Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Towards personalised molecular feedback for weight loss

Towards personalised molecular feedback for weight loss Background: Numerous diets, apps and websites help guide and monitor dietary behaviour with the goal of losing weight, yet dieting success is highly dependent on personal preferences and circumstances. To enable a more quantitative approach to dieting, we developed an integrated platform that allows tracking of life-style information alongside molecular biofeedback measurements (lactate and insulin). Methods: To facilitate weight loss, participants (≥18 years) omitted one main meal from the usual three-meal routine. Daily caloric intake was restricted to ~1200KCal with one optional snack ≤250KCal. A mobile health platform (personalhealth.warwick.ac.uk) was developed and used to maintain diaries of food intake, weight, urine collection and volume. A survey was conducted to understand participants’ willingness to collect samples, motivation for taking part in the study and reasons for dropout. Results: Meal skipping resulted in weight loss after a 24 h period in contrast to 3-meal control days regardless of the meal that was skipped, breakfast, lunch or dinner (p < 0.001). Common reasons for engagement were interest in losing weight and personal metabolic profile. Total insulin and lactate values varied significantly between healthy and obese individuals at p = 0.01 and 0.05 respectively. Conclusion: In a proof of concept study with a meal-skipping diet, we show that insulin and lactate values in urine correlate with weight loss, making these molecules potential candidates for quantitative feedback on food intake behaviour to people dieting. Keywords: Obesity, Insulin, Lactate, Weight loss, Biomarkers, App development Background of developing heart disease, stroke, type 2 diabetes and Obesity is a global epidemic with increasing incidence certain types of cancers among other conditions [2, 3]. rates in developed and developing nations. For example, Thus, health-based concerns should be excellent motiv- the past 10 years have seen an approximate doubling in ating factors to lose weight. However, achieving weight the number of obese adults in the UK alone with loss is hard work and failure is demoralising. Most typ- body-mass-index (BMI; defined as weight divided by the ical weight loss programs include following a diet regime square of height, in kg/m ) values > 30. Male obesity (with or without drugs such as appetite suppressants), a rates rose from 13.2 to 24.4% and from 16.4 to 25.1% in fitness regime or a combination of these approaches. women over the period 1993 to 2012 [1]. Besides body Technological support available includes diet trackers image considerations, being overweight or obese can and activity monitors. Recording of eating patterns has raise blood pressure and concentrations of cholesterol, been recognized as an effective step in managing obesity as well as cause insulin resistance. This increases the risk [4, 5]. While the traditional paper version of the com- monly used dietary questionnaire is considered tiring [6], there are numerous computer-assisted versions [7, 8]as * Correspondence: j.klein-seetharaman@warwick.ac.uk Systems Biology and Biomedicine, Division of Metabolic and Vascular well as apps and websites available for personal tracking of Health, Medical School, University of Warwick, Gibbet Hill, Coventry CV4 7AL, food intake [9, 10]. However, regardless of interface, UK mis-reporting of food intake is a well-documented Institute for Digital Healthcare, Warwick Manufacturing Group, University of Warwick, CV4 7A, Coventry, UK © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Tejpal et al. BMC Obesity (2019) 6:20 Page 2 of 15 problem [6, 11], as it depends on self-awareness, honesty Besides food composition, another factor amenable to and motivation of the user [5]. Often, unconscious bias of behaviour change is meal timing. Different studies have self-observation leads to under-realization of foods eaten suggested that consuming a regular breakfast is associ- [12]. Energy expenditure on the other hand can potentially ated with lower body weight [27–30]. There is also a be tracked without bias using activity monitors, but they plethora of metabolic studies that support the notion do not provide a direct link to weight loss. Even if a device that eating breakfast is preferable over eating dinner due that accurately measures caloric intake and expenditure to a phenomenon referred to as “evening diabetes” in was widely available, the information may not be sufficient which insulin sensitivity is higher early during the day to motivate a user to make changes in their behaviour that making it more likely to burn fat [31]. A crossover study would result in weight loss. This is evidenced by the fact comparing days with breakfast and lunch versus days that even when meticulously keeping records of food with lunch only showed different effects on clock gene intake, individuals still find it hard to lose weight [13]. expression in healthy and type 2 diabetic subjects, while This is because the relationship between caloric intake skipping of breakfast showed an altered response for and weight loss is not linear [14]. As a result, current clock gene profiles in both groups [32]. On the other approaches to lose weight loss generally do not work hand several studies indicate that breakfast eaters and well [15], and the weight loss market is missing a skippers didn’t vary significantly in terms of body weight device that is more directly coupled to the desired and nutrient intake [33–36]. A study by Gill and Panda outcome, weight loss. The prerequisite for such a [37] has demonstrated the extent to which adults display device is a quantitative and science-based biomarker an irregular daily and weekly rhythm of eating-fasting, of weight loss that has the potential to provide a which could be manoeuvred to obtain desirable health biological feedback loop to the dieter. benefits. It might be possible to cater to such personal Quantitative biomarkers of weight loss do not exist preferences, at least to some extent, without compromis- yet, but biomarkers have been considered for a number ing weight loss success. For example, from a metabolic of related areas. For example, urinary metabolic markers perspective it may be most efficient for weight loss to for cardiovascular disease, blood pressure and adiposity skip meals other than breakfast, especially dinner. How- have been identified [16, 17]. Several metabolomics stud- ever, from a behavioural and/or cultural perspective, ies involving untargeted proton ( H) nuclear magnetic breakfast and lunch may be the easiest meals to skip, resonance spectroscopy and ion exchange chromatog- and dinner the hardest. Weighing between these alter- raphy on obese human and mice urine samples have nate decision making strategies requires quantitative identified metabolites associated with BMI and adiposity analysis. Finally, to develop a tool for personalised feed- [16, 18]. Angiotensin converting enzyme (ACE) has been back there is also a need to understand the willingness shown through blood profiling for protein and steroid of individuals to collect samples in order to obtain meta- hormones to be an important predictor for sustained bolic biomarker information. weight loss [19]. Biomarkers for different foods include Here, our goal was to lay the foundation for a molecu- NMNA, a niacin-related (vitamin B3) metabolite marker lar feedback approach to assist dieting efforts. To this for coffee drinking [20], proline betaine for citrus fruit end, we aimed at obtaining a better understanding of the consumption [21] and O-acetyl carnitine for red meat complex interplay between possible molecular bio- intake [22]. Thus, while biomarkers have been consid- markers of dieting behaviour, individuals' personal ered for several related purposes, none have been tested preferences, their willingness to collect information, the yet as a means to provide molecular feedback to drive to act on the collected information and weight loss someone following a diet. success. Specifically, we asked 52 dieters to record meal The requirements for dieting feedback are more timing and composition using an electronic diary inter- immediate than the general adiposity markers, or the face, along with collection of urine samples, and weight. long-term ACE weight loss marker. Our goal is to meas- The urine samples were analysed for lactate and insulin ure a molecular marker that directly or indirectly reports concentrations and the information was integrated. Our on dieting behaviour so that we can use it for feedback. results provide a strong proof of concept that each We have chosen insulin and lactate because these are molecule studied, but especially insulin, can be used for both known to vary with consumption of glucose and biofeedback on dieting behaviour. other food, as well as caloric intake overall [16, 23–25]. Caloric restriction is one of the most widely used strat- Methods egies for losing weight [26, 27] but it requires strong will Study design and population power, and thus identifying biomarkers of weight loss Participants were recruited by website, flyer and newsletter and using them for dieting feedback may make caloric advertisement at the University of Warwick and through restriction easier on dieters. word of mouth. To be eligible for participation, individuals Tejpal et al. BMC Obesity (2019) 6:20 Page 3 of 15 had to be 18 years or older and clinically healthy. Sample collection and storage Exclusion criteria included diagnosed diabetes and Most samples were dropped off at the departmental pregnancy. The study was approved by the Warwick front desk and the investigators collected the samples Medical School Ethics committee BSREC (protocol and transferred them to -80 °C upon notification by the identification REGO-2014-1318). Everyone who receptionist. In some cases, samples were handed over expressed interest in the study was referred to details to investigators directly by participants. on the website personalhealth.warwick.ac.uk.Those who remained interested, met with the PI of the Mobile application study, and were provided with instructions on how to The mobile and web application was developed using use the interface. We also provided them with the Intel XDK, a framework that allows building of urine sample collection kit and instructions on how cross-platform mobile applications, including Android, to use it and how to transfer samples to the investi- iOS and Windows along with a web browser interface gators. All participants provided informed written (http://personalhealth.warwick.ac.uk). The program- consent at this time. Not all participants who were ming was using standard web technologies such as given the information and sample collection kit actu- HTML, Javascript and CSS. A responsive web design ally entered data and returned samples. Sample tubes approach was adopted in implementing the applica- were labelled with numbers that anonymized the sam- tion to enable device specific optimal user experi- ples. Data entered into the digital health platform was ences. Front end browser and mobile app matched to molecular information only through the functionality was built using AngularJS, a Javascript sample tube numbers. framework. All back end support for the mobile app Participants were asked to collect data for at least one and the web interface were built using Java and 24 h period. In total, participants provided data for 149 MariaDB database server. Back end automation for days. The average number of 24 h periods was 3 days/ providing users with graphical feedback was programmed participant, and the maximum number of data was using R. All data communication between the server and entered by one participant who provided 24 periods of mobile and web applications was through HTTPs. A 24 h data each. built-in interface between our server and the FATsecret A control day was defined as a day in which partici- database (https:\\platform.fatsecret.com) was developed so pants provided samples and data where they consumed that study participants could seamlessly search for nutri- a regular 3-meal diet. An intervention day was defined tion information without leaving our website or app. as a day in which they provided samples and data where Where users did not retrieve data from the FATsecret they skipped at least one meal. Since on average only 3 database, investigators carried out the searches subse- days of samples and data was provided, a washout period quently. FATsecret provides free access to their database was not necessary. through an API (up to 5000 calls a day were allowed). The database is very large and allows retrieval of nutritional information for the majority of food and drink items (see Demographics data collection below). All participants were given the option to enter informa- tion about their age, BMI, sex, weight and ethnicity at Meal plans the time of sign-up to the digital health platform. The Participants were asked to provide data and samples for at majority of participants provided this information and least one control day and at least one diet day (see this was used to carry out demographic analysis. In some definition above). For diet days, participants were cases, BMI was calculated by the investigators from recommended to skip lunch (L) or dinner (D), but were weight and height entries directly. Participants were not also allowed to skip breakfast (B). They were also asked to specifically recruited to target any particular BMI. How- restrict the total caloric intake for the day to less than ap- ever, for analysis purposes, the data was also divided by proximately 1200KCal. An optional snack (S) of <250KCal grouping individuals according to their BMI values. Indi- was allowed once in a day. B, L, D, S were defined viduals with a BMI of 18.5–24.9 kg/m were considered according to i) calorie intake (<250KCal for S, normal; 25.0–29.9 kg/m were overweight; and those >250KCal for B, L, D) summed over nearby entries above 30.0 kg/m and greater were obese [1]. The goal of (within 30 min) and ii) timing, being B before noon, the study was not to compare obese with non-obese L between noon and 3 pm and D after 4 pm (Fig. 1a). participants, but a separation of the data into BMI A control day had three or more meals on that day. groups was useful to identify BMI as a source of According to these specifications we defined different variation and to delineate possible trends in the data that meal plans to systematically describe the participants’ may relate to BMI. preferences:i)B,L,[S]; ii) B,D, [S]; iii)L,D, [S];iv) Tejpal et al. BMC Obesity (2019) 6:20 Page 4 of 15 The caloric intake was calculated based on the meal information provided by the user. The individuals either used the pre-specified meal options from the digital health database or entered detailed descriptions of their prepared meals. The food (in grams) breakdown in carbohydrate, fat and protein was obtained from the fat- secret database (available at https:\\platform.fatsecret. com). For analysis purposes, it was further converted into KCal by multiplying with 4, 4 and 9 per gram of carbohydrates, protein and fat, respectively. FatSecret was chosen as it had a mature API that could be used by our website and app to calculate nutrition information without maintaining our own nutrition database. In addition to offering a large and curated database of com- mon foods and packaged products, FatSecret also offers a professional interface for health practitioners to moni- tor app usage of their patients [38]. It has also been pre- viously used in several studies as a caloric counter to understand effects of tracking information (on mobile applications) on weight loss [38–40]. Entries in the database were used to determine every participants’ caloric patterns, meal plan preferences and to compare the different plans to the weight changes ob- served over the respective 24 h period along with insulin and lactate profiles. In all cases, only those records were analysed that contained a minimum of two meal records for the day (from 00:00 h to 12:00 h day + 1). No instruc- tions were given to participants with respect to food composition, and any food item was allowed. Some of the food options chosen by participants are depicted in Fig. 1a. The fasting times for diet days were obtained by sub- tracting the time of the second meal from the first meal of the day. Similarly, the overnight fasting period was obtained except that only those days that had following day weight information were considered. Insulin measurements using mesoscale luminescence assay Insulin in samples was measured using a Mesoscale Discovery Human Insulin Kit containing (catalogue number: K151 BZC-2) 96 well plates coated with insulin antibodies obtained from Mesoscale Inc. (www.mesoscale. Fig. 1 Study Design. a Meal plan layout for the participants in a com). The assay was performed according to the manufac- 24 h period. b Flow diagram of the study design. c Comparison between demographic features of the study participants and the UK turer’s instructions. The plates were analysed on a SEC- population. Overall numbers for the UK population was the TOR Imager 6000 system. All samples (urine and plasma) arithmetic middle between the male and female values based on were centrifuged prior to analysis at 12000 rpm for 5 min the assumption that the distribution of male and females in the at room temperature. Insulin calibrators (supplied by the statistics was approximately 50%. In this study, there was more kit) were run in duplicate to generate an 8-point standard females than males, so the overall number was obtained directly from the raw data curve covering the 0–50,000 pg/mL range. The standard curve was modelled using least squares fitting algorithms so that signals from standards with known concentrations B or L or D, [S] for dieting days; and v), B,L,D, [S] of insulin can be used to calculate insulin concentrations or > 3 meals for control days, where [S] denotes op- in samples. The MSD Discovery Workbench® analysis tional snack intake (Fig. 1b). software was used to calculate the concentration of insulin Tejpal et al. BMC Obesity (2019) 6:20 Page 5 of 15 in samples. The software uses a 4-parameter logistic The survey was divided into three parts: model and includes a 1/Y2 weighting function. This allows Section I: Motivation. Several options were provided to for a better fit of data over a wide dynamic range (3–4 the participants such as interest in losing weight, diet, logs), particularly at lower insulin concentrations. The metabolic profile, health platform and being involved in wide dynamic range of the assay allowed for the quantifi- medical research. These parameters were analysed cation of insulin in urine without the need for dilution nor together and when separated into 10-year bin-sized age concentration. groups. Section II: Dropout. The survey was used to identify Lactate measurement the reasons for dropout such as difficulty of diet and Lactate dehydrogenase is the enzyme responsible for sample collection, time consuming, complicated health interconversion of lactate to pyruvate following reduc- platform. The participants were also provided a free text tion of nicotinamide adenine dinucleotide (NAD) to its field to enter other factors that contributed to dropout. reduced form (NADH). To measure lactate, the reaction Section III: Feedback. The participants provided feed- needs excess NAD. To force the reaction to completion back on the health platform and sample collection. Their in this direction, it is necessary to trap the formed pyru- personal input was requested on suggestions for the vate with hydrazine present in glycine-hydrazine buffer. platform’s improvement. All samples were centrifuged prior to analysis at 12000 The information provided by the individuals was used rpm for 15 min at room temperature. The experiments to calculate the weighted average of each contributing were conducted in 96 well solid black fluorescence factor for motivation to participate, reasons for dropout plates (Thermo-Scientific, catalogue #: 634-0006). The and feedback about the study. stock solution of 10 mM of lactate was reconstituted in glycine-hydrazine buffer (0.6 M glycine and 0.5 M Statistical analysis hydrazine, pH 9.2) bought from Sigma-Aldrich, UK. This Statistical analyses were performed using IBM SPSS was used to prepare standard reactions in the range of Statistics 24 and R. Association between different vari- 25-100 μM lactate concentrations. A reaction mixture ables was calculated using bivariate Pearson Coefficient stock solution containing 10 mg NAD with 2.0 mL glycine analyses. Nonparametric Mann-Whitney U test was per- buffer, 4.0 mL water 0.1 mL L-lactate dehydrogenase formed in some cases as indicated. One-way analysis of (Sigma-Aldrich, UK). 20 μl of standard (in duplicates) or variance (ANOVA) was used to compare values from sample and 130 μl of reaction mixture were added to each control days with values from dieting days. well. The plate was then incubated at 37 °C for 15 min. The fluorescence was read using a Perkin Elmer Wallac Results 1420 Victor2 Microplate Reader with excitation at Study design and data collection 345-355 nm and emission at 450-460 nm. Initially, 146 individuals recruited by flyer and newsletter advertisement expressed interest in our study. Of these, Survey design 52 individuals became study participants (77% females A short self-administered questionnaire link was sent via and 23% males) who provided data and samples. They email to everyone who originally expressed an interest in received access to a web application at personalhealth. the study (including but not limited to those who actu- warwick.ac.uk, as well as an app “Digital Health Plat- ally enrolled), inviting them to participate in the survey. form” for android and apple devices available in google The survey was anonymous and the data from the play and iTunes stores, respectively. Through this plat- survey was not linked or compared with the data en- form, they entered life-style related data, including tered by participants during the weight loss study. The weight, food and drink intake, exercise, and urine sample questionnaire (Additional file 1:FigureS4) wasde- collection details. The mobile health platform creates a signed using Google forms. A total of 48 people sub- timeline of the logs or events that are entered by the mitted the study questionnaire, all of whom had user. This electronic information is sent to a web server consented to participate in the survey and stated pre- that allows users to store their information securely and vious experience on entering data onto the platforms. access it anywhere using either a web browser based The individuals participating in the survey were not interface or a native mobile application from their smart linked to their identification in the platform, as the phones or tablets. In addition to being a tool for logging survey google document and digital health platform time and other parameters, the application also serves to were independent of each other. The participants seamlessly share information between the user and the were provided with several options under each section analyst. It allowed researchers and cohort group partici- (see below) along with an option to enter other fac- pants to register, and manage the logistics of data collec- tors contributing to the study. tion. Researchers obtained analysis files in anonymized Tejpal et al. BMC Obesity (2019) 6:20 Page 6 of 15 fashion only through the website administrator. Ease of indicates meals, green snacks and blue/brown low calorie use and cross-platform support were the most important drinks (including water and coffee). Most entries for among the factors considered in the design of the health caloric intake of >250KCal (i.e. a meal) were observed in platform. just one hour from 6 to 7 pm (Fig. 2e). Many breakfast (B) Urine samples were used to measure insulin and “meals” were low in calories and were therefore classified lactate concentrations which were uploaded onto the here as snacks (S), see below. Caloric intake is significantly platform. Participants collected samples and life-style different for males and females (p = 0.05, Additional file 1: data for control and diet days (see Methods). Figure S2). Demographics of study participants Diet behaviour: meal plan preferences According to the UK Health and Social Care Informa- Out of the 52 participants, at least one entry with tion Centre, the prevalence of overweight individuals in two meals (see Methods) was available for 43 people. the UK population is age- and gender-dependent, with Since participants could freely chose the number of 9% (male) and 13% (female) in the 16–24 age group and days they participated in the study, the number of 13% (male) and 35% (female) in the 50–69 age group days for which data was available varied for each [1]. A similar pattern characterized the participants in participant. The majority followed the study plan for our study (Additional file 1: Figure S1). Grouping partic- 1–2 days, while one participant collected data for up ipants by age showed that the number of overweight to 24 days. Thus, the 43 individuals collectively pro- study participants was lowest among younger adults vided data for 147 days consisting of both, control (28) (20–29 year old group, Additional file 1: Figure S1a), and diet (119) days. Participants were given a rela- increasing through middle age (ages 30–59, Additional tively free choice in meal plans, with the only restric- file 1: Figure S1b-d), and only reducing among the oldest tions being the omission of one main meal and the participants (ages 60–69, Additional file 1: Figure S1e). total caloric intake as described in Methods The meal The majority of study participants were in the normal plan choices made by participants on the 147 days is and overweight groups (Additional file 1: Figure S1f). shown graphically in Fig. S3a. BL, the meal plan that The mean BMI of 27.2 kg/m observed in the UK popu- would be metabolically optimal from a theoretical lation [1] parallels that of 27.0 ± 5 kg/m (mean ± stand- perspective (see Introduction), or the slightly modified ard deviation) in our study. Similarly, the weight and BLS meal plan, were followed only on 10 days. 19 days height values split by gender also mirror those of the UK corresponded to the LD plan, while the largest num- population (Fig. 1c). This indicates that the sample of 52 ber of 46 days was in the BD or BDS category. BDS participants is a good representation of the UK popula- was followed on 39% of the dieting days, and was tion. The mean BMI of males and females in our study thus the most popular meal choice, while the B [S] are 26.0 and 28.0 kg/m , respectively, which indicated plan accounted for only 8% of the dieting days. A that they were significantly different (p < 0.001) within graph of the spread of meal timing of individuals our study group (Additional file 1: Figure S2). shows that participants followed similar eating pat- terns for all days if they provided samples and data Diet behaviour: caloric intake pattern for more than one day (Additional file 1:Figure The timings of health platform entries on the 149 days S3b). Another frequently followed meal plan was of data entered by participants show a wide spread from that of the single meal: 46 days had only one meal B, 7 am - midnight on a 24 h scale (Fig. 2a), with only night L or D (sometimes plus optional snack, B[S] or L[S] time (midnight to 7 am) receiving very few entries, in or D[S]). This large number likely arose from the line with previous observations [37]. There was a higher fact that we classified what participants may have percentage of total entries on the health platform in the thought of as “meals” as snacks based on the mornings and evenings, namely 33 and 32% of the total 250KCal cut-off. In total, there were 26 control days entries, respectively (Fig. 2b). Many of the morning en- (18% of the 147 days), where people have had at tries were weight and urine sample collections. When least three meals (BLD, BLD [S] or more). On these only food entries are plotted, entries cluster in the morn- control days, caloric intake was significantly higher ing (around 7 am), at lunchtime (around 1 pm), and in (p < 0.01) than on diet days, as expected, although the evening, peaking at 6 pm (Fig. 2c). When entries were there were many days of low calorie intake among quantified by calories consumed, one can see that the lar- control days. Notably, individuals did not lose weight gest calorie intake was in the evening, with 22 and 51% of on those caloric restricted control days, suggesting the total calories recorded from 7 to 11 am and 4 to 9 pm, that meal timing plays an important role, perhaps respectively (Fig. 2d). A more detailed break-down of en- more than caloric intake for losing weight tries as % food events per hour is shown in Fig. 2e. Purple (Additional file 1:FigureS3c). Tejpal et al. BMC Obesity (2019) 6:20 Page 7 of 15 Fig. 2 More calories are consumed at dinner and breakfast and dinner combination were more popular than breakfast and lunch. a Polar plot of all entries of each individual plotted against the time of day (angular axis). Data from 52 individuals are shown. 24 h rose plots showing (b) percentage of total entries from individuals, (c) percentage of ingestion events and (d) % of calories consumed. e Percentage of food events in 1 h bins. The radial axis for each rose plot shows % of events Dieting success by study participants’ meal plan choice kg, and no change =0 kg. Figure 3a shows the % of partici- Weight change data was available for only 43 out of the pants with weight change in each of these groups. One 52 participants for at least one 24 h period, reducing the can clearly see that all diet meal plans resulted more often total of 147 days to 126 days. For ease of analysis, we in weight loss as compared to the control days. Figure 3b grouped the weight change values into 3 groups: weight shows the more detailed split into sub-groups taking loss when the weight difference between the beginning whether or not a snack was eaten into account. Overall and end of the 24 h period was > 0 kg, weight gain for < 0 there does not appear to be a negative consequence of Tejpal et al. BMC Obesity (2019) 6:20 Page 8 of 15 Fig. 3 Weight loss is associated with fasting time and consumption of calories. a Effect of dieting on weight with respect to different meal groups. N/A refers to the days for which weight loss data is not available. Weight change is defined as weight loss (any change > 0 kg), weight gain (any change < 0 kg), and no change (=0 kg). One-way ANOVA analysis comparing control with other meal groups show significant difference at p = 0.01** (p < .001). b Effect of dieting on weight with respect to different meal plan subgroups. c Plot of total caloric intake against weight difference. Pearson’sR = − 0.21 correlation is significant at the 0.05 level (p < 0.05). d Plot of overnight fasting time against weight difference. Pearson’s R = -0.21 correlation is significant at the 0.05 level (p < 0.05) having the additional snack, although the size of the data Correlation R = -0.21, p = 0.016) (Fig. 3d),i.e.the is too small to ascertain the statistical significance of this longer the fasting the greater the weight loss. statement. Because we do have meal plan information for 21 days without weight change information, we included a fourth group “NA” (purple) in Fig. 3a, b for completeness. Motivation: reasons to participate in the study We conducted one-way analysis of variance (ANOVA) A survey was conducted to understand the reasons why comparing the control group with each of the other people were interested in a study that involved both, groups in Fig. 3a (i.e. BL, BD, LD and B or L or D). Each weight loss and urine sample collection (Additional file 1: group was significantly different from the control group Figure S4). Interest in losing weight, involvement in (p < 0.01). This indicates that skipping a meal results in research and knowledge of metabolic profile were the weight loss irrespective of which meal of the day is main drivers behind participation. There may be a skipped. Comparing weight loss with total caloric intake difference in motivation for different age groups, as showed an inverse relation with Pearson Correlation the 20–29 and 40–49 year age-group more often significant at p =0.05 (Fig. 3c). Finally, on days when reported interest in their metabolic profile (33%), participants achieved weight loss, the length of over- while the 30–39, 50–59 and 60–69 year age groups night fasting periods was inversely correlated to were more motivated by losing weight (32, 32 and weight loss expressed as negative kg values (Pearson 20% respectively). However, because of the small Tejpal et al. BMC Obesity (2019) 6:20 Page 9 of 15 number of participants, we cannot ascertain if these lactate and insulin concentrations. In particular, the differences are statistically significant. weight difference (expressed as negative kg) showed a correlation with total calorie intake, which was signifi- Dropout analysis cant at R = 0.04 (p < 0.05). Total insulin and total lac- Dropout rates in weight loss studies have been a prom- tate were positively correlated to the total calorie inent concern when promoting lifestyle and dietary intake (p < 0.001, R = 0.35 and R = 0.03, respectively). changes in overweight and obese populations, as well as Fasting, total, last, following day and maximum affecting the validity and generalisation of conclusions in amounts for insulin and lactate had significant correl- weight loss studies [41]. In our study, we observed a ation with carbohydrate, fat and protein content in similar trend. At the first meeting, people were informed the meals (Fig. 4d, e). of the study requirements. At this stage, of the 146 adults who had shown an interest, 70 dropped out, leav- Biomarker and BMI ing only 76 individuals who provided written consent for Because BMI was correlated with a number of param- participation in the study and received sample collection eters (Fig. 5), we investigated if pre-defined BMI kits. Of these, 52 participants actually provided urine groups differed in correlation of parameters (Fig. 5b). samples and life-style information through our online/ Segregation of the data into different BMI groups mobile platforms. Thus, the dropout rate after the first showed loss of correlation between weight loss and meeting of 48% reduced to 16% when comparing to the other parameters in the obese and overweight groups initial number of people interested in the study, and 34% while being sustained in the healthy group. Particu- when comparing to the previous step (Fig. 1b). larly, the insulin biomarker profiles in the overweight To identify the reasons for dropout in our study, and obese group are dampened in comparison to the we designed a number of questions (Additional file 1: healthy group (Fig. 6a).Total,fasting,last,following Figure S4). Busy schedule, complicated samples day and maximum insulin values were significantly collection and loss of motivation correspond to 25, 21 higher in the obese group in comparison to healthy and 18% of the reasons chosen by people who partici- individuals (Fig. 6b). Also, total and last lactate pated in the survey, respectively. Apart from amounts increased in obese people in comparison to pre-defined reasons, individuals also entered their per- the healthy group (Additional file 1: Figure S7a). Fur- sonal hurdles through a free text option. Participants thermore, total, maximum and minimum lactate found it hard to follow caloric restriction guidelines values were higher in obese than in overweight indi- duetotheir active worklife or thepsychological viduals (Additional file 1: Figure S7b), in accordance stress given by the word “diet”. The fear of eating with previous findings of increased lactate levels in more after a day of dieting also made people drop obese individuals [42]. out from the study. In addition, since the individuals in our study were UK based, they found it difficult to maintain the food diaries as the fatsecret database we used was an American foods database. Biomarker and weight loss Because weight loss is the desired outcome for most Molecular insulin and lactate biomarker correlate with dieters, we plotted weight change versus biomarkers life-style data in Fig. 6c. Because of inaccuracies inherent to meas- The urine samples collected by the participants were uring weight, we grouped the weight change values used to measure insulin and lactate concentrations. into 4 groups: weight loss > 0 .5kg, weight loss 0.1– These values were then used to extract a total of 23 0 .5kg, weight gain and no change. One can see very parameters relating to biomarker profiles or lifestyle data clearly that total insulin values vary most dramatically entered (Additional file 1: Figure S5). The in the weight gain group, and are overall higher in cross-correlation matrix of all the 23 extracted parame- the no weight and weight gain categories. Similar pat- ters from biomarker profiles and the digital health terns were also observed for fasting, last, following platform are shown for the complete cohort in Fig. 5a day and maximum insulin values. This graph thus for an overall summary. The weight difference showed a emphasizes that insulin values, even individual ones, positive correlation with BMI while a negative correl- as opposed to all values collected over a 24 h period, ation with carbohydrates, fat, lactate before second meal are potentially useful biomarkers for immediate feed- of the day and total calories was observed. Furthermore, back on dieting behaviour, with low values being as expected, the total lactate and insulin parameters likely predictive of weight loss, information which can were strongly correlated with other parameters such as only be obtained the day following a diet, too slow to first, last, maximum, minimum and following day be sufficiently motivating. Tejpal et al. BMC Obesity (2019) 6:20 Page 10 of 15 Fig. 4 Individual correlation plots of selected parameters. a Weight difference versus total calories. b Total insulin versus total calories. c Total lactate versus total calories. d Insulin parameters correlation with nutritional parameters: Panel I. Carbohydrate. Panel II. Fat. Panel III. Protein. e Lactate parameters correlation with nutritional parameters, as in (d). Significant correlations with macronutrient content were marked by ** or *, when significant at p = 0.01** and p = 0.05*, respectively Discussion long-term, and cannot be used for immediate feedback The escalating obesity epidemic that may in part even be to dieters. The present paper fills this gap. For the first related to the recent decline in life expectancy in the time, we demonstrate, that metabolic markers can be USA [43, 44] requires novel approaches suitable to help used in conjunction with food intake behaviour and have people lose weight. In this paper, we describe the first the potential to predict weight loss. Thus, a person on a attempt at developing quantitative, molecular feedback diet, in the future, can measure their insulin (or to a mechanisms for people dieting. While biofeedback is lesser extent, lactate) values and make a decision if it is well established to be successful in diabetes [45], it has acceptable to eat another meal that day, or what type of not been studied in people with no obvious signs of a meal it should be. Our current study has provided the disease. Our approach also differs from previous efforts proof of concept that biomarker measurements can be at identifying biomarkers of sustained weight loss which used in this context. Limitations of our study are the had for example identified ACE levels, amongst others short term nature of the diet (24 h periods, as opposed [19]. While extremely useful, this information is to more realistic weeks/months of dieting) and the Tejpal et al. BMC Obesity (2019) 6:20 Page 11 of 15 Fig. 5 Correlation plot of measured variables. a The correlation (or lack thereof) between the parameters is shown for 147 days. Correlations between the parameters were scaled from 1.0 to − 1.0. Blue indicates positive correlation while red indicates negative correlation. X indicates no correlation between the two parameters. b Correlation plot of measured variables for heathy individuals with BMI up to 25 (panel I), in the overweight category with BMI in the range 25–30 (panel II) and the obese category with BMI > 30 (panel III) length and cost of the assay of insulin, and the need for We are currently in the process of developing a rapid, urine samples. Thus, both assays for urine require a la- cheap and home-based sensor for insulin and lactate boratory setting, making it not yet feasible to conduct a [46], which will enable us to address these limitations in long-term study or investigate the effect on behaviour. the future. As the majority of participants only provided Tejpal et al. BMC Obesity (2019) 6:20 Page 12 of 15 Fig. 6 Insulin response is dependent on BMI. a Spread of total, fasting, last, following day insulin and total calories of all the participants in comparison to BMI. b Comparison of insulin parameters among healthy, overweight and obese participants. Significance levels are marked as follows: *p = 0.05, **p = 0.01, p = < 0.001. c Weight loss is associated with low insulin values. Weight change was grouped into four groups, no weight difference, weight gain or weigh loss between 0.1–0.5 kg and > 0.5 kg. Significance levels are marked as follows: *p < 0.05; **p < 0.01 data for 2–3 times 24 h periods, a long term trial is by the large disparity between male and female partici- needed to demonstrate if similar conclusions can be pants (77% female, 23% male). Consequently, there may reached over longer periods of dieting. be a sampling bias because participants were not chosen Our study was intended as a proof of concept to dem- at random and they might exhibit different lifestyles. onstrate if molecular measurements may provide useful Since most participants worked or studied at the Univer- information during dieting efforts. The most useful in- sity of Warwick, participants cannot be considered rep- formation for a dieter is weight loss. Thus, the main resentatives of the UK population (although some of the purpose of the study was to identify if there may be any demographics were similar), nor could the conclusions correlation between molecular data and weight loss. Be- necessarily be extrapolated to people from other coun- cause this was an observational study with a relatively tries. Another source of sampling bias introduced by the small number of participants (52), the treatments (which observational nature of the study is that participants meal to skip and on what day) were not assigned were given the liberty to choose what days to diet, as randomly. Thus, the protocol of measurement, as well as well as what meals to skip. This has resulted in different sampling may cause the study not to be representative meal plans to be followed for different number of days of the general population. Sources of sampling bias by the individuals. Therefore, there is another instance could be due to this being a volunteer sample, as well as of non-probability sampling, thus creating a possibility a convenience sample imposed by the requirement to for statistical bias. There are also sources of response transfer urine samples to the laboratory for measure- bias because participants were asked to record their data ments. The bias associated with this was made evident in an app, this means that participants may forget or Tejpal et al. BMC Obesity (2019) 6:20 Page 13 of 15 neglect to record data. Also, participants could have en- additional access to local based food information da- tered incorrect values for meal calories, thus indicating tabases need to be included (such as TESCO/Sains- voluntary response bias. Another form of response bias, bury’s basket for UK users). Increase in more unique to this study, was improper measurements user-friendliness of the app could also help to target bias by the participants. Users were asked to record their a wider audience. With the wide-spread use of weight, as well as collect samples of their urine. Incorrect smartphones and tablets, apps that run on these de- sample storage, and errors in measuring urine volume, im- vices have become a structural part of our lives [47]. properly weighing themselves, or using a poor scale could 74% of European and 73% of American adolescents all result in inaccurate entries in the digital health platform. use a smartphone on a regular basis [47]. With the Individuals entered information on the health platform for increase in abundance of such technologies came the 147 days but weight information was only provided for development of fitness and health apps that can pro- 126 days. The missing weight information for those 20 days vide behavioural interventions [47, 48]. However, could have affected the dieting success and biomarker Alleyetal. (2017),haveshown that there are only levels and meal plan choice. Incorrect use of weighing ma- 25 apps that directly target sedentary behaviour, chine, height measurement by the participants could have physical activity and/or diet. No app so far provides resulted in misclassification of individuals in BMI groups. personalised feedback using molecular measurement Finally, no conclusions on causation was intended or can information. This is the gap the approach described be inferred due to the high likelihood of confounding vari- in this paper is aiming to fill, which we hope could ables. One such variable is the fact that some individuals help target behaviour change techniques in individ- recorded data on consecutive days, while others on single uals, or in obesity clinics, weightwatcher programs days separated by days without data entries. There could and other organizations that aim to assist individuals have been an effect on some of the measurements after or patients making life-style changes. consecutive days of skipping meals. This, in turn, might have affected the conclusions of the study. Another pos- Conclusions sibly confounded variable was which meal was omitted. In this study, we have investigated a molecular feedback For example, participants may have skipped a meal where approach to assist dieting efforts and behavioural re- they regularly ate a lot as opposed to skipping a meal sponses of people using a web- and mobile-based applica- where they regularly eat less. Also, other daily activities tion to assist weight loss efforts. We found that skipping a might also have had an effect on an individual’sweight loss meal in a day regardless of which one, while also record- and thus been confounded with other variables in the ing all food and exercise events that day and collecting study. While the platform contained an entry form for urine samples for subsequent molecular profiling, resulted physical activity, few entries were made. In summary, in consistent weight loss for that day, in comparison to the present study contained a number of sources for control days in which any number of meals was allowed. potential bias that can be addressed in future efforts. Insulin and lactate values show correlations to BMI, Most importantly, the data collected provides us with caloric patterns and weight differences. In particular, low the necessary information to design a larger study in insulin and lactate values are likely predictive of weight which we can randomly assign participants to meal loss. This could be sufficiently motivating to dieters, a hy- plans over a longer period of time. Given that the last pothesis that needs to be tested in a future behavioural insulin and lactate measurements of the day are the study and over longer periods of dieting. most informative, a future study can restrict sample collection to these samples, allowing for data collec- Additional files tion over an entire diet period which normally takes place over several weeks. Once these molecular mea- Additional file 1: Supplementary Information. The file includes surements can be carried out by participants directly figures supporting the results of the paper. (DOCX 1132 kb) at home, recruiting participants not only from the Additional file 2: Raw Data. The file includes the raw data collected university campus would allow broadening of the during the study. (XLSX 46 kb) participant profiles. Extensions to the study can also include improve- Abbreviations ments to our digital health platform. Our current ACE: Angiotensin converting enzyme; B: Breakfast; BMI: Body-mass-index; D: Dinner; L: Lunch; NAD: Nicotinamide adenine dinucleotide; S: Snack app provides the setting that allows recording of life-style related data, including weight, food and Acknowledgements drink intake, exercise, and urine sample collection We would like to acknowledge help from Gail Calvert and Curtis Nicholson details. It also provides automation for the analysis in the initial setup of the chemiluminescence detection and for discussion of of the data. To broaden the use of the app sources of statistical bias with Roshan Klein-Seetharaman. Tejpal et al. BMC Obesity (2019) 6:20 Page 14 of 15 Ethical approval and consent to participate 13. Teixeira PJ, Silva MN, Mata J, Palmeira AL, Markland D. Motivation, self- The study was approved by Warwick Medical School Ethics committee determination, and long-term weight control. Int J Behav Nutr Phys Act. BSREC (protocol identification REGO-2014-1318) and all participants provided 2012;9:22–9. informed written consent about their participation in the study. 14. Gong Z, Gong Z. Modeling the relationship between body weight and energy intake: a molecular diffusion-based approach. Biol Direct. 2012;7:19. 15. Elobeid MA, Padilla MA, McVie T, Thomas O, Brock DW, Musser B, et al. Funding Missing data in randomized clinical trials for weight loss: scope of the The study was funded by University of Warwick. problem, state of the field, and performance of statistical methods. PLoS One. 2009;4(8):e6624. Availability of data and materials 16. Elliott P, Posma JM, Chan Q, Garcia-Perez I, Wijeyesekera A, Bictash M, et al. All data generated or analysed during this study are included in this Urinary metabolic signatures of human adiposity. Sci Transl Med. published article (and its Additional files 1 and 2). 2015;7(285):285ra62. 17. Molitor J, Brown IJ, Chan Q, Papathomas M, Liverani S, Molitor N, et al. Authors’ contributions Blood pressure differences associated with optimal macronutrient intake JKS carried out conceptualization of the study; ST, NS and JKS designed the trial for heart health (OMNIHEART)-like diet compared with a typical methodology; ST, NS, VM, JPI, KM and JKS performed the investigation; American diet. Hypertension. 2014;64(6):1198–U86. Writing of the original draft was done by ST, NS and JKS; All authors assisted 18. Du F, Virtue A, Wang H, Yang X-F. Metabolomic analyses for atherosclerosis, with revision of the manuscript. All authors read and approved the final diabetes, and obesity. Biomarker Research. 2013;1(1):17. manuscript. 19. Wang P, Holst C, Andersen MR, Astrup A, Bouwman FG, van Otterdijk S, et al. Blood profile of proteins and steroid hormones predicts weight change Consent for publication after weight loss with interactions of dietary protein level and glycemic Not applicable. index. PLoS One. 2011;6(2):e16773. 20. Lang R, Wahl A, Stark T, Hofmann T. Urinary N-methylpyridinium and trigonelline as candidate dietary biomarkers of coffee consumption. Mol Competing interests Nutr Food Res. 2011;55(11):1613–23. The authors declare that they have no competing interest. 21. Heinzmann SS, Brown IJ, Chan Q, Bictash M, Dumas ME, Kochhar S, et al. Metabolic profiling strategy for discovery of nutritional biomarkers: proline betaine as a marker of citrus consumption. Am J Clin Nutr. 2010;92(2):436–43. Publisher’sNote Springer Nature remains neutral with regard to jurisdictional claims in 22. Mozaffarian D, Hao T, Rimm EB, Willett WC, Hu FB. Changes in diet and published maps and institutional affiliations. lifestyle and long-term weight gain in women and men. N Engl J Med. 2011;364(25):2392–404. Received: 1 September 2018 Accepted: 7 March 2019 23. Bressanello D, Liberto E, Collino M, Reichenbach SE, Benetti E, Chiazza F, et al. Urinary metabolic fingerprinting of mice with diet-induced metabolic derangements by parallel dual secondary column-dual detection two- dimensional comprehensive gas chromatography. J Chromatogr A. References 2014;1361(Supplement C):265–76. 1. Moody RS, Alison. Adult obesity and overweight. 2014. healthsurvey.hscic. 24. Moore SC, Matthews CE, Sampson JN, Stolzenberg-Solomon RZ, Zheng W, gov.uk Cai Q, et al. Human metabolic correlates of body mass index. Metabolomics 2. Smith AG, Singleton JR. Obesity and hyperlipidemia are risk factors for early : Official J Metabol Soc. 2014;10(2):259–69. diabetic neuropathy. J Diabetes Complicat. 2013;27(5):436–42. 25. Jourdan C, Petersen A-K, Gieger C, Döring A, Illig T, Wang-Sattler R, et al. 3. Tejpal S, Bastie C, Seetharaman JK. Lemon juice: a potential source of Body fat free mass is associated with the serum metabolite profile in a angiotensin converting enzyme antagonism for weight loss and insulin population-based study. PLoS One. 2012;7(6):e40009. resistance. Proc Nutr Soc. 2018;77(OCE4):E213. 26. Gotthardt JD, Bello NT. Meal pattern alterations associated with intermittent 4. Liu B, Young H, Crowe FL, Benson VS, Spencer EA, Key TJ, et al. fasting for weight loss are normalized after high-fat diet re-feeding. Physiol Development and evaluation of the Oxford WebQ, a low-cost, web-based Behav. 2017;174:49–56. method for assessment of previous 24 h dietary intakes in large-scale 27. Gudzune KA, Doshi RS, Mehta AK, Chaudhry ZW, Jacobs DK, Vakil RM, et al. prospective studies. Public Health Nutr. 2011;14(11):1998–2005. Efficacy of commercial weight loss programs: an updated systematic review. 5. Liu J, Johns E, Atallah L, Pettitt C, Lo B, Frost G, et al., editors. An Intelligent Ann Intern Med. 2015;162(7):501–12. Food-Intake Monitoring System Using Wearable Sensors. 2012 Ninth 28. Affinita A, Catalani L, Cecchetto G, De Lorenzo G, Dilillo D, Donegani G, et International Conference on Wearable and Implantable Body Sensor al. Breakfast: a multidisciplinary approach. Ital J Pediatr. 2013;39(1):44. Networks; 2012 9-12 May 2012. 29. Neumann LB, Dunn A, Johnson D, Adams DJ, Baum IJ. Breakfast 6. Moshfegh AJ, Rhodes DG, Baer DJ, Murayi T, Clemens JC, Rumpler WV, et al. macronutrient composition influences thermic effect of feeding and fat The US Department of Agriculture Automated Multiple-Pass Method oxidation in Young women who habitually skip breakfast. Nutrients. reduces bias in the collection of energy intakes. Am J Clin Nutr. 2016;8(8):490. 2008;88(2):324–32. 30. Megson M, Wing R, Leahey TM. Effects of breakfast eating and eating 7. Vereecken CA, Covents M, Matthys C, Maes L. Young adolescents' nutrition frequency on body mass index and weight loss outcomes in adults enrolled assessment on computer (YANA-C). Eur J Clin Nutr. 2005;59(5):658–67. in an obesity treatment program. J Behav Med. 2017:1–7. 8. 24th European Congress on Obesity (ECO2017), Porto, Portugal, May 17–20, 31. Jarrett RJ, Keen H. Diurnal variation of Oral glucose tolerance: a 2017: Abstracts. Obesity Facts 2017;10(suppl 1)(Suppl. 1):1–274. possible pointer to the evolution of diabetes mellitus. Br Med J. 9. Inc A. Health & fitness - App store Downloads on iTunes. 2016 Available 1969;2(5653):341–4. from: https://itunes.apple.com/us/genre/ios-health-fitness/id6013?mt=8. 32. Jakubowicz D, Wainstein J, Landau Z, Raz I, Ahren B, Chapnik N, et al. Accessed 16 Mar 2016. Influences of breakfast on clock gene expression and postprandial Glycemia 10. Obesity – Android Apps on Google play. Available from: https://play.google. in healthy individuals and individuals with diabetes: a randomized clinical com/store/search?q=obesity&c=apps&docType=1&sp=CAFiCQoHb2 trial. Diabetes Care. 2017;40(11):1573–9. Jlc2l0eXoFGADAAQKKAQIIAQ%3D%3D%3AS%3AANO1ljJqCJM&hl=en_GB. Accessed 16 Mar 2016. 33. Nooyens ACJ, Visscher TLS, Schuit AJ, van Rossum CTM, Verschuren WMM, 11. Conway JM, Ingwersen LA, Vinyard BT, Moshfegh AJ. Effectiveness of the US van Mechelen W, et al. Effects of retirement on lifestyle in relation to Department of Agriculture 5-step multiple-pass method in assessing food changes in weight and waist circumference in Dutch men: a prospective intake in obese and nonobese women. Am J Clin Nutr. 2003;77(5):1171–8. study. Public Health Nutr. 2007;8(8):1266–74. 12. Scisco JL, Muth ER, Dong Y, Hoover AW. Slowing bite-rate reduces energy 34. Dhurandhar EJ, Dawson J, Alcorn A, Larsen LH, Thomas EA, Cardel M, et al. intake: an application of the bite counter device. J Am Diet Assoc. The effectiveness of breakfast recommendations on weight loss: a 111(8):1231–5. randomized controlled trial. Am J Clin Nutr. 2014;100(2):507–13. Tejpal et al. BMC Obesity (2019) 6:20 Page 15 of 15 35. Mohammadi D. The great breakfast myth. New Scientist. 2016; 229(3066):39–41. 36. Zhang L, Cordeiro SL, Liu J, Ma Y. The association between breakfast skipping and body weight, nutrient intake, and metabolic measures among participants with metabolic syndrome. Nutrients. 2017;9(4):384. 37. Gill S, Panda S. A smartphone app reveals erratic diurnal eating patterns in humans that can be modulated for health benefits. Cell Metab. 2015;22(5):789–98. 38. Turner-McGrievy GM, Wilcox S, Boutté A, Hutto BE, Singletary C, Muth ER, et al. The dietary intervention to enhance tracking with Mobile devices (DIET Mobile) study: a 6-month randomized weight loss trial. Obesity (Silver Spring, Md). 2017;25(8):1336–42. 39. Franco RZ, Fallaize R, Lovegrove JA, Hwang F. Popular Nutrition-Related Mobile Apps: A Feature Assessment. JMIR mHealth and uHealth. 2016;4(3):e85-e. 40. Turner-McGrievy G, Tate D. Tweets, apps, and pods: results of the 6-month Mobile pounds off digitally (Mobile POD) randomized weight-loss intervention among adults. J Med Internet Res. 2011;13(4):e120-e. 41. Gunnarsdóttir I, Sigurgeirsdóttir GK, Thórsdóttir I. Predictors of dropping out in a weight loss intervention trial. Ann Nutr Metab. 2010;56(3):212–6. 42. Newby FD, Wilson LK, Thacker SV, DiGirolamo M. Adipocyte lactate production remains elevated during refeeding after fasting. Am Jo Physiol-Endocrinol Metabol. 1990;259(6):E865–E71. 43. Walls HL, Backholer K, Proietto J, McNeil JJ. Obesity and trends in life expectancy. J Obes. 2012;2012:107989. 44. Crimmins EM PS, Cohen B. National Research Council (US) Panel on Understanding Divergent Trends in Longevity in High-Income Countries Washington (DC): National Academies Press (US); 2011. Available from: https://www.ncbi.nlm.nih.gov/books/NBK62367/. 45. McGinnis RA, McGrady A, Cox SA, Grower-Dowling KA. Biofeedback-assisted relaxation in type 2 diabetes. Diabetes Care. 2005;28(9):2145. 46. Sanghera N, Anderson A, Nuar N, Xie C, Mitchell D, Klein-Seetharaman J. Insulin biosensor development: a case study. Int J Parallel, Emergent and Distribut Sys. 2017;32(1):119–38. 47. Schoeppe S, Alley S, Rebar AL, Hayman M, Bray NA, Van Lippevelde W, et al. Apps to improve diet, physical activity and sedentary behaviour in children and adolescents: a review of quality, features and behaviour change techniques. Int J Behav Nutri Phys Activ. 2017;14:83. 48. Middelweerd A, Mollee JS, van der Wal CN, Brug J, te Velde SJ. Apps to promote physical activity among adults: a review and content analysis. Int J Behav Nutri Phys Activ. 2014;11:97. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png BMC Obesity Springer Journals

Towards personalised molecular feedback for weight loss

Loading next page...
 
/lp/springer-journals/towards-personalised-molecular-feedback-for-weight-loss-1yjPDcxNZ0

References (51)

Publisher
Springer Journals
Copyright
Copyright © 2019 by The Author(s).
Subject
Medicine & Public Health; Endocrinology; Public Health
eISSN
2052-9538
DOI
10.1186/s40608-019-0237-5
Publisher site
See Article on Publisher Site

Abstract

Background: Numerous diets, apps and websites help guide and monitor dietary behaviour with the goal of losing weight, yet dieting success is highly dependent on personal preferences and circumstances. To enable a more quantitative approach to dieting, we developed an integrated platform that allows tracking of life-style information alongside molecular biofeedback measurements (lactate and insulin). Methods: To facilitate weight loss, participants (≥18 years) omitted one main meal from the usual three-meal routine. Daily caloric intake was restricted to ~1200KCal with one optional snack ≤250KCal. A mobile health platform (personalhealth.warwick.ac.uk) was developed and used to maintain diaries of food intake, weight, urine collection and volume. A survey was conducted to understand participants’ willingness to collect samples, motivation for taking part in the study and reasons for dropout. Results: Meal skipping resulted in weight loss after a 24 h period in contrast to 3-meal control days regardless of the meal that was skipped, breakfast, lunch or dinner (p < 0.001). Common reasons for engagement were interest in losing weight and personal metabolic profile. Total insulin and lactate values varied significantly between healthy and obese individuals at p = 0.01 and 0.05 respectively. Conclusion: In a proof of concept study with a meal-skipping diet, we show that insulin and lactate values in urine correlate with weight loss, making these molecules potential candidates for quantitative feedback on food intake behaviour to people dieting. Keywords: Obesity, Insulin, Lactate, Weight loss, Biomarkers, App development Background of developing heart disease, stroke, type 2 diabetes and Obesity is a global epidemic with increasing incidence certain types of cancers among other conditions [2, 3]. rates in developed and developing nations. For example, Thus, health-based concerns should be excellent motiv- the past 10 years have seen an approximate doubling in ating factors to lose weight. However, achieving weight the number of obese adults in the UK alone with loss is hard work and failure is demoralising. Most typ- body-mass-index (BMI; defined as weight divided by the ical weight loss programs include following a diet regime square of height, in kg/m ) values > 30. Male obesity (with or without drugs such as appetite suppressants), a rates rose from 13.2 to 24.4% and from 16.4 to 25.1% in fitness regime or a combination of these approaches. women over the period 1993 to 2012 [1]. Besides body Technological support available includes diet trackers image considerations, being overweight or obese can and activity monitors. Recording of eating patterns has raise blood pressure and concentrations of cholesterol, been recognized as an effective step in managing obesity as well as cause insulin resistance. This increases the risk [4, 5]. While the traditional paper version of the com- monly used dietary questionnaire is considered tiring [6], there are numerous computer-assisted versions [7, 8]as * Correspondence: j.klein-seetharaman@warwick.ac.uk Systems Biology and Biomedicine, Division of Metabolic and Vascular well as apps and websites available for personal tracking of Health, Medical School, University of Warwick, Gibbet Hill, Coventry CV4 7AL, food intake [9, 10]. However, regardless of interface, UK mis-reporting of food intake is a well-documented Institute for Digital Healthcare, Warwick Manufacturing Group, University of Warwick, CV4 7A, Coventry, UK © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Tejpal et al. BMC Obesity (2019) 6:20 Page 2 of 15 problem [6, 11], as it depends on self-awareness, honesty Besides food composition, another factor amenable to and motivation of the user [5]. Often, unconscious bias of behaviour change is meal timing. Different studies have self-observation leads to under-realization of foods eaten suggested that consuming a regular breakfast is associ- [12]. Energy expenditure on the other hand can potentially ated with lower body weight [27–30]. There is also a be tracked without bias using activity monitors, but they plethora of metabolic studies that support the notion do not provide a direct link to weight loss. Even if a device that eating breakfast is preferable over eating dinner due that accurately measures caloric intake and expenditure to a phenomenon referred to as “evening diabetes” in was widely available, the information may not be sufficient which insulin sensitivity is higher early during the day to motivate a user to make changes in their behaviour that making it more likely to burn fat [31]. A crossover study would result in weight loss. This is evidenced by the fact comparing days with breakfast and lunch versus days that even when meticulously keeping records of food with lunch only showed different effects on clock gene intake, individuals still find it hard to lose weight [13]. expression in healthy and type 2 diabetic subjects, while This is because the relationship between caloric intake skipping of breakfast showed an altered response for and weight loss is not linear [14]. As a result, current clock gene profiles in both groups [32]. On the other approaches to lose weight loss generally do not work hand several studies indicate that breakfast eaters and well [15], and the weight loss market is missing a skippers didn’t vary significantly in terms of body weight device that is more directly coupled to the desired and nutrient intake [33–36]. A study by Gill and Panda outcome, weight loss. The prerequisite for such a [37] has demonstrated the extent to which adults display device is a quantitative and science-based biomarker an irregular daily and weekly rhythm of eating-fasting, of weight loss that has the potential to provide a which could be manoeuvred to obtain desirable health biological feedback loop to the dieter. benefits. It might be possible to cater to such personal Quantitative biomarkers of weight loss do not exist preferences, at least to some extent, without compromis- yet, but biomarkers have been considered for a number ing weight loss success. For example, from a metabolic of related areas. For example, urinary metabolic markers perspective it may be most efficient for weight loss to for cardiovascular disease, blood pressure and adiposity skip meals other than breakfast, especially dinner. How- have been identified [16, 17]. Several metabolomics stud- ever, from a behavioural and/or cultural perspective, ies involving untargeted proton ( H) nuclear magnetic breakfast and lunch may be the easiest meals to skip, resonance spectroscopy and ion exchange chromatog- and dinner the hardest. Weighing between these alter- raphy on obese human and mice urine samples have nate decision making strategies requires quantitative identified metabolites associated with BMI and adiposity analysis. Finally, to develop a tool for personalised feed- [16, 18]. Angiotensin converting enzyme (ACE) has been back there is also a need to understand the willingness shown through blood profiling for protein and steroid of individuals to collect samples in order to obtain meta- hormones to be an important predictor for sustained bolic biomarker information. weight loss [19]. Biomarkers for different foods include Here, our goal was to lay the foundation for a molecu- NMNA, a niacin-related (vitamin B3) metabolite marker lar feedback approach to assist dieting efforts. To this for coffee drinking [20], proline betaine for citrus fruit end, we aimed at obtaining a better understanding of the consumption [21] and O-acetyl carnitine for red meat complex interplay between possible molecular bio- intake [22]. Thus, while biomarkers have been consid- markers of dieting behaviour, individuals' personal ered for several related purposes, none have been tested preferences, their willingness to collect information, the yet as a means to provide molecular feedback to drive to act on the collected information and weight loss someone following a diet. success. Specifically, we asked 52 dieters to record meal The requirements for dieting feedback are more timing and composition using an electronic diary inter- immediate than the general adiposity markers, or the face, along with collection of urine samples, and weight. long-term ACE weight loss marker. Our goal is to meas- The urine samples were analysed for lactate and insulin ure a molecular marker that directly or indirectly reports concentrations and the information was integrated. Our on dieting behaviour so that we can use it for feedback. results provide a strong proof of concept that each We have chosen insulin and lactate because these are molecule studied, but especially insulin, can be used for both known to vary with consumption of glucose and biofeedback on dieting behaviour. other food, as well as caloric intake overall [16, 23–25]. Caloric restriction is one of the most widely used strat- Methods egies for losing weight [26, 27] but it requires strong will Study design and population power, and thus identifying biomarkers of weight loss Participants were recruited by website, flyer and newsletter and using them for dieting feedback may make caloric advertisement at the University of Warwick and through restriction easier on dieters. word of mouth. To be eligible for participation, individuals Tejpal et al. BMC Obesity (2019) 6:20 Page 3 of 15 had to be 18 years or older and clinically healthy. Sample collection and storage Exclusion criteria included diagnosed diabetes and Most samples were dropped off at the departmental pregnancy. The study was approved by the Warwick front desk and the investigators collected the samples Medical School Ethics committee BSREC (protocol and transferred them to -80 °C upon notification by the identification REGO-2014-1318). Everyone who receptionist. In some cases, samples were handed over expressed interest in the study was referred to details to investigators directly by participants. on the website personalhealth.warwick.ac.uk.Those who remained interested, met with the PI of the Mobile application study, and were provided with instructions on how to The mobile and web application was developed using use the interface. We also provided them with the Intel XDK, a framework that allows building of urine sample collection kit and instructions on how cross-platform mobile applications, including Android, to use it and how to transfer samples to the investi- iOS and Windows along with a web browser interface gators. All participants provided informed written (http://personalhealth.warwick.ac.uk). The program- consent at this time. Not all participants who were ming was using standard web technologies such as given the information and sample collection kit actu- HTML, Javascript and CSS. A responsive web design ally entered data and returned samples. Sample tubes approach was adopted in implementing the applica- were labelled with numbers that anonymized the sam- tion to enable device specific optimal user experi- ples. Data entered into the digital health platform was ences. Front end browser and mobile app matched to molecular information only through the functionality was built using AngularJS, a Javascript sample tube numbers. framework. All back end support for the mobile app Participants were asked to collect data for at least one and the web interface were built using Java and 24 h period. In total, participants provided data for 149 MariaDB database server. Back end automation for days. The average number of 24 h periods was 3 days/ providing users with graphical feedback was programmed participant, and the maximum number of data was using R. All data communication between the server and entered by one participant who provided 24 periods of mobile and web applications was through HTTPs. A 24 h data each. built-in interface between our server and the FATsecret A control day was defined as a day in which partici- database (https:\\platform.fatsecret.com) was developed so pants provided samples and data where they consumed that study participants could seamlessly search for nutri- a regular 3-meal diet. An intervention day was defined tion information without leaving our website or app. as a day in which they provided samples and data where Where users did not retrieve data from the FATsecret they skipped at least one meal. Since on average only 3 database, investigators carried out the searches subse- days of samples and data was provided, a washout period quently. FATsecret provides free access to their database was not necessary. through an API (up to 5000 calls a day were allowed). The database is very large and allows retrieval of nutritional information for the majority of food and drink items (see Demographics data collection below). All participants were given the option to enter informa- tion about their age, BMI, sex, weight and ethnicity at Meal plans the time of sign-up to the digital health platform. The Participants were asked to provide data and samples for at majority of participants provided this information and least one control day and at least one diet day (see this was used to carry out demographic analysis. In some definition above). For diet days, participants were cases, BMI was calculated by the investigators from recommended to skip lunch (L) or dinner (D), but were weight and height entries directly. Participants were not also allowed to skip breakfast (B). They were also asked to specifically recruited to target any particular BMI. How- restrict the total caloric intake for the day to less than ap- ever, for analysis purposes, the data was also divided by proximately 1200KCal. An optional snack (S) of <250KCal grouping individuals according to their BMI values. Indi- was allowed once in a day. B, L, D, S were defined viduals with a BMI of 18.5–24.9 kg/m were considered according to i) calorie intake (<250KCal for S, normal; 25.0–29.9 kg/m were overweight; and those >250KCal for B, L, D) summed over nearby entries above 30.0 kg/m and greater were obese [1]. The goal of (within 30 min) and ii) timing, being B before noon, the study was not to compare obese with non-obese L between noon and 3 pm and D after 4 pm (Fig. 1a). participants, but a separation of the data into BMI A control day had three or more meals on that day. groups was useful to identify BMI as a source of According to these specifications we defined different variation and to delineate possible trends in the data that meal plans to systematically describe the participants’ may relate to BMI. preferences:i)B,L,[S]; ii) B,D, [S]; iii)L,D, [S];iv) Tejpal et al. BMC Obesity (2019) 6:20 Page 4 of 15 The caloric intake was calculated based on the meal information provided by the user. The individuals either used the pre-specified meal options from the digital health database or entered detailed descriptions of their prepared meals. The food (in grams) breakdown in carbohydrate, fat and protein was obtained from the fat- secret database (available at https:\\platform.fatsecret. com). For analysis purposes, it was further converted into KCal by multiplying with 4, 4 and 9 per gram of carbohydrates, protein and fat, respectively. FatSecret was chosen as it had a mature API that could be used by our website and app to calculate nutrition information without maintaining our own nutrition database. In addition to offering a large and curated database of com- mon foods and packaged products, FatSecret also offers a professional interface for health practitioners to moni- tor app usage of their patients [38]. It has also been pre- viously used in several studies as a caloric counter to understand effects of tracking information (on mobile applications) on weight loss [38–40]. Entries in the database were used to determine every participants’ caloric patterns, meal plan preferences and to compare the different plans to the weight changes ob- served over the respective 24 h period along with insulin and lactate profiles. In all cases, only those records were analysed that contained a minimum of two meal records for the day (from 00:00 h to 12:00 h day + 1). No instruc- tions were given to participants with respect to food composition, and any food item was allowed. Some of the food options chosen by participants are depicted in Fig. 1a. The fasting times for diet days were obtained by sub- tracting the time of the second meal from the first meal of the day. Similarly, the overnight fasting period was obtained except that only those days that had following day weight information were considered. Insulin measurements using mesoscale luminescence assay Insulin in samples was measured using a Mesoscale Discovery Human Insulin Kit containing (catalogue number: K151 BZC-2) 96 well plates coated with insulin antibodies obtained from Mesoscale Inc. (www.mesoscale. Fig. 1 Study Design. a Meal plan layout for the participants in a com). The assay was performed according to the manufac- 24 h period. b Flow diagram of the study design. c Comparison between demographic features of the study participants and the UK turer’s instructions. The plates were analysed on a SEC- population. Overall numbers for the UK population was the TOR Imager 6000 system. All samples (urine and plasma) arithmetic middle between the male and female values based on were centrifuged prior to analysis at 12000 rpm for 5 min the assumption that the distribution of male and females in the at room temperature. Insulin calibrators (supplied by the statistics was approximately 50%. In this study, there was more kit) were run in duplicate to generate an 8-point standard females than males, so the overall number was obtained directly from the raw data curve covering the 0–50,000 pg/mL range. The standard curve was modelled using least squares fitting algorithms so that signals from standards with known concentrations B or L or D, [S] for dieting days; and v), B,L,D, [S] of insulin can be used to calculate insulin concentrations or > 3 meals for control days, where [S] denotes op- in samples. The MSD Discovery Workbench® analysis tional snack intake (Fig. 1b). software was used to calculate the concentration of insulin Tejpal et al. BMC Obesity (2019) 6:20 Page 5 of 15 in samples. The software uses a 4-parameter logistic The survey was divided into three parts: model and includes a 1/Y2 weighting function. This allows Section I: Motivation. Several options were provided to for a better fit of data over a wide dynamic range (3–4 the participants such as interest in losing weight, diet, logs), particularly at lower insulin concentrations. The metabolic profile, health platform and being involved in wide dynamic range of the assay allowed for the quantifi- medical research. These parameters were analysed cation of insulin in urine without the need for dilution nor together and when separated into 10-year bin-sized age concentration. groups. Section II: Dropout. The survey was used to identify Lactate measurement the reasons for dropout such as difficulty of diet and Lactate dehydrogenase is the enzyme responsible for sample collection, time consuming, complicated health interconversion of lactate to pyruvate following reduc- platform. The participants were also provided a free text tion of nicotinamide adenine dinucleotide (NAD) to its field to enter other factors that contributed to dropout. reduced form (NADH). To measure lactate, the reaction Section III: Feedback. The participants provided feed- needs excess NAD. To force the reaction to completion back on the health platform and sample collection. Their in this direction, it is necessary to trap the formed pyru- personal input was requested on suggestions for the vate with hydrazine present in glycine-hydrazine buffer. platform’s improvement. All samples were centrifuged prior to analysis at 12000 The information provided by the individuals was used rpm for 15 min at room temperature. The experiments to calculate the weighted average of each contributing were conducted in 96 well solid black fluorescence factor for motivation to participate, reasons for dropout plates (Thermo-Scientific, catalogue #: 634-0006). The and feedback about the study. stock solution of 10 mM of lactate was reconstituted in glycine-hydrazine buffer (0.6 M glycine and 0.5 M Statistical analysis hydrazine, pH 9.2) bought from Sigma-Aldrich, UK. This Statistical analyses were performed using IBM SPSS was used to prepare standard reactions in the range of Statistics 24 and R. Association between different vari- 25-100 μM lactate concentrations. A reaction mixture ables was calculated using bivariate Pearson Coefficient stock solution containing 10 mg NAD with 2.0 mL glycine analyses. Nonparametric Mann-Whitney U test was per- buffer, 4.0 mL water 0.1 mL L-lactate dehydrogenase formed in some cases as indicated. One-way analysis of (Sigma-Aldrich, UK). 20 μl of standard (in duplicates) or variance (ANOVA) was used to compare values from sample and 130 μl of reaction mixture were added to each control days with values from dieting days. well. The plate was then incubated at 37 °C for 15 min. The fluorescence was read using a Perkin Elmer Wallac Results 1420 Victor2 Microplate Reader with excitation at Study design and data collection 345-355 nm and emission at 450-460 nm. Initially, 146 individuals recruited by flyer and newsletter advertisement expressed interest in our study. Of these, Survey design 52 individuals became study participants (77% females A short self-administered questionnaire link was sent via and 23% males) who provided data and samples. They email to everyone who originally expressed an interest in received access to a web application at personalhealth. the study (including but not limited to those who actu- warwick.ac.uk, as well as an app “Digital Health Plat- ally enrolled), inviting them to participate in the survey. form” for android and apple devices available in google The survey was anonymous and the data from the play and iTunes stores, respectively. Through this plat- survey was not linked or compared with the data en- form, they entered life-style related data, including tered by participants during the weight loss study. The weight, food and drink intake, exercise, and urine sample questionnaire (Additional file 1:FigureS4) wasde- collection details. The mobile health platform creates a signed using Google forms. A total of 48 people sub- timeline of the logs or events that are entered by the mitted the study questionnaire, all of whom had user. This electronic information is sent to a web server consented to participate in the survey and stated pre- that allows users to store their information securely and vious experience on entering data onto the platforms. access it anywhere using either a web browser based The individuals participating in the survey were not interface or a native mobile application from their smart linked to their identification in the platform, as the phones or tablets. In addition to being a tool for logging survey google document and digital health platform time and other parameters, the application also serves to were independent of each other. The participants seamlessly share information between the user and the were provided with several options under each section analyst. It allowed researchers and cohort group partici- (see below) along with an option to enter other fac- pants to register, and manage the logistics of data collec- tors contributing to the study. tion. Researchers obtained analysis files in anonymized Tejpal et al. BMC Obesity (2019) 6:20 Page 6 of 15 fashion only through the website administrator. Ease of indicates meals, green snacks and blue/brown low calorie use and cross-platform support were the most important drinks (including water and coffee). Most entries for among the factors considered in the design of the health caloric intake of >250KCal (i.e. a meal) were observed in platform. just one hour from 6 to 7 pm (Fig. 2e). Many breakfast (B) Urine samples were used to measure insulin and “meals” were low in calories and were therefore classified lactate concentrations which were uploaded onto the here as snacks (S), see below. Caloric intake is significantly platform. Participants collected samples and life-style different for males and females (p = 0.05, Additional file 1: data for control and diet days (see Methods). Figure S2). Demographics of study participants Diet behaviour: meal plan preferences According to the UK Health and Social Care Informa- Out of the 52 participants, at least one entry with tion Centre, the prevalence of overweight individuals in two meals (see Methods) was available for 43 people. the UK population is age- and gender-dependent, with Since participants could freely chose the number of 9% (male) and 13% (female) in the 16–24 age group and days they participated in the study, the number of 13% (male) and 35% (female) in the 50–69 age group days for which data was available varied for each [1]. A similar pattern characterized the participants in participant. The majority followed the study plan for our study (Additional file 1: Figure S1). Grouping partic- 1–2 days, while one participant collected data for up ipants by age showed that the number of overweight to 24 days. Thus, the 43 individuals collectively pro- study participants was lowest among younger adults vided data for 147 days consisting of both, control (28) (20–29 year old group, Additional file 1: Figure S1a), and diet (119) days. Participants were given a rela- increasing through middle age (ages 30–59, Additional tively free choice in meal plans, with the only restric- file 1: Figure S1b-d), and only reducing among the oldest tions being the omission of one main meal and the participants (ages 60–69, Additional file 1: Figure S1e). total caloric intake as described in Methods The meal The majority of study participants were in the normal plan choices made by participants on the 147 days is and overweight groups (Additional file 1: Figure S1f). shown graphically in Fig. S3a. BL, the meal plan that The mean BMI of 27.2 kg/m observed in the UK popu- would be metabolically optimal from a theoretical lation [1] parallels that of 27.0 ± 5 kg/m (mean ± stand- perspective (see Introduction), or the slightly modified ard deviation) in our study. Similarly, the weight and BLS meal plan, were followed only on 10 days. 19 days height values split by gender also mirror those of the UK corresponded to the LD plan, while the largest num- population (Fig. 1c). This indicates that the sample of 52 ber of 46 days was in the BD or BDS category. BDS participants is a good representation of the UK popula- was followed on 39% of the dieting days, and was tion. The mean BMI of males and females in our study thus the most popular meal choice, while the B [S] are 26.0 and 28.0 kg/m , respectively, which indicated plan accounted for only 8% of the dieting days. A that they were significantly different (p < 0.001) within graph of the spread of meal timing of individuals our study group (Additional file 1: Figure S2). shows that participants followed similar eating pat- terns for all days if they provided samples and data Diet behaviour: caloric intake pattern for more than one day (Additional file 1:Figure The timings of health platform entries on the 149 days S3b). Another frequently followed meal plan was of data entered by participants show a wide spread from that of the single meal: 46 days had only one meal B, 7 am - midnight on a 24 h scale (Fig. 2a), with only night L or D (sometimes plus optional snack, B[S] or L[S] time (midnight to 7 am) receiving very few entries, in or D[S]). This large number likely arose from the line with previous observations [37]. There was a higher fact that we classified what participants may have percentage of total entries on the health platform in the thought of as “meals” as snacks based on the mornings and evenings, namely 33 and 32% of the total 250KCal cut-off. In total, there were 26 control days entries, respectively (Fig. 2b). Many of the morning en- (18% of the 147 days), where people have had at tries were weight and urine sample collections. When least three meals (BLD, BLD [S] or more). On these only food entries are plotted, entries cluster in the morn- control days, caloric intake was significantly higher ing (around 7 am), at lunchtime (around 1 pm), and in (p < 0.01) than on diet days, as expected, although the evening, peaking at 6 pm (Fig. 2c). When entries were there were many days of low calorie intake among quantified by calories consumed, one can see that the lar- control days. Notably, individuals did not lose weight gest calorie intake was in the evening, with 22 and 51% of on those caloric restricted control days, suggesting the total calories recorded from 7 to 11 am and 4 to 9 pm, that meal timing plays an important role, perhaps respectively (Fig. 2d). A more detailed break-down of en- more than caloric intake for losing weight tries as % food events per hour is shown in Fig. 2e. Purple (Additional file 1:FigureS3c). Tejpal et al. BMC Obesity (2019) 6:20 Page 7 of 15 Fig. 2 More calories are consumed at dinner and breakfast and dinner combination were more popular than breakfast and lunch. a Polar plot of all entries of each individual plotted against the time of day (angular axis). Data from 52 individuals are shown. 24 h rose plots showing (b) percentage of total entries from individuals, (c) percentage of ingestion events and (d) % of calories consumed. e Percentage of food events in 1 h bins. The radial axis for each rose plot shows % of events Dieting success by study participants’ meal plan choice kg, and no change =0 kg. Figure 3a shows the % of partici- Weight change data was available for only 43 out of the pants with weight change in each of these groups. One 52 participants for at least one 24 h period, reducing the can clearly see that all diet meal plans resulted more often total of 147 days to 126 days. For ease of analysis, we in weight loss as compared to the control days. Figure 3b grouped the weight change values into 3 groups: weight shows the more detailed split into sub-groups taking loss when the weight difference between the beginning whether or not a snack was eaten into account. Overall and end of the 24 h period was > 0 kg, weight gain for < 0 there does not appear to be a negative consequence of Tejpal et al. BMC Obesity (2019) 6:20 Page 8 of 15 Fig. 3 Weight loss is associated with fasting time and consumption of calories. a Effect of dieting on weight with respect to different meal groups. N/A refers to the days for which weight loss data is not available. Weight change is defined as weight loss (any change > 0 kg), weight gain (any change < 0 kg), and no change (=0 kg). One-way ANOVA analysis comparing control with other meal groups show significant difference at p = 0.01** (p < .001). b Effect of dieting on weight with respect to different meal plan subgroups. c Plot of total caloric intake against weight difference. Pearson’sR = − 0.21 correlation is significant at the 0.05 level (p < 0.05). d Plot of overnight fasting time against weight difference. Pearson’s R = -0.21 correlation is significant at the 0.05 level (p < 0.05) having the additional snack, although the size of the data Correlation R = -0.21, p = 0.016) (Fig. 3d),i.e.the is too small to ascertain the statistical significance of this longer the fasting the greater the weight loss. statement. Because we do have meal plan information for 21 days without weight change information, we included a fourth group “NA” (purple) in Fig. 3a, b for completeness. Motivation: reasons to participate in the study We conducted one-way analysis of variance (ANOVA) A survey was conducted to understand the reasons why comparing the control group with each of the other people were interested in a study that involved both, groups in Fig. 3a (i.e. BL, BD, LD and B or L or D). Each weight loss and urine sample collection (Additional file 1: group was significantly different from the control group Figure S4). Interest in losing weight, involvement in (p < 0.01). This indicates that skipping a meal results in research and knowledge of metabolic profile were the weight loss irrespective of which meal of the day is main drivers behind participation. There may be a skipped. Comparing weight loss with total caloric intake difference in motivation for different age groups, as showed an inverse relation with Pearson Correlation the 20–29 and 40–49 year age-group more often significant at p =0.05 (Fig. 3c). Finally, on days when reported interest in their metabolic profile (33%), participants achieved weight loss, the length of over- while the 30–39, 50–59 and 60–69 year age groups night fasting periods was inversely correlated to were more motivated by losing weight (32, 32 and weight loss expressed as negative kg values (Pearson 20% respectively). However, because of the small Tejpal et al. BMC Obesity (2019) 6:20 Page 9 of 15 number of participants, we cannot ascertain if these lactate and insulin concentrations. In particular, the differences are statistically significant. weight difference (expressed as negative kg) showed a correlation with total calorie intake, which was signifi- Dropout analysis cant at R = 0.04 (p < 0.05). Total insulin and total lac- Dropout rates in weight loss studies have been a prom- tate were positively correlated to the total calorie inent concern when promoting lifestyle and dietary intake (p < 0.001, R = 0.35 and R = 0.03, respectively). changes in overweight and obese populations, as well as Fasting, total, last, following day and maximum affecting the validity and generalisation of conclusions in amounts for insulin and lactate had significant correl- weight loss studies [41]. In our study, we observed a ation with carbohydrate, fat and protein content in similar trend. At the first meeting, people were informed the meals (Fig. 4d, e). of the study requirements. At this stage, of the 146 adults who had shown an interest, 70 dropped out, leav- Biomarker and BMI ing only 76 individuals who provided written consent for Because BMI was correlated with a number of param- participation in the study and received sample collection eters (Fig. 5), we investigated if pre-defined BMI kits. Of these, 52 participants actually provided urine groups differed in correlation of parameters (Fig. 5b). samples and life-style information through our online/ Segregation of the data into different BMI groups mobile platforms. Thus, the dropout rate after the first showed loss of correlation between weight loss and meeting of 48% reduced to 16% when comparing to the other parameters in the obese and overweight groups initial number of people interested in the study, and 34% while being sustained in the healthy group. Particu- when comparing to the previous step (Fig. 1b). larly, the insulin biomarker profiles in the overweight To identify the reasons for dropout in our study, and obese group are dampened in comparison to the we designed a number of questions (Additional file 1: healthy group (Fig. 6a).Total,fasting,last,following Figure S4). Busy schedule, complicated samples day and maximum insulin values were significantly collection and loss of motivation correspond to 25, 21 higher in the obese group in comparison to healthy and 18% of the reasons chosen by people who partici- individuals (Fig. 6b). Also, total and last lactate pated in the survey, respectively. Apart from amounts increased in obese people in comparison to pre-defined reasons, individuals also entered their per- the healthy group (Additional file 1: Figure S7a). Fur- sonal hurdles through a free text option. Participants thermore, total, maximum and minimum lactate found it hard to follow caloric restriction guidelines values were higher in obese than in overweight indi- duetotheir active worklife or thepsychological viduals (Additional file 1: Figure S7b), in accordance stress given by the word “diet”. The fear of eating with previous findings of increased lactate levels in more after a day of dieting also made people drop obese individuals [42]. out from the study. In addition, since the individuals in our study were UK based, they found it difficult to maintain the food diaries as the fatsecret database we used was an American foods database. Biomarker and weight loss Because weight loss is the desired outcome for most Molecular insulin and lactate biomarker correlate with dieters, we plotted weight change versus biomarkers life-style data in Fig. 6c. Because of inaccuracies inherent to meas- The urine samples collected by the participants were uring weight, we grouped the weight change values used to measure insulin and lactate concentrations. into 4 groups: weight loss > 0 .5kg, weight loss 0.1– These values were then used to extract a total of 23 0 .5kg, weight gain and no change. One can see very parameters relating to biomarker profiles or lifestyle data clearly that total insulin values vary most dramatically entered (Additional file 1: Figure S5). The in the weight gain group, and are overall higher in cross-correlation matrix of all the 23 extracted parame- the no weight and weight gain categories. Similar pat- ters from biomarker profiles and the digital health terns were also observed for fasting, last, following platform are shown for the complete cohort in Fig. 5a day and maximum insulin values. This graph thus for an overall summary. The weight difference showed a emphasizes that insulin values, even individual ones, positive correlation with BMI while a negative correl- as opposed to all values collected over a 24 h period, ation with carbohydrates, fat, lactate before second meal are potentially useful biomarkers for immediate feed- of the day and total calories was observed. Furthermore, back on dieting behaviour, with low values being as expected, the total lactate and insulin parameters likely predictive of weight loss, information which can were strongly correlated with other parameters such as only be obtained the day following a diet, too slow to first, last, maximum, minimum and following day be sufficiently motivating. Tejpal et al. BMC Obesity (2019) 6:20 Page 10 of 15 Fig. 4 Individual correlation plots of selected parameters. a Weight difference versus total calories. b Total insulin versus total calories. c Total lactate versus total calories. d Insulin parameters correlation with nutritional parameters: Panel I. Carbohydrate. Panel II. Fat. Panel III. Protein. e Lactate parameters correlation with nutritional parameters, as in (d). Significant correlations with macronutrient content were marked by ** or *, when significant at p = 0.01** and p = 0.05*, respectively Discussion long-term, and cannot be used for immediate feedback The escalating obesity epidemic that may in part even be to dieters. The present paper fills this gap. For the first related to the recent decline in life expectancy in the time, we demonstrate, that metabolic markers can be USA [43, 44] requires novel approaches suitable to help used in conjunction with food intake behaviour and have people lose weight. In this paper, we describe the first the potential to predict weight loss. Thus, a person on a attempt at developing quantitative, molecular feedback diet, in the future, can measure their insulin (or to a mechanisms for people dieting. While biofeedback is lesser extent, lactate) values and make a decision if it is well established to be successful in diabetes [45], it has acceptable to eat another meal that day, or what type of not been studied in people with no obvious signs of a meal it should be. Our current study has provided the disease. Our approach also differs from previous efforts proof of concept that biomarker measurements can be at identifying biomarkers of sustained weight loss which used in this context. Limitations of our study are the had for example identified ACE levels, amongst others short term nature of the diet (24 h periods, as opposed [19]. While extremely useful, this information is to more realistic weeks/months of dieting) and the Tejpal et al. BMC Obesity (2019) 6:20 Page 11 of 15 Fig. 5 Correlation plot of measured variables. a The correlation (or lack thereof) between the parameters is shown for 147 days. Correlations between the parameters were scaled from 1.0 to − 1.0. Blue indicates positive correlation while red indicates negative correlation. X indicates no correlation between the two parameters. b Correlation plot of measured variables for heathy individuals with BMI up to 25 (panel I), in the overweight category with BMI in the range 25–30 (panel II) and the obese category with BMI > 30 (panel III) length and cost of the assay of insulin, and the need for We are currently in the process of developing a rapid, urine samples. Thus, both assays for urine require a la- cheap and home-based sensor for insulin and lactate boratory setting, making it not yet feasible to conduct a [46], which will enable us to address these limitations in long-term study or investigate the effect on behaviour. the future. As the majority of participants only provided Tejpal et al. BMC Obesity (2019) 6:20 Page 12 of 15 Fig. 6 Insulin response is dependent on BMI. a Spread of total, fasting, last, following day insulin and total calories of all the participants in comparison to BMI. b Comparison of insulin parameters among healthy, overweight and obese participants. Significance levels are marked as follows: *p = 0.05, **p = 0.01, p = < 0.001. c Weight loss is associated with low insulin values. Weight change was grouped into four groups, no weight difference, weight gain or weigh loss between 0.1–0.5 kg and > 0.5 kg. Significance levels are marked as follows: *p < 0.05; **p < 0.01 data for 2–3 times 24 h periods, a long term trial is by the large disparity between male and female partici- needed to demonstrate if similar conclusions can be pants (77% female, 23% male). Consequently, there may reached over longer periods of dieting. be a sampling bias because participants were not chosen Our study was intended as a proof of concept to dem- at random and they might exhibit different lifestyles. onstrate if molecular measurements may provide useful Since most participants worked or studied at the Univer- information during dieting efforts. The most useful in- sity of Warwick, participants cannot be considered rep- formation for a dieter is weight loss. Thus, the main resentatives of the UK population (although some of the purpose of the study was to identify if there may be any demographics were similar), nor could the conclusions correlation between molecular data and weight loss. Be- necessarily be extrapolated to people from other coun- cause this was an observational study with a relatively tries. Another source of sampling bias introduced by the small number of participants (52), the treatments (which observational nature of the study is that participants meal to skip and on what day) were not assigned were given the liberty to choose what days to diet, as randomly. Thus, the protocol of measurement, as well as well as what meals to skip. This has resulted in different sampling may cause the study not to be representative meal plans to be followed for different number of days of the general population. Sources of sampling bias by the individuals. Therefore, there is another instance could be due to this being a volunteer sample, as well as of non-probability sampling, thus creating a possibility a convenience sample imposed by the requirement to for statistical bias. There are also sources of response transfer urine samples to the laboratory for measure- bias because participants were asked to record their data ments. The bias associated with this was made evident in an app, this means that participants may forget or Tejpal et al. BMC Obesity (2019) 6:20 Page 13 of 15 neglect to record data. Also, participants could have en- additional access to local based food information da- tered incorrect values for meal calories, thus indicating tabases need to be included (such as TESCO/Sains- voluntary response bias. Another form of response bias, bury’s basket for UK users). Increase in more unique to this study, was improper measurements user-friendliness of the app could also help to target bias by the participants. Users were asked to record their a wider audience. With the wide-spread use of weight, as well as collect samples of their urine. Incorrect smartphones and tablets, apps that run on these de- sample storage, and errors in measuring urine volume, im- vices have become a structural part of our lives [47]. properly weighing themselves, or using a poor scale could 74% of European and 73% of American adolescents all result in inaccurate entries in the digital health platform. use a smartphone on a regular basis [47]. With the Individuals entered information on the health platform for increase in abundance of such technologies came the 147 days but weight information was only provided for development of fitness and health apps that can pro- 126 days. The missing weight information for those 20 days vide behavioural interventions [47, 48]. However, could have affected the dieting success and biomarker Alleyetal. (2017),haveshown that there are only levels and meal plan choice. Incorrect use of weighing ma- 25 apps that directly target sedentary behaviour, chine, height measurement by the participants could have physical activity and/or diet. No app so far provides resulted in misclassification of individuals in BMI groups. personalised feedback using molecular measurement Finally, no conclusions on causation was intended or can information. This is the gap the approach described be inferred due to the high likelihood of confounding vari- in this paper is aiming to fill, which we hope could ables. One such variable is the fact that some individuals help target behaviour change techniques in individ- recorded data on consecutive days, while others on single uals, or in obesity clinics, weightwatcher programs days separated by days without data entries. There could and other organizations that aim to assist individuals have been an effect on some of the measurements after or patients making life-style changes. consecutive days of skipping meals. This, in turn, might have affected the conclusions of the study. Another pos- Conclusions sibly confounded variable was which meal was omitted. In this study, we have investigated a molecular feedback For example, participants may have skipped a meal where approach to assist dieting efforts and behavioural re- they regularly ate a lot as opposed to skipping a meal sponses of people using a web- and mobile-based applica- where they regularly eat less. Also, other daily activities tion to assist weight loss efforts. We found that skipping a might also have had an effect on an individual’sweight loss meal in a day regardless of which one, while also record- and thus been confounded with other variables in the ing all food and exercise events that day and collecting study. While the platform contained an entry form for urine samples for subsequent molecular profiling, resulted physical activity, few entries were made. In summary, in consistent weight loss for that day, in comparison to the present study contained a number of sources for control days in which any number of meals was allowed. potential bias that can be addressed in future efforts. Insulin and lactate values show correlations to BMI, Most importantly, the data collected provides us with caloric patterns and weight differences. In particular, low the necessary information to design a larger study in insulin and lactate values are likely predictive of weight which we can randomly assign participants to meal loss. This could be sufficiently motivating to dieters, a hy- plans over a longer period of time. Given that the last pothesis that needs to be tested in a future behavioural insulin and lactate measurements of the day are the study and over longer periods of dieting. most informative, a future study can restrict sample collection to these samples, allowing for data collec- Additional files tion over an entire diet period which normally takes place over several weeks. Once these molecular mea- Additional file 1: Supplementary Information. The file includes surements can be carried out by participants directly figures supporting the results of the paper. (DOCX 1132 kb) at home, recruiting participants not only from the Additional file 2: Raw Data. The file includes the raw data collected university campus would allow broadening of the during the study. (XLSX 46 kb) participant profiles. Extensions to the study can also include improve- Abbreviations ments to our digital health platform. Our current ACE: Angiotensin converting enzyme; B: Breakfast; BMI: Body-mass-index; D: Dinner; L: Lunch; NAD: Nicotinamide adenine dinucleotide; S: Snack app provides the setting that allows recording of life-style related data, including weight, food and Acknowledgements drink intake, exercise, and urine sample collection We would like to acknowledge help from Gail Calvert and Curtis Nicholson details. It also provides automation for the analysis in the initial setup of the chemiluminescence detection and for discussion of of the data. To broaden the use of the app sources of statistical bias with Roshan Klein-Seetharaman. Tejpal et al. BMC Obesity (2019) 6:20 Page 14 of 15 Ethical approval and consent to participate 13. Teixeira PJ, Silva MN, Mata J, Palmeira AL, Markland D. Motivation, self- The study was approved by Warwick Medical School Ethics committee determination, and long-term weight control. Int J Behav Nutr Phys Act. BSREC (protocol identification REGO-2014-1318) and all participants provided 2012;9:22–9. informed written consent about their participation in the study. 14. Gong Z, Gong Z. Modeling the relationship between body weight and energy intake: a molecular diffusion-based approach. Biol Direct. 2012;7:19. 15. Elobeid MA, Padilla MA, McVie T, Thomas O, Brock DW, Musser B, et al. Funding Missing data in randomized clinical trials for weight loss: scope of the The study was funded by University of Warwick. problem, state of the field, and performance of statistical methods. PLoS One. 2009;4(8):e6624. Availability of data and materials 16. Elliott P, Posma JM, Chan Q, Garcia-Perez I, Wijeyesekera A, Bictash M, et al. All data generated or analysed during this study are included in this Urinary metabolic signatures of human adiposity. Sci Transl Med. published article (and its Additional files 1 and 2). 2015;7(285):285ra62. 17. Molitor J, Brown IJ, Chan Q, Papathomas M, Liverani S, Molitor N, et al. Authors’ contributions Blood pressure differences associated with optimal macronutrient intake JKS carried out conceptualization of the study; ST, NS and JKS designed the trial for heart health (OMNIHEART)-like diet compared with a typical methodology; ST, NS, VM, JPI, KM and JKS performed the investigation; American diet. Hypertension. 2014;64(6):1198–U86. Writing of the original draft was done by ST, NS and JKS; All authors assisted 18. Du F, Virtue A, Wang H, Yang X-F. Metabolomic analyses for atherosclerosis, with revision of the manuscript. All authors read and approved the final diabetes, and obesity. Biomarker Research. 2013;1(1):17. manuscript. 19. Wang P, Holst C, Andersen MR, Astrup A, Bouwman FG, van Otterdijk S, et al. Blood profile of proteins and steroid hormones predicts weight change Consent for publication after weight loss with interactions of dietary protein level and glycemic Not applicable. index. PLoS One. 2011;6(2):e16773. 20. Lang R, Wahl A, Stark T, Hofmann T. Urinary N-methylpyridinium and trigonelline as candidate dietary biomarkers of coffee consumption. Mol Competing interests Nutr Food Res. 2011;55(11):1613–23. The authors declare that they have no competing interest. 21. Heinzmann SS, Brown IJ, Chan Q, Bictash M, Dumas ME, Kochhar S, et al. Metabolic profiling strategy for discovery of nutritional biomarkers: proline betaine as a marker of citrus consumption. Am J Clin Nutr. 2010;92(2):436–43. Publisher’sNote Springer Nature remains neutral with regard to jurisdictional claims in 22. Mozaffarian D, Hao T, Rimm EB, Willett WC, Hu FB. Changes in diet and published maps and institutional affiliations. lifestyle and long-term weight gain in women and men. N Engl J Med. 2011;364(25):2392–404. Received: 1 September 2018 Accepted: 7 March 2019 23. Bressanello D, Liberto E, Collino M, Reichenbach SE, Benetti E, Chiazza F, et al. Urinary metabolic fingerprinting of mice with diet-induced metabolic derangements by parallel dual secondary column-dual detection two- dimensional comprehensive gas chromatography. J Chromatogr A. References 2014;1361(Supplement C):265–76. 1. Moody RS, Alison. Adult obesity and overweight. 2014. healthsurvey.hscic. 24. Moore SC, Matthews CE, Sampson JN, Stolzenberg-Solomon RZ, Zheng W, gov.uk Cai Q, et al. Human metabolic correlates of body mass index. Metabolomics 2. Smith AG, Singleton JR. Obesity and hyperlipidemia are risk factors for early : Official J Metabol Soc. 2014;10(2):259–69. diabetic neuropathy. J Diabetes Complicat. 2013;27(5):436–42. 25. Jourdan C, Petersen A-K, Gieger C, Döring A, Illig T, Wang-Sattler R, et al. 3. Tejpal S, Bastie C, Seetharaman JK. Lemon juice: a potential source of Body fat free mass is associated with the serum metabolite profile in a angiotensin converting enzyme antagonism for weight loss and insulin population-based study. PLoS One. 2012;7(6):e40009. resistance. Proc Nutr Soc. 2018;77(OCE4):E213. 26. Gotthardt JD, Bello NT. Meal pattern alterations associated with intermittent 4. Liu B, Young H, Crowe FL, Benson VS, Spencer EA, Key TJ, et al. fasting for weight loss are normalized after high-fat diet re-feeding. Physiol Development and evaluation of the Oxford WebQ, a low-cost, web-based Behav. 2017;174:49–56. method for assessment of previous 24 h dietary intakes in large-scale 27. Gudzune KA, Doshi RS, Mehta AK, Chaudhry ZW, Jacobs DK, Vakil RM, et al. prospective studies. Public Health Nutr. 2011;14(11):1998–2005. Efficacy of commercial weight loss programs: an updated systematic review. 5. Liu J, Johns E, Atallah L, Pettitt C, Lo B, Frost G, et al., editors. An Intelligent Ann Intern Med. 2015;162(7):501–12. Food-Intake Monitoring System Using Wearable Sensors. 2012 Ninth 28. Affinita A, Catalani L, Cecchetto G, De Lorenzo G, Dilillo D, Donegani G, et International Conference on Wearable and Implantable Body Sensor al. Breakfast: a multidisciplinary approach. Ital J Pediatr. 2013;39(1):44. Networks; 2012 9-12 May 2012. 29. Neumann LB, Dunn A, Johnson D, Adams DJ, Baum IJ. Breakfast 6. Moshfegh AJ, Rhodes DG, Baer DJ, Murayi T, Clemens JC, Rumpler WV, et al. macronutrient composition influences thermic effect of feeding and fat The US Department of Agriculture Automated Multiple-Pass Method oxidation in Young women who habitually skip breakfast. Nutrients. reduces bias in the collection of energy intakes. Am J Clin Nutr. 2016;8(8):490. 2008;88(2):324–32. 30. Megson M, Wing R, Leahey TM. Effects of breakfast eating and eating 7. Vereecken CA, Covents M, Matthys C, Maes L. Young adolescents' nutrition frequency on body mass index and weight loss outcomes in adults enrolled assessment on computer (YANA-C). Eur J Clin Nutr. 2005;59(5):658–67. in an obesity treatment program. J Behav Med. 2017:1–7. 8. 24th European Congress on Obesity (ECO2017), Porto, Portugal, May 17–20, 31. Jarrett RJ, Keen H. Diurnal variation of Oral glucose tolerance: a 2017: Abstracts. Obesity Facts 2017;10(suppl 1)(Suppl. 1):1–274. possible pointer to the evolution of diabetes mellitus. Br Med J. 9. Inc A. Health & fitness - App store Downloads on iTunes. 2016 Available 1969;2(5653):341–4. from: https://itunes.apple.com/us/genre/ios-health-fitness/id6013?mt=8. 32. Jakubowicz D, Wainstein J, Landau Z, Raz I, Ahren B, Chapnik N, et al. Accessed 16 Mar 2016. Influences of breakfast on clock gene expression and postprandial Glycemia 10. Obesity – Android Apps on Google play. Available from: https://play.google. in healthy individuals and individuals with diabetes: a randomized clinical com/store/search?q=obesity&c=apps&docType=1&sp=CAFiCQoHb2 trial. Diabetes Care. 2017;40(11):1573–9. Jlc2l0eXoFGADAAQKKAQIIAQ%3D%3D%3AS%3AANO1ljJqCJM&hl=en_GB. Accessed 16 Mar 2016. 33. Nooyens ACJ, Visscher TLS, Schuit AJ, van Rossum CTM, Verschuren WMM, 11. Conway JM, Ingwersen LA, Vinyard BT, Moshfegh AJ. Effectiveness of the US van Mechelen W, et al. Effects of retirement on lifestyle in relation to Department of Agriculture 5-step multiple-pass method in assessing food changes in weight and waist circumference in Dutch men: a prospective intake in obese and nonobese women. Am J Clin Nutr. 2003;77(5):1171–8. study. Public Health Nutr. 2007;8(8):1266–74. 12. Scisco JL, Muth ER, Dong Y, Hoover AW. Slowing bite-rate reduces energy 34. Dhurandhar EJ, Dawson J, Alcorn A, Larsen LH, Thomas EA, Cardel M, et al. intake: an application of the bite counter device. J Am Diet Assoc. The effectiveness of breakfast recommendations on weight loss: a 111(8):1231–5. randomized controlled trial. Am J Clin Nutr. 2014;100(2):507–13. Tejpal et al. BMC Obesity (2019) 6:20 Page 15 of 15 35. Mohammadi D. The great breakfast myth. New Scientist. 2016; 229(3066):39–41. 36. Zhang L, Cordeiro SL, Liu J, Ma Y. The association between breakfast skipping and body weight, nutrient intake, and metabolic measures among participants with metabolic syndrome. Nutrients. 2017;9(4):384. 37. Gill S, Panda S. A smartphone app reveals erratic diurnal eating patterns in humans that can be modulated for health benefits. Cell Metab. 2015;22(5):789–98. 38. Turner-McGrievy GM, Wilcox S, Boutté A, Hutto BE, Singletary C, Muth ER, et al. The dietary intervention to enhance tracking with Mobile devices (DIET Mobile) study: a 6-month randomized weight loss trial. Obesity (Silver Spring, Md). 2017;25(8):1336–42. 39. Franco RZ, Fallaize R, Lovegrove JA, Hwang F. Popular Nutrition-Related Mobile Apps: A Feature Assessment. JMIR mHealth and uHealth. 2016;4(3):e85-e. 40. Turner-McGrievy G, Tate D. Tweets, apps, and pods: results of the 6-month Mobile pounds off digitally (Mobile POD) randomized weight-loss intervention among adults. J Med Internet Res. 2011;13(4):e120-e. 41. Gunnarsdóttir I, Sigurgeirsdóttir GK, Thórsdóttir I. Predictors of dropping out in a weight loss intervention trial. Ann Nutr Metab. 2010;56(3):212–6. 42. Newby FD, Wilson LK, Thacker SV, DiGirolamo M. Adipocyte lactate production remains elevated during refeeding after fasting. Am Jo Physiol-Endocrinol Metabol. 1990;259(6):E865–E71. 43. Walls HL, Backholer K, Proietto J, McNeil JJ. Obesity and trends in life expectancy. J Obes. 2012;2012:107989. 44. Crimmins EM PS, Cohen B. National Research Council (US) Panel on Understanding Divergent Trends in Longevity in High-Income Countries Washington (DC): National Academies Press (US); 2011. Available from: https://www.ncbi.nlm.nih.gov/books/NBK62367/. 45. McGinnis RA, McGrady A, Cox SA, Grower-Dowling KA. Biofeedback-assisted relaxation in type 2 diabetes. Diabetes Care. 2005;28(9):2145. 46. Sanghera N, Anderson A, Nuar N, Xie C, Mitchell D, Klein-Seetharaman J. Insulin biosensor development: a case study. Int J Parallel, Emergent and Distribut Sys. 2017;32(1):119–38. 47. Schoeppe S, Alley S, Rebar AL, Hayman M, Bray NA, Van Lippevelde W, et al. Apps to improve diet, physical activity and sedentary behaviour in children and adolescents: a review of quality, features and behaviour change techniques. Int J Behav Nutri Phys Activ. 2017;14:83. 48. Middelweerd A, Mollee JS, van der Wal CN, Brug J, te Velde SJ. Apps to promote physical activity among adults: a review and content analysis. Int J Behav Nutri Phys Activ. 2014;11:97.

Journal

BMC ObesitySpringer Journals

Published: May 6, 2019

There are no references for this article.