Fat Distribution Within the Pancreas According to Diabetes Status and Insulin Traits Free

Fatty pancreas is the most common disease of the pancreas and occurs in approximately one in five people in the general population (1,2). Earlier studies have established that liver fat can predict risk of developing diabetes (3) and that a decrease in liver and intrapancreatic fat content has an important role in the remission of diabetes (4). The 2008 “twin cycle” hypothesis postulated that fat accumulation in the liver would exacerbate insulin resistance, which would then result in impaired regulation of glucose homeostasis by the pancreas and the eventual onset of type 2 diabetes (5). The predictions of the twin cycle hypothesis have recently been confirmed, and there has been reasonable interest in investigating whether intrapancreatic fat could have an important role in the development of type 2 diabetes (6). In a 2014 large cross-sectional study, investigators found that people with fatty pancreas disease have significantly increased insulin resistance (2), and in our comprehensive 2017 meta-analysis we reported that people with fatty pancreas are more than two times more likely to have type 2 diabetes (7).

Although the importance of accurate measurement of intrapancreatic fat in investigations involving people with diabetes is now appreciated, it is important to note that in the studies included in the 2017 meta-analysis (7) there were several methods used to measure intrapancreatic fat and this led to a considerable variation in findings. Transabdominal ultrasound, endoscopic ultrasound, computed tomography, MRS, and chemical shift–encoded MRI were all used. Of these, MRI has emerged as the optimal nonionizing method for noninvasive intrapancreatic fat quantification (1). Further, the standardized magnetic resonance image biopsy (“MR-opsy”) technique of measuring intrapancreatic fat by placing regions of interest in the head, body, and tail is now advocated (8). It has been shown to have higher accuracy and reproducibility compared with other techniques (e.g., conventional freehand drawing), which may often be subject to errors due to the inclusion of surrounding entities such as the duodenum, splenic vein, inferior vena cava, and visceral fat. These anatomical structures can make accurate estimation of intrapancreatic fat difficult and potentially result in overestimation of fat content. Importantly, the lack of consistency in results could also be due to studies failing to account for the potential for uneven fat distribution among the head, body, and tail regions of the pancreas. Uneven distribution of intrapancreatic fat was first suggested in a 1995 crude analysis of 80 computed tomography scans from a multihospital database (9). This question is still unsettled in the absence of large studies that use modern techniques and take into account possible confounders (10).

With a large prospectively accrued population-based cohort and by using comprehensive adjustment for key covariates, and the MR-opsy technique to quantify intrapancreatic fat, we aimed to investigate the differences in intrapancreatic fat among the head, body, and tail regions of the pancreas, as well as the differences in regional intrapancreatic fat according to diabetes status and insulin traits.

This was a cross-sectional study conducted at the University of Auckland, with participants enrolled as four individual cohorts, and approved by the New Zealand Health and Disability Ethics Committee (13/STH/182, 16/STH/23, 17/NTA/172, 18/NTB/1). The investigation included adults (at least 18 years old) from the general population of Auckland who provided written informed consent to undergo MRI of the pancreas. Individuals were excluded if they had history of pancreatic cancer or other malignancy, bariatric or pancreatic surgery, chronic pancreatitis or any other pathology of the pancreas detected on imaging, or type 1 diabetes; had received endoscopic or radiological procedures involving the pancreas; or had chronic liver disease. Individuals were also excluded if they had been involved in a weight loss program or had received specialized dietary advice, used systemic corticosteroids, or had metal implants or implanted electronic devices such as cardiac pacemakers. Further, pregnant or currently breastfeeding women as well as individuals who had contraindications to undergoing MRI (e.g., people with end-stage renal failure, congestive heart failure, or psychiatric disorders) were excluded. None of the study participants had history of acute inflammatory or infectious diseases that required medical attention within 6 months prior to MRI.

All individuals underwent imaging at the Centre for Advanced Magnetic Resonance Imaging (CAMRI) at the University of Auckland. Height and weight, for calculation of BMI, were measured prior to scan acquisition with standardized protocols, with individuals wearing light clothing and without headgear and footwear. All scans were conducted on a 3 Tesla MAGNETOM Skyra scanner VE 11A (Siemens Healthineers, Erlangen, Germany). The same imaging protocol was used for all participants, who lay in the supine position and held their breath at end-expiration for 11 s. An axial T1-weighted volumetric interpolated breath-hold Dixon sequence was used (11). The parameters for the sequence included true form abdomen shim mode, field view 500 × 400 mm, echo time 2.46 ms and 3.69 ms, repetition time 5.82 ms, flip angle 9°, pixel bandwidth 750 Hz, signal average 1, slice thickness 5 mm, and matrix 512 × 410, with use of partial Fourier and parallel imaging with total acceleration factor 2.8. In-phase, out-of-phase, fat-only, and water-only images were created and exported from the scanner as DICOM files. In addition, MRS with a 20 mm × 20 mm × 20 mm voxel placed in the right lobe of the liver was used (to quantify liver fat) as previously described (12).

In the current study, we used a modified MR-opsy method (8) to quantify the regional intrapancreatic fat of participants. Two candidate slices located in the center of the pancreas were chosen from the out-of-phase images with use of MicroDicom software. ImageJ software (National Institutes of Health, Bethesda, MD) was then used to place regions of interest on the head, body, and tail regions of the pancreas (Fig. 1). Regions of interest were placed close to the corresponding pancreas diameters, as detailed in our comprehensive systematic literature review (13). The pancreas head diameter was the anteroposterior diameter measured in line with the right-most point of the confluence of the superior mesenteric and splenic veins. The pancreas body diameter was the greatest anteroposterior diameter in line with the left lateral border of the lumbar vertebrae. The pancreas tail diameter was a line perpendicular to the organ midline at a point 20 mm from the distal-most point of the pancreas in the slice. Attention was paid to avoid the inclusion of irrelevant organs and tissues, such as the duodenum, splenic vein, inferior vena cava, and visceral fat. The histogram feature of ImageJ was then used to quantify the fat content according to the signal intensity of the fat-only images. The same process was repeated for the two candidate slices, and the results were averaged. Two independent assessors measured the intrapancreatic fat of each participant, and the results of the two assessors were averaged. For evaluation of the concordance of measurements between the two assessors, intraclass correlation coefficients (ICC) were calculated. An ICC of >0.90 was deemed to signify excellent agreement (11).

Fasting venous blood samples were collected and centrifuged at 4°C. Aliquots were prepared and stored until use at −80°C. Blood analyses were conducted separately for each of the four individual cohorts, with use of the same laboratory methods. Hexokinase colorimetric assay was used to measure fasting plasma glucose, and boronate affinity chromatography assay was used to analyze HbA_1c. The MILLIPLEX MAP Human Metabolic Hormone Magnetic Bead Panel, based on Luminex xMAP technology (Merck KGaA, Hesse, Germany), was used to measure fasting insulin. HOMA of insulin resistance (HOMA-IR) was calculated with the following formula: fasting glucose (mg/dL) × fasting insulin (μU/mL) / 405. HOMA of β-cell function (HOMA-β) was calculated as follows: 360 × fasting insulin (μU/mL) / (fasting glucose (mg/dL) − 63) (14). Individuals in the highest tertiles of HOMA-IR and HOMA-β were deemed to have insulin resistance and insulin deficiency, respectively (15). With use of the American Diabetes Association criteria, individuals were categorized based on a single fasting blood sample according to diabetes status as follows: normoglycemia (HbA_1c <5.7% [<39 mmol/L] or fasting plasma glucose <100 mg/dL [<5.6 mmol/L]), prediabetes (HbA_1c 5.7–6.4% [39–47 mmol/L] or fasting plasma glucose 100–125 mg/dL [5.6–6.9 mmol/L]), or diabetes (HbA_1c ≥6.5% [≥48 mmol/mol] or fasting plasma glucose ≥126 mg/dL [≥7.0 mmol/L]) (16).

For assessment of differences in intrapancreatic fat deposition of the head, body, and tail of the pancreas, repeated-measures ANOVA was used. Data were presented as P and F values from the linear terms of the within-subjects contrasts. Where the overall regional difference was statistically significant, pairwise comparisons with Bonferroni adjustment were conducted and Bonferroni-adjusted P values were presented. Linear regression analyses were then used to compare differences in intrapancreatic fat deposition of the three pancreatic regions according to diabetes status (i.e., normoglycemia, prediabetes, and diabetes). Linear regression analyses were also used to compare differences in intrapancreatic fat deposition of the three pancreatic regions according to tertiles of three insulin traits (i.e., HOMA-IR, HOMA-β, and fasting insulin). Diabetes status and insulin traits were treated as independent variables, whereas regional intrapancreatic fat depositions in the head, body, and tail of the pancreas were treated as dependent variables. Data were presented as β-coefficients, SEs, and P values. Subgroup analyses were conducted according to the use of antidiabetes medications (for the purpose of treating diabetes). Four statistical models were built: unadjusted (model 1); adjustment for age, sex, and ethnicity (model 2); adjustment for age, sex, ethnicity, and BMI (model 3); and adjustment for age, sex, ethnicity, BMI, and liver fat (model 4). Statistical analyses were performed with SPSS, version 27.0, for macOS (IBM, Armonk, NY). P < 0.001 for the repeated-measures ANOVA analyses and P < 0.05 for the linear regression analyses were deemed statistically significant.

The data sets generated during the current study are available from the corresponding author on reasonable request. No applicable resources were generated or analyzed during the current study.

A total of 368 individuals met all of the eligibility criteria, of whom 159 were men (43.2%) and 209 were women (56.8%). Of participants, 121 individuals were European Caucasian (32.9%), 179 were Asian (48.6%), and 68 were of other ethnicities (18.5%). The latter-most category included (but was not limited to) 54 Māori individuals. There were 208 participants with normoglycemia (56.5%), 117 participants with prediabetes (31.8%), and 43 with type 2 diabetes (11.7%) in the study cohort. The latter-most category included 27 individuals with new-onset diabetes and 16 individuals with preexisting diabetes. Diabetes was treated with oral antidiabetes medications alone in 17 individuals and with insulin in 14 individuals. None of the participants had BMI <18 kg/m², and 88 (23.9%) participants had BMI ≥30 kg/m². Other characteristics of the study participants are presented in Table 1. There was excellent interassessor concordance of the intrapancreatic fat measurements, as the ICC of the total intrapancreatic fat percentage was 0.98 (95% CI, 0.97–0.98) (Fig. 2A), the ICC of the head fat percentage was 0.92 (95% CI 0.91–0.94) (Fig. 2B), the ICC of the body fat percentage was 0.90 (95% CI 0.88–0.92) (Fig. 2C), and the ICC of the tail fat percentage was 0.92 (95% CI 0.90–0.94) (Fig. 2D).

The minimum pancreatic head fat percentage was 2.4%, and the maximum was 14.5%. For the body region, the minimum and maximum were 2.3% and 14.4%, respectively, and for the tail region the minimum and maximum were 2.4% and 14.7%. Intrapancreatic fat variability (as evidenced by the SD) was 2.6%, 2.5%, and 2.4% in the head, body, and tail of the pancreas. Intrapancreatic fat across the three regions of the pancreas had a consistent pattern of associations with the studied covariates (Supplementary Table 1). In the unadjusted analysis, the overall difference in intrapancreatic fat deposition between the three regions of the pancreas was statistically significant (F = 286.027, P < 0.001). The post hoc pairwise comparisons showed a significant difference between the head and body regions (Bonferroni-adjusted P < 0.001), the head and tail regions (Bonferroni-adjusted P < 0.001), and the body and tail regions (Bonferroni-adjusted P < 0.001). After adjustment for covariates, none of the overall regional differences in intrapancreatic fat deposition were statistically significant (F = 1.258, P = 0.263, in model 2; F = 3.578, P = 0.059, in model 3; and F = 2.392, P = 0.123, in model 4).

Diabetes status explained 5.6% of the variance in pancreatic head fat. In the model 1 analyses, the head region of the pancreas had significant differences in fat percentage in comparing of both prediabetes with normoglycemia (β = 1.185, P < 0.001) and diabetes with normoglycemia (β = 1.315, P = 0.002). In the model 2, model 3, and model 4 analyses, the head region of the pancreas had no significant differences in fat percentage in comparisons of the different diabetes statuses (Table 2). Use of antidiabetes medications did not materially influence the results (Supplementary Table 2).

Diabetes status explained 6% of the variance in pancreatic body fat. In the model 1 analyses, the body region of the pancreas had significant differences in fat percentage in comparisons of both prediabetes with normoglycemia (β = 1.167, P < 0.001) and diabetes with normoglycemia (β = 1.431, P = 0.001). In the model 2 analyses, the body region of the pancreas had significant differences in fat percentage in comparisons of prediabetes with normoglycemia (β = 0.629, P = 0.029). However, in comparisons of diabetes with normoglycemia, the differences were not statistically significant (Table 2). In the model 3 and model 4 analyses, the body region of the pancreas had no significant differences in fat percentage in comparisons of the different diabetes statuses (Table 2). The use of antidiabetes medications did not materially influence the results (Supplementary Table 2).

Diabetes status explained 7% of the variance in pancreatic tail fat. In the model 1 analyses, the tail region of the pancreas had significant differences in fat percentage in comparisons of both prediabetes with normoglycemia (β = 1.221, P < 0.001) and diabetes with normoglycemia (β = 1.480, P < 0.001). In the model 2 analyses, the tail region of the pancreas had significant differences in fat percentage in comparisons of prediabetes with normoglycemia (β = 0.671, P = 0.014). However, in comparisons of diabetes with normoglycemia, the differences were not statistically significant (Table 2). In the model 3 and model 4 analyses, the tail region of the pancreas had no significant differences in fat percentage in comparisons of the different diabetes statuses (Table 2). The use of antidiabetes medications did not materially influence the results (Supplementary Table 2).

HOMA-IR explained 5.4% of the variance in pancreatic head fat. In the model 1 analyses, the head region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 0.870, P = 0.015) and tertile 3 with tertile 1 (β = 1.477, P < 0.001). In the model 2 analyses, the head region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 0.732, P = 0.025) and tertile 3 with tertile 1 (β = 1.245, P < 0.001). In the model 3 and model 4 analyses, the head region of the pancreas had no significant differences in fat percentage in comparisons of the HOMA-IR tertiles (Table 3). The use of antidiabetes medications did not materially influence the results (Supplementary Table 3).

HOMA-IR explained 7.4% of the variance in pancreatic body fat. In the model 1 analyses, the body region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 1.154, P = 0.001) and tertile 3 with tertile 1 (β = 1.690, P < 0.001). In the model 2 analyses, the body region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 1.084, P = 0.001) and tertile 3 with tertile 1 (β = 1.604, P < 0.001). In the model 3 analyses, the body region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 0.637, P = 0.037) and tertile 3 with tertile 1 (β = 0.746, P = 0.021). In the model 4 analyses, the body region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 0.634, P = 0.044) and tertile 3 with tertile 1 (β = 0.713, P = 0.033). In stratification of the data for antidiabetes medication use, results of the models 1–4 analyses were statistically significant only for the individuals who did not use antidiabetes medications (Supplementary Table 3).

HOMA-IR explained 6.7% of the variance in pancreatic tail fat. In the model 1 analyses, the tail region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 1.091, P = 0.002) and tertile 3 with tertile 1 (β = 1.566, P < 0.001). In the model 2 analyses, the tail region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 1.022, P = 0.001) and tertile 3 with tertile 1 (β = 1.485, P < 0.001). In the model 3 analyses, the tail region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 0.684, P = 0.023) and tertile 3 with tertile 1 (β = 0.810, P = 0.011). In the model 4 analyses, the tail region of the pancreas had significant differences in fat percentage in comparisons of both tertile 2 with tertile 1 (β = 0.631, P = 0.042) and tertile 3 with tertile 1 (β = 0.786, P = 0.017). In stratification of the data for antidiabetes medication use, results of the models 1–4 analyses were statistically significant only for the individuals who did not use antidiabetes medications (Supplementary Table 3).

HOMA-β explained 0.3% of the variance in pancreatic head fat. In the model 1, model 2, model 3, and model 4 analyses, the head region of the pancreas had no significant differences in average fat percentage in comparisons of both tertile 2 with tertile 1 and tertile 3 with tertile 1 (Table 3). The use of antidiabetes medications did not materially influence the results (Supplementary Table 3).

HOMA-β explained 0.2% of the variance in pancreatic body fat. In the model 1, model 2, model 3, and model 4 analyses, the body region of the pancreas had no significant differences in average fat percentage in comparisons of both tertile 2 with tertile 1 and tertile 3 with tertile 1 (Table 3). The use of antidiabetes medications did not materially influence the results (Supplementary Table 3).

HOMA-β explained 0.1% of the variance in pancreatic tail fat. In the model 1, model 2, model 3, and model 4 analyses, the tail region of the pancreas had no significant differences in average fat percentage in comparisons of both tertile 2 with tertile 1 and tertile 3 with tertile 1 (Table 3). The use of antidiabetes medications did not materially influence the results (Supplementary Table 3).

See Supplementary Material.

This study of 368 individuals is one of the largest prospective general population–based studies on intrapancreatic fat determined with the use of MRI and also the largest study to investigate the associations between regional intrapancreatic fat and insulin traits. Besides, the study improved on previous studies by strictly adhering to the American Diabetes Association criteria for diagnosing diabetes and prediabetes (16). Three main findings resulted from the current study. First, in the overall cohort, there were no significant differences in fat deposition of the three regions of the pancreas after adjustment for covariates. Second, there were no significant differences in regional intrapancreatic fat according to diabetes status after adjustment for covariates. Third, indices of insulin resistance (HOMA-IR and fasting insulin) were associated with intrapancreatic fat in the tail and body regions in both unadjusted and adjusted analyses. By contrast, there was no significant association between regional intrapancreatic fat and index of insulin secretion (HOMA-β) in either unadjusted or adjusted analyses. Further, HOMA-β explained 0.1% of the variance in pancreatic tail fat, whereas HOMA-IR explained 6.7% of it.

Only a handful of studies in the literature both had a large sample size (at least 200 participants) and used robust manual quantification of intrapancreatic fat based on MRI—the gold standard for noninvasive quantification of this fat depot (1). Our finding of no significant differences in fat deposition in the three regions of the pancreas is in line with the results of a 2017 German prospective study of 385 individuals, similarly categorized into three subgroups by diabetes status (17). Both studies were population based (hence, limiting the possibility of a selection bias), and in both a single 3 Tesla scanner was used for determination of intrapancreatic fat in all participants. Our finding contradicts the results of a 2020 U.S. retrospective study of 276 individuals where significant differences in intrapancreatic fat in the three regions were found (10). However, that study did not adjust for covariates and represented a retrospective analysis of a multihospital database of scans (from different scanners that used varying magnetic field strengths). Previous studies showed that age, sex, ethnicity, and BMI can influence body fat deposition (18,19), including a 2015 German prospective study where investigators found that both age (P < 0.001) and BMI (P < 0.001) were positively associated with intrapancreatic fat (20). This emphasizes the importance of analyses of intrapancreatic fat that account for these covariates in order to yield unbiased results. In our study, we used four progressive statistical models throughout each analysis, with age, sex, ethnicity, BMI, and liver fat accounted for in the model with most adjustments. The results of our study also further improve on the findings of the 2020 U.S. study, as their technique was markedly different from that of the MR-opsy, which had been shown to be a highly accurate method of quantifying intrapancreatic fat (8). Due to the potential for the three MR-opsy regions of interest to be placed in a number of possible locations within each of the three pancreatic regions, this technique of quantifying intrapancreatic fat could potentially be unreliable if there was considerable heterogeneity of intrapancreatic fat (similar to what appears to be characteristic of liver fat) (21). Results from our MRI study have established, however, that there are no significant regional differences in intrapancreatic fat. This finding is in alignment with a histopathology and autopsy study of the pancreas in 50 people (including 7 with diabetes) shortly after death, with authors reporting no significant differences in fat deposition in the three regions of the pancreas (22). These congruent findings lend further support for the use of the MR-opsy method as the designated choice for noninvasive quantification of intrapancreatic fat. Also notable is that during the quantification of intrapancreatic fat in the current study, we used the average measurements of two assessors in order to yield robust results, which is another improvement from the 2020 U.S. and 2017 German studies (which only included a second assessor for a subset of the sample [10,17]) as well as the 2015 German study (which only included one assessor [20]). Also, the interassessor reliability of our measurements was determined using ICCs, which were between 0.90 and 0.92 for all three regions (deemed to signify excellent agreement) (11). This compares favorably with the findings of the 2020 U.S. study, which reported ICCs of only 0.58, 0.66, and 0.70 for the head, body, and tail, respectively (10).

Our finding of no significant differences in regional intrapancreatic fat according to diabetes status on adjusted analysis suggests that a relationship between regional intrapancreatic fat deposition and diabetes status is unlikely. Our results are consistent with results of the above 2015 German population-based study of 1,241 individuals with no significant association found between intrapancreatic fat and diabetes status (20). As that study was based in a particular region of Germany (20), its findings might not be generalizable to populations outside of Western Europe. Our study is based on a more ethnically diverse cohort that included 33% European Caucasian and 49% Asian participants. Also important is that 12% of participants in our cohort had type 2 diabetes defined according to the American Diabetes Association criteria (16). This prevalence is similar to the overall prevalence of diabetes in the general populations of the U.S. and China, which have estimates of 13% and 11% of adults having type 2 diabetes, respectively (23,24). By contrast, diabetes status was self-reported in the 2015 German study and the prevalence of diabetes in the study cohort was 5.6% (20). Also notable is that the German study was conducted with a 1.5 Tesla MRI scanner (20), which provides lower spatial resolution (and, correspondingly, lower accuracy of intrapancreatic fat measurements) than the 3 Tesla scanner used in the current study (25).

To the best of our knowledge, there is no other population-based study with specific examination of the relationship between insulin traits and regional intrapancreatic fat (26). On analysis of HOMA-IR and fasting insulin, we found consistent significant positive associations (in both unadjusted and adjusted analyses) between these indices of insulin resistance and intrapancreatic fat in the tail and body of the pancreas (but not the head of the pancreas). A 2020 animal study demonstrated significantly increased insulin resistance (as evidenced by serum insulin) in the mice that had their peripancreatic fat surgically removed (vs. sham controls) prior to a 16-week-long high-fat feeding, suggesting a negative association between insulin resistance and peripancreatic fat (27). These mice were also characterized by significantly increased hepatic fat but not intrapancreatic fat. However, although peripancreatic fat and intrapancreatic fat are closely located, the two depots differ both structurally and functionally (27,28). Also, there are numerous morphological differences between human and mouse pancreas (29). In one human MRI study, investigators found a positive association between insulin resistance and total intrapancreatic fat (30). Specifically, they reported that individuals with high HOMA-IR had greater intrapancreatic fat than insulin-sensitive individuals. The results from our study suggest that the pathophysiological processes of increased intrapancreatic fat deposition and insulin resistance may occur initially in the tail and body of the pancreas rather than the head. The tail of the pancreas was also observed as the first region to shrink in people with type 1 diabetes (31). Given that the highest density of islets was reported in the tail of the pancreas (32), we consider that β-cell dysfunction may initialize in the tail. This has clinical relevance, as individuals who are identified with progressive increase in intrapancreatic fat from the tail to the body and then the head regions of the pancreas on successive scans may be at risk for developing insulin resistance and could therefore benefit from screening and follow-up for diseases that stem from increased insulin resistance. The relevance to prevention or early disease detection is further supported by the findings of our subgroup analysis (Supplementary Table 3), which showed that the above associations held true only for individuals who did not receive any antidiabetes medication.

From our analysis of HOMA-β, there was no significant association of this index of insulin secretion with regional intrapancreatic fat. Most previously published studies on intrapancreatic fat and insulin secretion cannot be compared with the current study in a meaningful way, as, apart from being much smaller, those studies either used inferior methods (such as MRS [33–37] or computed tomography [38,39]) for noninvasive measurement of intrapancreatic fat in humans or were conducted in animals (40). Our results on HOMA-β corroborate the findings of a population-based MRI study with no significant association of HOMA-β with fatty pancreas disease (2). That study, however, included only Hong Kong Chinese, only adjusted for two covariates, and did not investigate regional intrapancreatic fat (2). Our study of regional intrapancreatic fat improves on that study with involvement of an ethnically diverse population and adjustment for five key covariates. Two smaller MRI studies of the associations between total intrapancreatic fat and indices of insulin secretion (derived from an oral glucose tolerance test) had conflicting findings (41,42). The results of one study (n = 116) did not show a significant association (41), whereas the results of the other study (n = 51) did (42). It is worth noting that, while the former study used statistical models with adjustment for age and body composition, the latter did not. Last, the findings of the current study are in line with the results of a 2017 human histopathology study with no significant correlation between intrapancreatic fat and β-cell mass (as well as no differences between the three regions of the pancreas in terms of intrapancreatic fat) (43).

There are several limitations of the current study that should be acknowledged. First, as this was a cross-sectional study, it was not possible to draw conclusions about causality between intrapancreatic fat and the studied variables. As, to date, there have been no prospective longitudinal studies of intrapancreatic fat (determined with the use of MRI) and insulin traits (1), such studies will be necessary in the future to investigate whether the significant relationship between intrapancreatic fat in the tail/body and insulin resistance observed in the current study is causal. Second, while we adjusted our analyses for five covariates (which makes the current study one of the most thoroughly adjusted studies on intrapancreatic fat determined with the use of MRI [1]), residual confounding might have affected the studied associations. Unfortunately, this type of bias is known to bedevil observational studies and only randomized controlled trials are positioned to negate it fully. Third, genetic factors were shown to be associated with body fat distribution but were not investigated in the current study. However, to date, there has been no study to conclusively demonstrate a genetic predisposition specifically to high intrapancreatic fat. In fact, results of a 2020 genome-wide association study showed no significant association between genome-wide polygenic risk score and total intrapancreatic fat and investigators of a 2021 genome-wide association study did not find a significant association between 36 favorable variants and total intrapancreatic fat (44,45). Fourth, as the present population-based study was conducted in New Zealand, most of the study participants were of European or Asian descent or the indigenous people of New Zealand (1). People of Black African descent may need to be specifically included in future studies, as they may have a different pattern of association between intrapancreatic fat and insulin traits than other racial groups (46). Fifth, data on habitual dietary intake and physical activity were not available. However, while these factors are known to affect deposition of fat in various organs, to the best of our knowledge there has been no study to demonstrate that these factors affect differentially fat deposition in the three regions of the pancreas. Sixth, given that the current study only included participants of at least 18 years of age, it was not possible to extrapolate our findings to pediatric settings. As intrapancreatic fat increases throughout childhood (1), it is possible that regional intrapancreatic fat distribution in children is different, and further studies of pediatric populations will be necessary to investigate this. Seventh, HOMA-IR and HOMA-β are rather simple measures of insulin resistance and insulin secretion, reflecting largely hepatic insulin resistance and basal insulin secretion. The gold standard for determining insulin traits, hyperglycemic-euglycemic clamp, was not used in the current study. This was because only the use of proxies was feasible in this population-based study of hundreds of participants who underwent MRI. Last, although the MR-opsy technique used in the current study is considered to be a highly reliable method of quantifying pancreatic fat (8), the gold standard is still histopathology. However, biopsying the pancreas is not an easily accessible option and an in vivo study conducted in this way would have been very difficult (if not unethical) in a general population setting.

In conclusion, there were no significant differences in intrapancreatic fat deposition in the head, body, and tail regions of the pancreas after adjustment for covariates. This held true consistently for individuals with diabetes, prediabetes, and normoglycemia. Intrapancreatic fat in the tail and body regions showed a significant positive association with indices of insulin resistance (consistently in both unadjusted and adjusted analyses), which likely implicates the tail region of the pancreas in the initial development of insulin resistance and ensuing disorders.

Funding. This study was funded by the Royal Society of New Zealand (Rutherford Discovery Fellowship to M.S.P.) and the New Zealand High-Value Nutrition National Science Challenge, Ministry for Business, Innovation and Employment (grant 3710040).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. I.R.S., J.C., and J.K. contributed to patient recruitment. L.S.-H., I.R.S., and J.K. contributed to data acquisition. L.S.-H. and J.C. contributed to analysis and interpretation of data. L.S.-H. drafted the manuscript. I.R.S., J.C., J.K., S.D.P., and M.S.P. contributed to revision of the manuscript. M.S.P. supervised the study. All authors read and agreed to the submitted version of the manuscript. M.S.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.