6, we assessed linearity by investigating the associations after grouping EDC exposure into categories. Even though the significant findings we found using continuous EDCs were confirmed to be linear, some associations

showed signs of non-monotonicity (e.g. blood pressure in chapter 4, or BPF and type 2 diabetes in chapter 6). A limiting factor for nonlinear association analyses is that they require a relatively large sample size. Although this is often hard to obtain in EDC-focussed research due to necessary biological samples (e.g. urine, blood), repeated measurements and biochemical analysis, we do acknowledge this as a limitation in this thesis and recommend the inclusion of nonlinear association analysis in future research.

The use of big data in disentangling complex multifactorial diseases

Complex diseases such as type 2 diabetes are characterized by the vast amount of risk variables with which they have been associated over the years making them complex, multifactorial diseases. In the past, association analyses between risk variables and complex diseases were performed in studies which were often solely designed for that specific analysis. However, this approach changed after the introduction of large, population-based cohort studies. These studies, such as NHANES, the UK Biobank and Lifelines, assess a plethora of different risk variables in sample sizes often exceeding 100,000 individuals. In contrast to traditional studies, cohort studies are not designed with one clear hypothesis in mind, but rather allow researchers to formulate their research question based on the available data. Although these cohorts facilitate a whole new area of research, several problems arise when combined with traditional scientific approaches. First of all, many studies report the association between one or a small number of risk variables and a disease in the same cohort. A simple search on “UK biobank” AND diabetes in PubMed resulted in 316 studies, and showed associations with body mass index, genetics, night shift work, commuting, body anthropometry, and grip strength in the titles of the first ten hits. Even though each of these studies investigated only one or few associations between a potential risk variable and the development of diabetes, the large amount of tests being

Figure 3. Dose-response curves. Associations between a concentration and a response can occur A.) linear, in which an increase in concentration is associated with a significant increase in response. Non-monotonic dose-response curves include B.) U-shaped curves in which concentrations at the low and high extremes are significantly associated with a response, C.) inversed U-shaped curves in which non-extreme concentrations are significantly associated with an increase in response.

9

discoveries. Moreover, it is unknown how the reported risk variables are correlated. For example, a high body-mass index may very well be highly correlated with grip strength and thus the association may be explained by an equal denominator; Yet both studies describe their findings in unrelated stories. In Chapter 8, we proposed novel, data-driven methodologies which we applied to the extensive dataset of the Lifelines cohort study to make use of its full potential while taking multiple testing and correlations between risk variables into account. Moreover, we were able to find key variables for risk prediction, which led to sparser models compared to literature while not reducing discrimination. These methods should be used in future studies to fully explore the available data for risk variables and their applicability in risk prediction in a systematic, data-driven way. As large, population-based cohorts with a wide set of phenotypic information are needed for these approaches, we encourage the formation of and easy access to such studies.

We used data-driven, machine learned approaches to investigate risk variables for the development of type 2 diabetes. However, many variations of this analysis can shed light on the complex development of this disease. For example, the same methods could be used to assess risk variables in the development of pre-diabetes, a diagnosis for which the treatment is much debated. Further, one should assess robust risk variables for the different complications of type 2 diabetes which could lead to a more targeted prevention approach. Moreover, differences in risk variables between complications could provide insights in the pathophysiology of different complications. Next to diabetes, these methodologies could provide valuable insights in other diseases with a complex etiology such as obesity, cardiovascular diseases and cancer. We therefore recommend the application of described methodologies beyond type 2 diabetes.

Conclusion

In this thesis, we observed that the Dutch population is ubiquitously exposed to a wide range of parabens, bisphenols and phthalates, and that the exposure has been declining over recent years. Further, we demonstrated the necessity of repeated measurements in prospective association studies, in which we found that exposure to high EDC concentrations was associated with obesity but not type 2 diabetes. When broadening our scope of risk variables for the development of type 2 diabetes, we discovered that many variables showed robust but modest associations, and only a handful were fit for risk prediction.

References

1. Frederiksen H, Aksglaede L, Sorensen K, et al. Bisphenol A and other phenols in urine from Danish children and adolescents analyzed by isotope diluted TurboFlow-LC-MS/MS. Int J Hyg Environ Health. 2013 Nov;216(6):710–20.

2. Frederiksen H, Jørgensen N, Andersson A-M. Parabens in urine, serum and seminal plasma from healthy Danish men determined by liquid chromatography-tandem mass spectrometry (LC-MS/ MS). J Expo Sci Environ Epidemiol. 2011 May;21(3):262–71.

3. Frederiksen H, Jørgensen N, Andersson AM. Correlations between phthalate metabolites in urine, serum, and seminal plasma from young Danish men determined by isotope dilution liquid chromatography tandem mass spectrometry. J Anal Toxicol. 2010;34(7):400–10.

4. Song Y, Hauser R, Hu FB, Franke AA, Liu S, Sun Q. Urinary concentrations of bisphenol A and phthalate metabolites and weight change: a prospective investigation in US women. Int J Obes . 2014 Dec;38(12):1532–7.

5. Bellou V, Belbasis L, Tzoulaki I, Evangelou E. Risk factors for type 2 diabetes mellitus: An exposure-wide umbrella review of meta-analyses. PLoS One. 2018 Mar 20;13(3):e0194127.

6. Fukata H, Miyagawa H, Yamazaki N, Mori C. Comparison of Elisa- and LC-MS-Based Methodologies for the Exposure Assessment of Bisphenol A. Toxicol Mech Methods. 2006;16(8):427–30.

7. Koch HM, Gonzalez-Reche LM, Angerer J. On-line clean-up by multidimensional liquid chromatography-electrospray ionization tandem mass spectrometry for high throughput quantification of primary and secondary phthalate metabolites in human urine. J Chromatogr B Analyt Technol Biomed Life Sci. 2003;784(1):169–82.

8. Silva MJ, Samandar E, Preau JL, Reidy JA, Needham LL, Calafat AM. Quantification of 22 phthalate metabolites in human urine. J Chromatogr B Analyt Technol Biomed Life Sci. 2007;860(1):106–12. 9. Kato K, Silva MJ, Needham LL, Calafat AM. Determination of 16 phthalate metabolites in urine

using automated sample preparation and on-line preconcentration/high-performance liquid chromatography/tandem mass spectrometry. Anal Chem. 2005;77(9):2985–91.

10. Heffernan AL, Thompson K, Eaglesham G, et al. Rapid, automated online SPE-LC-QTRAP-MS/MS method for the simultaneous analysis of 14 phthalate metabolites and 5 bisphenol analogues in human urine. Talanta. 2016;151:224–33.

11. Ye X, Kuklenyik Z, Needham LL, Calafat AM. Automated on-line column-switching HPLC-MS/MS method with peak focusing for the determination of nine environmental phenols in urine. Anal Chem. 2005;77(16):5407–13.

12. Zhou X, Kramer JP, Calafat AM, Ye X. Automated on-line column-switching high performance liquid chromatography isotope dilution tandem mass spectrometry method for the quantification of bisphenol A, bisphenol F, bisphenol S, and 11 other phenols in urine. J Chromatogr B Analyt Technol Biomed Life Sci. 2014;944:152–6.

13. Schindler BK, Esteban M, Koch HM, et al. The European COPHES/DEMOCOPHES project: towards transnational comparability and reliability of human biomonitoring results. Int J Hyg Environ Health. 2014 Jul;217(6):653–61.

14. Moos RK, Angerer J, Dierkes G, Brüning T, Koch HM. Metabolism and elimination of methyl, iso- and n-butyl paraben in human urine after single oral dosage. Arch Toxicol. 2016 Nov;90(11):2699–709.

9

15. Völkel W, Colnot T, Csanády GA, Filser JG, Dekant W. Metabolism and kinetics of bisphenol a in humans at low doses following oral administration. Chem Res Toxicol. 2002 Oct;15(10):1281–7. 16. Anderson WA, Castle L, Scotter MJ, Massey RC, Springall C. A biomarker approach to measuring

human dietary exposure to certain phthalate diesters. Food Addit Contam. 2001 Dec;18(12):1068–74. 17. Christensen KL, Lorber M, Koch HM, Kolossa-Gehring M, Morgan MK. Population variability of

phthalate metabolites and bisphenol A concentrations in spot urine samples versus 24- or 48-h collections. J Expo Sci Environ Epidemiol. 2012;22(6):632–40.

18. Johns LE, Cooper GS, Galizia A, Meeker JD. Exposure assessment issues in epidemiology studies of phthalates. Environ Int. 2015 Dec;85:27–39.

19. Preau JL Jr, Wong L-Y, Silva MJ, Needham LL, Calafat AM. Variability over 1 week in the urinary concentrations of metabolites of diethyl phthalate and di(2-ethylhexyl) phthalate among eight adults: an observational study. Environ Health Perspect. 2010 Dec;118(12):1748–54.

20. LaKind JS, Naiman DQ. Temporal trends in bisphenol A exposure in the United States from 2003-2012 and factors associated with BPA exposure: Spot samples and urine dilution complicate data interpretation. Environ Res. 2015;142:84–95.

21. Koch HM, Rüther M, Schütze A, et al. Phthalate metabolites in 24-h urine samples of the German Environmental Specimen Bank (ESB) from 1988 to 2015 and a comparison with US NHANES data from 1999 to 2012. Int J Hyg Environ Health. 2017 Mar;220(2 Pt A):130–41.

22. LaKind JS, Pollock T, Naiman DQ, Kim S, Nagasawa A, Clarke J. Factors affecting interpretation of national biomonitoring data from multiple countries: BPA as a case study. Environ Res. 2019 Jun;173:318–29.

23. Frederiksen H, Jensen TK, Jorgensen N, et al. Human urinary excretion of non-persistent environmental chemicals: an overview of Danish data collected between 2006 and 2012. Reproduction. 2014;147(4):555–65.

24. Frederiksen H, Nielsen O, Koch HM, et al. Changes in urinary excretion of phthalates, phthalate substitutes, bisphenols and other polychlorinated and phenolic substances in young Danish men; 2009-2017. Int J Hyg Environ Health. 2020 Jan;223(1):93–105.

25. Ye X, Pierik FH, Hauser R, et al. Urinary metabolite concentrations of organophosphorous pesticides, bisphenol A, and phthalates among pregnant women in Rotterdam, the Netherlands: the Generation R study. Environ Res. 2008;108(2):260–7.

26. Philips EM, Jaddoe VWV, Asimakopoulos AG, et al. Bisphenol and phthalate concentrations and its determinants among pregnant women in a population-based cohort in the Netherlands, 2004-5. Environ Res. 2018;161:562–72.

27. Philips EM, Jaddoe VWV, Deierlein A, et al. Exposures to phthalates and bisphenols in pregnancy and postpartum weight gain in a population-based longitudinal birth cohort. Environ Int. 2020 Jul 31;144:106002.

28. COMMISSION REGULATION (EU) No 358/2014 [Internet]. Official Journal of the European Union. 2014 [cited 2020 Jun 18]. Available from: https://eur-lex.europa.eu/legal-content/EN/TXT/ HTML/?uri=CELEX:32014R0358&from=EN

29. EFSA Panel on Food Contact Materials, Enzymes, Flavourings And Processing. Scientific Opinion on the risks to public health related to the presence of bisphenol A (BPA) in foodstuffs. EFSA Journal. 2015 Jan 1;13(1):3978.

30. Ye X, Wong L-Y, Kramer J, Zhou X, Jia T, Calafat AM. Urinary Concentrations of Bisphenol A and Three Other Bisphenols in Convenience Samples of U.S. Adults during 2000-2014. Environ Sci Technol. 2015 Oct 6;49(19):11834–9.

31. COMMISSION REGULATION (EU) 2016/2235 [Internet]. Available from: https://eur-lex.europa. eu/legal-content/EN/TXT/PDF/?uri=CELEX:32016R2235&from=EN

32. COMMISSION REGULATION (EU) 2018/2005 [Internet]. Available from: https://eur-lex.europa. eu/legal-content/EN/TXT/PDF/?uri=CELEX:32018R2005&from=EN

33. Directive 2011/65/EU [Internet]. Available from: https://eur-lex.europa.eu/legal-content/EN/TXT/ PDF/?uri=CELEX:32015L0863&from=FR

34. Rochester JR, Bolden AL. Bisphenol S and F: A Systematic Review and Comparison of the Hormonal Activity of Bisphenol A Substitutes. Environ Health Perspect. 2015;123(7):643–50.

35. Aubert N, Ameller T, Legrand J-J. Systemic exposure to parabens: pharmacokinetics, tissue distribution, excretion balance and plasma metabolites of [14C]-methyl-, propyl- and butylparaben in rats after oral, topical or subcutaneous administration. Food Chem Toxicol. 2012 Mar;50(3-4):445–54.

36. Janjua NR, Frederiksen H, Skakkebaek NE, Wulf HC, Andersson A-M. Urinary excretion of phthalates and paraben after repeated whole-body topical application in humans. Int J Androl. 2008 Apr;31(2):118–30.

37. Bhagwat VM, Ramachandran BV. Malathion A and B esterases of mouse liver-I. Biochem Pharmacol. 1975 Sep 15;24(18):1713–7.

38. Christensen KLY, Lorber M, Koslitz S, Brüning T, Koch HM. The contribution of diet to total bisphenol A body burden in humans: results of a 48 hour fasting study. Environ Int. 2012 Dec 1;50:7–14.

39. Rudel RA, Gray JM, Engel CL, et al. Food packaging and bisphenol A and bis(2-ethyhexyl) phthalate exposure: findings from a dietary intervention. Environ Health Perspect. 2011 Jul;119(7):914–20. 40. Koch HM, Lorber M, Christensen KLY, Pälmke C, Koslitz S, Brüning T. Identifying sources of

phthalate exposure with human biomonitoring: results of a 48h fasting study with urine collection and personal activity patterns. Int J Hyg Environ Health. 2013 Nov;216(6):672–81.

41. Serrano SE, Braun J, Trasande L, Dills R, Sathyanarayana S. Phthalates and diet: a review of the food monitoring and epidemiology data. Environ Health. 2014;13(1):43.

42. National Research Council. Phthalates and Cumulative Risk Assessment. 2008.

43. Dirtu AC, Geens T, Dirinck E, et al. Phthalate metabolites in obese individuals undergoing weight loss: Urinary levels and estimation of the phthalates daily intake. Environ Int. 2013 Sep;59:344–53. 44. Colacino JA, Harris TR, Schecter A. Dietary intake is associated with phthalate body burden in

a nationally representative sample. Environ Health Perspect. 2010 Jul;118(7):998–1003.

45. Trasande L, Sathyanarayana S, Jo Messito M, S Gross R, Attina TM, Mendelsohn AL. Phthalates and the diets of U.S. children and adolescents. Environ Res. 2013 Oct;126:84–90.

46. Lind PM, Roos V, Rönn M, et al. Serum concentrations of phthalate metabolites are related to abdominal fat distribution two years later in elderly women. Environ Health. 2012 Apr 2;11:21.

9

47. LaKind JS, Idri F, Naiman DQ, Verner M-A. Biomonitoring and Nonpersistent Chemicals-Understanding and Addressing Variability and Exposure Misclassification. Curr Environ Health Rep. 2019 Mar;6(1):16–21.

48. Starling AP, Engel LS, Calafat AM, et al. Predictors and long-term reproducibility of urinary phthalate metabolites in middle-aged men and women living in urban Shanghai. Environ Int. 2015 Nov;84:94–106.

49. Townsend MK, Franke AA, Li X, Hu FB, Eliassen AH. Within-person reproducibility of urinary bisphenol A and phthalate metabolites over a 1 to 3 year period among women in the Nurses’ Health Studies: a prospective cohort study. Environ Health. 2013 Sep 13;12(1):80.

50. Sun Q, Cornelis MC, Townsend MK, et al. Association of urinary concentrations of bisphenol A and phthalate metabolites with risk of type 2 diabetes: a prospective investigation in the Nurses’ Health Study (NHS) and NHSII cohorts. Environ Health Perspect. 2014 Jun;122(6):616–23. 51. Shu X, Tang S, Peng C, et al. Bisphenol A is not associated with a 5-year incidence of type 2

diabetes: a prospective nested case-control study. Acta Diabetol. 2018 Apr;55(4):369–75. 52. Salamanca-Fernández E, Iribarne-Durán LM, Rodríguez-Barranco M, et al. Historical exposure

to non-persistent environmental pollutants and risk of type 2 diabetes in a Spanish sub-cohort from the European Prospective Investigation into Cancer and Nutrition study. Environ Res. 2020 Jun;185:109383.

53. Rancière F, Botton J, Slama R, et al. Exposure to Bisphenol A and Bisphenol S and Incident Type 2 Diabetes: A Case-Cohort Study in the French Cohort D.E.S.I.R. Environ Health Perspect. 2019 Oct;127(10):107013.

54. Kolatorova L, Sramkova M, Vitku J, et al. Parabens and their relation to obesity. Physiol Res. 2018 Nov 28;67(Suppl 3):S465–72.

55. Kang H-S, Kyung M-S, Ko A, et al. Urinary concentrations of parabens and their association with demographic factors: A population-based cross-sectional study. Environ Res. 2016 Apr;146:245–51. 56. Quirós-Alcalá L, Buckley JP, Boyle M. Parabens and measures of adiposity among adults and

children from the U.S. general population: NHANES 2007-2014. Int J Hyg Environ Health. 2018 May;221(4):652–60.

57. Trasande L, Attina TM, Blustein J. Association between urinary bisphenol A concentration and obesity prevalence in children and adolescents. JAMA. 2012 Sep 19;308(11):1113–21.

58. Liu B, Lehmler H-J, Sun Y, et al. Bisphenol A substitutes and obesity in US adults: analysis of a population-based, cross-sectional study. Lancet Planet Health. 2017 Jun;1(3):e114–22.

59. Mouneimne Y, Nasrallah M, Khoueiry-Zgheib N, et al. Bisphenol A urinary level, its correlates, and association with cardiometabolic risks in Lebanese urban adults. Environ Monit Assess. 2017 Oct 23;189(10):517.

60. James-Todd TM, Huang T, Seely EW, Saxena AR. The association between phthalates and metabolic syndrome: the National Health and Nutrition Examination Survey 2001-2010. Environ Health. 2016;15:52.

61. Hatch EE, Nelson JW, Qureshi MM, et al. Association of urinary phthalate metabolite concentrations with body mass index and waist circumference: a cross-sectional study of NHANES data, 1999-2002. Environ Health. 2008 Jun 3;7:27.

62. Stahlhut RW, van Wijngaarden E, Dye TD, Cook S, Swan SH. Concentrations of urinary phthalate metabolites are associated with increased waist circumference and insulin resistance in adult U.S. males. Environ Health Perspect. 2007 Jun;115(6):876–82.

63. Wang B, Li M, Zhao Z, et al. Urinary bisphenol A concentration and glucose homeostasis in non-diabetic adults: a repeated-measures, longitudinal study. Diabetologia. 2019 Sep;62(9):1591–600. 64. Williams G, Harrold JA, Cutler DJ. The hypothalamus and the regulation of energy homeostasis:

lifting the lid on a black box. Proc Nutr Soc. 2000 Aug;59(3):385–96.

65. Jandacek RJ, Anderson N, Liu M, Zheng S, Yang Q, Tso P. Effects of yo-yo diet, caloric restriction, and olestra on tissue distribution of hexachlorobenzene. Am J Physiol Gastrointest Liver Physiol. 2005 Feb;288(2):G292–9.

66. Imbeault P, Tremblay A, Simoneau J-A, Joanisse DR. Weight loss-induced rise in plasma pollutant is associated with reduced skeletal muscle oxidative capacity. American Journal of Physiology-Endocrinology and Metabolism. 2002 Mar 1;282(3):E574–9.

67. Malarvannan G, Van Hoorenbeeck K, Deguchtenaere A, et al. Dynamics of persistent organic pollutants in obese adolescents during weight loss. Environ Int. 2018 Jan;110:80–7.

68. Artacho-Cordón F, Arrebola JP, Nielsen O, et al. Assumed non-persistent environmental chemicals in human adipose tissue; matrix stability and correlation with levels measured in urine and serum. Environ Res. 2017;156:120–7.

69. Wang L, Asimakopoulos AG, Kannan K. Accumulation of 19 environmental phenolic and xenobiotic heterocyclic aromatic compounds in human adipose tissue. Environ Int. 2015;78:45–50.

70. Mes J, Coffin DE, Campbell DS. Di-n-butyl-and di-2-ethylhexyl phthalate in human adipose tissue. Bull Environ Contam Toxicol. 1974 Dec;12(6):721–5.

71. Mathieu-Denoncourt J, Wallace SJ, de Solla SR, Langlois VS. Influence of Lipophilicity on the Toxicity of Bisphenol A and Phthalates to Aquatic Organisms. Bull Environ Contam Toxicol. 2016 Jul;97(1):4–10.

72. Abbasi A, Peelen LM, Corpeleijn E, et al. Prediction models for risk of developing type 2 diabetes: systematic literature search and independent external validation study. BMJ. 2012 Sep 18;345:e5900.

73. Noble D, Mathur R, Dent T, Meads C, Greenhalgh T. Risk models and scores for type 2 diabetes: systematic review. BMJ. 2011 Nov 28;343:d7163.

74. Staples CA, Peterson DR, Parkerton TF, Adams WJ. The environmental fate of phthalate esters: A literature review. Chemosphere. 1997;35(4):667-749

75. Lagarde F, Beausoleil C, Belcher SM, et al. Non-monotonic dose-response relationships and endocrine disruptors: a qualitative method of assessment. Environ Health. 2015 Feb 11;14:13. 76. Hill CE, Myers JP, Vandenberg LN. Nonmonotonic Dose-Response Curves Occur in Dose Ranges That