Quantification, Variability, and Reproducibility of Basal Skeletal Muscle Glucose Uptake in Healthy Humans Using ¹⁸F-FDG PET/CT

DISCUSSION

The main goal of this study was to assess test–retest characteristics of SMGU under basal conditions and variability within and between subjects. Furthermore, the validity of a simplified approach without blood input function (i.e., SUV) was compared with advanced kinetic modeling strategies.

SUV measurements are widely used because of the method’s simplicity, but many well-known factors might affect the accuracy of SUV measurements, including patient weight, glucose and insulin levels, and uptake period. Image-derived input function with venous blood sampling requires recalibration of venous samples into arterial input (for ¹⁸F-FDG, venous and arterial concentrations become similar after 20 min) and blood region of interest, which is free from spill-in and spill-out (17,18). This method is successful when the descending aorta is used as a source of image-derived input function though it is complicated in the case of gluteal/quadriceps muscle glucose utilization because only small vessels are within the field of view (19). In the case of arterial plasma image-derived input function, kinetic modeling with the 2T3K model is usually used. However, the more complex 3T5K model, which is more physiologic (it does not assume an instant cell membrane transport), can also be applied.

Previous investigations of SMGU reported a TRT of 17% for SUVs, but subjects were scanned with a time interval of about 9 mo (20). Therefore, it was difficult to attribute this difference to either the accuracy of the measurement itself or the change in glucose metabolism over time or to both factors. In our study, interval between both scans was only 24 h, and metabolic conditions (diet, physical activity, etc.) were controlled as much as possible to rule out interference resulting in a TRT of less than 10%. On average, test–retest SUV measurements of liver and skeletal muscle were reproducible and remained within 5%, whereas difference for any given subject remained within 10%. These results are concordant with reports in the literature underscoring that SUV can serve as an indicative physiologic parameter in those tissues. Such a tight coefficient of variation allows a study with smaller sample size to be performed, a key consideration especially in early clinical studies to evaluate relatively small changes in treatment effect. As an example, a 7% (or 9%) increase in glucose uptake in skeletal muscle could be detected with 80% power, 1-side α = 0.05, with 8 (or 6, respectively) subjects in a placebo-controlled crossover design. In contrast, myocardial SUV measurements under standardized conditions showed both a high intersubject and a high intrasubject variability.

The advantage of using the more complex 3-tissue model is based on a more reliable separation of 3 compartments, which enables studying physiologic processes that occur between individual compartments. In the case of interventions (e.g., pharmacologic agents, exercise), ¹⁸F-FDG uptake might change and the 3-tissue model may provide insight into the underlying mechanism. For example, increased ¹⁸F-FDG uptake may be due to increased cellular permeability (k₃), higher rate of metabolization of ¹⁸F-FDG inside the cell (k₅), or higher blood flow (K₁). All these changes relate to K_i, and the driver of the effect cannot be deducted from SUVs only.

This complex 3T5K model has been evaluated to assess SMGU in patients with obesity or type 2 diabetes mellitus (21,22). Regardless of the high number of fit parameters, a 3T5K model provided good fits and glucose uptake rates (Ki) that were significantly different between patient groups. In addition, the effect of weight-loss on Ki was assessed and compared between groups (23). Our data are in agreement with previously reported kinetic parameters for healthy volunteers; however, TRT was unknown (21).

The 3T5K model was preferred based on spectral analysis as well as by Akaike information criterion in our and previous studies in most but not all subjects (15). In the study by Bertoldo et al., the 3-tissue model lost its superiority with insulin intervention, which affected mainly k₃ (cell-membrane influx) as expected but also k₅ (15). In the case of a significantly lower or negligible k₅ than k₃, an extra complication of the model is not justified. However, under normal conditions with comparable values for k₃ and k₅ (though k₃ is still bigger than k₅) a 3-tissue model is preferred over more simplified kinetic models.

Nevertheless, the lack of standardization in patient preparation of previous reported studies led to increased variability in modeling results (15,21–23). For example, Williams et al. and Bertoldo et al. reported a variation of K₁ in healthy volunteers from 0.043 ± 0.013 to 0.019 ± 0.002 mL/min, respectively, using the same 3T5KS model (21). In our study, average K₁ values for gluteal and quadriceps muscle were 0.050 ± 0.018 and 0.052 ± 0.015, respectively, similar to the values reported by Williams et al. (15,21).

Test–retest measurements for individual modeling parameters (Fig. 4) demonstrated that not only is Ki highly reproducible but also all compartments and other kinetic parameters for subjects 2, 3, 5, and 6. We observed more discrepancy in kinetic parameters for subjects 1 and 4 even though only small changes in SUVs were measured. For subjects 1 and 4, a 25% increase and 12% decrease in Ki, respectively, were observed for test–retest values for gluteal muscles whereas K₁/k₂ ratios did not change much, indicating that changes are mainly occurring in intracellular and bound compartments. For subject 1, a decreased concentration of ¹⁸F-FDG within cells was compensated by increased phosphorylation and the opposite is true for subject 4. These findings can be attributed either to a change in muscle metabolism between two scans or to a modeling bias, but distinguishing between both is difficult because fits used have the best Akaike information criterion scores. The fact that final glucose uptake did not change much is an argument in favor of modeling bias, but on the other hand deviation of some kinetic parameters could be compensated by the other mechanism to maintain glucose consumption in muscle tissue.

The 2T4K model had fewer parameters to fit than a 3-tissue model and therefore may become useful in the case the quality of the scan does not allow a reliable fit to the 3T5K model (e.g., noisy data). Our results showed a good correlation between Vt values from the 2T4K model and Ki (Fig. 6A). So this model can be used for glucose uptake estimations though interpretation of kinetic parameters remains uncertain.

A classic ¹⁸F-FDG 2-tissue-irreversible model provided a poor fit for the measured time–activity curves. Regardless of good test–retest reproducibility, Ki values were overestimated compared with the 3-tissue model (Supplemental Table 1). From these results, we concluded that a 2T3K model is less recommended for quantifying SMGU.

We also evaluated the correlation between SUVs, remaining ¹⁸F-FDG in blood, and different kinetic parameters. As mentioned before, SUV is a semiquantitative parameter that ignores the actual concentration of tracer in blood and assumes that tracer delivery to a target organ is similar for a particular injected activity. However, that muscle uptake itself can affect tracer concentration in blood is one of the key elements in washout. Figure 5D showed a negative correlation between ¹⁸F-FDG in blood and SUV of gluteal muscle meaning that SUVs are smoothed, high glucose metabolism clears blood faster, and SUVs are underestimated. The latter is also illustrated by the dynamic range of measured SUVs (0.52–0.75) versus Ki from the 3T5KS model (0.0011–0.0022). As a result, muscle uptake regulates ¹⁸F-FDG levels in blood and therefore it is not unexpected that a correlation between Ki and SUV is not pronounced (Fig. 5B).

We questioned whether it was possible to improve the quantification of glucose uptake without arterial blood sampling and to find a parameter that is more adequate than SUV. On the basis of our data, we suggest the use of Vts of almost true equilibrium, which is observed 60 min after tracer administration. A good correlation between Ki and the measured volume-of-interest activity (which is almost stable between 60 and 90 min) divided by blood activity was demonstrated (Fig. 6B), and this approach requires only 1 blood sample. In practice, this blood sample can be venous because at that time (60 min after injection) tracer concentration in the artery and vein is almost the same. This approach is less invasive and can be recommended for clinical use.

Another possibility, which does not require blood sampling but an additional static PET scan, is based on the observation that tracer activity in the liver is linearly correlated to blood activity (Fig. 6A).

Both approaches provide Vt values that correlate better to Ki (R² = 0.64 and 0.71, respectively, vs. R² = 0.47 for SUV) and are therefore preferred over SUV measurements, with the minor complication of taking at most 1 additional venous blood sample.

Finally, SMGU values are similar to the published results in a healthy population; however, we cannot directly compare our values, which are about 5–10 times lower, to those reported under hyperinsulinemic euglycemic clamp (7,24,25).