文献 Low-Grade Dysplastic Nodules Revealed as the Tipping Point during Multistep Hepatocarcinogenesis by Dynamic Network Biomarkers.p

G C A TT A C GG C A T genesArticleLow-Grade Dysplastic Nodules Revealed as theTipping Point during Multistep Hepatocarcinogenesisby Dynamic Network BiomarkersLina Lu 1, Zhonglin Jiang 1, Yulin Dai 2 and Luonan Chen 1,3,*123*Key Laboratory of Systems Biology, CAS Center for Excellence in Molecular Cell Science, Innovation Centerfor Cell signaling Network, Institute of Biochemistry and Cell Biology, Shanghai Institutes for BiologicalSciences, Chinese Academy of Sciences, Shanghai 200031, China; (L.L.); (Z.J.)Center for Precision Health, School of Biomedical Informatics, The University of Texas Health Science Centerat Houston, 7000 Fannin St., Suite 820, Houston, TX 77030, USA; School of Life Science and Technology, Shanghai Tech University, Shanghai 201210, ChinaCorrespondence: These authors contributed equally to this work.Received: 29 July 2017; Accepted: 1 October 2017; Published: 13 October 2017Abstract: Hepatocellular carcinoma (HCC) is a complex disease with a multi-step carcinogenicprocess from preneoplastic lesions, including cirrhosis, low-grade dysplastic nodules (LGDNs),and high-grade dysplastic nodules (HGDNs) to HCC. There is only an elemental understandingof its molecular pathogenesis, for which a key problem is to identify when and how the criticaltransition happens during the HCC initiation period at a molecular level. In this work, for the rsttime, we revealed that LGDNs is the tipping point (i.e., pre-HCC state rather than HCC state) ofhepatocarcinogenesis based on a series of gene expression proles by a new mathematical modeltermed dynamic network biomarkers (DNB)a group of dominant genes or molecules for thetransition. Different from the conventional biomarkers based on the differential expressions of theobserved genes (or molecules) for diagnosing a disease state, the DNB model exploits collectiveuctuations and correlations of the observed genes, thereby predicting the imminent disease stateor diagnosing the critical state. Our results show that DNB composed of 59 genes signals thetipping point of HCC (i.e., LGDNs). On the other hand, there are a large number of differentiallyexpressed genes between cirrhosis and HGDNs, which highlighted the stark differences or drasticchanges before and after the tipping point or LGDNs, implying the 59 DNB members serving asthe early-warning signals of the upcoming drastic deterioration for HCC. We further identied thebiological pathways responsible for this transition, such as the type I interferon signaling pathway,Janus kinasesignal transducers and activators of transcription (JAKSTAT) signaling pathway,transforming growth factor (TGF)- signaling pathway, retinoic acid-inducible gene I (RIG-I)-likereceptor signaling pathway, cell adhesion molecules, and cell cycle. In particular, pathways relatedto immune system reactions and cell adhesion were downregulated, and pathways related to cellgrowth and death were upregulated. Furthermore, DNB was validated as an effective predictor ofprognosis for HCV-induced HCC patients by survival analysis on independent data, suggestinga potential clinical application of DNB. This work provides biological insights into the dynamicregulations of the critical transitions during multistep hepatocarcinogenesis.Keywords: dynamic network biomarkers; hepatocellular carcinoma; low-grade dysplastic nodules;tipping pointGenes 2017, 8, 268; doi:10.3390/genes8100268 /journal/genesGenes 2017, 8, 268 2 of 161. IntroductionHepatocellular carcinoma (HCC) is the sixth most common cancer worldwide and the thirdleading cause of cancer-related deaths around the world 1. Hepatocellular carcinoma is clinicallycharacterized by a high incidence rate and very poor prognosis 2. Currently, it is generally acceptedthat persistent hepatitis B virus (HBV) or hepatitis C virus (HCV) infections is the primary cause ofchronic liver disease leading to HCC 3. Hepatitis C virus infection is the main risk factor in westerncountries and Japan. Despite the progress made in numerous treatments, the survival rate of HCCpatients remains low because HCC is not easily detected prior to the advanced stage. Thus, it is ofutmost importance to clinically diagnose early HCC.In recent years, the concept of multi-step human hepatocarcinogenesis has been well documented 46.The liver injury induced by HCV produces a progressive inammatory milieu that results in a cycleof necrosis and regeneration leading to liver cirrhosis. Subsequently, cirrhosis patients often presentdysplastic nodules. These lesions which are conrmed as precancerous lesions of HCC are classiedas low-grade dysplastic nodules (LGDNs) and high-grade dysplastic nodules (HGDNs) based onpresence of cytologic and architectural atypia 7. Although the morphology of these nodules is notsufcient to support a diagnosis of malignant tumor, these nodules are closely correlated with theoccurrence of HCC. And, the HGDNs are more likely transformed into HCC than LGDNs based onclinical, pathological, molecular genetics, and radiological assessments 811. The sequence of HCCinitiation and progression is shown in Figure 1A, but the precise molecular events and their regulatorynetworks that underlie HCC formation remain largely unknown.Recently, a novel, model-free approach based on nonlinear dynamic theory, termed dynamicnetwork biomarkers (DNB), was developed to detect critical transitions or tipping points during theprogression of complex diseases 12,13. Generally, a disease progression can be divided into threestages, i.e., normal state, critical state (or the tipping point), and disease state (Figure 2A). After thetipping point moves gradually from the normal state, the system drastically deteriorates to a diseasestate. Specically, DNB is a group of molecules (i.e., genes, RNAs, proteins, or metabolites) withstrongly collective uctuations. Based on nonlinear dynamical theory, DNB appears only at the tippingpoint of a homeostatic system and the molecules in DNB are strongly correlated and also fluctuated justbefore the critical transition (i.e., the tipping point or pre-disease state). Quantitative criteria for the DNBcan be obtained by measuring the differential correlations and deviations of molecular expressionsrather than the differential expressions adopted in the traditional methods. In contrast to the “diseasediagnosis” by traditional biomarkers, DNB is for “disease prediction” (i.e., for the pre-disease diagnosisas the early-warning signals of the disease state). If the state of a system passes over the tipping pointto the disease state, it becomes very difcult to reverse to the normal state even by advanced medicaltreatment. Therefore, it is crucial to identify the pre-disease state so as to prevent the irreversibledeterioration of the disease. In addition to complex diseases, DNB theory had also been applied todetect the tipping points in cell fate decisions and immune checkpoint blockade processes 14,15.Given the difculty to diagnose early HCC, it is a key problem to identify when and how thetipping point or the critical transition happens at a molecular level. In this work, from stage-wise geneexpression proles of HCC initiation (i.e., normal, cirrhosis, LGDNs, HGDNs, and very early HCC),we identied the tipping point or pre-HCC state of HCV-induced HCC by DNB model. The obtainedDNB formed a specic module with 59 genes at the LGDNs stage to signal the tipping point just beforethe drastic deterioration in HCC progression. We also partially revealed molecular mechanism on theHCC initiation by functional analysis of DNB, which both provides biological insights into the dynamicregulations of the critical transitions and opens a new way for the identication of therapeutic targets.We further identied biological pathways responsible for the critical transition, including severalpathways in immune system reactions, cell growth and death, and cell adhesion. And furthermore,DNB was validated as an effective predictor of prognosis for HCC patients by survival analysis onindependent data.Genes 2017, 8, 268 3 of 16 Figure 1. The progression of hepatitis C virus (HCV)-inducedinduced hepatocellular carcinoma (HCC) and gene expression proling. (A) A schematic diagram shows the HCV-induced HCC development. (B) Agene expression profiling. (A) A schematic diagram shows the HCVinduced HCC development. (B) three-dimentional image shows principal component analysis (PCA) for clustering 48 samples alongA threedimentional image shows principal component analysis (PCA) for clustering 48 samples HCC development. Each small spot represents the principal component (PC) score along the top threealong HCC development. Each small spot represents the principal component (PC) score along the principle components for each sample. (C) Unsupervised hierarchical clustering of 48 tissue samplestop three principle components for each sample. (C) Unsupervised hierarchical clustering of 48 tissue using the Pearson correlation coefcient (PCC) distance. Similar to (B), low-grade dysplastic nodulessamples using the Pearson correlation coefficient (PCC) distance. Similar to (B), lowgrade dysplastic (LGDNs) samples are not grouped together but dispersed into cirrhosis and high-grade dysplasticnodules (LGDNs) samples are not grouped together but dispersed into cirrhosis and highgrade nodules (HGDNs) groups. n: Normal; ci: Cirrhosis; ld: Low-grade dysplastic nodules; hd: High-gradedysplastic nodules (HGDNs) groups. n: Normal; ci: Cirrhosis; ld: Lowgrade dysplastic nodules; hd: dysplastic nodules; ve: Very early HCC. The colored bars mark clusters: black, normal; red, cirrhosis;Highgrade dysplastic nodules; ve: Very early HCC. The colored bars mark clusters: black, normal; dark blue, HGDNs; light blue, Very early HCC. Dispersed LGDNs samples are highlighted in green.red, cirrhosis; dark blue, HGDNs; light blue, Very early HCC. Dispersed LGDNs samples are highlighted in green. normalized expression values processed by marker aided selection (MAS) method were downloaded. The dataset includes the expression proles of 75 tissue samples (10 normal liver tissues, 13 cirrhosis liver tissues, 10 LGDNs, 7 HGDNs, 8 very early HCC, 10 early HCC, 7 advanced HCC and 10 very advanced HCC) from 48 patients with HCV infection representing the stepwise carcinogenic process from preneoplastic lesions to HCC. The earlier ve stages were selected to study pathogenesis of HCC. Genes 2017, 8, 268Genes 2017, 8, 268 4 of 164 of 16 Figure 2. A brief mathematical model of dynamic network biomarkers (DNB) theory and its analysis results. (A) Three stages during HCC progression. A normal state is a relatively healthy stage in which the disease is under control, whereas the pre-HCCHCC state or the critical state at the tipping point is the limit of the normal state just before the transition of the disease. After the tipping point, the system drastically deteriorates to the disease state. The DNB method can identify the pre-HCCHCC state at the tipping point by using the signals shown in (B). (B) Dynamic network biomarkers as a network signals the emergenceof ofthe thecritical criticaltransition. transition.When Whenthe system the systemapproaches approachesthe pre-HCC the prestate,HCCDNB state,members DNB satisfymembersthe satisfythree conditions. the three Theconditions.expression Theof expressionDNB members of DNBbecome membersstrongly becomeuctuate strongly(high standardfluctuate deviations),(high standard thesedeviations),DNB members and theseare highly DNB correlated,members meanwhile,are highly thecorrelated,correlations meanwhile,between DNB the memberscorrelationsand betweenother non-DNB DNB members anddecrease. other nonHere,DNBedge memberswidth corresponds decrease. Here,to the edgecorrelation width betweencorrespondsa pair toof thenodes, correlationand node betweencolor correspondsa pair of nodes,to the andstandard node colordeviation correspondsof a node. to the(C)Scoresstandardof candidatedeviation ofDNBs a node.in every (C) Scoresstage. ofThe candidatescore in theDNBsLGDNs in everystage stage.is obviously The scorehigher in thethan LGDNsother stages,is therefore,obviouslythe highermolecule than moduleother stages,in LGDNs therefore,is considered the moleculeas the moduleDNB and inthe LGDNs is consideredcorrespond asto the tippingDNB andpoint the duringLGDNsthe correspondHCC progression. to the tipping point during the HCC progression. 2. MaterialsGiven theand difficultyMethods to diagnose early HCC, it is a key problem to identify when and how the tipping point or the critical transition happens at a molecular level. In this work, from stagewise 2.1.geneGene expressionExpression profilesDatasets of HCC initiation (i.e., normal, cirrhosis, LGDNs, HGDNs, and very early HCC),The wegene identifiedexpression the tippingproles pointfor orDNB preHCCanalysis statewere of HCVobtainedinducedfrom HCCthe byGene DNB Expressionmodel. The Omnibusobtained DNBdatabase formed(GEO, a specific/geo/ module with 59 genes at the LGDNs) under stageaccession to signalID GSE6764. the tippingAnd pointthe just before the drastic deterioration in HCC progression. We also partially revealed molecularmechanism on the HCC initiation by functional analysis of DNB, which both provides biologicalinsights into the dynamic regulations of the critical transitions and opens a new way for theidentification of therapeutic targets. We further identified biological pathways responsible for thecritical transition, including several pathways in immune system reactions, cell growth and death,Inandthis celldataset, be Andsets furthermore,without corresponding DNB was validatedgene symbols as an wereeffectiveexcluded predictorduring of prognosisour analysis, for HCC patients by survival analysis on independent data. while multiple probe sets mapping to the same gene were averaged as the expression values. And forthe probe sets mapping to more than one gene, we took the rst annotation.2. Materials and Methods As an independent dataset for validation, a cohort of HCC patients was subjected to survivalanalysis. The dataset was deposited in International Cancer Genome Consortium (ICGC) database2.1. Gene Expression Datasets (/) provided by RIKEN (project code: LIRI-JP) including 260 donors 16. Both the geneexpressionThe genepro lesexpression(FPKM data)profilesand forclinical-pathological DNB analysis wereinformation obtained werefromdownloaded. the Gene ExpressionThe data fromOmnibuspatients databasewithout (GEO,HCV /geo/)infection were not considered in this understudy.accession ID GSE6764. And the normalized expression values processed by marker aided selection (MAS) method were downloaded. 2.2.The Identicationdataset includesof Dynamic the expressionNetwork profilesBiomarkers of 75 tissue samples (10 normal liver tissues, 13 cirrhosis liverIt tissues,has been 10 indicatedLGDNs, 7that HGDNs,the progression 8 very earlyof manyHCC, chronical10 early HCC,diseases 7 advanced(e.g., cancer) HCCis andnot 10always very smoothadvancedbut HCC)there fromis an abrupt48 patientschange withafter HCVa system infectionstate representingpasses over thea critical stepwisestate carcinogenicpre-disease processstate, from preneoplastic lesions to HCC. The earlier five stages were selected to study pathogenesis of HCC. In this dataset, probe sets without corresponding gene symbols were excluded during our Genes 2017, 8, 268 5 of 16resulting in the drastic transition or serious deterioration to a disease state. However, in contrast tothe signicant difference between normal state and disease state in terms of molecular concentrations(or differential expressions of proteins or genes), there is generally no signicant difference betweennormal state and pre-disease state. Hence, traditional molecular biomarkers or methods may failto diagnose the pre-disease state. To overcome this problem, based on nonlinear dynamical theory,the DNB method was proposed to detect the pre-disease state or critical state by exploring uctuationinformation of the measured omics data, rather than the information of traditional differentialexpressions 12. In brief, DNB is a group of molecules satisfying the following three requirementswhen the system approaches the critical state:Condition 1: The DNB members are closely correlated to each other, i.e., their average Pearsoncorrelation coefcient (PCCin) in an absolute value becomes very high.Condition 2: The DNB members lose correlations with other non-DNB members, i.e., the averagePCC (PCCout) between DNB members and non-DNB members becomes very low.Condition 3: The DNB members are highly uctuated, i.e., their average standard deviation(SDin) becomes very high.Based on nonlinear dynamical theory, whenever DNB satisfying all above three criteria appears,the system is at the tipping point. Therefore, the three conditions are considered as the generic propertiesto detect early-warning signals of the pre-disease state or critical state. Note that DNB is a functionalmodule, which signals the imminent transition or deterioration from the normal state to the diseasestate, and, therefore, is considered to be causally related to the initiation and progression of thedisease 12,13,17,18.The three conditions can be combined into a single composite index (CI) to quantitatively detectthe DNB as follows: SD PCCCI = in inPCCout(1)where SDin and PCCin are the average standard deviation and average absolute PCC of all moleculesin DNB, corresponding to Conditions 3 and 1, respectively. The PCCout is the average absolute PCCbetween molecules inside and outside of DNB, corresponding to Condition 2.Assume that we collect molecular proles of several samples (e.g., gene expression data) in eachstage t during the disease progression proc