β-cell function in black South African women: Associations with insulin clearance and ectopic fat

Background: The role of ectopic fat, insulin secretion and clearance in the preservation of β-cell function in black African women with obesity who typically present with hyperinsulinemia is not clear. We aim to examine the associations between disposition index (DI, an estimate of β-cell function), insulin secretion and clearance and ectopic fat deposition. Methods: This is a cross-sectional study of 43 black South African women (age 20-35 years) with obesity (BMI 30-40 kg/m 2 ) and without type 2 diabetes that measured the following: DI, insulin sensitivity (S I ), acute insulin response (AIRg), insulin secretion rate (ISR), hepatic insulin extraction and peripheral insulin clearance (frequently-sampled intravenous glucose tolerance test); pancreatic and hepatic fat, visceral adipose tissue (VAT) and abdominal subcutaneous adipose tissue (aSAT) volume (magnetic resonance imaging), intramyocellular (IMCL) and extramyocellular fat content (EMCL) (magnetic resonance spectroscopy). Results: DI correlated positively with peripheral insulin clearance before (β 55.80, p=0.002) and after adjusting for hepatic insulin extraction. Higher DI was associated with lower VAT, pancreatic fat and soleus fat, but VAT explained most of the variance in DI (32%). Additionally, higher rst phase ISR (p=0.033) and lower hepatic insulin extraction (p=0.022) associated with lower VAT, independent from S I, rather than with ectopic fat. Conclusion: Peripheral insulin clearance emerged as an important correlate of DI, independent from hepatic insulin extraction . However, VAT was the main determinant of a lower DI above ectopic fat depots. Importantly, VAT, but not ectopic fat, was associated with both lower insulin secretion and higher hepatic insulin extraction, independent from S I , and may provide a novel explanation of these ndings in black South African women with obesity.

For the determination of hepatic, pancreatic and skeletal muscle fat, VAT and aSAT obtained by MRI, a fat fraction map was created, calculated as the fat signal over the sum of the water and fat signals. A region of interest (ROI) was drawn on 7 consecutive slices in the right lobe of the liver. In the soleus and tibialis anterior muscles, an ROI was drawn on 7 consecutive slices with a method adapted from Machann et al. [20]. Pancreatic fat was determined by drawing one circular 1 cm 2 ROI in the head, body and tail of the pancreas to calculate the average [21]. The methods for determining VATand aSAT volumes were previously published [22]. The above mentioned methods, to quantify hepatic [23], pancreatic [24] and skeletal muscle fat [25] and VAT and aSAT [26], have previously been reported to have high precision.

Determination of S I and β-cell function
A FSIGT allows for the determination of the acute insulin response after an intravenous glucose load [27]. After 20 minutes insulin is infused to suppress endogenous insulin production which enables the evaluation of the effect of insulin on glucose disappearance. A FSIGT was used in this study because it provides measures of the AIRg as well as S I and it is less labour intensive compared to the clamp method and it does not require a steady state. Additionally, the FSIGT shows good correlation to the hyperglycemic clamp [28]. The FSIGT was conducted after an overnight fast (10-12 h). Glucose (50% dextrose, 11.4 g/m 2 body surface area) was infused at 0 min over a 60 s period. At 20 min, human insulin (0.02 unit/kg; NovoRapid, Novo Nordisk Limited, Cape Town, SA) was infused over 5 min at a constant rate. Thirty-two blood samples were drawn over 240 minutes for the determination of plasma glucose and serum insulin and C-peptide.

Biochemical analyses
Fasting whole blood was collected and analysed for HbA1c using high-performance liquid chromatography (Meharini Diagnostics, Florence, Italy). Plasma glucose concentrations was measured using a colorimetric assay (Randox (Pty) Ltd, Gauteng, South Africa). Serum insulin and C-peptide concentrations were determined by an immunochemiluminometric assay (IMMULITE 1000 immunoassay system, Siemens Healthcare (Pty) Ltd, Midrand, South Africa).

Mathematical modelling
The minimal model of glucose kinetics was used to calculate S I , AIR g and glucose effectiveness [29]. S I is a measure of the fractional disappearance of glucose for a given insulin concentration achieved by uncoupling the glucose and insulin responses. Glucose effectiveness is a measure of the ability of glucose to enhance its own uptake. AIRg is the incremental area under the insulin curve in response to intravenous glucose over the rst 10 minutes, and re ects the rst phase insulin response. DI is a measure of the insulin response to glucose relative to the prevailing level of S I (AIRg x S I ) which gives an index of whether the insulin response is adequate for the level of S I [1]. The ISR was determined by C-peptide deconvolution using a 2-compartmental model with standard kinetic parameters for individuals with obesity [30]. The incremental area-under curve (AUC), above baseline, for ISR, glucose, insulin and Cpeptide were calculated using the trapezoidal rule for the rst 10 minutes and the entire 240 min of the FSIGT, and are referred to as the rst phase AUC (ISR rst phase ) and the total AUC (ISR total ), respectively. FE L and CLp were calculated using a method previously described by Polidori et al. [8]. Hepatic insulin clearance were described using both linear and saturable kinetics. The linear model estimated a single parameter for hepatic insulin clearance (FE L ) while the saturable model estimated two parameters (Vmax and Km). The linear model described hepatic insulin clearance well (fractional standard deviation <5%) in 91% (39/ 43) of participants. In addition, the linear model described the plasma insulin values with a mean normalized root square error (NMRSE) ±SD of 8.3 ±2.14%. The parameters of the models were estimated using WinSAAM [31].

Statistical analysis
Data were analysed using STATA 12 .0 (College Station, TX, USA). Data analysis included those that completed a FSIGT (n=43). The cohort was divided into tertiles based on the DI, calculated as S I x AIRg: DI Low , DI Intermediate and DI High . Normally and non-normally distributed data were expressed as mean ± standard deviation (SD) and median (25-75 th percentile), respectively. Differences between DI tertiles was determined using one-way ANOVA with a Bonferroni post hoc test or Kruskal Wallis test with the Dunn post hoc test for normally and non-normally distributed data, respectively. Correlations were conducted using Pearson (normally distributed variables) and Spearman (nonnormally distributed variables). Linear regression was performed with the transformed variables to determine the association between DI components and central fat depots, adjusted for S I , where applicable.

Results
Subject Characteristics of the overall sample and by DI tertiles A hyperbolic relationship between S I and AIRg, the main components of DI, is presented in Figure 1 for the overall sample (A) and in each of the DI tertiles (B). The cohort had a median age of 23 (IQR 21-27) years and a mean BMI of 33.2 ±2.8 kg/m 2 . Participant characteristics by DI tertile, as an estimate of β-cell function, are displayed in Table 1.
The group with the highest DI, were the most insulin sensitive (post hoc p=0.001) and had the highest glucose effectiveness (post hoc p<0.001) compared to the group with the lowest DI. AIRg and ISR rst phase showed no decline across DI tertiles (post hoc p=0.060 and p=0.071, respectively). In addition the group with the highest DI had the highest CLp (post hoc p=0.001) and lowest FE L (post hoc p=0.049) compared to those in the lowest DI tertile. Further, the highest DI had the lowest pancreatic and hepatic fat, and the lowest VAT and VAT-aSAT ratio, compared to those with the lowest DI.
Associations of DI and its components with CLp and FE L Further, in univariate analysis, DI was positively associated with CLp and inversely with FE L which explained 23.4% and 16.8% of the variance in DI, respectively (Table 2). However, when both CLp and FE L were placed in the model only CLp remained a signi cant positive determinant of DI. S I was not associated with FE L (p=0.065) and CLp (p=0.881) in the univariate model ( Table 2). The association between S I and FE L became signi cant when adjusted for CLp (p=0.038). AIRg was associated with CLp in the univariate analysis but after adjusting for FE L and S I this association was diminished ( Table 2).

Contributions of FE L and ISR rst phase to AIRg
In addition, we used linear regression models to determine the relative contributions of FE L and ISR rst phase to AIRg, without and with adjustment for S I (Table 3). In the univariate models, ISR rst phase and FE L explained 81.9% and 57.6%, of the variance in AIRg, respectively. S I explained 32% of the variance in AIRg (data not shown). In the multivariate model, FE L and ISR rst phase were independently associated with AIRg, adjusted for S I. DI and its components -associations with body fat distribution and ectopic fat DI was not associated with fat mass (%), leg fat mass or aSAT (data not shown) but was negatively associated with VAT and the VAT-aSAT ratio (β -167.4, p<0.001). The unadjusted associations between DI and ectopic fat depots and VAT, are shown in Figure 2. DI was inversely correlated with pancreatic fat, total soleus fat and both soleus IMCL and EMCL. DI was not associated with tibialis anterior fat (β -2.59, p=0.887). In univariate analysis, VAT explained 32% of the variance in DI, while soleus IMCL explained 18.7%, soleus EMCL 10.9% and pancreatic fat 12%. Adjusting for age and fat mass (%) in multivariable models did not alter the models. Accordingly, the negative associations of DI with pancreatic fat (β -62.9, p=0.030), VAT (β -0.037, p<0.001), soleus fat (β -50.0, p=0.049), soleus IMCL (β -41.3, p=0.012) and soleus EMCL (β -32.0, p=0.047) remained. We explored the associations between DI and ectopic fat, independent of VAT and found that the associations of DI with pancreatic fat, hepatic fat, soleus fat and soleus IMCL were ameliorated to non-signi cance (p>0.05) (data not shown). Instead, VAT remained a signi cant determinant of DI independent of any of the ectopic fat depots (data not shown).
The components of DI (S I , AIRg, ISR rst phase , FE L and CLp) were not associated with fat mass (%) or ectopic fat sites (p>0.05) (data not shown). Rather, signi cant associations were found between the components of DI and central fat measures (Table 4). A lower S I was associated with higher VAT but not aSAT, while higher AIRg was associated with higher aSAT and lower VAT-aSAT. When adjusting for S I , a higher AIRg was associated with only a lower VAT and VAT-aSAT. Only after adjusting for S I, an inverse association emerged between ISR rst phase and VAT. We further showed that FE L was positively associated with VAT only after adjusting for S I . CLp was not associated with any central fat depots (data not shown).

Discussion
The correlates of β-cell function, a critical factor in the pathogenesis of T2D, are incompletely understood, especially in black African populations who present with a phenotype of low S I, hyperinsulinemia and low ectopic fat deposition.
Our study extend existing evidence by demonstrating that DI was positively associated with CLp. Notably, the major correlate of AIRg was ISR rst phase , above insulin clearance. VAT emerged as the strongest correlate of DI, above and independent of pancreatic fat and soleus fat. Additionally, we showed that a lower VAT also associated with a higher ISR rst phase and lower FE L . Thus, our ndings suggest that VAT is a more important correlate of a lower DI than ectopic fat in this cohort, not only through its association with a lower S I , but also through its relation with the AIRg downstream components, ISR and FE L .
DI associates with peripheral insulin clearance Our study, not only distinguishes between FE L and CLp, which has only been done in a few studies in adults without T2D [6,8,32], but we also demonstrated for the rst time a positive association between DI and CLp. The reason for this relationship is unclear. Nevertheless, considering DI is based on the product of S I and AIRg, we can postulate two scenarios. Firstly, a higher DI may be due to a hyperinsulinemia relative to the the level of S I . However, hyperinsulinemia has been associated with lower insulin clearance in both adipose tissue [33] and muscle [34], due to reduced a nity of insulin receptors at these sites, and is therefore an unlikely explanation for our nding of a higher CLp relative to higher DI. Secondly, a higher DI could be due to a greater S I in relation to the level of AIRg. In this scenario, a higher CLp may be due to enhanced binding of insulin to insulin receptors in peripheral tissues. Although a positive association has been demonstrated between S I and hepatic insulin clearance [35], no association between insulin internalization, a measure of insulin clearance, and S I was observed in rat adipocytes [33], whereas the association between S I and insulin clearance in muscle is unknown. However, in support of the second scenario, we showed that those with the highest DI and CLp were also more insulin sensitive, but only exhibited a slightly higher AIRg compared to those with lower DI and CLp. Nevertheless, CLp was not associated with S I in our study, but rather with the insulin secretory component. Further, we may also consider that the observed association between DI and CLp is in compensation for a lower FE L , but this association remained independent of FE L . We also evaluated the associations of CLp with ectopic fat deposition as possible explanation for the association with DI, but found that CLp was not associated with skeletal muscle fat. Skeletal muscle fat has been associated with reduced S I [36], which may also affect CLp in the muscle. Further study is justi ed to evaluate the relationship between CLp and S I in the muscle, considering that the muscle is only secondary to the kidney in the proportion of insulin cleared in the periphery [37]. Of note, CLp in our study is much higher compared to another study conducted in black American women [6]. However, the acute insulin response in our cohort was twice as high as that observed in Piccinini et al, and may contribute to this discrepancy.

Associations between DI and ectopic fat
Preservation of β-cell function is critical for delaying the onset of T2D. We, therefore, investigated, rstly, whether ectopic fat accumulation may explain the variance in DI and secondly, we assessed the associations of ectopic fat with the components of DI. Our study demonstrated that a higher DI was associated with a lower pancreatic and soleus fat but these associations were not independent from VAT. A similar observation was found in overweight African American and Hispanic adolescents (13 to 25 years old) without T2D [38]. In contrast a positive association between DI and pancreatic fat, adjusted for BMI and VAT, was shown in black African American women [13]. However, this study included participants with and without T2D, which may explain the incongruent ndings. Our study extends the literature by showing that pancreatic fat was not associated with ISR rst phase or with FE L in black Africans.
Notably, a positive association was found between pancreatic fat and VAT in our study. Our ndings therefore suggest that pancreatic fat may only be a marker of VAT accumulation and may not be detrimental to β-cell function in this cohort.

Contribution of hepatic insulin clearance and ISR to AIRg
The ability to maintain DI and prevent deteriorating glucose tolerance depends on the balance between AIRg and S I [1]. However, there is no consensus on the mechanism of maintaining a higher AIRg, which is frequently observed in black African populations [3,4]. Some studies reported that a lower hepatic insulin clearance alone is responsible for a higher AIRg [39,40], while others showed that both lower hepatic insulin clearance and higher ISR contribute towards a higher AIRg [4,6]. Notably, AIRg may be out of proportion for the level of S I, and to assess the relative contribution of ISR rst phase and insulin clearance to AIRg in this context, we need to adjust for S I . Accordingly, a lower FE L and higher ISR rst phase were the main independent contributors towards a higher AIRg, but ISR rst phase explained more of the variance in AIRg. Lowering hepatic insulin clearance is an important compensatory mechanism to reduce the strain on the pancreatic β-cells, which has been shown in canines [41]. Indeed, we also noted a higher DI is associated with a lower FE L which may be explained by the negative association between AIRg and FE L . Further, we showed that a higher ISR associates with AIRg, independently of lower FE L and S I , suggesting that despite a lower FE L , pancreatic βcells may continue to secrete insulin at a higher rate, which may be detrimental to the longevity of the β-cell in this cohort.

Associations of ISR and hepatic insulin clearance with ectopic fat
The current study further examined the associations between the components of AIRg and ectopic fat. Previously, reduced hepatic insulin clearance and increased insulin secretion, have been associated with hepatic fat accumulation [42]. An association between fasting hepatic insulin clearance and hepatic fat was studied in African American women [15], but the association between stimulated hepatic insulin clearance, which is a more physiological response, and hepatic fat has not been previously investigated in black African populations without T2D . We showed that hepatic fat was not associated with FE L and ISR in black South African women with obesity.
Therefore, in black African populations, hepatic fat may not be an important correlate of hepatic insulin clearance and insulin secretion prior to T2D, and also not in those with early T2D [43]. Instead, we found that a lower ISR rst phase and higher FE L were associated with higher VAT. This highlights a novel mechanism to explain the association between AIRg and VAT since to the best of our knowledge, no previous study has evaluated the effect of central fat depots on FE L . In addition, evidence on the associations between central fat depots and ISR, determined by C-peptide deconvolution are also limited in black African populations [11]. We add to the literature by showing a negative association between VAT and ISR rst phase , independent of S I , in premenopausal black South African women.
Interestingly, black African women have lower VAT compared to other ethnicities [3,4], but they have a greater propensity to increase VAT over time, with the greatest increase occurring in the 20-29-year age group [44]. However, further studies are needed to con rm the effect of VAT on ISR in this cohort.

Associations between S I and ectopic fat
We also assessed the association between S I and ectopic fat and found no association between S I and skeletal muscle or hepatic fat. These ndings suggest that these ectopic fat depots are not important correlates of S I in our study. However, we should consider that we measured whole body S I , which could have diluted the associations with hepatic and soleus fat. Indeed, a small South African study found that in black South African women, hepatic fat was associated with hepatic S I but not peripheral S I [14]. Furthermore, total soleus fat and soleus IMCL were associated with peripheral S I, but not hepatic S I [14]. Nevertheless, lower S I was associated with higher VAT which may contribute to the observed inverse association between DI and VAT in this cohort.
A major strength of this study is that ISR and both hepatic and peripheral clearance were determined using mathematical modelling, which has never been done before in an African cohort. Further we assessed pancreatic, hepatic and skeletal muscle ectopic fat depots, as well as VAT and aSAT volumes using MRI. However, limitations of this study were that we did not determine glucose tolerance, and we did not distinguish between hepatic and peripheral S I . We have not corrected for multiple correlation which may have produced false positive associations.
However, the focus was to explore the possible signi cant associations between variables and therefore we wanted to minimize false negative associations. In addition, these ndings are applicable to premenopausal, black South African women with obesity and may not be extrapolated to men or other races.
In conclusion, an original nding from this study was that DI was associated with CLp, independent from FE L . Further, while both FE L and ISR rst phase independently contributed towards hyperinsulinemia, ISR rst phase was more important.
Ectopic fat was not an important independent correlate of DI and its components. Rather, a key nding was that higher VAT was the principal correlate of a lower DI, above other ectopic fat depots. Additionally, the associations of higher VAT on the downstream components of DI, a lower ISR rst phase and higher FE L may contribute to lower DI prior to T2D onset but requires further elucidation. Accordingly, the prevention of VAT accumulation, especially in young black African women should be an important target for β-cell preservation.

Availability of data and materials
The data and materials used to support the ndings in this study is available from the corresponding author upon a reasonable request.

Ethics approval and consent to participate
Ethical approval was obtained from the University of Cape Town Human Research Ethics Committee (reference number 799/2015). This study formed part of a randomized controlled trial that was registered in the Pan African Clinical Trial Registry (PACTR201711002789113). All study participants provided written informed consent to participate in the study. The study was conducted in accordance to ethical principles outlined in the Declaration of Helsinki.

Consent for publication
All the study participants provided written informed consent to participate in the study.      Hyperbolic association between insulin sensitivity (SI) and acute insulin response to glucose (AIRg) for overall sample (A) and by disposition index (DI) tertiles (B) Figure 2