1 INTRODUCTION

The regional growth of adipose tissue (AT) is a major risk factor for cardiometabolic diseases such as hypertension and type 2 diabetes. Simultaneous comparisons of the effects of upper- and lower-body compartments with type 2 diabetes indicate harmful effects of upper-body AT accumulation and protective effects of lower-body AT accumulation.1 Imaging methods such as computed tomography (CT) can identify different ATs and muscle groups within scanned regions, allowing for assessment of tissue-specific associations with cardiometabolic disease. Upper body abdominal subcutaneous (ASAT) and visceral (VAT) ATs are both associated with increased risk of hypertension and diabetes, with VAT associations tending to be stronger.2-6 Within the thigh, results are more mixed. Thigh subcutaneous AT (TSAT) is not significantly associated with hypertension,5, 6 while TSAT and muscle appear to be mostly protective against type 2 diabetes and related biomarkers.7-12 The impact of thigh muscle may even have opposing associations seemingly due to effect modification by obesity status.9, 13 Lower body intermuscular AT (IMAT) is generally positively associated with type 2 diabetes status7, 8, 10, 14-17 and with hypertension.5

It is important to note that the above studies may have lacked CT scans of either the abdomen or the thigh, or if both were present, that final models may not include all measured abdominal and thigh tissues. The exclusion of metabolically relevant tissues can lead to an incomplete picture of the associations of regional tissue accumulation with health. Additionally, individuals with African ancestry are underrepresented in studies including both abdominal and thigh CT scans. African ancestry individuals have different AT distributions (greater ASAT18-20 and IMAT,21 lower VAT18-20) when compared to European ancestry counterparts. There are also racial/ethnic differences in the contributions of specific tissues to cardiometabolic health, with ASAT having greater importance compared to VAT in African ancestry populations.22, 23

A further limitation of previous studies is in the analytic treatment of body imaging data. Compositional data are defined as components which sum to a whole, and as such, they have an inherent correlation structure so that in order to hold size constant, an increase in one component must come at the expense of at least one other component.24 Therefore, ignoring this inherent correlational structure can result in biased and misleading estimates. Compositional data are more appropriately modeled after application of a compositional data analysis (CoDA) transformation, which effectively removes this correlation structure.24

Body composition CT data are compositional data, as they investigate components (various tissues) which sum up to a whole (the total scanned area). However, while health sciences fields such as high-throughput sequencing,25 physical activity,26 and diet27 are applying compositional approaches to their data, few studies28, 29 have applied compositional methods to body composition data. This analysis uses a cohort of African Caribbean men that has both abdominal and thigh CT scans. A CoDA approach was used to compare the simultaneous associations of abdominal (ASAT and VAT) and thigh (TSAT, thigh IMAT, and thigh muscle) body tissues with hypertension and type 2 diabetes. The hypothesis was that ASAT, VAT, and thigh IMAT would be positively associated, while TSAT and thigh muscle would be negatively associated, with hypertension and type 2 diabetes.

2 METHODS 2.1 Study population

All men in this analysis were from the Tobago Health Study, which has been previously described.30 Briefly, the Tobago Health Study is a population-based, prospective cohort study of community-dwelling men aged 40 years and older, residing on the Caribbean island of Tobago, Trinidad and Tobago. Men from Tobago are of homogeneous African ancestry with low European admixture (<6%).31 Participants in the Tobago Health Study were recruited without regard to health status and men were eligible if they were ambulatory and not terminally ill. The baseline visit occurred from 2004 to 2007 and recruited 2482 men; of these, a random subset (N = 1725) attended the first follow-up visit from 2010 to 2014. Men used in the current analysis attended an ancillary study visit from 2014 to 2018, when a convenience sub-sample of N = 768 participants from the prior visit had CT scans of the abdomen and mid-thigh for ectopic AT assessment. Exclusion from the current analysis included non-African Caribbean ethnicity by self-report (N = 67), missing CT scans in the abdomen or in one or both thighs (N = 31), missing covariate data (N = 53), being underweight (N = 4), and non-fasting serum samples (N = 1). Two individuals were also excluded for improper serum handling that led to glucose degradation. The final analytical sample included 610 individuals. Written informed consent was obtained from each participant using forms and procedures approved by the University of Pittsburgh Institutional Review Board, the U.S. Surgeon General's Human Use Review Board, and the Tobago Division of Health and Social Services Institutional Review Board. This study was completed in accordance with the Declaration of Helsinki.

2.2 Computed tomography scans

CT scans were performed at the Calder Hall Medical Clinic, Tobago. Abdominal and thigh volumes were assessed on 3 mm thick slices and 500 mm display field of view from scans acquired using a GE dual slice, high-speed NX/I CT scanner (GE Medical Systems) with 120 KVp, 250 mA, 0.7 s gantry speed, and pitch of 1.5:1. For participants with body weight greater than 200 lbs, the mA was increased to 300. CT contrast was not used. Only one CT scanner was used, and a single individual collected the scans for all participants. Scans were electronically transmitted to the central CT reading center at Vanderbilt University Medical Center (VUMC) where image analysis and quality control were performed.

Image analysis was performed using a semi-automated method. Briefly, images were analyzed using a dedicated imaging processing workstation with custom-programmed subroutines (OsiriX, Pixmeo) and a dedicated pen computing display (Cintiq, Wacom Technology Corporation). A radiologist-trained analyst manually traced anatomical boundaries (skin, muscular fascia, muscle, bone, and peritoneum) in CT scans. Tissue attenuation thresholds of −190 to −30 Hounsfield Units (HU) were used to distinguish AT voxels in these defined regions and tissue attenuations of −29 to 160 HU were used to distinguish lean muscle voxels. For each tissue, the volume (mm3) was calculated.

Abdominal VAT and ASAT were measured from CT scans of three contiguous slices of 3 mm thickness centered at L4-L5. A lateral scout image was used to determine the z-axis location of the L4-L5 intervertebral space and that location and the slice immediately above and the slice immediately below were used to reconstruct a 9-mm-thick single block of images. VAT was defined as AT located within the peritoneal cavity; ASAT was defined as AT located beneath the skin and superficial to the abdominal muscular fascia. The remaining non-VAT and non-ASAT tissues (e.g., organs, bone, abdominal muscle, abdominal IMAT) were not separately measured at the L4-L5 intervertebral space, and so these remaining tissues were combined to form a third “Other abdominal tissue” group.

TSAT, thigh IMAT, thigh muscle, and thigh bone volumes were measured from CT scans of 10 contiguous slices of 3 mm thickness at the mid-thigh level in both legs. An anterior-posterior scout scan of the entire femur was used to localize the mid-thigh position, and that location and the four slices immediately above and five slices immediately below were used to reconstruct a 30-mm-thick single block of images. Hand-drawn boundaries were traced at the medulla, cortex, thigh muscles, fascia, and skin in three of the 10 slices; boundaries were imputed over the remaining slices and verified for accuracy by the trained analyst. Bone volume was identified as the cortical volume. Lean muscle volume was defined as the sum of the adductors, hamstrings, and quadriceps muscles across both thighs. TSAT was defined as AT located between the skin and the muscle fascia, and IMAT was defined as AT located within thigh muscle groups.

2.3 Generation of compositions and additive log ratio transformation

Two separate compositions were created: abdominal and thigh. The abdominal composition was comprised of VAT, ASAT, and the ‘Other’ remaining abdominal tissues. Similarly, thigh composition was comprised of TSAT, IMAT, muscle, and bone.

The additive log ratio (ALR) transformation is described in greater detail elsewhere.24 Briefly, for a composition made up of D components (x1, x2, …, xD), the ALR transformation generates D-1 terms where each term is the log of the ratio of each component to a referent component, for example, log(x1/xD), log(x2/xD), …, log(xD-1/xD). For the abdominal composition, the ‘Other’ tissue component was used as the referent; for the thigh composition, the bone component was used as a referent. A log2 transformation was applied to these ratios such that interpretation of coefficients is for a two-fold increase in the ratio of the numerator tissue compared to its respective referent component.

2.4 Outcome definitions: Hypertension and type 2 diabetes categories

Systolic (SBP) and diastolic (DBP) blood pressures were measured 3 times in a seated position with 10 min of rest in between readings using an automated sphygmomanometer (Omron); the average of the last two readings was used for this analysis. Hypertension was defined using ACC/AHA 2017 criteria,32 and individuals who were on an antihypertensive medication were assigned Stage 2 Hypertension regardless of SBP or DBP.

Fasting serum glucose and insulin measures were measured at the Advanced Research and Diagnostics Laboratory (ARDL), University of Minnesota. Fasting serum glucose was measured using an enzymatic procedure (interassay CV: 1.3%–1.8%), and fasting serum insulin was measured using a Sandwich immunoassay procedure (interassay coefficient of variation: 3.1%) (assays manufacturer: Roche Diagnostics). Insulin resistance was estimated using the HOMA-IR equation.33 Diabetes categories were defined based on American Diabetes Association (ADA) fasting glucose criteria.34 Individuals taking antidiabetic medications were classified as “Type 2 Diabetes” regardless of measured fasting glucose.

2.5 Other measures

Standing height was measured to the nearest 0.1 cm using a wall-mounted stadiometer. Body weight was recorded to the nearest 0.1 kg without shoes on a balance beam scale. BMI was calculated from body weight and standing height (kg/m2); obesity status was defined as normal weight (18.5–24.9 kg/m2), overweight (25–30 kg/m2), or obese (>30 kg/m2). Information on current smoking [yes/no], number of hours walked per week, watching 14 or more hours of television (TV) per week [yes/no], current intake of alcohol of more than four drinks per week [yes/no], family history of hypertension or diabetes [yes/no] and medication use were assessed using standardized interviewer-administered questionnaires. Lipid-modifying medications were defined as the use of a statin, ezetimibe, or a combination of the two. Self-reported information on walking was recorded as walking is the predominant form of physical activity on the island of Tobago. Men were asked to bring all prescription medications taken in the past 30 days to their clinic visit.

2.6 Statistical analyses

Population characteristics were reported overall and stratified by obesity status; p-values for linear trend were reported, with linear contrasts used for continuous variables and Cochrane-Armitage trend test used for categorical variables. Ternary plots for abdominal and thigh compositions were generated using the package ‘compositions’35 in R version 3.5.2,36 and the mean compositions for each hypertension and diabetes category was plotted over the population distribution. Age-adjusted Pearson correlations were reported between the ALR-transformed components, BMI, and continuous risk factors. Linear regressions were performed for continuous risk factor outcomes (SBP, DBP, glucose, insulin, and HOMA-IR); log transformations were applied to non-normal distributions. Ordinal logistic regression models were performed for hypertension and diabetes categories; a partial proportional odds models with unequal slopes for lipid-modifying medications was chosen for the diabetes category model after rejection of the score test and empirical cumulative logit plots indicated that this variable was the only one violating the proportional odds assumption. All models were adjusted for age, BMI, family histories of diabetes, drinking 4+ alcoholic drinks per week, current smoking, watching TV ≥ 14 h per week, hours walked per week for exercise, taking lipid-modifying medications, total measured abdominal and thigh volumes, and the ALR-transformed abdominal and thigh compositions; the continuous biomarker models were additionally adjusted for antihypertensive or antidiabetic medication use. Interactions of abdominal or thigh tissues with BMI were assessed and visualized using the PROCESS macro37; interaction models with hypertension and type 2 diabetes outcomes used a dichotomized version of these outcomes. Statistical significance was based on α = 0.05, and analyses were performed using SAS 9.4 software (SAS Institute, Inc.).

2.7 Sensitivity analyses

Three sets of sensitivity analyses were performed. In the first sensitivity analysis, models only included either the abdominal composition or the thigh composition, but not adjusting simultaneously for both regions.

In the second set of sensitivity analyses, HU-based estimates to generate abdominal muscle and IMAT components, as there were no direct measures of abdominal muscle/IMAT at the L4-L5 intervertebral space. Abdominal muscle and IMAT were estimated without manual tissue tracing in the area between the peritoneal cavity and the muscular fascia using attenuation tissue thresholds defining AT (−190 to −30 HU) and lean muscle (−29 to 160 HU). Two individuals had missing data, leading to a sample size of 608 individuals for this sensitivity analysis. Abdominal compositions in the sensitivity analyses now consisted of ASAT, VAT, abdominal IMAT, abdominal muscle, and remaining ‘Other’, such that two new log ratio terms (IMAT:Other and Muscle:Other) were included in models. Models were otherwise constructed as indicated in the main analyses.

In the third sensitivity analysis, thigh muscle attenuation was included as a surrogate measure for intramuscular fat accumulation.38 Thigh muscle attenuation was defined as the average HU across measured thigh muscle volumes; a lower average HU reflects greater fatty infiltration.

3 RESULTS 3.1 General baseline characteristics

Overall population characteristics and characteristics stratified by obesity status are displayed in Table 1. Men had a median age of 62 and mean BMI of 27.7 kg/m2. About 75.5% of the men had stage 1 or stage 2 hypertension, while 23% of the men had type 2 diabetes; more than half of individuals with hypertension or diabetes were on a medication for that disease.

TABLE 1. Population characteristics, overall, and by BMI category Variable Mean (SD), Median (IQR), or N(%) Overall (N = 610) Normal weight (N = 177) Overweight (N = 266) Obese (N = 167) p-value Demographic and lifestyle factors Age (years) 62.0 (57.0, 68.0) 63.0 (58.0, 71.0) 62.0 (57.0, 69.0) 60.0 (56.0, 65.0) 0.0003 Weight (kg) 85.5 (15.4) 70.6 (7.0) 84.2 (7.9) 103.4 (12.3) <0.0001 Height (cm) 175.5 (6.7) 176.0 (6.6) 175.4 (6.9) 174.9 (6.3) 0.1186 BMI (kg/m2) 27.8 (4.7) 23.2 (21.7, 24.2) 27.4 (26.2, 28.3) 32.9 (30.9, 35.4) <0.0001 Current smoker [N(%)] 44 (7.2%) 16 (9.0%) 19 (7.1%) 9 (5.4%) 0.1906 Drinks 4+ alcoholic beverages per week [N(%)] 75 (12.3%) 20 (11.3%) 36 (13.5%) 19 (11.4%) 0.9699 Watches TV ≥ 14 h per week [N(%)] 294 (48.2%) 84 (47.5%) 128 (48.1%) 82 (49.1%) 0.7609 Walking for exercise (hours per week) 1.9 (0.0, 5.0) 1.5 (0.0, 4.5) 2.1 (0.0, 5.0) 1.5 (0.0, 5.0) 0.5095 On lipid-modifying medications [N(%)] 79 (13.0%) 18 (10.2%) 33 (12.4%) 28 (16.8%) 0.0696 Has family history of type 2 diabetes [N(%)] 340 (55.7%) 90 (50.8%) 148 (55.6%) 102 (61.1%) 0.0564 Has family history of hypertension [N(%)] 331 (54.3%) 79 (44.6%) 145 (54.5%) 107 (64.1%) 0.0003 Cardiometabolic disease measures Fasting glucose (mg/dL) 89.0 (81.0, 102.0) 87.0 (79.0, 97.0) 89.0 (82.0, 102.0) 93.0 (83.0, 115.0) 0.0008 Fasting insulin (µU/mL) 9.0 (5.8, 14.0) 5.7 (4.0, 7.7) 9.0 (6.3, 13.2) 15.0 (11.5, 19.8) <0.0001 HOMA-IR 2.2 (1.3, 3.5) 1.3 (0.9, 1.8) 2.1 (1.4, 3.1) 3.7 (2.5, 5.5) <0.0001 Type 2 diabetes categories [N(%)] <0.0001 Normal glucose 401 (65.7%) 136 (76.8%) 176 (66.2%) 89 (53.3%) Impaired fasting glucose 70 (11.5%) 13 (7.3%) 31 (11.7%) 26 (15.6%) Type 2 diabetes 139 (22.8%) 28 (15.8%) 59 (22.2%) 52 (31.1%) Antidiabetic medication use [N(%)] 106 (17.4%) 23 (13.0%) 46 (17.3%) 37 (22.2%) 0.0251 SBP (mmHg) 142.0 (21.8) 134.0 (120.5, 152.0) 139.5 (126.5, 156.0) 145.0 (19.4) 0.0013 DBP (mmHg) 79.7 (12.2) 74.5 (68.0, 82.5) 79.9 (11.6) 83.5 (12.0) <0.0001 Hypertension categories [N(%)] <0.0001 Normal 68 (11.2%) 33 (18.6%) 28 (10.5%) 7 (4.2%) Elevated 82 (13.4%) 35 (19.8%) 36 (13.5%) 11 (6.6%) Stage 1 76 (12.5%) 21 (11.9%) 34 (12.8%) 21 (12.6%) Stage 2 384 (63.0%) 88 (49.7%) 168 (63.2%) 128 (76.7%) Antihypertensive medication use [N(%)] 244 (40.0%) 43 (24.3%) 110 (41.4%) 91 (54.5%) <0.0001 Body composition tissue measures ASAT volume (cm3) 181.8 (129.2, 245.7) 101.8 (49.3) 188.2 (50.9) 308.3 (100.3) <0.0001 VAT volume (cm3) 86.1 (52.3, 125.0) 44.6 (26.7, 68.4) 92.4 (39.6) 138.6 (56.6) <0.0001 Other abdominal volume (cm3) 312.8 (46.6) 288.9 (37.8) 308.5 (40.6) 345.2 (46.3) <0.0001 Total abdominal volume (cm3) 581.5 (485.6, 690.6) 442.8 (69.8) 590.0 (79.4) 792.0 (139.7) <0.0001 TSAT volume (cm3) 341.2 (229.9, 485.0) 204.0 (111.9) 361.6 (133.7) 587.1 (228.3) <0.0001 Thigh IMAT volume (cm3) 118.4 (50.5) 80.3 (36.8) 120.4 (37.4) 143.6 (118.5, 187.7) <0.0001 Thigh muscle volume (cm3) 1068.4 (172.8) 951.1 (139.6) 1079.3 (144.0) 1175.2 (171.4) <0.0001 Thigh bone volume (cm3) 44.3 (41.6, 47.9) 42.8 (40.5, 45.4) 44.6 (5.1) 46.3 (4.8) <0.0001 Total thigh volume (cm3) 1608.9 (326.0) 1278.2 (176.0) 1605.9 (174.8) 1964.2 (254.4) <0.0001 Note: Continuous p-values: linear regression predicting the characteristic (for parametric), or Joncheere-Terpstra Test (for nonparametric). Categorical p-values: Cochrane-Armitage trend test for binary variables, or Mantel-Haenszel Chi-square test for ordinal variables. Abbreviations: ASAT, abdominal subcutaneous adipose tissue; BMI, body mass index; DBP, diastolic blood pressure; IMAT, intermuscular adipose tissue; SBP, systolic blood pressure; TSAT, thigh subcutaneous adipose tissue; TV, television; VAT, visceral adipose tissue.

Ternary plots (Figures 1 and 2) were constructed to show overall abdominal and thigh composition distributions in the population, as well as the mean compositions for each of the cardiometabolic disease categories. Ternary plots are read such that the closer an individual is plotted towards a particular corner, the greater that individual's composition is comprised of that component (with a corner being completely 100% that composition). In the abdominal compositions (Figures 1 and 2, left panels), individuals in higher cardiometabolic disease categories appeared to have a greater %ASAT, and a slight shift to having a greater %VAT. In the thigh compositions (Figures 1 and 2, right panels), individuals in higher cardiometabolic disease categories appeared to have greater %TSAT and a slight shift towards having a greater %IMAT.

Ternary plots of abdominal (left) and thigh (right) compositions, with average composition by hypertension category overlaid. VAT, visceral adipose tissue; ASAT, abdominal subcutaneous adipose tissue; IMAT, (thigh) intermuscular adipose tissue; TSAT, thigh subcutaneous adipose tissue

Ternary plots of abdominal (left) and thigh (right) compositions, with average composition by diabetes category overlaid. VAT, visceral adipose tissue; ASAT, abdominal subcutaneous adipose tissue; IMAT, (thigh) intermuscular adipose tissue; TSAT, thigh subcutaneous adipose tissue

3.2 Association of tissue depots with anthropometric measures and diabetes categories

Age-adjusted Pearson correlations (Table 2) were performed to investigate associations between ALR-transformed abdominal and thigh components, BMI, and continuous risk factor measures. BMI was most strongly correlated with ASAT and TSAT (r = 0.69 and 0.70, respectively; all p < 0.001), and moderately correlated with VAT and IMAT components (r = 0.58 and 0.61, respectively; all p < 0.001); similar correlations were also observed between tissue components and their respective total measured region (abdomen or thigh). Interrelationships among all AT components were high, with some of the highest correlation coefficients being between ASAT, TSAT, and IMAT (r = 0.80–0.89; all p < 0.0001). Despite these higher correlations, multicollinearity was not identified when investigating condition indices and variance proportions in regression models.

TABLE 2. Age-adjusted Pearson partial correlation coefficients for ALR-transformed components, BMI, and continuous risk factors ASATa VATa TSATa Thigh IMATa Thigh musclea BMI (kg/m2) Abdominal volume (cm3) Thigh volume (cm3) ASATa 1 0.73** 0.89** 0.80** 0.14* 0.69** 0.72** 0.70** VATa 0.73** 1 0.64** 0.61** 0.13* 0.58** 0.68** 0.53** TSATa 0.89** 0.64** 1 0.82** 0.20** 0.70** 0.71** 0.77** Thigh IMATa 0.80** 0.61** 0.82** 1 0.29** 0.61** 0.64** 0.69** Thigh musclea 0.14* 0.13* 0.20** 0.29** 1 0.27** 0.19** 0.35** BMI (kg/m2) 0.69** 0.58** 0.70** 0.61** 0.27** 1 0.91** 0.87** SBP (mmHg) 0.21** 0.19** 0.16** 0.14* 0.04 0.18** 0.18** 0.14* DBP (mmHg) 0.24** 0.22** 0.18** 0.16** 0.04 0.24** 0.24** 0.20** Log glucose 0.16** 0.16* 0.09* 0.06 −0.04 0.18** 0.13* 0.005 Log insulin