In terms of public health, the management of the NASH epidemic is a priority. However, NASH is ... circulating miRNAs we
A NEW METHOD INCLUDING THE QUANTIFICATION OF CIRCULATING MIRNAS ALLOWS THE EFFICIENT IDENTIFICATION OF NASH PATIENTS AT RISK WHO SHOULD BE TREATED Arun Sanyal *1 , Genevieve Cordonnier2 , John Brozek2 , Alice Roudot2 , Sylvie Deledicque2 , Martin Barbazanges2 , Emilie Praca2 , 2 2 2 3 4 5 6 2 2 Fouad Ben Sudrik , Sophie Megnien , Remy Hanf , Bart Staels , Pierre Bedossa , Vlad Ratziu , , Dean Hum , Raphael Darteil 1
Virginia Commonwealth University, Richmond, United States, 2GENFIT, Loos, 3Université Lille 2, INSERM U1011, European Genomic Institute for Diabetes (EGID), Institut Pasteur de Lille, Lille, 4Department of Pathology, Hôpital Beaujon, University Paris-Denis Diderot, 5Université Pierre et Marie Curie, Hôpital Pitié Salpêtrière, 6Institute of Cardiometabolism and Nutrition (ICAN), INSERM, UMRS 938, Paris, France
FIRST BIOSTATISTICAL APPROACH THE MEDIAN ALGORITHM
Availability of a wide range of miRNA profiling technologies based on sequence specificity (Taq Man assays, NGS, miRNA arrays). High degree of specificity and sensitivity. Sequence homology among species facilitates the translation from preclinical to clinical studies.
miR-200a is implicated in the regulation of α-SMA activity and affects the proliferation of TGF-β-dependent activation of hepatic stellate cells (Sun et al., 2014)
miR-34a miR-34a is upregulated in the liver of NAFLD patients and reflects histological steatohepatitis severity (Cheung et al., 2008; Cermelli et al., 2011) miR-34a regulates steatosis by targeting the PPARα-related pathway in NAFLD (Ding et al., 2015) miR-34a promotes Hepatic Stellate Cell activation (Yan et al., 2015)
0.6
0.8
1 - Specificity
1.0
0.2
0.4
0.6
1 - Specificity
0.8
0.0 0.6
1 - Specificity
0.8
1.0
0.0
0.4
0.6
1 - Specificity
0.8
1.0
1.0
0.2
0.8 0.6 0.4
Sensitivity
0.6
Steatosis Grade AUC = 0.56 (95% CI: 0.48 - 0.63)
0.2
FLI AUC = 0.58 (95% CI: 0.51 - 0.66)
0.0 0.0
1.0
0.6 0.2
0.2 0.0
0.4
1.0
0.2
0.8 0.0
0.4
Sensitivity
0.8
0.8 0.6 0.4
Sensitivity
1.0 0.8
FIB-4 AUC = 0.72 (95% CI: 0.65 - 0.79)
0.2
0.0
0.4
1.0
1.0
1.0 0.8 0.6
0.6
0.0
0.0
1.0
0.2
0.4
0.6
1 - Specificity
0.8
1.0
0.0
0.2
0.4
0.6
1 - Specificity
0.8
1.0
(A) Overview of the biostatistical method used to identify and verify the Bootstrap algorithm. Selected variables: alpha-2-Macroglobulin (A2M), miR-34a, miR-200a, Glycated hemoglobin (HbA1c) and amino terminal type III procollagen peptide (P3NP). (B) Performances of the Bootstrap algorithm: median Area Under the Receiver Operating Characteristic (ROC) curve (AUC) of 0.82 (95% CI: 0.76 – 0.87), total accuracy of 75%, sensitivity of 75%, specificity of 76%, PPV of 72% and NPV of 79%.
1.0
25000
3A
CONCLUSION
3B
0.8
Clinical data from the GOLDEN-505 cohort
1.0
2 Biostatistical Approaches Bootstrap Algorithm 1.0
1 - Specificity
0.8
0.6 0.4
0.6
Sensitivity
0.4
0.6
Sensitivity
1.0
0.0
0.2
0.4
0.6
1 - Specificity
0.8
1.0
0.0
0.2
0.4
0.6
1 - Specificity
0.8
1.0
0.2
0.6
Bootstrap algorithm with miRNA Median algorithm with miRNA NAFLD Fibrosis Score ELF FibroTest Fibrometre S FLI Steatosis Grade FIB-4
0.2
0.4
0.0
0.2
Sensitivity
0.8
1.0
0.0
0.4
Median algorithm AUC = 0.82 (95%: 0.72 - 0.91)
0.8
Median Algorithm
95% of ≈27,000 Verifications
0.2
miR-200a expression is associated with the progression of liver fibrosis both in human and mouse studies (Murakami et al., 2011)
0.8
1 - Specificity
1.0
1C
Adapted from Baffy, 2015
miR-200a
0.6
(C) Comparison with existing scoring systems. The Bootstrap algorithm displays much better accuracy and robustness in identifying the NASH patient that should be treated as shown by the comparison of the AUC of each score.
0.8
miR-21 miR-34a miR-221/222 miR-122
0.8
0.2
(A) Overview of the biostatistical method used to identify the Median algorithm. (B) Frequency of apparition of each variable among the approximately 27,000 algorithms. Only the variables displaying a frequency over 50% were retained into the algorithm. Selected variables: alpha-2-Macroglobulin (A2M), miR-200a, miR-34a, Glycated hemoglobin (HbA1c) and amino terminal type III procollagen peptide (P3NP). (C) Performances of the Median algorithm: median Area Under the Receiver Operating Characteristic (ROC) curve (AUC) of 0.82 (95%: 0.72 – 0.91), total accuracy of 79%, sensitivity of 81%, specificity of 77%, PPV of 74% and NPV of 83%.
0.0
miRNAs are protected by the formation of protein-miRNA complexes.
miR-21 miR-122 miR-34a miR-221/222 3 miR-128-3p miR-200a
1 - Specificity
0.2
miRNAs are packaged in membranous vesicles such as exosomes or apoptotic bodies which offer protection against RNase activity.
miR-21 miR-34a miR-122 miR-451
0.6
0.4
RESULTS OVERVIEW
0.0
Circulating miRNAs are highly stable in serum and plasma samples :
miR-33a/b miR-34a miR-185 miR-199a-5p
20000
HCC
miRNAs are involved in the pathophysiology of chronic diseases, by impacting on gene regulation. miRNAs are released from cells upon tissue injury during progression of disease.
15000
0.4
0.0
Fibrosis/ Cirrhosis
10000
0.2
0.2
0.6
Steatosis Steatohepatitis
5000
0.4
Healthy liver
0.0
FibroTest AUC = 0.69 (95% CI: 0.62 - 0.76)
Frequency
Sensitivity
miRNAs are present in a wide range of biofluids: blood, urine, cerebrospinal fluid…
1.0
0.8 0.2
0
Several miRNAs have been identified play key roles the development steatosisfrequent and its For both biostatistical approaches, a stepwise logistic regression method was toapplied toinselect the ofmost progression to steatohepatitis, fibrosis, cirrhosis, and hepatocellular carcinoma (Figure 1). One of the variables included in the algorithms. most lipid-responsive miRNAs in the liver is miR-34a, which is heavily upregulated in mice kept on
CIRCULATING MIRNAS AS BIOMARKERS
0.6 0.0
Two independent biostatistical approaches (Median and Bootstrap) were used to generate thousands of cohorts from the initial patient population in order to assure the translatability of the results to the global NAFLD/NASH population. 4. Specific miRNAs Associated with the Progression of NAFLD
high-fat diet and its expression levels in humans correlate with the severity of NASH [31,32]. Overexpression of miR-34a results in hepatocellular apoptosis [32]. A major target of miR-34a is the NAD-dependent deacetylase Sirtuin-1 (SIRT1), which has a key role in energy homeostasis by activating pivotal transcription factors such as peroxisome proliferator-activated receptor-alpha (PPARα) and liver X receptor (LXR), while it has an inhibitory effect on PPAR-gamma coactivator-1alpha (PGC-1α), sterol regulatory element-binding protein 1c (SREBP1c), and farnesoid X receptor (FXR) [33]. Silencing of miR-34a restores the expression of SIRT1 and PPARα, resulting in activation of the metabolic sensor AMP-activated protein kinase (AMPK) and the activation of various PPARα target genes, suggesting a fundamental role of miR-34a in the deregulation of lipid metabolism associated with NAFLD [34].
Bootstrap algorithm AUC = 0.82 (95%: 0.76 - 0.87)
0.0
In addition to the 100+ variables that were measured at baseline, the systematic measurements of 9 different circulating miRNAs were also introduced into the data set.
0.8
Fibrometre S AUC = 0.71 (95%: 0.64 - 0.78)
0.0
0.2
Variables
Data and plasma samples from the 274 biopsy proven NASH patients included in the GOLDEN-505 phase IIb trial with Elafibranor were used for this study. This patient cohort is extremely well-characterized, including a complete anthropometric and biochemical data set, centralized biopsy reading, and covers a wide spectrum of NASH disease activity and severity.
Sensitivity
METHODS
0.6
1 - Specificity
0.8
50%
Constant A2M miR-200a miR-34a HbA1c P3NP
Sensitivity
1B
2B
1.0
Verif ied on 27 000 verif ication sets
3. Aberrant miRNA Profiles in Experimental and Human NAFLD
increased or decreased abundance associated with human NAFLD or experimental NAFLD induced by dietary or genetic manipulations [22,24–27]. By virtue of their ability to modulate multiple metabolic and signaling pathways, miRNAs appear to be involved in all stages of NAFLD. Thus, deregulation of miRNAs has been associated with altered lipid and glucose metabolism, oxidative stress, inflammation, and pathways of hepatocellular survival and proliferation [22,24,25,27]. A notable limitation of these studies is that while global miRNA sequencing is increasingly performed, most of currently available reports have been based on microarrays with a limited set of probes and cannot account for all miRNAs potentially involved in a given experimental or observational paradigm. Diet-induced obesity in mice results in the differential expression of 6% of total miRNAs [24,28]. High-fat diet administered to rats leads to differentially expressed miRNAs including upregulation of miR-146, miR-152, and miR-200 family members with predicted target genes regulating ion and protein transport, cell adhesion, and migration [29]. Importantly, these changes can be similarly observed in human hepatocytes and immortalized liver cell lines exposed to various fatty acids and pro-inflammatory cytokines [29]. Human observations provide additional evidence for aberrant miRNAs in obesity, insulin resistance, diabetes, and NAFLD [30]. In one of the earliest observations in humans, hepatic miRNA profiles of subjects with NASH and the metabolic syndrome by using a microarray of 474 human miRNAs were compared to healthy controls and 46 differentially expressed miRNA species were identified of which 23 were upregulated and 23 were downregulated [31]. Predicted targets of these miRNAs included genes that regulate lipid metabolism, inflammation, oxidative stress, and apoptosis. Intriguingly, however, individual histological features of NAFLD severity showed no correlations with changes in the expression level of these miRNAs [31].
0.4
0.4
1 Median algorithm using most f requent variables
0.2
ELF AUC = 0.69 (95% CI: 0.62 - 0.76)
0.2
Those patients at risk were defined according to FDA, EMA, and KOL’s consensus as having a NAFLD J. Clin. Med. 2015, 4, page–page Activity Score (NAS) of 4 or more and a fibrosis grade of 2 or 3.
NAFLD Fibrosis Score AUC = 0.68 (95% CI: 0.60 - 0.75)
0.0
27 000 algorithms (regression)
0.0
1.0
The first goal of this work was to develop a new simple, rapid, reliable and non-invasive diagnostic method, in order to screen and identify NASH patients that should be treated without requiring a liver biopsy.
The second objective was to compare the performances of this new method with those oftherapeutic existing MicroRNAs have recently emerged as novel biomarkers and potential targets in the management of NAFLD. Differential miRNA expression has identified a number of miRNAs with scoring systems.
Robustness on 1000 Bootstrap cohorts 95% CI
27 000 Verif ication sets (1/3)
0.0
27 000 Training sets (2/3)
0.4
AIM
0.2
27 000 Partitioning of GOLDEN-505 cohort
Bootstrap algorithm AUC = 0.82 (95%: 0.76 - 0.87)
Sensitivity
Bootstrap = Resampling by replacement
0.2
Optimal algorithm
0.0
The efficient management of the NASH patient population thus requires new non-invasive diagnostic tools dedicated to the screening and the detection of NASH patients at risk of liver outcomes from the large population of NAFLD/NASH patients.
106 simulations realized 106 partitions Analyses performed on all homogenous partitions (≈27,000)
Sensitivity
1000 bootstrap cohorts
0.4
Partitioning: Training set (2/3) and Verification set (1/3)
0.6
There is currently no approved medicine for this indication.
Sensitivity
0.8
NASH diagnosis necessitates an invasive procedure, the liver biopsy, that can only be performed by an experienced person,
Clinical data from the GOLDEN-505 cohort
0.4
Clinical data from the GOLDEN-505 cohort
2C
0.6
2A
0.4
1A
0.4
NASH is a silent, asymptomatic disease,
Sensitivity
In terms of public health, the management of the NASH epidemic is a priority. However, NASH is currently under-diagnosed since:
SECOND BIOSTATISTICAL APPROACH THE BOOTSTRAP ALGORITHM 1.0
BACKGROUND
Same variables A2M + miR-200a + miR-34a + HbA1c + P3PN
0.0
0.2
0.4
0.6
0.8
1.0
1 - Specificity
(A) The two independent biostatistical approaches indentified the same 5 variables including the same 2 miRNAs in their corresponding algorithm. (B) The Median and the Bootstrap algorithms showed better performances than existing scoring systems.
The two parallel biostatistical approaches, allowing the identification of the patients that should be treated, independently identified the same two specific miRNA species within the top 3 most frequent markers of NASH. The variables identified with the Bootstrap algorithm are the same as those identified with the Median algorithm. This observation strengthens the statistical relevance and confidence in these variables as predictive of NASH stages and severity. Additionally, a comparative study demonstrates that these 2 algorithms are more Powerful than existing scoring systems to identify the patient that should be treated. The results provide also the proof of the added value of miRNAs as diagnostic markers in NASH. Next steps: This work is still in progress, and the 2 algorithms will be improved through the implementation of new variables including recently discovered miRNA candidates. All the algorithms will be validated using independent cohorts.
REFERENCES Angulo Index - NAFLD fibrosis score: Angulo P, Hui JM, Marchesini G, Bugianesi E, George J, Farrell GC, Enders F, Saksena S, Burt AD, Bida JP, Lindor K, Sanderson SO, Lenzi M, Adams LA, Kench J, Therneau TM, Day CP (2007) The NAFLD fibrosis score: a noninvasive system that identifies liver fibrosis in patients with NAFLD. Hepatology 45: 846-854 | ELF: Guha IN, Parkes J, Roderick P, Chattopadhyay D, Cross R, Harris S, Kaye P, Burt AD, Ryder SD, Aithal GP, Day CP, Rosenberg WM (2008) Noninvasive markers of fibrosis in nonalcoholic fatty liver disease: Validating the European Liver Fibrosis Panel and exploring simple markers. Hepatology 47: 455-460 | Fibrotest: Ratziu V, Massard J, Charlotte F, Messous D, Imbert-Bismut F, Bonyhay L, Tahiri M, Munteanu M, Thabut D, Cadranel JF, Le Bail B, de Ledinghen V, Poynard T (2006) Diagnostic value of biochemical markers (FibroTest-FibroSURE) for the prediction of liver fibrosis in patients with non-alcoholic fatty liver disease. BMC Gastroenterol 6: 6 | Fibrometer S: Cales P, Laine F, Boursier J, Deugnier Y, Moal V, Oberti F, Hunault G, Rousselet MC, Hubert I, Laafi J, Ducluzeaux PH, Lunel F (2009) Comparison of blood tests for liver fibrosis specific or not to NAFLD. J Hepatol 50: 165-173 | FIB-4: Vallet-Pichard A, Mallet V, Nalpas B, Verkarre V, Nalpas A, Dhalluin-Venier V, Fontaine H, Pol S (2007) FIB-4: an inexpensive and accurate marker of fibrosis in HCV infection. Comparison with liver biopsy and fibrotest. Hepatology 46: 32-36. | FLI - Fatty Liver Index: Bedogni G, Bellentani S, Miglioli L, Masutti F, Passalacqua M, Castiglione A, Tiribelli C (2006) The Fatty Liver Index: a simple and accurate predictor of hepatic steatosis in the general population. BMC Gastroenterol 6: 33 | Steatosis Grade or SteatoTest: Poynard T, Ratziu V, Naveau S, Thabut D, Charlotte F, Messous D, Capron D, Abella A, Massard J, Ngo Y, Munteanu M, Mercadier A, Manns M, Albrecht J (2005) The diagnostic value of biomarkers (SteatoTest) for the prediction of liver steatosis. Comp Hepatol 4: 10
