Female-specific pancreatic cancer survival from CT imaging of visceral fat implicates glutathione metabolism in solid tumors

Structured Abstract

Rationale and Objectives

To identify if body composition, assessed with preoperative CT-based visceral fat ratio quantification as well as tumor metabolic gene expression, predicts sex-dependent overall survival (OS) in patients with pancreatic ductal adenocarcinoma (PDAC).

Materials and Methods

This was a retrospective analysis of preoperative CT in 98 male and 107 female patients with PDAC. Relative visceral fat (rVFA; visceral fat normalized to total fat) was measured automatically using software and corrected manually. Median and optimized rVFA thresholds were determined according to published methods. Kaplan Meier and log-rank tests were used to estimate OS. Multivariate models were developed to identify interactions between sex, rVFA, and OS. Unsupervised gene expression analysis of PDAC tumors from The Cancer Genome Atlas (TCGA) was performed to identify metabolic pathways with similar survival patterns to rVFA.

Results

Optimized preoperative rVFA threshold of 38.9% predicted significantly different OS in females with a median OS of 15 months (above threshold) vs 24 months (below threshold; p=0.004). No significant threshold was identified in males. This female-specific significance was independent of age, stage, and presence of chronic pancreatitis (p=0.02). Tumor gene expression analysis identified female-specific stratification from a five-gene signature of glutathione S-transferases. This was observed for PDAC as well as clear cell renal carcinoma and glioblastoma.

Conclusion

CT-based assessments of visceral fat can predict pancreatic cancer OS in females. Glutathione S-transferase expression in tumors predicts female-specific OS in a similar fashion.

Introduction

Of all cancers, pancreatic cancer has the seventh highest incidence with an estimated 57,600 diagnoses and the fourth highest mortality with an estimated 50,550 deaths in 2023 (1). Pancreatic ductal adenocarcinoma (PDAC) is a histologic subtype of pancreatic cancer that accounts for more than 90% of all pancreatic cancers (2) and is an aggressive cancer with a 5 year relative survival of 12% (1). Over the past decades, the mortality due to pancreatic cancer has been relatively unchanged despite advances in cancer treatments (1). Because of the lethal progression of this disease and the inability to achieve significant increases in survival, new avenues need to be explored that delve into mechanisms that underlie disease progression and that may aid in better prognostic indicators and druggable targets. Sex differences in cancer outcomes may provide important insights. In fact, the National Institutes of Health has issued a directive to incorporate sex as a biological variable into all phases of basic, translational and clinical investigation (3), and there are as yet no standardized approaches.

In multiple cancers throughout the body, males are characterized by both increased incidence and mortality compared to females (1, 4). This is true of pancreatic cancer, where males have not only increased incidence across the globe, but increased mortality (1, 5). Although this phenomenon is not completely understood, recent developments suggest that metabolism, a hallmark of tumorigenesis, may play an important role (6, 7). Not only are there sex differences in nutrient utilization that are present across all stages of development (8–11), but there are also sex differences in body composition. Males carry more visceral fat and premenopausal females carry more subcutaneous fat (7, 12–14). Visceral fat, in particular, is important in the context of cancer incidence and prognosis. Several studies have shown that visceral fat by virtue of its auto- and paracrine signaling plays a role in carcinogenesis and tumorigenesis, particularly in pancreatic adenocarcinoma where it may impact outcomes after surgical resection (15–20). However, this is complicated by the fact that while some previous work indicates that increased visceral fat may worsen outcomes in pancreatic cancer, other work suggests that severely decreased visceral fat can result in worsened outcomes (21, 22). This discrepancy merits further investigation to determine if a sex-specific analysis can provide a clearer picture given the already understood sex differences in adiposity. Further, to control for the potential effects of subcutaneous fat, which has been shown to be antagonistic to the metabolic effects of visceral fat in diabetes (23, 24), a measure that better isolates the effect of visceral fat should be used in that analysis.

Previous work has demonstrated differences in sex-specific overall survival with renal cell carcinoma and diffuse large B cell lymphoma CT-based relative visceral obesity measurements. In those studies, females with increased “male-pattern” visceral fat percentages had significantly worse survival that all other groups (7, 31). Moreover, sex-specific stratification of survival with tumor metabolism itself has also been observed in multiple solid tumors by our group as well using metabolic gene expression signatures that could identify metabolic pathway differences in male and female tumors and their association with survival (6, 32, 33). In this study, we wanted to determine if these techniques could be extrapolated to shed light on patients with PDAC using computed tomography (CT) based visceral fat quantification on a homogeneous, single-institutional cohort as well as potential connections to tumor metabolism using an independent cohort from The Cancer Genome Atlas (TCGA). Here, we present data further demonstrating the utility of using visceral fat and tumor metabolism to stratify cancer patient outcomes on a sex-dependent basis.

Materials and Methods

Software

The license for the fat assessment tool was provided by Vital images, but the data for the study were obtained independently by the authors and are under their control for publication.

Imaging Datasets

The retrospective single center study was approved by our respective institutional review board (IRB). Inclusion criteria included: (i) treatment-naive patients, (ii) diagnosis of PDAC in a location suitable for a Whipple procedure (localized to the pancreatic head) (iii) a preoperative pancreas protocol CT examination at our institution prior to surgical resection, (iv) presence of either R0 or R1 surgical resection margins, and (v) with at least five years of follow-up data or follow-up until death. These criteria were selected to provide a long-term homogenous dataset of patients and to ensure that all necessary pathologic, clinical and imaging data was available for the analysis. Six patients were excluded from this analysis due to insufficient field of view that precluded accurate quantification of fat (more than 10% of the body wall collimated at the level of the umbilicus). In total, 205 patients (98 male, 107 female) were analyzed by imaging segmentation from preoperative CT examinations and clinical data.

Clinical, Surgical, and Pathologic Data

Clinical, surgical, and pathologic data from the 205 study patients described above were obtained from a retrospective review of a prospectively maintained institutional database. Clinical data used in this study included patient sex, age at diagnosis, height, weight, body mass index (BMI), length of postoperative follow-up, and overall survival (OS). Surgical and pathologic data included pathologic stage, surgical margins, maximum tumor diameter, perineural and/or venous invasion, TNM stage, and evidence of chronic pancreatitis on histopathology. CT imaging data of the primary tumor were not used to assess these or any other characteristics, and imaging was solely used for fat segmentation. Preoperative serum CA19–9 levels, a clinical biomarker for PDAC, were available in 143 patients (79 males, 64 females).

CT-based Fat Segmentation

Fat segmentation was performed as previously published (7, 25). CT examinations in Digital Imaging and Communications in Medicine format were transferred to a workstation equipped with the Vitrea Fat Measurement Application (Vital Images, Minnetonka, Minn). Subcutaneous and visceral fat area (SFA and VFA) at the level of the umbilicus were mapped by using thresholds from −150 to −50 HU. The level of the umbilicus was manually selected, and the subcutaneous and visceral fat were initially identified by the Vitrea software. Errors from software-defined areas were corrected manually and overseen by one of the authors, a radiologist subspecializing in abdominal imaging. Errors included incorrect contours of the subcutaneous and visceral fat and identifying fat attenuation in colonic stool, epidural fat, and bone marrow fat as visceral fat (these latter three areas were manually excluded from visceral fat after manual corrections). Total fat area (TFA) was summed from absolute SFA and VFA. Relative VFA (rVFA) and relative SFA (rSFA) were calculated as percentage of TFA (i.e., rVFA = VFA/TFA). By using this method, rVFA and rSFA become complementary (rVFA + rSFA = 100%), and any analyses that investigate both of these normalized quantities are redundant. The rVFA therefore represents a normalized fat ratio that takes into account potential confounding effects from subcutaneous fat and overall adiposity (7)(Figure 1).

Figure 1. Case examples of women above (A-C) and below (D-F) the relative visceral fat area (rVFA) of 38.9%, with high rVFA in women associated with significantly worse OS.

A-C. 74-year-old woman with a BMI of 25 who initially presented with obstructive jaundice managed with biliary stenting, and now undergoing multiphase pancreas protocol CT. Contrast-enhanced obliqued coronal (A) axial (B, C) CT images at the level of the pancreatic head (A, B) and umbilicus (C). An ill-defined pancreatic ductal adenocarcinoma head mass (black arrows in A and B) measuring up to 2.7 cm causes biliary and pancreatic duct obstruction (yellow arrowheads in A) managed by a common bile duct stent. The patient had a rVFA of 40.1% (above the 38.9% cutoff), note the proportion of visceral fat and subcutaneous fat (represented as VF and SF) at both the level of the pancreatic mass (B) and the study level/level of fat quantification (C). The patient had an overall survival of 2.2 months.

D-F. 73-year-old woman with a BMI of 31 who presented with obstructive jaundice undergoing characterization of a previously imaged pancreatic mass with multiphase pancreas protocol CT. Contrast-enhanced obliqued coronal (D) axial (E, F) CT images at the level of the pancreatic head (D, E) and umbilicus (F). An ill-defined pancreatic ductal adenocarcinoma head mass (black arrows in D and E) measuring up to 2.6 cm causes biliary and pancreatic duct obstruction (yellow arrowheads in D) managed by a common bile duct stent. The patient had a rVFA of 24.2% (below the 38.9% cutoff). The patient had an overall survival of 117 months.

In C and F, the blue area represent subcutaneous fat (SF C and F) in and the red area represents visceral fat (VF C and F) (after correction/manual removal of fat attenuation in colonic stool and epidural fat).

Statistical Analysis of CT-based Segmentation

Continuous variables were reported using median (range) and categorical variables as proportions unless otherwise specified. Comparisons between sexes for continuous and categorical variables were performed using the Mann-Whitney test or chi squared test (or Fisher’s exact testing for smaller cell counts), respectively. Pathologic data that were missing from pathology reports in a subset of patients (i.e. lymphatic invasion, venous invasion, and perineural invasion; See Table 1) were labeled as “cannot be assessed” and were not used in the statistical analysis. Correlations between metabolic metrics were assessed using Pearson correlation coefficient (r). R software-based biomarker cutoff optimization algorithm was used to determine the optimal rVFA value required to maximally stratify male and female overall survival (OS)(30). OS was analyzed using the Kaplan-Meier method and differences were assessed using the log-rank test.

Table 1.

Clinical, surgical, and pathologic characteristics.

Variable	Total n = 205	Male n = 98	Female n = 107	P-value
Age (y)	67.0[36.0–86.9]	65.7[36.0–84.5]	67.8[37.5–86.9]	0.331
Race				0.199
White	189 (92.2)	96 (89.7)	93 (94.9)
Black	16 (0.8)	11 (10.3)	5 (5.1)
Tumor size (cm)	2.8 [1.0–8.0]	3.0[1.1–8.0]	2.8[1.0–7.0]	0.648
Preoperative CA19-9 (U/mL)	199.5[10.0–15,255.0]	143.0[11.0–10,951.0]	299.3[10.0–15255.0]	0.032
Grade				0.099
Well differentiated	3 (1.5)	0 (0)	3 (3.1)
Moderately differentiated	89 (43.6)	51 (49.1)	38 (38.8)
Poorly Differentiated	112 (54.9)	55 (51.9)	57 (58.1)
Surgical margins				0.450
Free of tumor	103 (50.2)	57 (53.2)	46 (47.0)
< 1 mm	37 (18.0)	16 (15.0)	21 (21.4)
Positive	65 (31.7)	34 (31.8)	31 (31.6)
Lymphatic invasion				0.502
Cannot be assessed	8 (4)	4 (4)	4 (4)
No	56 (27)	29 (30)	27 (25)
Suspicious	1 (0)	0 (0)	1 (1)
Yes	140 (68)	65 (66)	75 (70)
Venous invasion				0.769
Cannot be assessed	8 (4)	4 (4)	4 (4)
No	73 (36)	36 (37)	37 (35)
Yes	124 (60)	58 (59)	66 (62)
Perineural invasion				0.818
Cannot be assessed	6 (3)	4 (4)	2 (2)
No	21 (10)	9 (9)	12 (11)
Yes	178 (87)	85 (87)	93 (87)
T stage				0.642
T1	4 (2)	3 (3)	1 (1)
T2	11 (5)	6 (6)	5 (5)
T3	190 (93)	98 (92)	92 (94)
N stage				0.604
N0	42 (20)	20 (19)	22 (22)
N1	163 (80)	87 (81)	76 (78)
Stage				0.867
IA	2 (1)	1 (1)	1 (1)
IB	4 (2)	2 (2)	2 (2)
IIA	36 (18)	18 (17)	18 (19)
IIB	161 (79)	86 (80)	75 (77)
III	1 (0)	0 (0)	1 (1)
Evidence of chronic pancreatitis				0.649
No	63 (32)	34 (33)	29 (30)
Yes	136 (68)	68 (67)	68 (67)

Values presented as median [range] or N (%).

Cox proportional hazards regression was used to assess the risk the factors of interest (e.g., rVFA) after controlling for age, evidence of chronic pancreatitis and stage. Assumptions for the proportional hazard regression were verified using cumulative sums of martingale-based residuals.

Statistical analyses were performed by using Prism 5.04 (GraphPad Software, La Jolla, Calif) and SAS v9.4 (SAS Institute, Cary, NC) software. Two-tailed statistical tests were performed where applicable, and p ≤ 0.05 was considered to indicate a statistically significant difference.

Molecular Analysis of Solid Tumors

The results published here are in whole or part based upon data generated by The Cancer Genome Atlas (TCGA) Research Network (http://cancergenome.nih.gov/). Normalized RNA gene expression data (RNA-Seq by Expectation-Maximization; RSEM) of pancreatic adenocarcinoma tumors (26) were downloaded from TCGA (TCGA-PAAD) using the cBioPortal for Cancer Genomics (https://www.cbioportal.org/)(27, 28). In total, there were 177 datasets (97 male, 80 female). In addition, data from clear cell renal cell carcinoma (TCGA-KIRC; 325 male, 183 female) and glioblastoma (TCGA-GBM; 63 male, 43 female) RSEM gene expression values were transformed to sex-specific Z-scores, as previously published (6, 7). Metabolic pathway data were obtained from the Kyoto Encyclopedia of Genes and Genomes (KEGG; https://www.genome.jp/kegg/). Gene symbols of enzymes within the following metabolic superpathways were downloaded: Carbohydrate Metabolism (350 genes), Lipid Metabolism (388 genes), and Amino Acid Metabolism (360 genes).

To identify metabolic pathways whose overall expression resulted in sex-specific outcomes, our previously published algorithm was used (6). Briefly, hierarchical clustering was performed based on the (1 - pairwise spearman correlation) as the distance matrix, and the dendrogram was divided into two groups (29). This algorithm was applied to the z-transformed RNASeq data on a sex-specific basis and the two groups of male and female patients obtained from each analysis were plotted on Kaplan-Meier curves with significance obtained by the logrank test.

Genes that were significantly enriched in the female amino acid clusters were based upon a two tailed Mann-Whitney U test comparison of RSEM values between both clusters. Genesets that were significantly enriched in each cluster were analyzed using the Gene Set Enrichment Analysis web-based tool (GSEA; https://www.gsea-msigdb.org/) and the canonical KEGG pathway gene set. Heatmaps were visualized using Morpheus (https://software.broadinstitute.org/morpheus). Characterization of survival patterns from the five glutathione S-transferases was performed using a method previously published by our group (6). Briefly, sex-specific Z-scores were scaled from Z=1 to Z=2 in 0.25 step increments individually for each gene and logrank tests were performed to identify significance. Data were plotted on Kaplan-Meier curves. For each cancer type, a Z-score threshold to divide the low vs high GST groups was selected based upon the maximum number of GST enzymes that were significant for that threshold (i.e., Z=2 for TCGA-PAAD, Z=1.75 for TCGA-KIRC, and Z=1 for TCGA-GBM).

Results

Patient Characteristics

For the CT-based analysis portion of the study, the population consisted of 98 males and 107 females with a mean age of 66.3 ± 10.0 years. Baseline patient and clinical-pathologic characteristics stratified by sex are summarized in Table 1. Out of all the clinical variables, the only variable that was statistically significant between the sexes was the preoperative CA19–9. Females had a higher median CA19–9 (299.3 U) compared to males (143.0 U; p=0.032). When the effects of clinical variables (including sex) on OS were investigated, there was no difference in the median OS between males (median OS = 15.6 months) and females (median OS = 18.0 months; p = 0.457). However, there were significant differences in OS in patients with positive nodes at surgery (median OS = 17.4 months) compared to node negative patients (median OS = 47.9 months; p = 0.003). In addition, patients with evidence of chronic pancreatitis on surgical pathology had a significantly better OS (median OS = 24.0 months) compared to those without chronic pancreatitis (median OS = 13.9 months; p=0.013).

Visceral Fat Analysis

First, the effects of rVFA on OS was evaluated among male and female PDAC patients separately. Using the median rVFA as a threshold value, which for females was 32.5%, only females displayed significant differences in OS. Females with rVFA greater than 32.5% had a median OS of 19 months compared to females with rVFA less than 32.5% with a median OS of 27 months (p=0.006). In comparison, males were not significantly stratified by the median rVFA (p=0.562). Optimized thresholding was also performed to identify maximum differences in OS. The optimized rVFA threshold for females (38.9%) displayed similar results; females with rVFA greater than 38.9% had a median OS of 15 months compared to females with rVFA less than rVFA with a median OS of 26 months (p=0.004). Differences in OS for males were not significant despite optimized thresholding (p=0.065; Figure 2).

Figure 2. Comparison of optimized and median rVFA thresholds on PDAC OS.

A-B. Median rVFA threshold calculations to stratify male and female patients demonstrating improved stratification of females, but not males. C-D. Optimized rVFA threshold calculations used as a comparison to median thresholds demonstrating retention of female-specific stratification. P-value is derived from the log-rank test. OS is Overall Survival. PDAC is pancreatic ductal adenocarcinoma. rVFA is relative visceral fat area.

The rVFA was then compared to the BMI. As opposed to rVFA, there were no sex differences in BMI (Table 2). In addition, neither median nor optimized thresholds for the BMI identified significant differences in OS in males or females (p=0.257 for males, p=0.197 for females).

Table 2.

Fat quantification data derived from preoperative multiphase pancreas-protocol abdominal CT.

Variable	Total n = 205	Male n = 98	Female n = 107	P-value
BMI (kg/m2)	26.7[13.5–49.4]	26.3[19.0–38.1]	27.2[13.5–49.40	0.809
SFA (cm2)	244.3[4.4–817.2]	224.5[25.9–542.1]	267.4[4.4–817.2]	0.032
VFA (cm2)	153.0[11.2–437.7]	180.2[18.3–437.7]	133.3[11.2–333.9]	<0.001
TFA (cm2)	406.1[26.4–1141.0]	397.1[44.2–875.6]	407.3[26.4–1141.0]	0.889
rVFA (%)	38.7[15.0–76.6]	43.1[15.9–69.4]	32.5[15.0–76.6]	<0.001

Values presented as median [range].

Because of the sex difference in preoperative CA19–9 (Table 1), correlations between CA19–9 and rVFA were also tested. No significant correlations were found in either males (r=0.183; p=0.107) or females (r=−0.110, p=0.388).

A multivariate model was developed that adjusted for nodal stage, chronic pancreatitis, tumor diameter, and age. The optimized rVFA threshold of 38.9% remained a significant predictor for OS in females. There was a striking 209% increase in the risk of death in women with rVFA >38.9%. However, rVFA did not independently predict increased risk of death in males. This sex-related difference was further highlighted in models that showed that there was a significant interaction between sex and rVFA (p=0.022, Table 3).

Table 3.

Cox Proportional Hazards model for N stage, evidence of chronic pancreatitis, maximum tumor diameter, age, and optimized rVFA thresholds.

Risk factor	Hazard Ratio	95% Hazard Ratio Confidence Limits	P-value
N stage			0.036
0 (reference)	-	-
1	1.634	(1.047, 2.632)
Evidence of Chronic Pancreatitis			0.003
Yes (reference)	-	-
No	1.740	(1.209, 2.480)
Max tumor diameter (cm)	1.252	(1.068, 1.457)	0.005
Age	1.017	(0.999, 1.034)	0.060
rVFA (optimized threshold)			0.022
for Females	2.085	(1.200, 3.536)
for Males	0.910	(0.558, 1.486)

Tumor Metabolic Gene Expression Analysis

Female-specific stratification with rVFA has also been observed in clear cell renal carcinoma as diffuse large B-cell lymphoma by our group (7, 31). Moreover, sex-specific stratification with tumor metabolism has also been observed in multiple solid tumors by our group as well (6, 32, 33). We wanted to determine if there were pancreatic cancer metabolic gene expression signatures that could stratify female pancreatic cancer patients in a similar manner. Gene expression signatures from three major nutrient metabolic superpathways (carbohydrate metabolism, 350 genes; lipid metabolism, 388 genes; and amino acid metabolism, 360 genes) with associated OS data from male and female PDAC tumor were downloaded from The Cancer Genome Atlas (TCGA-PAAD). Unsupervised hierarchical clustering of gene expression was performed or each of the three metabolic superpathways in males and females separately and differences in OS were measured for the two highest order hierarchical clusters in each analysis. Strikingly, amino acids were the only superpathway to significantly stratify patients, and these patients were female (male p=0.79 and female p=0.04). Females in hierarchical cluster 1 had a median OS of 15.3 months versus 19.8 months for cluster 2 (Figure 3A). In comparison, lipids and carbohydrates were not significant in either sex (male p=0.19 and female p=0.14 for lipids; male p=0.28 and female p=0.07 for carbohydrates). To discern which amino acid genes were driving the differences between the two groups, the genes most significantly enriched (FDR-corrected p-value <0.01) in each group were identified (Figure 3B) and were analyzed for pathway enrichment with Gene Set Enrichment Analysis (GSEA). Interestingly, the two most significantly enriched amino acid subpathways in Group 1 females with poorer OS were glutathione metabolism and drug metabolism (Figure 3C). In fact, there were only five genes common to both of these subpathways, all of which were glutathione S-transferases (GST): GSTP1, GSTK1, MGST2, MGST3, and GSTO1. Glutathione metabolism is a critical pathway in tissues, as it helps reduce oxidative stress and promote cell viability (34)(Figure 3D). GST enzymes are also very important, as they are involved in the detoxification of metabolites, toxins and xenobiotics (e.g., chemotherapy) through the attachment of the chemicals to glutathione for excretion and could represent an mechanism for therapeutic resistance and worse OS (34, 35).

Figure 3. Unsupervised gene expression analysis of amino acid metabolism from pancreatic adenocarcinomas in TCGA reveals female-specific stratification from glutathione and drug metabolism.

A. Survival analysis of the two highest order clusters (groups) in amino acid metabolism. B. Heatmap of the most significantly enriched amino acid genes (FDR corrected p-value <0.01) in each group. C. Gene Set Enrichment Analysis of the most significantly enriched genes in Group 1 identifying glutathione and drug metabolism as the two most significantly enriched pathways in Group 1. D. Schematic of glutathione metabolism in cells. Glutathione S-Transferase (GST) enzymes that were present in the top two pathways are in red. GR is glutathione reductase, GPX is glutathione peroxidase. GS-X is glutathione conjugated to a toxin or xenobiotic.

To identify which, if any, of the five GST enzymes were driving the worse OS in the Group 1 females, the effect of the expression of each of the GST enzymes on OS was assessed. Of the five GSTs, overexpression of GSTK1, MGST2, and MGST3 had significantly worse female-specific survival in pancreatic adenocarcinomas; median OS in the female high GST expression group was 12.9 months versus 19.8 months in the low GST group (Figure 4A–B, Table 4). Because female-specific outcomes based on visceral fat were previously observed with clear cell renal carcinoma by our group(7), GST expression data for these tumors were also analyzed, further confirming female-specific stratification that was driven by GSTP1, GSTK1, and MGST3 (Figure C-D, Table 4). Finally, as sex differences in brain tumor metabolism has also been discovered by our group (6, 33), analysis of glioblastoma tumors showed the same female-specific stratification driven by MGST2 and GSTK1 (Figure E-F, Table 4).

Figure 4. TCGA-based tumor gene expression analysis of five GST enzymes reveal female-specific stratification in solid tumors.

A-B. Pancreatic adenocarcinoma (TCGA-PAAD); a Z-score threshold of 2.0 for GSTK1, MGST2 and MGST3 GST enzymes divided low vs high GST groups. C-D. Clear cell renal carcinoma (TCGA-KIRC); a Z-score threshold of 1.75 for GSTP1, GSTK1, and MGST3 GST enzymes divided low vs high GST groups. E-F. Glioblastoma (TCGA-GBM); a Z-score threshold of 1.0 for GSTK1 and MGST2 GST enzymes divided low vs high GST groups. Thresholds and associated p-values for individual genes are listed in Table 4.

Table 4.

Gene expression Z-score threshold analysis for the five GST enzymes across three TCGA solid tumor datasets. The threshold used for Kaplan Meier graphing in each tumor is boxed. All significant p-values are highlighted.

TCGA-PAAD p-values	Z=1	Z=1.25	Z=1.5	Z=1.75	Z=2
GSTP1 female	n.s.	n.s.	n.s.	n.s.	n.s.
GSTP1 male	n.s.	n.s.	n.s.	n.s.	n.s.
GSTK1 female	n.s.	0.002	0.002	0.002	0.002
GSTK1 male	n.s.	n.s.	n.s.	n.s.	n.s.
MGST2 female	n.s.	n.s.	n.s.	n.s.	0.04
MGST2 male	n.s.	n.s.	n.s.	n.s.	n.s.
MGST3 female	n.s.	n.s.	n.s.	0.016	0.008
MGST3 male	n.s.	n.s.	n.s.	n.s.	n.s.
GSTO1 female	n.s.	n.s.	n.s.	n.s.	n.s.
GSTO1 male	n.s.	n.s.	n.s.	n.s.	n.s.
TCGA-KIRC p-values	Z=1	Z=1.25	Z=1.5	Z=1.75	Z=2
GSTP1 female	n.s.	n.s.	0.005	0.024	0.01
GSTP1 male	n.s.	n.s.	n.s.	n.s.	n.s.
GSTK1 female	n.s.	n.s.	n.s.	0.046	n.s.
GSTK1 male	n.s.	n.s.	n.s.	n.s.	n.s.
MGST2 female	n.s.	n.s.	n.s.	n.s.	n.s.
MGST2 male	n.s.	n.s.	n.s.	n.s.	n.s.
MGST3 female	n.s.	n.s.	n.s.	0.017	0.017
MGST3 male	n.s.	n.s.	n.s.	n.s.	n.s.
GSTO1 female	n.s.	n.s.	n.s.	n.s.	n.s.
GSTO1 male	n.s.	n.s.	n.s.	n.s.	n.s.
TCGA-GBM p-values	Z=1	Z=1.25	Z=1.5	Z=1.75	Z=2
GSTP1 female	n.s.	n.s.	n.s.	n.s.	n.s.
GSTP1 male	n.s.	n.s.	n.s.	n.s.	n.s.
GSTK1 female	0.017	0.017	n.s.	n.s.	n.s.
GSTK1 male	n.s.	n.s.	n.s.	n.s.	n.s.
MGST2 female	0.015	n.s.	0.01	0.035	0.004
MGST2 male	n.s.	n.s.	n.s.	n.s.	n.s.
MGST3 female	n.s.	n.s.	n.s.	n.s.	n.s.
MGST3 male	n.s.	n.s.	n.s.	n.s.	n.s.
GSTO1 female	n.s.	n.s.	n.s.	n.s.	n.s.
GSTO1 male	n.s.	n.s.	n.s.	n.s.	n.s.

n.s.=not significant.

Discussion

PDAC is a lethal cancer with a 5-year relative survival rate of 12% (1). Over the past many decades, the mortality due to pancreatic cancer has been relatively stable despite advances in cancer treatments (1). Because metabolism is not only a hallmark of tumorigenesis but also sex-dependent, it is therefore possible that sex differences in metabolism can be leveraged to improve outcomes in this patient population. Herein, we report that preoperative assessment of visceral fat in patients with PDAC has sex-dependent importance in the assessment of OS. In this patient population, both median and optimized rVFA thresholds are able to identify subgroups of females, but not males, with significantly different OS.

In prior studies, the effects of obesity (measured with BMI and visceral fat) have been inconsistently linked to oncologic surgical outcomes. Previous studies have investigated BMI, visceral obesity, and OS in PDAC but do not report survival differences with visceral fat between males and females (16–20). Okumura et al. reported a cohort of more patients (301 patients) compared to the current study and found that overall and recurrence-free survival were associated with high visceral fat (16). Although this study used sex-specific cut-offs in the initial visceral fat determination (i.e., determining a high vs low visceral fat designation), it did not report sex-specific outcomes for the effects of visceral fat. Bian et al. found that a high proportion of visceral fat (visceral to subcutaneous adipose tissue area ratio) was an independent risk factor for mortality (18). Other studies have found different results with visceral fat. One study that evaluated a cohort of 408 pancreatic cancer patients found that cachectic patients with low BMI and low visceral fat had greater 90-day mortality (21). In addition, in a cohort of 100 PDAC patients, Clark et al. found that BMI and cross sectional visceral fat area did not correlate with tumor progression or overall survival in patients with PDAC undergoing resection (22). Although these discrepancies cannot be immediately reconciled, this study, as opposed to other published reports (i) controls for sex as an independent factor in survival and (ii) uses a normalized quantity for visceral fat (i.e. rVFA) that controls for the effects of subcutaneous fat that are antagonistic to visceral fat and have been implicated in protective metabolic effects against diabetes (23, 24).

Our work underscores the need to analyze patients separately by biological sex, in keeping with guidelines by the National Institutes of Health (3). This has broad implications for diagnosis and therapy in the cancer community and support the cumulative body of literature that metabolism is a hallmark of tumorigenesis. Of note, the female-specific impact of rVFA on OS in this cohort of PDAC patients is very similar to our prior observations in patients with renal cell carcinoma and in lymphoma (7, 31) (both less aggressive in comparison with PDAC), suggesting that similar metabolic mechanisms may underlie these sex differences in survival, independent of tumor type. In fact, the identification that glutathione metabolism, specifically GST enzyme expression is associated with female-specific poor outcomes in pancreatic and renal cancers and glioblastoma has potentially significant implications for the interplay between tumor metabolism and obesity. It is well known that obesity is associated with increased oxidative stress and that glutathione, one the major cellular antioxidants, has the ability to regulate obesity as well as inflammation, energy metabolism, and insulin sensitivity (36). In fact, sex differences in glutathione metabolism have been documented across a spectrum of diseases throughout the human body (37). Of potential relevance to our findings, the risk of type 2 diabetes is associated with GST gene polymorphisms in a sex-specific manner (38) suggesting that additional metabolic abnormalities in these patients (e.g., diabetes) may also be associated with sex-specific outcomes. Although the specific mechanism underlying the association of these genes and pathways with female cancer patient outcomes is currently unknown, data suggest that the X chromosome may exert a certain level of control over regulation of glutathione metabolism (39) as well as a significant impact on obesity (40–42).

Because CT imaging is ubiquitous in the routine care of these patients, quantitative body composition assessment can be performed as an alternative to the commonly used BMI that does not provide the same level of sex-dependent outcomes information. Further, with the advent of Current Procedural Terminology Category III codes for imaging analyses (43), body composition metrics can now be integrated easily into clinical practice and billing in a way that is trackable and allows for further supportive evidence to be gathered. As artificial intelligence integration into image processing software continues to advance, the ease of incorporating predictive advanced body composition metrics into daily clinical practice will increase, furthering our ability to tailor care based on predicted outcomes.

Limitations of this study include the retrospective nature of this study and the sample size. Although this was a relatively homogeneous pre-surgical population with PDAC, this homogeneity restricted the study’s sample size and the potential generalizability of these results to larger populations. As an example, given that many patients with PDAC are diagnosed at advanced stages, it is unknown how or if this study’s data can be applied to patients with locally advanced disease or metastasis at presentation. Another limitation of the study is that it uses abdominal fat as the sole body composition biomarker. Other studies have shown that sarcopenia and sarcopenic obesity are independent prognostic factors of poor outcomes in pancreatic cancer patients (44–46), and our study did not quantify serial or baseline muscle mass to assess its sex-specific effects on patient outcomes. Thus, measurements of sarcopenia may be able to complement visceral fat analysis for a broader depiction of the sex-specific body composition changes that are associated with worse outcomes in future research.

Another limitation of this study is the use of single slice body composition analysis software as opposed to volumetric artificial intelligence-aided CT segmentation. Recently, volumetric artificial intelligence-aided CT segmentation has become a more validated technique that has been shown to have more reproducibility and higher accuracy than single slice methods used in this study (47–49). An advantage of these newer artificial intelligence algorithms is that they can investigate fat texture as well as quantity (50) which can provide even more information that may have implications for pancreatic cancer as well. Future research could explore the rapidly expanding capacity of artificial intelligence in this space.

Despite the limitations, this study does suggest new avenues for investigation relevant to patient care, one of which is in the interaction of visceral fat metabolism and PDAC patient management. Increased physical activity and exercise is increasingly recognized as a “therapeutic” to not only reduce cancer risk, but also reduce treatment side effects and improve quality of life and outcomes in multiple cancers, including pancreatic cancer (51–53). In addition, as metabolism is a well-characterized druggable target for other diseases, namely diabetes, and suggest that FDA-approved pharmaceuticals could be rapidly repurposed off-label for PDAC patients to target both tumor nutrient metabolism and visceral obesity. For example, new obesity pharmaceuticals reduce visceral fat in diabetic patients, such as glucagon-like peptide 1/glucose-dependent insulinotropic polypeptide/gastric inhibitory peptide (GLP-1/GIP) agonists and sodium glucose cotransporter 2 (SGLT2) inhibitors (54, 55). SGLT2i, in particular, have documented anti-cancer effects in numerous animal models (56, 57). The use of such drugs, perhaps along with exercise, could be the subject of large prospective multi-institutional clinical trials that would provide the statistical power to identify the impact of sex and metabolism on PDAC survival.

In summary, these data provide further evidence of the potential utility of using metabolic markers obtained from CT imaging and tumor gene expression to better predict outcomes in cancer patients on a sex-specific basis. The utility of this personalized paradigm is further strengthened by the fact that pharmaceuticals and dietary regimens exist that can alter the activity of metabolic pathways in the tumor and in the individual and result in the loss of visceral obesity. A clinical workflow of this nature could advance precision medicine initiatives by treating males and females separately resulting in more substantial gains in outcomes for individuals with pancreatic cancer.

Abbreviations:

PDAC

pancreatic ductal adenocarcinoma

CT

computed tomography

OS

overall survival

SFA

subcutaneous fat area

VFA

visceral fat area

TFA

total fat area

rVFA

relative visceral fat area

rSFA

relative subcutaneous fat area

GST

Glutathione S-Transferase