Study design and participants
This was an NIH-funded ancillary study conducted in a subset of participants enrolled in the CALERIE 2 (NCT00427193) trial at Pennington Biomedical Research Center. CALERIE 2 was a multicenter randomized clinical trial that tested the efficacy of a 24-month intervention targeting a sustained 25% CR compared to a comparator group that ate AL13,20. The ancillary study protocol for this study was approved by the institutional review boards at Pennington Biomedical Research Center (Baton Rouge, LA) and Columbia University Irving Medical Center (New York, NY). All participants provided written, informed consent for study procedures prior to data collection, which was performed according to relevant guidelines and regulations.
Individuals in CALERIE 2 were healthy with a body mass index (BMI) between 22.0 to less than 28.0 kg/m2 at enrollment. Males were aged between 21 and 50 years and females were aged between 21 and 47 years to minimize the effects of menopause onset during the trial. Participants were randomized with a 2:1 allocation to either the CR or AL group, with randomization stratified by sex and BMI. Details about the intervention have previously been reported13. Briefly, the CR intervention targeted an immediate and sustained 25% restriction of energy intake from baseline energy requirements as determined by doubly labeled water24,25 for the first 12 months followed by 12 months of subsequent weight maintenance. A mathematical model predicting weekly changes in body mass was used to guide adherence to the intervention and CR participants followed an intensive behavioral intervention26,27. For the ancillary study analysis, we only included individuals that were “adherent” given their assigned treatment groups; that is, individuals that showed > 5% decrease in body mass at both follow-up time points if they were in the CR group, and individuals that showed < 5% change in body mass at both time points if they were in the AL group28. This criterion resulted in exclusion of two participants and a resulting sample size of 42 participants.
Assessments
Demographics, anthropometric variables, measures of body composition using dual energy X-ray absorptiometry (DXA), and sleeping energy expenditure in a whole room indirect calorimeter were collected as part of the parent study.
Demographic and anthropometric variables
Demographic variables including age and biological sex were collected at baseline. Height was measured in duplicate at baseline without shoes to the nearest 0.5 cm using a wall-mounted stadiometer. Body mass was measured in duplicate in the morning following an overnight fast at baseline, midpoint (12 months), and end (24 months) of the trial using an electric scale (Scale Tronix 5200; Welch Allyn) with participants wearing a hospital gown and no shoes. The weight of the gown was subtracted to obtain true metabolic body mass. BMI was calculated as body mass divided by meters squared.
Dual-energy X-ray absorptiometry for assessment of body composition
Body composition (i.e., fat mass and fat-free mass) was assessed using DXA at baseline, 12 months, and 24 months (Hologic 4500A, Delphi W or Discovery A scanners). Scans were performed following standardized procedures for subject positioning and scan mode, and all scans were analyzed at a central reading center by blinded analysts.
Magnetic resonance imaging for tissue and organ size
At baseline, midpoint (month 12), and end (month 24) of the trial, whole-body MRI scans were acquired in the Biomedical Imaging Core at Pennington Biomedical using a 3.0 T Sigma Excite system (General Electric, Milwaukee, WI). MRI images were acquired from the tips of the fingers with arms stretched above the head to the bottom of the feet29,30. A 3D gradient echo sequence with a repetition time of 3.5 ms and an echo time of 1.7 ms was used for acquisition of the whole-body MRI scans using a torso phase array coil. The acquisition matrix was 380 by 192, and the slice thickness of the contiguous axial slice was 3.4 mm. For brain imaging, axial contiguous brain MRI scans were acquired using a spin echo sequence with a slice thickness of 5 mm, a repetition time of 3500 ms, and an echo time of 98 ms. Following acquisition, images were segmented for adipose tissue, skeletal muscle, brain, liver, kidneys, and residual lean tissues (i.e., bones, tendons, connective tissue, and the digestive tract) at the Image Analysis Core Laboratory of New York Obesity Nutrition Research Center by a blinded experienced MRI technician using the image analysis software SliceOmatic 5.0 (Tomovision Inc., Montreal, Canada)31. Tissue compartment volume was calculated as previously described32. The intraclass correlation coefficient for volume rendering of brain, liver, kidneys, skeletal muscle, and adipose tissue for the same scan by the same analysts at the Image Analysis Core Laboratory of New York Obesity Nutrition Research Center is 0.95–0.9929,33,34. MRI volume estimates were converted to tissue mass by using the assumed density of 1.04 kg/L for skeletal muscle, liver, and kidneys; 1.03 kg/L for brain; and 0.92 kg/L for adipose tissue29,35,36. The images could not be segmented for heart because the MRI sequences were not cardiac gated so there was no clear separation between blood and cardiac muscle. To approximate heart mass, we used a previously published equation with trunk lean mass derived from DXA23,37: heart mass (kg) = 0.012 * trunk lean mass (kg)1.0499.
Energy expenditure assessments
Energy expenditure was assessed in a whole room indirect calorimeter as previously described25. Participants stayed in the room for 23.25 h and were provided with three meals and one snack at scheduled intervals with the instruction to consume all their food within 30 min. Meals were tailored such that individuals met the caloric target that aligned with their treatment group assignment; that is, participants in the CR group were fed 25% calories less than their baseline energy requirements, whereas participants in the AL group were fed to maintain energy balance28. Sleeping energy expenditure was calculated from VO2 and VCO2 according to the Weir equation38 between 2 and 5 am when motion detectors in the chamber were recording no activity. We specifically decided to use sleeping energy expenditure as a robust measure of resting energy expenditure given its repeatability of over 95% across repeated measures and precision with an estimated measurement error of only ~ 2%39.
Energy expenditure prediction
We derived three different prediction equations for sleeping energy expenditure at baseline. Firstly, we used multiple linear regression with sex, age, and body mass. Secondly, we built a model using sex, age, and fat mass and fat-free mass derived from DXA28. Thirdly, we employed previously established18 and validated19 values for tissue-specific energy expenditure to predict sleeping energy expenditure from sex, age, and adipose tissue, skeletal muscle mass, brain mass, liver mass, kidney mass, heart mass and residual lean mass from MRI. For comparison, and because we did not have a measurement of all organs included in the previously validated equation (i.e., heart mass), we also developed our own model using MRI-measured skeletal muscle, adipose tissue mass, and the combined mass of organs and tissues available (i.e., brain, liver, kidney, spleen, and residual lean tissue), in addition to age and sex.
Metabolic adaptation (MA)
Predicted values for sleeping energy expenditure were generated from each linear regression equation developed at baseline (body mass, DXA and MRI) for month 12 (the midpoint) and month 24 (end of the trial) using the actual body mass and composition values at those time points. The differences in the measured and predicted energy expenditure values (termed “residuals”) while accounting for the baseline error were calculated as a measure of metabolic adaptation (also termed adaptive thermogenesis); i.e., a change in energy expenditure that was not explained by the observed changes in tissue mass.
Statistical analysis
The primary outcome of this analysis was to more precisely quantify MA by considering changes in organ sizes as assessed via MRI in comparison to simpler models including either fat mass and fat-free mass derived from DXA, or body mass alone. All analyses were conducted in R (version 4.3.2). Baseline participant characteristics are presented as means (standard deviation) or N (percent) for continuous or categorical variables, respectively. Changes in outcome variables from baseline were analyzed using a linear mixed model with fixed effects for time and treatment group, the interaction thereof, and a random effect for participant to account for repeated measures. Within- and between-group effect estimates alongside corresponding 95% confidence intervals as derived from the model are reported for these analyses. Residuals (i.e., the difference between measured and predicted energy expenditure values) were derived from the different prediction equations while accounting for the error at baseline. These residual values were compared against zero to determine if statistically significant metabolic adaptation occurred. This was done with linear mixed models with fixed effects for the method of prediction and time as well as the interaction thereof, and a random effect for participant. Additionally, the different prediction methods were modeled as a continuous variable representing the increase in granularity of tissue assessment, and the interaction between this slope and the treatment group were assessed using a linear mixed model. Furthermore, individuals were classified categorically as having metabolic adaptation (or not) if the residual (i.e., the unexplained decrease in SleepEE) was greater than the measurement error of SleepEE, estimated at 2%39. Finally, the magnitude of metabolic adaptation was calculated as a percentage of SleepEE at each respective time point.