Introduction

Nonalcoholic fatty liver disease (NAFLD) is the most common cause of liver disease worldwide, affecting approximately 20 % of the general population, based on different assessment methods such as ultrasonography, liver biopsy, or liver test abnormalities [1]. In the USA, some studies based on MRI evaluation estimate a prevalence of 31 %, while others 47 % based on ultrasonography findings [2, 3]. While the prevalence of other liver diseases has been stable over the last 20 years, NAFLD has increased remarkably over the last years along with the prevalence of metabolic syndrome [4]. With the current and rising prevalence of NAFLD, diagnostic tools not only for disease assessment but also for follow-up are required.

The natural history of NAFLD depends on severity of the liver damage and population risk factors, such as obesity and diabetes [5]. Disease spectrum ranges from the mild benign form with simple steatosis (NAFL) to the more severe form of the disease in 10–20 % of these patients, known as nonalcoholic steatohepatitis (NASH), characterized by steatosis, inflammation, and hepatocyte ballooning (apoptosis), leading to fibrosis [1, 6]. NASH has a more severe disease course, in which patients are more susceptible to develop liver cirrhosis and hepatocellular carcinoma [7, 8], so identifying patients at risk of end-stage liver disease is essential during evaluation and management [9]. In addition, patients with NASH have increased overall mortality, cardiovascular diseases, and liver-related complications [10]. Although some studies suggest that not only NASH but also NAFL can progress to liver fibrosis, the annual progression rate in patients with steatohepatitis in liver biopsy is higher than simple steatosis (one stage of progression over 7.1 vs. 14.3 years, respectively) [11–13]. Moreover, the prognosis of patients with NASH appears to be directly related to fibrosis stages, which can predict both overall and disease-specific mortality. Ekstedt et al. [14] demonstrated that patients with fibrosis 3–4, irrespective of NAS, had increased mortality, with a HR 3.3.

At present, liver biopsy is the current way to evaluate the presence of NASH and assess the severity of liver damage; however, it is an invasive and expensive procedure, with known sampling variability and risks of complications such as bleeding, bile leaks, and death is rare cases. Therefore, noninvasive methods to distinguish between NAFLD and NASH and to predict fibrosis stage are imperatively required.

A number of serum biomarkers have been studied in NASH, with the most well-studied biomarker being cytokeratin-18 (CK18) fragments. CK18 is a keratin-containing protein involved in cytoskeleton cell formation. During hepatocyte apoptosis, it is cleaved by caspases and released in circulation as fragments (M30 and M65), which are easily detectable and used as markers of apoptosis to identify patients with NASH. CK18 M30 fragment is more related to apoptosis than M65, as the latter can be detected during tissue necrosis, and may be released in circulation despite caspase activation [15].

The first study addressing its importance was presented by Wieckowska et al. [16] when a pilot study demonstrated that caspase-generated CK18 fragments were higher in patients with NAFLD compared to controls and able to differentiate between low and high fibrosis and inflammation stages. Since then, numerous studies suggest a wide range of potential usefulness of this test as a noninvasive tool in NASH diagnosis, to differentiate NASH from simple steatosis and as a marker of disease severity and response to treatment [17–19]. In addition, recent studies suggest models combining CK18 with different biomarkers to improve diagnostic performance [20, 21].

Although previous studies [1, 22] showed that CK18 is a clinical relevant biomarker and has sufficient diagnostic accuracy in detection or exclusion of NASH, recent publications showed poorer test performance of CK18 with a low sensitivity as a diagnostic test for NASH [23, 24]. The differences in test performance are seen between studies using the same ELISA kits and between studies that use different kits. There are a number of CK18 fragment ELISA kits commercially available. However, there does not seem to be standardization between the kits, or data available comparing the test performance of different kits. Different studies on the test performance of CK18 as a NASH biomarker have varying results, raising the question of whether the different findings are due in part to differences in the test kits.

The aim of this study is to compare serum measurements of the CK18 M30 fragments by two different CK18 M30 ELISA kits using the same cohort of patients to test inter-test reliability.

Materials and Methods

Patient Population

The subjects for this study were enrolled in an NAFLD patient registry at Beth Israel Deaconess Medical Center (BIDMC) from 2009 to 2014. The BIDMC NAFLD patient registry is a prospective study of subjects with biopsy-proven NAFLD. Patients with other chronic liver diseases or consumption of >20 g alcohol daily were excluded from the registry. Data on patient demographics, medical history, and from their physical examination were obtained at the enrollment of the study. At the time of this study, there were 183 patients enrolled in the NAFLD registry study. The study has been approved by the BIDMC institutional review board.

Liver Biopsy and Histological Assessment

Ultrasound-guided liver biopsy was performed within 3 months of the baseline visit and evaluated by an experienced liver pathologist, unaware of subjects’ clinical information, using the NASH Clinical Research Network Scoring System developed by Kleiner et al. [25]. Advanced fibrosis and NASH were defined by liver biopsy findings. Advanced fibrosis was defined as fibrosis stages 3–4. According to Kleiner et al. [25] NAFLD activity score (NAS score) was also considered for diagnosis of NASH when >5. We defined severe hepatocyte ballooning as having a hepatocyte ballooning score of 2 and severe lobular inflammation as having score of 2–3. For subsequent analyses, we studied the role of CK18 in prediction of NASH and advanced fibrosis, as these are the most important clinical predictors related to disease severity and mortality [11, 13].

Clinical Biochemistry and Measurements of CK18 Fragments

Laboratory tests and collection of serum were performed at enrollment. Baseline serum was stored at −80 °C. Routine blood tests including complete blood count, chemistry, liver function tests, albumin, and lipid panel were processed at the BIDMC clinical laboratory.

Serum CK18 M30 fragment was measured using two different ELISA kits: Human cytokeratin 18®, Biotang (Test 1) and M30-Apoptosense, PEVIVA®, DiaPharma (Test 2). All procedures were done according to kit manufacturer’s instructions. Intra-assay variability was evaluated by using the intra-assay variability coefficient (CV%) for both tests. CV% was calculated by using the standard deviation of the means after bootstrap analysis from each test, divided by the overall mean of each group. CV% <10 was considered acceptable.

Test 1

CK18 concentration in the sera was measured using the Human CK18 ELISA kit (Biotang, Lexington, MA, USA). Antihuman KRT18 mAb is derived from hybridization of mouse FO myeloma cells with spleen cells from BALB/c mice immunized with recombinant human KRT18 amino acids 79-430 purified from Escherichia coli. This kit is a presentation of using antibody against CK18 full protein, able to detect all fragments generated after caspase activation. Thus, high levels of this biomarker are directly related to M30 fragment levels, the main fragment released during the apoptosis process.

The ELISA was performed following protocol provided by the manufacturer. In short, 10 µL of serum samples was pipetted into the pre-coated well along with standards. After incubation, the protein antigen and a biotinylated monoclonal antibody specific for target protein were simultaneously incubated. The streptavidin–peroxidase followed with a substrate solution was added after washing to induce a colored reaction. Colorimetric absorbance was read at 450 nm on a SpectraMax 190 Microplate reader (Molecular Devices, Sunnyvale, CA, USA). Serum CK18 concentration was expressed in U/L (1 U/L = 1.24 pM recombinant protein standard).

Test 2

CK18 levels were estimated by using in vitro immunoassay M30-Apoptosense ELISA kit (PEVIVA®, DiaPharma). This one-step test is designed to measure soluble caspase-cleaved keratin 18 fragments in serum. K18Asp396 neo-epitope is termed M30 antigen. An antibody directly against it is used in this test.

One single researcher was responsible for all techniques, following manufacturer’s instructions. Absorbance was read at 450 nm. Serum concentration was expressed in U/L; also, all values were according to standard curve ranges.

Statistical Analysis

Data were analyzed using a statistical software program (IBM® SPSS® Statistics, version 22.0). Two-tailed Student’s t test and ANOVA were calculated for pairwise comparison of continuous variable from both tests. Categorical variables were expressed in percentages and analyzed using Chi-square test or Fisher’s exact test when applicable. Statistical significance was defined when p < 0.05, using two-tailed tests. Results from CK18 ELISA tests were compared using Pearson’s correlation test or partial correlation Pearson tests controlling for the effect size. Advanced fibrosis and NASH prediction was estimated for each CK18 test using binary logistics regression analysis, and acquired confidential intervals were compared with those using bootstrap technique. R 2, variables’ significance, and odds ratio were compared between models from both tests. The performance of each CK18 tests for prediction of NASH and advanced fibrosis was determined using area under receiver operating characteristic curve (AUROC).

Results

Patient Characteristics

The main demographic, clinical, and laboratory characteristics of the patients are presented in Table 1. Serum CK18 M30 measurements were taken using both Test 1 and Test 2 on a total of 172 subjects. Only Test 2 was performed on the serum of an additional 11 patients who were enrolled after Test 1 was performed, bringing the total number of patients with Test 2 done to 183. There was no difference in baseline characteristics between groups of patients who had both T1 and T2 performed versus those who only had T2 performed. NASH was diagnosed in 49 % of patients, according to criteria proposed by the NASH Clinical Research Network [25]. Patient age, ethnicity, and race were similar in patient with and without NASH. While 60 % of the total cohort was male, there were more women in NASH group than in the non-NASH group (47 vs. 34 %, p = 0.024). Mean body mass index (BMI), ALT, and AST were higher in patients with NASH, otherwise diabetes and hypertension had similar frequencies.

Table 1 Patient characteristics
Full size table

The histological features of subjects are presented in Table 2. Of the 49 % patients who had NASH diagnosed, 15.5 % had advanced fibrosis (grades 3 or 4); 31.6 % presented with severe inflammation (grade 2) and 24.6 % with severe ballooning (grade 2).

Table 2 Histological findings
Full size table

CK18 Levels from Both Tests and Relation to Liver Histology Findings

Intra-assay variability was considered acceptable for both tests. CV% for Test 1 and Test 2 was 9.2 and 6.0, respectively.

While the mean serum CK18 level measured by Test 2 was significantly higher in patients with NASH, advanced fibrosis, or severe ballooning, the mean level measured by Test 1 was not significantly different between the different groups (see Table 3). CK18 level measured by Test 2 did not differ between low and severe inflammation, although a trend suggesting this difference was noticed (p = 0.053).

Table 3 CK18 concentration according to different liver histological parameters
Full size table

One-way ANOVA and post hoc test were used to identify differences in CK18 levels among the various stages of steatosis, inflammation, ballooning, and fibrosis. Significant differences were found just in CK18 from Test 2, except among inflammation stages. Test 1 was ineffective in distinguishing stages of any parameters.

There were no significant differences for Test 2 and no differences for Test 1 concerning inflammation stages. Steatosis: significant differences, using Test 2, between grade 1 versus 3 (p = 0.026); ballooning: grade 0 versus 2 (p = 0.037), 1 versus 2 (p = 0.049); and fibrosis: grade 0 versus 2, 3, and 4 (p = 0.002, 0.005, 0.002).

Poor Inter-test Reliability Between CK18 Kits

There was no significant correlation between measurements from the two tests using the Pearson’s correlation (p = 0.86, r = 0.01) or the partial correlation Pearson tests controlling for the effect size of fibrosis, NAS score, and hepatocyte ballooning (p = 0.65, 0.89, and 0.81, respectively).

Binary logistics regression was carried out to assess serum level of CK18 measured by both tests as a predictor of NASH and advanced fibrosis. The results are given in Table 4. Serum CK18 level from Test 2 was a significant predictor of NASH and advanced fibrosis, whereas CK18 level from Test 1 was not.

Table 4 Logistics regression model for prediction of different liver histology parameters according to Test 1 and Test 2
Full size table

To assess the role of CK18 as a diagnostic tool for the prediction of NASH and advanced fibrosis, the receiver operating characteristic curves (ROCs) were constructed and are shown in Fig. 1. Again, Test 1 performed very poorly for both the prediction of NASH and advanced fibrosis with the area under the ROC (AUROC) of 0.513 (95 % CI 0.425–0.601) for NASH and 0.517 (95 % CI 0.408–0.627) for advanced fibrosis. Test 2 performed better, with AUROC of 0.638 (95 % CI 0.555–0.722) for NASH and 0.676 (95 % CI 0.571–0.782) for advanced fibrosis.

Fig. 1
figure 1

a A comparison of CK18 by Test 1 (gray line) and by Test 2 (black line) in predicting NASH. b A comparison of CK18 by Test 1 (gray line) and by Test 2 (black line) in predicting advanced fibrosis (stages 3–4). The predicted AUROC with its 95 % CI was shown below each panel

Full size image

Table 5 lists the overall sensitivity, specificity, positive and negative predictive values, and positive and negative likelihood ratios of CK18 by Test 2 for NASH and advanced fibrosis. All these values were calculated using the optimal cutoff points of CK18 levels after Youden’s index analysis (356.18 and 395.97 U/L for NASH and advanced fibrosis diagnosis, respectively). Test 2 has poor sensitivity for both NASH and advanced fibrosis, showing that CK18 as a biomarker alone will miss half of the NASH cases.

Table 5 CK18 by Test 2 performance for the diagnosis of NASH and advanced fibrosis
Full size table

Discussion

The aim of this study was to assess the inter-test reliability between two CK18 ELISA kits in a cohort of patients with biopsy-proven NAFLD and to evaluate the test performance of this biomarker in the prediction of NASH and advanced fibrosis in patients with NAFLD. This study highlights the absence of correlation between the results from two commercially available CK18 kits.

Given the high prevalence of NAFLD and the variable severity and disease course, a simple, quick inexpensive biomarker for risk stratification and to monitor progression is very appealing. Given the previous data suggesting CK18 as a practical and reproducible noninvasive biomarker in NASH [17, 22, 26], it is increasingly used in clinical trials. Our study confirms that serum levels of CK18 fragments are significantly higher in NASH and advanced fibrosis (p = 0.001, p = 0.001, respectively) when Test 2 is used. However, our results showed that Test 1 was essentially useless as a biomarker in NAFLD.

Although statistical differences were demonstrated between these groups, low sensitivity (Test 1 50 % and Test 2 60 %) and specificity (Test 1 79 % and Test 2 75 %) showed that this biomarker alone was insufficiently accurate as a predictor of NASH and advanced fibrosis. Previous studies, using the same Test 2 for the diagnosis of NASH, achieved sensitivity of 56–77 % and specificity of 63–92 %, demonstrating a wide range despite using the same kit [22–24]. Using the optimal serum CK18 cutoffs for our cohort (356.18 U/L for NASH and 395.97 U/L for advanced fibrosis diagnosis), the accuracy of Test 2 is 64.6 and 71.8 %, respectively, resulting in the misclassification of 35.4 and 28.2 % patients for NASH and advanced fibrosis, respectively.

The AUROC was estimated to assess the role of CK18 M30 as a diagnostic tool for the prediction of NASH and advanced fibrosis (0.64 and 0.67, respectively). Although the multicenter validation study published by Feldstein et al. showed higher AUROC for prediction of NASH (0.83), most of recent studies demonstrated lower values ranging from 0.53 to 0.63 for NASH and 0.53 to 0.68 for fibrosis prediction [23, 24, 27]. One possible explanation for this contradictory result is the different percentages of subjects with NASH among studies—49 % of our cohort had NASH, compared to just 19 % of the previous validation cohort. All of the studies presented above used the same kit, suggesting that besides variability between kits, other factors such as differences in the study cohorts and test reliability may be at play.

Recent studies [23, 24] report poor performance of CK18 fragments as a biomarker to identify NASH and advanced fibrosis in NAFLD patients supported by low AUROC, sometimes similarly to conventional biochemistry evaluation like AST and ALT. Chan et al. [23] highlight the limited utility of M30 in the detection of NASH based on low AUROC for NASH (0.59), lower than the AUROC achieved by ALT (0.64) and AST (0.75) in the same cohort of patients. Similarly, Cusi et al. [24] concluded that although CK18 M30 has a reasonable sensitivity for NAFLD and any stage of fibrosis (68 and 85 %), its low specificity (58 and 54 %, respectively) rendered it a poor test for screening and staging NASH. Also, the same study reinforced the low AUROC for prediction of NASH (0.65) and the presence of fibrosis (0.68).

The discordance between the different studies of CK18 test performance as a biomarker of NASH diagnosis may be influenced by different ethnic population, size of cohorts, prevalence of diabetes, and other comorbidities and unbalance between forms of disease presentation (simple steatosis, NAFLD, and NASH). All these factors together prevent a clear interpretation about the discordance among studies.

In our study, the noticeable lack of correlation between the two tests in the same cohort highlights the importance of standardization of all the available kits and may explain the differences in findings between different studies. However, given the wide variability between two different kits in the same cohort shown by our study, we suggest that the kit’s influence is significant and should not be undervalue.

CK18 kits are only available as a research tool and not yet evaluated for clinical purpose in the setting of NASH disease. Thus, this method did not undergo through the extensive protocols involved during regulatory entities validation and approval. Further analysis is needed to determine batch-to-batch variability and variability with repeated freeze/thaw cycles. This is particularly important in larges studies that work with large cohorts through long periods of follow-up, where tests will also be run in multiple batches.

Further studies on the standardization of CK18 kits are needed to determine which kit is most reliable and has the best test performance as a biomarker in NAFLD. In addition, combinations of several serologic markers with better performance to identify NASH or advanced fibrosis are a promising alternative [20, 21]. While serum CK18 fragments is insufficiently accurate as a biomarker in NAFLD, its measurement using the right ELISA kit remains promising as an important tool for NAFLD evaluation.

This study does have some limitations. The average biopsy length (12.4 ± 6.5 cm) might be considered low according to guidelines parameters [9, 25], but also as the main end point of our study is to evaluate CK18 M30 test performance; thus, correlations between biomarker levels were not affected by histological features. Our study used histopathological examination of liver biopsies as a reference to stages liver fibrosis, which may underestimate severity due to sample error and intra-observer variability. Another weakness of this study is not have a control group; comparisons were made between NASH versus non-NASH groups and patients with and without advanced features in liver biopsy.

This study is the first study comparing serum CK18 fragment levels between different kits in a large group of well-characterized NAFLD patients with biopsy-proven NASH. In summary, our findings show that while serum CK18 M30 level using the right test kit has a role in the noninvasive assessment of NAFLD, there are significant variations between ELISA kits, which could greatly bias the results. There is a need for standardization of the multiple available CK18 ELISA kits.