Metabolic Effects of Gastric Bypass Surgery: Is It All About Calories?

Obesity presents a major challenge to our society, due to comorbidities such as type 2 diabetes (T2D), cardiovascular disease, and nonalcoholic fatty liver disease. Bariatric surgery efficiently induces rapid and sustainable weight loss, also resulting in a rapid weight loss–independent remission of T2D (1) and improvements in cardiovascular disease risk factors (2) and nonalcoholic fatty liver disease (3). In the long term, T2D remission is likely a result of massive weight loss; however, the exact mechanism is not completely understood. Various factors may underlie the beneficial effects of bariatric surgery, including improved insulin secretion and action (4), increased β-cell mass (5), altered blood flow (6), bile acid (7), and gut microbiota composition (8). Despite the fact that reduced caloric intake results in T2D remission (9), we and others have shown that this may not be the only factor underlying bariatric surgery–elicited T2D remission (10,11).

While the majority of studies on T2D remission have focused on hormonal changes (12), a few studies investigated the effects on metabolism using metabolomics (13–21). Metabolomics enables a more comprehensive description of metabolism as compared with measurements of the one single metabolite glucose, which is assessed in most studies. Metabolomic studies have investigated both short-term (4–15 days) (19,20) and long-term (1–12 months) (13–19) consequences of Roux-en-Y gastric bypass surgery (RYGB) and sleeve gastrectomy. Collectively, these studies show effects on metabolism of all macronutrients, but do not resolve whether bariatric surgery elicits effects that differ from those observed after caloric restriction.

Very-low-calorie diet (VLCD) regimens for only 8 weeks induce T2D remission, which is then sustained for several weeks (22,23). This shared feature of bariatric surgery and VLCD suggests that the effect of RYGB on T2D remission could be caused by caloric restriction. In support of this, two recent studies found similar effects of RYGB and VLCD on weight loss, β-cell function, and insulin sensitivity (24,25). Despite these similarities, RYGB causes a more rapid T2D remission when compared with VLCD (26), as well as having different effects on the plasma metabolome (27). The lack of consensus may be explained by relatively small sample sizes in relation to the interindividual variability (21). A potential confounder in these studies is that dietary interventions tend to suffer from poor compliance (28), which is less expected after RYGB, in which food intake is physically limited. Hence, studies taking the interindividual variation into account are needed to resolve whether RYGB elicits unique effects or if it is just an extreme form of caloric restriction.

In the current study, we aimed to assess the immediate metabolic response to RYGB and compare that with that of calorie restriction (VLCD) in the same patients, thereby eliminating interindividual variation.

The study design has been described previously (11) and is summarized in Fig. 1A. Briefly, 19 age- and weight-matched Caucasian women with a BMI >35 kg/m² were enrolled in the study. Nine subjects were normoglycemic (NG; HbA_1c <6.0%, fasting glucose <6.1 mmol/L), and 10 were diagnosed with T2D at baseline (Supplementary Tables 1 and 2). Prior to surgery, patients were given weight-loss counseling and followed a VLCD regimen (858 kcal/day), aiming at losing 5% of body weight. Average age was 43 ± 6.3 years, BMI 39.8 ± 3.3 kg/m², and median HbA_1c 5.9% (interquartile range 3.5%). Median duration of T2D was 2.0 (interquartile range 2.75) years. RYGB was performed in a laparoscopic fashion, standardized to create a small (15–30 mL) gastric pouch, a 150-cm-long Roux-limb, and a 60-cm biliopancreatic limb. After surgery, patients were allowed sipping liquids and received 1,500–2,000 mL Ringer solution over 18–20 h. The patients were on a liquid diet for 2 weeks, followed by 2 weeks of purée, and a gradual introduction of solid foods. Expected caloric intake following surgery is 300 kcal/day the first few days, increasing to ∼1,500 kcal/day after 6 weeks.

Mixed-meal tests (MMTs; 220 kcal, 5 g fat, 28 g carbohydrates, and 14 g protein) (Modifast) were performed at each study occasion (i.e., before the intervention) (−4 weeks, baseline), after 4 weeks of VLCD (1 day prior to surgery; −1 day, diet), 1 day postsurgery (+1 day, surgery), and 6 weeks after RYGB (+6 weeks, recovery) (Fig. 1A). Blood samples were collected in prechilled EDTA tubes at several time points between 0 and 90 min after mixed-meal intake using a cubical vein catheter. Plasma was stored at −80°C until analysis. The study was approved by the Human Ethical Committee in Lund, Sweden, and adhered to the Declaration of Helsinki.

Metabolomics and lipidomics were conducted in plasma samples collected at 0, 30, and 90 min, reflecting the largest variation in glucose, insulin, glucose-dependent insulinotropic polypeptide (GIP), and glucagon-like peptide 1, during the MMT (11). Metabolites were extracted from 40 μL plasma (29) and analyzed on an Agilent 1,290 infinity UPLC connected to an Agilent 6550 iFunnel Q-TOF (Agilent Technologies, Santa Clara, CA), operated in positive and negative electrospray ionization mode. Metabolites of intermediate polarity were analyzed as previously described (29). Polar metabolites were separated on an Acquity BEH Amide column (1.7 µm, 2.1 × 100 mm) (Waters Corporation, Milford, MA), with an Acquity BEH Amide precolumn (1.7 µm, 2.1 × 5 mm) (Waters Corporation). Mobile phase A was 5 mmol/L ammonium formate and 0.1% formic acid, and B was acetonitrile/water (9:1; v/v) with 5 mmol/L ammonium formate and 0.1% formic acid, with the gradient: 0–2 min, 100% B; 2–5 min, 100–80% B; 5–9 min, 80–70% B; 9–11 min, 70–40% B; 11–14 min, 40% B; and 14–15 min, 40–95% B.

Lipids were extracted from 40 µL plasma with 150 µL methanol and 750 µL methyl tertbutyl ether (29,30). Following evaporation, the upper phase was reconstituted in 50 µL isopropanol/acetonitrile (90:10; v/v). Four microliters of sample was injected on an Acquity UPLC (Waters Corporation), equipped with an Acquity CSH C18 column (1.7 µm, 2.1 × 100 mm) (Waters Corporation) and an Acquity CSH C18 precolumn (1.7 µm, 2.1 × 5 mm) (Waters Corporation). Mobile phase A was acetonitrile/water (60:40; v/v), and B was isopropanol/acetonitrile (90:10; v/v), both with 10 mmol/L ammonium formate with the gradient: 0–13 min, 40–100% B; and 13–15 min, 100% B. Detection was performed in positive electrospray ionization on a Xevo 2G QTof MS (Waters Corporation).

Samples were analyzed in two batches, matched for glycemic status, using constrained randomization (31). Hence, samples from the same individual were analyzed in blocks, with randomization within and between blocks. Quality control samples, created by mixing aliquots from all samples, were analyzed eight times prior to the first injection and then every 10th injection. Metabolites were identified by tandem mass spectrometry using in-house libraries and the MassHunter METLIN Metabolite PCDL (Agilent Technologies). Leucine and isoleucine eluted as a single peak and were quantified as a single signal. Metabolites were integrated and manually confirmed in MassHunter Profinder B.06.00 (Agilent Technologies). Lipids were identified by tandem mass spectrometry using in-house libraries and integrated using MZmine 2.2.3 (32).

When the same metabolite was detected by multiple methods, data were kept from the analysis showing the lowest percent coefficient of variation of indicated metabolite in the quality control samples. Statistical analyses were conducted in R3.3.3. Metabolite and lipid values were log2-transformed, and missing data (≤15%) were imputed using k-nearest-neighbor averaging (impute.knn, impute); metabolites with >15% missing data were excluded from further analysis. Common variance compensation (ComBat, SVA) was used to remove batch-to-batch variation. For principal component analysis (PCA; prcomp, stats), data were mean centered and scaled to unit variance. Linear mixed-effect models (LMMs), using the individual as blocking factor and including the interaction between study occasion and glycemic state, were fitted to the data via REML (lmer, lme4), followed by general linear hypotheses and multiple comparisons for parametric models (glht, multcomp). Outcomes of the LMMs were visualized in heat maps using hierarchical clustering (heatmap.2, gplots). The effect of VLCD was defined as differences between samples collected after 4 weeks of VLCD (1 day before RYGB; −1 day) versus baseline (−4 weeks). The immediate effect of RYGB was defined as the difference between samples obtained at first meal after surgery (+1 day) versus 1 day before RYGB (−1 day), whereas the 6-week effect of RYGB was defined as the difference between samples collected at 6 weeks’ recovery after RYGB (+6 weeks) versus 1 day before RYGB (−1 day). The cumulative effect of the intervention (i.e., the collective effect of VLCD and RYGB) was defined as the difference between samples obtained at 6 weeks’ recovery after RYGB (+6 weeks) versus baseline (−4 weeks) (Fig. 1A). A Tukey test was used post hoc to evaluate the significance of study occasion and glycemic state. We evaluated associations between metabolite data and clinical markers assessing an individual’s glycemic status by building linear models on the scaled data (lmFit, limma) using empirical Bayes statistics for differential expression (ebayes, limma). Significance was defined as q < 0.05 using multiple testing adjustments according to the false discovery rate method.

The data sets generated and analyzed during this study are available for download from the MetaboLights data repository.

Anthropometric and clinical data for the patients included in this study have been reported previously using an unpaired analysis design (11). HbA_1c levels were lower in NG compared with T2D (Supplementary Table 1). VLCD resulted in significantly reduced weight (Supplementary Fig. 1) and BM, and improved HOMA of β-cell function (HOMA-β) (Supplementary Table 3). Six weeks after surgery, BMI and weight decreased further, but no additional effect on HOMA-β was observed (Supplementary Fig. 1 and Supplementary Table 3). Fasting GIP levels increased immediately after surgery. The cumulative effect of the intervention elicited a decrease in fasting GIP levels, weight, and BMI (Supplementary Fig. 1 and Supplementary Table 3). All 10 patients with T2D were in a state of remission, as judged from normal HbA_1c levels and absence of medication, for 4 years after RYGB, after which medication was restarted in 1 patient.

We determined relative levels of 104 plasma metabolites using two complementary platforms (Fig. 1B and Supplementary Table 4). First, we applied PCA to obtain an overview of the metabolite profiles in the fasting state. The score scatter plot revealed samples to cluster according to study occasion along principal component 2 (Fig. 1C and Supplementary Fig. 2). Samples collected 1 day after RYGB (+1 day) were most distant from those collected at baseline (−4 weeks), followed by samples collected after VLCD (−1 day) and samples collected after 6 weeks of recovery (+6 weeks) (Supplementary Table 5).

The PCA loading plot revealed substantial alterations in multiple metabolite classes, including lipids (Supplementary Fig. 2). Therefore, we investigated lipids in more detail using a dedicated platform, yielding data on relative levels of 72 lipids, including phosphatidylcholines (PCs), lyso-PCs, cholesteryl esters, sphingomyelins (SMs), and triglycerides (TGs). A PCA calculated on the lipid profiles in the fasting state revealed that samples collected at baseline (−4 weeks) clustered far from all other occasions along principal component 2 (Fig. 1D and Supplementary Fig. 2). Samples collected after VLCD (−1 day), 1 day after RYGB (+1 day), and after 6 weeks of recovery (+6 weeks) clustered together (Supplementary Table 5), indicating that lipid profiles were predominantly affected by VLCD, but not by RYGB.

Next, we examined alterations in fasting metabolite and lipid levels elicited by VLCD and RYGB using LMMs. VLCD resulted in a dramatic increase in levels of acylcarnitines, essential amino acids, and the ketone body 3-hydroxybutyrate and reduced levels of basic amino acids (Fig. 2A, Supplementary Fig. 3, and Supplementary Table 6). All changes remained significant after adjustment for insulin levels and the majority also after adjustment for weight loss (Supplementary Table 6). Notably, PCs and TGs with shorter fatty-acyl chain length were decreased and those with longer chain length increased (P < 0.05) (Fig. 2A and Supplementary Table 6). The majority of changes in lipids remained significant after adjustment for insulin levels, whereas those showing increased levels were significant also after adjustment for weight loss (Supplementary Table 6).

The 6-week effect of RYGB was associated with reduced levels of 6 out of the 20 detected acylcarnitines and 3-hydroxybutyrate (Fig. 2A and Supplementary Table 6). Changes in other polar metabolite classes were less systematic, and levels of very few lipids were affected (Fig. 2A and Supplementary Table 6). All effects remained significant after adjustment for insulin levels and most also after adjustment for weight loss (Supplementary Table 6). Hence, the 6-week effect of RYGB was associated with alterations in levels of acylcarnitines and 3-hydroxybutyrate, as well as SM 36:0, which were opposite to those observed after VLCD.

As RYGB can induce remission of T2D within days after surgery (1), we additionally examined the immediate effect of RYGB. The immediate response to RYGB elicited reduced levels of medium-chain acylcarnitines, whereas levels of the short-chain acylcarnitine 2:0 and carnitine were increased (Fig. 2A and Supplementary Table 6). Levels of most amino acids and 3-hydroxybutyrate were increased, purines were decreased, and few lipids were altered (Fig. 2A and Supplementary Table 6). All effects remained significant after adjustment for insulin levels and the majority also after adjustment for weight loss. Thus, the immediate effect of surgery was opposite to that of VLCD for several metabolites, mainly acylcarnitines, whereas the effect was similar regarding others, such as 3-hydroxybutyrate.

Finally, we investigated the cumulative effect of the intervention, which is similar to the comparisons examined in the majority of previous studies (21). Levels of medium- and long-chain acylcarnitines were increased, whereas those of the short-chain acylcarnitine 3:0 and carnitine were decreased (Fig. 2A and Supplementary Table 6). In addition, levels of 27 out of 72 detected phospholipids and TGs were altered (Fig. 2A and Supplementary Table 6). Overall, the cumulative effect was similar to the effect of the VLCD, including the reduction in PCs and TGs with shorter fatty-acyl chain length, and an increase in those with longer chain length (P < 0.05). The majority of changes remained significant after adjustment for insulin levels, whereas changes in levels of long-chain acylcarnitines and 11 lipids remained significant also after adjustment for weight loss (Supplementary Table 6).

As changes in acylcarnitines were a predominant feature in all comparisons, we investigated this class of metabolites in more detail. VLCD increased levels of total acylcarnitine, with the effect being more pronounced for medium- and long-chain compared with short-chain and free carnitine (Fig. 2B). Conversely, total acylcarnitine levels were decreased as an immediate and 6-week effect of RYGB, which was mainly driven by changes in short- and long-chain acylcarnitines (Fig. 2B and Supplementary Table 6). The VLCD and cumulative effects on acylcarnitine levels were subtle, in particular for medium- and long-chain acylcarnitines, in T2D as compared with NG (Fig. 2B and Supplementary Table 7). The immediate and 6-week effects of surgery were similar in both groups (Fig. 2B).

Both the VLCD and cumulative effect of the intervention on levels of the branched-chain amino acid (BCAA) (iso)leucine and lipids depended on the glycemic status (Supplementary Table 7). In particular, the increase of highly unsaturated lipids was less pronounced in T2D as compared with NG (Supplementary Table 7).

The targeted data analysis reveals that most changes occur as an effect of VLCD. Similar results were observed in an untargeted analysis (Supplementary Fig. 4 and Supplementary Table 8). Out of 719 detected features, 107 were affected by VLCD, 26 as an immediate effect of RYGB and 27 as an effect of RYGB. The cumulative effect of the intervention was associated with alterations in 58 features, out of which 40 were governed by VLCD and only 7 by RYGB.

Having established that diet and surgery elicit unique changes in the metabolite and lipid profiles, we investigated whether these changes are associated with clinical variables for assessment of glycemic status using fasting metabolite levels from all occasions in linear models adjusted for age and BMI. Insulin levels were negatively associated with levels of several acylcarnitines, HOMA-β was negatively associated and glucose was positively associated with levels of long-chain acylcarnitines, and the Matsuda index was positively associated with short- and medium-chain acylcarnitines (Fig. 2C and Supplementary Table 9). TGs with shorter acyl chains and few unsaturations showed positive associations with HOMA-β, insulin, GIP, and glucose levels and negative associations with glucagon-like peptide 1 and the Matsuda index, whereas those with longer acyl chains and higher degree of unsaturation were not correlated (Fig. 2D and Supplementary Table 9), suggesting that TGs harbor detailed information on glycemic state. Associations for all variables, except for HOMA-β, remained after adjustment for weight loss.

Finally, we evaluated the response to an MMT for each metabolite, estimated as the incremental area under the curve (iAUC). VLCD elicited decreased iAUC for the majority of acylcarnitines and several phospholipids (Fig. 3A and B and Supplementary Table 10). All changes remained significant after adjustment for insulin levels, whereas mainly acylcarnitines remained significant after adjustment for weight loss (Supplementary Table 10).

The iAUCs for the majority of metabolites and lipids remained largely unchanged as an effect of RYGB. Only the iAUC of acylcarnitine 3:0 was increased, an effect remaining significant after additional adjustment for weight loss or insulin levels (Supplementary Table 10).

The immediate effect of RYGB elicited increased iAUCs of dietary purines, as well as short- and medium-chain acylcarnitines, and reduced iAUCs for several amino acids (Fig. 3A and Supplementary Table 10). The majority of alterations remained significant after adjustment for insulin levels and changes in purines and medium-chain acylcarnitines also after adjustment for weight loss (Supplementary Table 10).

The cumulative effect of the intervention resulted in significantly decreased iAUCs for several acylcarnitines and lipids (Fig. 3A and B and Supplementary Table 10). The changes in iAUCs for most acylcarnitines remained significant after adjustment for insulin levels or weight loss (Supplementary Table 10).

None of the observed effects differed significantly between NG and T2D (Supplementary Table 11). At a nominal level, however, alterations in the iAUC for several acylcarnitines, BCAAs, aromatic amino acids ([iso]leucine, phenylalanine, and tyrosine), and phospholipids elicited by the VLCD and the cumulative effect of the intervention were significantly different between T2D and NG (Fig. 3B and Supplementary Table 11).

It is well established that both VLCD and RYGB result in improved glycemia (4). RYGB has gained enormous attention for its ability to induce rapid T2D remission (1), which was observed in all patients with T2D in this study. VLCD has a transient effect on T2D with respect to improved fasting plasma glucose, HbA_1c, and β-cell function (9,22,23,33). As presurgical caloric restriction is a safety prerequisite for undergoing RYGB, most studies assessing the impact of RYGB on various end points have in fact studied the combined effect of caloric restriction and RYGB (24,25). In this study, we assessed the impact of both types of interventions separately by using tight sampling intervals. Furthermore, the paired study design allowed us to eliminate interindividual variation. Our data show that the initial caloric restriction achieved by VLCD resulted in very distinct alterations in the metabolome, as compared with those later elicited by RYGB.

VLCD resulted in a marked increase in acylcarnitine levels, whereas RYGB, both immediately and after a 6-week recovery period, resulted in a smaller, but opposite, effect. Consequently, acylcarnitine levels were still above basal levels in the postoperative phase. This is in line with a previous study showing acylcarnitine levels to be elevated 15 days postsurgery, which were diminished to preoperative levels after 90 days (20). To some extent, these changes may be due to increased caloric and carbohydrate intake over time. However, whereas these changes have previously been attributed to RYGB, our data rather suggest that they result from the preoperative caloric restriction. An early study comparing RYGB with VLCD revealed both RYGB and VLCD to increase total acylcarnitine levels, whereas short (C3 and C5) acylcarnitines decreased only after RYGB (27). These results are in line with the effects of VLCD and the cumulative effect of the intervention observed in this study. Presumably, increased levels of acylcarnitines are caused by increased lipolysis in combination with excretion of acylcarnitines to avoid CoA trapping (34) and accumulation of toxic fatty acids (35). The reduction in acylcarnitine levels observed as an effect of RYGB may reflect a stress response (36), in line with observations of acylcarnitines to be elevated after the preoperative dietary restriction and then to decrease after induction of anesthesia (37). In addition, we found that the immediate effect of RYGB was associated with increased levels of 3-hydroxybutyrate, which has also been previously reported as an effect of anesthesia (38), but also may be caused by the preoperative fasting. Whereas anesthesia previously has been shown to reduce levels of most amino acids (38), our data reveal a more complex pattern. It must be noted that our samples are collected after and not during anesthesia. Hence, some of the immediate effects of surgery are likely to be associated with surgical stress, fasting, and anesthesia, whereas others are likely the result of the RYGB per se.

The VLCD-induced increase in acylcarnitine levels was less pronounced in patients with T2D when compared with NG individuals. Multiple studies have shown levels of acylcarnitines to be elevated in T2D and insulin resistance (39). Hence, a starvation-associated increase in acylcarnitines could blunt the reduction of these intermediates after normalization of insulin signaling, leading to normalization in levels of the metabolites. This is further supported by our findings of acylcarnitine levels being inversely associated with the Matsuda index. The lack of association with HOMA of insulin resistance implicates an effect on postprandial glucose disposal, which is perturbed in obesity, rather than on fasting insulin resistance (40,41). This notwithstanding, both indices may have shortcomings when applied on RYGB and during a negative energy balance.

We observed profound changes in lipid metabolism following VLCD, but less pronounced changes in response to RYGB. Notably, PCs and TGs with shorter fatty-acyl chain lengths were decreased after VLCD, whereas lipids with longer fatty-acyl chains were increased. These findings could explain some of the improvement in glucose homeostasis, as it is coherent with several previous findings in lipid metabolism associated with glycemic control and future risk of T2D (42–44). With respect to specific lipids, TG 48:1 and 50:0, which previously have been associated with an increased future risk of T2D (45), were positively correlated with insulin and glucose levels.

Our findings contradict previous studies in which similar changes have been observed but attributed to the RYGB (16–20) instead of the preoperative VLCD. Moreover, our finding of carbon number and unsaturation-dependent alterations in lipids contradict previous studies finding a coherent decrease in PC and phosphatidylethanolamine levels after RYGB (17–19). Our observation of decreased levels of long saturated SMs and increased levels of long polyunsaturated SMs is in agreement with a previous study (17).

We found a few weak (0.01 < q < 0.05), but unique, immediate, and 6-week effects of RYGB on the metabolite profile. However, these alterations, involving an unsystematic response in BCAAs and aromatic amino acids, elevated levels of arginine and taurine, and changes in presumably diet-derived metabolites such as caffeine and trans-cinnamic acid, do not support a beneficial effect on glycemia (45–49). Nonetheless, a coordinated decrease in levels of amino acid–derived short-chain acylcarnitines (C3–C5) may suggest a shift in substrate preference for oxidative metabolism.

Some limitations of the study should be mentioned. Firstly, patients were followed for a relatively short period of time, precluding prediction of long-term effects and potentially introducing carryover effects. Secondly, all samples except for baseline samples were taken very close to the interventions, with patients in a catabolic state, which was initiated during the VLCD and maintained throughout the subsequent interventions. The catabolic state is expected to remain for at least 12 months, as patients continue to lose weight in the first 12 months after surgery (50). Retesting at steady state would be interesting and will be attempted in the future, but then metabolic regulation occurring during T2D remission will be concealed. Finally, the number of individuals included in this study was fairly small, all were women, and all were in a state of T2D remission, precluding analyses stratified for remission. However, by combining constrained randomization with dependent analysis of effects, we could minimize the impact of individual and technical variation, thereby improving the power of our analysis (31).

Overall, our data suggest that the preoperative dietary restriction elicits most of the metabolic changes that have previously been attributed to RYGB. Our data are in concordance with more recent studies showing that weight loss either by surgery or VLCD elicits similar effects on glucose homeostasis, despite documented changes of incretin/insulin levels after RYGB (24,25). Moreover, some of the metabolic changes that associated with the immediate effect of surgery are likely to result from the surgical procedure and not RYGB per se.

Acknowledgments. The authors thank Anita Jonsson (Department of Surgery, Kalmar County Hospital) for technical assistance with MMTs.

Funding. This work was supported by the Påhlssons Foundation, the Crafoord Foundation, the Novo Nordisk Foundation, the Swedish Diabetes Foundation, the Hjelt Foundation, the Per and Ulla Schyberg Foundation, the Diabetes Wellness Foundation Sweden, the Horizon 2020 Program (T2Dsystems), the Swedish Research Council (Dnr 2017-00862, Dnr 521-2012-2119, Linnaeus grant, Dnr 349-2006-237, and Strategic Research Area Exodiab, Dnr 2009-1039), the Swedish Foundation for Strategic Research (Dnr IRC15-0067), and ALF.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. K.H., J.B., M.A.M., and C.B.A. performed experiments. K.H. analyzed the data and wrote the first draft of the manuscript. J.H., L.G., N.W., and P.S., planned and conceived the study. All authors contributed in writing the final version of the manuscript. P.S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.