The Ecologic Validity of Fructose Feeding Trials: Supraphysiological Feeding of Fructose in Human Trials Requires Careful Consideration When Drawing Conclusions on Cardiometabolic Risk

Background Select trials of fructose overfeeding have been used to implicate fructose as a driver of cardiometabolic risk. Objective We examined temporal trends of fructose dose in human controlled feeding trials of fructose and cardiometabolic risk. Methods We combined studies from eight meta-analyses on fructose and cardiometabolic risk to assess the average fructose dose used in these trials. Two types of trials were identified: (1) substitution trials, in which energy from fructose was exchanged with equal energy from other carbohydrates and (2) addition trials, in which energy from fructose supplemented a diet compared to the diet alone. Results We included 64 substitution trials and 16 addition trials. The weighted average fructose dose in substitution trials was 101.7 g/day (95% CI: 98.4–105.1 g/day), and the weighted average fructose dose in addition trials was 187.3 g/day (95% CI: 181.4–192.9 g/day). Conclusion Average fructose dose in substitution and addition trials greatly exceed national levels of reported fructose intake (49 ± 1.0 g/day) (NHANES 1977–2004). Future trials using fructose doses at real world levels are needed.


Introduction
With the increase in high-fructose corn syrup (HFCS) consumption since 1970s, there has been rising interest in the role of sugars toward the development of cardiometabolic diseases (1). Particular attention has focused on the "fructose hypothesis, " which suggests that the metabolic and endocrine responses unique to fructose are the main drivers in the etiology of obesity, diabetes, and cardiometabolic risk (2,3). While this perspective is well supported by lower quality evidence from ecological studies (4) and animal models (5-7), it is not well supported by the highest level of evidence from controlled trials in humans (8)(9)(10)(11)(12)(13).
A main limitation of these trials has been the use of extreme levels of fructose feeding not representative of real world conditions. The present analysis aims to quantify the dose of fructose used in trials assessing the effects of fructose and cardiometabolic risk, and compare it to national levels of fructose consumption in the United States at the average and 95th percentile levels of intake based on the    National Health and Nutrition Examination Survey (NHANES 1977(NHANES -2004 (14).

Materials and Methods
We collated studies previously identified in a series of metaanalyses and systematic reviews of the effects of fructose on various cardiometabbolic endpoints (8)(9)(10)(11)(12)(13). We included controlled dietary trials across all populations investigating the effect of fructose on fasting blood lipids (Chiavaroli et al., unpublished study), postprandial triglycerides (13), blood pressure (9), glycemic control (Cozma et al., unpublished study), uric acid (11), non-alcoholic fatty liver disease (NAFLD) (12), body weight using mixed forms of fructose (solid, liquid, mixed) (10), and body weight from fructose-containing sugars-sweetened beverages only (Choo et al., unpublished study). Trials lasting <7 days, using intravenous administration or possessing unsuitable endpoints or comparators were excluded. Two types of trials were identified for the purposes of this analysis-substitution trials, in which fructose was exchanged for equal amounts of energy from other carbohydrates, or addition trials, in which a control diet was supplemented with additional energy from fructose compared to the control diet alone without the excess energy. Duplicate studies between metaanalyses were removed, and fructose dose data were extracted from each study when available and reported in grams per day. A weighted average fructose dose used across all studies was calculated according to the sample size of each trial, and reported as a mean and 95% confidence interval.

Results
The search and selection process can be found in Figure 1. A total of 16,673 reports were identified between all meta-analyses, and 203 reports (267 trials) were included after excluding reports based on title and abstract. After combining eligible trials and removal of duplicates from the meta-analyses, 64 substitution trials (1235 participants) and 16 addition trials (197 participants) were included in this analysis.  -maltose (3%), galactose (2%), and HFCS (1%) and comparators in all addition trials were diet alone.

Discussion
This analysis, which combined the trials identified from eight meta-analyses, aimed to examine the trends of fructose dose in controlled dietary trials assessing the effects of fructose on various cardiometabolic outcomes. We identified 64 substitution trials, in which fructose was provided in isocaloric substitution for other carbohydrate sources (usually starch), and 16 addition trials, in which fructose supplemented diets with excess energy compared to the same diets without the excess energy. The average weighted fructose dose was 101.7 g/day (95% CI: 98.4-105.1 g/day) in substitution trials from 1966 to 2014, whereas the average weighted fructose dose was nearly twice as high at 187.3 g/day (95% CI: 181.4-192.9 g/day) in the 16 addition trials from 1980 to 2013.
There were differences observed in the temporal trends between substitution and addition trials. Most substitution trials were conducted in 1980s with a resurgence that followed in 2010s. The reason for this pattern is unclear. A growing interest in fructose trials early on may have reflected the initial interest in fructose as a potentially beneficial alternative sweetener (69)(70)(71). By controlling for energy, substitution trials provided a rigorous study design, which allowed for the assessment of whether fructose had a unique set of metabolic or endocrine responses beyond its energy across a wide dose range. The emergence of the addition trials in 2000s may have grown out of the consistent lack of effect or even the benefit (glycemic control) seen in the substitution trials (8) and the concern stimulated by the ecological analysis of Bray et al. (4) linking fructose from HFCS with the epidemic of overweight and obesity. The recent resurgence of substitution trials in 2010s appears to have been to reconcile the role of energy from that Irrespective of any control for energy, the levels of intake observed in the available trials has been well beyond population levels of consumption. Compared to levels of reported fructose intake assessed by the National Health and Examination Survey in the United States (NHANES 1977(NHANES -2004, the doses used in both the substitution and the addition trials exceeded the average and 95th percentile levels of fructose consumption (49 ± 1.0 and 87 ± 4.0 g/day, respectively). Furthermore, all addition trials used doses of fructose above the 95th percentile of reported intake, with the weighted average dose more than double that amount. While the present analysis suggests that these trials using supraphysiological doses of fructose feeding are not representative of levels normally consumed in the diet, the important caveat remains that underreporting from national population intake surveys, such as NHANES, may underestimate the actual amount of fructose consumed (72). However, taking into consideration the potential for underreporting when interpreting calculated trial means compared to reported population means, if an estimated level of 50% underreporting were present (average and 95th percentile fructose intake of 100 and 172 g/day, respectively), the fructose dose in substitution trials would reach levels representative of true dietary intake [101.7 g/day (95% CI: 98.4-105.1 g/day)], while supraphysiological doses of fructose in addition trials would still persist [187.3 g/day (95% CI: 181.4-192.9 g/day)]. Another important consideration is that fructose consumption has been changing with time in NHANES. HFCS (a main proxy for fructose consumption) availability has been declining since it peaked in 1999 (73). Variability of fructose consumption over time should be taken into consideration when predicting the true population average intake.
The implications of our findings suggest a potential lack of ecological validity when drawing conclusions from addition trials using unrealistically high doses of fructose. As with the excess consumption of any macronutrient, an adverse effect on cardiometabolic risk factors may be irrelevant under levels of normal dietary consumption and lead to unnecessary concern and confusion regarding the safety of fructose. Two trial designs have helped to clarify whether adverse effects relate to excess energy (either from fructose or any macronutrient in general) or specific metabolic and endocrine properties inherent to fructose itself. In a series of systematic reviews and meta-analyses of controlled trials to determine the effect of fructose on various cardiometabolic outcomes, a consistent signal for harm has only been shown in the addition trials (8-10, 12, 13). Substitution trials have failed to show differences in body weight (10), fasting triglycerides (74), postprandial triglycerides (13), uric acid (9), glucose, insulin (8), or markers of NAFLD (12) with improvements seen in blood pressure (9) and glycemic control (8,75). These findings hold even under conditions of overfeeding as long as the excess energy is matched. The one exception may be for an effect on fasting triglycerides at a high dose threshold as seen in some subgroup analyses (76,77). Taken together, these findings suggest that fructose appears to be a determinant of cardiometabolic risk only in as much as it contributes to excess energy in the diet.

Conclusion
Most trials on fructose and cardiometabolic risk have used doses of fructose well beyond reported population levels of intake. While such high doses may be useful for determining a cause-effect relationship, replication of these studies using fructose doses closer to dietary levels are warranted and could help to establish a threshold beyond which excess energy from fructose demonstrate a signal for harm under real world conditions.