Development and validation of a novel circulating fibroblast activation protein - based predictive model to improve fibrosis risk stratification in metabolic liver disease population

1 Introduction

Fatty liver disease is becoming an increasingly common metabolic disorder and currently affects one quarter of the global population^{1, 2}. The pathogenesis of fatty liver disease encompasses multi-factorial causes such as abnormal lipid metabolism, glucose dysregulation, gut microbiome alterations and genetic variants³. These abnormal physiological changes promote chronic inflammation that can lead to hepatic fibrosis, which is the most potent driver towards end-stage liver disease and hepatocellular carcinoma. Moreover, obesity, type 2 diabetes mellitus (T2DM) and insulin resistance drive this disease^{4, 5}. Early-stage metabolic fatty liver disease and hepatic fibrosis is usually asymptomatic^{6, 7}. However, the risk of liver-related morbidity and mortality increases dramatically in parallel with the progression into advanced hepatic fibrosis and cirrhosis^{8, 9}. As the presence of advanced fibrosis is the major determinant of liver-related complications, accurately identifying those at risk in the context of increasing metabolic associated fatty liver disease prevalence has become an urgent clinical need for hepatologists, diabetologists and primary care physicians.

Histological assessment with liver biopsy is becoming less frequently used due to its invasive and inherent sampling bias. New 3D imaging techniques highlight the latter limitation by starkly revealing the uneven localisation of fibrosis¹⁰. Elastography techniques are not widely accessible in the community. Therefore, non-invasive blood tests have come to the forefront for the diagnosis of advanced liver fibrosis in patients with metabolic disorders. The two most commonly used models are the NAFLD fibrosis score (NFS)¹¹ and the Fibrosis index-4 (FIB-4)¹². However, despite being recommended for advanced fibrosis screening, these tests require further optimisation^{13, 14}. Recently, the European Association for the Study of the Liver (EASL) has proposed a three-step strategy including non-invasive tests (NITs) to screen for advanced fibrosis¹⁵. This algorithm has been further validated for discriminating advanced fibrosis, where it displays superior accuracy to a single test algorithm⁸. Despite the limitations of these NITs, the application of low-cost and simple sequential NITs appears more appropriate to screening in the community setting than higher cost alternatives.

Fibroblast activation protein alpha (FAP) is a dimeric type 2 trans-membrane glycoprotein with proteolytic activity that degrades fibroblast growth factor 21 (FGF21)^{16, 17}, neuropeptide Y¹⁸ and denatured collagens^{17, 19, 20}, and activates α2-antiplasmin¹⁹. FAP is near-absent from healthy tissues but highly expressed by activated stellate cells and myofibroblasts in cirrhosis^20–22. Moreover, FAP is heavily glycosylated (∼30% by mass), which may confer resistance to degradation that explains the persistence of soluble FAP in circulation (cFAP)²³. We previously reported our specific enzyme assay of FAP^{24, 25}, and have shown that increased cFAP activity associates with advanced fibrosis (F3-F4)²⁵. Additionally, cFAP activity decreases following liver transplantation²⁶. Thus, cFAP activity is a potential serum biomarker for liver fibrosis that derives from the cell types that produce and regulate most of the liver extracellular matrix (ECM)^{17, 19}. Moreover, FAP colocalizes with ECM proteins in liver fibrosis^{22, 27, 28}. Thus, it is important to explore the utility of measuring FAP when assessing fibrosis. FAP can be measured by multiple approaches including specific substrate catalytic assay, antibody capture catalytic assay or ELISA^{24, 29–31}. Our robust, single-step in-house FAP assay that uses a FAP-specific fluorescent substrate²⁴ strongly correlates with the commercial ELISA. ^{24, 26, 29–31}

In this study, the primary aim was to develop a NIT-based fibrosis algorithm that incorporates our specific, quantitative cFAP enzyme activity assay in patients with metabolic associated fatty liver disease to identify advanced fibrosis, and then to validate the performance of this algorithm upon a cohort that mimics a community population. We also examined whether the two most prominent FAP-specific substrates^{24, 29} generate comparable quantitative data on FAP abundance in serum samples.

2 Method

3 Results

3.3 Application of cFAP for hepatic fibrosis detection

The cFAP activity was significantly associated with fibrosis (supplemental table 2). In particular, cFAP of patients with F3-F4 was significantly greater (P<0.0001) than patients without fibrosis (F0) in the training cohort (figure 1A-B), suggesting an ability to discriminate between absence of fibrosis and presence of advanced fibrosis. Significant differences in cFAP were also observed in inter-stage comparisons with moderate liver fibrosis in the validation cohort.

• Download figure
• Open in new tab

Figure 1.

Associations of cFAP activity, FAP Index and FIB-4 with fibrosis staging. The cFAP activity (U/L) (A, B), FAP Index (C, D) and FIB-4 (E, F) segregated according to fibrosis stage are displayed for training cohort (n=160) (A, C, E) and validation cohort (n = 332) (B, D, F). Dot plots with mean ± SEM. Differences between groups were determined using a one-way ANOVA with Tukey’s post-hoc test. Significant differences are indicated with asterisks to indicate degree of difference: * p-value<0.05, ** p-value<0.01, *** p-value<0.001, **** p-value<0.0001.

3.4 Model development for the detection of advanced fibrosis

Multivariable logistic regression revealed age, T2DM, ALT and cFAP (table 2) as predictors of advanced fibrosis. We therefore derived FAP Index using the following formula:

View this table:

• View inline
• View popup

Table 2. Associations between clinical parameters and advanced fibrosis in training cohort. ^a Univariate logistic regression; ^b Chi-square test.

The Hosmer-Lemeshow test ξ² was 3.066 (P = 0.93), indicating that FAP Index was well fitted. In the training cohort, the AUROC for FAP Index in predicting advanced fibrosis was 0.875 (95% CI: 0.813 – 0.938; figure 2A). Based on the ROC, a dual cut-off strategy was applied to optimise FAP Index for ruling in and out advanced fibrosis. The low cut-off was chosen at −1.681 (sensitivity = 0.84, specificity = 0.75), and the high cut-off at 0.953 (sensitivity = 0.34, specificity = 0.99). Patients with a score lying in between these low and high cut-off scores were designated indeterminate.

• Download figure
• Open in new tab

Figure 2.

(A) The receiver operating characteristics curve for the FAP Index in the training cohort (AUROC = 0.875 with 95% CI: 0.757 – 0.927). (B) The AUROC for the FAP Index is comparable with FIB-4 and NFS. Receiver operating characteristics curve of test scores in the training cohort subset including FAP Index (AUROC = 0.842 with 95% CI: 0.757-0.927), FIB-4 (AUROC = 0.894 with 95% CI: 0.825-0.963) and NFS (AUROC = 0.779 with 95% CI: 0.67-0.889). Statistical significance between ROC curves was tested by DeLong test.

The association of FAP Index with fibrosis severity (figure 1C-D) displayed better differentiation across fibrosis stages than with cFAP alone and comparable with FIB-4 (figure 1E-F).

3.5 Comparison of FAP Index to existing NITs

Only patients with complete data for FAP Index, NFS and FIB-4 were included, which was designated as the training sub-cohort (n=87). The training sub-cohort was compared with its original cohort (supplemental table 2) and was older (57 vs 52, P=0.03), had lower BMI (34 vs 38, P = 0.02) and had higher insulin levels (15 vs 11, P=0.02) and cholestatic liver enzymes. Notably, there was no difference in the prevalence of advanced fibrosis. The FAP Index accuracy was numerically superior to NFS and similar to FIB-4 (AUROC 0.842, 0.779 and 0.894, respectively), but without a statistically significant difference between FAP Index and FIB-4 or NFS (figure 2B). The NPVs of FAP Index, NFS and FIB-4 were 95%, 92.6% and 95.9%, respectively, with corresponding specificities of 97.7%, 80.6% and 93.8%. The sensitivities of FAP Index, NFS and FIB-4 were between 80 and 85%. Use of FAP Index and FIB-4 as standalone tests resulted in 39.1% and 32.2% of patients classified as indeterminate, respectively, whereas NFS generated 49.4% indeterminate (figure 3A-C, table 3). FAP Index, when considered, as a continuous variable, correlated with both FIB-4 and NFS (P < 0.001; supplemental Material S2.3; supplemental figure 4)

• Download figure
• Open in new tab

Figure 3.

Classification analyses in training cohort. Fibrosis stages classification as single test or sequential tests, as a function of subgroups defined by: A. FAP Index classification. B. FIB-4. C. NAFLD Fibrosis Score. D. FAP Index followed by FIB-4. E. FAP Index followed by NAFLD Fibrosis Score. F. Fibrosis stages as a function of subgroups defined by FIB-4 followed by FAP Index. G. NAFLD Fibrosis Score followed by FAP Index.

View this table:

• View inline
• View popup
• Download powerpoint

Table 3. Summary table for these classification analyses of all tests in the training subcohort (n=87), reported for NPV, PPV, proportion of indeterminate, sensitivity and specificity. *: Sensitivity and Specificity were calculated with the indeterminate results excluded.

3.6 FAP Index validation

The AUROC for FAP Index to predict advanced fibrosis in the validation cohort was 0.841 (figure S5). In the validation cohort, FAP Index and FIB-4 produced similar outcomes, with specificity above 98%, which were superior to NFS (specificity 64.4%). The NPVs for all three tests were above 90% (figure 4A-C, table 4). The FIB-4 sensitivity was 37.5%, while sensitivities for FAP Index and NFS were over 50%. Importantly, all tests exhibited large proportions of indeterminate: 32.6% for FAP Index, 24.1% FIB-4 and 47.5% for NFS. FAP Index was significantly correlated with FIB4 and NFS in the validation cohort (P < 0.0001; supplemental figure 4).

• Download figure
• Open in new tab

Figure 4.

Classification analysis in the validation cohort. Fibrosis stages classification as a single test or sequential tests, as a function of subgroups defined by A. FAP Index classification. B. FIB-4. C. NAFLD Fibrosis Score. D. FAP Index followed by FIB-4. E. FAP Index followed by NAFLD Fibrosis Score. F. FIB-4 followed by FAP Index. G. NAFLD Fibrosis Score followed by FAP Index.

View this table:

• View inline
• View popup
• Download powerpoint

Table 4. Summary table for these classification analyses of all tests in the validation cohort, reported for NPV, PPV, proportion of indeterminate, sensitivity and specificity. * Sensitivity and Specificity were calculated with the indeterminate results excluded.

3.7 Sequential application of NITs

Next, sequential application of each possible combination of NITs was performed, with the primary aim to minimize the proportion of patients who were indeterminate without compromising diagnostic accuracy (figure 5). All combinations and sequences of NITs generated similar NPVs, above 90%, in both training cohort (table 3) and validation cohort (table 4). In the training cohort, FAP Index applied as either the 1^st line or 2^nd line NIT in combination with either NFS or FIB-4 reduced the proportion of indeterminate to 20.7% (figure 3D-G; table 3). In the validation cohort, FAP Index used in combination with NFS resulted in 14.9% indeterminates, whereas FAP Index with FIB-4 in any combination led to 16.8% indeterminates (table 4, figure 4D-G). In both cohorts, the relative reduction in the indeterminate population caused by adding the second line test reached 50%. However, combinations of FAP Index with FIB-4 followed by FAP Index maintained specificity of ∼95% in both cohorts (98% in validation cohort) compared to NFS combinations with FAP Index where specificity was as low as 78% in the validation cohort (tables 3 and 4). This is reflected in figure 4, where although NFS combinations classified more patients as high risk, up to 80% did not have histologic advanced fibrosis. Importantly, applying a step-wise test approach did not influence the sensitivity and PPV compared with applying a single test. Sensitivity was ∼75% in the training cohort, and ∼50% in the validation cohort. As expected in the validation cohort where the prevalence of advanced fibrosis was lower, NPV increased and PPV reduced compared to the training sub-cohort. Overall, the combination of FAP Index followed by FIB-4 generated the optimal diagnostic metrics, sufficiently reducing indeterminate results while maintaining an excellent NPV and acceptable PPV compared to single NITs.

• Download figure
• Open in new tab

Figure 5.

Flowchart for model validation in all cohorts, using two sequential tests to further classify the indeterminate patients from the 1^st line screening test.

4 Discussion

In this study, we demonstrated that circulating fibroblast activation protein (cFAP), a marker of activated stellate cells and myofibroblasts, can be incorporated into an algorithm and applied following NFS or FIB-4 to greatly diminish the number of indeterminate outcomes without compromising sensitivity or specificity for advanced hepatic fibrosis. This novel NIT, which we call FAP index, was examined in a training cohort and validated in a pooled cohort of individuals with metabolic fatty liver disease with variable prevalence of biopsy proven advanced fibrosis, type 2 diabetes, obesity and metabolic syndrome. The FAP Index utilises cFAP and three routinely obtained parameters; age, T2DM status and ALT. We have demonstrated the combination of FAP Index followed by FIB-4 provides a significant reduction in indeterminate results without compromising diagnostic accuracy.

Of the variables included in FAP Index, ALT is a well-established biomarker for the severity of liver injury; age is a known risk factor for progression for liver disease,⁴² T2DM and insulin resistance are drivers of metabolic fatty liver disease progression, and insulin resistance increases intrahepatic oxidative stress^{3, 5, 43, 44}. The presence of these variables suggests a dysmetabolic state with liver injury resulting in continual hepatic stellate cell (HSC) activation. We have shown that FAP is strongly associated with the presence of advanced fibrosis^{20–22, 25}, and FAP is found in and on activated HSC and myofibroblasts and contributes to collagen turnover^{20, 21}. Moreover, new FAP/⁴²radionuclide-based 3D imaging methods have shown that intrahepatic FAP strongly aligns with fibrosis severity¹⁰. Serum FAP drops following liver transplantation²⁶, which suggests that liver is the origin of increased circulating FAP in patients with advanced fibrosis.

In the training sub-cohort, FAP Index showed excellent accuracy in predicting the presence of advanced fibrosis, with an AUROC of 0.875 (95% CI: 0.813-0.938), while FIB-4 and NFS achieved comparable AUROCs of 0.899 (95% CI: 0.833-0.965) and 0.781 (95% CI: 0.672-0.89) respectively. Our findings demonstrated that the FAP Index has a strong discriminating capability among cohorts of varied patient composition and prevalence of advanced fibrosis. However, the percentage classified as indeterminate was relatively large as standalone NIT, which is an established issue for FIB-4 and NFS.

In contrast, sequentially using FAP Index with either FIB-4 or NFS was shown to dramatically reduce the number of indeterminate results by between 30% and 70%, depending on the NIT combination. ^{13, 45} Notably, the two-step strategy did not significantly alter the NPV, PPV nor specificity when using FAP Index as either 1^st line test or 2^nd line test in combination with either FIB-4 or NFS. Indeed, our group and others have shown that sequential screening tests that employ dual cut-offs can the increase diagnostic accuracy of NITs and thereby reduce the number of biopsy referrals^46–48. Similarly, a recent evaluation of the current EASL algorithm by Patel and colleagues found that a stepwise strategy can improve diagnostic accuracy⁸. However, the use of FIB-4 or NFS as first line NITs resulted in approximately 35% and up to 65% indeterminate results, respectively, in both the diabetes clinic and primary care cohorts analysed. Furthermore, up to 84% needed further assessment (ie indeterminate or high-risk classification) when using NFS or FIB-4 ⁸. In contrast, our two-step algorithm of FAP Index followed by FIB-4 resulted in 39% in the training subcohort and only 21% in the validation cohort who would need further investigation by the same criteria (ie indeterminate or high-risk classification). This reduction compared to the Patel study was in the context of similar advanced fibrosis prevalence in our training subcohort of 27%, however only 48 patients underwent biopsy in the Patel cohorts.

Another advantage of the sequential combination of FAP Index with FIB-4 was the high specificity. It is essential to make sure that false positives and, particularly in the context of screening, that false negatives are minimized. Our validation cohort had a prevalence of advanced fibrosis similar to the community of 11%^{35, 36} and using FAP Index followed by FIB-4 resulted in 3.6% (9/249) false negatives and 35.7% (5/14) false positives. This strategy provided the lowest rate of false negative diagnoses and comparable false positives to FIB-4 and NFS, in addition to the dramatic reduction in indeterminate results. This suggests that FAP Index has potential benefit in community screening algorithms over current standard of care.

FAP Index has potential as a readily applicable first line test as it contains variables that are easily obtained, and cFAP measurement requires only a standard fluorescence measurement device, so its cost is likely to be comparable with AST and ALT. Similarly, the measurement of ordinal cFAP in FAP Index could be readily modified into a point-of-care lateral flow test, or as a reflex test during the measurement of liver biochemistry as part of clinical pathology services.

Community demand for screening for hepatic fibrosis is rising, particularly in diabetes clinics, so inexpensive, simple liver fibrosis testing is needed. Although biopsy is considered as the most accurate diagnostic tool in clinical guidelines, it has unavoidable pitfalls such as accessibility, invasiveness, the need for specialised staff, variability between practitioners, and economic impacts. While elastography is a useful alternative to biopsy, accuracy drops when BMI exceeds 44, and it is influenced by operator-dependent variables and barriers still exist in terms of accessibility and/or affordability⁴⁹. Therefore, the potential for FAP Index as an accurate blood test and point-of-care test in routine screening could alleviate total health care costs. More importantly, the enzyme activity of soluble FAP was found to have high stability with extended ultracold storage ²⁴ (supplemental figure2), which means that the cFAP assay is amenable as a subsequent test after an initial FIB4 or NFS is completed. Thus, FAP Index as a reflex test would not require an additional blood collection, which further increases convenience and reduces health care expenditures.

Following a successful clinical trial⁵⁰, the U.S. Food and Drug Administration recently approved a thyroid hormone receptor beta (THR-β) agonist⁵¹ as the first therapy for non-cirrhotic steatohepatitis and advanced stages of liver fibrosis. This further motivates the need for more accurate triage of patients with advanced fibrosis. NFS and FIB-4 are currently regarded as standard of care for fibrosis screening, however, these NITs rely upon parameters that are strongly associated with liver function, liver injury and inflammation, rather than directly with fibrosis per se^{11, 12}. As a result, these tests can yield large proportions of misclassifications among patients at high risk of metabolic fatty liver disease¹⁴. Liver fibrosis is dependent on the intensity of tissue remodelling and myofibroblast activation. Therefore, including a simple biomarker such as FAP that directly reflects the fibrotic process is advantageous.

A major strength of this study was the robust training and validation cohorts. The prevalence of fibrosis in the training cohort was similar to that of hepatology clinic populations, while the validation cohort mimics community screening in an at-risk population, albeit with high frequencies of obesity and diabetes. We have demonstrated that our FAP assay is robust and activity can be measured via a robust and simple one-step enzyme assay. We used validated methodology to arrive at the FAP Index and have performed comprehensive statistical analysis to examine the influence of using different NITs alone or in combination. The study limitations were sample selection, sample size and non-centralised histologic fibrosis assessment. Comparison of NITs in the training cohort were only available in a subset of patients, however, the findings were replicated in the validation cohort.

5 Conclusion

We have developed and validated FAP Index; a simple and novel non-invasive test for triaging the risk of advanced fibrosis in metabolic associated fatty liver disease. FAP Index demonstrated excellent diagnostic accuracy with an AUROC of 0.875 as a standalone test. However, sequential use of FAP Index followed by FIB-4 yielded more favourable results with a reduction of indeterminately classified patients by one third compared to FIB-4 alone in both training and validation cohorts. This was achieved while maintaining an NPV of >95% and having the lowest percentage of false negative diagnoses of any NIT strategy examined. Moreover, FAP Index used sequentially with NFS compared to NFS alone generated even more striking results. FAP Index includes a direct marker of fibrogenesis (cFAP), which is simple and inexpensive to measure with potential application as a point of care or reflex test. Therefore, FAP Index demonstrates several advantages over the current first-line standard of care NITs for triaging risk of advanced fibrosis due to MAFLD and should be developed for clinical use.

FINANCIAL SUPPORT

This project was funded under Australian National Health and Medical Research Council project grant 1105238 (MDG, GWM, WWB), a Rebecca L Cooper Medical Research Foundation (MDG) grant, Centenary Institute Foundation (MDG) and a Gilead Fellowship (AM). JG is supported by: the Robert W. Storr Bequest to the Sydney Medical Foundation, University of Sydney; a National Health and Medical Research Council of Australia (NHMRC) Program Grant (APP1053206); Project, Ideas and Investigator grants APP2001692, APP1107178, APP1108422 and APP1196492; and a Cancer Institute NSW grant (2021/ATRG2028). LAA received research grant funding from the Hollywood Private Hospital Research Foundation and the Sir Charles Gairdner Osborne Parke Health Care Group.

AUTHOR CONTRIBUTIONS

ZVW: study concept and design, statistical analysis, analysis and interpretation of data, drafting the manuscript; BBB: acquisition of data, review and manuscript editing, study supervision; TP: study concept, statistical analysis; WWB, SEP, JHSL, PVDV, JG, LAA: Supply study materials, critical revision of the manuscript; GO, ME: acquisition of study samples; KB: statistical analysis and interpretation of data; HEZ: study concept and design, statistical analysis; GWM: acquisition of study samples, obtained funding; AM: study concept and design, critical revision of the manuscript for important intellectual content, study supervision; MDG: study concept and design, obtained funding, critical revision of the manuscript for important intellectual content, study supervision.

Competing interests

All authors declared to have no conflict of interest to this project.

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

Patient consent for this publication, Consent was not required from patients.

Ethics approval

The approvals were HREC_X18-0241 in Sydney Local Health District, 2019/ETH02319 in Westmead Hospital, RGS01287 in Sir Charles Gairdner Hospital, and 195/15 in The Alfred Hospital. All participants provided written informed consent and their serum samples and information were handled in accordance with ethics approvals and attendant regulations of each hospital.

Supplemental material

This content has been supplied by the authors.