Pancreatic cancer is the seventh leading cause of cancer-related deaths worldwide, climbing to the fourth in developed countries (Sung et al., 2021). The notably low 5-year overall survival rate, at approximately 10%, positions pancreatic cancer to potentially become the second leading cause of cancer-related deaths in the United States by 2030 (Park et al., 2021). Pancreatic ductal adenocarcinoma (PDAC), an exocrine cell tumor, represents the most prevalent type accounting for over 90% of all pancreatic cancer cases. The high mortality rate of PDAC can be attributed to the lack of effective biomarkers for detection and challenges in diagnostic imaging. In addition, existing chemotherapy treatments for PDAC, such as FOLFIRINOX, which is comprised of folinic acid, 5-fluorouracil, irinotecan, and oxaliplatin, or gemcitabine-based regimen, prolong patient survival only by a few months (Conroy et al., 2011; Von Hoff et al., 2013). This underscores the urgent need for novel targets for diagnosis and therapy. PDAC cells exhibit a rewired metabolism that supports their proliferation in their environment with low oxygen and scarce nutrient (Encarnación-Rosado and Kimmelman, 2021). These cells show diverse shifts in energy resource use, including increased glucose uptake and glycolysis (Ying et al., 2012), non-canonical use of glutamine into tricarboxylic acid (TCA) cycle for redox homeostasis (Son et al., 2013), elevated levels of lysosomal autophagy (Yang et al., 2011), and macropinocytosis (Kamphorst et al., 2015). Such metabolic adaptations might offer avenues for therapeutic intervention (Suzuki et al., 2020).

Numerous studies have sought to identify cancer biomarkers employing omics approaches, focusing on diagnosis, prognosis, and chemotherapy resistance (Hristova and Chan, 2019; Cheong et al., 2022). Still, many of these studies lack validation with orthogonal approaches and/or often present simple association without causality establishment. For PDAC, although many diagnostic or prognostic biomarkers have been proposed representing lncRNA (Li et al., 2014), miRNA (Guo et al., 2018), mRNA (Boyd et al., 2023), protein (Kim et al., 2017; Nam et al., 2022), metabolite (Mayerle et al., 2018) or multi factor panels (Yang et al., 2020), most of these studies have been conducted using blood samples. While this may be due to the practical difficulty of obtaining many PDAC tissues, serum or plasma reflects systemic changes that may not be specific to PDAC. In this context, a tissue biomarker with relevance across mRNA, protein, and metabolite levels along with the causal relationship in PDAC is in need to understand the altered metabolism of PDAC and to explore potential therapeutic strategies.

Taurine, or 2-aminoethanesulfonic acid, is a semi-essential, non-proteinogenic amino acid and is one of the most abundant amino acids in eukaryotes, constituting up to 0.1% of human body weight (Ripps and Shen, 2012). Taurine plays pivotal roles in various biological processes, including bile acid conjugation, cardiovascular protection, central nervous system function, and anti-oxidation (Lourenco and Camilo, 2002). A recent study showed that taurine concentrations in humans inversely correlated with age-related diseases such as hypertension, inflammation, and type 2 diabetes, suggesting a potential link between taurine deficiency and aging (Singh et al., 2023). For cancer, taurine has exhibited inhibitory effects on various cancer cells, including lung (Tu et al., 2018), colon (Zhang et al., 2014), nasopharynx (He et al., 2018), and cervix (Li et al., 2019), indicating a potentially broad implication in oncological processes. Notably, taurine has also been identified as a diagnostic marker of breast cancer, with its low serum levels in affected individuals and high-risk groups (El Agouza et al., 2011). Despite these studies on taurine, its role in PDAC has not been explored.

ADO, or 2-aminoethanethiol dioxygenase, is a thiol dioxygenase in humans that adds two oxygen atoms to cysteamine, forming hypotaurine (Dominy et al., 2007), which is further oxidized to taurine. This reaction could occur either enzymatically or non-enzymatically (Grove and Karpowicz, 2017). In both plant and animal cells, ADO has been shown to facilitate the oxidation of N-terminal cysteinyl residues into sulfinic acid in proteins, promoting proteasomal degradation (Masson et al., 2019). For cancer, ADO protein has been associated with high-grade, more severe glioblastomas, indicating its potential role in glioma progression or severity (Gao et al., 2016). Other than that, the roles of ADO in cancer have not been studied.

In this study, we investigated the PDAC-specific metabolic pathways by applying multi-omics on PDAC tissues and explored their association with patients’ clinical phenotypes.

Patients and clinical samples

Pancreatic tissues were collected at Asan Medical Center (Seoul, Korea) from 2011 to 2020. A total of 108 samples (54 pairs of each tumor and matched normal tissues) were collected from patients diagnosed with PDAC. Written informed consent had been obtained from the patients. The study protocol was approved by the institutional review board (IRB No.: 2019-0185). Tissue samples were snap-frozen in liquid nitrogen, transferred to cryotubes and stored at −80°C. Tissue samples were transported to the measurement laboratory in boxes filled with dry ice and stored at −80°C until metabolomics analysis.

LC-HRMS metabolomics analysis

For metabolite extraction, 40 mg of tissue samples were weighed, snap-frozen in liquid nitrogen, and pulverized. Samples were lysed in a 600 μL of methanol/chloroform (2:1) followed by three cycles of vortexing, freezing in liquid nitrogen, and thawing at room temperature. After adding 400 μL of chloroform/water (1:1) mixture, samples were centrifuged at 15,000×g, 4°C for 20 min. The upper phase containing metabolites was separated, dried in a vacuum evaporator, and redissolved in 100 μL of acetonitrile/water (1:1) for analysis, with same volume of samples pooled for quality control (QC) samples. For the metabolite extraction from cells, incubated cells were lysed and processed following our previously reported protocol (Oh et al., 2023).

For untargeted liquid chromatography-high resolution mass spectrometry (LC-HRMS) analysis, a Q Exactive Plus Hybrid Quadrupole-Orbitrap Mass Spectrometer coupled with Vanquish UHPLC (Thermo Scientific, Waltham, MA, USA) was used. Chromatic separation occurred on a BEH amide column (Waters 186004801, Milford, MA, USA) under column at 40°C and autosampler at 4°C. The QC sample was injected every ten samples. Two microliters of analytes were injected and eluted with a mobile phase of 10 mM ammonium acetate in water (A, pH 10.0) and 80% acetonitrile (B, pH 10.0). Gradient conditions were as follows: 0-2 min 100% B, 2-7.5 min gradient 100-60% B, 7.5-16 min 60% B, 16.5-20 min gradient 60-100% B, under flow rate of 0.2 mL/min at 0-16.5 min, and 0.3 mL/min at 16.5-20 min. Both positive and negative modes for tissue samples, and only positive mode for cell extracts were used, detecting mass range of 70-900.

LC-HRMS data were analyzed using MZmine 2.53, where chromatograms were generated and peaks isolated, identifying compounds via PubChem and the Human Metabolome Database. The sum of peak intensities from four features in ESI(+) mode was calculated as the intensity of taurine in tissue samples. The taurine peak from cell samples was detected using Thermo Xcalibur software (Thrermo Scientific, Waltham, MA, USA) with base peak m/z=122.022 and normalized against the total chromatogram peak sum.

Immunohistochemistry of tissue microarray

Two human tissue microarrays (PA1001c and PA2072a, US Biomax, Derwood, MD, USA) with normal pancreas (45 cores, 19 cases) and PDAC (258 cores, 99 cases) were used to assess taurine levels, with tumor stage, metastasis and grade details provided by the manufacturer. For immunostaining, 8-μm sections underwent deparaffinization, antigen retrieval in citrate buffer (pH 6.0), peroxidase quenching with 3% H₂O₂ in phosphate-buffered saline (PBS), water rinse, and blocking with goat serum. Primary antibody incubation was done at 4°C for two days using a 1:30 anti-taurine antibody (AB5022, Sigma-Aldrich, St. Louis, MO, USA), followed by biotinylated secondary antibodies (1:60, BA-1000, Vector Laboratories, Newark, CA, USA) for 2 h at room temperature. Sections were then washed, treated with avidin-biotin complex, developed with 3,3′-diaminobenzidine, and counterstained with hematoxylin. One human tissue microarray (PA1001c, US Biomax) with normal pancreas (20 cores, 10 cases) and PDAC (78 cores, 39 cases) was used to examine the expression level of ADO. Immunostaining for ADO was performed in same protocol as taurine with modified antibody reaction, with an anti-ADO antibody (1:50, sc-515318, Santa Cruz, Dallas, TX, USA) and a 1:100 dilution of secondary antibody. All staining was scored by four levels (0 for no, 1 for weak, 2 for moderate, and 3 for strong staining).

Bioinformatics screening on transcriptomic and proteomic data

The gene expression and the patient’s survival information were from the UCSC Xena website (The Cancer Genome Atlas [TCGA] TARGET The Genotype-Tissue Expression [GTEx] cohort, https://xenabrowser.net). In PDAC, tumor tissues (n=179) were all from TCGA, and among the corresponding normal tissues (n=171), 4 were from TCGA and the others were from GTEx. Gene expression profiles across various cancer types were analyzed through the GEPIA web server (http://gepia.cancer-pku.cn) (Tang et al., 2017). The protein expression data were downloaded from the Proteomic Data Commons website (Clinical Proteomic Tumor Analysis Consortium [CPTAC] cohort https://pdc.cancer.gov/pdc/cptac-pancancer). In PDAC, tumor (n=137) and normal (n=74) tissues were compared for protein expression related to taurine metabolism. Comparisons of protein expression profiles across various cancer types were conducted using the UALCAN portal (https://ualcan.path.uab.edu) (Chandrashekar et al., 2022).

Cell culture and transfection

PANC-1, a human pancreatic cancer cell line, was purchased from Korean Cell Line Bank. Cells were grown in DMEM (LM 011-01, Welgene, Gyeongsan, Korea) supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin, at 37°C incubator. Three siRNAs (control, ADO-KD1, and ADO-KD2) were generated by Bioneer (Daejeon, Korea), and the sequences for ADO silencing followed those from a previous study (Shen et al., 2021) and listed in Supplementary Table 1. For transfection, PANC-1 cells (1.2×10⁶) were seeded in a 60-mm plate, incubated overnight, and then transfected with 60 picomoles of siRNA using Lipofectamine RNAiMAX (Invitrogen, Waltham, MA, USA) according to the manufacturer’s protocol. After 24 h, transfected cells were transferred 96-well or 6-well plates and cultured for two days more for cell growth assay or RNA and metabolite extraction.

Quantitative real-time PCR (qPCR) and western blot analysis

Total RNA was extracted using Easy-Spin Total RNA Extraction Kit (iNtRON Biotechnology, Seongnam, Korea), and cDNA was synthesized with the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Waltham, MA, USA), all according to the manufacturer’s protocols. The primer sequences are listed in Supplementary Table 2. qPCR was performed using TOPreal SyBR Green qPCR Premix (Enzynomics, Daejeon, Korea) on an Applied Biosystems Prism 7300 system. Western blot experiment was conducted using primary antibodies against ADO (16479-1-AP, Proteintech, Rosemont, IL, USA), β-actin (sc-47778, Santa Cruz), p65 (sc-8008, Santa Cruz), phospho-p65 (#3033, CST, Danvers, MA, USA), MEK1 (#9124, CST), phospho-MEK1 (Thr286) (#9127, CST), phospho-MEK1/2 (Ser221) (#2338, CST), ERK1/2 (#9102, CST), and phospho-ERK1/2 (#4377, CST).

Cell growth assay and clonogenic assay

Transfected PANC-1 cell (1.0×10⁴ cells per well) were seeded into 96-well plates and incubated for 6 h for cell adhesion. After the cells were adhered, 10 µL of CCK solution (DonginLS, Seoul, Korea) was added to each well, followed by 2 h incubation. The measurement of absorbance at 450 nm was conducted daily for six days. For the clonogenic assay, 100 cells were seeded into 6-well plate. The cells were cultured for 20 days in DMEM supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin. Once colonies had formed, the plate was washed with PBS and stained with 0.5% crystal violet solution for 30 min. Following staining, the plate was gently rinsed and left to dry before imaging was conducted.

Animal experiment

Six-week-old female BALB/c mice were procured from Orient Bio (Gapyeong, Korea). All animal procedures were conducted following institutional guidelines and received approval from the Seoul National University Institutional Animal Care and Use Committee (Approval No. 2006241-2E). Prior to experimentation, the mice underwent a two-week acclimatization period. For tumor induction, PANC-1 cells, transfected with either control siRNA or ADO-KD1, were suspended in PBS and combined with an equal volume of Matrigel (Corning 354248, Corning, NY, USA). Two hundred microliters of mixture containing 5.0×10⁶ cells were injected subcutaneously into the right flank of the mice. Following a month post-injection, body weight and tumor growth were monitored every ten days. Tumor volume (in mm³) was calculated using the formula: [length (mm)×width (mm)×width (mm)]/2.

Statistical analysis

Descriptive statistical analyses, encompassing paired, Student’s, and Welch’s t-test, Mann-Whitney U test, and receiver operating characteristic (ROC) analysis, were executed using GraphPad Prism 9.3 (GraphPad, Boston, MA, USA). Kaplan-Meier analysis was conducted in the Kaplan-Meier Plotter (https://kmplot.com). All statistical tests were two-sided, and p<0.05 was considered statistically significant. For multivariate analysis, LC-HRMS data underwent normalization through mean centering and Pareto scaling in SIMCA-P 11.0 (Umetrics, Umea, Sweden).

Participant characteristics

The demographic and clinical characteristics of the patients in this study are presented in Table 1. Fifty-four pairs of normal pancreas and tumor tissues from patients with PDAC were obtained during surgery. Of these 54 patients, 38 and 16 were diagnosed as stage I-II and III-IV, respectively [stage IA (n=4), IB (n=14), IIA (n=6), IIB (n=14), III (n=11), and IV (n=5)]. Classification of tumor stages was according to the 8th edition of the American Joint Committee on Cancer (Chun et al., 2018). Among the 54 patients, a total of 16 (29.6%) individuals either experienced no recurrence of the cancer or were subject to censoring during the study period. Conversely, 38 (70.4%) patients encountered cancer recurrence.

Table 1 Demographic and clinical characteristics of patients (n=54)

Characteristics		n (%)
Gender	Male	30 (55.6)
	Female	24 (44.4)
Age	<60	24 (44.4)
	≥60	30 (55.6)
Pre-operational CA19-9	<35 U/mL	18 (33.3)
	≥35 U/mL	36 (66.7)
AJCC stage	IA	4 (7.4)
	IB	14 (25.9)
	IIA	6 (11.1)
	IIB	14 (25.9)
	III	11 (20.4)
	IV	5 (9.3)
Tumor grade	1 (Well differentiated)	5 (9.3)
	2 (Moderately differentiated)	35 (64.8)
	3 (Poorly differentiated)	6 (11.1)
	4 (Undifferentiated)	1 (1.9)
	N/A	7 (13.0)
Lymphovascular invasion	No	21 (38.9)
	Yes	32 (59.3)
	N/A	1 (1.9)
Perineural invasion	No	12 (22.2)
	Yes	42 (77.8)
Adjuvant chemotherapy	No	15 (27.8)
	Yes	39 (72.2)
Recurrence	No or censored	16 (29.6)
	Yes	38 (70.4)

CA19-9, carbohydrate antigen 19-9; AJCC, American Joint Committee on Cancer; N/A, not available.

Discovery of taurine as a PDAC marker via untargeted metabolomics analysis

We performed untargeted metabolomics analysis on 54 pairs of normal pancreas and pancreatic tumor samples using liquid chromatography-high resolution mass spectrometry (LC-HRMS). Comprehensive analysis of tissue metabolic profiles was conducted, followed by multivariate statistical analysis to discern differences between normal and tumor samples. The unsupervised multivariate principal component analysis (PCA) score plot demonstrated close clustering of quality control (QC) samples, indicating instrumental stability (Fig. 1A). An orthogonal projections to latent structure-discriminant analysis (OPLS-DA) model was obtained using one predictive (Pp) and two orthogonal components (Po) to discriminate the normal and PDAC samples and to identify the relevant biomarkers. The OPLS-DA score plot showed a good separation between normal and tumor groups with good predictive values (R2=0.737, Q2= 0.543; Fig. 1B). Among 296 features detected from both ESI(+) and ESI(−) modes of LC-HRMS, 109 features exhibited a significant decrease (p<0.05) and 24 features a significant increase (p<0.05) in tumor samples upon univariate analysis (Fig. 1C). Twelve features were identified as tumor markers, while 54 features were identified as normal markers. A complete list of identified marker features is listed in Supplementary Table 3. Out of 12 identified features enriched in tumor, five features belonged to taurine, and were significantly higher in tumor group (Supplementary Fig. 1). In a comparison of 54 matched pairs, the amount of total taurine was significantly higher in tumor than in normal tissues, (fold change=2.51, p<0.0001; Fig. 1D), and the ROC analysis yielded an area under the curve (AUC) of .749 (95% CI=0.657 to 0.840; Fig. 1E).

Figure 1. Discovery of taurine as a PDAC marker via untargeted metabolomics analysis. (A) PCA score plot of metabolite features with principal component 1 and 2 (PC1, PC2). Samples from normal tissues (n=54, blue), tumor tissues (n=54, red), and quality control (QC, n=11, black) are plotted. (B) OPLS-DA score plot illustrating the clear separation between normal and tumor samples, achieved with one predictive (Pp) and two orthogonal components (Po). (C) Volcano plot derived from LC-HRMS data, showcasing the differential metabolic features between normal and tumor samples. Features of taurine are distinctly marked in red. Two-sided Student’s t-test or Welch’s t-test was used following the assessment of variance homogeneity with an F-test. (D) Taurine levels in paired tumor samples (n=54) compared to normal samples (n=54). Two-sided paired t-test was used. A.U.=arbitrary unit. (E) ROC curve analysis demonstrating the diagnostic performance of taurine levels from 108 samples (normal and tumor, each n=54). AUC=area under the curve. ****p<0.0001.

Elevated taurine level in PDAC and its association with prognosis

To validate the elevated taurine level in an independent group, we used pancreatic tissue microarrays that include both normal pancreas (47 cores, 19 cases) and PDAC tissues (258 cores, 99 cases) obtained commercially. In addition, we employed an orthogonal modality for measuring taurine level (IHC using antibody) to avoid a potential bias from the metabolomics approach. The IHC staining showed a significant elevation of taurine level in the PDAC tissues (p<0.001; Fig. 2A), confirming our metabolomics results. Next, we tested if the elevated taurine level bears any relation with clinical information. In terms of PDAC stages, taurine levels were significantly higher in PDAC stage I and II (p<0.001 and p=0.001, respectively) and maintained the elevated trend in PDAC with combined stage III-IV (p=0.09; Fig. 2B). There was no significant difference in the staining score between stage I, stage II, and stage III-IV PDAC. In our clinical cohort, taurine levels showed higher trend in stage I (p=0.056) and significant increase in other stages (stage II, p=0.002; combined stage III-IV, p<0.001; Fig. 2C). Moreover, taurine levels were elevated in both primary (p<0.001) and metastatic tumors (p=0.003; see Fig. 2B) in the tissue microarray. Consistently, our clinical cohort exhibited elevated taurine levels in both primary (p<0.001) and metastatic tumors (p<0.001; see Fig. 2C). In terms of tumor grade, both the tissue microarray and our clinical cohort demonstrated an elevated level of taurine in PDAC tissues, starting from the low-grade tumors (see Fig. 2B, 2C). These findings from two independent cohort using orthogonal measurement modality show that taurine levels are elevated from the early stages of PDAC and sustain this elevation throughout its progression. Then, we conducted a Kaplan-Meier analysis based on taurine levels in the metabolomics cohort of 54 PDAC patients (Fig. 2D). For overall survival (OS), the high taurine group trended towards a shorter survival post-operation than low taurine group but did not meet statistical significance (Log-rank p=0.20, Hazard Ratio=1.58 [0.79-3.17]). Notably, however, for disease-free survival (DFS), the high taurine group (n=28) showed significantly more and faster onset of cancer recurrence (Log-rank p=0.023, Hazard Ratio=2.15 [1.09-4.21]) than the low taurine group (n=26). Overall, taurine levels were elevated in PDAC tissues from early stages and associated with poorer disease-related prognosis.

Figure 2. Elevated taurine level in PDAC and its association with prognosis. (A) Representative images of immunohistochemistry (IHC) of taurine in PDAC tissue microarrays, comprising normal (n=47, 19 cases) and tumor tissues (n=258, 99 cases) (scale bar=30 µm) (Left), and IHC scores of taurine (Right). Taurine staining was scored by 4 levels: 0 for none, 1 for weak, 2 for moderate, and 3 for strong staining. Two-sided Mann Whitney U test was used. (B) Taurine levels of tissue microarrays used in (A) by stage (Left) (normal, n=47; stage I, n=99; stage II, n=105; stage III and IV, n=54), metastatic status (Middle) (normal, n=47; primary, n=174; metastasis, n=84), and tumor grade (Right) (normal, n=47; grade 1, n=57; grade 2, n=103; grade 3, n=64). Two-sided Mann Whitney U test was used. (C) Taurine levels of patient cohort by stage (Left) (normal, n=54; stage I, n=18; stage II, n=20; stage III and IV, n=16), metastatic status (Middle) (normal, n=54; primary, n=49; metastasis, n=5), and tumor grade (Right) (normal, n=54; grade 1, n=5; grade 2, n=35; grade 3 and 4, n=7). Two-sided Mann Whitney U test was used. A.U.=arbitrary unit. (D) Kaplan-Meier analysis of PDAC patients based on taurine levels in overall survival (Left) (high taurine, n=29; low taurine, n=25), and disease-free survival (Right) (high taurine, n=28; low taurine, n=26). The data is from our patient cohort. Data are presented as the mean ± SD. *p<0.05, **p<0.01, and ***p<0.001.

Upregulated ADO expression and its prognostic association in PDAC identified through multi-omic data analysis on the taurine metabolism pathway

To identify the gene responsible for the elevation of taurine in PDAC, we first conducted bioinformatics screening focused on genes associated with the taurine metabolism pathway. Utilizing the KEGG (Kanehisa and Goto, 2000) (Kyoto Encyclopedia of Genes and Genomes) pathway database, 16 genes implicated in taurine metabolism were identified. Additionally, a taurine transporter gene (SLC6A6) was selected, totaling 17 genes chosen for downstream analysis (Fig. 3A). Then, the expression levels of these gene were compared in PDAC and normal tissues using the transcriptomic (TCGA-PAAD and GTEx) and proteomic (CPTAC-PAAD) datasets. Among the 17 genes, seven genes exhibited significant differences in mRNA expression between normal and PDAC tissues, while six genes displayed significantly different protein expression between the two groups (Fig. 3B). We could find three genes that were different in both mRNA and protein expression levels (ADO, GGT1, and GGT5). As GGT1 exhibited opposite trends in mRNA and protein levels, with the mRNA level increasing and protein decreasing, it was not considered any further. Of the remaining two genes (ADO and GGT5), ADO is involved in taurine synthesis, catalyzing the cysteamine-to-hypotaurine conversion, whereas GGT5 is involved in taurine degradation to 5-glutamyltaurine. As the degradation steps are mediated by many other isozymes of GGT family (GGT1 to GGT7), and GGT5 has limited contribution, we selected ADO as the target for further investigation and its expression levels in normal and PDAC tissues are shown in Fig. 3C.

Figure 3. Upregulated ADO expression and its prognostic association in PDAC identified through multi-omic data analysis. (A) Seventeen genes related to taurine metabolism pathway (11 genes for taurine synthesis, 5 genes for taurine degradation, and 1 gene for taurine transport) for bioinformatics screening. (B) Bioinformatics screening of 17 genes related to taurine metabolism pathway in transcriptomic (TCGA-GTEx) and proteomic (CPTAC) databases. Number of genes with significantly different expression between normal and tumor (Left) and their profiles (Right) (red: higher expression in tumor, blue: higher expression in normal). (C) Violin plots of ADO mRNA expression (Left) (normal, n=171; tumor, n=179 from TCGA-GTEx) and protein expression (Right) (normal, n=74; tumor, n=137 from CPTAC) levels in normal versus tumor tissues in PDAC. Two-sided Welch’s t-test was used. Violin plots are presented with the median and quartiles. TPM=transcripts per million. (D) Kaplan-Meier analysis of PDAC patients based on ADO mRNA expression levels in overall survival (high ADO, n=46; low ADO, n=131) and disease-free survival (high ADO, n=49; low ADO, n=20) from TCGA database. (E) Representative images of immunohistochemistry (IHC) of ADO in PDAC tissue microarrays, comprising normal (n=20, 10 cases) and tumor tissues (n=78, 39 cases) (scale bar=30 µm) (Left), and IHC scores of ADO (Right). ADO staining was scored by 4 levels: 0 for none, 1 for weak, 2 for moderate, and 3 for strong staining. Two-sided Mann Whitney U test was used. (F) ADO protein expression levels of tissue microarrays used in (E) by stage (Left) (normal, n=20; stage I, n=12; stage II, n=24; stage III and IV, n=42), metastatic status (Middle) (normal, n=20; primary, n=40; metastasis, n=38), and tumor grade (Right) (normal, n=20; grade 1, n=23; grade 2, n=31; grade 3, n=19). Two-sided Mann Whitney U test was used. *p<0.05, ***p<0.001 and ****p<0.0001.

Then, we analyzed the clinical relevance of ADO in the transcriptomic and proteomic cohorts. In the TCGA-GTEx mRNA cohort encompassing 33 cancer types, elevated ADO mRNA expression in tumor relative to normal tissues was observed in six cancers: large B-cell lymphoma, esophagus, pancreas, melanoma, stomach, and thymus. Conversely, testicular cancer exhibited lower ADO expression than its normal counterpart (Supplementary Fig. 2A). In the CPTAC proteomic cohort, within the 10 cancer available cancer types, only PDAC demonstrated a significantly elevated level of ADO protein compared to normal tissues (Supplementary Fig. 2B). This suggests relatively specific presence of ADO in PDAC. Subsequently, Kaplan-Meier analysis was conducted based on the ADO levels in the TCGA cohort (Fig. 3D). For overall survival (OS), the high ADO group (n=46) showed significantly shorter survival period (Log-rank p=0.037, Hazard Ratio=1.61 [1.03-2.52]) compared to the low ADO group (n=131). In disease-free survival (DFS), the high ADO group (n=49) showed significantly more rapid and frequent cancer recurrence (Log-rank p=0.035, Hazard Ratio=3.08 [1.03-9.21)) than the low ADO group (n=20).

On the IHC analysis of independent tissue microarray comprising normal pancreas (20 cores, 10 cases) and PDAC tissues (78 cores, 39 cases). The IHC staining revealed a significant increase in ADO protein levels in PDAC tissues (p<0.0001; Fig. 3E), aligning with our bioinformatics findings. We then examined the association between ADO levels and clinical characteristics. In terms of PDAC stages, ADO levels were significantly higher in PDAC stage I, II (p<0.0001 for both) and maintained the elevated trend in PDAC with combined stage III-IV (p<0.0001; Fig. 3F). Additionally, increased ADO levels were observed in both primary and metastatic PDAC samples (p<0.0001; Fig. 3F). ADO levels were also significantly higher across all PDAC grades (p<0.0001 for grades 1, 2, and 3; Fig. 3F) Overall, ADO showed higher expression in PDAC tissues from early stage and was correlated with more unfavorable prognosis.

Inhibition of ADO mitigates pancreatic cancer cell growth in vitro

As the above data suggests association of the ADO-Taurine metabolic pathway with PDAC, we tested its causal involvement using genetic manipulation. For that, two distinct siRNAs targeting ADO were transfected into PANC-1, a pancreatic cancer cell line. ADO expression was downregulated by 82% and 46% in ADO-KD1 (p=0.005) and ADO-KD2 (p=0.012), compared to control siRNA, respectively (Fig. 4A). To see the ADO-Taurine relationship, we also measured the taurine levels in the knockdown cells and found that it was decreased by 78% only in cells treated with ADO-KD1 (p=0.008), but not in those with ADO-KD2 (Fig. 4B). This confirms the ADO-Taurine connection and suggests that a substantial (>50%) decrease in ADO expression is required to lower the taurine level. Then, we tested the effect of the lower ADO-Taurine level on the cancer cell growth. Only ADO-KD1 with taurine-lowering activity inhibited cell growth, whereas taurine-neutral ADO-KD2 did not (Fig. 4C). Consistently, a clonogenic assay showed that ADO-KD1 significantly reduced the tumorigenicity in the treated cells, demonstrating the long-term impact of ADO knockdown (Fig. 4D). To explore the downstream targets of ADO-Taurine, we examined the expression of NF-κB and mitogen-activated protein kinase kinase (MEK) - extracellular signal-regulated kinase (ERK), both implicated in the active progression of PDAC (Pedersen et al., 2017; Li et al., 2018). In PANC-1 cells, ADO knockdown did not result in altered NF-κB (p65 and phosphorylated p65) expression levels; however, it led to a decrease in phosphorylated MEK (Fig. 4E) while no significant changes were observed in total ERK or phosphorylated ERK levels (Supplementary Fig. 3). Overall, these results suggest the clinically suggested ADO-Taurine connection also bears causal relation to PDAC cell growth with possible downstream effect on MEK.

Figure 4. Inhibition of ADO mitigates pancreatic cancer cell (PANC-1) growth in vitro. (A) Relative expression of ADO in PANC-1 cells after control, ADO-KD1, and ADO-KD2 siRNA transfection (n=3). Two-sided Welch’s t-test was used. (B) Relative level of taurine in PANC-1 cells after control, ADO-KD1, and ADO-KD2 transfection (n=5). Two-sided Welch’s t-test was used. A.U.=arbitrary unit. (C) (Left) Cell proliferation of PANC-1 after control, ADO-KD1, and ADO-KD2 siRNA transfection was detected by CCK assay (n=8). (Right) Relative cell proliferation to day 0. Two-sided Welch’s t-test was used. O.D.=optical density. (D) Clonogenic assay of PANC-1 cells after control and ADO-KD1 siRNA transfection (20-day cultivation, n=3). (E) Expression of NF-κB (p65 and phosphorylated p65) and MEK (MEK1 and phosphorylated MEK1/2) proteins in PANC-1 cells after control and ADO-KD1 siRNA transfection, as detected by western blot assay. Data are presented as the mean ± SD. *p<0.05, **p<0.01 and ***p<0.001.

ADO knockdown reduces PDAC tumorigenesis in vivo

To see if the ADO-Taurine’s causal involvement in PDAC cell growth is also manifested in vivo, we xenografted the ADO-KD1 siRNA-treated cells into mice and observed the tumor formation. Consistent with the in vitro findings, the tumors originating from the ADO-knockdown cells exhibited reduced growth compared to those from control cells (Fig 5A, 5B). In addition, ADO-knockdown cell-derived tumors were not established as robustly as control cell-derived tumors (tumors formed in 5/10 and 10/10 mice, respectively, Fig. 5C). There was no difference in body weight between the two groups of mice throughout the experiment (Fig. 5D). Additionally, the ADO protein levels were found to be reduced in the ADO-KD1 tumors (Fig. 5E), indicating a prolonged suppression of ADO in vivo. Taken together, our findings indicate that the ADO-Taurine axis is causally involved in tumorigenesis and progression of PDAC.

Figure 5. ADO knockdown reduces PDAC tumorigenesis in vivo. (A) Tumor growth of BALB/c mice (n=10 per group) xenografted with control or ADO-KD1 siRNA-transfected PANC-1 cells. The tumor volume was measured every 10 days with a caliper. Two-sided Welch’s t-test was used. (B) Photographic image of grafted tumors in (A). The tumors were extracted after 70 days. (C) Tumor establishment in xenograft mice after siRNA transfection. The ratio of tumor formation in each control and ADO-KD1 group is shown. (D) Body weight of mice during the experimental period. Two-sided Student’s t-test was used. Data are presented as the mean ± standard error. (E) Expression of ADO in control and ADO-KD1 of PANC-1 graft tumor tissues. Data were obtained from five graft tumors in each group. *p<0.05.

In this study, we identified an elevated level of taurine in PDAC and established its relationship with cancer recurrence. Importantly, we showed the causal involvement of taurine metabolism in PDAC tumorigenesis through the ADO gene. As taurine has been recognized for its cancer-inhibiting properties (Ma et al., 2022), previous metabolomics studies on PDAC also reported taurine levels. Still, taurine levels in these studies, ranging approximately 1.5 to 3.0 times higher than those in control tissues, were presented in tables of detected metabolites and were not noted any further (Kaur et al., 2012; Kamphorst et al., 2015; Battini et al., 2017). Although one observational study did present high levels of taurine in pancreatic cancer and pre-cancerous pancreatic intraepithelial neoplasia (PanIN) as a possible PDAC biomarker (Wang et al., 2014), it was limited by the small sample size (7 pairs of normal and PDAC human tissues) and it lacked an analysis correlating these findings with patients’ clinical attributes. Therefore, our observation of a 2.51-fold increase in taurine not only corroborates those earlier findings but also extends them by showing prognostic value of the taurine level and identifying the target enzyme associated with the altered taurine metabolism in PDAC tissues. Numerous studies have identified prognostic markers in PDAC, ranging from individual gene (Chen et al., 2021) and protein (Roa-Peña et al., 2019) to multi-panel model (Kim et al., 2021). Still, there has not been a single metabolite marker demonstrating prognostic value. Once set up properly, a metabolite level is generally cheaper and easier to measure with a simple modality than proteins and genes, such as thin layer chromatography. As our study linked a single metabolite taurine to PDAC recurrence prognosis, it should be another significant example in PDAC marker studies. The suggested ADO-Taurine axis may serve as an effective prognostic marker, either independently or in conjunction with established biomarkers like CA19-9 (Bauer et al., 2013). While our study focused on taurine metabolism, it’s important to consider the potential links between taurine and glutathione, both derived from cysteine. Our extraction method, involving quick freezing, grinding, and multiple freeze-thaw cycles, can lead to the oxidation of glutathione, complicating accurate measurement. Thus, precise detection of glutathione levels in PDAC tissues requires targeted extraction methods that preserve glutathione integrity. Future study using targeted method is needed for assessment of glutathione in PDAC.

ADO is a gene that catalyzes the conversion of cysteamine to hypotaurine, an immediate precursor of taurine. Shen et al. (2021) identified a significant correlation between ADO and hypotaurine levels in distinguishing low- from high-grade gliomas. Their study further revealed that the ADO-Hypotaurine axis stimulates NF-κB pathway activation, subsequently increasing the secretion of the cytokine CCL20, thereby sustaining and enhancing the ‘cancer stemness’ of glioma cells. In our experiments with pancreatic cancer cells, knocking down ADO did not alter NF-κB expression levels. However, we noted a reduction in phosphorylated MEK, with no change in ERK or phosphorylated ERK levels (Supplementary Fig. 3), suggesting a complex underlying mechanism. This finding indicates that the ADO-Taurine axis might influence PDAC cell growth through MEK-related pathways, potentially non-ERK downstream targets of MEK. While the Raf-MEK-ERK signaling pathway is known to be pivotal in PDAC (Collisson et al., 2012; Ozkan-Dagliyan et al., 2020), our results suggest that the role of ADO in regulating PDAC proliferation may involve additional or alternative MEK-related mechanisms. Further research is needed to elucidate the exact downstream effectors and signaling cascades involved in this process.

Interestingly, our bioinformatics analysis revealed that ADO mRNA levels are reduced in testicular germ cell tumors (TGCT) compared to normal testicular tissue, contrasting with our findings in PDAC. This difference highlights the tissue-specific roles of ADO and taurine metabolism in cancer. Taurine plays a crucial role in normal testicular function, being essential for sperm physiology and fertility (Li et al., 2023). The high taurine concentration in normal testicular tissue suggests that ADO expression might be higher in normal testis to maintain taurine synthesis, whereas in TGCT, the loss of normal cellular functions could lead to decreased ADO expression. This contrasting pattern underscores the complexity of metabolic reprogramming in different cancer types and emphasizes the need for tissue-specific approaches in cancer metabolism research.

In conclusion, our study underscores the pivotal role of the ADO-Taurine axis in PDAC’s altered metabolism, highlighting its significance in the disease’s progression and potential as a prognostic marker. Our findings, derived from multi-omics analyses and validated through in vitro and in vivo experiments, reveal that elevated taurine levels, regulated by ADO, correlate with poor prognosis in PDAC patients (Fig. 6). This work not only opens new avenues for understanding the metabolic intricacies of PDAC but also proposes the ADO-Taurine pathway as a novel therapeutic target. Further investigations into this metabolic pathway may unveil more profound insights and contribute to the development of new therapeutic interventions for PDAC.

Figure 6. Overview of study design and key findings.

This work was supported by the National Research Foundation (NRF) grant funded by the Korean government (NRF-2021R1A5A2031612 to S-S.H.), a grant (2021IP0020-1 to S.C.K.) from Asan Institute for Life Sciences, Asan Medical Center, and a Inha University Grant (to K.H.J).

The authors declare that they have no conflict of interest.

Hoonsik Nam and Woohyung Lee contributed equally to this study.

Conception and design: S.C.K. and S.P. Development of methodology: H.N., W.L., S-S.H., S.C.K., and S.P. Clinical sample procurement: W.L. and S.C.K. Acquisition of data: H.N., W.L., Y.J.L., J-M.K., and K.H.J. Analysis and interpretation of data: H.N., W.L., Y.J.L., J-M.K., and K.H.J. Writing, review, and/or revision of manuscript: All authors. All authors read and approved the final manuscript.