We thank Saisho, Butler, and Butler (1) for their interest in our study (2). They confirm and extend our findings by providing additional data on pancreatic fat content in individuals with and without type 2 diabetes using different technologies, i.e., computed tomography (CT) to measure pancreas volume and morphometric analyses of intrapancreatic fat spaces to estimate fat in postmortem pancreas sections (3).
As repeatedly stated in our paper (2), proton magnetic resonance spectroscopy (1H-MRS) cannot discriminate between intracellular fat in adipocytes versus nonadipocytes. Similar to Saisho, Butler, and Butler, we reasoned that since islets constitute only ∼2% of the pancreas mass, the fat signal obtained by 1H-MRS is to the greatest extent derived from infiltrating adipose tissue, fatty replacement of damaged tissue, stromal cells, or a combination.
A possible explanation of why Saisho, Butler, and Butler did not find differences in pancreatic fat between type 2 diabetic and nondiabetic subjects and why they regard pancreatic fat as yet another indicator of obesity could be the difference in the methods used and, consequently, the definition of pancreatic fat (3). They used an initial window level on CT images such that both parenchyma and surrounding/infiltrating fat are visible, which they defined as the total pancreas. Subsequently, the fat signal was suppressed by choosing another fixed window setting such that only the signal of primary parenchyma remained (which the authors termed “pancreas-volume-free-of-fat”) (3). The fat volume obtained through this method lies primarily outside the parenchyma, which is, not surprisingly, associated with adiposity. This method differs crucially from our 1H-MRS method, which measures fat within the pancreatic parenchyma. The magnetic resonance images used for positioning the volume of interest for localized spectroscopy have excellent contrast between parenchyma and surrounding fat (2), which allows concentration on parenchymal tissue only. Consequently, in the case of almost fat-free parenchyma, the fat content of the pancreas can be very low despite possible presence of surrounding fat, as illustrated by two nondiabetic subjects with fat contents of only 1.0 and 1.4%. These cannot simply be discarded as outliers, since the measurements were of high spectral quality and reproducible. Interestingly, we found no association between 1H-MRS–measured BMI-related liver fat and pancreatic fat (3).
Saisho, Butler, and Butler used contrast enhancement in the CT scans, which may influence CT density, thereby potentially causing additional intersubject variability when a fixed window level for outlining the parenchymal volume is used. Furthermore, they did not use the actual pixel density, which would take into account parenchymal fat (as done in the cited study by Kodama et al. [4]) but instead assigned each pixel to be either fat or parenchyma, which they mention as a potential source of error (3).
We concur with Saisho, Butler, and Butler that histological measurements of pancreatic fat have serious limitations, since the removal of fat during tissue processing may not be fully proportional to the actual fat content premortem (2). Finally, their studies preclude conclusions regarding the existence of a link between pancreatic fat and β-cell function in humans, as no such measurements were performed. Taken together, we respect that the findings of Saisho, Butler, and Butler may not be in full agreement with ours;however, this discrepancy is due to differences in the methods used, all of which have their own limitations.
