RAB19, SERPINB9P1, and Pancreatitis in Patients Taking Azathioprine in Routine Clinical Practice: Genome and Transcriptome‐Wide Association Studies

Abstract

Azathioprine is used to treat several inflammatory and autoimmune diseases. However, its use is limited by serious adverse events, including acute pancreatitis. Prior studies have found an association between the HLA region and thiopurine‐induced acute pancreatitis (TIAP); however, in clinical practice, many patients with pancreatitis do not meet this strict criterion. We aimed to identify additional genes associated with azathioprine‐related pancreatitis using genome‐wide and transcriptome‐wide association studies (GWAS and TWAS) by broadening the definition of pancreatitis. We conducted a retrospective study of azathioprine users with inflammatory conditions. We used electronic health records linked to genomic data from BioVU (Vanderbilt's biobank) and replicated the results using NIH's All of Us. The primary outcome was acute pancreatitis, and the secondary outcome was pancreatic injury. Sixteen patients with pancreatitis and 2085 control subjects were included from BioVU; the All of Us cohort included < 20 patients with pancreatitis and 847 control subjects. The GWAS analysis (adjusted for 10 principal components of genetic ancestry, sex, age, and azathioprine indication) in the BioVU cohort found an association between pancreatic injury and rs2948386 in RAB19 (OR = 3.47, P = 1.46E⁻⁸), which was replicated in All of Us (OR = 2.70, P = 4.18E⁻³). We also conducted a TWAS adjusting for the same factors above and found a significant association between genetically predicted pancreatic expression of SERPINB9P1 with pancreatic injury (BioVU: effect size = 0.42, P = 1.48E‐5; All of Us: effect size = 0.48, P‐value = 0.01). In summary, we identified two new genetic associations for azathioprine‐related pancreatic injury: the predicted expression of SERPINB9P1 and a SNP in RAB19.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Azathioprine is an immunosuppressant used to treat inflammatory conditions including glomerulonephritis, inflammatory bowel disease, and autoimmune rheumatologic conditions. Thiopurine‐induced pancreatitis is not related to TPMT and NUDT15. Pancreatitis is a dose‐independent, idiosyncratic side effect of azathioprine.

• WHAT QUESTION DID THIS STUDY ADDRESS?

We sought to identify novel loci/genes associated with pancreatitis/pancreatic injury in patients taking azathioprine using high‐throughput methods such as GWAS and TWAS.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This study uncovered two genes associated with pancreatitis and pancreatic injury in patients taking azathioprine: RAB19 and SERPINB9P1. The effects of these genes may be mediated through pathways involving autophagy (RAB19) and DNA repair (SERPINB9P1).

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

Elucidating variants that confer additional risk has the potential to help identify novel mechanisms of azathioprine side effects and could contribute to personalized use of the drug.

Acute pancreatitis is a common, potentially life‐threatening inflammatory condition. Over 2.8 million cases of acute pancreatitis were diagnosed globally in 2019. ¹ The overall mortality rate for those who are diagnosed with acute pancreatitis is estimated to be 5%. ² Approximately 20% of patients with acute pancreatitis progress to severe acute pancreatitis, which can lead to multiorgan failure. ³ In the United States, over 275,000 patients are hospitalized for pancreatitis each year and incur annual healthcare costs exceeding $2.6 billion. ⁴ Most cases of acute pancreatitis are due to alcohol and gallstones. ⁵ Other causes include hypertriglyceridemia, autoimmune conditions, infection, and medications. ⁶

One medication known to cause pancreatitis is azathioprine. It is an immunosuppressant used to treat a variety of inflammatory and immune‐mediated diseases, including inflammatory bowel disease (IBD), rheumatic conditions, and some forms of glomerulonephritis. A systematic review of Crohn's demonstrated that the risk of pancreatitis in patients taking thiopurines is higher than placebo/control (3.80% vs. 0.2%). ⁷ When it occurs, it necessitates permanently discontinuing azathioprine, given the high risk of recurrent pancreatitis with continued use or re‐challenge. ⁸ Further, thiopurine‐induced pancreatitis is not related to TPMT and NUDT15. ⁹ , ¹⁰ Pancreatitis is independent of dose, which suggests the cause is not linked to the drug's metabolic pathway. ⁷ , ¹¹ , ¹² The underlying mechanisms driving pancreatitis in patients who are taking azathioprine are unclear, and currently, there are no definitive ways to predict who will develop pancreatitis. Studies have shown that patients taking azathioprine for IBD are more susceptible to azathioprine‐induced pancreatitis when compared to patients taking azathioprine for other inflammatory or rheumatic diseases. ¹³ , ¹⁴ The reason why this population is at increased risk remains elusive, but the increased risk has prompted studies to focus almost exclusively on this population. ¹⁵ In fact, the only GWAS that has investigated the association between azathioprine and pancreatitis was conducted exclusively in patients with IBD. ¹⁶ The study used the Liverpool Causality Assessment Tool to define cases and accepted only definite or probable cases of pancreatitis. ¹⁶ A broader definition would help capture more people at risk for this side effect. Furthermore, research focused on untangling the genetic and phenotypic architecture underlying pancreatitis in patients taking azathioprine for a variety of conditions would be more generalizable. The primary analysis of this study was to identify genetic risk factors for pancreatitis in users of azathioprine for diverse inflammatory and autoimmune conditions. For our secondary analysis, we used a less restrictive definition of pancreatitis: pancreatic injury.

METHODS

Data sources

The discovery cohort was assembled from BioVU, a large genomic biobank linked to deidentified longitudinal EHRs for individuals receiving care at a tertiary care academic medical center, Vanderbilt University Medical Center. ¹⁷ , ¹⁸ , ¹⁹ BioVU data include clinical care notes, medication usage, demographic characteristics, diagnostic and procedure codes, and laboratory results.

The validation cohort was All of Us ²⁰ ; it is one of the largest, most diverse, and most broadly accessible datasets assembled to date; it has over 849,000 participants enrolled. ²¹ As with BioVU, the genomic data are linked to deidentified longitudinal data including demographics, comorbidities, medications, lifestyle factors (e.g., smoking), and laboratory tests. All of Us curates and cleans its datasets extensively, which is sourced from EHRs, fitness trackers, and survey questions.

Study population

Discovery cohort

We assembled the BioVU cohort using natural language processing (NLP) in the EHR to gather possible users of azathioprine. Next, we excluded individuals without high‐quality DNA and then conducted a review of records on the remaining patients to confirm the use of azathioprine. We excluded patients with organ transplant as the primary indication for azathioprine and anyone with a history of pancreatitis prior to azathioprine initiation cohort entry. Because the genetic data used for imputation and the GTEx project primarily includes individuals of European ancestry, ²² we restricted the cohort to individuals whose reported race was White (Figure S1 ). To better reflect real‐world clinical settings, we used race as recorded in clinical records. There is a strong correlation between self‐reported or attributed race and genetic ancestry in BioVU. ²³ , ²⁴ , ²⁵

We gathered demographic and clinical variables through record review, and data were entered into a REDCap database. Demographic variables included sex, age at end of follow‐up (see below), and initial weight (kg). For weight, we allowed measurements taken up to 1 year prior to azathioprine initiation or within 3 days after starting, but we prioritized the closest measurement. Clinical variables included initial azathioprine dose (mg) and azathioprine indication, categorized as (i) IBD and (ii) other rheumatologic conditions (e.g., systemic lupus erythematosus, other connective tissue disorders).

Validation cohort

All of Us curates and cleans demographic and clinical variables by sourcing EHRs; they are provided to the researcher in table format. To assemble the All of Us cohort, we conducted database queries and restricted individuals with reported White race, taking azathioprine or mercaptopurine. Mercaptopurine is structurally and functionally similar to azathioprine, as azathioprine is first converted into mercaptopurine. ²⁶ We excluded patients with a history of organ transplant. We collected similar demographic and clinical variables in the All of Us cohort as BioVU; however, All of Us did have ~75% missingness for initial dose. All of Us has a cell suppression limit of 20; therefore, any numbers less than 20 will be presented as < 20.

Follow‐up

The cohort entry was the date of the first prescription fill. Follow‐up continued until the earliest of the following dates: (i) day of azathioprine discontinuation, (ii) loss of follow‐up, (iii) predefined end of the study period (BioVU: December 31, 2018; All of Us: October 1, 2023), (iv) 90 days after the last confirmed azathioprine dose, (v) death, or (vi) outcome.

Outcome

In BioVU, for our primary outcome of acute pancreatitis, we first gathered possible cases by identifying individuals with either (1) an ICD‐9 or ICD‐10 code for acute pancreatitis (Table S1 ) or (2) an amylase or lipase at least three times the upper limit of normal (amylase: 90 U/L; lipase: aged < 18 = 70 U/L, aged ≥ 18 = 156 U/L). Next, we conducted a record review to confirm pancreatitis cases using the American College of Gastroenterology Guidelines. ²⁷ We defined the secondary outcome, pancreatic injury, as anyone with an amylase or lipase two times the upper limit of normal. For both definitions, we confirmed that azathioprine initiation preceded any laboratory values or ICD codes. All reviewers were blind to the genetic data throughout the review process. Individuals who were classified as having pancreatic injury but did not meet the criteria of acute pancreatitis were not included as controls in any of the analyses.

For All of Us, we did not have access to record notes. The definition of pancreatic injury remained the same, but we had a separate definition for acute pancreatitis: either (1) an ICD‐9 or ICD‐10 code for pancreatitis (Table S1 ) or (2) an amylase or lipase at least three times the upper limit of normal.

Genotyping, imputation, and quality control

For BioVU data, genotyping was completed using Illumina Infinium Expanded Multi‐Ethnic Genotyping Array plus custom content data (VUMC BioVU MEGA^EX). The VUMC BioVU MEGA^EX, as part of a larger institutional initiative, underwent quality control; specifically, it filtered out variants with a call rate < 95%, unexplained relatedness, lack of concordance in a HapMap Mendel/Concordance Evaluation, or with gender discrepancies between genetics and EHR. We applied additional quality control, including removing SNPs with minor allele frequency < 1% and individuals with a call rate < 90%. We imputed the data using previously validated and published methods. ²³ Briefly, we first prepared the data using the McCarthy Tools and then imputed genetic variants using the Michigan Imputation Server (HRC version r1.1 reference panel); we phased using Eagle. ²⁸ , ²⁹ , ³⁰

In addition to GWAS, we conducted a transcriptome‐wide association study (TWAS). This method tests the association between the genetically predicted expression of a gene and a phenotype. For the TWAS analysis, we imputed gene expression (in liver and pancreatic tissue) using SPrediXcan (GTEx v8) with MASHR version 8 weights for imputation. ³¹ , ³²

When All of Us participants enrolled, blood samples were drawn for genomic analysis. ³³ The All of Us genomic database contains over 310,00 genotyping array samples, over 245,000 short‐read whole genome sequencing, and over 11,300 long‐read whole genome sequencing. The All of Us Genome Center and Data and Research Center developed a research pipeline for their genetic data. Quality control steps are applied before releasing the data for research use. These steps are applied independently and across samples. A detailed report about their pipeline can be found on their website. ³⁴

Statistical analysis

Categorical variables are presented as frequencies and percentages, while continuous variables are presented as medians with interquartile ranges (IQR) or means with standard deviation (SD). To compare demographic and clinical characteristics between cases and controls, Fisher's exact or Pearson's Chi‐squared tests were used for categorical variables, and Wilcoxon's rank sum tests were used for continuous variables.

We conducted genome‐wide and transcriptome‐wide association studies (GWAS, TWAS) for acute pancreatitis and pancreatic injury, adjusting for 10 principal components of ancestry (PCA), sex, age at event, and azathioprine indication. Acknowledging the relatively small overall cohort and possible imbalance in cases vs. controls, we used the Firth logistic regression method, which is a penalized likelihood‐based estimation.

GWAS analyses for both cohorts were conducted using PLINK (version 2). To account for multiple testing, GWAS statistical significance for the discovery cohort (BioVU) was set at P < 5 × 10⁻⁸ and TWAS significance was set at P < 5 × 10⁻⁶. We set the statistical significance for the validation cohort (All of Us) at P < 0.05.

The BioVU portion of the study was reviewed by the Institutional Review Board from Vanderbilt University Medical Center (#180498) and the University of Miami (#20221355), which categorized them as non‐human subjects research. The use of data from All of Us is covered under a single IRB and is considered not human subjects research. ³⁵

RESULTS

Baseline characteristics and outcomes

Discovery cohort

Natural language processing identified 10,271 possible azathioprine users. From them, 5937 either had DNA available for genotyping or had already been genotyped. After internal quality control, we received data from 5064 patients. After excluding non‐users and applying other inclusion and exclusion criteria, 2127 eligible azathioprine users remained (35.4% male; mean age: 44.5 years, SD: 17.2) (Figure S1 ).

A total of 16 individuals (0.76%) developed pancreatitis and 42 individuals (1.97%) developed pancreatic injury; the remaining 2085 patients qualified as controls. Table 1 provides demographic and clinical comparisons of acute pancreatitis or pancreatic injury cases vs. controls. There were no statistically significant differences among cases vs. controls with respect to age at index date, initial weight, sex, or indication in either primary or secondary analysis. The initial azathioprine dose was significantly higher among cases vs. controls (pancreatitis: 131.7 ± 75.3 vs. 86.6 ± 50.2, P = 0.036; pancreatic injury: 109.8 ± 61.5 vs. 86.6 ± 50.2, P = 0.021).

Table 1.

Demographic and clinical characteristics of azathioprine users who developed pancreatitis (ICD code or amylase/lipase three times upper limit of normal) or pancreatic injury (amylase/lipase two times upper limit of normal) vs. controls (those without evidence of prior pancreatic compromise) in the BioVU cohort

Pancreatitis	Case (n = 16)	Control (n = 2085)	P‐value
Age in years, median [IQR]	47.7 [24.1–56.7]	44.6 [30.9–57.9]	0.507
Male sex, n (%)	8 (50.0%)	740 (35.5%)	0.344
Baseline weight (kg), mean ± SD	81.6 ± 14.9	79.0 ± 23.1a	0.518
Baseline dose (mg/day), mean ± SD	131.7 ± 75.3b	86.6 ± 50.2c	0.036
Indications
Inflammatory bowel disease, n (%)	9 (56.2%)	837 (40.1%)	0.292
Other autoimmune, n (%)	7 (43.8%)	1267 (60.8%)	0.258
Follow‐up months, median [IQR]	1.33 [0.9–29.2]	23.2 [4.0–64.4]	0.008

Pancreatic injury	Case	Control	P‐value
	(n = 42)	(n = 2085)
Age in years, median [IQR]	44.4 [25.1–56.4]	44.6 [30.9–57.9]	1.197
Male sex, n (%)	13 (31.0%)	740 (35.5%)	0.656
Baseline weight, mean ± SD	71.8 ± 16.9d	79.0 ± 23.1a	0.024
Baseline dose, mean ± SD	109.8 ± 61.5b	86.6 ± 50.2c	0.021
Indications
Inflammatory bowel disease, n (%)	23 (54.8%)	837 (40.1%)	0.080
Other autoimmune, n (%)	20 (47.6%)	1267 (60.8%)	0.117
Follow‐up month, median [IQR]	3.7 [0.8–29.5]	23.2 [4.0–64.4]	< 0.001

^a n = 2013.

^b n = 15.

^c n = 2070.

^d n = 41.

Validation cohort

In the All of Us cohort, we identified 857 azathioprine users who met the inclusion criteria. The majority were females (69.5%) with a mean age of 49.0 years ± 16.5. There were < 20 cases of acute pancreatitis or pancreatic injury. Age at index, initial weight, sex, or indication did not differ between cases and controls in either the primary or secondary analysis (Table S2 ).

GWAS results

Discovery cohort

Adjusting for 10 PCA, sex, age at event, and indication, the GWAS identified 32 SNPs associated with acute pancreatitis in BioVU (Figure 1 ; Table 2 ). For the secondary outcome of pancreatic injury, the BioVU‐adjusted GWAS identified 8 SNPs significantly associated with pancreatic injury (Table 3; Figure S2 ).

Figure 1.

Analyses of pancreatitis adjusted for 10 principal components of ancestry, sex, age at event, and indication in BioVU: (a) GWAS Manhattan plot with the chromosome position on the x axis and log₁₀ of association P‐value on the y axis. The red line indicates the Bonferroni corrected P‐value. (b) QQ plot (genomic inflation factor = 1.0654).

Table 2.

Significant SNPs from pancreatitis GWAS analysis in BioVU cohort. Analysis is adjusted for 10 principal components of ancestry, sex, age at event, and indication

Chr	POS	RSID	GENE	REF	ALT	MAF	OR	P
1	71758154	rs115416871		G	A	0.01	17.81	3.25E‐08
1	238818794	rs16837283	RYR2	T	C	0.09	6.55	3.56E‐08
3	179980191	rs112560658		C	A	0.01	17.85	3.87E‐08
4	61286605	rs114106156		C	A	0.01	19.70	1.35E‐08
5	40488659	rs75236196	C6	T	C	0.07	7.25	1.51E‐08
5	40494153	rs77091262		G	A	0.07	7.21	1.66E‐08
5	40498181	rs78870278		C	T	0.07	7.56	8.04E‐09
5	40498331	rs149137776		G	A	0.07	7.68	6.07E‐09
5	40549622	rs113700968		G	A	0.07	7.21	1.69E‐08
5	40592753	rs79546139	C6 and PTGER4	A	G	0.07	7.54	8.55E‐09
5	40597004	rs75627498		T	C	0.07	7.51	9.33E‐09
5	40599286	rs16870146	LOC105374737	C	A	0.06	7.95	1.47E‐08
5	40603203	rs16870155	LOC105374737	C	T	0.06	8.13	1.10E‐08
5	40605039	rs113854143	LOC105374737	C	T	0.06	8.13	1.10E‐08
5	40608160	rs16870171	C6, PTGER4	A	G	0.06	8.17	1.06E‐08
5	40608885	rs79770300	LOC105374737	T	G	0.06	8.17	1.06E‐08
5	40612802	rs112596015	LOC105374737	G	A	0.06	8.41	5.64E‐09
5	40620153	rs78257075	LOC105374737	G	C	0.05	8.84	4.05E‐09
5	40620290	rs78311621	LOC105374737	A	G	0.06	8.40	5.85E‐09
6	33027782	rs530948827		T	C	0.01	20.02	3.05E‐08
6	116506494	rs117620011	COL10A1 and NT5DC1	A	G	0.01	22.25	1.16E‐08
7	108842895	rs78275279		A	G	0.03	14.47	1.90E‐09
7	108939568	rs117753839		G	A	0.03	13.94	2.86E‐09
7	109012048	rs74662268		G	A	0.02	18.28	4.07E‐09
7	109059687	rs144401393		G	C	0.03	14.29	2.69E‐09
8	51135288	rs151252309	SNTG1 and LOC105375829	A	G	0.02	15.28	3.05E‐08
8	126992614	rs72728319		A	G	0.05	9.02	4.51E‐08
10	86058851	rs138986187		G	A	0.02	17.96	8.43E‐09
10	86176424	rs147318267	CCSER2	T	A	0.01	18.44	2.58E‐09
10	86251297	rs181321593	CCSER2	G	T	0.02	17.08	4.26E‐09
10	86288498	rs143005444		T	C	0.01	18.47	2.14E‐09
16	11645068	rs79245913	LITAF	C	T	0.02	16.88	1.34E‐09

Table 3.

Significant SNPs from pancreatic injury GWAS analysis in BioVU cohort. Analysis is adjusted for 10 principal components of ancestry, sex, age at event, and indication

CHR	POS	RSID	GENE	REF	ALT	MAF	OR	P
1	23051264	rs72653622	EPHB2	G	A	0.02	8.59	1.84E‐08
5	25132974	rs200407865	LINC02228	G	T	0.01	12.60	9.76E‐09
7	140125750	rs2948386	RAB19	T	C	0.26	3.47	1.46E‐08
9	5525874	rs147662134	CD274, PDCD1LG2, PLGRKT	T	C	0.06	4.80	2.49E‐08
18	41222682	rs113528748	NA	A	G	0.01	10.67	1.67E‐08
18	41481974	rs117940420	LOC105372088	T	G	0.01	11.12	1.05E‐08
18	41619694	rs142640276	NA	C	G	0.01	11.61	6.61E‐09
18	41710279	rs191224981	NA	T	C	0.01	10.21	2.77E‐08

Validation cohort

There was one SNP associated with acute pancreatitis in BioVU (rs115416871) that replicated in All of Us (Table 4 ; P = 2.31E‐3); it is not associated with any known genes. For pancreatic injury, we identified one SNP in RAB19, rs2948386, from BioVU that replicated in All of Us (Table 4 ; P = 4.18E‐3).

Table 4.

Significant SNPs from GWAS found in BioVU that replicated in the All of Us (AOU) Cohort. Analysis is adjusted for 10 principal components of ancestry, sex, age at event, and indication

	CHR	POS	RSID	Gene	BioVU					AOU
					REF	ALT	MAF	OR	P	REF	ALT	MAF	OR	P
Acute pancreatitis	1	71758154	rs115416871	NA	G	A	0.01	17.81	3.25E‐08	G	A	0.01	7.23	2.31E‐03
Pancreatic injury	7	140125750	rs2948386	RAB19	T	C	0.26	3.47	1.46E‐08	T	C	0.19	2.70	4.181E‐03

TWAS results

We performed a TWAS by calculating the predicted expression levels of genes in both pancreatic and liver tissue. In the BioVU dataset, focusing on pancreatic tissue, we found the predicted expression of SERPINB9P1 was significantly associated with pancreatitis and pancreatic injury. When the analysis included liver tissue, the predicted expression of SERPINB9P1 was above our predefined significance threshold (effect size: 0.80; P‐value: 5.18E‐05). The results for pancreatic injury were successfully replicated in the All of Us dataset (Table 5 ). All the analyses were adjusted for 10 PCAs, sex, age at event, and indication.

Table 5.

Significant associations between the SERPINB9P1 gene and the phenotypes. Predicted expression is using pancreatic and liver tissue. Results are adjusted for 10 principal components of ancestry, sex, age at event, and indication

Phenotype	Tissue	Bio VU		AOU
		Effect size	P‐value	Effect size	P‐value
Pancreatitis	Pancreas	0.62	4.14E‐06	0.22	2.50E‐01
Pancreatic injury	Pancreas	0.42	1.48E‐05	0.48	1.11E‐02
Pancreatic injury	Liver	0.80	5.18E‐05	1.35	3.00E‐04

DISCUSSION

In this study, we aimed to identify genetic predictors of azathioprine‐related acute pancreatitis and pancreatic injury among patients taking azathioprine for a variety of conditions. A single variant in RAB19 (rs2948386) within the BioVU cohort was associated with pancreatic injury and replicated in the cohort All of Us Cohort (Table 4 ). Additionally, we found that the genetically predicted expression of SERPINB9P1 was associated with both pancreatitis and pancreatic injury in BioVU, and this association was replicated in the All of Us cohort for pancreatic injury (Table 5 ).

RAB19 encodes a protein that interacts with ATG16, a core component of autophagy machinery. ³⁶ Autophagy is a cellular process responsible for degrading damaged or unwanted organelles and proteins. Previous GWAS studies have shown an association between ATG16 variants and the autophagy pathway with susceptibility to Crohn's disease. ³⁷ , ³⁸ , ³⁹ Studies in drosophila demonstrate that genetic variations in RAB19 predispose to the development of IBD, ³⁶ and RAB19 is also a key protein in primary ciliogenesis. ⁴⁰ However, there has been no clear mechanism underlying any potential role of RAB19 in pancreatitis identified to date.

SERPINB9P1 is a long non‐coding RNA (lncRNA) member of the serine protease inhibitor family, which plays important roles in regulating blood clotting, inflammation, immune response, and complement activation. ⁴¹ To our best knowledge, there is no information regarding how this gene could be linked to pancreatitis.

The previous pancreatitis GWAS included 2207 patients with IBD who were of European ancestry; it identified a variant in HLA‐DQA1‐HLA‐DRB1 associated with TIAP (172 cases). The SNP, rs2647087, conferred a nearly threefold higher odds of TIAP. Patients who were heterozygous for rs2647087 had a 9% risk of TIAP, whereas homozygotes had a 17% risk of TIAP. ¹⁶ The study restricted cases to those who developed TIAP within 3 months of thiopurine initiation. The HLA haplotype was later confirmed in a candidate SNP study in IBD patients (cases = 13; controls = 360) from a tertiary care center in London, Canada, ⁴² and recently a paper demonstrated that prescreening for the variant resulted in an 11‐fold reduction in the incidence of azathioprine‐induced pancreatitis. ⁴³ While our results indicated a trend toward higher risk for pancreatitis associated with variants in the same SNP, these results in BioVU (pancreatic injury: OR = 1.38, P‐value = 0.15; pancreatitis: OR = 1.36, P‐value = 0.35) were not statistically significant. The GWAS platform had two SNPs in the discovery cohort that have been associated with thiopurine‐induced pancreatitis in prior research (rs2647047 and rs12415432). ¹⁶ In our study, the point estimate for rs2647047 and rs12415432 suggested a trend toward higher risk, but the results were statistically non‐significant (OR = 1.28, P = 0.42 and OR = 1.50, P = 0.49, respectively). This may be due to limited power.

Our study extends the current literature by (i) analyzing both pancreatic injury and pancreatitis, (ii) including patients with other immune‐mediated conditions apart from IBD, (iii) ensuring fidelity of the exposure in our discovery cohort by confirming that patients were active users of azathioprine, and (iv) leveraging TWAS for functional analyses. We also demonstrate the use of an EHR data repository linked to genomic data to identify novel variants and new gene‐trait associations without relying solely on large‐scale GWAS analyses. This study is also a proof‐of‐concept of the feasibility of the use of real‐world data linked to genomic data to identify novel variants. Our study is not without limitations. In the BioVU cohort, we were able to conduct record review to confirm cases of pancreatic injury/pancreatitis occurred during azathioprine use; however, this was not possible in the validation cohort. Like the Heap et al. study, ¹⁶ our study was not a multi‐ethnic study, and thus future studies are needed in non‐European ancestry populations. Lastly, given that we were not able to confirm the SNP in HLA (rs2647087), it is possible that we were underpowered, given that pancreatic injury/pancreatitis are still rare events.

In conclusion, we identified a novel SNP in RAB19 (rs2948386) associated with pancreatic injury, and that the predicted expression of SERPINB9P1 was linked to both pancreatitis and pancreatic injury. We can hypothesize that the effects of RAB19 may be mediated through alterations in autophagy, and the role of SERPINB9P1 likely involves DNA repair; however, further studies are needed to identify the underlying mechanisms.

AUTHOR CONTRIBUTIONS

S.C.S., L.L.D., P.N., T.M.‐F., P.S., J.M., E.J.P., and C.P.C. wrote the manuscript; S.C.S., T.S.R., L.L.D., and J.Z. designed the research; T.S.R., L.L.D., and A.L.D. performed the research; L.L.D., J.Z., A.L.D., and P.N. analyzed the data; A.M.H., R.T., W.‐Q.W., N.J.C., C.M.S., and Q.F. contributed new reagents/analytical tools.

CONFLICT OF INTEREST

The authors declared no competing interests for this work.

FUNDING INFORMATION

This study was supported by National Institute of General Medical Sciences (NIGMS) grants R01GM126535, R01GM109145, and R35GM131770, and by the Rheumatology Research Foundation K‐supplement and R01 bridge awards. CPC was supported by National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) grant R01AR073764 and The Vanderbilt Institute for Clinical and Translational Research (VICTR) grant 2UL1TR000445‐06 from supported by the National Center for Advancing Translational Sciences (NCATS). The datasets used for the analyses described were obtained from Vanderbilt University Medical Center's BioVU and the All of Us Research Program. BioVU projects are supported by institutional funding, private agencies, and federal grants. The grants include the National Institutes of Health (NIH)‐funded Shared Instrumentation Grant S10OD017985 and S10RR025141; CTSA grants UL1TR002243, UL1TR000445, and UL1RR024975. Genomic data are also supported by investigator‐led projects that include U01HG004798, R01NS032830, RC2GM092618, P50GM115305, U01HG006378, U19HL065962, and R01HD074711; and additional funding sources available at https://victr.vumc.org/biovu‐funding/. The All of Us Research Program is supported by the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026553; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA #: AOD 16037; Federally Qualified Health Centers: HHSN 263201600085U; Data and Research Center: 5 U2C OD023196; Biobank: 1U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1U24 OD023163; Communications and Engagement: 3 OT2 OD023205; 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276. In addition, the All of Us Research Program would not be possible without the partnership of its participants. The funding sources had no role in the collection, analysis, or interpretation of data, writing of the manuscript, or decision to submit for publication.

Supporting information

Data S1.

DATA AVAILABILITY STATEMENT

This study used data from the All of Us Research Program’s Controlled Tier Dataset v7, available to authorized users on the Researcher Workbench.