|Home | About | Journals | Submit | Contact Us | Français|
The American Society for Parenteral and Enteral Nutrition (A.S.P.E.N.) Research Workshop, “Using Nutrigenomics and Metabolomics in Clinical Nutrition Research” was held on January 21, 2012 in Orlando, Florida. The conference brought together experts in human nutrition who use nutrigenomic and metabolomic methods to better understand metabolic individuality and nutrition effects on health. We are beginning to understand how genetic variation and epigenetic events alter requirements for, and responses to foods in our diet (the field of nutrigenetics/nutrigenomics and epigenetics). At the same time, methods for profiling almost all of the products of metabolism in plasma, urine and tissues (metabolomics) are being refined. The relationships between diet and nutrigenomic-metabolomic profiles, and between these profiles and health, are being elucidated, and this will dramatically alter clinical practice in nutrition 1.
One of the characteristics of nutrition research often is that there are significant variations in response to a nutrition intervention, and these result in large standard errors. This variance makes it harder to prove that the nutrition intervention had biological significant effects. One of the assumptions made by nutrition scientists is that people are metabolically similar; however it is becoming apparent that this may not be the case and there is significant metabolic individuality. This, in part, underlies the recent interest in individualized nutrition. For the clinical nutrition researcher it is important to realize that combining metabolic responders with non-responders to an intervention is an important source of variance in the study data, and may explain why studies sometimes result in divergent conclusions. If we could find biomarkers that identify the responders from the non-responders, we could appropriately group study subjects, not combine them, and greatly reduce variance in data, thereby enhancing identification of biologically significant effects using smaller numbers of study subjects. It would be best to separate responders from non-responders before a study is initiated; however, if the appropriate samples are collected during the study, it is possible to retrospectively analyze the data using appropriate grouping of subjects and thereby resuscitate an apparently failed clinical study.
Metabolism can be modified by many factors including diet, other environmental exposures and drugs and body composition. Possibly the most important modifiers of metabolism are genetic and epigenetic 2, 3. Whatever the source of variation in metabolism, this variation should be detected by metabolomic profiling which identifies a footprint of metabolism that is composed of the many small molecules generated by metabolic pathways. Thus, a clinical nutrition researcher should collect data on diet, exposures and body composition, as well as collect samples of DNA for genetic analyses and of biological fluids for metabolomic profiling (usually plasma and/or urine are used, but any fluid can be analyzed as appropriate to the study problem). With the help of a bioinformatician one can usually differentiate responders from non-responders in a clinical study.
In the old days, nutrition scientists measured a small set of targeted metabolites and extrapolated from these data to predict what was changing in metabolism. We often looked at what we knew best to find only what we expected to find; akin to looking under the streetlamp for lost keys. Modern technology makes it possible to measure hundreds of small molecules in a single sample (less than half a mL) of plasma or urine 4. This untargeted metabolite profiling permits scientists to use many more data points in developing their estimate of changes in metabolism, and it permits them to look at pathways they never suspected would be changed by an intervention. There are two main platforms available for performing metabolic profiling, and most laboratories use one or the other. Nuclear Magnetic Resonance methods have the advantage that little sample processing is required, but sensitivity can be a problem. Molecules present in low concentrations are often not seen. Mass spectrometry methods require some form of separation of the metabolites using chromatography before entry into the mass spectrometer, and are quite sensitive. Usually a metabolomic profiling platform will split the sample and pass it through a gas chromatography-mass spectrometer and a liquid chromatography mass spectrometer because these two separation methods complement each other for isolating the widest variety of small molecules. Once the data is generated, rather complicated analyses need to be performed to interpret the data, and this is best done by a team that consists of a bioinformatics expert and a metabolism expert.
The field of nutrigenetics/nutrigenomics is growing rapidly. For now, it is sensible to focus on two aspects that can be practically applied today in clinical nutrition research. Single nucleotide polymorphisms (SNPs; stated simply these are spelling errors in the genetic code) are very common, with more than 1 million identified 5. It is estimated that every person has about 50,000 SNPs. Some portion of these alters the expression or function of genes at critical steps in metabolism, and these changes result in metabolic inefficiencies that underlie metabolic variation between people. Current commercial technology allows us to measure a million or more of these SNPs in a study subject, but this may not be wise unless you are studying tens of thousands of subjects 1. This is because of biostatistical corrections that must be made when you make multiple comparisons. For clinically sized studies (tens to hundreds of subjects) it is better to preselect a small number of SNPs to be studied based on suspected underlying mechanisms.
It is now possible to measure epigenetic marks on genes 6. These marks influence whether genes are expressed. When DNA cytosines are methylated, they attract capping proteins that block of the binding site for transcription factors, and this blocks gene expression. This inhibitory signal can be reinforced by epigenetic marks on the proteins around which DNA is coiled (histones) as well as by microRNAs that bind to the gene and prevent it from being transcribed. Soon we will understand enough about epigenetics to study how these marks contribute to metabolic variation in human clinical nutrition studies. The major problem impeding progress is that these marks differ tissue by tissue. While SNPs are the same in all DNA, epigenetic marks on genes in peripheral white cells may not reflect the marks present in liver or brain or heart. It is a rare study that can obtain samples from such target tissues, and most have to extrapolate from blood sample data. A second problem for epigenetics is that these marks may change during development, so a single measure in time may not fully explain metabolic variation.
When the dietary requirement for choline in humans was studied, it was found that most men and postmenopausal women had to eat a diet containing choline or they developed liver and/or muscle damage that resolved when they were fed again with a choline-containing diet 7. However, less than half of premenopausal women developed liver or muscle damage when deprived of choline 7. This finding led to studies that determined that the pathway in liver for endogenous synthesis of phosphatidylcholine (a source of choline) is induced by estrogen; premenopausal women have extra capacity to make their own phosphatidylcholine and thus need to eat less choline in the diet 8. Why then did almost half of premenopausal women still need to eat choline? They had SNPs in genes of choline and folate metabolism that made theses metabolic pathways inefficient9, 10. The most common SNP that made women require choline in the diet was a SNP in the gene responsible for endogenous synthesis of phosphatidylcholine (PEMT); women with this SNP could not turn this gene on with estrogen 11. Thus, like men who have little estrogen, these women had to eat choline to prevent liver damage. More that 70% of women in North Carolina had one minor allele for this SNP, and approximately 20% had two minor alleles; the increase in relative risk for choline deficiency imparted by two copies of the minor allele was 24 fold 9.
Thinking about the above experiment, it would have been quite a puzzle if the data from young women were combined and analyzed because there were really two groups of women, responders and non-responders to the diet restriction. When combined there would be large standard deviations of the means, and many subjects would need to be studied to determine if there was a statistical difference between pre- and postmenopausal women and men in dietary requirements for choline. However, the ability to separate responders (those with the SNPs that increased the requirement for dietary choline) from non-responders (those women who did not need to eat choline) markedly reduced data variance, and made it easy to detect significant differences in a study of 80 subjects.
As noted above, these SNPs in genes of 1-carbon metabolism created metabolic inefficiencies that should be present even when a person is not challenged by being fed a low choline diet. These inefficiencies should be detected by metabolomic profiling. In the above study on human choline requirements, plasma samples at baseline (before any diet manipulation) were subjected to metabolomic profiling, and it was possible to predict with high confidence which subjects would develop liver damage when deprived of choline 12. Even when on a diet adequate in choline, these individuals had a group of metabolite changes caused by the SNPs in genes. In addition, metabolic profiling can be useful when studying response to an intervention because it provides a picture of metabolism with a greater scope than is usually obtained by measuring a few targeted metabolites. In the choline study, after subjects were fed a low choline diet there were expected changes in choline metabolites in plasma (they dropped), but there were unexpected changes in other metabolites – for example, changes in acylcarnitines suggested that mitochondrial function was disturbed during choline deficiency.
In summary, nutrigenetic and metabolomic methodology have great utility for scientists studying human nutrition. Every clinical study should plan to collect appropriate samples so that these methods can be employed to refine data analysis. This approach may reduce the “fuzziness” associated with human nutrition studies and may identify new mechanistic pathways because we are looking beyond the light shed by the streetlamp we were trained under.
Epigenetics is the study of mitotically heritable alterations in gene expression potential that are not caused by DNA sequence alterations 13. By stably regulating gene expression potential in differentiated tissues, epigenetic mechanisms such as DNA methylation play a critical role in mammalian development. In the last decade it has increasingly been recognized that dysregulation of epigenetic mechanisms may play an important role in human disease 14, 15. Indeed, nearly any disease with a genetic basis could also have an epigenetic basis. The inherent tissue-specificity of epigenetic gene regulation, however, presents a major obstacle to an improved understanding of the epigenetic basis of human disease 16. To determine if genetic variation is associated with a specific disease, any easily obtainable DNA sample – such as from peripheral blood – is sufficient, since essentially all cells in the body contain the same DNA. If, on the other hand, one wishes to determine if epigenetic variation is associated with, say, Alzheimer's disease or type-2 diabetes, epigenetic marks in peripheral blood DNA may be completely irrelevant.
Mouse studies of ‘metastable epialleles’ (MEs) suggest the potential to bypass in some cases this obstacle of tissue specificity. At murine MEs DNA methylation is established stochastically (i.e., randomly), even among genetically identical individuals 17, and interindividual epigenetic variation is influenced by maternal nutrition before and during pregnancy 18-21. Moreover, interindividual epigenetic variation at MEs occurs systemically, affecting all tissues 19, 20. The best characterized murine ME is the agouti viable yellow (Avy) locus, which affects coat color and body weight regulation. Whereas two genetically identical Avy heterozygous mice appear indistinguishable at birth, differences in DNA methylation at Avy can subsequently cause one to become yellow, hyperphagic, and obese while the other grows up with a normal brown coat and lean body type. One could take a few drops of blood from each of several newborn Avy heterozygous mice, however, and by measuring DNA methylation at Avy predict with absolute certainty which will become obese and which will be lean as adults. Hence, although the obesity of yellow Avy mice is caused by dysregulated agouti expression in the hypothalamus 22, the epigenetic lesion causing this misexpression is detectable in peripheral blood! The implications for human epigenetic epidemiology are obvious. Interindividual epigenetic variation at human MEs, detectable in peripheral blood DNA, may enable inferences about epigenetic dysregulation in internal organs and cell types of relevance to disease.
We performed a two-tissue epigenomic screen to identify MEs in humans 23. Using a DNA methylation microarray approach, we performed 4 interindividual comparisons, in each case comparing the same two healthy Caucasian adults using both peripheral blood and hair follicle DNA (representing the mesodermal and ectodermal lineages of the early embryo, respectively). Gene regions that exhibited concordant interindividual variation in both tissues were candidate MEs. Candidates were validated by testing for interindividual variation in DNA methylation in three tissues representing all three germ layers of the early embryo (liver, kidney, and brain) in post-mortem samples from Vietnamese accident victims. Moreover, several of the loci demonstrated substantial interindividual epigenetic variation among monozygotic twins, providing further evidence that this epigenetic variation occurs stochastically.
Seasonal variation in maternal dietary intake and nutritional status in rural Gambia, West Africa 24, provided a “natural experiment” by which to test whether the establishment of epigenotype at these loci is affected by maternal nutrition. In collaboration with Andrew Prentice and colleagues at the London School of Hygiene and Tropical Medicine, we obtained peripheral blood DNA samples from children (average age 9 years) who were conceived during either the rainy or the dry season in West Kiang, The Gambia (n=25 per season). Based on the mouse studies, we anticipated that individuals conceived during the nutritionally challenged rainy season – when villagers are running low on staple foods from the previous crop – would have lower DNA methylation at the MEs. We found just the opposite. At all five putative MEs tested, DNA methylation was significantly higher in children conceived during the rainy season 23. (This seemingly contradictory result may be due to the high availability of folate-rich leafy vegetables during the rainy season 25.)
In summary, our results show that epigenetic metastability does occur in humans. At select human genomic loci, establishment of systemic interindividual epigenetic variation occurs stochastically and is influenced by maternal nutrition before and during pregnancy. We anticipate that the identification of more such loci will highlight excellent candidate genes at which to study associations among early nutrition, epigenetic regulation, and human disease.
The progress of genomics, fostered by the Human Genome Project, has been spectacular. In just one decade, we have gone from having a rough draft of the Human Genome to being close to achieving the “$1,000 Genome.” The ability to conduct genome wide association studies (GWAS) using denser gene arrays has made feasible to conduct comprehensive genomic analysis in hundreds of thousands of individuals and the identification of hundreds of new loci associated with most common genetic disorders. Nevertheless, the translation of these findings into practical applications is still lacking. Moreover, most of the genetic variability attributed to Cardiovascular Disease (CVD)-related risk factors remains unaccounted, suggesting that additional genetic variants and genetic mechanisms need to be identified. Some have proposed that the “missing heritability” will be found buried into the epigenome or by conducting whole genome sequencing and identifying a myriad of new rare mutations associated with the phenotypes of interest 26. Another plausible explanation for this apparent “missing heritability” may be the presence of gene by environment interactions27. Among then the most relevant to health may be those resulting from gene-diet interactions that, has highlighted earlier (overview, Steve Zeisel), once properly identified and characterized could dramatically alter the translation of nutrition research into clinical practice.
One of the traditional contributions of nutrition research to public health has consisted on the definition of optimal dietary recommendations aimed to preventing disease and to promoting optimal health. For this purpose, and based on the best scientific evidence available at each time in history, several dietary guidelines have been implemented to improve the health of the general population and of those at high risk for specific diseases. However, past and current dietary guidelines have been based for the most part in observational epidemiological evidence and have not considered the dramatic differences in the individual's response to changes in nutrient intake [see above, Nutrigenetics and metabolomics can help define responders and non-responders in clinical nutrition studies (Steven Zeisel)]. These differences in response may greatly affect the efficacy of these recommendations at the individual level.
The mechanisms responsible for the interindividual differences in response to intervention, and particularly dietary response, are far from being fully understood28. Nevertheless, although the presence of a genetic component has been proposed for several decades, only recently researchers began examining these nutrient-gene interactions at the molecular level.
The different response of traditional CVD risk factors to diet depending on the particular characteristics of an individual is not a new observation but has already been widely observed and documented for decades. Having admitted that each individual may respond differently to the same diet, it becomes crucial to identify the factors defining such differential response.
The traditional approach to the identification of genetic factors implicated in differential dietary response has been based on the candidate gene approach. More recently with the ability to agnostically interrogate the entire genome, we can begin to identify unsuspected genes and biological pathways. However, a major barrier to the progress of Nutrigenomics towards the personalization of recommendations for CVD prevention has been the lack of consistency of the reported interactions across populations and studies29. Nevertheless, there are some cases that can be used as proof of concept for this approach. The one summarized below demonstrates a significant interaction between a functional polymorphism in the apolipoprotein A2 locus (APOA2), dietary saturated fat (SFA) and obesity risk.
Apolipoprotein A-II (APOA2) plays an ambiguous role in lipid metabolism, obesity, and atherosclerosis. Initially, we studied the association between a functional APOA2 promoter polymorphism (-265T>C) and plasma lipids (fasting and postprandial), anthropometric variables, and food intake in 514 men and 564 women who participated in the Genetics of Lipid Lowering Drugs and Diet Network (GOLDN) study30. We obtained fasting and postprandial (after consuming a high-fat meal) measures, including lipoprotein particle concentrations by proton nuclear magnetic resonance (NMR) spectroscopy and estimated dietary intake by use of a validated questionnaire.
We found recessive effects for this SNP that were homogeneous by sex. Thus, -265C/C subjects had statistically higher body mass index (BMI) than did carriers of the T allele and their odds ratio for obesity was 1.70 (P = 0.039). This could be explained in part by the fact that C/C individuals had a total energy intake that was statistically higher [mean (SE) 9371 (497) vs 8456 (413) kJ/d, P = 0.005] than in T allele carriers. This association remained statistically significant even after adjustment for BMI. Moreover, we found no associations with fasting lipids and only some associations with HDL subfraction distribution in the postprandial state. Therefore, this initial analysis demonstrated that the -265T>C SNP was consistently associated with food consumption and obesity, suggesting a new role for APOA2 in regulating dietary intake.
Next, we investigated the potential interaction among the APOA2 -265T>C SNP, food intake, and BMI. For this purpose, we carried out cross-sectional, follow-up (20 years), and case-control analyses in 3 independent populations 31. We analyzed gene-diet interactions between the APOA2 -265T>C SNP and SFA intake on BMI and obesity in 3462 individuals from 3 populations in the United States: the Framingham Offspring Study (1454 whites), the GOLDN Study (1078 whites), and Boston-Puerto Rican Centers on Population Health and Health Disparities Study (930 Hispanics of Caribbean origin).
The prevalence of the C/C genotype in study participants ranged from 10.5% to 16.2%. Most interesting, we identified statistically significant interactions between the APOA2 -265T>C SNP and SFA regarding BMI in all 3 populations. Thus, the magnitude of the difference in BMI between the individuals with the C/C and T/T+T/C genotypes differed by SFA. A mean increase in BMI of 6.2% (P = 0.01) was observed between genotypes with high- (> or =22 g/d) but not with low-SFA intake in all studies. Likewise, the C/C genotype was significantly associated with higher obesity prevalence in all populations only in the high-SFA stratum. Meta-analysis estimations of obesity for individuals with the C/C genotype compared with the T/T+T/C genotype were an odds ratio of 1.84 (P < 0.001) in the high-SFA stratum, but no association was detected in the low-SFA stratum (odds ratio, 0.81; P = 0.18). Therefore, we identified a gene-diet interaction influencing BMI and obesity that was strongly replicated in 3 independent populations.
In order to increase the level of evidence, we extended our findings to European and Asian populations 32. For this purpose, we did a cross-sectional study in 4602 subjects from two independent populations: a high-CVD risk Mediterranean population (n = 907 men and women; aged 67 ± 6 years) and a multiethnic Asian population (n = 2506 Chinese, n = 605 Malays and n = 494 Asian Indians; aged 39 ± 12 years) participating in a Singapore National Health Survey. In this case, the frequency of C/C subjects differed among populations (1-15%). Most important, we confirmed the recessive effect of the APOA2 SNP and replicated the APOA2-SFA interaction on body weight. In Mediterranean individuals, the C/C genotype was associated with a 6.8% greater BMI in those consuming a high (P = 0.018), but not a low (P = 0.316) SFA diet. Likewise, the C/C genotype was significantly associated with higher obesity prevalence in Chinese and Asian Indians only, with a high-SFA intake (P = 0.036). We also found a significant APOA2-SFA interaction in determining insulin resistance in Chinese and Asian Indians (P = 0.026). Therefore, the influence of the APOA2 -265T>C SNP on body-weight-related measures was also modulated by SFA in these Mediterranean and Asian populations.
Furthermore, we investigated some of the behavioral and hormonal mechanisms underlying our previous findings33. For this purpose, we evaluated relationships between APOA2 and obesity risk with particular focus on patterns of eating and ghrelin. The design was cross-sectional and we investigated overweight and obese subjects (n=1225) in five weight loss clinics in southeastern Spain. Our data show that C/C subjects were more likely to exhibit behaviors that impede weight loss and less likely to exhibit protective behaviors. Plasma ghrelin for C/C subjects consuming low SFA was lower compared with (1) C/C subjects consuming high SFA, (2) T/T+T/C carriers consuming low SFA and (3) T/T+T/C carriers consuming high SFA (all P<0.05). In summary, the APOA2 -265 T/C SNP was associated with eating behaviors and dietary modulation of plasma ghrelin, providing some mechanistic basis for our previous findings.
An example of the use of the more agnostic knowledge generated by GWAS in combination with the large sample size available from consortia is presented next. The focus of this work was the interaction between genetic variants associated with glucose homeostasis and type 2 diabetes and zinc intake. Zinc is an essential micronutrient that is important for β-cell function and glucose homeostasis. Therefore, we tested the hypothesis that zinc intake could influence the glucose-raising effect of specific variants 34. For this purpose, we conducted a 14-cohort meta-analysis to assess the interaction of 20 genetic variants known to be related to glycemic traits and zinc metabolism with dietary zinc intake (food sources) and a 5-cohort meta-analysis to assess the interaction with total zinc intake (food sources and supplements) on fasting glucose levels among individuals of European ancestry without diabetes.
Our analyses revealed a significant association of total zinc intake with lower fasting glucose levels (β-coefficient ± SE per 1 mg/day of zinc intake: -0.0012 ±0.0003 mmol/L, summary P value = 0.0003), while the association of dietary zinc intake was not significant. Moreover, we identified a nominally significant interaction between total zinc intake and the SLC30A8 rs11558471 variant on fasting glucose levels (β-coefficient ± SE per A allele for 1 mg/day of greater total zinc intake: -0.0017 ± 0.0006 mmol/L, summary interaction P value = 0.005). This suggests a stronger inverse association between total zinc intake and fasting glucose in individuals carrying the glucose-raising A allele compared with individuals who do not carry it. Therefore, our results suggest that higher total zinc intake may attenuate the glucose-raising effect of the rs11558471 SLC30A8 (zinc transporter) variant. Our findings also support evidence for the association of higher total zinc intake with lower fasting glucose levels.
These examples provide some proof of concept about the potential use of gene-diet interactions to identify individuals with differential response to dietary factors. However, in practice, we need to provide a more complete and clinically relevant picture. Most important, we still need to demonstrate whether personal risk information will trigger changes towards a healthier diet considering that most people tend to connect healthy eating with eating less and with less pleasurable food35. Therefore, we need evidence-based data to ensure that the knowledge generated by nutrigenetic science is properly implemented and scrutinized. Furthermore, as nutrition becomes increasingly integrated with preventive medicine, it is essential that dieticians and medical practitioners as well as geneticists are properly educated in the field of nutrigenetics/nutrigenomics. Therefore, it is essential to prove the initial hypothesis that individual approaches can improve public health better than global recommendations, and this benefit is available to the entire population and not only restricted to those with higher socioeconomic status and education 36.
In metabolic diseases such obesity and diabetes skeletal muscle fails to respond appropriately to the master counter-regulatory hormone, insulin, resulting in impaired glucose disposal after a meal. The onset of this “insulin resistant” condition is intimately associated with generalized increases in adiposity as well as ectopic lipid deposition within the muscle and other non-adipose tissues 37. A major quest in this field has been to identify specific lipid molecules that universally discriminate insulin responsive versus resistant states. To this end, our laboratory has employed a targeted metabolomics approach to survey several two-state models of insulin sensitivity. The methods used focused on quantifying approximately 150 intermediary metabolites measured in serum, urine and tissue extracts38. Results of these analyses identified a subclass of lipid-derived metabolites, known as the acylcarnitines, that correlated negatively with glucose tolerance 39, 40. Most of the even chain acylcarnitines are formed as metabolic byproducts of incomplete fatty acid β-oxidation and are derived from their respective acyl-CoA intermediates by a family of carnitine acyltransferases that reside principally in mitochondria. Our interpretation of the muscle acylcarnitine profiles was informed by experiments in which substrate flux, mitochondrial function and metabolic capacity were assessed by several complementary methods. In aggregate, the findings established a strong connection between mitochondrial bioenergetics and insulin action while raising new questions regarding the roles of incomplete β-oxidation and acylcarnitines as potential biomarkers and/or mediators of metabolic disease.
The observation that tissue acylcarnitines accumulate in several instances of insulin resistance tempts speculate that these metabolites might act as “lipotoxic” culprits. However, this suggestion is at odds with evidence that production of carnitine conjugates helps to avert mitochondrial dysfunction, owing at least in part to regeneration of free CoA 41, 42. Presuming that acylcarnitine production and efflux actually benefit mitochondrial function, we considered the possibility that prolonged exposure to lipid stress disrupts this defense mechanism by compromising carnitine availability. Consistent with this prediction, we uncovered a recurrent signature of carnitine diminution in multiple rodent models of overnutrition, metabolic disease and aging 42, 43. The decline in free carnitine in obese and/or aged rodents was associated with increased whole body fat oxidation, muscle accumulation of long chain acylcarnitines, a corresponding fall in short chain acylcarnitine species, and impaired substrate switching from fatty acid to pyruvate when assessed in isolated mitochondria. Subsequent studies showed that obesity-related derangements in mitochondrial fuel selection were reversed by dietary supplementation with L-carnitine, in parallel with improved glucose tolerance and insulin responsiveness 42-44. Interestingly, the antidiabetic effects of L-carnitine were accompanied by a shift in whole body fuel preference towards glucose oxidation (8); a surprising result given the prominent role of this nutrient in permitting mitochondrial import and beta-oxidation of long chain acyl-CoAs.
The initial step in fat oxidation is executed by carnitine palmitoyltransferase 1 (CPT1), which catalyzes the reversible transesterification of long chain acyl-CoA with carnitine. The long chain acylcarnitine product of CPT1 traverses the inner mitochondrial membrane via carnitine/acylcarnitine translocase (CACT) and is then delivered to CPT2, which regenerates acyl-CoA on the matrix side of the membrane where β-oxidation occurs. Importantly however, in addition to its requisite role in fatty acid oxidation, carnitine also facilitates mitochondrial efflux of excess carbon fuels. Thus, in the event that rates of substrate catabolism exceed energy demand, accumulating acyl-CoA intermediates are converted back to their membrane permeant acylcarnitine counterparts, which readily exit the organelle and tissue. Fitting with the latter function, carnitine supplemented rodents had robust increases in tissue efflux and urinary excretion of acetylcarnitine 42, 43. This specific metabolite derives from acetyl-CoA via the action of carnitine acetyltransferase (CrAT), a mitochondrial matrix enzyme that converts short chain CoA species to their corresponding acylcarnitine esters. Acetyl-CoA holds a prominent position in intermediary metabolism as the universal end product of fatty acid, glucose and amino acid oxidation. As its major metabolic fate, acetyl-CoA typically enters the tricarboxylic acid (TCA) cycle where it drives production of reducing equivalents that in turn fuel ATP synthesis by the electron transport chain. During conditions of lipid surplus, a rise in the mitochondrial pool of acetyl-CoA results in feedback inhibition of pyruvate dehydrogenase (PDH), the enzyme complex that connects glycolysis to glucose oxidation 45. We therefore surmised that increased flux through the CrAT reaction might serve to mitigate lipid-induced suppression of PDH. Direct experimental evidence that CrAT activity can indeed impact glucose homeostasis came from studies wherein the metabolic consequences of overexpressing the enzyme were examined in primary human skeletal myocytes. As predicted, the resulting enhancement of acetylcarnitine production and efflux increased cellular PDH activity, glucose oxidation and glucose uptake 43, thus mimicking the therapeutic actions of L-carnitine supplementation. Our current working model suggests that carnitine buffers intramitochondrial imbalances between acyl-CoA load and TCA cycle flux, thereby affording protection against nutrient-induced mitochondrial stress.
In summary, targeted metabolomics analyses pointed to a heretofore unappreciated role for carnitine and CrAT in regulating skeletal muscle glucose disposal; and this interpretation was subsequently corroborated by direct experimental evidence from animal and cell-based models. Ongoing investigations are now testing the anti-diabetic potential of L-carnitine therapy in human subjects with impaired glucose tolerance. Taken together, this work provides an example of how metabolomics approaches can be used to identify potential sites of metabolic dysfunction while also serving as a guide for more traditional, hypothesis-driven nutrition research.
Metabolomics measures metabolic phenotypes that are the net result of genomic, transcriptomic, and proteomic variability, therefore providing the most integrated profile of biological status. The pathological development and the drug intervention of diabetes mellitus (DM) involve altered expression of downstream low molecular weight metabolites including lipids and amino acids, and carbohydrates such as glucose. Currently a small number of markers used for clinical assessment of type 1 and 2 DM treatment may be insufficient to reflect global variations in pathophysiology.
Metabolomic biomarker discovery is a young research area that carries great hopes for both medicine and the nutritional sciences, particularly for the early detection of well-characterized metabolic disorders such as DM. The principal concept of metabolomic biomarker research is to identify key metabolites (other than glucose) differing in a control and a diabetic group, with diagnostic or prognostic abilities.
In this workshop, we would like to discuss the clinical application of metabolomics by means of two metabolomic studies of diabetes conducted by our group. Recently we performed a metabolomics study on a new phenotype of DM, fulminant type 1 diabetes mellitus. Fulminant T1DM is newly discovered as a subtype of T1DM (first reported in 2000), and is defined as the acute destruction of pancreatic beta cells as well as alpha cells, leading to extremely rapid progression of hyperglycemia and ketoacidosis 46. The prevalence of FT1DM was estimated to be 8.9% in all T1DM patients and 0.2% in newly-diagnosed all diabetic patients 47. Although there is no report on the prevalence of fulminant T1DM in China, the number may be comparable given a similar genetic background and lifestyles between the two East Asian populations.
The rapid progression of hyperglycemia and ketoacidosis of fulminant T1DM leads to almost total destruction of beta cells within a few days (typically less than one week) and thus, a high death rate if appropriate therapies are not in place. To date, the pathogenesis of this disease has not been established. Additionally, there is no early detection method for such a rapidly progressing disorder.
Four groups of age-matched human subjects, healthy controls (n=20), T1DM (n=6 + 20, with and without ketoacidosis), T2DM (n=20), and Fulminant T1DM patients (n=6) were recruited by the Shanghai 6th People's Hospital affiliated to Shanghai Jiao Tong University. Only male subjects were selected in the current study. The diagnostic criteria for Fulminant T1DM were based on published literature 48, and included all of the following: (1) occurrence of diabetic ketosis or ketoacidosis within 7 days after the onset of hyperglycemic symptoms (elevation of urinary and/or serum ketone bodies at first visit); (2) plasma glucose ≥16.0 mM and glycated hemoglobin level (HbA1c) < 8.5% at first visit; and (3) fasting serum C-peptide level <0.3 ng/ml (<0.10 nmol/l) and <0.5 ng/ml (<0.17 nmol/l) after intravenous glucagon (or after meal) load at onset. Venous blood samples were taken from individuals after overnight fasting for at least 10 h and the serum samples were obtained in the normal manner. Aliquots of serum samples were stored at -80 °C until metabolic analysis was performed.
Serum samples were prepared, chemically derivatized, and measured with a LECO's gas chromatography time of flight mass spectrometry (GC-TOFMS), following our published protocols 49. GC/MS data files were pretreated as previously described 50. The mean-centered and autoscaled data were then introduced into the SIMCA-P 11.5 Software (Umetrics, Umeå, Sweden) for multivariate statistical analysis. Principal component analysis (PCA) was used to obtain an overview of metabolic variations among the different groups. Orthogonal projections to latent structures discriminant analysis (OPLS-DA), a supervised pattern recognition approach, was utilized to construct a predictive model to identify key metabolites differentially expressed the each disease phenotype.
Fulminant T1DM subjects exhibited distinct metabolic profile in the scores plots compared to healthy controls, T1DM, and T2DM, indicating a robust identification through serum metabolite expression levels. A panel of serum metabolites that are differentially expressed in fulminant T1DM was identified as potential diagnostic markers for fulminant T1DM. The differential metabolites between fulminant T1DM and classic T1DM (with and without ketoacidosis) mostly overlapped with those derived from comparison between fulminant T1DM and healthy controls, which include significantly altered ketone bodies and free fatty acids. These metabolites reflect the perturbed metabolism of ketone bodies and fatty acids under the condition of insulin deficiency. Only those differential metabolites that are distinct in fulminant T1DM were chosen as potential biomarkers.
The study demonstrated that metabolomics profiling can contribute to the development of a panel of biomarkers for more sophisticated classification of the diabetic diseases, which may ultimately serve as an early diagnostic approach for the fulminant T1DM.
In the second example, a metabolomic study was performed to determine metabolic variations associated with type 2 diabetes (T2DM) and the drug treatments on 74 patients who were newly diagnosed with T2DM and received a 48 week treatment of a single drug – repaglinide, metformin or rosiglitazone49. Fasting overnight and 2h postprandial blood serum of patients were collected at 24 and 48 weeks to monitor the biochemical indices (FPG, 2hPG, HbA1c, etc.). Gas chromatography/mass spectrometer coupled with multivariate statistical analysis was used to identify the alteration of global serum metabolites associated with T2DM as compared to healthy controls and responses to drug treatment. Significantly altered serum metabolites in diabetic subjects include increased valine, maltose, glutamate, urate, butanoate and long-chain fatty acid (C16:0, C18:1, C18:0, octadecanoate and arachidonate), and decreased glucuronolactone, lysine and lactate. All of the three treatments were able to down-regulate the high level of glutamate to a lower level in serum of T2DM patients, but rosiglitazone treatment was able to reverse more abnormal levels of metabolites, such as valine, lysine, glucuronolactone, C16:0, C18:1, urate, and octadecanoate, suggesting that it is more efficient to alter the metabolism of T2DM patients than the other two drugs.
There are on the order of 10 times more microbial cells in the human the body than human cells51. Within the genomes of those microbial cells, there are on the order of 100 times more genes than genes encoded by the human genome. The metabolomic capacity of the microbial community is thought to substantially exceed that of the liver 52. Determining how the microbial community impacts human phenotypes will require the execution of metagenome-association studies in which next generation sequencing is used to link the state of the microbial community to human health and disease phenotypes. In order to understand the extent of normal variation in the microbial community, the Human Microbiome Project (HMP) is performing a survey of 18 tissue types in over 200 healthy individuals 53. An initial analysis of this healthy cohort is encouraging in suggesting the complexity of the microbial community is not infinite and therefore that metagenome-association studies may be tractable. For ~5,000 samples within the HMP cohort, 454 sequencing technology targeting the V3-V5 region of the 16S rRNA gene was used to generate over 30,000,000 sequences 54, 55. If we cluster these sequences into operational taxonomic units (or OTUs, defined as groups of sequences with an average percent identity of 97%) with a clustering strategy 56 that requires that an OTU consist of sequences that are seen often enough that they can form consensus sequences, we find that a very modest number of V3-V5 OTUs (695) can account for nearly all (~98%) of the V3-V5 sequences in the HMP cohort 57. Moreover, when we compare the consensus sequences that represent these 695 OTUs to an existing database of full-length sequences, nearly all of them have a very high quality match (>97% identity) 57. Taken together these data suggest that, at least viewing the microbial community through the lens of taxa defined by 97% average sequence identity, the complexity of the human associated microbiome is not infinite but rather consists of a small number of taxa that have been seen in multiple cohorts. In a metagenome association study, each taxa will be considered in a null hypothesis that the taxa is not associated with phenotpyes of interest. If there were vast numbers of taxa, we would need vast sample sizes in order to evaluate large numbers of null hypotheses. The relatively modest numbers of taxa discovered within the more than 5,000 V3-V5 samples within the human microbiome project suggests that, by contrast, modest sample sizes will be required to see significant effects. Moreover, since nearly all of the V3-V5 OTUs have been previously observed in other studies, we can have some confidence that the results of a metagenome association study performed on one cohort may be informative for other cohorts.
While there appear to be only a modest number of taxa in the non-rare biosphere, there is still tremendous individual variation within these taxa 54. Nearly all V3-V5 OTUs within the HMP cohort show two to three orders of magnitude in variation in different people 58. That is, in samples from some people a particular OTU may represent nearly all of the sequences in the sample, while in other people that OTU may represent less than 0.01% of the sequences or be completely absent58. This variation is seen in all tissue types with both V1-V3 and V3-V5 primers 58. These data tell us that while the number of “parts” that make up the microbiome is limited, these “parts” exist at very different levels of abundance in different people. This is consistent with the observation that each person has a distinct microbiome that is stable over time59, 60. This individual variation provides a great challenge to metagenomic association studies: it is an open question as to whether the high degree of individual variation will confound attempts to link individual or sets of taxa to health and disease phenotypes across multiple subjects.
One intriguing idea that has recently been introduced in an attempt to reduce the complexity of individual variation in the gut microbial community is the concept of enterotypes 61. Arumugam et al. 61 argue that despite individual variation, people can be clustered into two or three distinct types based on their gut community. The enterotype concept is attractive because it offers a substantially simplification of metagenome associations studies; if subjects can be easily classified into enterotypes, hypothesis testing can proceeed based on which enterotype classification subjects were assigned rather than on other more complex and more variable characteristics of the microbial community. In the HMP cohort, there was some support for the enterotypes concept when classifying gut taxa at the genus level but not at the more detailed OTU level 58 suggesting that individual variation at the sub-genus levels could potentially confound enterotypes. The appropriate level of taxonomic resolution for metagenome association studies remains an open question62 and it will be interesting to see if future metagenomic association studies have more success when working at more or less refined levels of the taxonomic tree.
While the technologies that enable metagenomic association studies are still undergoing rapid change, metagenomic association studies with modest samples sizes are finding some success in establishing statistically significant associations. For example, a recent study showed several taxa that were associated with the tendency of subjects to develop fatty liver on a low choline diet59. When this taxanomic information was combined with host SNP information for a gene involved in choline synthesis, a simple model could be established that was nearly perfectly correlated with the degree of fatty liver observed in response to a low choline diet 59. While this sort of successful metagenomic associated study is encouraging, it remains to be seen how reproducible such observations will be across multiple cohorts. Next-generation sequencing technology is continuing to rapidly evolve and this will make obtaining metagenomic profiles both faster and less expensive, reducing the barriers to achieving substantial sample sizes on multiple future cohorts. In particular, the increasing read-length of the Illumina platform is making it possible to achieve great sequencing depth on 16S-rRNA gene based community profiling for only a few dollars a sample63. As sequencing costs continue to decline and bioinformatics pipelines become more refined, we are moving towards a future where a personalized view of each individual's microbiome may become part of the individual genetic background that drives personalized medicine.
The 2012 ASPEN Research Workshop “Using Nutrigenomics and Metabolomics in Clinical Nutrition Research” was supported by Award Number U13DK064190 from the National Institute of Diabetes and Digestive and Kidney Diseases and by awards from Abbott Nutrition and GenoVive. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Diabetes and Digestive and Kidney Diseases or the National Institutes of Health. Support for this work also was provided by grants from the National Institutes of Health (DK55865, DK56350) to S.H.Z, from the March of Dimes Foundation (1-FY08-392), NIH (R01-DK081557), and USDA (CRIS 6250-51000-055) to R.A.W, from DK089312, AG028930, a Claude Pepper Older Americans Center grant AG028716 and the American Diabetes Association to D.M.M.