Serum keratin 19 (CYFRA21-1) links ductular reaction with portal hypertension and outcome of various advanced liver diseases

Background Keratins (Ks) represent tissue-specific proteins. K18 is produced in hepatocytes while K19, the most widely used ductular reaction (DR) marker, is found in cholangiocytes and hepatic progenitor cells. K18-based serum fragments are commonly used liver disease predictors, while K19-based serum fragments detected through CYFRA21-1 are established tumor but not liver disease markers yet. Since DR reflects the severity of the underlying liver disease, we systematically evaluated the usefulness of CYFRA21-1 in different liver disease severities and etiologies. Methods Hepatic expression of ductular keratins (K7/K19/K23) was analyzed in 57 patients with chronic liver disease (cohort i). Serum CYFRA21-1 levels were measured in 333 Austrians with advanced chronic liver disease (ACLD) of various etiologies undergoing hepatic venous pressure gradient (HVPG) measurement (cohort ii), 231 French patients with alcoholic cirrhosis (cohort iii), and 280 hospitalized Germans with decompensated cirrhosis of various etiologies (cohort iv). Results (i) Hepatic K19 levels were comparable among F0–F3 fibrosis stages, but increased in cirrhosis. Hepatic K19 mRNA strongly correlated with the levels of other DR-specific keratins. (ii) In ACLD, increased serum CYFRA21-1 associated with the presence of clinically significant portal hypertension (CSPH; HVPG ≥ 10 mmHg) (OR = 5.87 [2.95–11.68]) and mortality (HR = 3.02 [1.78–5.13]; median follow-up 22 months). (iii) In alcoholic cirrhosis, elevated serum CYFRA21-1 indicated increased risk of death/liver transplantation (HR = 2.59 [1.64–4.09]) and of HCC (HR = 1.74 [1.02–2.96]) over the long term (median follow-up 73 months). (iv) In decompensated cirrhosis, higher serum CYFRA21-1 predicted 90-day mortality (HR = 2.97 [1.92–4.60]) with a moderate accuracy (AUROC 0.64), independently from established prognostic scores. Conclusions Hepatic K19 mRNA and serum CYFRA21-1 levels rise in cirrhosis. Increased CYFRA21-1 levels associate with the presence of CSPH and reliably indicate mortality in the short and long term independently of conventional liver biochemistry markers or scoring systems. Hence, the widely available serum CYFRA21-1 constitutes a novel, DR-related marker with prognostic implications in patients with different settings of advanced liver disease.


Background
Keratins are the intermediate filaments of epithelial cells that are expressed in a tissue-specific manner and therefore constitute widely used biomarkers [1,2]. Keratins are subdivided into type I proteins (including K1-K8) and type II proteins (including K9-K20), and both types assemble to form heteropolymers. As a consequence, each cell type has a characteristic type I-type II keratin repertoire [3]. For example, keratins 8/18 (K8/K18) are ubiquitously produced in single-layered epithelia while K7/K19 are found in some, but not all, simple epithelial cells [1,2]. In the liver, K19 constitutes the most widely used histologic marker of ductal/ductular reaction (DR) as it is expressed in cholangiocytes and hepatic progenitor cells (HPC) but not in adult hepatocytes [1,4].
Upon tissue injury, keratins are released into the serum and therefore are well-known disease markers [5]. K18-based serum markers pose the most prominent keratin biomarkers as they represent useful surrogates of hepatocellular injury [5,6]. Among them, M30/M65 are among the most widely used severity markers in multiple liver disorders [5]. In contrast, while serum CYFR A21-1 is widely available in routine laboratories as a tumor marker [7,8], its usefulness in liver disease has not been systematically studied.
The liver possesses two fundamentally different strategies to recover from parenchymal losses, i.e., the division of mature hepatocytes and the progenitor cellbased DR [9,10]. Division of mature hepatocytes is typically seen after surgical liver resection, but may become impaired during chronic liver diseases, especially at stages with advanced liver scarring (fibrosis). In the latter, DR constitutes the predominant mode of parenchymal replenishment [9,10]. DR is driven by HPC that can differentiate to both biliary and hepatocellular lineage [11]. Since regeneration typically reflects the rate of parenchymal loss, the extent of DR mirrors the severity of the underlying liver disease [9,10] and thus may predict the disease course [10,12].
In line, multiple studies showed that the amount of K7/K19-positive cells correlates with the severity of liver disease and is of prognostic relevance [12][13][14][15][16][17]. However, these studies only evaluated the amount of DR-specific keratins in tissue sections, which are less commonly available due to the increasing use of non-invasive methods for diagnosing and staging liver fibrosis and portal hypertension. Hence, the aim of this study was to test whether a serum marker reflecting the extent of DR would be a valuable tool for prognostication. Thus, we systematically analyzed hepatic K19 expression in liver biopsies from various chronic liver diseases and assessed the significance of CYFRA21-1 serum levels in three further, well-characterized cohorts of patients with different settings of advanced liver disease.

Analysis of human liver samples
(Cohort i) 57 liver samples from patients with liver disease of different etiologies and 13 control samples were assessed for hepatic expression of the ductular keratins K19, K7, and K23 (Additional file 1: Table S1). Further details on patient selection and their molecular analysis are given in the supplement (Additional file 1).

Analysis of CYFRA 21-1 levels in patients with different liver disease stages
Three well-characterized cohorts were used to assess serum CYFRA21-1 in relation to liver disease severity and clinical outcomes: (cohort ii; Table 1) 333 Austrians with advanced chronic liver disease (ACLD; as defined by a hepatic venous pressure gradient (HVPG) ≥ 6 mmHg [18]) of various etiologies undergoing HVPG measurement, (cohort iii; Table 2) 231 French patients with alcoholic cirrhosis, and (cohort iv; Table 3) 280 Germans hospitalized for decompensation of cirrhosis. Further details are given in the supplement (Additional file 1).

Statistical analysis
Continuous variables were given as median with interquartile range (IQR) and were compared by the Mann-Whitney U test or Kruskal-Wallis test. Categorical variables were reported as absolute (n) and relative (%) frequencies, and contingency tables were analyzed with chi-squared tests.
The optimal cut-off values for CYFRA 21-1 as an indicator of CSPH or survival were determined using a cohort-specific Youden's index in receiver operating characteristic (ROC) analysis. Areas under the receiver operating characteristic curve (AUROC) were calculated. Univariate and multivariable analyses of risk factors for outcomes such as mortality were conducted using Cox regression analyses to calculate hazard ratios (HRs). Distributions among groups were assessed by univariable logistic regression analyses to calculate odds ratios (ORs), and forward-stepwise multiple logistic regression analyses were used to test for independent associations.
Time-to-event variables were displayed using Kaplan-Meier curves, and differences in event-free survival (or HCC occurrence, respectively) were tested using the logrank test. For the analysis of transplant-free survival, death or transplant was considered as endpoints and patients were right-censored at the end of the observation period.
HRs and ORs were given with their corresponding 95% confidence intervals (CI) [in brackets].
Correlations between clinical variables and serum biomarkers or elastography parameters were assessed by calculating Spearman's rank correlation (Spearman's rho) coefficients.
Nominal P values were reported for all statistical tests. Due to the exploratory nature of most parts of our study, we applied two-sided significance levels of P < .05 for all tests without correction for multiple testing. Statistical analyses were performed using the SAS System Package version 8.02 (SAS Institute, Cary, NC, USA) and SPSS version 23 (IBM, Armonk, NY, USA), and graphs were created with SPSS or Prism 5 (GraphPad, La Jolla, CA, USA).

Results
Hepatic K19 levels are upregulated in patients with liver cirrhosis First, we analyzed the hepatic expression of ductular keratins (K7/19/23) in 57 patients with different liver fibrosis stages and disease etiologies (Additional file 1: Table S1). K19 mRNA levels were comparable among patients without fibrosis (F0) and with mild to advanced fibrosis stages (F1-F3), but elevated in cirrhotic individuals (F4; Fig. 1a). Notably, the increase seemed to be unrelated to the etiology of liver disease, since comparable K19 mRNA levels were seen in samples from patients with NAFLD, ALD, and chronic hepatitis C (Additional file 1: Fig. S1A). Similar results were obtained with the ductular keratins K7/K23 (Fig. 1b, c; Additional file 1: Fig. S1B, C). Moreover, the expression of K7/K19 and K19/K23 strongly correlated with each other (Fig. 1d, e). The mRNA results were confirmed by immunohistochemistry and immunoblotting against K19 revealing a higher number of K19-positive cells and K19 protein levels in cirrhotic vs. control livers ( Fig. 1f and Additional file 1: Fig. S2).
Taken together, levels of ductular keratins were elevated in patients with cirrhosis most likely as a consequence of a prominent DR.
Collectively, increased serum CYFRA21-1 levels were strongly and independently associated with the presence of CSPH and poor survival in patients with ACLD. Of note, CYFRA21-1 provided prognostic information in the subgroup of patients with CSPH, highlighting its prognostic relevance beyond being a non-invasive surrogate of CSPH.

CYFRA21-1 indicates poor long-term survival in alcoholic cirrhosis
To validate the long-term prognostic usefulness of serum CYFRA21-1 levels, we turned to a French cohort of 231 patients with alcoholic cirrhosis, who were followed up longitudinally for HCC development (Table 3). During the median follow-up of 73 months, 98 patients died or received a liver transplant (non-survivors), while 133 patients survived without the need for a liver transplant (transplant-free survivors; Table 3). While transplant-free survivors and non-survivors did not differ in their demographic characteristics (Table 3), non-survivors more often developed hepatocellular carcinoma (47.4% vs. 6.8%, P < .0001; Table 3). 44.9% of non-survivors reached the pre-defined endpoint due to HCC while 51.0% did so due to other liver-related reasons. The baseline CYFRA21-1 serum level was higher in non-survivors vs. survivors (Table 3). A comparison of patients with high versus low serum CYFRA21-1 using the median (5.26 ng/mL) as a cut-off revealed that patients with higher CYFRA21-1 levels had more often ascites, had higher Child-Pugh scores, and reached the pre-defined endpoint of HCC more frequently (Additional file 1: Table S3).
Finally, higher CYFRA21-1 serum levels were associated with a somewhat higher occurrence of HCC (univariate HR = 1.74 [1.02-2.96], log-rank P = .039; Fig. 3c). The positive predictive value and negative predictive value of the median CYFRA21-1 levels of 5.26 ng/mL for HCC occurrence were 22% and 74%, respectively. Using the same Cox's proportional hazards model as for mortality, higher serum CYFRA21-1 levels were not significantly associated with HCC occurrence (Additional file 1: Table S4).
Taken together, in patients with alcoholic cirrhosis, serum CYFRA21-1 was an independent indicator of overall and liver-related long-term mortality.

CYFRA21-1 indicated short-term survival in patients with decompensated cirrhosis and ascites
To determine the predictive ability of serum CYFRA21-1 levels for short-term mortality in decompensated cirrhosis, we analyzed a cohort of patients hospitalized due to hepatic decompensation with ascites (Table 4). Within 90 days of follow-up, 88 patients (31.4%) either died or were transplanted. These "non-survivors" were significantly older, more commonly had HCC or spontaneous bacterial peritonitis, and displayed higher MELD, Child-Pugh, and acute-on-chronic liver failure (ACLF) scores than transplant-free survivors (Table 4). Non-survivors had higher median CYFRA21-1 serum levels than survivors (Table 4). In the entire cohort, CYFRA21-1 only weakly correlated with ALT, AST, and MELD, while it did not significantly correlate with markers of liver synthetic function (e.g., INR and albumin; Additional file 1: Table S5). The CYFRA21-1 serum values were 8.5 ± 12.5 ng/mL in cirrhotic patients without ACLF and 8.2 ± 9.0 ng/mL, 12.4 ± 13.4 ng/mL, and 19.1 ± 14.8 ng/mL in patients with ACLF grades 1, 2, and 3, respectively (Additional file 1: Figure S5).
To estimate the diagnostic accuracy of serum CYFR A21-1 for 90-day transplant-free survival, we performed a ROC analysis. The AUROC for CYFRA 21-1 alone was 0.64 with an optimal cut-off being 12.19 ng/mL (sensitivity  Figure S6). High serum CYFRA21-1 at the established cut-off indicated patients with a low 90-day transplant-free survival in the corresponding Kaplan-Meier curve analysis (Fig. 4a).
Notably, CYFRA21-1 levels in the highest quartile conferred a particularly bad prognosis (Fig. 4b). While the composite scores MELD (AUROC 0.70) and ACLF grade (AUROC 0.68) performed numerically better than CYFRA 21-1 alone, the combination of MELD with CYFRA 21-1 numerically improved the AUROC to 0.73 (Additional file 1: Figure S6). In multivariable Cox regression models, high serum CYFRA21-1 (unadjusted HR 2.97 [1.92-4.60]; P < .0001) remained a robust indicator of death or transplant Fig. 2 Serum CYFRA21-1 as an indicator of clinically significant portal hypertension (CSPH) and transplant-free liver-related survival in patients with portal hypertension due to advanced chronic liver disease (ACLD; cohort ii). Three hundred thirty-three patients (median follow-up of 22 months) were analyzed (cohort ii). a Correlation of serum CYFRA21-1 and hepatic venous pressure gradient (HVPG). Data from patients with hepatic decompensation or liver-related death (events) are depicted in red color whereas the data from patients without hepatic decompensation or liver-related death (eventfree) are depicted in blue color. b Serum CYFRA21-1 levels in three strata of ACLD patients with HVPG of 6-9 mmHg (no CSPH), 10-15 mmHg, and ≥ 16 mmHg. The serum CYFRA21-1 cut-off of 3.90 ng/mL was used to generate Kaplan-Meier curves (CYFRA21-1 < 3.90 ng/mL in blue, CYFRA21-1 ≥ 3.90 ng/mL in red) for the following outcomes: c the cumulative transplant-free, liver-related survival in the overall cohort and d the cumulative transplant-free, liver-related survival in patients with CSPH (HVPG ≥ 10 mmHg). Unadjusted hazard ratios (HRs) with their 95% confidence intervals [in brackets] and log-rank P values are shown in the corresponding graphs  Table S6). Similar results were obtained when CYFRA21-1 was assessed as a dichotomized and a continuous variable (Additional file 1: Table S6). Sensitivity analyses showed that the prognostic ability of serum CYFRA21-1 remained strong across several patient subgroups such as patients with/without ACLF, high/low MELD score, and with/without HCC (Fig. 4c).

Discussion
In our study, we evaluated the diagnostic relevance and prognostic significance of the K19-based CYFRA21-1 as a novel, direct serum marker of DR in different liver disease settings. We demonstrate that hepatic K19 expression remains unaltered in non-cirrhotic liver disease patients of various etiologies, which is in line with our previous findings [19]. In contrast, the expression of hepatocellular keratins K8/K18 rises already in intermediate fibrosis stages [19]. We also observed a strong correlation between the hepatic levels of K19 and the other DR keratins (K7/K23), thereby strongly suggesting that the detected elevated hepatic K19 expression mirrors an increased DR. Notably, DR is known to be particularly prominent in cirrhosis where regeneration through mitosis of mature hepatocytes becomes impaired [9,10]. Future studies correlating the hepatic expression of K19 with the serum levels of CYFRA21-1 in a well-characterized cohort of patients with various stages of chronic liver disease (i.e., non-advanced and advanced) are warranted.
The major aim of our study was to evaluate the pathophysiologic associations underlying the prognostic implications of CYFRA21-1 levels. Notably, CYFRA21-1 displayed only weak correlations with the hepatocellular injury markers AST/ALT or M30/M65 and no/minimal  correlation with liver synthesis parameters (Additional file 1: Table S5) suggesting that CYFRA21-1 mirrors liver disease severity rather than the synthetic ability of cirrhotic livers. This is in line with the current understanding of DR, that is thought to parallel liver disease severity [10,11] as well as the studies employing immunohistochemical staining for ductular keratins, that also detected a correlation with liver disease severity [12,17]. The extent of portal hypertension is an excellent indicator of liver disease severity and therefore has wellestablished prognostic implications [18]. Consequently, the predictive usefulness of CYFRA21-1 might be in part due to the fact that it correlates with the degree of portal hypertension. A correlation between DR and the extent of portal hypertension has been noted previously in a small series of patients with schistosomiasis (i.e., a cause of non-cirrhotic portal hypertension) undergoing direct portal pressure gradient [20] and is not surprising, since DR is known as the prevailing mode of regeneration in advanced/end-stage liver disease [10,11]. However, portal hypertension due to schistosomiasis differs in several aspects from the etiologies of chronic liver diseases that were evaluated by HVPG measurement in our study.
Regarding prognostic implications, in all three cohorts of patients with ACLD (cohorts ii-iv), higher CYFRA21-1 values were associated with reduced survival. This association was independent of the underlying etiology, the disease severity (i.e., compensated or decompensated cirrhosis), and a variety of potentially relevant confounders (e.g., demographic parameters). Importantly, the associations also remained robustly significant after adjustment for the MELD score, Child-Pugh score, and ACLF grade, indicating its incremental benefit over classical scoring systems. In line with this clear and consistent association of serum CYFRA 21-1 with liver-related survival, multiple studies demonstrated the usefulness of histological DR quantification in the long-term prognosis of chronic liver disease. These typically relied on the immunohistochemical detection of ductular keratins K7/K19, as an indicator of the disease course [12][13][14][15]. Our findings extend these observations and demonstrate that DR products are useful not only as histological surrogates, but may also constitute attractive serum markers. This scenario is particularly attractive given that CYFRA21-1 is already widely available in the clinical routine, and thus, our findings are easy to implement. Quantitative measures are shown as median (interquartile range) or as an absolute count (n) and relative frequency (%). Abbreviations: HCC hepatocellular carcinoma, SBP spontaneous bacterial peritonitis, MELD model for end-stage liver disease, ACLF acute-on-chronic liver failure, ALT alanine aminotransferase, AST aspartate aminotransferase, INR international normalized ratio. *Including 38 patients who were lost to follow-up after a median of 14 days. **Non-survivors include patients who died (n = 78) or received a liver transplantation (n = 10) within 90 days after study recruitment Notably, elevated CYFRA21-1 levels also associated with both HCC occurrence and HCC-unrelated liver-related mortality. While the association with HCC occurrence was rather weak, K19 is abundantly produced in liver cancer stem cells and delineates a HCC subgroup with poor prognosis [21][22][23].
In all three cohorts of patients with ACLD, the median CYFRA21-1 serum levels were mostly above the upper limit of normal for the general population (e.g., median CYFRA21-1 levels in healthy controls were described as of 1.5 ng/mL [24]). While K19 is produced in several non-hepatic tissues [1,25], these data suggest that the release of K19 from cirrhotic livers exceeds the K19 liberation from other organs. However, our study design cannot fully exclude the contribution of non-hepatic sources of K19. CYFRA21-1 does not appear to have a broad diagnostic range as a majority of values were below 6 ng/mL. While continuous increases of serum CYFRA21-1 were significantly and independently associated with survival, the corresponding ORs and AUROCs were rather low (Additional file 1: Table 2, S4, and S6). Of note, using a cut-off determined by a cohort-specific Youden index (between 3.90 and 12.19 ng/mL), CYFRA21-1's association with portal hypertension and survival was not only independent but also clear (i.e., ORs between 2 and 7). Given the steady hepatic K19 expression in noncirrhotic fibrosis stages (cohort i) and the increasing relevance of DR as a primary source of regeneration with increasing parenchymal loss and the concomitant tissue remodeling [9,10], CYFRA21-1 does not appear to be discriminative in non-advanced chronic liver disease. Instead, CYFRA21-1's diagnostic utility lies rather in its association with CSPH and its independent survival prognostication when the serum levels are rather high. One might speculate that CYFRA21-1 is of particular prognostic relevance in decompensated cirrhosis and/or ACLF with a strong DR. However, further studies are warranted to determine CYFRA21-1's utility in prospective cohorts of patients with advanced and non-advanced chronic liver disease.

Conclusions
While CYFRA21-1 is a well-established and widely available serum marker of several non-hepatic tumors [5], our Fig. 3 Serum CYFRA21-1 as an indicator of long-term death and HCC occurrence in patients with alcoholic cirrhosis (cohort iii). Two hundred thirty-one patients with cirrhosis due to long-term alcohol misuse (median follow-up of 73 months) were analyzed. The median serum CYFRA21-1 value was used to generate Kaplan-Meier curves (CYFRA21-1 < 5.26 ng/mL in blue, CYFRA21-1 ≥ 5.26 ng/mL in red) for the following outcomes: a the overall transplantation-free survival, b the liver-related, not HCC-related transplantation-free survival, and c the occurrence of HCC. Unadjusted hazard ratios (HRs) with their 95% confidence intervals [in brackets] and log-rank P values are shown in the corresponding graphs data strongly suggest that it can be repurposed to serve as the first direct serological marker of DR. Regardless of the underlying etiology, serum CYFR A21-1 levels are consistently elevated in different ACLD cohorts and are linked to the severity of portal hypertension. Furthermore, serum CYFRA21-1 levels appear to be a stable marker independent of conventional liver biochemistry markers and prognostic scores (e.g., MELD) pointing to an added benefit in terms of prognostication and pathophysiological insight. Importantly, CYFRA21-1 does not only reliably predict both the short-and long-term mortality of ACLD but also holds a great promise to promote our understanding of the biological importance of DR.