Metabolomics of Type 1 and Type 2 Diabetes: Insights into Risk Prediction and Mechanisms

Abstract

Purpose of review:

Metabolomics enables rapid interrogation of widespread metabolic processes making it well-suited for studying diabetes. Here, we review the current status of metabolomic investigation in diabetes, highlighting its applications for improving risk prediction and mechanistic understanding.

Recent findings:

Findings of metabolite associations with type 2 diabetes risk have confirmed experimental observations (e.g., branched-chain amino acids) and also pinpointed novel pathways of diabetes risk (e.g., dimethylguanidino valeric acid). In type 1 diabetes, abnormal metabolite patterns are observed prior to the development of autoantibodies and hyperglycemia. Diabetes complications display specific metabolite signatures that are distinct from the metabolic derangements of diabetes and differ across vascular beds. Lastly, metabolites respond acutely to pharmacologic treatment, providing opportunities to understand inter-individual treatment responses.

Summary:

Metabolomic studies have elucidated biological mechanisms underlying diabetes development, complications, and therapeutic response. While not yet ready for clinical translation, metabolomics is a powerful and promising precision medicine tool.

Introduction

Despite substantial progress over recent decades in uncovering the pathophysiology and mechanisms underlying type 1 (T1D) and type 2 diabetes (T2D), and improved treatment options, their incidence and prevalence continue to rise(1). Diabetes is defined by hyperglycemia and almost invariably results from progressive loss of pancreatic β-cell mass and/or function(2). However, there are many different processes that culminate in the β-cell destruction and dysfunction that characterize diabetes and this pathophysiological heterogeneity may have implications for treatment and prognosis. Molecular signatures of metabolic dysfunction in diabetes are therefore highly relevant to identify disease subtypes with distinct risk profiles and treatment responses and to identify risk of diabetes at earlier, potentially more modifiable, stages. Quantification of small molecule “metabolites” in the blood (“metabolomics”) is a promising avenue for metabolic interrogation due to its proximal role in reflecting metabolic dysfunction at the organ tissue level(3) prior to the development of overt hyperglycemia(4). Accordingly, metabolomic investigation for risk prediction and mechanistic understanding has been an active area of diabetes research for the last decade, since the advent of high-throughput metabolomic techniques capable of measuring hundreds of small molecules in a single blood sample(5). In this review, we provide an overview of human metabolomic investigation into the development, progression of complications, and response to therapy of both T1D and T2D, seeking to highlight mechanistic insights that have emerged from these human studies (Figure 1).

Figure 1. Metabolomics as a snapshot of diabetes pathophysiology.

Diabetes is the culmination of metabolic insults that originate from a combination of genetic and environmental factors. These factors cause metabolic dysfunction, inflammation, and autoimmunity (in the case of type 1 diabetes), and eventually lead to overt hyperglycemia and diabetes, of which the potential microvascular and macrovascular complications are well-described. At each stage of this process, distinct molecular signatures can be identified in the blood through metabolomic profiling. Shown here is an overview of select metabolite pathways that have been linked with the risk of incident diabetes in human epidemiologic studies. Higher levels of metabolites in red are associated with higher diabetes risk and metabolites in blue are inversely related to diabetes risk. Each metabolite displayed here is also involved in numerous other biological pathways that are not shown in this figure for the sake of simplicity.

Metabolites and risk of incident diabetes

Type 2 diabetes

High-throughput metabolite profiling techniques have been applied over the last decade to better understand mechanisms and risk markers for T2D. Initial efforts demonstrated associations of branched-chain amino acids (BCAAs; e.g., valine, leucine, isoleucine) and aromatic amino acids (e.g., phenylalanine, tyrosine) with incident diabetes(6) and with its precursor states, insulin resistance (IR) and obesity(3). The observational finding of higher circulating BCAAs and aromatic amino acids being related to higher risk of developing T2D has been well-replicated(3, 7–13), and is in concert with mechanistic studies that have elucidated the pathobiological links between BCAAs and diabetes. Higher levels of circulating BCAAs are a consequence of decreased rates of oxidation in adipose tissue and reduced hepatic BCAA uptake from inactivation of the branched-chain ketoacid dehydrogenase (BCKDH) complex(14). While it remains uncertain whether BCAAs are truly causal in the development of diabetes, as opposed to a reflection of other metabolic processes, several lines of evidence point to potential causal contributions. Accumulation of BCAAs in the circulation may activate mammalian target of rapamycin (mTOR), which increases gluconeogenesis through downregulation of Akt signaling(15). It has also been demonstrated that higher circulating BCAA levels occur from over nutrition-induced activation of the transcription factor carbohydrate-response element binding protein (ChREBP)(16). ChREBP drives the expression of enzymes involved in de novo lipogenesis, an increasingly recognized contributor to diabetes pathobiology(17). Further support of a causal mechanism for BCAAs in diabetes is provided from human genetics studies demonstrating that genetic variants related to higher circulating BCAA levels (including modulators of the BCKDH complex) are also associated with a higher risk of incident diabetes(18).

In addition to BCAAs, numerous other metabolites have been linked to diabetes and IR (summarized in Table 1). Glutamine and glutamate, for example, have been independently associated with IR and T2D(19, 20). Glutamine is involved in regulating insulin secretion and pancreatic β-cell function, while glutamate promotes gluconeogenesis, glucagon release, and other processes relevant to energy metabolism(21, 22). Accordingly, glutamine is inversely associated with diabetes risk, while glutamate is directly related(4, 23). Molecular substrates of the tricarboxylic acid (TCA) cycle have also been shown to be indicative of IR and diabetes risk. Higher levels of lactate, pyruvate, and alanine are related to obesity(3) and T2D(24). These associations partly reflect mitochondrial dysfunction and reduced oxidative capacity exhibited by skeletal muscle in the setting of IR(25), but lactate may also contribute to T2D development directly via stimulation of hepatic gluconeogenesis and by interfering with glucose uptake(26).

Table 1.

Summary of metabolite associations with diabetes risk and with diabetes complications.

Metabolite class/pathway	Example metabolite(s)	Associated phenotypes	Adverse direction	Potential mechanisms and biological implications
Branched-chain & aromatic amino acids	Branched-chain: Val, Leu, Ile	T2D, T1D, IR, retinopathy	↑ (↓ for tyrosine and retinopathy)	From diet or microbial metabolism; ↑ levels in blood from ↓ adipose and liver metabolism; lead to ↑ gluconeogenesis (via mTOR), ↑ de novo lipogenesis (via ChREBP)(15, 16)
	Aromatic: Phe, Tyr
Glutamate/glutamine cycle	Glutamate	T2D, T1D, retinopathy	↑ (↓ for retinopathy)	Glutamate promotes gluconeogenesis, muscle proteolysis, inflammation (via glutathione synthase)(20) Glutamine regulates insulin secretion, pancreatic β-cell function(4); contributes to cellular metabolism (anaplerosis)
	Glutamine		↓ (↑ for retinopathy)
Amino acid	Glycine	T2D, IR, hepatic steatosis	↓	Glycine involved in many homeostatic processes; protective; depletion may reflect oxidative stress and ↑ gluconeogenesis(9)
Glycolysis & cellular respiration	Lactate, pyruvate, alanine	T2D, T1D, kidney disease, CVD	↑	Pyruvate formed via glycolysis, leads to aerobic or anaerobic (lactate producing) respiration; alanine a substrate for TCA, Cahill cycles; ↑ levels due to mitochondrial dysfunction, ↓ oxidative capacity(3, 25)
ADMA metabolism	DMGV	Hepatic steatosis, fitness, T2D, CVD	↑	Formed by transamination of ADMA; dynamic to lifestyle changes(28, 29)
	ADMA	Retinopathy, CVD		ADMA is an inhibitor of nitric oxide (NO) synthase, noted to be elevated in diabetes
NO metabolism	Arginine, citrulline	Retinopathy	↑	Arginine and citrulline are NO precursors; ↑ levels in microvascular complications reflects altered NO metabolism
Lysine metabolism	Lysine	T2D, CVD	↑ (T2D)/ ↓ (CVD)	Amino acid, theorized to be elevated in T2D as response to upregulation of 2-AAA (for which it is precursor)
	2-alpha aminoadipic acid (2-AAA)	T2D	↑	Function is incompletely understood; produced by oxidative stress; promotes adipogenesis(33)
Vitamin C metabolism	Dehydroascorbic acid (DHAA)	T1D	↑	Oxidized form of vitmin C; substrate for glutathione; may inhibit insulin secretion(42)
Hydroxy acids and derivatives	3,4 dihydroxybutyric acid (DHBA)	Retinopathy	↑	Related to ketone bodies, may be higher in high fat diet; related to dyslipidemia(55)
Triglyceride metabolism	Triacylglycerols (TAGs) with ↓ carbon and double bond numbers	Hepatic steatosis, T2D, T1D, retinopathy, neuropathy, kidney disease	↑	Hypothesized to primarily reflect IR in the liver, may contribute to lipotoxicity
Additional lipid species	Phospthatidylcholines (PCs) and lysoPCs	T2D, kidney disease	↑/↓	Similar to TAGs, ↓carbon number and double bond content are linked with adverse outcomes; choline containing lipids, components of cell membrane and lipid metabolism(9)
	Acylcarnitines	T2D, neuropathy, CVD	↑	Reflection of impaired mitochondrial function, reduced oxidative capacity

This table is not an exhaustive list of all metabolites that have been associated with diabetes or its complications. It is intended to highlight metabolites discussed in this review and to demonstrate their links to relevant metabolic pathways.

Abbreviations: Val, valine; Leu, leucine; Ile, isoleucine; Phe, phenylalanine; Tyr, tyrosine; T2D, type 2 diabetes; T1D, type 1 diabetes; IR, insulin resistance; TCA, ticarboxylic acid; DMGV, dimethylguanidino valeric acid; ADMA, asymmetric dimethylarginine;

An important strength of metabolomic interrogation is the ability to measure both known molecular compounds and chemical structures with unknown curation, enabling novel discovery. In one such example, untargeted metabolomics was used to identify novel metabolite associations with liver fat. Through genetic analyses, the top metabolite association was identified as dimethylguanidino valeric acid (DMGV)(27). DMGV has subsequently been shown to be an important marker of cardiometabolic health that is associated with incident diabetes(27), poor cardiorespiratory fitness(28, 29), and future CVD(30). DMGV is formed via transamination of asymmetric dimethylarginine (ADMA) and it is theorized that higher circulating levels in cardiometabolic disease may be a reflection of greater transamination by alanine glyoxylate aminotransferase 2 in the context of reduced competition from β-aminoisobutyric acid (a protective compound linked to adipose tissue browning and leptin response)(27, 31). The metabolite 2-alpha aminoadipic acid (2-AAA) is another poorly understood metabolic intermediary that was identified via metabolomic quantification to be related to future T2D risk(32). Further investigation demonstrated that 2-AAA promotes adipogenesis, contributing to lipotoxicity, and may impair insulin signaling across different tissues(33).

Abnormalities in circulating lipid levels are well described to occur in the pathogenesis of diabetes, and elevated triglyceride levels are frequently observed in individuals with IR(34). Through profiling of individual triglyceride (triacylglycerol; “TAG”) species, investigators have demonstrated that TAGs with lower carbon number and double bond content (e.g., saturated or monounsaturated fatty acids) are more closely related to IR and diabetes risk(35). After a glucose challenge, TAGs with lower carbon number and double bond content tend to decrease, while other TAGs increase, likely reflecting the direct effects of insulin. Acylcarnitines may also be increased in IR due to impaired mitochondrial function and reduced beta-oxidation(36). Several additional lipid species have been associated with diabetes risk (e.g., phosphatidylcholines(9, 37, 38) and lysophosphatidylcholines(35)), but their mechanisms remain unclear.

Type 1 diabetes

T1D is speculated to typically result from specific factors (e.g., autoantibodies directed to the pancreatic β-cells), and is characterized by insulin deficiency. In the setting of insulin withdrawal in patients with T1D, the expected increase in proteolysis, lipolysis, ketogenesis, and other catabolic functions are observed and reflected by elevated circulating levels of BCAAs, glycerol, and beta-hydroxybutyrate(39). Intriguingly, however, it is increasingly recognized that widespread metabolic dysfunction may be apparent prior to the development of autoantibodies in T1D. Lamicchane and colleagues characterized circulating metabolite profiles of children who went on to develop T1D(40). They observed amino acid dysregulation to precede detectable islet autoantibodies, with lower levels of glutamic acid, aspartic acid, and tryptophan observed even in infancy in individuals who developed T1D. Metabolites of microbial origin also differed between individuals who developed T1D and controls, consistent with evolving evidence for a contribution of the gut microbiome to the pathogenesis of T1D(40). Abnormalities in circulating lipid profiles may also precede T1D diagnosis, with higher levels of several sphyngomyelins and lower levels of specific TAG and phosphatidylcholine species observed in individuals who eventually developed diabetes(41). In infants at increased risk of developing T1D, distinct metabolomic signatures comprised of dehydroascorbic acid (DHAA), amino acids, BCAAs and fatty acids were associated prospectively with autoantibody development(42). These findings suggest the possibility that autoimmunity may be a response to earlier metabolic dysfunction in T1D(43), presenting new opportunities for screening and prevention.

Metabolite profiling for predicting and understanding diabetes complications

It is well established that complications from diabetes—especially CVD, the leading cause of death in diabetic individuals—are not solely a consequence of elevated blood glucose levels(44–46). Additional metabolic insults occurring in the setting of diabetes are likely to contribute to complications and may vary across different tissues and organ systems. Metabolomic investigation is being increasingly utilized to uncover risk markers and mechanisms underlying diabetes complications, with potential applications for screening, prevention, and possible therapeutic modulation(44).

Retinopathy

While chronicity of disease, hyperglycemia, and blood pressure control are established risk factors for diabetic retinopathy (DR), there remains substantial unexplained variation in who develops DR despite similar levels of these clinical factors(47). Metabolites have shown promise both as biomarkers of increased risk for DR and as molecular snapshots that can be used to improve understanding of mechanisms underlying its pathogenesis. For example, circulating levels of ADMA and dimethylarginine (DMA) are higher in individuals with diabetes and DR versus those with diabetes but no DR(48). A unique advantage of metabolomic interrogation is that levels in the blood can be compared with levels in other complementary biofluids. In this case, higher ADMA levels have also been reported in the aqueous humor of individuals with severe diabetic retinopathy(49). As higher circulating levels of ADMA can limit NO bioavailability through inhibition of NO synthase, these observational findings build on experimental evidence suggesting a key role of deranged nitric oxide (NO) metabolism in the pathogenesis of DR(50). Higher blood levels of the NO precursors arginine and citrulline have also been described in individuals with DR, which may further reflect dysregulated NO metabolism(51). Moreover, arginine metabolism may be integral to the development of DR, as experimental model systems suggest arginase as a biological mediator of DR through its effects of reducing NO bioavailability and increasing oxidative stress(52).

Numerous other metabolites have been associated with DR severity in mostly cross-sectional studies, including amino acids (higher phenylalanine(53), lower histidine(53) and asparagine(54)) and several oxidated lipid species(54). While many of these studies have focused on patients with T2D, circulating metabolite levels have also been linked to DR in patients with T1D. In a study of 648 individuals with T1D, 3,4 dihydroxybutyric acid (DHBA) and several triglyceride species were associated with poorer DR grading(55). DHBA is a regulator of lipid metabolism and these findings, therefore, indicate that adverse lipid metabolism may be involved in DR development in T1D.

Recent efforts have sought to leverage multi-metabolite scores, often developed through machine learning approaches, to improve clinical prediction of DR(54, 56). While the initial data are promising, these models require validation against future (rather than cross-sectional) outcomes and broad validation across different population before they can be considered for clinical use.

Intriguingly, many of the metabolite associations that have been observed with DR risk so far are distinct from the metabolites that have been identified for future IR or diabetes itself. One notable example is the aromatic amino acid tyrosine. While higher tyrosine levels predict future diabetes development in apparently healthy individuals(6), lower tyrosine levels have been associated with increased risk of DR in individuals with T2D enrolled in the ADVANCE trial(53). Similarly, glutamine/glutamate metabolism has been related to DR(57), with lower glutamate and higher glutamine demonstrating associations with DR(58), which is opposite to the directionality observed for these metabolites and the risk of diabetes. While the biological processes underlying the different metabolomic signatures of diabetes and DR are not yet fully elucidated, metabolomic investigation of DR has clearly implicated separate metabolic mechanisms for diabetes and for DR, with important implications for screening/treatment(59).

Neuropathy

Similar to DR, investigations of metabolomic markers of diabetic neuropathy (DN) have largely implicated distinct metabolic signatures (and biological pathways) when compared to diabetes itself. Largely, these studies have focused on lipid metabolism given the extensive observational and experimental data establishing dysregulated lipid metabolism as being integral to the development of DN(60, 61). Circulating levels of several specific lipid species have been related to DN in diabetic individuals: higher 1-deoxydihydroceramides, potentially reflecting serine deficiency(62); higher ursodeoxycholate, which modulates lipid absorption(63); lower long-chain acylcarnitines(63), among others. These findings from lipidomic/metabolomic analyses support prior experimental models implicating dysregulation of peripheral nerve lipid metabolism as being central to neurotoxicity and DN pathogenesis(64, 65). Reflecting the accumulating evidence linking lipid metabolism to DN, an initial clinical trial of dietary supplementation with omega-3 polyunsaturated fatty acids (PUFA) to prevent lipotoxicity of sensory nerves in DN has shown improvement in DN symptoms and more definitive clinical trials are pending(66).

Kidney disease

Metabolic dysregulation is known to play an important role in the pathogenesis of diabetic kidney disease (DKD)(67). Accordingly, metabolites reflecting diverse biological processes have been related to DKD in mostly small cross-sectional studies(68). Lower levels of pyruvate and higher levels of glycolytic and TCA-cycle metabolites in individuals who develop DKD are in concert with findings from kidney biopsy samples implicating glycolytic flux as a potential mediator of DKD(69). In these studies, upregulation of enzymes and substrate for glycolysis was observed in individuals who did not develop DKD despite longstanding diabetes, suggesting that the ability to increase glycolysis in response to hyperglycemia was integral to glomerular cell integrity(69, 70). Altered tryptophan metabolism, reflected by higher levels of L-tryptophan and serotonin, has also been described in individuals with DKD, although the mechanisms are less clear(71). Dysfunctional lipid metabolism may contribute to progressive kidney disease in diabetes(72). In a study of 79 individuals with chronic kidney disease of whom ≈50% had diabetes, lower circulating levels of diacylglycerols (DAGs), cholesteryl esters, and phosphatidylcholines, and higher monoacylglycerols and phosphatidylethanolamine were associated with progression to end-stage renal disease(72).

Profiling of blood metabolites can be complemented by urinary metabolite profiling to provide additional information regarding the kidney’s metabolic responses in diabetes. This has been recently applied to identify urinary biomarkers of progression to DKD in T1D(73). Investigators evaluated 51 urinary metabolites from 24-hour urine samples in >2000 individuals with T1D in the Finnish Diabetic Nephropathy study and identified seven metabolites to be associated with progression of DKD: higher BCAAs (leucine, valine, isoleucine), pseudouridine, threonine, and 2-hydroxyisobutyrate, and lower citrate(73). The findings of higher urinary BCAA levels preceding DKD in T1D is intriguing and supports the hypothesis that development of IR contributes to DKD in T1D(73, 74).

Cardiovascular disease

It is well established that the high risk of CVD events in individuals with diabetes is not fully explained by hyperglycemia and that additional mechanisms such as inflammation, platelet hyperreactivity, endothelial dysfunction, autonomic nerve dysfunction, and membrane instability in various cell types play important roles(75–77). Additional biomarkers are, therefore, necessary to improve prediction of CVD events in individuals with diabetes and to improve preventative strategies.

In >700 individuals with T2D followed for over 5 years, a number of metabolites were associated with future CVD risk. These included associations of CVD with higher levels of DMGV (which has also been associated with hepatic steatosis and incident diabetes), N,N2-dimethylguanosine (a nucleoside derivative linked with cardiometabolic disease and CVD risk in those without diabetes(78)), homocitrulline, 1-methyladenosine, acylcarnitine C10:3, and urobilin and lower levels of lysine, hippurate, tryptophan, and threonine(79). Blood levels of BCAAs and aromatic amino acids have also been positively associated with the risk of CVD(13). Therefore, CVD-associated metabolites in diabetes have mostly implicated similar pathways as being involved with the pathogenesis of CVD and diabetes in individuals without diabetes (e.g., BCAAs, DMGV, N,N2-dimethylguanosine), while also indicating several metabolites with distinct associations that might reflect more complex biological regulation. Lysine, for example, is directly associated with diabetes risk, but is inversely associated with CVD risk in diabetes(80). Additional studies in larger numbers of individuals across both T1D and T2D are necessary to identify biomarkers and pathways most closely linked with CVD risk.

Dynamic changes in the metabolome in response to perturbation or therapy

A unique aspect of the circulating metabolome is that it can change rapidly in response to shifts in environmental factors (e.g., external stressors) or physiological changes. This feature can be leveraged to examine the metabolic response to a discrete stressor (e.g., glucose challenge or exercise) to evaluate disease susceptibility and biological pathways. It can also be used to provide insight into treatment responses, which may potentially enable the development of individualized and tailored, precision approaches to diabetes treatment.

Metabolic response to perturbation

Substantial changes are observed in the circulating metabolome in response to the increased metabolic demands imposed by acute exercise(29, 81). Favorable shifts in many metabolites previously linked to T2D risk (e.g., BCAAs, glutamate, DMGV, TAGs) are observed with an acute bout of exercise and these salutary changes may be blunted in the setting of obesity, implicating improved insulin sensitivity with exercise (and relative IR in obesity) as a potential unifying mechanism(29, 81). Indeed, chronic exercise training also exerts favorable responses in certain metabolites (e.g., DMGV and BAIBA(28, 31)) pointing to improvement in their representative metabolic pathways with habitual exercise. Future studies are needed to clarify optimal exercise training regimens for diabetes prevention and potential treatment.

The circulating metabolome also changes in response to a dietary stress, such as an oral glucose tolerance test (OGTT). The post-OGTT metabolic response is complex, with changes in metabolites with known roles in glucose and insulin metabolism as well as in metabolites with less-defined roles(82). In addition, large inter-individual differences can be observed in response to the same meal(83, 84). This observation was used in a recent study to identify novel genetic loci associated with the circulating lipid response to a mixed meal challenge(85). In turn, these genetic loci were also linked to body composition and clinical diabetes in large genetic consortia. Taken together, these findings suggest that inter-individual differences in the metabolic response to dietary perturbation can be assessed through metabolomic changes, and that these studies may improve understanding of diabetes susceptibility.

Weight loss and lifestyle interventions

Metabolites signatures demonstrate promise for uncovering the metabolic effects of different weight loss treatment strategies and, potentially, for targeting specific weight loss approaches towards patients that might derive the most benefit. For example, in obese individuals without diabetes, higher levels of BCAAs predict the degree of improvement in IR with weight loss(86). In the Diabetes Prevention Program, heterogeneity in response to lifestyle interventions versus metformin was observed across different metabolomic signatures, with higher levels of several phospholipids at baseline identifying individuals who appeared to benefit most from lifestyle modifications(87). Metabolite changes with different weight loss treatment strategies have also been investigated. The fall in levels of BCAAs and 2-hydroxybutyric acid following gastric bypass surgery correlate with improvements in IR, indicating potential mechanisms for metabolic benefits of surgical weight loss therapies(86, 88). Moreover, BCAA levels are noted to decrease more rapidly with surgical obesity therapies rather than weight loss from caloric restriction, mirroring the cardiometabolic health benefits reported for bariatric surgery(89).

Metabolic response to drug therapy for T2D

It is increasingly recognized that there are different subgroups of T2D that may benefit from different pharmacologic treatment approaches(90). Metabolites offer an attractive option for use in precision phenotyping as they (1) may enable the assigning of specific therapies to appropriate subgroups; (2) change rapidly in response to medication treatment; and, (3) can be assessed serially. While such approaches are not yet ready for clinical use, recent advances in employing metabolomic profiling to understand distinct metabolic effects of T2D medications or differences in inter-individual treatment responses make this an intriguing prospect.

In the Diabetes Prevention Program, heterogeneity in response to metformin treatment (in addition to variation in the effect of lifestyle modifications) was observed according to baseline metabolite levels(87). Notably, individuals with higher levels of AMP (a master cell energy regulator and potential mediator of metformin’s positive effects(91)) appeared to benefit more from metformin therapy, suggesting the potential of using metabolomics to assign specific drugs to specific patients(87). In a separate study, specific metabolites were observed to predict the therapeutic response to metformin or sulfonylureas(92). Higher levels of liver metabolites 2-hydroxybutanoic acid and 3-hydroxybutanoic acid were associated with a larger decrease in HbA1c with metformin treatment versus sulfonylurea treatment over the next 5 years(92). The acute metabolic effects of metformin have also been evaluated with a metabolomic approach to better understand its beneficial effects revealing changes in 36 metabolites after a single dose. Metabolites reflecting lipid metabolism, BCAA metabolism, energy homeostasis, and gut microbiota functions were among those with the greatest change in response to metformin administration(93).

Substantial recent interest has focused on using metabolomics to unravel the metabolic effects of sodium/glucose cotransporter-2 inhibitors (SGLT2i) due to their benefit across a broad array of cardiometabolic outcomes and uncertainty regarding their underlying mechanisms. In one study, untargeted metabolomics was performed on 25 individuals with T2D and CVD treated with empagliflozin for one month(94). After one month of treatment, 162 of 1269 assayed metabolites were altered including enrichment of TCA cycle intermediates that were not explained by increased rates of glycolysis(94). In a separate study, after 12 weeks of treatment with dapagliflozin, enhanced glycine degradation (better mitochondrial function), TCA cycle (energy metabolism), L-carnitine biosynthesis (energy metabolism), and super-pathway of citrulline metabolism (NO synthase and endothelial function) were identified in plasma and kidneys of individuals with T2D(95). Overall, these studies have demonstrated that the beneficial effects of SGLT2i may be partly explained by optimization of fuel substrate for end organs (such as the heart and kidneys). This effect of SGLT2i is theorized to act largely through SIRT and AMPK/mTOR, which stimulate a fasting/starvation metabolic response that results in preferential shift towards more optimal energy sources(96).

Metabolic response to drug therapy for T1D

The metabolic response to glycemic control in T1D is complex, with many metabolite levels corresponding to hyperglycemia itself, but metabolic profiles remain altered in T1D (when compared to controls without diabetes) even in the context of optimal blood glucose control(97). Levels of BCAAs and alanine appear to correspond with blood glucose levels in T1D, similar to T2D(98). With poor glycemic control, BCAAs, glucogenic amino acids such as glycine, serine, arginine, ketoacids, and others are noted to be elevated. While appropriate glycemic control can normalize certain pathways (e.g., nucleotide metabolism, lysine, histidine-glutamate-glutamine, cortisol, urea cycle), others remain elevated, suggesting widespread metabolic dysfunction in T1D beyond what can be explained by insulin deficiency alone(97).

Conclusions and future directions

Human metabolomic studies have identified numerous blood metabolites to be related to disease pathogenesis, risk of complications, and therapeutic response in both T1D and T2D (Table 1). In several instances (e.g., BCAAs), these findings point to pathways that may be biological mediators of diabetes and have served to either corroborate prior work in non-human experimental models or to discover novel pathways for further mechanistic interrogation. In this review, we have highlighted features of circulating metabolites that render them especially powerful tools for translational investigation in diabetes: (1) untargeted metabolomics can be used for discovery of novel compounds and biology; (2) metabolites are dynamic with controlled perturbations (e.g., exercise or dietary challenge) and may therefore provide a “readout” of the physiological response to stress; (3) metabolites can also be measured in complementary biofluids reflecting features of key end-organs (e.g., eyes, kidneys); and, (4) metabolites respond to lifestyle and pharmacologic interventions, and can therefore be used to guide precision therapeutics.

While metabolomic investigation in diabetes has already yielded important insights, the translation of these findings to clinical care will require several additional steps. Prior to clinical use, risk prediction models will need to be validated in larger samples, with representation across the age rage and encompassing racial and ethnic diversity(99). These prediction models will benefit from uniform approaches to metabolite quantification with assays developed for clinical use (rather than discovery). Studies of larger numbers of individuals are necessary also to explore the potential of using metabolomics to help define specific T2D subgroups that have implications for prognosis and treatment response(90). Metabolomics are likely to be an important part of the toolkit for precision diabetes care, but approaches that integrate multiple “omic” techniques (e.g., genomics, transcriptomics, epigenomics, microbiome, proteomics(100)) may have the greatest potential to advance the screening, prevention, and treatment of patients with diabetes.

Abbreviations:

BCAA

branched-chain amino acids

BCKA

branched-chain ketoacid

TCA

tricarboxylic