Autotaxin activity predicts transplant-free survival in primary sclerosing cholangitis

Autotaxin has been associated with liver disease severity and transplant-free survival. This study aimed to validate autotaxin as a biomarker in two cohorts of Norwegian large-duct PSC patients, one discovery panel (n = 165) and one validation panel (n = 87). Serum activity of autotaxin was measured in diluted sera by a fluorometric enzymatic assay. Patients reaching an end-point, liver transplantation or death, (discovery panel: n = 118 [71.5%]; validation panel: n = 35 [40.2%]), showed higher autotaxin activity compared with the other patients, P < 0.001 and P = 0.004, respectively. Kaplan-Meier survival analyses showed a strong association between increasing autotaxin activity and shorter liver transplant-free survival (discovery panel: P < 0.001, validation panel: P = 0.001). There was no relationship between autotaxin activity and the presence of inflammatory bowel disease or occurrence of hepatobiliary malignancy. In a multivariable analysis, high autotaxin activity was associated with an increased risk of liver transplantation or death (hazard ratio 2.03 (95% confidence interval 1.21–3.40), P < 0.01), independent from Mayo risk score, an in-house enhanced liver fibrosis score and interleukin-8 in serum. In conclusion, increased serum autotaxin activity is associated with reduced liver transplant-free survival independent from Mayo risk score and markers of inflammation and fibrosis.


Patients and Methods
patient population and data collection. In the discovery panel, 165 patients (74% male) with large-duct PSC were included, while the validation panel comprised 87 patients with large-duct PSC (82% male). Detailed demographic data for both panels can be found in Table 1. The median observation time was 6.58 years (range 0.01-24.36) from time of sampling for the merged panel.
Diagnosis of PSC was based on typical findings on magnetic resonance cholangiography or endoscopic retrograde cholangiography according to established criteria 15 . Patients were recruited at admission to Oslo University Hospital Rikshospitalet in the period from 1992-2006 in the discovery panel and 2008-2011 in the validation panel. Regarding the end-point of liver transplantation or death, patients were followed until 31 st of December 2016. Blood samples were included into the Norwegian PSC Research Centre (NoPSC) Biobank (Oslo, Norway). The first cholangiography compatible with PSC defined the time of diagnosis. Duration of disease was defined as the time from diagnosis to serum sampling. Clinical and demographic information was acquired from patient records and research databases, including laboratory data, history of ascites, encephalopathy, variceal bleeding, inflammatory bowel disease, colorectal or hepatobiliary malignancy and medication at the time of serum extraction. Inflammatory bowel disease was diagnosed based on endoscopic and histological findings, according to accepted criteria 16 . Cause of death was extracted from the Cause of Death Registry (Oslo, Norway, reference . Revised Mayo risk score was calculated according to the established algorithm 17 , in addition to Model For End-Stage Liver Disease (MELD) score 18 and Aspartate aminotransferase (AST) to Platelet Ratio Index (APRI) score, all calculated from data at sample time 19 . In-house eLF score. Commercial kits were used to analyze serum levels of TIMP-1 and hyaluronic acid (both R&D Systems), and intact N-terminal PIIINP was analyzed by radioimmunoassay (UniQ PIIINP RIA; Orion Diagnostica, Espoo, Finland) in frozen serum samples. Inter-and intra-assay coefficients of variation were <10%. The algorithm used in the ADVIA Centaur XP ELF test, combining TIMP-1, hyaluronic acid, and PIIINP values, was subsequently used to calculate in-house ELF scores (ELF score = 2.278 + 0.851 ln(C hyaluronic acid ) + 0.751 ln(C PIIINP ) + 0.934 ln(C TIMP1 )).
Autotaxin activity assay. Stored sera, from sample time, thawed up to two times were analyzed for autotaxin activity by a fluorometric enzymatic assay. Autotaxin activity was quantified in (at least 20-fold diluted) sera. Briefly, serum samples were incubated with a buffer containing 500 mmol/L of NaCl, 5 mmol/L of MgCl 2 , 100 mmol/L of Tris (pH = 9.0), and 0.05% Triton X-100 for 60 minutes at 37 °C. Parallel incubations were performed in the presence and absence of 1 mmol/L of LPC (14:0). Subsequently the amount of liberated choline www.nature.com/scientificreports www.nature.com/scientificreports/ was determined by enzymatic fluorimetry using choline oxidase (2 U/mL), horseradish peroxidase (1.6 U/mL), and homovanillic acid as substrates for peroxidase. After the addition of both enzymes in a buffer (consisting of 20 mmol/L of CaCl 2 , 2 mmol/L of HVA, 50 mmol/L of 3-[N-morpholino] propanesulfonic acid [pH = 8.0], and 0.1% Triton X-100), the increase in fluorescence was monitored at 37 °C using a NOVOstar analyzer (excitation 320 nm and emission 405 nm; BMG Labtech GmbH, Offenburg, Germany). The (endogenous) amount of choline present in the sample without the addition of LPC was subtracted from the amount measured in the presence of LPC. Inter-assay variance was less than 15%, and intra-assay variance below 10%.

ethics.
The study was carried out in accordance with the Declaration of Helsinki. Written informed consent was obtained from all study participants. The Regional Committee for Medical and Health Research Ethics of South-Eastern Norway approved the study (reference 2011/2572). statistical analyses. Data are presented as median (range) unless stated otherwise, except survival (in years) which is presented as mean (95% CI). When dichotomizing autotaxin levels in high/low categories, the previously published cutoff (>7.5 nmol mL −1 min −1 ) defined by Wunsch et al. was used 14 . Comparisons were done with Mann-Whitney U tests. Liver transplant-free survival was analyzed using Kaplan-Meier plots, and univariate and multivariate Cox regression. Variables with P < 0.05 in the univariate analysis were included in the final multivariate Cox regression model, except age, AST, bilirubin and albumin as they are included in Mayo risk score. The MELD score was highly right-skewed even after log-transformation and had a high rate of missing data; thus, we used the Mayo risk score in the Cox regression analyses. For correlations, Spearman's rank correlation test was applied. Statistical analyses were performed using SPSS (version 24; SPSS, Inc., Chicago, IL). P-values < 0.05 were considered statistically significant. 74% and 82% of the patients were male, median age at sampling was 41.0 and 38.5 years, 83% and 78% had concomitant IBD, 52% and 31% underwent liver transplantation and 32% and 8% died during follow-up in the exploration and validation panel, respectively (Table 1). There was one (0.6%) case of gall bladder cancer and 2 cases of hepatocellular carcinoma (1.2%) in the exploration panel, 18 cases (11%) and 8 cases (9%) of cholangiocarcinoma in the exploration and validation panel, respectively.
During follow-up, 118 (71.5%) and 35 (40.2%) patients reached an end-point, liver transplantation or death, in the discovery and validation cohorts, respectively. Patients who reached an end-point during follow-up showed higher ATX activity compared with the other patients in both cohorts (P < 0.001 and P < 0.004 in the discovery and validation cohort, respectively, Fig. 1a,b).
When stratifying patients according to autotaxin tertiles, there was a strong association between increasing autotaxin activity and shorter liver transplant-free survival in both cohorts (Fig. 1c,d). In a previous study, the optimal cut-off of autotaxin activity to distinguish patients with a higher risk of end-points was 7.5 nmol mL −1 min −1 14 . Repeating the analysis with this cut-off in the present study yielded similar results, with high autotaxin activity concentrations showing reduced liver transplant-free survival compared with the low autotaxin group in both cohorts (Fig. 1e, 6.80 [6.07-7.53] years, P < 0.001). In a merged panel (n = 252) a cut-off of 7.5. nmol mL −1 min −1 yielded a sensitivity of 35% and specificity of 85% for predicting liver transplantation or death, with corresponding positive predictive value of 81% and negative predictive value of 49%. A dedicated AUROC suggested an optimal cut-off in the present population (using Youden's index) of 6.4 nmol mL −1 min −1 , with corresponding sensitivity of 54% and specificity of 78%, but with similar positive (79%) and negative (52%) predictive values.
Thirty of the patients with death as endpoint died of liver related causes (75.0%), while nine were registered with non-hepatic causes of death (22.5%, data missing for n = 1). Re-doing the above analyses with liver-related deaths only (censoring patients dying from other causes) did not have major impact on the results (data not shown).
Autotaxin and biochemical characteristics. Autotaxin activity was similar in the discovery and validation panels, mean 6.6 nmol mL −1 min −1 (SD 3.6) and 7.2 nmol mL −1 min −1 (SD 3.8), respectively (P = 0.28). Despite some baseline differences (Table 1), we therefore merged the discovery and validation panels to increase power when investigating the relationship between clinical and biochemical characteristics.

Autotaxin activity and clinical characteristics.
In the merged cohort, 30 (11.9%) PSC patients were diagnosed with hepatobiliary malignancy during follow-up. Autotaxin activity was similar in patients who developed cancer during follow-up and those who did not ( ). In a small subgroup n = 32, severity of disease using the cholangiography-based Amsterdam score was evaluated in a previous study 21 . Patients with advanced PSC had numerically higher autotaxin activity than mild PSC (8.9 vs. 7.7), but this was not statistically significant (P = 0.51).
Modest significant correlations were found between autotaxin and PSC duration (rho = 0.184, P = 0.003), and clinical risk scores; Mayo risk score (rho = 0.408, P < 0.001) and MELD score (0.233, P = 0.006), also see Table 2. Data on ursodeoxycholic acid (UDCA) treatment was available for 86 patients from the validation cohort (99%), out of which 35 (41%) were treated at the time of sampling. Patients treated with UDCA had higher autotaxin levels compared to other patients (7.48   www.nature.com/scientificreports www.nature.com/scientificreports/

Autotaxin activity and new biomarkers of inflammation and fibrosis in PSC. Recent studies have
suggested that markers of inflammation (IL-8) and liver fibrosis (ELF test) predict liver transplant-free survival in PSC 11, 21,22 . Data on IL-8 and an in-house enhanced liver fibrosis score (ELF-score) were available from the previous studies 21,22 , and both correlated with autotaxin activity (Fig. 3). There was a correlation between IL-8 and the in-house ELF-score (rho: 0.656, P < 0.001), autotaxin and IL-8 (rho: 0.427, P < 0.001) and autotaxinand ELF-score (rho: 0.481 P < 0.001). When dividing ELF and IL-8 in high and low by median, and autotaxin in high and low by a cut-off at 7.5 nmol mL −1 min −1 , as defined by Wunsch et al. 14 , high levels were associated with reduced liver transplant-free survival for all markers (Fig. 4). Re-doing these analyses with liver-related deaths only, did not influence the results (data not shown).
Notably, Mayo risk score had a better ability to discriminate patients with and without liver transplantation or death during follow-up (AUC 0.81 vs. 0.66, P < 0.001, Fig. 5a). The same was apparent for in-house ELF (AUC 0.79, P < 0.001), but not IL-8 (AUC 0.71, P = 0.25), see Fig. 5b. However, when analyzed using Cox regression, high autotaxin as defined by the validated cut-off 7.5 nmol mL −1 min −1 was also associated with an increased risk of liver transplantation or death, independent from Mayo risk score, the in-house ELF score and IL-8 (HR 2.03 (95%CI 1.21−3.40), P = 0.007), see Table 3. Adding disease panel (discovery or validation) as covariate, or doing analyses on liver-related deaths only, did not influence the results (data not shown).

Discussion
In this cross-sectional study of autotaxin activity in two independent cohorts of PSC patients, we found a strong correlation between increasing autotaxin and shorter liver transplantation free survival. Autotaxin activity correlated with disease severity, as measured by Mayo risk score, but there was no association with IBD or cancer. Autotaxin correlated moderately with new biomarkers in PSC, including IL-8 and ELF-score, but in multivariate analyses high autotaxin was an independent predictor of liver transplant-free survival.
A key observation in the present study is that we show in two independent cohorts of in total 252 large-duct PSC patients that high autotaxin activity is associated with reduced liver transplant-free survival with a moderate HR of 2.0 in multivariable regression, corroborating a previous finding by Wunsch et al., who investigated autotaxin in 115 PSC patients 14 . Similar correlations between autotaxin activity and liver biochemistries and clinical parameters like Mayo risk score and MELD score were also seen. This highlights autotaxin activity as a biomarker of potential clinical value, in particular in combination with other markers. Its relationship with disease activity and response to treatment in drug trials should clearly be investigated to define its utility, in line with recent trials investigating markers of fibrosis 23,24 .
The mechanisms causing elevated autotaxin activity in PSC are not clear. Autotaxin has been shown to predict survival, disease stage and cirrhosis also in non-cholestatic conditions 8,25 , while the value of autotaxin as a serum marker of fibrosis varies between phenotypes, being a better predictor in viral hepatitis compared to non-alcoholic fatty liver disease 26 . In other studies of viral hepatitis, autotaxin also correlates with liver stiffness measured with transient elastography 27 . Autotaxin has also been reported to inversely correlate with albumin levels in chronic liver diseases 7,28 , in line with the present data. Mechanistic studies of autotaxin in PSC are scarce. In murine models of fibrosis, increased autotaxin expression in hepatocytes has been observed both after liver injury caused by carbon tetrachloride and after viral infection 29 . Genetic or pharmacological inhibition of autotaxin subsequently reduced liver fibrosis in the model, suggesting a direct link to fibrosis. However, whether hepatocyte expression can account for the increased autotaxin activity observed during cholestasis and cholestatic pruritus in PSC is not known. The negative correlation between autotaxin activity and albumin on the one hand, and positive correlation with INR on the other hand suggest that autotaxin activity is not affected by liver synthesis function. We also make the novel observation that autotaxin activity only moderately correlates with the new PSC biomarkers IL-8 and an in-house ELF, whereas autotaxin activity still independently predicts survival. This suggests that autotaxin is a marker of aspects of the PSC pathogenesis reflecting more than just inflammation or  Table 2. Correlations between ATX activity and biochemical markers and clinical risk scores in the merged cohort (n = 252). APRI, aspartate aminotransferase to platelet ratio index; AST, aspartate aminotransferase; ALT, alanine aminotransferase; ALP, alkaline phosphatase; CRP, C-reactive protein; INR, international normalized ratio; MELD, model for end-stage liver disease.
www.nature.com/scientificreports www.nature.com/scientificreports/ fibrosis. Biomarkers of inflammation, fibrosis and pruritus might independently predict prognosis and underline the potential utility of multi-marker panels to account for the complexity of PSC. This should be investigated in subsequent studies.
Another important observation in the present study is the lack of correlation between autotaxin and biliary tract cancer. This is contradictory to studies describing an emerging role of LPA and autotaxin in malignant diseases 6 , e.g. hematological malignancies 30 and pancreatic cancer 31 . Autotaxin has been proposed to be an important mediator in cancer cell growth and metastasis 6 . It has also been speculated that it is a link between liver disease and hepatocellular carcinoma 29,32 , but data from human studies are conflicting 33 . Little data is available on cholangiocarcinoma (CCA), and there were no signs of association between autotaxin and biliary cancer in our cohort. One possible explanation is if autotaxin only reflects the presence of tumor without being able to predict later tumor development. Still, the observation contrasts other novel PSC biomarkers, which seem CCA-related. Examples include associations observed between CCA in PSC and elevated ELF 22 , increased soluble CD14 10 and increased prevalence of IgA anti-GP2 34 , providing further evidence of unique characteristics of autotaxin as biomarker.
A peculiar observation was the increased autotaxin activity observed in PSC patients receiving treatment with UDCA. Importantly, data on UDCA use was only available in the smaller validation cohort (n = 87) and an association does not imply causation. In the study by Wunsch et al. autotaxin activity was higher in PBC patients who did not respond to UDCA treatment, compared to other patients 14 . It has been hypothesized that the lack of response to UDCA in some PBC patients may indicate a failure to upregulate alk-SMase, another nucleotide pyrophosphatase/phosphodiesterase (NPP) enzyme similar to autotaxin, by UDCA in these patients, leading to increased levels of autotaxin 35 . A more likely hypothesis explaining increased autotaxin activity in UDCA-treated PSC patients in the present study may be that PSC patients in Norway usually receive UDCA only on specific indication, i.e. flares, pruritus or gallstones, and it may be that patients receiving UDCA had more pruritus or  www.nature.com/scientificreports www.nature.com/scientificreports/ even more serious disease. However, it should be noted that measures of disease severity were not convincingly different in patients with and without UDCA (Mayo risk score and liver transplant-free survival).
The strengths of the present study are the large sample size, the inclusion of a validation cohort, and the long follow-up period, although the cross-sectional design makes us unable to make assumptions about cause end  www.nature.com/scientificreports www.nature.com/scientificreports/ effect, which could be improved by repeated sampling during disease activity or after endoscopic intervention. The inclusion from a tertiary care center suggest that the patients have a more severe phenotype than the general population. The limitations also include missing data for some covariates, reducing the number of patients in the multivariate analyzes. We lack data on pruritus, a possible key confounder. There is also a possibility that patients included in the discovery and validation panels could have inherent differences in liver transplantation-free survival due to changes in clinical practice which could introduce a bias, but adding cohort as covariate did not influence the effect of autotaxin. Despite these limitations, it is reasonable to firmly conclude that increased serum autotaxin activity is associated with reduced liver transplant-free survival in PSC patients. This association seems to be independent from Mayo risk score and markers of inflammation and fibrosis, and with no relationship with cholangiocarcinoma development. Autotaxin is therefore an interesting candidate as clinical biomarker, which should be prioritized for follow-up in prospective and interventional studies.

Data Availability
The dataset used in the present study is available from the corresponding author on reasonable request.  Table 3. Cox regression analysis in the combined panel (n = 252). *All variables with P < 0.05 in the univariate analysis were included in the multivariate model, except age, AST, bilirubin and albumin as they are included in the Mayo risk score. Complete data for all variables in the model were available for n = 157 patients in the combined panel. † Autotaxin >7.5 nmol mL −1 min −1 as defined by Wunsch et al. 14 . # Right-skewed distribution, transformed by the natural logarithm before analyses. ‡ Defined as INR ≥ 1.2 or Normotest < 70. PSC, primary sclerosing cholangitis; IBD, inflammatory bowel disease; ELF, enhanced liver fibrosis; INR, international normalized ratio; AST, aspartate aminotransferase; ALT, alanine aminotransferase; ALP, alkaline phosphatase; CRP, C-reactive protein; IL, interleukin; HR, hazard ratio.