The subjects included 93 adults who participated in very low carbohydrate (LC) 122232425 and 70 who participated in low fat (LF) dietary studies 122 designed to examine the effects on weight loss, body composition, and other metabolic responses related to cardiovascular disease in the Human Performance Laboratory at the University of Connecticut (Table 1). The subjects were free of diabetes, cardiovascular, respiratory, gastrointestinal, thyroid and other metabolic disease. They were weight stable (± 2 kg) the month prior to starting the study. Subjects were not allowed to use nutritional supplements (except a daily multi-vitamin/mineral), or take medications to control blood lipids or glucose. The majority of subjects were sedentary and all were instructed to maintain the same level of physical activity throughout the study. The LC diet intervention was 12 weeks for the majority of subjects, but in some cases the duration was shorter (4–6 wk). Before and after LC and LF, body mass was determined in the morning after an overnight fast on a calibrated digital scale with subjects in light clothing and not wearing shoes. All subjects signed an informed consent document approved by the University of Connecticut Institutional Review Board.

Whole body composition was assessed using DXA (Prodigy™, Lunar Corporation, Madison, WI). Analyses were performed by the same technician (blinded as to dietary intervention) using commercial software (enCORE version 6.00.270). Percent body fat was calculated as soft tissue fat mass divided by the sum of soft tissue fat mass, soft tissue lean body mass, and bone mineral content. Coefficients of variation for lean body mass, fat mass, and bone mineral content on repeat scans with repositioning on a group of men and women in our laboratory were 0.4, 1.4, and 0.6%, respectively.

To identify genes involved in physiological responses to carbohydrate versus fat ingestion, we canvassed physiological pathways and biological processes relevant to the regulation of adiposity, including but not limited to eating behavior, digestion and absorption, hormonal signaling in the prandial and postprandial period, the regulation of fuel distribution and processing, and the control of adipocyte size and proliferation. From these categories, we selected 28 genes to represent three categories: food intake, energy homeostasis, and adipocyte regulation. Table 2 provides summary information relevant to the genes and SNPs. The role for each gene is described below.

All statistical analysis was performed using the R Statistics Language and Environment 525354. Covariates were analyzed using multiple linear regression, and selected using the stepwise procedure. The change in relative fat mass, Δ%BF, was calculated as the difference between post-diet and baseline measurements of %BF. To test for association with SNP genotypes, the residual of Δ%BF after covariate analysis was tested using linear regression on the SNP genotypes. SNP genotype was coded quantitatively as a numerical variable indicating the number of minor alleles: 0 for major homozygotes, 1 for heterozygotes, and 2 for minor homozygotes. The F-statistic p-value for the SNP variable was used to evaluate the significance of association. To test the validity of the p-values, we also performed an independent calculation of the p-values using permutation testing. The ranking of the first three SNPs were identical under permutation and F-statistic analyses (data not shown). To account for the multiple testing of 53 SNPs, we calculated adjusted p-values using Benjamini and Hochberg's false discovery rate (FDR) procedure 555657.

We use a locally smoothed function of the SNP frequency as it varies with Δ%BF to visually represent the nature of an association. LOESS (LOcally wEighted Scatter plot Smooth) is a method to smooth data using a locally weighted linear regression 5859. At each point in the LOESS curve, a quadratic polynomial is fitted to the data in the vicinity of that point. The data are weighted such that they contribute less if they are further away, according to the tricubic function

w i = ( 1 − | x − x i d ( x ) | 3 ) 3 , MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGacaGaaiaabeqaaeqabiWaaaGcbaGaem4DaC3aaSbaaSqaaiabdMgaPbqabaGccqGH9aqpdaqadaqaaiabigdaXiabgkHiTmaaemaajuaGbaWaaSaaaeaacqWG4baEcqGHsislcqWG4baEdaWgaaqaaiabdMgaPbqabaaabaGaemizaqMaeiikaGIaemiEaGNaeiykaKcaaaGccaGLhWUaayjcSdWaaWbaaSqabeaacqaIZaWmaaaakiaawIcacaGLPaaadaahaaWcbeqaaiabiodaZaaakiabcYcaSaaa@446C@

where x is the abscissa of the point to be estimated, the x_i are the data points in the vicinity, and d(x) is the maximum distance of x to the x_i.

The intake of total dietary energy at baseline did not differ between groups, nor did absolute amounts of protein, carbohydrate, and fat (Table 4). Relative to total energy intake, the intakes of protein, carbohydrate, and fat energy at baseline for LC and LF, expressed as percent of total energy, were 15.9%, 50.2%, and 32.6; and 16.6%, 47.7%, and 34.9%; respectively.

Dietary intervention decreased mean total energy intake to 1470 kcal in LF and to 1705 kcal/day in LC (p < 0.01), representing mean changes from baseline of -32.5% and -26.5%, respectively. During intervention, carbohydrate intake averaged 212 g/day (57% of total energy) in LF compared to 53 g/day (12% of total energy intake) in LC. Protein intake averaged 94 g/day (19% of total energy) in LF and 115 g/day (28% of total energy) in LC. Fat intake averaged 37 g/day (23% of total energy) in LF and 112 g/day (59% of total energy) in LC. All between-group differences in macronutrient gram intake were significant (p < 1^e-10, Table 4).

There were no differences in body size or composition between groups at baseline. With diet intervention, the change in total body, fat, and lean masses were significantly greater in LC (Table 5).

Figure 1 depicts the baseline and changes in % body fat profiles in the LF and LC study populations. The distributions are approximately normal and the baseline percent body fat levels for LF (35.4 ± 7.8%) and LC (34.5% ± 7.3%) were not different between groups. Overall LC induced an absolute decrease in % body fat of 2.98 ± 2.62%, reflecting losses of 4.8 kg of fat mass and 1.42 kg of lean mass, compared to a % fat decrease of 1.92 ± 1.63% in LF (p < 0.01), reflecting losses of 3.18 kg for fat mass and 0.70 kg for lean mass. Men outnumbered women in both study groups and had lower baseline % body fat in both groups (33.7 vs. 37.6 LF, and 32.9 vs. 38.0 LC) (Table 1). The change in % body fat (Δ%BF) due to diet was greater for men versus women (Table 5). For physiogenomics analyses, we employed gender as a covariate to account for the difference.

Table 6 lists the results of the association tests, comparing LF and LC groups. A single SNP rs322695 in the RARB gene was significantly associated with Δ%BF for both the LF and LC interventions (p < 0.0001 and p < 0.0121, respectively). SNPs in the HNMT (rs3100722, p < 0.002) and PFKL genes (rs2838549, p < 0.002) were significant only for the LF group. Conversely, the rs5950584 SNP in the AGTR2 gene was significant only for the LC group (p < 0.0001). The FDR-corrected p values yield an estimate of the false positive rate. The following SNPs, RARB rs322695, HNMT rs1269140, and PFKL rs2838549 and associations in LF, and the AGTR2 SNP rs5950584 association in LC are clearly significant after the FDR correction (p < 0.005, p < 0.041, p < 0.041, and p < 0.041, respectively). All remaining genes showed no significant association in either treatment group, and no gene showed significance for both diet treatments after adjusting for multiple tests.

Figure 2 shows the top three markers (according to p value) related to LF response. The first panel shows the LOESS curve for SNP rs322695 of the RARB gene. The frequency of the minor allele increases as the Δ%BF response to the LF diet becomes less pronounced (i.e., no loss of relative fat mass). The minor allele is completely absent among subjects with the largest decreases in %BF, and the frequency is <10% among subjects whose decrease in %BF exceeded 1%. This finding indicates a strong association between the RARB marker and response to LF. Similar patterns are seen for HNMT rs3100722 and PFKL SNP rs2838549. The SNPs, RARB SNP rs322695, HNMT SNP rs3100722, and PFKL SNP rs2838549, are considered "torpid" markers for responsiveness to the LF diet.

Figure 3 shows the top three markers associated with Δ%BF through LC. The first panel in Figure 3 shows the LOESS curve for SNP rs5950584 of the AGTR2 gene. The frequency of the minor allele is 30% in the Δ%BF range that is less than -5%, and the minor allele is completely absent among subjects with subtle change in Δ%BF (i.e., no loss of relative fat). A similar pattern is seen with the GYS2 SNP rs1478290. These two SNPs are considered reactive markers. The third panel shows the RARB SNP rs322695 response. The minor allele shows a higher frequency in the subject with little response in Δ%BF, indicating that the RARB SNP rs322695 is a torpid marker of responsiveness to LC diet also.

We observed a significant relationship between the RARB SNP rs322695 and change in relative fat to both diets after accounting for nongenetic factors. The relationship was significant after correction for multiple testing in mediation of the LF Δ%BF profiles (p < 0.005). This observation is noteworthy in light of the putative role of the retinoic acid system in insulin resistance 60. The RARB SNP rs322695 is found 100 kb upstream and ~600 kb from the adjacent THR (thyroid hormone receptor) gene. We have attributed this SNP to RARB because it is the next gene downstream, and regulatory elements such as enhancers have been found several hundred kb from their transcription start sites. However, it cannot be excluded that the SNP effect may be unrelated to RARB protein expression, and rather involve an as yet uncharacterized protein or non-coding RNA 61. We are unaware of other polymorphisms in the region or near RARB that are related to body weight or fat regulation in humans and believe this finding to be novel.

The widespread genomic distribution of retinoic acid response elements (RARE) suggests participation in many physiological pathways 626364. Recently Morgan et al. 40 described a neural hypothesis for regulation of annual cycles of body fattening in animals in which changes in the expression of RAR in the arcuate nucleus of the brain were noted. Vitamin A deficiency preferentially decreases hepatic RARB expression 65, and regulates hepatic glucose metabolic enzymes 66. Given that hepatic enzyme energy regulation occurs in part through RARB signaling, the present finding allows the hypothesis that the rs322695 variant affects such regulation.

In addition to the RARB gene, LF Δ%BF profiles had relationships through PFKL and one SNP in HNMT not found with LC. The HNMT SNP rs3100722, found at intron 2, and the PFKL SNP rs2838549, located in intron 8, showed significant relationships (p < 0.041) after correction for multiple tests) second in strength only to the RARB gene. The HNMT inhibitor metoprine suppresses feeding in mice 67. The present finding enables us to hypothesize a role for HNMT in the regulation of food intake. Hepatic phosphofructokinase, a key regulatory enzyme for glycolysis encoded by PFKL, is responsive to macronutrient changes 68 and is regulated by Vitamin A 66. Genetic loci near PFKL have been associated with bipolar affective disorder 69, but to our knowledge, no SNPs are known to modulate diet response.

The strongest association (p < 0.003, after correction for multiple tests) was found in the LC group with SNP rs5950584 in the angiotensin II receptor, type 2 (AGTR2). This SNP is found ~4.5 kb upstream in the promoter region. The AGTR2 gene has a metabolic role that is in contrast to the vascular role of the angiotensin II type 1 receptor (AGTR1 gene), which we previously found linked to statin-associated elevations in serum creatine kinase 17. Humans and mice express AGTR2 in muscle and adipose tissue, and evidence supports the existence of a functional renin angiotensin II system within adipose tissue 70. Mice lacking the AGTR2 receptor are resistant to the adipocyte hypertrophy and muscle cell insulin resistance induced by high fat, hypercaloric feeding 71. AGTR2-dependent angiotensin II signaling thus could account for LC unresponsiveness 71. AGTR2 polymorphisms were reported to modulate left ventricular mass, through the polymorphism known as G1675A (rs1403543) in a European population 72, and the T-A combination derived from G/T rs5193 and G/A rs5194 SNPs in a Cantonese population 73.

In the present study, the AGTR2 SNP rs5950584 serves as a reactive marker for the LC diet response. The AGTR2 gene is X-linked, thus men can have only one of either the T (major) or the G (minor) allele, while women may have zero, one, or two copies of either allele. Men lost significantly more fat than women, which together with the X-linked nature of the gene might lead to a false positive result. However, gender was highly significant in our covariate model, and the association test was performed adjusting for it, which excludes such a confounding effect.

Another association was found between LC response and the SNP rs1478290 in the GYS2 gene. The GYS2 product is glycogen synthase 2, a key regulator of hepatic glucose storage that is increased by feeding 74. The GYS2 SNP rs1478290 is found in the promoter region. We previously found a different GYS2 SNP, rs2306179, located in intron 5, to be associated with body weight loss in response to LC diet 5. The present study, conducted in a larger number of subjects, found also that a variant in the GYS2 gene is a response marker for the change in relative fat induced by LC.

The number of significant associations overall suggests a complex system through which regulation of the %BF phenotype is accomplished. Our list of genes, which is comparable to other studies to date 15, is exploratory, not comprehensive, and is representative of food intake, energy homeostasis, and adipocyte regulation. Virtually none of the genes in the present study were examined by previous diet studies 1315. Nevertheless, the present results showed a larger number of gene associations with Δ%BF profiles in relation to the LF diet compared to the LC diet. Based simply on the numbers of genes found to be significantly associated, we infer that the low fat diet utilizes a greater number of signal inputs into the regulation of relative body fat compared to the low carbohydrate diet. Even after the most stringent correction for multiple testing, RARB remained significantly associated with LF Δ%BF and AGTR2 remained associated with LC Δ%BF.

Our findings run counter to those of a recent study 15 that reported no significant SNP associations with changes in BMI following carbohydrate or fat restriction despite using large cohorts. Methodological differences between the studies are worth noting. The present study used DXA to precisely assess Δ%BF as opposed to relying on BMI. Second, different genes were studied. Here, the selection of genes was designed to reflect Δ%BF phenotype variability. In contrast, the SNPs examined by Sorenson et al. 15 were those demonstrated in previous reports to affect obesity and not diet response per se. Macronutrient compositions of the intervention diets were quite dichotomous in the present study, especially for carbohydrate (~12% and 57% of total energy as carbohydrate in the LC and LF arms, respectively, in the present study; vs. 42% and 57%, respectively, in the study of Sorenson et al. 15). We believe a greater imbalance in the macronutrient combination increases the strength of the dietary stimuli, which accentuates differences in the physiological signal intensities associated with each diet, increasing the chances for detection of physiogenomic relationships.

The prevention and treatment of obesity could be more efficient if dietary recommendations were carried out based on knowledge of innate individual characteristics that are genetically predictable. Moreover, adherence to dietary advice may increase when the advice is personalized. DNA-guided regimens allow healthcare providers to tailor therapies based on individual patient characteristics, rather than the average diet responses applied in longitudinal studies 375. A dietary treatment with a higher average chance for success, for example, may be desirable for those patients who inherited an ensemble of DNA markers with the most responsive factors and the least torpid factors for it. We interpret the coefficients associated with each significant gene marker in Table 4 as either having responsive or torpid effects to facilitate fat loss. The ability to match a diet regimen's genetic "contour" with the DNA profile of an individual patient marks the beginning of high-resolution personalized nutritional medicine for the treatment and prevention of obesity.

The number of subjects studied, the list of genes analyzed, and in some cases the frequency of the minor allele in SNPs of interest, were relatively small. Thus the results will need to be validated before clinical application. Nevertheless, the RARB and AGTR2 results are clearly significant even when fully and most conservatively corrected for multiple comparisons. The p-values were confirmed by non-parametric permutation analysis, excluding non-normal distribution as a source of false positives. In addition, the RARB result was found in the LF group and independently confirmed in the LC group. Although we assessed only a small sample of SNPs in the genome, this study included 53 SNPs from 28 genes representing a larger ensemble of genetic probes than other published diet studies. The data for the present study were compiled from a series of studies in whom the majority of subjects were of European ancestry. We lack power to detect ethnogeographic associations or confounding effects. We did not consider men and women separately but accounted for the effect of gender through its use as a covariate. Future studies investigating a sex difference might be useful. Finally, both diets were hypocaloric. It is possible that the genes shown to exert their effects through both diets, actually do so through the commonality of caloric restriction.