Variants in Genes Controlling Oxidative Metabolism Contribute to Lower Hepatic ATP Independent of Liver Fat Content in Type 1 Diabetes

Nonalcoholic fatty liver disease (NAFLD) tightly relates to insulin resistance, the hallmark of type 2 diabetes (1). Only recently, type 1 diabetes has been also linked to NAFLD with similar adverse outcomes compared with patients with type 2 diabetes (2). Likewise, the role of insulin resistance in the development of type 1 diabetes has gained more interest (3–5). Insulin resistance in type 1 diabetes was described more than 30 years ago (4) and originally accounted for by long-term glucose toxicity (6). But mitochondrial function may also be impaired as shown by lower muscle ATP synthase flux (7). Interestingly, muscle ATP synthesis inversely correlates with hepatocellular lipid content (HCL) (8), suggesting a tight link between liver and muscle energy metabolism.

In overt type 2 diabetes, lower hepatic concentrations of energy-rich substrates such as inorganic phosphate (Pi) and ATP synthesis associate with increased fat content and insulin resistance in the liver (9,10). However, hepatic respiratory capacity was transiently elevated and followed by augmented production of lipid peroxides in the NOD mouse, a model of human type 1 diabetes (11). Recent studies provided further evidence for enhanced hepatic energy metabolism and oxidative stress in human obesity and NAFLD (12,13). It is therefore unclear whether hepatic energy homeostasis is altered only in the context of diabetes-related NAFLD and which factors contribute to such abnormalities.

Specifically, the impact of inherited or acquired factors possibly involved in hepatic energy metabolism has not been investigated. Genetic variants in regulators of oxidative phosphorylation like peroxisome proliferator–activated receptor-γ and ∆ (genes PPARG and PPARD), PPARG coactivator-1α (gene PPARGC1A), respiratory chain complex 1 (gene NDUFB6), and fat mass– and obesity-associated gene (gene FTO) also modulate glucose and lipid metabolism (14–18). Several cytokines, such as adiponectin, fetuin A, and fibroblast growth factor 21 (FGF21), can be altered in insulin-resistant states, but also in human (19–21) and rodent type 1 diabetes (11).

Thus, this study tests the hypothesis that hepatic energy metabolism is already lower in young nonobese near-normoglycemic patients with type 1 diabetes than in age- and BMI-matched healthy individuals and that it associates with impaired hepatic and peripheral insulin sensitivity independent of liver fat content. Highly sensitive magnetic resonance spectroscopy (MRS) was used to assess hepatic energy metabolism from hepatic ATP concentrations (22), which tightly correlate with unidirectional flux through hepatic ATP synthase (23).

Fifty-five patients with type 1 diabetes of the German Diabetes Study (GDS), a prospective observational study aiming at characterization of subphenotypes and monitoring of disease progression (clinicaltrials.gov registration no. NCT01055093) (24), fulfilled the following inclusion criteria: 1) age 18–69 years, 2) known diabetes duration of <12 months, and 3) type 1 diabetes diagnosis based on ketoacidosis with immediate insulin substitution, detection of at least one islet cell–directed autoantibody (islet cell autoantibody, glutamic acid decarboxylase autoantibody, or islet antigen-2 antibody), or C-peptide below the detection limit. Exclusion criteria were pregnancy, malignancies, and severe chronic diseases, including liver disease other than NAFLD. Healthy normoglycemic individuals from the GDS, the German National Cohort Study (25), and a pilot study at the German Diabetes Center (clinicaltrials.gov registration no. NCT01736202) served as the control group (CON) (n = 57). These volunteers fulfilled the inclusion and exclusion criteria of GDS except for the presence of diabetes. Particularly, they did not have a family history of diabetes and had normal glucose tolerance based on a standard 75-g oral glucose tolerance test. All participants were screened, including medical history, physical examination, and routine blood test. Self-reported alcohol consumption was <20 g/day. Informed consent was obtained from all volunteers prior to inclusion after approval of the trial by the ethics board of Heinrich Heine University.

All 112 participants had fasting blood samples and ³¹P/¹H-MRS measurements. All 55 patients with type 1 diabetes and 20 of all 57 CON, matched for age and BMI to 20 patients with type 1 diabetes of the larger patient group, underwent the modified Botnia clamp test on a separate day. They refrained from physical activity and any alcohol ingestion for 3 days prior to visits and remained fasted overnight for 10–12 h before the tests. Patients with type 1 diabetes withdrew their regular insulin dose in the morning of the tests.

The modified Botnia clamps were performed as previously described and validated (26). In brief, an intravenous infusion of [6,6-²H₂]glucose was started at −120 min and continued until +240 min. At 0 min, an intravenous glucose bolus (1 mg/kg body weight [BW] in a 30% [weight for volume] solution containing 1.98% [6,6-²H₂]glucose) was administered followed by the clamp test using an insulin dose of 1.5 mU * (BW in kg)⁻¹ * min⁻¹ (Insuman Rapid; Sanofi, Frankfurt, Germany) from +60 until +240 min. Blood glucose concentrations were maintained at 5 mmol/L using a variable intravenous 20% glucose infusion.

Blood samples were immediately chilled and centrifuged and supernatants stored at −20°C until analysis. Venous blood glucose concentrations were measured with the Biosen C-Line glucose analyzer (EKF Diagnostics, Barleben, Germany) (27). Serum triglycerides (TGs), cholesterol, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and γ-glutamyl transpeptidase (GGT) were analyzed using a Cobas c311 analyzer (Roche Diagnostics, Mannheim, Germany) (27). Free fatty acids (FFAs) were quantified enzymatically (Wako, Neuss, Germany) in samples containing orlistat to prevent ex vivo lipolysis (27). Serum C-peptide, insulin, and plasma glucagon were measured radioimmunometrically (Millipore, St. Charles, MO) (27). Circulating FGF21, fetuin A, selenoprotein P (SepP), and high-molecular-weight adiponectin were determined by ELISA (13,27). Peroxidation of endogenous lipids and nucleic acid oxidation were assessed from thiobarbituric acid reactive substances (TBARS) and 8-hydroxydeoxyguanosine (8-OHdG), respectively (11,13). Protein carbonyl products and oxidized LDL (oxLDL) were measured with ELISA (Biocat, Heidelberg, Germany; intra-assay coefficient of variation: 1.6 and 2.5%, respectively). Antioxidative capacity was assessed colorimetrically from measurement of plasma glutathione (GSH) (USBio, Boston, MA; intra-assay coefficient of variation: 1.6%). Atom percent enrichment of ²H in glucose was determined after deproteinization and derivatization to the aldonitrile-pentaacetate as previously described (26).

Genomic DNA was extracted from whole blood with the Qiagen Blood and Tissue Kit (Qiagen, Hilden, Germany). Genotyping was performed using real-time PCR–based allelic discrimination with TaqMan predesigned single nucleotide polymorphism (SNP) genotyping assays and chemistry (Thermo Fisher Scientific, Darmstadt, Germany) on a StepOnePlus Real-Time PCR System (Thermo Fisher Scientific).

All measurements were performed in a 3-T magnetic resonance scanner (Achieva 3T; Philips Healthcare, Best, the Netherlands) using a 14-cm circular ³¹P surface transmit-receive coil (Philips Healthcare) for ³¹P-MRS and the built-in ¹H whole-body coil for localization and ¹H-MRS. Absolute quantification of γATP and Pi was performed as previously described (22). Data from localized ¹H-MRS were analyzed to assess fat content as previously described (28), and absolute concentrations were expressed as percent HCL relative to water content. Concentrations of phosphorus metabolites were corrected for the volume captured by lipid droplets within hepatocytes (9). Phosphocreatine contamination in the liver spectra was avoided by an MR localization technique excluding signals from abdominal wall muscle. Truncal subcutaneous and visceral adipose tissue contents were assessed by MRI as previously described (27).

Whole-body insulin sensitivity was calculated as M value from the mean glucose infusion rate during the last 30 min of the clamp test. Rates of glucose appearance were calculated by dividing the tracer infusion rate times tracer enrichment by percent of tracer enrichment in plasma and subtracting tracer infusion rate (26). Endogenous glucose production (EGP) was calculated from the difference between rates of glucose appearance and mean glucose infusion rates. Hepatic insulin sensitivity was calculated as percent EGP suppression by insulin during clamp steady state (iEGP).

Normally distributed data, given as means ± SEM, were compared with the two-tailed unpaired Student t test. Nonnormally distributed data, given as median (25th; 75th percentiles), were compared with the Mann-Whitney U test. Pearson correlation analyses were used to describe the associations between parameters of anthropometry, glycemic control, insulin sensitivity, and lipid availability with hepatic ATP or Pi concentrations in the total study population. AST, ALT, HCL, FFAs, TGs, high-sensitivity C-reactive protein (hsCRP), oxLDL, and M values were not normally distributed and were therefore log transformed before analyses. Partial correlation analysis was performed by adjusting first for diabetes diagnosis and then for study, age, sex, and BMI. Multivariable linear regression analysis unadjusted as well as adjusted for age, sex, and BMI as potential confounders was used to assess interactions between variant allele carrier status and hepatic ATP concentrations. Differences and correlations were considered significant at P < 0.05. All statistical analyses were performed using SAS version 9.4 software (SAS Institute, Cary, NC).

Participants of the whole cohort had comparable age, sex, BMI, fasting lipids, and hsCRP (Table 1). Patients with type 1 diabetes exhibited higher fasting blood glucose and HbA_1c than CON but had near-normal glucometabolic control and short known diabetes duration. Waist circumference was higher and AST was slightly lower in patients with type 1 diabetes. Among the age- and BMI-matched subgroups, whole-body insulin sensitivity was 44% lower in the patients with type 1 diabetes (M value 7.5 mg * kg BW⁻¹ * min⁻¹ [6.3; 10.4] vs. 13.4 mg * kg BW⁻¹ * min⁻¹ [9.1; 16.5] in CON, P = 0.001), whereas basal EGP (bEGP) and iEGP were comparable to CON (Supplementary Table 1). Differences in waist circumference and glycemic control were similar to those of the whole cohort.

Hepatocellular ATP content was 15% lower in patients with type 1 diabetes than in CON (2.27 ± 0.07 vs. 2.66 ± 0.07 mmol/L in CON, P < 0.001), whereas Pi concentrations were comparable (1.94 ± 0.07 vs. 2.03 ± 0.07 mmol/L, P = 0.49) (Fig. 1A and B). The subgroup comparison confirmed the results of lower hepatic ATP concentration (2.30 ± 0.10 vs. 2.72 ± 0.12 mmol/L, P = 0.018) but comparable hepatic Pi and HCL content in type 1 diabetes compared with CON (Supplementary Table 1).

Across the whole cohort, HCL content did not differ between type 1 diabetes and CON (0.48% [0.26; 1.47] vs. 0.48% [0.20; 1.34], P = 0.55) (Fig. 1C). The distribution of the individual HCL values covered a range from <1 to 13% across the groups, with values >5.56% in four patients with type 1 diabetes and two healthy humans.

Patients with type 1 diabetes exhibited higher fetuin A (305 ± 7 vs. 279 ± 7 mg/mL, P = 0.01) (Fig. 1D), whereas FGF21, SepP (Fig. 1E and F), and adiponectin (2,359 ng/mL [1,387; 5,213] vs. 2,681 ng/mL [1,608; 4,052] in CON, P = 0.88) levels were similar to CON. OxLDL was 39% higher in patients with type 1 diabetes (43.9 μg/mL [24.3; 110.0] vs. 31.4 μg/mL [17.1; 49.3], P = 0.038) (Fig. 2A), whereas TBARS, 8OH-dG, and GSH levels (Fig. 2B–D) and protein carbonyl products (21.5 nmol/mL [19.2; 35.3] vs. 22.6 nmol/mL [17.7; 35.3] in CON, P = 0.62) were not different from CON.

In an unadjusted analysis across all participants, hepatic ATP related negatively to HbA_1c, fasting glucose, and oxLDL (Table 2). After adjustments for diabetes status, study, sex, age, and BMI, these correlations were no longer present. There were no associations between hepatic phosphorus metabolites and FFAs, TGs, total cholesterol, or hsCRP. Hepatic Pi correlated negatively with BMI (r = −0.20, P = 0.03), which remained significant after adjustment for diabetes status (data not shown). In the patients with type 1 diabetes, HCL content correlated positively with FGF21 (r = 0.67, P < 0.001), GGT (r = 0.56, P < 0.001), TGs (r = 0.55, P < 0.001), and visceral fat content (r = 0.41, P < 0.01) but negatively with M value (r = −0.52, P < 0.001) and adiponectin (r = −0.63, P < 0.001), which remained significant after adjustment for age, sex, and BMI (data not shown), whereas no association was observed between hepatic ATP and peripheral or hepatic insulin sensitivity (Supplementary Table 2).

Among the patients with type 1 diabetes, carriers of the PPARG type 2 diabetes protective G allele (rs1801282) and noncarriers of PPARGC1A type 2 diabetes high-risk A allele (rs8192678) had 21% (P = 0.018) and 13% lower (P = 0.028) hepatic ATP concentrations compared with the corresponding noncarriers and carriers, respectively (Table 3). Multivariate linear regression analysis showed that the effect of variant allele carrier status of PPARG on hepatic ATP content is different at different carrier status of PGC1A SNP, or alternatively, the effect of variant allele carrier status of PGC1A on hepatic ATP content is different at different carrier status of PPARG (P = 0.03 unadjusted, P = 0.04 adjusted for age, sex, and BMI) (Fig. 3). In carriers of the minor G allele in PPARD (rs2267668), related to reduced response in mitochondrial function to lifestyle intervention (18), no differences in hepatic ATP and Pi concentrations were found. The high-risk A allele in FTO (rs9939609) did not influence hepatic ATP and Pi levels (Table 3). Carriers of the A allele of NDUFB6 gene polymorphism (rs540467) also exhibited no differences in hepatic phosphorous metabolites.

The main finding of this study is that patients with recent-onset type 1 diabetes already exhibit impaired hepatic energy metabolism independent of liver fat content, but in the setting of increased peripheral insulin resistance and circulating fetuin A levels. Absolute concentrations of hepatic ATP may be further modulated by variations in genes known to control mitochondrial biogenesis. Of note, lower hepatic ATP levels in these near-normoglycemic patients are present early in the course of the disease, when hepatic insulin resistance and NAFLD are absent.

Recent data revealed that biopsy-proven NAFLD has similar liver-related adverse outcomes in type 1 diabetes compared with patients with type 2 diabetes (2). However, the risk factors and methods for early detection in type 1 diabetes remained unknown. Decreased hepatic ATP levels have been previously reported not only in obesity (29) but also in alcoholic hepatitis (30) and during postsurgical recovery in liver cirrhosis (31). There is evidence for a metabolic shift from oxidative phosphorylation to glycolysis at an early stage of liver injury, supporting the central role of mitochondrial function in liver adaptation to disease (32). Lower hepatic ATP could result from compromised production, i.e., oxidative phosphorylation, or from increased utilization by energy-consuming processes like lipogenesis. The patients of the current study exhibited low HCL in line with previous studies in humans (33) and comparable circulating lipids, excluding enhanced lipogenesis. Breakdown of ATP would also result in increased Pi levels, suggesting a reduced phosphorylation state (34), as observed in obesity (29), virus-induced cirrhosis (35), and hepatic malignancies (36). In contrast to type 2 diabetes (9), patients with type 1 diabetes had unchanged hepatic Pi concentrations. Of note, ATP content is a major discriminator for the degree of liver injury in NAFLD (37). In the current study, hepatic ATP concentrations were corrected for hepatocellular lipid volume (22) and still remained lower in type 1 diabetes, indicating reduced energy metabolism independent of liver fat content. The absence of any clinical or biochemical signs of liver disease in these patients suggests that impaired switching between energy-demanding and energy-producing processes, but not loss of functional hepatocytes, is responsible for the observed abnormality in hepatic energy status.

Human as well as mouse models showed that glycemia, lipid availability, and oxidative stress are major determinants of hepatic mitochondrial function, with the latter considered at the same time a cause and a consequence of alterations in hepatic energy metabolism. Previous studies investigating hepatic energy metabolism with ³¹P/¹H-MRS have shown reduced hepatic ATP and Pi concentrations as well as lower flux through ATP synthase in type 2 diabetes (9,10). Our finding of decreased hepatic ATP content in type 1 diabetes (Fig. 1) shows that abnormal liver energy homeostasis is also a feature of the early course of type 1 diabetes and could therefore be associated with common hallmarks like hyperglycemia or muscle and hepatic insulin resistance (38). Previous studies demonstrated similar results and correlations of hepatic absolute ATP concentrations and hepatic ATP synthase flux with insulin resistance and liver fat content in patients with type 2 diabetes (9,10). Thus, hepatic absolute ATP content likely reflects ATP synthase flux, further suggesting that the lower ATP concentrations reflect diminished hepatic ATP synthesis in type 1 diabetes. In muscle of nonobese patients with type 1 diabetes, insulin-stimulated flux through muscle ATP synthase is reduced and associates with long-term glycemic control, as measured by HbA_1c (7). In the current study, hepatic ATP associates negatively with fasting glycemia but only before adjustment for diabetes status, age, sex, and BMI, suggesting that other factors might be at play.

Recent studies in mouse models of type 1 diabetes described greater mitochondrial biogenesis by increased transcript levels of PGC1α and TFAM and respiratory chain complex activity as adaptation to the increased lipid and glucose flux at diabetes onset and compensation for mitochondrial energetic deficit due to enhanced gluconeogenesis (11,39). Liver biopsy samples were not available in the current study, rendering direct measurements of mitochondrial function and biogenesis impossible. Depletion of hepatic ATP in humans with type 1 diabetes could result from chronically increased rates of energy-consuming processes, such as increased and unrestrained hepatic gluconeogenesis under conditions of portal hypoinsulinemia (33). Nevertheless, the patients with type 1 diabetes of this study exhibited no increase in fasting EGP, suggesting no relevant increase in gluconeogenic flux. Higher hepatic or systemic lipids could also impair hepatic mitochondrial function by lipotoxicity (40), but again operation of this mechanism is unlikely in the absence of changes in plasma TGs, FFAs, and HCL.

Recent clinical trials in obese humans confirmed the concept of hepatic mitochondrial adaptation and mitochondrial flexibility at early stages of NAFLD but reported declining hepatic respiratory capacity along with augmented oxidative stress in nonalcoholic steatohepatitis (12,13). Higher systemic levels of oxidative stress and lipid peroxidation, along with the reduced antioxidant capacity in type 1 diabetes (41), could contribute to impaired mitochondrial function and reduced hepatic ATP content. The current study found no differences between lipid peroxidation assessed by TBARS, DNA oxidative damage measured by 8-OHdG, and protein carbonyl products or antioxidant defense, as assessed from GSH concentrations, between humans with and without type 1 diabetes. This could be due to the short duration of disease and excellent glucometabolic control of the patients. Interestingly, oxLDL reflecting susceptibility of LDL to oxidative modification was increased in patients with type 1 diabetes of the current study compared with CON, which has previously been linked to reduced plasma antioxidant levels and atherosclerosis in type 1 diabetes (42). Indeed, the finding that hepatic ATP concentrations correlated negatively with oxLDL supports a close relationship between early changes in oxidative stress and impaired hepatic energy metabolism but does not allow conclusions as to causality. Moreover, association with oxLDL was lost after adjustment for status, sex, age, and BMI, suggesting that observed changes are not necessarily related.

In patients with type 2 diabetes, hepatic insulin resistance is an independent predictor explaining lower absolute hepatic ATP and Pi contents (9) and ATP synthesis (10). In the current study, hepatic ATP concentrations did not correlate with any anthropometric or insulin sensitivity measures in patients with type 1 diabetes despite substantially lower peripheral insulin sensitivity and higher waist circumference compared with CON. Of note, no relation between subcutaneous and visceral fat content and hepatic ATP concentrations was detected, implying that fat mass and depot distribution do not play a major role for hepatic energy homeostasis in type 1 diabetes. Moreover, comparable iEGP between the two subgroups and the lack of association with hepatic ATP further suggest no effect of hepatic insulin sensitivity on hepatic energy status at this early stage of type 1 diabetes development. This implies that factors different from fat mass and insulin sensitivity underlie the abnormal hepatic energy metabolism.

A number of circulating adipo-/hepatokines are known to be altered and play a role in insulin resistance and lipid homeostasis in humans with type 1 diabetes (19–21) and rodent models (11). Indeed, we confirmed higher fetuin A concentrations in patients with type 1 diabetes, as previously observed in rodent models of type 1 diabetes (11) and in human type 1 diabetes, where it relates to early markers of atherosclerosis and obesity (20). Concentrations of FGF21, a modulator of mitochondrial oxidative phosphorylation (43), did not differ between patients and CON, in contrast to one previous study, probably due to lower age and markedly worse glycemic control of those patients (21). Also, concentrations of the hepatokine SepP, which may be increased in obesity and NAFLD (44) and interfere with insulin signaling by inhibiting phosphorylation of key mediators in energy metabolism such as protein kinase B and AMP-activated protein kinase (45), were unchanged in patients with type 1 diabetes.

Abnormalities in glucose and energy homeostasis have also been related to SNPs in genes regulating metabolism. The common polymorphism Pro12Ala (rs1801282) of the PPARG gene associates with reduced type 2 diabetes risk through modulation of the production and release of adipose-derived insulin-sensitizing factors (14) and represents an important genetic link between type 2 and type 1 diabetes (46). The PPARγ coactivator-1α regulates oxidative phosphorylation, and the variant serine-encoding allele (Gly482Ser) in the PPARGC1A gene associates with increased obesity and oxidative stress risk (47). Unexpectedly, we found reduced hepatic ATP concentrations in Ala allele carriers (PPARG rs1801282) and in Ser allele noncarriers (PPARGC1A rs8192678). As hepatic energy metabolism is augmented in obesity and insulin resistance (12), our present finding might suggest that in nonobese patients with type 1 diabetes with normal hepatic insulin sensitivity, these type 2 diabetes protective gene variants contribute to the absence of any increase in hepatic ATP concentrations.

Finally, a variant in the NDUFB6 gene, encoding a subunit of complex I of the respiratory chain, associates with increased type 2 diabetes risk, relates to insulin resistance (48), and predicts the response of muscle energy metabolism to exercise training (16,17). However, no difference in hepatic phosphorous metabolites and HCL between carriers of the A allele in NDUFB6 polymorphism rs540467 and noncarriers was found (Table 3).

In conclusion, hepatic ATP concentrations are reduced independently of hepatic lipid content and influenced by variants in genes controlling oxidative metabolism in type 1 diabetes. Our data underline the importance of liver mitochondrial function and energy homeostasis as early and sensitive markers in metabolic liver disease progression.

The GDS Group consists of Michael Roden (speaker), Hadi Al-Hasani, Annette Buyken, Juergen Eckel, Gerd Geerling, Christian Herder, Andrea Icks, Joerg Kotzka, Oliver Kuss, Eckhard Lammert, Jesper Lundbom, Karsten Muessig, Peter Nowotny, Wolfgang Rathmann, Julia Szendroedi, and Dan Ziegler.

Acknowledgments. The authors thank Andrea Nagel, Nicole Achterath, Ulrike Partke, Gabi Gornitzka, Kai Tinnes, and Myrko Eßer (Institute for Clinical Diabetology, German Diabetes Center) for their excellent help with the experiments.

Funding. This work was supported by the Ministry of Innovation, Science and Research of the State of North Rhine-Westphalia and the German Federal Ministry of Health. This study was supported in part by a grant from the Federal Ministry of Education and Research (BMBF) to the German Center for Diabetes Research, by a grant from the Helmholtz Alliance Imaging and Curing Environmental Metabolic Diseases (ICEMED), and by the Schmutzler-Stiftung. Part of this project was conducted in the context of the pretest studies of the German National Cohort (www.nationale-kohorte.de). These were funded by the BMBF (Förderkennzeichen 01ER1001A-I) and supported by the Helmholtz Association as well as by the participating universities and institutes of the Leibniz Association.

The funding sources had no input in the design and conduct of this study, in the collection, analysis, and interpretation of the data, or in the preparation, review, or approval of the article.

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

Author Contributions. S.G. wrote the manuscript and researched and collected data. A.B. contributed to the manuscript and collected data. K.K., C.H., P.N., S.K., G.G., B.Kl., B.Kn., P.B., and J.L. collected data. K.S. and H.A.-H. contributed to the manuscript and data analysis. J.S. and M.R. designed the study, researched data, and contributed to the manuscript. M.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.