Insulin resistance in skeletal muscle is a long-recognized hallmark feature of type 2 diabetes mellitus and obesity. Although type 2 diabetes generally only occurs when pancreatic beta cells are incapable of secreting enough insulin to maintain normoglycemia, this most often occurs in the setting of insulin resistance (DeFronzo et al., 2015). Numerous investigators using in vivo approaches in humans, rodent models, or in vitro experiments have attempted to define the molecular nature of the changes responsible for insulin resistance in skeletal muscle. Studies too numerous to cite completely have described molecular defects in insulin as action and fuel selection (Kahn and Cushman, 1985; Mandarino et al., 1986; Kelley and Mandarino, 1990; Cusi et al., 2000; Kelley and Mandarino, 2000; Patti et al., 2003). Despite learning a great deal of molecular physiology from these studies, this approach usually does not consider the overall physiological context of the skeletal muscle in which such defects may be found.

A different approach to understanding insulin action in skeletal muscle is to start from first principles, in this case from the observation that the energy state of resting and mildly exercising skeletal muscle is held far from equilibrium by the action of creatine kinase. Nonequilibrium thermodynamic theory arises from properties of thermodynamic equilibrium and can explain the mechanism and consequences of the extent to which skeletal muscle is able to maintain a high energy state when energy demand increases. In skeletal muscle, this energy state can be estimated as the energy made available by ATP hydrolysis, often termed ΔG_ATP. Mitochondria maintain this energy at rest and defend it from falling too steeply under conditions of energy demand, such as muscle contraction or other energy consuming processes, which might include insulin action. Insulin at physiologic concentrations measurably increases oxygen consumption in leg muscle, demonstrating an increase in energy demand (Kelley and Mandarino, 1990; Kelley et al., 1992; Mandarino et al., 1993; Mandarino et al., 1996).

It is well known that a higher ATP energy state is linearly related to power output of the contractile apparatus (Jeneson et al., 1995; Westerhoff et al., 1995). This relationship would apply not only to muscle contraction but also to the many processes that require movement within a cell, and we recently proposed that this principle could apply to transport processes like insulin-stimulated, kinesin-powered movement of GLUT4 vesicles along microtubules (Mandarino and Willis, 2023) due to similarities between kinesins and myosins (Vale and Milligan, 2000). Insulin sensitivity, therefore, could be a function of the ability of the cell to maintain the energy state of ATP, which depends mitochondrial functional capacity (Paganini et al., 1997; Glancy et al., 2008).

Thermodynamic variables can be measured in vivo in human skeletal muscle using calibrated ³¹P-MRS at rest or during recovery from muscle contraction (Kemp et al., 2007; Kemp et al., 2015; Sleigh et al., 2016). ³¹P-MRS allows estimation of the free energy of ATP noninvasively and continuously at rest, during exercise, and during recovery from exercise. ³¹P-MRS also can be used to estimate the rate of ATP synthesis (J_ATP) during recovery from exercise by monitoring the recovery of the concentration of phosphocreatine (Kemp et al., 2007; Kemp et al., 2015; Sleigh et al., 2016). PCr recovery from exercise follows a monoexponential time course, allowing estimation of the time constant parameter τ (Kemp et al., 2007; Kemp et al., 2015; Sleigh et al., 2016). However, ³¹P-MRS does not provide concentrations of energy phosphates unless spectra are calibrated to a standard, and unphosphorylated creatine is not visible. Because absolute concentrations of energy phosphate compounds such as PCr, ATP, ADP, as well as phosphate (Pi) are required to estimate thermodynamic variables, muscle biopsies at rest can be performed to assay [ATP], [PCr], and [Pi], allowing calculation of [ADP], the concentration of which is too low to be observed with MRS. Total creatine also can be assayed in muscle biopsies, providing all the absolute concentrations needed to calculate the energy of ATP (ΔG_ATP) and the rate of ATP synthesis (J_ATP). In skeletal muscle, mitochondria are responsible for most ATP production and maintain a high energy state of ATP. Increased ATP demand and the increased rate of ATP hydrolysis cause ΔG_ATP to fall in a nearly linear manner (Jeneson et al., 1995; Westerhoff et al., 1995). In response, J_ATP rises. The slope of this relationship depends on the cellular content of functional mitochondria (Dudley et al., 1987). Higher mitochondrial functional content enables better maintenance of ΔG_ATP (Paganini et al., 1997; Glancy et al., 2008).

A linear electrical circuit analog of the nearly linear kinetics of the relationship between the rate of oxidative phosphorylation and the energy from ATP has been used for over 30 years to simplify nonequilibrium thermodynamic considerations in skeletal muscle during non-steady state conditions when ΔG_ATP is changing (Meyer, 1989). There is theoretical and experimental evidence indicating that this model predicts linear relationships between 1) ΔG_ATP and J_ATP (the “force-flow” relationship), and 2) between the slope of the force-flow relationship and the functional mitochondrial content or “conductance” of the system. Establishment of these linear relationships is a key to validating this model in the context of the questions posed here. Conductance closely reflects mitochondrial functional content (Paganini et al., 1997; Glancy et al., 2008), while capacitance is proportional to the total creatine concentration, and affects the time course of changes during non-steady state conditions (Meyer, 1989). Finally, τ, the time constant of recovery of phosphocreatine after exercise, is the third variable needed to completely determine this model linear system. All of these parameters were estimated in the present study using a combination of ³¹P-MRS and biochemical assays in muscle biopsies. So, these studies can provide complete and detailed information about skeletal muscle nonequilibrium thermodynamics and the functional content of mitochondria needed to maintain a high energy state of the cell, which will drive energy-consuming processes more robustly (Jeneson et al., 1995). The reader interested in a more detailed theoretical treatment of calculations of thermodynamic parameters and derivation of the non-steady state treatment of nonequilibrium thermodynamics, including the meaning of the theoretical constructs of conductance and capacitance can refer to the Supplementary Material (Davis and Davis-Van Thienen, 1989; De Saedeleer and Marechal, 1984; Glancy et al., 2013; Golding et al., 1995; Harris et al., 1992; Hasselbach and Oetliker, 1983; Jeneson et al., 2009; Kushmerick, 1998; Masuda et al., 1990; Nicholls and Ferguson, 2013; Pate et al., 1998; Rottenberg, 1973; Walter et al., 1997; Westerhoff et al., 1981; Westerhoff and Van Damm, 1987; Willis et al., 2016).

We refer to mitochondrial protein abundance, determined using proteomics, as mitochondrial protein content, a biochemical measure of the abundance of all mitochondrial proteins, to distinguish it from mitochondrial functional content, which can differ from mitochondrial protein content to the extent that molecular changes in mitochondria may either increase or decrease the ability of a given mass of mitochondria to influence the rate of oxidative phosphorylation. The functional advantages of both higher mitochondrial protein and mitochondrial functional content can be seen in muscle adapted to endurance exercise, where there is a better ability to maintain the energy of ATP (Paganini et al., 1997), fuel selection that favors lipid over carbohydrate oxidation (Holloszy and Coyle, 1984; Helge et al., 2007), and better fatigue resistance (Fitts et al., 1975).

Many processes activated by insulin, such as GLUT4 translocation, require energy from ATP hydrolysis and therefore would have rates that are dependent on the energy of ATP. Compared to our knowledge of muscle contraction, we know much less about the impact of skeletal muscle nonequilibrium thermodynamics and mitochondrial functional content on skeletal muscle metabolism at rest or during insulin stimulation, especially in the setting of insulin resistance. Therefore, it is important to consider whether functional mitochondrial content and the capacity to maintain a higher ATP energy influences the rates of insulin-dependent processes, as recently conjectured (Mandarino and Willis, 2023). The primary aim of this investigation therefore was to use ³¹P-MRS, muscle biopsies, euglycemic clamps, and exercise tests in the context of the linear model to explore whether and to what extent nonequilibrium thermodynamic parameters and skeletal muscle mitochondrial functional content are related to insulin sensitivity and fuel selection (Mandarino and Willis, 2023). Finally, we measured VO_2max because of its well-described positive association with insulin sensitivity that likely is independent of mitochondrial content and more dependent on the cardiorespiratory system and delivery of oxygen to muscle and the capacity for muscle blood flow (Lundby et al., 2017). Linear models were used to determine which of these variables independently predict insulin sensitivity determined using a euglycemic clamp or fuel selection during mild exercise (Barakati et al., 2022).

Because skeletal muscle insulin resistance often precedes the development of beta cell failure and type 2 diabetes mellitus, the underlying causes of insulin resistance have been the subject of intense interest. Many investigators have used a molecular approach to try to discover the biochemical and molecular factors that lead to insulin resistance. Such studies, many of which have been performed in humans, have provided valuable information about differences in insulin signal transduction, regulation of glucose transport, hexokinase and glucose uptake, glycogen synthesis, mitochondrial function, and gene expression and protein abundance measures (Kahn and Cushman, 1985; Mandarino et al., 1990; Mandarino et al., 1995; Shulman, 1999; Cusi et al., 2000; Kelley and Mandarino, 2000; Petersen and Shulman, 2002; Patti et al., 2003). Many other studies have used genetic manipulation of rodents, usually at a single gene level, to create models of altered insulin action (Wasserman and Ayala, 2005). All of these studies are predicated on the idea that by understanding the molecular details of a system, properties of the whole system can be deduced. However, there are alternate ways of understanding a system that start with fundamental principles of physiology, such as the laws of thermodynamics, and attempt to describe the system based on these principles rather than, or in accompaniment with, a description of biochemical or molecular processes. Both molecular and physiological approaches have value and should be interpreted together, yet to our knowledge, nearly all studies of the origins of skeletal muscle insulin resistance mainly have used a molecular or biochemical approach. In the present study, we asked the question of whether basic thermodynamic principles (nonequilibrium thermodynamics) can be used to explain variance in insulin sensitivity and changes in skeletal muscle fuel preference. To accomplish this, we used calibrated ³¹P-MRS to assess thermodynamic variables in skeletal muscle at rest and during recovery from mild exercise coupled with euglycemic, hyperinsulinemic clamps to assess insulin sensitivity and graded exercise tests to determine fuel selection in working muscle. The participants in the present study purposely were selected to provide a wide range of body composition, insulin action and exercise characteristics. We chose this approach in order to provide the best opportunity to determine relationships between these characteristics and the thermodynamic and fuel selection variables that we measured. Although selection of discrete groups of participants is commonly used in such studies, often the selection of discrete groups can mask the fact that these are continuous variables and participants fall along a continuous spectrum. Our approach in fact resulted in the revelation of novel relationships.

We used “calibrated” ³¹P-MRS to accomplish the purposes of this study (Kemp et al., 2007). Concentrations of energy phosphates that we calibrated to resting [ATP] assayed biochemically in skeletal muscle biopsies for each participant corresponded well to values previously reported using biopsies or ³¹P-MRS (Kemp et al., 2007; Lanza et al., 2011; Ripley et al., 2018). This also was the case for total creatine assayed in biopsies. Both resting [ATP] and total creatine were slightly lower than the often used “standard” assumptions of 8.2 mM [ATP] and 42.5 mM TCr in resting human muscle (Kemp et al., 2007; Kemp et al., 2015; Meyerspeer et al., 2020). What was most important regarding these findings, though, was that resting [ATP] varied from 5.33–8.2 mM and TCr ranged from 24.1–49.7 mM. Without calibration, this high level of variability would have had considerable impact on the calculation of individual values for energy phosphates, total creatine, J_ATP, and ΔG_ATP. Errors produced in these variables by using standard assumptions or other constant values likely would have led to an inability to determine whether thermodynamic considerations were related to insulin sensitivity or fuel selection. The importance of using calibrated values has been emphasized by Kemp and coworkers (2007, 2015) and more recently has been explored by Ripley and colleagues, who used a phosphate phantom and the water signal to calibrate energy phosphates and total creatine, respectively (Ripley et al., 2018).

We used a weighted bag system (Sleigh et al., 2016) to drive PCr down by 15%–25% and followed recovery of [PCr] to determine tau and estimate the rate of oxidative phosphorylation (J_ATP). Along with the corresponding energy state of the cell (ΔG_ATP). During the two exercise periods [ATP] remained nearly constant while [PCR] fell and [Pi] rose in a nearly equimolar fashion. The time course of [PCr] recovery, which is a function of the rate of oxidative phosphorylation (J_ATP), is characterized by the time constant τ of a monoexponential function (Meyerspeer et al., 2020). The value of τ was around 30 s in the current study, similar to what has been reported (Conley et al., 2000; Schrauwen-Hinderling et al., 2007a; Lanza et al., 2011; Cuthbertson et al., 2014), and was reproducible, with a coefficient of variation of around 13%. Since τ is estimated independently of calibrated values for energy phosphates or creatine, it is a commonly measured variable and provides some insight into muscle oxidative characteristics. However, in the absence of independent estimates of mitochondrial conductance or muscle capacitance (TCr), the value of τ potentially is misleading (see Supplementary Material for a more detailed discussion of this point). Thus, calibration of MRS signals is essential to assess the role nonequilibrium thermodynamics might play in insulin sensitivity or fuel selection.

To test the hypotheses regarding the relationship between the thermodynamic properties of muscle and insulin sensitivity or fuel selection, it is critical that our data conform to the linear model of nonequilibrium thermodynamics (Meyer, 1989). We showed that both the ΔG_ATP:J_ATP relationship and the relationship between the slope of that relationship and apparent conductance of the linear model (TCr/τ) were highly linear. In addition, the relationship between TCr measured in biopsies and apparent capacitance of the model also was linear and supports the concept that TCr is proportional to the capacitance of the electrical circuit analog model under non-steady state conditions of ΔG_ATP, such as during recovery from exercise.

We have conjectured that, on a purely theoretical basis, the ability of muscle to maintain a high ΔG_ATP in the face of a rise in energy demand could positively influence the rates of insulin-stimulated energy-consuming reactions (Mandarino and Willis, 2023). Supporting this notion, a rise in plasma insulin concentration increases oxygen consumption across the leg under resting conditions (Kelley et al., 1990; Kelley and Mandarino, 1990; Kelley et al., 1992; Kelley et al., 1993; Mandarino et al., 1996), indicating that insulin action induces a mild increase in energy demand. If lower mitochondrial content requires a lower ΔG_ATP to meet energy demand, energy demand might be met at the expense of slower rates of insulin-stimulated reactions (Mandarino and Willis, 2023). For example, force production and power output by contractile proteins decline with ΔG_ATP (Jeneson et al., 1995; Westerhoff et al., 1995). To the extent that motor proteins may share similar characteristics with proteins like kinesins (Vale and Milligan, 2000), we speculated that the rate of transit of GLUT4 vesicles along cytoskeletal tracks could be sensitive to ΔG_ATP. We used a simple linear nonequilibrium thermodynamic model to confirm that high conductance in the mitochondrial oxidative pathway predicts higher insulin sensitivity. This could indicate that higher mitochondrial functional content predicts insulin sensitivity. This finding is consistent with the concept that a higher energetic state of muscle may drive one or more insulin-stimulated energy-requiring processes at a faster rate, leading to higher insulin-stimulated glucose disposal during a euglycemic clamp. The current results do not provide direct proof of this idea, but the results provide a compelling argument to pursue additional exploration of this hypothesis. These findings appear to be related to the results of Koch and Britton (2022) that rats selected for running capability are healthier, leaner, more insulin sensitive, and long-lived than rats selected for poor running ability.

The second purpose of this study was to determine if the ability of mitochondria to maintain ΔG_ATP in skeletal muscle dictates fuel selection during mild exercise. A lower ΔG_ATP would be predicted to lead to higher carbohydrate oxidation at the expense of fat oxidation through elevations in [ADP], [Pi], and potentially [AMP] (Connett, 1989; Stanley and Connett, 1991). Lower ΔG_ATP during mild exercise would follow from a lower content of functional mitochondria that lessens the ability of muscle to defend ΔG_ATP during exercise or potentially any increase in energy demand (see Supplementary Material). The present results show that mitochondrial protein content is the best linear predictor of the fraction of fuel accounted for by lipid (r = 0.74). This is not direct evidence in favor of this mechanism but provides an argument for performing more mechanistic studies in the future to test this hypothesis.

A number of previous studies used ³¹P-MRS in vivo in humans to determine the relationship between oxidative phosphorylation defects and insulin sensitivity (Conley et al., 2000; Newcomer et al., 2001; Petersen et al., 2003; Petersen et al., 2004; Schrauwen-Hinderling et al., 2007a; Schrauwen-Hinderling et al., 2007b; Szendroedi et al., 2007; Lanza et al., 2011; Ripley et al., 2018; Toledo et al., 2018). Several studies used the saturation transfer technique (Pi →ATP) to estimate rates of oxidative phosphorylation (Petersen et al., 2003; Petersen et al., 2004; Szendroedi et al., 2007; Kacerovsky-Bielesz et al., 2009). However, this technique does not accurately measure oxidative phosphorylation (Balaban and Koretsky, 2011; From and Ugurbil, 2011; Kemp and Brindle, 2012), leaving those findings open to reinterpretation. Other studies of this topic employing ³¹P-MRS techniques used assumptions, rather than measurements, of the concentrations of ATP and total creatine (TCr) in muscle (Schrauwen-Hinderling et al., 2007a; Schrauwen-Hinderling et al., 2007b). Although these studies, mainly examining recovery of phosphocreatine after exercise, have revealed interesting information regarding the relationships between oxidative phosphorylation and insulin sensitivity, the lack of calibration of energy phosphates or creatine concentrations prohibits their use in determining a role for nonequilibrium thermodynamics in insulin action or fuel selection. Two other studies have used calibrated MRS to understand oxidative phosphorylation changes in aging (Conley et al., 2000) and type 2 diabetes (Ripley et al., 2018). Ripley et al. (2018) used ³¹P-MRS calibrated with a phosphate phantom for energy phosphate concentrations, with quantification of creatine using the water signal, to calculate [ADP] in resting skeletal muscle from insulin resistant obese and type 2 diabetic patients finding no difference in [ADP] between obese and type 2 diabetic volunteers. Rather, these investigators found lower [PCr] to be correlated with several factors including mitochondrial density determined from biopsies. We confirmed this inverse correlation between [PCr] mitochondrial protein content determined using proteomics (r = −0.42, p < 0.10). The calibrated [ATP] in that study (6–7 mM) was similar to our biopsy-determined value of about 6.8 mM, somewhat below a commonly used “standard” assumption of 8.2 mM. The reasons for these differences in [ATP] are unclear. Unfortunately, neither Conley et al. (2000) nor Ripley et al. (2018) put their findings into the context of nonequilibrium thermodynamics. Nevertheless, the results of Ripley et al. (2018) provide the valuable demonstration that to study mitochondrial function and content in vivo in human muscle requires validly calibrated values for concentrations of energy phosphates and total creatine. A unique aspect of the present study is that we used calibrated ³¹P-MRS to study the relationship of metabolic flows to thermodynamic forces and found that functional mitochondrial content and efficient oxidative energy transfer predict insulin sensitivity.

We also used a new label-free proteomics approach to calculate an index of mitochondrial protein content using lysates of whole muscle protein. This simple index was taken to be the sum of all individual abundance levels associated with proteins assigned to mitochondria. Nearly 500 mitochondrial proteins out of a total of over 4000 total proteins were quantified in this analysis. The use of hundreds rather than a single mitochondrial protein as an index of mitochondrial protein content should provide a more accurate measurement of mitochondrial content. However, mitochondrial functional content, measured as conductance in the electrical circuit linear model of nonequilibrium thermodynamics may be a better indicator of mitochondrial function than mere mitochondrial protein content with respect to insulin sensitivity. Even so, mitochondrial protein content was a good predictor of fat oxidation during mild exercise and citrate synthase activity was related to the apparent conductance of the model we used.

In conclusion, the results of the present study show that insulin resistance and a preference for carbohydrate oxidation in skeletal muscle can be explained by a lower free energy state that likely results from lower mitochondrial functional and protein content, as recently conjectured (Mandarino and Willis, 2023). The latter, in turn, may be a consequence of lower physical activity, leading to reduced mitochondrial functional content, although this study does not directly address the question of chronic physical activity. A number of studies have used various aspects of ³¹P-MRS to assess mitochondrial function in healthy and insulin resistant humans, but to our knowledge, this is the first study to use calibrated ³¹P-MRS to assess the energy state of muscle at rest and during mild exercise in the context of insulin resistance in the context of a model of skeletal muscle thermodynamics. These results show that strategies to raise skeletal muscle mitochondrial functional content, whether by increased physical activity, new pharmacological approaches, or even mitochondrial transplant (Guariento et al., 2021) likely would improve both insulin sensitivity and lipid oxidation and potentially reduce the risk of type 2 diabetes and cardiovascular disease.