Introduction

Hepatocellular carcinoma (HCC) is one of the most common causes of cancer-related death. So, development of novel drugs in HCC has become a challenge for biologists. Alpha-fetoprotein (AFP) computational secreted network construction and analysis of HCC is very useful to identify novel markers and potential targets for prognosis and therapy.

AFP (fold change = 6.065) is one out of 50 genes identified as significantly different in HCC vs no-tumor hepatitis/cirrhotic liver tissues. AFP is relevant to molecular function of transfer/carrier protein, and it is relevant to biological process of transport, developmental processes, mesoderm development, and oncogenesis (database for annotation, visualization, and integrated discovery (DAVID)). AFP’s relational study can also be seen in these papers [1–10]. However, the molecular mechanism concerning AFP secreted network construction in HCC has received little attention.

In this paper, we first identified the significant molecule AFP by significant analysis of microarrays (SAM), then constructed AFP up- and downstream network by gene regulatory network infer (GRNInfer), and further data-mined the AFP secreted cluster from 25 no-tumor hepatitis/cirrhotic liver tissues from HCC patients and 25 HCC in the same GEO Dataset GSE10140-10141 by using the DAVID. We gained the negative result of AFP secreted module through the net numbers of activation minus inhibition compared with no-tumor hepatitis/cirrhotic liver tissues and predicted possibly the decrease of AFP secreted module in HCC. Our integrative result showed that AFP secreted cluster of HCC contained both in human no-tumor hepatitis/cirrhotic liver tissues and HCC terms of secreted, extracellular region, extracellular region part, extracellular space, signal peptide, signal, disulfide bond, glycosylation site: N-linked (GlcNAc...) and glycoprotein (with AFP gene), only in HCC term of disease mutation (without AFP gene), and none in HCC terms of cell surface receptor linked signal transduction, neuroactive ligand–receptor interaction, cell–cell signaling, and pancreas (without AFP gene) compared with human no-tumor hepatitis/cirrhotic liver tissues; all the conditions are vital to the invasion of HCC. Therefore, we deduced the weaker AFP secreted function in HCC consistent with our above computation. We predicted AFP high expression localization within cells of HCC and without secretion to extracellular matrix. It would be necessary of AFP secreted function to decrease the invasion of HCC. AFP secreted interaction module construction in HCC can be a new route for studying the pathogenesis of HCC. Our construction of AFP secreted network may be useful to identify novel markers and potential targets for prognosis and therapy of HCC.

Materials and methods

Microarray data

We used microarrays containing 6,144 genes from 25 no-tumor hepatitis/cirrhotic liver tissues from HCC patients and 25 HCC in the same GEO Dataset GSE10140-10141 [11].

Gene selection algorithms

Fifty potential HCC molecular markers were identified using SAM. SAM is a statistical technique for finding significant genes in a set of microarray experiments. The input to SAM is gene expression measurements from a set of microarray experiments, as well as a response variable from each experiment. The response variable may be a grouping like untreated, treated (either unpaired or paired), and so on. SAM computes a statistic d i for each gene i, measuring the strength of the relationship between gene expression and the response variable. It uses repeated permutations of the data to determine if the expression of any genes is significantly related to the response. The cutoff for significance is determined by a tuning parameter delta, chosen by the user based on the false-positive rate. We normalized data by log2 and selected minimum fold change = 2.78. Here, we chose the 50 top-fold significant (a big difference compared with no-tumor hepatitis/cirrhotic liver tissues) genes under the false-discovery rate, and q values are 0%. The q value (invented by John Storey [12]) for each gene is the lowest false-discovery rate at which that gene is called significant. It is like the well-known p value, but adapted to multiple-testing situations.

Network establishment of candidate genes

The entire network was constructed using GRNInfer [13] and GVedit tools. GRNInfer is a novel mathematic method called gene network reconstruction tool based on linear programming and a decomposition procedure for inferring gene networks. The method theoretically ensures the derivation of the most consistent network structure with respect to all of the datasets, thereby not only significantly alleviating the problem of data scarcity but also remarkably improving the reconstruction reliability. Equation 1 represents all of the possible networks for the same dataset.

$$ J = \left( {X\prime - A} \right)U{\Lambda^{ - 1}}{V^T} + Y{V^T} = \mathop {J}\limits^\wedge + Y{V^T} $$
(1)

We established a network based on the 50 top-fold distinguished genes and selected parameters as lambda 0.0 because we used one dataset, threshold 0.000001. Lambda is a positive parameter, which balances the matching and sparsity terms in the objective function. Using different thresholds, we can predict various networks with different edge density.

Functional annotation clustering

The DAVID Gene Functional Clustering Tool provides typical batch annotation and gene-GO term enrichment analysis for highly throughput genes by classifying them into gene groups based on their annotation term co-occurrence [14, 15]. DAVID uses a novel algorithm to measure relationships among the annotation terms based on the degrees of their co-association genes to group similar annotation contents from the same or different resources into annotation groups. The grouping algorithm is based on the hypothesis that similar annotations should have similar gene members. The functional annotation clustering integrates the same techniques of kappa statistics to measure the degree of the common genes between two annotations, and fuzzy heuristic clustering to classify the groups of similar annotations according to kappa values. The tool also allows observation of the internal relationships of the clustered terms by comparing to the typical linear, redundant term report, over which similar annotation terms may be distributed among many other terms.

Results

Identification of HCC molecular markers

AFP is one out of 50 genes identified as high expression in HCC vs no-tumor hepatitis/cirrhotic liver tissues. The 50 significant genes included serine peptidase inhibitor Kazal type 1 (SPINK1); alpha-fetoprotein (AFP); P antigen family member 4 (PAGE4); v-myb myeloblastosis viral oncogene homolog-like 2 (MYBL2); cyclin-dependent kinase inhibitor 3 (CDKN3); thyroid stimulating hormone beta (TSHB); dickkopf homolog 1 (DKK1); ribonucleotide reductase M2 polypeptide (RRM2); S100 calcium binding protein P (S100P); actinin alpha 2 (ACTN2); trophinin associated protein (TROAP); ELAV-like 3 (ELAVL3); discs large homolog 7 (DLG7); NIMA-related kinase 2 (NEK2); lipocalin 2 (LCN2); histone cluster 1 H3 (HIST1H3H); ninjurin 2 (NINJ2); protease serine 1 (PRSS1); nucleolar and spindle associated protein 1 (NUSAP1); insulin-like growth factor 2 mRNA binding protein 3 (IGF2BP3); sema domain immunoglobulin domain short basic domain secreted 3B (SEMA3B); forkhead box M1 (FOXM1); regenerating islet-derived 3 alpha (REG3A); FLJ33790; secreted frizzled-related protein 4 (SFRP4); baculoviral IAP repeat-containing 5 (BIRC5); solute carrier family 17 member 7 (SLC17A7); breast cancer 1 (BRCA1); topoisomerase II alpha (TOP2A); nuclear receptor subfamily 5 group A member 1 (NR5A1); X-ray repair complementing defective repair in Chinese hamster cells 2 (XRCC2); E2F transcription factor 1 (E2F1); desmuslin (DMN); roundabout axon guidance receptor homolog 1 (ROBO1); aldehyde dehydrogenase 3 family memberA1 (ALDH3A1); non-SMC condensin I complex subunit H (NCAPH); budding uninhibited by benzimidazoles 1 homolog beta (BUB1B); homeobox A5 (HOXA5); pituitary tumor-transforming 1 (PTTG1); glycerophosphodiester phosphodiesterase domain containing 5 (GDPD5); thymidine kinase 1 (TK1); amelogenin Y-linked (AMELY); protein kinase cGMP-dependent type II (PRKG2); ubiquitin-conjugating enzyme E2C (UBE2C); sulfotransferase family cytosolic 1C member 2 (SULT1C2); glutamate receptor metabotropic 1 (GRM1); sex comb on midleg-like 2 (SCML2); TTK protein kinase (TTK); cyclin E2 (CCNE2); cell division cycle 20 homolog (CDC20).

AFP up- and downstream secreted network construction in human no-tumor hepatitis/cirrhotic liver tissues and HCC

In the no-tumor hepatitis/cirrhotic liver tissues, AFP upstream secreted network appeared that AMELY, LCN2, REG3A activate AFP, and DKK1, SFRP4, SPINK1 inhibit AFP, as shown in Fig. 1a, whereas in HCC, AFP upstream secreted network showed that LCN2, REG3A, SFRP4 activate AFP, and AMELY, DKK1 inhibit AFP, as shown in Fig. 1b.

Fig. 1
figure 1

AFP up- and downstream secreted network construction in human no-tumor hepatitis/cirrhotic liver tissues by infer (a, c). AFP up- and downstream secreted network construction in HCC by infer (b, d). Arrowheads represent activation; cycle represents inhibition

Full size image

In the no-tumor hepatitis/cirrhotic liver tissues, AFP downstream secreted network reflected that AFP activates PRSS1, REG3A, TSHB and inhibits AMELY, DKK1, as shown in Fig. 1c, whereas in HCC, AFP downstream secreted network appeared that AFP activates AFP, AMELY and inhibits DKK1, LCN2, PRSS1, SEMA3B, SPINK1, as shown in Fig. 1d.

Identification of AFP up- and downstream secreted cluster in human no-tumor hepatitis/cirrhotic liver tissues and HCC by DAVID

AFP secreted cluster terms both in human no-tumor hepatitis/cirrhotic liver tissues and HCC include extracellular region, secreted, extracellular region part, extracellular space, signal peptide, signal, disulfide bond, glycosylation site: N-linked (GlcNAc...) and glycoprotein (with AFP gene). Other AFP secreted cluster terms only in human no-tumor hepatitis/cirrhotic liver tissues containing cell surface receptor linked signal transduction, neuroactive ligand–receptor interaction (without AFP gene), as shown in Fig. 2a, c. However, AFP secreted cluster terms only in HCC cover disease mutation (without AFP gene), as shown in Fig. 2b, d.

Fig. 2
figure 2

AFP up- and downstream secreted cluster in human no-tumor hepatitis/cirrhotic liver tissues by DAVID (a, c). AFP up- and downstream secreted cluster by DAVID in HCC (b, d). Gray color represents gene-term association positively reported; black color represents gene-term association not reported yet

Full size image

In the no-tumor hepatitis/cirrhotic liver tissues, AFP upstream modules mainly include secreted AMELY, LCN2, REG3A, DKK1, SFRP4, SPINK1, AFP, etc., as shown in Fig. 2a. In HCC, AFP upstream modules mainly cover secreted LCN2, REG3A, SFRP4, AMELY, DKK1, AFP, etc., as shown in Fig. 2b. In the no-tumor hepatitis/cirrhotic liver tissues, AFP downstream modules mainly consist of secreted PRSS1, REG3A, TSHB, AMELY, DKK1, AFP, etc., as shown in Fig. 2c. In HCC, AFP downstream modules mainly contain secreted AFP, AMELY, DKK1, LCN2, PRSS1, SEMA3B, SPINK1, AFP, etc., as shown in Fig. 2d.

Discussion

We have already done some work in this relative field about gene network construction and analysis seen in our published papers [16–19]. Here, we first indentified the 50 top-fold significant genes in HCC by SAM. Then, we compared AFP up- and downstream gene numbers of activation and inhibition between no-tumor hepatitis/cirrhotic liver tissues and HCC (Table 1). We identified AFP secreted module based on the integration of infer and DAVID as follows.

Table 1 Up- and downstream gene numbers of activation and inhibition of each module with AFP gene in AFP secreted cluster between human no-tumor hepatitis/cirrhotic liver tissues and HCC
Full size table

In AFP secreted module of upstream network of human no-tumor hepatitis/cirrhotic liver tissues, our integrative result showed that AMELY, LCN2, and REG3A activate AFP, and DKK1, SFRP4, SPINK1 inhibit AFP, whereas in that of HCC, LCN2, REG3A, and SFRP4 activate AFP, and AMELY and DKK1 inhibit AFP. In AFP secreted module of downstream network of human no-tumor hepatitis/cirrhotic liver tissues, our integrative result reflected that AFP activates PRSS1, REG3A, and TSHB and inhibits AMELY and DKK1, whereas in that of HCC, AFP activates AFP and AMELY and inhibits DKK1, LCN2, PRSS1, SEMA3B, and SPINK1 (Figs. 1, 2; Table 2). AMELY has been reported to have molecular function of extracellular matrix, extracellular matrix structural protein, miscellaneous function, and structural protein, and it is relevant to biological process of developmental processes, skeletal development, and mesoderm development (DAVID). AMELY’s relational study also can be seen in these papers [20–24]. SEMA3B’s molecular function contains signaling molecule and membrane-bound signaling molecule, and it is relevant to biological process of signal transduction, cell surface receptor mediated signal transduction, receptor protein tyrosine kinase signaling pathway, developmental processes, neurogenesis, ectoderm development, tumor suppressor, and oncogenesis (DAVID). SEMA3B’s relational study also can be seen in these papers [25–29]. SPINK1 is relevant to molecular function of select regulatory molecule, protease inhibitor and serine protease inhibitor, and biological process of protein metabolism and modification, proteolysis (DAVID). SPINK1’s relational study also can be seen in these papers [30–33]. PRSS1 has been proved to be concerned with molecular function of protease and serine protease, and biological process of protein metabolism and modification, and proteolysis (DAVID). PRSS1’s relational study also can be seen in these papers [34–37]. We gained the negative result of AFP secreted module through the net numbers of activation minus inhibition compared with no-tumor hepatitis/cirrhotic liver tissues and predicted possibly the decrease of AFP secreted module in HCC.

Table 2 Activation and inhibition gene names of AFP up- and downstream secreted cluster in human no-tumor hepatitis/cirrhotic liver tissues and HCC
Full size table

Otherwise, our integrative result showed that AFP secreted cluster of HCC contained both in human no-tumor hepatitis/cirrhotic liver tissues and HCC terms of secreted, extracellular region, extracellular region part, extracellular space, signal peptide, signal, disulfide bond, and glycosylation site: N-linked (GlcNAc...) and glycoprotein (with AFP gene), only in HCC term of disease mutation (without AFP gene), and none in HCC terms of cell surface receptor linked signal transduction, neuroactive ligand–receptor interaction, cell–cell signaling, and pancreas (without AFP gene) compared with human no-tumor hepatitis/cirrhotic liver tissues; all the conditions are vital to the invasion of HCC (Fig. 2). As we all know, relationships of annotation terms in one cluster have similar annotation contents and correlative functions. Koelink et al. indicated that a slow AFP decrease is an important predictor of liver cancer development in further life [38]. Ishigami et al. observed that AFP-positive gastric cancer has an aggressive behavior with hematogenous metastasis [39]. Therefore, we deduced the weaker AFP secreted function in HCC consistent with our above computation. We predicted AFP high expression localization within cells of HCC and without secretion to extracellular matrix. It would be necessary of AFP secreted function to decrease the invasion of HCC.

In conclusion, we predicted AFP high expression localization within cells of HCC and without secretion to extracellular matrix and possibly the decrease of AFP secreted module in HCC compared with no-tumor hepatitis/cirrhotic liver tissues. It would be necessary of AFP secreted module to decrease the invasion of HCC. AFP secreted interaction module construction in HCC can be a new route to study the pathogenesis of HCC.