Plasma miRNA Biomarkers in Limited Volume Samples for Detection of Early-stage Pancreatic Cancer

Pancreatic cancer, presenting as pancreatic ductal adenocarcinoma (PDAC) in more than 90% cases, is the most deadly cancer by organ site with a 5-year survival rate of only approximately 9% (1). While only 2.9% of patients with distant PDAC survive 5 years, survival rates increase by over an order of magnitude to 37.4% when the disease is detected early (stages I and II; ref. 1). Recent findings have revealed a significant increase in 5-year survival of patients with stage IA disease from 44.7% in 2003 to 83.7% in 2012. This trend has been suggested to be the result of improved early diagnosis and detection (2). Only patients with early stage, localized PDAC (comprised of stages I and II), are candidates for curative surgical resection, thus detection of these early-stage cases is crucial to improving survival rates (3). However, early detection of PDAC remains challenging because most patients do not present with symptoms until after the disease has spread locally or to distant sites. Furthermore, given the location of the pancreas deep in the abdomen, palpating pancreatic tumors is difficult, and imaging may miss tumors depending on their size and location. Together, these facts underscore the need to develop minimally invasive liquid biopsy assays with adequate sensitivity and specificity to detect PDAC early.

The utility of carbohydrate antigen 19-9 (CA 19-9) as a diagnostic biomarker has been extensively studied in PDAC. While serum CA 19-9 levels have a sensitivity and specificity for PDAC of 79%–81% and 82%–90%, respectively, CA 19-9 is not recommended as a screening marker due to its low positive predictive value (0.5%–0.9%; refs. 4–6), particularly in asymptomatic populations (5). CA 19-9 is elevated in about 10% to 20% of patients with various benign pancreatobiliary conditions, which result in false positives (7–9). Conversely, 10% of the population lacking 1,4-fucosyl transferase due to germline mutations are unable to produce sialyl Lewis antigen epitopes and hence do not secrete CA 19-9, making the marker unusable for a subset of patients with pancreatic cancer due to their false negativity (6, 10). Despite these limitations, however, CA19-9 remains the most informative biomarker for a substantial group of patients with pancreatic cancer and no other biomarker has yet surpassed its performance. Among other promising biomarkers described for diagnosis of pancreatic cancer, most have revealed improved specificity and sensitivity in combination with CA19-9, which include cell-free DNA, metabolites, and various RNAs, including miRNAs.

miRNAs are small noncoding RNAs, approximately 22 nucleotides in length, that epigenetically modulate gene expression or translation by predominantly binding to the 3′ untranslated region of mRNAs resulting in miRNA deadenylation, target cleavage, or translational repression (11–13). It is estimated that miRNAs regulate at least 60% of mRNA transcripts and are conserved across species (14, 15). These small RNAs are stable when circulating in various body fluids (16), including blood, and have also been found ensconced in the protective lumen of circulating extracellular vesicles, such as exosomes (17). In addition, miRNAs regulate critical physiologic processes and are frequently deregulated in many diseases, including viral infections, immune-related diseases, neurodegenerative disorders, and cancer (18–20). Taken together, their deregulation in disease and stability in circulation make miRNAs promising circulating biomarkers of disease diagnosis and prognosis.

The utility of circulating miRNAs as biomarkers has been explored in many diseases, including in pancreatic cancer (21). A recent publication has reported compelling evidence about the diagnostic potential of a serum miRNA panel, developed with a neural network analysis, for detection of ovarian cancer with 100% specificity (22). The miRNA neural network outperformed the currently used gold standard biomarker for ovarian cancer, CA125, with a positive predictive value of 91.3% and negative predictive value of 78.6%.

Despite their biological significance due to involvement in critical disease relevant pathways and many publications demonstrating their promising performance as informative biomarkers of disease diagnosis and prognosis, there are multiple challenges to developing and transitioning circulating miRNA biomarker assays to clinical application. These include the relatively large sample volume, in hundreds of microliters, required to perform the currently available and commonly used assays for profiling of miRNAs.

In this study, we report successful application of a novel limited volume assay platform, Abcam Fireplex (23), to identify differentially abundant plasma miRNAs between early-stage pancreatic cancer cases and controls. This hydrogel particle-based miRNA assay platform was used for 20 μL of plasma to detect up to 68 miRNAs per sample in a 96-well plate format without requiring extensive prior miRNA isolation and purification steps (23). This assay was recently used to develop a plasma miRNA signature for detection of lung cancer and identify histologic subtypes (24). Here we demonstrate successful application of this technology in identifying circulating miRNA biomarkers for detection of early-stage resectable PDAC in limited volume plasma with good sensitivity and specificity.

The study was approved by the Institutional Review Boards at all participating institutions with written informed consent from enrolled patients. This study was conducted in accordance with the U.S. Common Rule regulations. PDAC cases were pathologically confirmed and plasma was collected from these patients at baseline, prior to receiving any treatment, including surgery. Patients with prior history of cancer or concurrent diagnosis with another cancer were excluded from this study. The main cohort included 88 plasma samples from 30 healthy controls and 58 patients with pathologically confirmed stage II PDAC cases from the TexGen repository, a consortium from the Texas Medical Center. Using the TexGen cohort, we identified promising candidate plasma miRNA markers and marker panels. We confirmed and validated the individual candidate markers and marker panel using two independent patient cohorts: one from the University of Pittsburgh (UPMC; Pittsburgh, PA), which included 25 control plasma samples with benign pancreatic disease and 34 stage II PDAC plasma samples, and the second from the NCI Early Detection Research Network (EDRN) pancreatic cancer reference set including plasma samples from18 healthy controls and 44 stage II cases.

We analyzed plasma samples in a 96-well plate using the Abcam Fireplex platform with their predesigned oncology panel (Abcam) in accordance with manufacturer instructions (23). In brief, samples were thawed at room temperature and placed on ice. A total of 20 μL of lysis buffer was placed in each well of a sterile 96-well plate and mixed with 20 μL of plasma. Plates were sealed and incubated for 45 minutes at 60°C while shaking (here and after always 750 revolutions per minute [RPM]). A total of 20 μL of lysed sample was aliquoted into a new plate along with the positive controls provided with the platform. Water served as negative controls. The remaining sample was stored at −20°C. Then 35 μL of particles from the oncology panel were added to a clean filter plate and connected to a vacuum manifold to remove storage buffer. After adding 25 μL each of hybridization buffer and lysed sample, the filter plate was covered and incubated for 60 minutes at 37°C while shaking. The samples were then rinsed twice with 1× Rinse A solution and resuspended in 1× labeling mix. The plates were incubated for 60 minutes at room temperature while shaking and stored overnight at −20°C. The next day, plates were thawed on ice and rinsed with Rinse B and Rinse A solutions. Then miRNAs were eluted from the probes by adding RNAse free water and incubated for 30 minutes at 55°C while shaking. A clean catch plate was inserted into the vacuum manifold and carefully aligned with the wells so that orientation of the samples could be properly suctioned into the correct wells. The wells were kept hydrated with Rinse A in the empty filter plate and stored at 4°C until needed. In a clean PCR plate, 30 μL of eluent and 20 μL of PCR master mix were added and subjected to PCR in a thermocycler with the following program: 1 cycle at 93°C for 15 seconds; 27 cycles of: 93°C for 5 seconds, 59°C for 15 seconds, 72°C for 60 seconds; 6 cycles of: 93°C for 5 seconds, 63°C for 15 seconds, 72°C for 60 seconds; 1 cycle of 94°C for 4 minutes; 4°C hold. After PCR, the Rinse A solution was removed under vacuum and 60 μL of hybridization buffer was added to each well of the filter plate. A total of 20 μL of PCR product was transferred to the filter plate and incubated for 30 minutes at 37°C with shaking while the remaining PCR product was frozen at −20°C. After removing the plate from the filter plate on the shaker, the PCR product in the wells of the plate was rinsed twice with Rinse A, added with 75 μL of reporter mix, covered and incubated at room temperature for 15 minutes while shaking. Then the wells were rinsed twice with Rinse A, mixed with run buffer and read using a BD Accuri cytometer (BD Biosciences). The FCS files were uploaded into the Fire Code software (Abcam) and analyzed by normalizing the mean fluorescence intensities using the three most stable miRNAs across all samples: with miR-17-5p, miR-20a-5p, and miR-93-5p. The customized oncology panel allowed us to measure relative abundances of 68 vendor designed preselected miRNAs (Supplementary Table S1).

Patient characteristics have been summarized using frequency tables and descriptive statistics. The χ² test or Fisher exact test was used to compare the discrete variable (such as sex and diabetes) differences between cases and controls. For continuous variables such as age, the Wilcoxon test was used. In the miRNA marker screening process, the method of Benjamini and Hochberg (1995) was used to produce the adjusted P values so that we could control the FDR under 10% (25). Logistic regression model was used to combine markers in the panel development. ROC curves were constructed. The AUC was estimated, and its 95% confidence interval (CI) was estimated using the bootstrapping method. The study used the TexGen cohort for marker discovery. The top three candidate miRNA markers were chosen for the panel development due to their reported clinical importance in published literature and their statistical significance in differentiating cancer cases versus controls. The performance of the marker panels was assessed using 10-fold cross-validation. The performance of the candidate markers and marker panels were further tested using two independent validation cohorts from the University of Pittsburgh (UPMC) and the NCI EDRN. All the analyses were performed using the statistical software R 3.3.3 (CRAN, RRID:SCR_003005, https://cran.r-project.org) and Stata release 16 (Stata, RRID:SCR_012763, https://www.stata.com).

TCGA-PAAD miRNA data for gene set enrichment analysis (GSEA) were downloaded from http://gdac.broadinstitute.org/; mRNA data for GSEA was downloaded from The Cancer Genome Atlas (TCGA) data portal and normalized using the quantile normalization technique. Both datasets were log₂ transformed for subsequent analysis. A total of 178 pancreatic adenocarcinoma cases with the miRNA profiling data were used for further analysis. The level of the 3mir panel (3mir-level) was calculated as the average log₂ expression of miR-34a-5p, miR-130a-3p, and miR-222-3p. All samples were divided into two groups by median into those with high 3miR-level and those with low 3miR-level. Then the datasets were analyzed using GSEA (26, 27). Signal-to-noise and 1,000 permutations of the genes were applied in the GSEA with the gene sets obtained from the MSigDB database v6.2 for all hallmark and oncogenic signatures.

Patient characteristics have been summarized in Table 1. In the training set (TexGen cohort), there were 58 stage II and 30 healthy control samples. Comparisons were performed for each individual miRNA marker present on the Fireplex oncology panel. The distribution for all miRNAs can be found in Supplementary Fig. S1. Eleven miRNAs were identified with significant differential abundance between stage II and control patients with FDR < 0.1 (25). Our top three candidate miRNAs included miR-34a-5p (P < 0.0001), miR-130a-3p (P = 0.0003), and miR-222–3p (0.0019; Table 2; Fig. 1). All three of these miRNAs were elevated in cancer versus controls. Neither sex, smoking, nor diabetes were significantly different between cases and controls.

The top three statistically significant differentially abundant miRNAs, discriminating between stage II versus control plasma samples, included: miR-34a-5p, miR-130a-3p, and miR-222-3p, with AUC values of 0.77 (95% CI, 0.66–0.87), 0.74 (95% CI, 0.63–0.84), and 0.70 (95% CI, 0.59–0.82; Table 2; Fig. 2A), respectively. Consistent with previous studies (28, 29), CA19-9 performed well in differentiating stage II cases versus control cases, with an estimated AUC = 0.89 (95% CI, 0.81–0.95). Combining CA 19-9 with selected individual miRNA markers, miR-34a-5p, miR-130a-3p, and miR-222–3p, improved the AUC values for each panel to 0.92 (95% CI, 0.86–0.97), 0.94 (95% CI, 0.89–0.98), and 0.92 (95% CI, 0.87–0.97), respectively (Table 3; Fig. 2B). Among these, the marker panel of CA19-9 with miR-130a-3p improved AUC significantly compared with that using CA19-9 alone (P = 0.0390). The performance of the combined CA19-9 and miRNA marker panels remained similar in estimated AUC when using 10-fold cross-validation. Using 10-fold cross-validation, CA 19-9 alone had an estimated AUC of 0.88 (95% CI, 0.81–0.95). When combined with CA 19-9, miR-34a-5p, miR-130a-3p, and miR-222–3p had AUCs of 0.91 (95% CI, 0.84–0.97), 0.94 (95% CI, 0.89–0.99), and 0.89 (95% CI, 0.82–0.96), respectively.

The three significant miRNAs from the training set, miR-34a-5p, miR-130a-3p, and miR-222-3p outperformed chance in the UPMC cohort, with significant differences between stage II and control patients (Table 2; Fig. 2C). Neither sex, smoking, nor diabetes were significantly different between cases and controls, consisting of patients with benign pancreatic disease, as described in Table 1. When combined with CA19-9, their estimated AUCs were 0.84 (95% CI, 0.72–0.93), 0.83 (95% CI, 0.72–0.93), and 0.83 (95% CI, 0.72–0.93), respectively (Table 3; Fig. 2D). However, none of the marker panels had statistically significant improvement in AUC compared with that of CA19-9 alone.

Next, we tested the performance of the three selected miRNA markers and their respective panels with CA19-9 using a blinded validation set of EDRN samples comprising of 44 stage II cases and 18 healthy controls. Marker miR-34a-5p showed significantly higher values for cases than controls (P = 0.0470), with an AUC of 0.66 (Table 2; Fig. 2E). When combined with CA 19-9, miR-130a-3p (Table 3; Fig. 2F) yielded an improved AUC of 0.87 (95% CI, 0.77–0.95).

Relatively stronger performance of the panels in discriminating stage II from controls in the training set (TexGen cohort) compared with UPMC cohort, could, in part, be a reflection of the fact that the TexGen training set control samples consisted of plasma from disease-free healthy individuals while for the UPMC cohort, control samples consisted of plasma from patients with benign pancreatic disease. Improved performance in the training set could also be due to differences in age ranges across cohorts (P < 0.001; Table 1). PDAC risk increases with age, hence this could affect performance in our validation sets, which included control patients of older age. In addition, given that the sample cohorts were from banked plasma collected at different institutions, which may have had different standard operating protocols (SOP) for sample collection, time to processing and storage, it is possible that varying performance of the candidate miRNAs in detecting resectable disease were partly due to absence of uniform SOPs for bio-banking at respective institutions that affects miRNA detection outcome (30, 31). It is imperative that future studies are designed to control for these potential variables by adhering to uniform SOPs for sample collection and storage across institutions. Furthermore, there can be discrepancies in clinical staging, particularly in unresected patients for whom pathologic staging is unavailable, where nodal involvement is often underestimated because of current imaging limitations (32–34).

To understand the significance of these three miRNAs in pancreatic cancer, we conducted GSEA of all hallmark genes for 178 pancreatic adenocarcinoma samples from TCGA. Interestingly, samples with high 3miR-levels showed significant enrichment in genes associated with glycolysis (P = 0.0139; Fig. 3A). Glycolysis is known to play an important role in cancers, especially in the context of those with Ras mutations (35–37), which are almost ubiquitous in PDAC (38). This finding suggests that high 3miR expression is associated with the genetic pathways facilitating aerobic glycolysis for energy metabolism in early-stage PDAC cells. We also conducted GSEA of oncogenic signatures to better understand the role of these three miRNAs in tumorigenesis. Results revealed that high 3miR expression is associated with gene sets that correlate with the activation of oncogenes, including KRAS and AKT and inactivation of the tumor suppressor gene, RB (Fig. 3B). Conversely, samples with low 3miR expression showed enrichment of genes that are downregulated in cells with TP53 loss-of-function mutations, also revealed in the hallmark GSEA results. In addition, Kyoto Encyclopedia of Genes and Genomes pathway analysis using DIANA-miRPath v3. 0 revealed that these miRNAs are involved in several pancreatic cancer-relevant pathways, such TGFβ signaling (P = 1.41 × 10^–3), phosphatidylinositol signaling (P = 1.72 × 10^–3), FOXO signaling (P = 5.73 × 10^–3), TP53 signaling (P = 0.0170), dorso-ventral axis formation (including NOTCH signaling; P = 0.0186), ERBB signaling (P = 0.0210), and axon guidance (P = 0.0307) pathway, among others (39).

A panel of miRNA biomarkers that can stratify patients with early-stage PDAC from control subjects, including those with benign disease, was developed using a novel hydrogel particle-based miRNA profiling platform (Abcam Fireplex) amenable to screening with limited volume body fluid samples. Development of a minimally invasive early detection circulating miRNA biomarker assay using limited volume plasma samples addresses the major challenge of large volume requirement for miRNA profiling in currently available conventional methods. This “proof of concept” study with the vendor designed, prefabricated assay platform for miRNA biomarker assay demonstrates the feasibility of the technology being translated to clinical practice in diagnostic settings to improve chances of curative clinical intervention and survival for patients with early-stage PDAC. Although the current assay development was restricted to 68 miRNAs on the Abcam Fireplex Oncology Panel, individual miRNAs (miR-34a-5p, miR-130a-3p, and miR-222-3p) together with the currently used biomarker for pancreatic cancer, CA 19-9, improved performance of the latter in discriminating patients with early-stage PDAC from controls. These miRNAs in combination with CA 19-9 yielded AUCs of 0.92, 0.94, and 0.92, respectively, in our training cohort, with miR-34a-5p and miR-222-3p also performing well in both validation cohorts.

All three candidate miRNAs have been previously reported to play roles in disorders of the pancreas and cancer. Importantly, miR-34a has been demonstrated to be a tumor suppressor miRNA in PDAC (40, 41) as well as a candidate serum biomarker for the disease (42). Previous studies have also reported that miR-34a-5p impairs PDAC progression through posttranscriptional regulation of SNAI1 and NOTCH1, inhibiting epithelial–mesenchymal transition (43). In addition, miR-34a-5p deficiency mediated loss of TP53 function in lung adenocarcinoma and head and neck cancer was shown to affect critical tumor-promoting pathways including cancer associated adrenergic transdifferentiation of sensory nerves (44). These findings have led to the idea of miR-34a is a potential therapeutic target (45, 46). Preclinical studies with miR-34a miRNA-mimics in mice with PDAC combined with PLK1 siRNA improved survival and decreased the rate of tumor growth (45). Another study has suggested that Genistein, which upregulates miR-34a promoting apoptosis and inhibiting cell growth in PDAC cell lines, may have therapeutic potential against PDAC (46). The GSEA result showing differential enrichment of TP53 hallmark signatures between high and low 3miR-expressing pancreatic adenocarcinoma cases corroborated previous finding. The level of miR-130a-3p was reported to be elevated in pancreatic islets from hyperglycemic donors and in islets from Goto-Kakizaki rats with type 2 diabetes (47, 48). An estimated 50% to 80% of patients with PDAC have diabetes or some form of glucose intolerance (49–51), and diabetes itself is a known risk factor for PDAC development (52–54). It would be interesting to determine whether circulating levels of miR-130a-3p could be used to identify which diabetic patients are predisposed to develop PDAC. Interestingly, miR-130a was found to be decreased in high-risk intraductal papillary mucinous neoplasm (55), but has not been assessed as a circulating biomarker in blood for such lesions. Finally, elevated levels of miR-222 in tumor tissues has been associated with poor prognosis for patients with pancreatic cancer (56, 57). Cell line studies have suggested that miR-222 may play a role in PDAC by targeting cyclin-dependent kinase inhibitor CDKN1C/P57, promoting cell proliferation and viability (58). Overall, GSEA results indicate that the three differentially abundant miRNAs in plasma from patients with early-stage PDAC, as candidate biomarkers of the disease, play roles in oncogenic signaling mediating downregulation of tumor suppressor genes, activation of glycolysis and cellular growth.

Although the above findings reflect mechanistic involvement of the three miRNAs in deregulating cancer relevant pathways, it is plausible that the three miRNAs do not comprehensively capture the complex and heterogeneously evolving aberrant proteomics landscape of early-stage cancer genomes. In addition, miRNA expression is regulated by diverse cellular and developmental contexts (59) and multiple miRNAs regulate overlapping target genes and their pathways (60). This suggests that highly sensitive and specific early detection PDAC miRNA biomarker signatures will include miRNAs involved in diverse pathways frequently deregulated in the disease. Development of such miRNA signatures would require unbiased global profiling of miRNAs in larger sample cohorts processed under controlled standard operating procedures. It will also be important that the data be analyzed with adequately powered statistical tools, such as neural network analysis. This tool recently helped generate a 7 miRNA panel from next-generation sequencing data for detecting epithelial ovarian cancer with high sensitivity and specificity (22). This panel outperformed the gold standard biomarker, CA-125, commonly used for diagnosing ovarian cancer (22). While such additional larger scale discovery and validation studies are warranted for developing a more robust early detection miRNA biomarker signature for PDAC, the current findings demonstrate that limited volume plasma samples, available in clinical diagnostic settings, can be effectively used to assay the differentially abundant miRNAs associated with early-stage resectable pancreatic cancer. Reliability of the assay platform, used in this study, was suggested by a recent study reporting that on average, approximately 70% of miRNAs detected by Fireplex match in directionality and significance to those detected by RNA sequencing (61). In the future, it will be important to include stage I cases as well as prospective prediagnostic samples besides those with benign disease and benign pancreatic neoplasms to assess performance of the miRNA biomarkers described in this study.

In summary, we have identified candidate circulating miRNA biomarkers from limited volume of plasma that can discriminate individuals with early-stage resectable pancreatic cancer from cancer-free controls, including those with benign pancreatic diseases. Having vetted this technology by using the commercially available Oncology Panel from Fireplex, we plan to use in the future a custom panel with miRNAs, deregulated in pancreatic cancer and/or involved in known pancreatic cancer–related pathways, such as those suggested by the GSEA findings in this study. Focusing on differentially abundant circulating miRNAs involved in deregulated biological pathways would be a logical approach to developing minimally invasive early detection biomarker assay for patients not expressing CA19-9 besides improving sensitivity and specificity of CA19-9 as a combined panel, similar to the proteomic markers representing a migratory signature in pancreatic cancer reported by us earlier (62). Further development of the assay platform for more robust plasma miRNA biomarker panel discriminating early-stage pancreatic cancer with high sensitivity and specificity is expected to have significant impact on improving clinical care and disease outcomes for patients with this recalcitrant malignancy.

R. Brand reports grants from NCI EDRN grant during the conduct of the study; grants from Immunovia and grants from Freenome outside the submitted work. C. DeCapite reports grants from NIH during the conduct of the study; personal fees from Immunovia outside the submitted work. R. Dittmar reports other from Dr. John J. Kopchick Fellowship from the University Of Texas MD Anderson UT Health Graduate School of Biomedical Sciences, grants from NIH NCI T32, grants from NIH NCI U01, and grants from Cancer Prevention Research Institute of Texas (CPRIT) during the conduct of the study. A. Killary reports a patent for US Patent filed pending and a patent for International Patents filed pending. K. Kirkwood reports grants from V Foundation and grants from NIH/NCI during the conduct of the study; personal fees from Alaunus and personal fees from Amplified Biosciences outside the submitted work. G. Longton reports grants from Fred Hutchinson Cancer Research Center during the conduct of the study. A. Maitra reports other from Cosmos Wisdom Biotechnology Ltd and other from Thrive Earlier Detection during the conduct of the study. S. Sen reports grants from NIH/NCI, grants from CPRIT, during the conduct of the study. No disclosures were reported by the other authors.

R.L. Dittmar: Conceptualization, investigation, methodology, writing–original draft. S. Liu: Formal analysis, visualization, writing–original draft. M.C. Tai: Formal analysis, writing–review and editing. K. Rajapakshe: Writing–review and editing. Y. Huang: Formal analysis, visualization, writing–review and editing. G. Longton: Formal analysis, visualization, writing–review and editing. C. DeCapite: Resources. M.W. Hurd: Resources. P.L. Paris: Writing–review and editing. K.S. Kirkwood: Writing–review and editing. C. Coarfa: Writing–review and editing. A. Maitra: Writing–review and editing. R.E. Brand: Resources, writing–review and editing. A.M. Killary: Resources, funding acquisition, writing–review and editing. S. Sen: Conceptualization, funding acquisition, writing–original draft.

This study was supported by funding from the University of Texas MD Anderson Cancer Center Pancreatic Cancer Moonshot Program; the Cancer Prevention Research Institute of Texas (RP180392, to A.M. Killary and S. Sen); the NCI at the NIH (1 U01 CA214263, to S. Sen and A.M. Killary, P50CA221707, U01CA196403, and U01CA200468, to A. Maitra, and U01 CA200466, to R.E. Brand). R.L. Dittmar is supported by John J. Kopchick Fellowship from the University Of Texas MD Anderson UT Health Graduate School of Biomedical Sciences and the National Institutes of Health Translational Genomics and Precision Medicine in Cancer T32 Training Program (5T32CA217789-03) from the NCI. This work was also supported in part by The Cancer Prevention Institute of Texas (CPRIT) RP170005, NIH P30 shared resource grant CA125123, and NIEHS grants 1P30ES030285 and 1P42ES0327725 (to C. Coarfa).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.