Hindawi Publishing Corporation BioMed Research International Volume 2015, Article ID 780357, 15 pages http://dx.doi.org/10.1155/2015/780357

Research Article Cofunctional Subpathways Were Regulated by Transcription Factor with Common Motif, Common Family, or Common Tissue Fei Su,1 Desi Shang,1 Yanjun Xu,1 Li Feng,1 Haixiu Yang,1 Baoquan Liu,2 Shengyang Su,3 Lina Chen,1 and Xia Li1 1

College of Bioinformatics Science and Technology, Harbin Medical University, Harbin 150081, China Department of Anatomy, Harbin Medical University, Harbin 150081, China 3 Department of Computer Science, University of Texas at Dallas, 800 W Campbell Road, Richardson, TX 75080, USA 2

Correspondence should be addressed to Lina Chen; [email protected] and Xia Li; [email protected] Received 26 July 2015; Revised 2 November 2015; Accepted 4 November 2015 Academic Editor: Stelvio M. Bandiera Copyright © 2015 Fei Su et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Dissecting the characteristics of the transcription factor (TF) regulatory subpathway is helpful for understanding the TF underlying regulatory function in complex biological systems. To gain insight into the influence of TFs on their regulatory subpathways, we constructed a global TF-subpathways network (TSN) to analyze systematically the regulatory effect of common-motif, commonfamily, or common-tissue TFs on subpathways. We performed cluster analysis to show that the common-motif, common-family, or common-tissue TFs that regulated the same pathway classes tended to cluster together and contribute to the same biological function that led to disease initiation and progression. We analyzed the Jaccard coefficient to show that the functional consistency of subpathways regulated by the TF pairs with common motif, common family, or common tissue was significantly greater than the random TF pairs at the subpathway level, pathway level, and pathway class level. For example, HNF4A (hepatocyte nuclear factor 4, alpha) and NR1I3 (nuclear receptor subfamily 1, group I, member 3) were a pair of TFs with common motif, common family, and common tissue. They were involved in drug metabolism pathways and were liver-specific factors required for physiological transcription. In short, we inferred that the cofunctional subpathways were regulated by common-motif, common-family, or common-tissue TFs.

1. Introduction The mechanism of transcriptional regulation of coding genes is one of the basic contents in systems biology. Transcriptional factors (TFs) are proteins that regulate several target genes by binding DNA motifs at the transcriptional level [1–5]. Some investigators have reported that TFs take part in many important biological functions and human diseases, such as cell differentiation, proliferation, immune response, apoptosis, cardiac diseases, and tumor development [6–9]. Dissecting the characteristics of TFs is helpful for understanding their regulatory function in complex biological systems. An increasing number of studies have demonstrated that TFs with similar motifs often (which were defined as common-motif TFs in the paper) recognize target genes with similar expression

patterns [10], and in a similar way, TFs with different motifs may directly show different functions [11]. Several researchers have revealed that TFs in the common family (were defined as common-family TFs in the paper) have strong homology and the TFs are functionally related [12]; TFs in the transcription factor family are functional redundancies [13]; and TF family members share a relatively high degree of similarity in genomic structure and gene arrangement [12–14]. Several studies have demonstrated that TFs in the common tissue or cell line (were defined as common-tissue TFs in the paper) play critical roles in disease progression [15]. Currently, the regulatory functional interpretation of TFs mainly relies on activating or inhibiting their target genes [16, 17]. Several studies have revealed the functions of TFs based on the enrichment analysis of their target genes from

2

BioMed Research International

a number of pathways [18–22]. Subpathways (local regions within an entire biological pathway) exert certain functions on practical biological processes, disease development, and drug treatment [23–27]. Subpathway recognition is crucial because it is capable of focusing on local variation, increasing our power to identify the underlying causal genes, and may give us more detailed explanations of the biological process and pathogenesis of human disease. In this study, we dissected the regulatory influence of common-motif, common-family, or common-tissue TFs on cofunctional subpathways. Cofunctional subpathways are defined as subpathways that belong to the same pathway or pathway class. To gain insight into the relationships between TFs and their regulatory subpathways, we constructed a bipartite network TF-subpathways network (TSN) to explore the regulatory influence of TFs on subpathways. Through one-way cluster analysis, the common-motif, common-family, or common-tissue TFs that regulated the same pathway classes tended to cluster together, and through Jaccard coefficient analysis, the subpathways regulated by the TFs with common motif, common family, or common tissue were functionally consistent at the subpathway, pathway, and pathway class levels. These findings help us to understand the mechanisms of the regulatory effects of TFs on subpathways and provide a more detailed picture of biological processes and complex human diseases.

2. Materials and Methods 2.1. Preparation of the Data 2.1.1. TF Information. Transfac Professional (licensed on December 15, 2012) [28] provided details of the relations between TFs and their target genes, verified by biologists. We acquired 4598 transcription relations between 492 human TFs and 1557 target genes. We extracted 1061 human TF binding motifs. We extracted 80 unique TFs belonging to 29 families. We obtained 343 unique TFs expressed on 122 unique human tissues or cell lines. 2.1.2. Subpathway Data. We used R package of iSubpathwayMiner [24] to find the subpathways in public Kyoto Encyclopedia of Genes and Genomes (KEGG) (http://www.kegg .jp/) pathways database [29]. According to the pathway classification in KEGG, all subpathways were grouped into cellular processes (CP), environmental information processing (EIP), human diseases (HD), metabolism (Met), and organismal systems (OS). 2.2. Identifying the Subpathways Regulated by TFs. Cumulative hypergeometric enrichment analysis was used for identification of subpathways, in which TF target genes were significantly enriched. The 𝑝 value can be calculated to evaluate the enrichment significance for that subpathway as follows: 𝑟−1

( 𝑥𝑡 ) ( 𝑚−𝑡 𝑛−𝑥 ) . 𝑚 (𝑛) 𝑥=0

𝑝=1− ∑

(1)

We supposed that the whole human genome had 𝑚 genes, and among those, 𝑡 genes were included in the subpathway. The number of target genes of one TF is 𝑛, of which 𝑟 are involved in the same subpathways. 2.3. Evaluating the Consistent Function of Subpathways. We proposed a hypothesis that one TF pair regulated the consistent functional subpathways if they had common motif, common family, or common tissue. We applied the Jaccard coefficient to evaluate the consistent function of subpathways for TF pairs with a common motif, common family, or common tissue. The Jaccard coefficient is 𝐽 (𝑋, 𝑌) =

|𝑋 ∩ 𝑌| . |𝑋 ∪ 𝑌|

(2)

Assume TF1 and TF2 are a pair of common-motif, common-family, or common-tissue TFs, where 𝑋 represents the subpathways regulated by TF1 and 𝑌 represents the subpathways regulated by TF2. 𝐽 is the number of intersections of 𝑋 and 𝑌 divided by the number of unions of 𝑋 and 𝑌. The range of Jaccard coefficient, 𝐽, is from 0 to 1. If TF1 and TF2 annotated the same subpathway set, then 𝐽 = 1, and if TF1 and TF2 annotated entirely different subpathways, then 𝐽 = 0. The higher Jaccard value means stronger similarity, while lower values mean weaker similarity, so we can use the Jaccard coefficient to evaluate the consistent functionality of subpathways regulated by a pair of TFs with common motif, common family, or common tissue. At the pathways or pathway classes level, Jaccard coefficient is also used to evaluate the consistent functionality of subpathways, where 𝑋 represents the pathways or pathway classes that include the subpathways regulated by TF1 and 𝑌 represents the pathways or pathway classes that include the subpathways regulated by TF2.

3. Results 3.1. TF-Subpathway Network (TSN). We constructed a global TSN based on TFs and subpathway data (Figure 1) and systematically analyzed the characteristics of TFs that affected subpathways. We obtained all terms of TF binding sites (𝑛 = 21 038) information from the authoritative Transfac database. After filtering, we obtained 4598 specific human transcriptional regulatory relations between 492 TFs and 1557 target genes, verified by biologists. For each TF, we used the “𝑘-cliques” subpathway identification method provided by the iSubpathwayMiner software package to identify significantly enriched subpathways based on the TF-regulated target gene set, when 𝑘 = 3 and 𝑝 < 0.01. 𝑘 = 3 meant that the distance among the genes in one subpathway was not greater than 3 to ensure that genes in the subpathways have highly similar functions. We obtained 1490 significant TFsubpathway associations between 89 TFs and 468 subpathways (Table S1 in Supplementary Material available online at http://dx.doi.org/10.1155/2015/780357). We constructed a bipartite network consisting of TFs and subpathways in which a TF and a subpathway were connected if the TF target genes were significantly enriched in the corresponding subpathways (Figure 2).

BioMed Research International

3

Identify TF target genes TFi

TF1

TFi

TFm Filter redundant terms and species

TF1 TFi

···

Gene

Gene

···

Experiments

Integrate human TFs and human target genes

Gene

···

Gene

Keep 4598 pairs of human TF target genes associations TFi target gene set

TF1 target gene set TFi Subpathway 3

Subpathway 1 Gene

Subpathway 3 TFi

TF1

(a)

Subpathway p

Subpathway 2 Identify subpathways

Subpathway 16 Subpathway 1 TFm Subpathway 9

Subpathway 18

Combine all TFssubpathway associations:

TFi target gene set

Subpathway 1 Subpathway n

TFm target gene set

Subpathway j

Hypergeometric Subpathway 3 test

Subpathway p

“k-clique” subpathway identification algorithm:

(1) Convert pathways to the undirected graths with genes as nodes (2) Mine all 3-cliques as subpathways (c) Transfac professional data build 21038 TF binding sites

(b)

Pathway class

Pathway class CREM JUND ETV4 NRL ATF1 HMGA1 GATA2 GATA1 GATA3 STAT5A STAT4 NR2F1 THRB TCF7 JUNB MITF TFAP2C KLF5 NFATC1 CEBPA FOXA2 REL SREBF1 IRF5 SRF NFE2L2 GLI3 GLI1 STAT3 STAT6 NR1I3 NR1I2 GATA6 EGR1 ELF1 SMAD2 SMAD3 SPI1 FOXA3 HNF1A LEF1 NR5A1 NR4A1 NR5A2 GATA4 HNF4A POU2F1 IRF7 IRF3 IRF1 CEBPB FOS STAT1 RELA JUN ESR1 TP73 SMAD4 NFYA TP63 RFX1 XBP1 RFX2 RFX3 CREB1 PAX6 AR STAT5B NFIC VDR TFAP2A YBX1 E2F1 KLF4 MYC YY1 ATF2 SP3 TP53 SP1

Common-family TFs

Common-motif TFs

GATA2 GATA1 ATF1 HMGA1 STAT5A STAT4 THRB ETV4 CREM FOSL1 JUND RUNX3 GATA3 NR2F1 TCF7 JUNB MITF GLI1 TFAP2C NFATC1 CEBPA FOXA2 REL SREBF1 IRF5 SRF GLI3 NFE2L2 STAT3 STAT6 ESR1 TP73 SMAD4 ELF1 SPI1 SMAD2 SMAD3 ZBTB7B EGR1 LEF1 TCF7L2 AR STAT5B TP63 NFYA NR5A1 NR4A1 NR5A2 GATA4 HNF4A POU2F1 NR1I3 NR1I2 GATA6 FOXA3 HNF1A RFX2 RFX1 RFX3 XBP1 CREB1 IRF7 IRF3 IRF1 STAT1 FOS CEBPB RELA JUN PAX6 VDR NFIC RBPJ TP53 TFAP2A E2F1 ZBTB7A KLF4 MYC YY1 ATF2 SP3 SP1

CREB AP 1 Ets type AP 1 CREB Fungal regulators GATA factors GATA factors GATA factors STAT STAT Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) TCF 1 AP 1 Ubiquitous bHLH ZIP factors AP 2 Developmental/cell cycle regulators NF AT C/EBP like factors Tissue specific regulators Rel/ankyrin Ubiquitous bHLH ZIP factors Interferon regulating factors Responders to external signals AP 1 Developmental/cell cycle regulators Developmental/cell cycle regulators STAT STAT Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) GATA factors Developmental/cell cycle regulators Ets type SMAD SMAD Ets type Tissue specific regulators Homeo domain only TCF 1 Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) GATA factors Steroid hormone receptors (NR3) POU domain factors Interferon regulating factors Interferon regulating factors Interferon regulating factors C/EBP like factors AP 1 STAT Rel/ankyrin AP 1 Steroid hormone receptors (NR3) p53 like SMAD Heteromeric CCAAT factors p53 like Regulators with a DNA recognition wing AP 1 Regulators with a DNA recognition wing Regulators with a DNA recognition wing CREB Paired plus homeo domain Steroid hormone receptors (NR3) STAT NF 1 Steroid hormone receptors (NR3) AP 2 csd Cell cycle controlling factors Developmental/cell cycle regulators Cell cycle controlling factors Ubiquitous factors AP 1 Ubiquitous factors p53 like Ubiquitous factors

Common-tissue TFs

Pathway class T cells Breast T cells Breast HeLa HepG2 Hep3B T cells HepG2 Liver MCF7 T cells Testis Heart Brain Kidney Liver HUVEC HeLa HeLa K562 HepG2 MCF7 MCF7 K562 MCF7 Lymph node Kidney Heart Liver Liver Caco HUVEC Hep3B K562 T cells Thymus Spleen Liver Bone marrow Spleen T cells Testis Thymus Intestine Bone marrow Liver Lymph node Lymphocytes Spleen T cells Testis Thymus Thymus T cells Heart Breast Thymus Lymphocytes HeLa Caco Breast Intestine Kidney Liver Spleen Liver Liver Brain Caco Intestine Hep3B Kidney Liver Testis Heart Brain Kidney Liver Lymphocytes Spleen Testis Thymus Spleen Lymph node Thymus Lymph node Bone marrow Spleen Thymus K562 HUVEC MCF7 HepG2 MCF7 HeLa MCF7 HeLa Brain Bone marrow Intestine Lymphocytes MCF7 Testis HeLa HUVEC HeLa MCF7

TCF7 TCF7 CREM ETV4 NR2F1 GATA2 GATA2 GATA3 GATA3 FOXA2 TFAP2C NFATC1 KLF5 SREBF1 SREBF1 SREBF1 SREBF1 REL REL SRF NFE2L2 NFE2L2 STAT3 FOS FOS ESR1 TP73 NR1I3 NR1I3 NR1I3 NR1I2 GATA6 EGR1 EGR1 EGR1 EGR1 ELF1 ELF1 SPI1 SPI1 SPI1 SPI1 SPI1 SPI1 AKNA AKNA AKNA AKNA AKNA AKNA AKNA AKNA AKNA LEF1 LEF1 TP63 TP63 TP63 NFYA NFYA HNF1A HNF1A HNF1A HNF1A HNF1A HNF1A FOXA3 NR4A1 NR4A1 GATA4 HNF4A HNF4A HNF4A HNF4A HNF4A CREB1 CREB1 CREB1 CREB1 CREB1 CREB1 CREB1 CREB1 IRF7 IRF7 IRF7 IRF5 IRF5 IRF5 IRF5 RELA RELA E2F1 STAT5B VDR VDR NFIC NFIC TP53 TP53 TP53 TP53 TP53 TP53 TFAP2A SP1 SP1 SP1

(d) Clustering

Figure 1: Continued.

4

BioMed Research International 100

40

60 40 Observed

20

80 60 40 Observed

20 0

0 0.035

0.045

20 Observed 10

0.04 0.05 0.06 0.07 0.08 0.09 0.10

Subpathway

0.30

0.35 0.40 Pathway class

Pathway

Probability density

800 600 400 200

30

0

0.055

Observed

250

100

200

80

150 100 Observed

50

Probability density

0.025

Probability density

Probability density

80

Probability density

Probability density

100

0

0 0.00 0.02 0.04 0.06 0.08 0.10 0.12

0.45

60 40 Observed

20 0

0.00

0.05 0.10 Pathway

Subpathway

0.15

0.20

0.25

0.30

0.35

0.40

Pathway class

Probability density

Probability density

800 600 400 200

Observed

200 150 100 Observed

50 0

0 0.000

0.010

0.020

0.030

Probability density

150 1000

100

50 Observed 0

0.00

0.02

0.04

Subpathway

0.06

0.08

Pathway

0.0

0.1

0.2 0.3 Pathway class

(e) Jaccard coefficient

Subpathway level Pathway level

Pathway class level

Path: 00120_15 Path: 00120_19 Path: 03320_2 Path: 00120_6 Path: 04610_2 Path: 00120_7 Path: 00120_9 Path: 00590_1

Path: 04060 Path: 00590

Path: 04610_3 Path: 04610_4 Path: 04610_5

Path: 00590_2

Path: 00830_2 Path: 00982_4 Path: 00982_5

Path: 04610_6 Path: 00982_9 Path: 00590_3 Path: 04610_8 Path: 00590_4 Path: 04950_4 Path: 00591_1 Path: 03320_1 Path: 00982_10

Path: 00980_1

Path: 00591

Path: 00830

Path: 03320

Path: 00980

Path: 04610

Path: 00982

Path: 04950

Path: 00980_2

(f) Jaccard coefficient of HNF4A and NR1I3

Figure 1: Continued.

Organismal systems Human diseases

Metabolism

0.4

0.5

BioMed Research International

5 4-Glutathionyl-CPA (inactive) 4-Keto-CPA

Dechloroethyl-CPA Chloroacetaldehyde

2.5.1.18

CYP3A4

CYP3A4

CYP3A4 CYP2C CYP2B6

Ifosfamide (prodrug)

Nornitrogen mustard (active) Carboxyphosphamide

4-OH-CPA 4-Hydroxyifosfamide

Aldophosphamide

Phosphamide mustard (active)

Acrolein Aldoifosfamide

CYP3A4 CYP3A4

Alcophosphamide 1.1.1.1 1.2.1.5

CYP2C CYP2B6

Cyclophosphamide (prodrug)

Excretion

Isophosphamide mustard (active) Carboxyphosphamide

CYP3A5

CYP2B6 4-Ketoifosfamide

Alcophosphamide

Chloroacetaldehyde Dechloroethylifosfamide (g) Drug metabolism-cytochrome P450

Figure 1: Schematic of the construction of the global TSN based on TF target genes and pathway structure data from KEGG. (a) We extracted 21038 TFs that were verified through biology experiment from Transfac Professional database and then filtered redundant terms and species and obtained 4598 pairs of human TF target gene associations. (b) We used the 𝑘-clique subpathway method to identify significantly enriched subpathways to generate TF-subpathway associations. (c) We combined these TF-subpathway associations to form the TSN. (d) We used oneway clustering between TFs and subpathways to analyze the regulatory effects of common-motif, common-family, and common-tissue TFs on cofunctional subpathways. (e) Jaccard coefficient was applied to evaluate the functional consistency of the common-motif, common-family, and common-tissue TFs at the subpathway, pathway, and pathway class levels. (f) HNF4A and NR1I3 were a pair of TFs with common motif, common family, and common tissue. Jaccard coefficient at subpathways level, pathways level, and pathway classes level. (g) HNF4A regulated subpathway 00982-10, and NR1I3 regulated subpathway 00982-4, and they have shared genes (CYP3A4, CYP3A5, and CYP2C).

The TSN was composed of 557 nodes (468 subpathways and 89 TFs) and 1490 edges (Figure 2(a)). We calculated the size of the giant component of TSN (Figure 2(b)). The giant component was the largest connected component of a network and measured local functional clustering when compared to random networks. The TSN network connected most TFs and subpathways into a highly interlinked giant component, with strong local clustering of TFs and subpathways. We constructed 1000 random networks by randomly shuffling the relations between the 1490 pairs of TFs and subpathways, while keeping the number of each TF and subpathway unchanged. We compared the giant component of the real TSN and 1000 random networks, which showed that the actual size of the giant component of the TSN (521) was significantly smaller (experiential 𝑝 = 0) than the average giant component of randomized networks (554.963). We paid close attention to the TF degree distribution. As shown in Figure 2(c), a small number of TFs had higher connectivity, which meant that these TFs regulated many subpathways. For example, SP1 had the highest degree (186) (Table S2). This might be because it could activate or inhibit the expression of a number of essential oncogenes and tumor suppressors, and SP1 took part in the main biological processes such as cell growth, apoptosis, differentiation, immune response, response to DNA damage, and chromatin remodeling [30, 31]. Tumor suppressor gene TP53 had higher

degree (72) than other TFs, and it is always present in many types of human cancer. For example, TP53 gene mutations are more frequent in cervical adenocarcinoma [32]. In addition, we also calculated the degree distribution of the subpathways (Figure 2(d)). We evaluated which kind of subpathway was preferentially regulated by TFs. We grouped all the subpathways into the categories of Met, OS, CP, HD, and EIP, according to the pathway classification of KEGG. The average degrees of the five subpathway groups were 4.0000, 2.6026, 5.2982, 3.0372, and 1.9250 (𝑝 = 2.77 × 10−11 by one-way ANOVA test), respectively. Therefore, the cellular processes subpathways had the greatest opportunity of being regulated by TFs, such as the cell cycle subpathway (path: 04110 18) and TP53 signaling pathway (path: 04115 1). This finding was consistent with a previous study that showed that TF preferred to regulate cellular process genes [33]. 3.2. Common-Motif TFs Tend to Regulate Cofunctional Subpathways. The common conserved sequence of the TF binding sites is called a motif. Some studies have indicated that motif sequence is associated with TF function. TFs with similar motifs often recognize target genes with similar expression patterns [10], and TFs with different motifs may directly influence different TF function [11], and TFs influence pathways initiated by specific DNA-binding motifs

6

BioMed Research International Path: 00140_23 Path: 00140_13 Path: 00140_22 Path: 00140_9 Path: 00590_4 Path: 00140_10

Path: 00140_19 Path: 00140_12

Path: 00140_6

Path: 00140_4

Path: 00140_7 Path: 00140_5

Path: 00140_20

Path: 00140_18

Path: 04350_13 Path: 04630_4

Path: 04650_12

HNF1A

Path: 04670_14

Path: 05200_43 Path: 04115_1

Path: 04670_15

Path: 04145_6 Path: 04145_2

Path: 04115_7

Path: 05332_1 RFX2

Path: 05219_4

CREB1

Path: 05330_2

Path: 05320_4

Path: 05218_8 RFX3

Path: 05220_12

XBP1

Path: 05330_1

Path: 05200_47 Path: 04650_7 Path: 05220_11 Path: 04620_9 Path: 05214_14 Path: 04145_5 Path: 05218_5 Path: 05210_10 Path: 05218_7 Path: 04620_17 Path: 05218_6 Path: 04620_22 Path: 05214_15 Path: 05220_9

Path: 05210_9

RFX1

ETV4 NR5A1

Path: 04620_16 Path: 04620_18

Path: 05150_17

Path: 05310_3 Path: 05150_14

Path: 05200_29

Path: 04060_21

Path: 04670_10

Path: 05214_13

STAT4

CREM

SMAD3 SPI1

Path: 05310_1 Path: 05330_4

Path: 05200_52

NR5A2

GATA4

Path: 05330_7

Path: 05320_3

Path: 05219_3

JUND FOSL1 NR4A1

Path: 04060_3 Path: 04630_3

Path: 04650_10 Path: 05140_4

Path: 05200_51

STAT3

ENO1

Path: 04060_41

Path: 05322_6

STAT5A

STAT5B

Path: 05320_1 Path: 05320_6

Path: 05220_8

STAT6 STAT1 YY1

Path: 04514_55 Path: 04650_13 Path: 04514_62

Path: 00140_14

Path: 00140_3 Path: 00140_8 Path: 00140_24

Path: 00140_16

Path: 00120_5

Path: 05416_4

Path: 04514_3

Path: 05322_5

Path: 04514_1 Path: 04940_1 Path: 04612_3 Path: 04672_5 Path: 04672_4

Path: 04664_3 Path: 04650_1 Path: 04650_15

Path: 04664_9 Path: 04722_19 SMAD4 Path: 04060_47 Path: 04060_32 Path: 04512_24 Path: 04512_10 Path: 04512_11 Path: 04630_2 Path: 04010_21 Path: 04512_4 Path: 04010_5 Path: 04514_57 Path: 04512_18 Path: 04512_23 Path: 04514_50 Path: 04012_5 Path: 04144_2 Path: 04512_13 Path: 04514_49 Path: 04512_9 Path: 04010_26 Path: 04060_56 Path: 04060_62 Path: 04010_15 Path: 04912_8 Path: 04060_22 Path: 04512_3 Path: 04060_43 Path: 04110_7 Path: 04020_2 Path: 04010_16

Path: 04115_2 Path: 04115_3

Path: 04060_51 Path: 04060_38

Path: 04210_20

Path: 04620_11

Path: 04520_3

Path: 04620_12 IRF1

Path: 04510_8

IRF3

Path: 04510_6 IRF7

Path: 04650_6

IRF5

Path: 04620_1 Path: 04110_15

Path: 04620_2

Path: 04110_17

Path: 04010_31

Path: 04110_18

Path: 04610_8 NFATC1

PAX6

SP1

E2F1

TP53

TP63

TFAP2C

CEBPB Path: 04110_20

ZBTB7A

Path: 04110_27

Path: 05211_2 SRF Path: 05160_12 Path: 05020_9 Path: 05223_7 Path: 05220_5 Path: 05142_10 Path: 05410_2 Path: 05131_5 Path: 05142_11 Path: 05212_1 Path: 05214_4 Path: 05140_6 Path: 05414_3 Path: 05140_10 Path: 05146_5

Path: 04110_3

Path: 04110_26

Path: 04110_25 Path: 04110_23

Path: 05200_42 Path: 05222_2 Path: 05222_7

TCF7L2 Path: 05221_1 LEF1

Path: 05200_31

YBX1

Path: 05210_12

Path: 05215_4

Path: 04916_8 Path: 05210_7 Path: 04916_11

Path: 04514_10 Path: 04514_78 Path: 04514_9

Path: 04672_2

Path: 04510_11

Path: 04340_2

Path: 00100_9 GLI3

NR1I2

FOXA3 Path: 05140_1

Path: 04920_11

Path: 00980_1 Path: 00590_1

Path: 05142_18 Path: 05215_1 Path: 05145_14 Path: 05014_4 Path: 05215_3

ZBTB7B

Path: 05212_11 Path: 05212_14 Path: 05146_1 Path: 05214_11

Path: 04340_1 Path: 00980_2

Path: 05142_17 Path: 05200_53 Path: 05223_3 Path: 05100_10 Path: 05214_21 Path: 05200_12 Path: 05332_2 Path: 05214_10 Path: 05200_9 Path: 05144_3 Path: 05200_11 Path: 05200_19 Path: 05200_20 Path: 05215_11 Path: 05214_3 Path: 05200_4 Path: 05414_2 Path: 05100_3 Path: 05218_2 Path: 05214_8 Path: 05200_38 Path: 05145_21 Path: 05200_24 Path: 05223_5 Path: 05214_17 Path: 05211_8 Path: 05100_2 Path: 05332_5 Path: 05200_48 Path: 05320_9 Path: 05200_15 Path: 05214_6 Path: 05200_3 Path: 05145_19 Path: 05200_6 Path: 05142_20 Path: 05145_8 Path: 05212_12 Path: 05213_6 Path: 05213_2 Path: 05330_5 Path: 05214_5 Path: 05200_8 Path: 05200_16 Path: 05214_1 Path: 05100_15 Path: 05214_12 Path: 05215_7 Path: 05200_10 Path: 05211_7 Path: 05016_8 Path: 05142_4 Path: 05020_8 Path: 05218_3 Path: 05222_4 Path: 05200_21 Path: 05416_1 Path: 05120_6 Path: 05131_6 ··· Path: 05211_1 Path: 05219_2 Path: 05214_2 Path: 05200_54

Path: 05216_4

Path: 05213_1

NFIC

Path: 05131_12 Path: 05215_12 Path: 05142_8

Path: 04940_4 Path: 05218_4 Path: 05212_13 Path: 05146_2 Path: 05200_5

SMAD2 NFYA

Path: 04623_5

Path: 05131_9

Path: 04622_3 Path: 04664_5 Path: 04062_19 Path: 04920_1 Path: 04664_2

POU2F1 GATA6

Path: 05200_56

Path: 05222_1

Path: 05200_50

GATA1

Path: 05160_3

Path: 05142_21

Path: 04650_22

Path: 00240_2

Path: 00591_1

Path: 00330_1 Path: 00140_11 Path: 00120_1 Path: 00350_3 Path: 00350_1 Path: 00240_5 Path: 00350_2 Path: 00140_1 Path: 00120_2 Path: 00590_3

GATA2 Path: 05200_17 Path: 05014_3 Path: 05215_10 Path: 05146_9 Path: 05160_15

Path: 04110_4

ATF1

Path: 04660_1

Path: 00240_17

Path: 00330_9 MYC

FOS

Path: 05140_9

Path: 04612_7

Path: 04650_23

Path: 00120_3

ATF2

KLF5

Path: 04110_16

Path: 04110_19

Path: 04350_11

Path: 04210_3

Path: 04670_9

AR

Path: 04722_24

REL

Path: 05220_6

Path: 00240_16

Path: 04610_4

MITF RELA

Path: 05140_13

Path: 04062_1

EGR1

SP3

Path: 04620_10

RBPJ

HMGA1

TXK

ESR1

Path: 04210_23 Path: 04810_32 Path: 04510_23 Path: 04110_12 Path: 04510_17 Path: 04110_1

Path: 04350_6

Path: 05222_9

Path: 04670_20

Path: 04610_3 Path: 04660_8

Path: 04110_8 Path: 04510_24 Path: 04110_22 Path: 04110_11

Path: 04062_2 Path: 04670_2

Path: 05160_8

ELF1 Path: 04722_13

Path: 04060_8

Path: 04620_13

JUN

Path: 04620_25

Path: 05160_17

Path: 04650_20 Path: 04910_2

Path: 04310_7

Path: 04520_9 KLF4

Path: 04622_5

Path: 05222_10 Path: 04620_15

Path: 04060_33

Path: 04010_10 Path: 04512_12 Path: 04512_27 Path: 04512_7 Path: 04512_5

Path: 04520_2 Path: 04520_7

Path: 04520_5

Path: 04810_29

Path: 05215_2 Path: 05120_3

Path: 04512_19 Path: 04514_56 Path: 04662_6 Path: 04512_22 Path: 04080_7 Path: 04010_30 Path: 04916_3 Path: 04630_1 Path: 04512_2 Path: 04512_1 Path: 04512_20 Path: 04512_8 Path: 04621_7 Path: 04512_6 Path: 04610_7

Path: 04510_13 Path: 04540_13

Path: 04621_4

Path: 05145_5

Path: 04512_14

Path: 04210_24 Path: 04510_7

Path: 05310_4 Path: 05310_5 Path: 05200_18 Path: 05131_10

Path: 04662_1

Path: 04650_14

Path: 04310_6

Path: 05160_6

Path: 04620_14

Path: 04664_1

Path: 04664_10

TP73 Path: 04115_4

GLI1

Path: 05217_2

Path: 00590_2 Path: 05200_57 Path: 00830_2

Path: 05217_1 Path: 05200_34

Path: 04510_20 Path: 04510_19

Path: 04510_25

Path: 04510_18 Path: 04510_1 Path: 04510_16 TFAP2A Path: 04510_4

Path: 04510_15

Path: 05144_9

Path: 04514_12

Path: 04514_11

Path: 00982_7 Path: 00982_5 Path: 00982_10 Path: 04950_4

Path: 05320_2

Path: 00982_4

AKNA Path: 04060_31

Path: 04510_5

Path: 05330_12

Path: 00480_1

Path: 05223_6

Path: 04612_4

Path: 04744_2 RUNX3

Path: 00480_4

Path: 03320_2

GATA3 NRL

Path: 05330_3

Path: 05310_2

Path: 05215_8 Path: 04510_10 Path: 05214_16

Path: 00480_6

Path: 04610_6

Path: 00982_9 Path: 00982_11

Path: 05214_7

Path: 04910_5

Path: 00982_3 Path: 00982_8

Path: 05322_7

Path: 05145_20 Path: 05416_3 Path: 05320_5

Path: 04660_20

NFE2L2

Path: 04910_6

TCF7

Path: 00480_2

JUNB THRB Path: 00982_2

HNF4A

Path: 00982_6

VDR

Path: 04610_2

NR1I3

Path: 03320_1 CEBPA

Path: 04080_2

Path: 00480_3 Path: 04350_12 Path: 00480_5 Path: 04350_10

NR2F1

Path: 00140_26

Path: 04610_5 SREBF1

FOXA2

Path: 00140_25

Path: 04060_54

Path: 00120_19

Path: 04012_14

Path: 00120_15

Path: 04012_11

Path: 00120_6

Path: 04012_1 Path: 04012_12

Path: 00120_9

Path: 00120_7 Path: 04012_7 Path: 04012_4

Pathway class Cellular processes Human diseases Metabolism Organismal systems Environmental information processing

Node side TF 6–50

1–5

51–100 101–150 151–186

Subpathway 1

2–5

6–10 11–15 16–17 Edge color

0.01–0.001 0.001–0.0001

0.0001–0

Observed

The number of subpathways

0.30 0.25 0.20 0.15 0.10 0.05 0.00

The number of TFs

Probability density

(a)

15 10 5 0

500 520 540 560 580 600 Size of giant component (b)

5

10 15 20 25 30 34 · · · 200 TF (deg.) (c)

1 2 3 4 5 6 7 8 9 10 11 12 13 15 17 Subpathway (deg.) (d)

Figure 2: The TSN is a bipartite network. (a) The circles and rectangles in the network represent TFs and subpathways, respectively. A pair of TFs and a subpathway were connected by an edge if the set of target genes of TF was significantly enriched in the corresponding subpathway. Node size was proportional to the degree of the node. Edges were colored according to the enrichment significance (𝑝 < 0.01) of associations between TF target genes and subpathways. (b) Component size distributions of TFs and subpathways in the TSN and random network. The actual size of the real network giant component was smaller than that of the 1000 randomly shuffled TFs (521 versus 554.9630, experiential 𝑝 = 0). (c) Bar plot of degree of TF distribution in the TSN. (d) Bar plot of degree of distribution of subpathways in TSN.

BioMed Research International

7 Pathway class

Common-motif TFs

GATA2 GATA1 ATF1 HMGA STAT5 STAT4 THRB ETV4 CREM FOSL1 JUND RUNX3 GATA3 NR2F1 TCF7 JUNB MITF GLI1 TFAP2 NFATC CEBPA FOXA2 REL SREBF RF5 SRF GLI3 NFE2L STAT3 STAT6 ESR1 TP73 SMAD4 ELF1 SPI1 SMAD2 SMAD3 ZBTB7 EGR1 LEF1 TCF7L AR STAT5 TP63 NFYA NR5A1 NR4A1 NR5A2 GATA4 HNF4A POU2F NR1I3 NR1I2 GATA6 FOXA3 HNF1A RFX2 RFX1 RFX3 XBP1 CREB1 RF7 RF3 RF1 STAT1 FOS CEBPB RELA JUN PAX6 VDR NFIC RBPJ TP53 TFAP2 E2F1 ZBTB7 KLF4 MYC YY1 ATF2 SP3 SP1

(b)

Path: 05160_6

Path: 05320_4

Path: 05322_6

Path: 05322_5

Path: 05330_2

Path: 05320_1

Path: 05310_3

Path: 05150_14

Path: 05330_1

Path: 05310_1

Path: 05416_4

Path: 05332_1

Path: 05330_4

Path: 04940_1

Path: 05320_3

Path: 05320_6

Path: 05330_7 Path: 05320_6 Path: 05320_3 Path: 05330_4 Path: 04940_1 Path: 05416_4 Path: 05332_1 Path: 05330_1 Path: 05310_1 Path: 05150_14 Path: 05310_3 Path: 05330_2 Path: 05320_1 Path: 05322_5 Path: 05320_4 Path: 05322_6 Path: 05160_6

RFX2 RFX1 RFX3 XBP1 CREB1 IRF7

Path: 05330_7

(a)

(c)

Figure 3: One-way clustering to TFs with common motif of the TSN. (a) One-way clustering between 83 TFs and 436 subpathways. The corresponding cell was colored orange if there was an edge between the TF and subpathway in the TSN. Subpathway labels were colored according to the pathway class colors used in Figure 2. (b) Zoom-in plot of part of the orange circled region in (a), showing the TFs with similar motif and pathway class. (c) Zoom-in plot of part of the subpathway labels in the lower-right region of (a).

[34]. To explore whether the common-motif TFs tended to regulate cofunctional subpathways, we extracted 1061 human DNA motifs from the Transfac database. We obtained 15 150 pairs of TFs with similar DNA motifs (using STAMP software (http://www.benoslab.pitt.edu/stamp) [35] with 𝑒 value < 0.01), of which 610 pairs (83 unique TFs in the TSN) had similar DNA motifs. These TFs annotated 436 subpathways, forming 1402 relations (Table S3). To examine whether common-motif TFs regulated the cofunctional subpathways, we applied one-way clustering to the 83 unique TFs and 436 unique subpathways (Figure 3(a)). The common-motif TFs that regulated the same pathway classes tended to group together. For example, IRF7 (the interferon regulatory factor 7) driven regulatory cascade in which genetic variation on chromosome 15q25 leads to

type 1 diabetes (pathway: 04940) [36] and XBP1 (X-box binding protein 1) induce pancreatic beta cell dysfunction and apoptosis of type 1 diabetes [37] (Figure 3(b)). That is, common-motif TFs tend to regulate the same pathway classes and lead to the same disease. To further investigate the effect of the common motif TFs on subpathway, we randomly shuffled the 610 pairs of common-motif TFs 1000 times, while keeping the number of each TF unchanged. The average Jaccard coefficient of the real common-motif TF pairs was significantly higher than the average value for the 1000 times randomly shuffled TF pairs at the subpathway level. The mean Jaccard coefficient of the real common-motif TF pairs was 0.0568 compared with 0.0399 for the randomly shuffled TF pairs (experiential 𝑝 = 0.007, Figure 4(a)). The average Jaccard coefficient of the

8

BioMed Research International

40

100

60 40 Observed

20

80 60 40 Observed

20

20 Observed 10

0.30

0.04 0.05 0.06 0.07 0.08 0.09 0.10

0.025 0.035 0.045 0.055 Jaccard at subpathway level

Jaccard at pathway level

(a)

400

Observed

0

250

100

200

80

150 100 Observed

50

0.45

60 40 Observed

20 0

0 0.00

0.00 0.02 0.04 0.06 0.08 0.10 0.12

0.40

(c)

Probability density

Probability density

600

0.35

Jaccard at pathway class level

(b)

800

200

30

0

0

0

Probability density

Probability density

80

Probability density

Probability density

100

Jaccard at subpathway level

0.05

0.10

0.15

0.20

Jaccard at pathway level

(d)

0.25

0.30

0.35

0.40

Jaccard at pathway class level

(e)

(f)

150 800 600 400 200

Observed

0

200 150 100 Observed

50 0

0.000 0.005 0.010 0.015 0.020 0.025 0.030 Jaccard at subpathway level (g)

Probability density

Probability density

Probability density

1000 100

50 Observed 0

0.00

0.02 0.04 0.06 0.08 Jaccard at pathway level (h)

0.0

0.1 0.2 0.3 0.4 Jaccard at pathway class level

0.5

(i)

Figure 4: Jaccard coefficient. (a) The average Jaccard coefficient of common-motif TF pairs in TSN (0.0568) was significantly greater than that of 1000 times randomly shuffled TFs (0.0399) at the subpathway level (experiential 𝑝 = 0.007). (b) The average Jaccard coefficient of real common-motif TF pairs was significantly higher than that of random TF pairs (0.1045 versus 0.0723, experiential 𝑝 = 0) at the pathway level. (c) The average Jaccard coefficient of real common-motif TF pairs was significantly higher than that of random TF pairs (0.3867 versus 0.3649, experiential, 𝑝 = 0.005) at the pathway class level. The result implied that TFs with similar DNA motifs regulated the cofunctional subpathway. (d) The average Jaccard coefficient of real common-family TF pairs in TSN (0.1239) was significantly higher than that of random TF pairs (0.005) at the subpathway level (experiential 𝑝 = 0). (e) The average Jaccard coefficient of real common-family TF pairs in TSN (0.1478) was significantly higher than that of random TF pairs (0.0402) at the pathway level (experiential 𝑝 = 0). (f) The average Jaccard coefficient of real common-family TF pairs in TSN (0.4088) was significantly higher than that of random TF pairs (0.2420) at the pathway class level (experiential 𝑝 = 0). The result implied that common-family TFs regulated the cofunctional subpathway. (g) The average Jaccard coefficient of real common-family TF pairs in TSN (0.0289) was significantly higher than that of random TF pairs (0.0048) at the subpathway level (experiential 𝑝 = 0). (h) The average Jaccard coefficient of real common-family TF pairs in TSN (0.0935) was significantly higher than that of random TF pairs (0.0386) at the pathway level (experiential 𝑝 = 0). (i) The average Jaccard coefficient of real common-family TF pairs in TSN (0.4779) was significantly higher than that of random TF pairs (0.2413) at the pathway class level (experiential 𝑝 = 0).

BioMed Research International real common-motif TF pairs was significantly higher than for the randomly shuffled TF pairs (0.1045 versus 0.0723, experiential 𝑝 = 0, Figure 4(b)) at the pathway level. The average Jaccard coefficient of the real common-motif TF pairs was significantly higher than that of the randomly shuffled TF pairs (0.3867 versus 0.3649, experiential 𝑝 = 0.005, Figure 4(c)) at the pathway class level. These results revealed a high degree of consistent functionality of subpathways regulated by the common-motif TFs. Thus, the TSN can help us explain the consistent function of the subpathways regulated by the common-motif TFs. For example, HNF4A (hepatocyte nuclear factor 4, alpha) and NR1I3 (nuclear receptor subfamily 1, group I, member 3) have similar DNAbinding motifs (using STAMP software [35] with 𝑒 value < 0.01) (Figure 5), and the Jaccard coefficient at the subpathway level was 0.1538 (Figure 5(a)); the Jaccard coefficient at the pathway level was 0.3333 (Figure 5(b)); the Jaccard coefficient at the pathway class level was 0.3333 (Figure 5(c)). They, respectively, regulated subpathways 00982 4 and s 0098210, and these subpathways belonged to the same drug metabolism-cytochrome P450 pathway (pathway: 00982, Figure 5(d)) [38]. HNF4A and NR1I3 regulate the drugmetabolizing enzyme cytochrome P450 3A4 (CYP3A4) expression. HNF4A is critically involved in the NR1I3mediated transcriptional activation of CYP3A4. CYP3A4 is thought to be involved in the metabolism of nearly 50% of all the drugs currently prescribed, and alteration in the activity or expression of this enzyme seems to be a key predictor of drug responsiveness and toxicity [38]. Recent studies have identified HNF4A sites in the CYP2C8 and CYP2C9 promoter. Silencing HNF4A reduces the constitutive expression of CYP2C8 and CYP2C9. CYP2C enzymes are expressed constitutively and comprise about 20% of the total cytochrome P450 in human liver [39]. Constitutive androstane receptor NR1I3 activates CYP2B6, CYP2C9, and CYP3A4 expression to increase metabolic capability [40]. 3.3. Common Family TFs Tend to Regulate Cofunctional Subpathways. TFs in a common family had similar structure. Some researchers have revealed functional redundancies of the TF family, and common-family TFs show functional similarity [12–14]. Whether common-family TFs tend to regulate the cofunctional subpathways was still unknown. We extracted 80 unique TFs belonging to 28 families from the Transfac database, and these TFs regulated 419 unique subpathways from five pathway classes in KEGG. Eighty unique TFs and 419 unique subpathways formed 1353 relations (Table S4). To test the functional consistency of subpathways regulated by the common-family TFs, we performed one-way clustering to the 80 TFs and 419 subpathways (Figure 6(a)). The TFs that regulated the same pathway classes tended to gather in the same families at the local level. For example, the steroid hormone receptors family of TFs (NR5A1, NR4A1, and NR5A2) regulated the steroid hormone biosynthesis pathway (pathway: 00140, Figure 6(b)). That is, commonfamily TFs tended to regulate the cofunctional subpathways.

9 This local clustering phenomenon may be interpreted by the small average number of TFs (2.86) in each family. To further analyze the effects of the common-family TFs on subpathways, we applied the Jaccard coefficient to evaluate the consistent functionality of subpathways regulated by pairs of TFs with common family. We randomly shuffled the family information of TFs 1000 times and kept the number of each TF (80 unique TFs) and family (28 unique families) unchanged. The average Jaccard coefficient of real common-family TF pairs was greater than that of the randomly shuffled TF pairs (0.1239 versus 0.0050, experiential 𝑝 = 0, Figure 4(d)) at the subpathway level. The average Jaccard coefficient of real common-family TF pairs was significantly greater than that of the randomly shuffled TF pairs (0.1477 versus 0.0402, experiential 𝑝 = 0, Figure 4(e)) at the pathway level. The average Jaccard coefficient of real common-family TF pairs was significantly greater than that of randomly shuffled TF pairs (0.4088 versus 0.2420, experiential 𝑝 = 0, Figure 4(f)) at the pathway class level. These results indicated that common-family TFs regulate cofunctional subpathways. For example, HNF4A and NR1I3 appeared in the common steroid hormone receptors TF family, and they, respectively, regulated subpathways 0098-24 and 00982-10, and these subpathways belonged to the same drug metabolism-cytochrome P450 pathway (Figure 5). 3.4. Common-Tissue TFs Tend to Regulate Cofunctional Subpathways. TFs are always expressed in specific tissues or cell lines. Common-tissue TFs have functional consistency [15]. We wanted to explore the consistent function of subpathways regulated by the common-tissue TFs. We obtained 1464 relations between 122 unique tissues or cell lines and 343 unique TFs, of which 47 unique TFs were from 45 unique tissues and enriched in 385 unique subpathways in the TSN. So, there are 3494 relations between 47 TFs and 385 subpathways (Table S5). We retained tissues or cell types that included at least three unique TFs and obtained 44 unique TFs belonging to 21 tissue types, and these TF target genes were located in 378 unique subpathways, forming 2339 relations between TFs and subpathways. To examine whether common-tissue TFs regulated the cofunctional subpathways, we applied one-way clustering to the 44 unique TFs and 378 unique subpathways (Figure 7(a)). The TFs that regulated the same pathway classes tended to gather in the same tissues at the local level. For example, SP1 and TP53 were expressed in MCF7 cells (Figure 7(b)), increasing binding of SP1 complex to the TP53 promoter region, which enhanced expression of tumor suppressor factor TP53 and led to breast cancer cell apoptosis (pathway: 04210) [41]. This means that cofunctional subpathways regulated by common-tissue TFs tended to form many small clusters. This local clustering phenomenon might be explained by the following reasons: (i) the average amount of TF (5.35) related to tissues or cell lines was small; (ii) many TFs belonged to different tissue or cell types; and (iii) some special tissue TFs had a fundamental biological function and took part in many basic functions [42].

10

BioMed Research International HNF4A and NR1R3 common motif 1 0

HNF4A and NR1R3 common family steroid hormone receptors family

2 (bits)

(bits)

2

1

2

3

4

5

6

HNF4A and NR1R3 common tissue

1 0

7

1

2

3

4

5

6

Path: 00120_15 Path: 00120_19 Path: 00120_6 Path: 00120_7 Path: 00120_9

Path: 03320_2

Path: 04610_3

Path: 04610_5

Path: 00590_2 Path: 00590_3

Path: 00982_5

Path: 04610_6 Path: 04610_8

Path: 00590_4

Path: 00590 Path: 00830_2 Path: 00982_4

Path: 04610_4

Path: 00590_1

Path: 04060

Path: 04610_2

Path: 00980_1

Path: 00982_9

Path: 00830

Path: 03320

Path: 00980

Organismal systems Human diseases

Metabolism

Path: 00982

Path: 04610

Path: 04950_4

Path: 00591_1

Path: 00591

Path: 04950

Path: 03320_1

Path: 00982_10

Path: 00980_2

(a) Jaccard coefficient at subpathway level is 0.1538

(b) Jaccard coefficient at pathway level is 0.3333 4-Glutathionyl-CPA (inactive) 4-Keto-CPA

Dechloroethyl-CPA Chloroacetaldehyde

1.2.1.5 Carboxyphosphamide

CYP3A4 CYP2C

4-OH-CPA 4-Hydroxyifosfamide

Nornitrogen mustard (active) Phosphamide mustard (active)

CYP2B6

Ifosfamide (prodrug)

Aldophosphamide Acrolein Aldoifosfamide

CYP2B6 CYP3A4

CYP3A4

Alcophosphamide 1.1.1.1

CYP2C Cyclophosphamide (prodrug)

Excretion

2.5.1.18 CYP3A4

CYP3A4

(c) Jaccard coefficient at pathway class level is 0.3333

Carboxyphosphamide

Isophosphamide mustard (active)

CYP3A5

CYP2B6 4-Ketoifosfamide

Alcophosphamide

Chloroacetaldehyde Dechloroethylifosfamide (d) Drug metabolism-cytochrome P450

Figure 5: Common-motif, common-family, and common-tissue TF pair. HNF4A and NR1I3 are a pair of TFs with a common motif, common family (steroid hormone receptors), and common tissue (liver). (a) Jaccard coefficient at the subpathway level of HNF4A and NR1I3 was 0.1538. (b) Jaccard coefficient at the pathway level was 0.3333. (c) Jaccard coefficient at the pathway class level was 0.3333. (d) HNF4A regulated subpathway 00982-10, and NR1I3 regulated subpathway 00982-4. Red genes represent the shared genes between subpathways 00982-10 and 00982-4. Pink genes represent the genes only in subpathway 00982-4. Purple nodes represent the genes in the entire pathway 00982.

To test the effects of common-tissue TFs on subpathways in a general sense, we calculated the Jaccard coefficient for pairs of common-tissue TFs. We randomly shuffled the tissue information of TFs (47 unique TFs and 45 unique tissues) 1000 times and kept the number of

each TF unchanged. The average Jaccard coefficient of real common-tissue TF pairs was significantly greater than that of randomly shuffled TF pairs at the subpathway level (0.0289 versus 0.0048, experiential 𝑝 = 0, Figure 4(g)). The average Jaccard coefficient of real common-tissue TF pairs

BioMed Research International

11

Common-family TFs

Pathway class CREB AP 1 Ets type AP 1 CREB Fungal regulators GATA factors GATA factors GATA factors STAT STAT Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) TCF 1 AP 1 Ubiquitous bHLH ZIP factors AP 2 Developmental/cell cycle regulators NF AT C/EBP like factors Tissue specific regulators Rel/ankyrin Ubiquitous bHLH ZIP factors Interferon regulating factors Responders to external signals AP 1 Developmental/cell cycle regulators Developmental/cell cycle regulators STAT STAT Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) GATA factors Developmental/cell cycle regulators Ets type SMAD SMAD Ets type Tissue specific regulators Homeo domain only TCF 1 Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) Steroid hormone receptors (NR3) GATA factors Steroid hormone receptors (NR3) POU domain factors Interferon regulating factors Interferon regulating factors Interferon regulating factors C/EBP like factors AP 1 STAT Rel/ankyrin AP 1 Steroid hormone receptors (NR3) p53 like SMAD Heteromeric CCAAT factors p53 like Regulators with a DNA recognition wing AP 1 Regulators with a DNA recognition wing Regulators with a DNA recognition wing CREB Paired plus homeo domain Steroid hormone receptors (NR3) STAT NF 1 Steroid hormone receptors (NR3) AP 2 csd Cell cycle controlling factors Developmental/cell cycle regulators Cell cycle controlling factors Ubiquitous factors AP 1 Ubiquitous factors p53 like Ubiquitous factors

CREM JUND ETV4 NRL ATF1 HMGA1 GATA2 GATA1 GATA3 STAT5A STAT4 NR2F1 THRB TCF7 JUNB MITF TFAP2C KLF5 NFATC1 CEBPA FOXA2 REL SREBF1 IRF5 SRF NFE2L2 GLI3 GLI1 STAT3 STAT6 NR1I3 NR1I2 GATA6 EGR1 ELF1 SMAD2 SMAD3 SPI1 FOXA3 HNF1A LEF1 NR5A1 NR4A1 NR5A2 GATA4 HNF4A POU2F1 IRF7 IRF3 IRF1 CEBPB FOS STAT1 RELA JUN ESR1 TP73 SMAD4 NFYA TP63 RFX1 XBP1 RFX2 RFX3 CREB1 PAX6 AR STAT5B NFIC VDR TFAP2A YBX1 E2F1 KLF4 MYC YY1 ATF2 SP3 TP53 SP1

(a)

Steroid hormone receptors (NR3)

NR5A1 NR4A1

Steroid hormone receptors (NR3)

NR5A2

Steroid hormone receptors (NR3)

GATA4

GATA factors

HNF4A

Steroid hormone receptors (NR3)

POU2F1 Path: 00140_11 Path: 00140_12 Path: 00140_14 Path: 00140_20 Path: 00140_19 Path: 00140_13 Path: 00140_16 Path: 00140_10 Path: 00140_6 Path: 00140_5 Path: 00140_3 Path: 00140_4 Path: 00140_9 Path: 00140_8 Path: 00140_18 Path: 00140_7 Path: 00980_2 Path: 00982_10 Path: 00590_4 Path: 00590_3 Path: 00590_1 Path: 00590_2 Path: 00591_1 Path: 00982_9 Path: 00982_4 Path: 00982_5

POU domain factors

(b)

Figure 6: One-way clustering to TFs with common family of the TSN. (a) One-way clustering between 80 TFs and 419 subpathways. The corresponding cell was colored orange if there was an edge between the TF and subpathway in the TSN. Subpathway labels were colored according to the pathway class colors used in Figure 2. (b) Zoom-in plot of part of the orange circled region in (a), showing the common family TFs and subpathways.

was significantly greater than that of randomly shuffled TF pairs at the pathway level (0.0935 versus 0.0386, experiential 𝑝 = 0, Figure 4(h)). The average Jaccard coefficient of real common-tissue TF pairs was significantly greater than that of randomly shuffled TF pairs at the pathway class level (0.4779 versus 0.2413, experiential 𝑝 = 0, Figure 4(i)). We conclude that common-tissue TFs tend to regulate the consistent functional subpathway at different functional levels. For example, HNF4A and NR1I3 were

expressed in the liver (Figure 5), where they, respectively, regulated subpathways 00982-4 and subpathways 00982-10, which belonged to the same drug metabolism-cytochrome P450 pathway. HNF4A and NR1I3 regulated CYP3A4, and CYP3A4 promoter activity was most pronounced in liver cells, indicating that CYP3A4 is a liver-specific factor that is required for physiological transcriptional response [38]. Constitutive androstane receptor NR1I3 is predominantly expressed in the liver and it activates CYP2B6, CYP2C9,

12

BioMed Research International

Common-tissue TFs

Pathway class

T cells Breast T cells Breast HeLa HepG2 Hep3B T cells HepG2 Liver MCF7 T cells Testis Heart Brain Kidney Liver HUVEC HeLa HeLa K562 HepG2 MCF7 MCF7 K562 MCF7 Lymph node Kidney Heart Liver Liver Caco HUVEC Hep3B K562 T cells Thymus Spleen Liver Bone marrow Spleen T cells Testis Thymus Intestine Bone marrow Liver Lymph node Lymphocytes Spleen T cells Testis Thymus Thymus T cells Heart Breast Thymus Lymphocytes HeLa Caco Breast Intestine Kidney Liver Spleen Liver Liver Brain Caco Intestine Hep3B Kidney Liver Testis Heart Brain Kidney Liver Lymphocytes Spleen Testis Thymus Spleen Lymph node Thymus Lymph node Bone marrow Spleen Thymus K562 HUVEC MCF7 HepG2 MCF7 HeLa MCF7 HeLa Brain Bone marrow Intestine Lymphocytes MCF7 Testis HeLa HUVEC HeLa MCF7

TCF7 TCF7 CREM ETV4 NR2F1 GATA2 GATA2 GATA3 GATA3 FOXA2 TFAP2C NFATC1 KLF5 SREBF1 SREBF1 SREBF1 SREBF1 REL REL SRF NFE2L2 NFE2L2 STAT3 FOS FOS ESR1 TP73 NR1I3 NR1I3 NR1I3 NR1I2 GATA6 EGR1 EGR1 EGR1 EGR1 ELF1 ELF1 SPI1 SPI1 SPI1 SPI1 SPI1 SPI1 AKNA AKNA AKNA AKNA AKNA AKNA AKNA AKNA AKNA LEF1 LEF1 TP63 TP63 TP63 NFYA NFYA HNF1A HNF1A HNF1A HNF1A HNF1A HNF1A FOXA3 NR4A1 NR4A1 GATA4 HNF4A HNF4A HNF4A HNF4A HNF4A CREB1 CREB1 CREB1 CREB1 CREB1 CREB1 CREB1 CREB1 IRF7 IRF7 IRF7 IRF5 IRF5 IRF5 IRF5 RELA RELA E2F1 STAT5B VDR VDR NFIC NFIC TP53 TP53 TP53 TP53 TP53 TP53 TFAP2A SP1 SP1 SP1

(a) MCF7 HeLa Brain Bone marrow Intestine Lymphocytes MCF7 Testis HeLa HUVEC HeLa MCF7 Path: 04540_13 Path: 04110_26 Path: 04520_7 Path: 04520_5 Path: 04144_2 Path: 04520_3 Path: 04810_29 Path: 04510_4 Path: 04510_24 Path: 04510_23 Path: 04510_20 Path: 04510_19 Path: 04510_13 Path: 04510_1 Path: 04110_18 Path: 04110_25 Path: 04510_8 Path: 04510_6 Path: 04510_10 Path: 04510_18 Path: 04210_23 Path: 04110_8 Path: 04110_20 Path: 04110_27 Path: 04210_20 Path: 04210_24 Path: 04510_7 Path: 04510_5 Path: 04510_25 Path: 04510_17 Path: 04510_15 Path: 04510_16 Path: 04115_1 Path: 04115_7 Path: 04115_4 Path: 04115_2 Path: 04115_3 Path: 04110_1 Path: 04110_7 Path: 04110_17 Path: 04110_19 Path: 04110_4 Path: 04110_3 Path: 04110_22 Path: 04110_23

NFIC NFIC TP53 TP53 TP53 TP53 TP53 TP53 TFAP2A SP1 SP1 SP1

(b)

Figure 7: One-way clustering to TFs with common tissue of the TSN. (a) One-way clustering between 44 unique TFs and 378 unique subpathways. The corresponding cell was colored orange if there was an edge between the TF and subpathway in the TSN. Subpathway labels were colored according to the pathway class colors used in Figure 2. (b) Zoom-in plot of part of the orange circled region in (a), showing the common tissue TFs and subpathways.

and CYP3A4 expression to increase metabolic capability [40].

4. Discussion and Conclusion To gain insight into the influence of TF characteristics on their regulatory subpathways, we constructed a TSN using the TF target gene associations and 𝑘-clique subpathway identification method. We used the 𝑘-clique (𝑘 = 3) method to divide large pathways into multiple subpathways. This

ensured that the identified TF-regulated subpathways were located in the TF-related local regions of the pathways and also increased the tendency for genes in a subpathway to share similar biological functions and be involved in similar biological processes. To date, most pathway identification methods have only identified entire pathways without considering the scale of these pathways. However, each pathway has obviously different scales. These differences in scale hinder the evaluation of TF-related pathways from global networks and the identification of TF-related subpathways that are

BioMed Research International usually located in the local regions of the pathways. By using the subpathway identification method, we ensured that each subpathway was represented on a small scale. Through our analyses of the TSN, we found that the cofunctional subpathways were regulated by the commonmotif, common-family, or common-tissue TFs. We performed one-way clustering on the TSN based on commonmotif, common-family, and common-tissue TFs. The results showed that all of the TFs with three types of characters that regulated the same pathway classes tended to cluster together. Furthermore, we applied the Jaccard coefficient to establish whether common-motif, common-family, and common-tissue TFs regulate cofunctional subpathways. Our results provided strong support for cofunctional subpathways that were regulated by TFs with three types of characters. For example, HNF4A and NR1I3 were a pair of TFs with a common motif, common family (steroid hormone receptors), and common tissue (liver). HNF4A and NR1I3 were expressed in the liver and, respectively, regulated subpathways 00982-4 and 00982-10, which belonged to the same drug metabolismcytochrome P450 pathway. HNF4A and NR1I3 regulated expression of the drug-metabolizing enzyme cytochrome P450 3A4 (CYP3A4) expression [28]. HNF4A is critically involved in the NR1I3-mediated transcriptional activation of CYP3A4, which is thought to be involved in the metabolism of nearly 50% of all the drugs currently prescribed. Alteration in the activity or expression of CYP3A4 seems to be a key predictor of drug responsiveness and toxicity. CYP3A4 promoter activity was most pronounced in liver cells, indicating that CYP3A4 is a liver-specific factor that is required for physiological transcriptional response [38]. Recent studies have identified HNF4A sites in the CYP2C8 and CYP2C9 promoter. Silencing HNF4A reduces constitutive expression of CYP2C8 and CYP2C9, as shown by quantitative polymerase chain reaction analysis [39]. Constitutive androstane receptor NR1I3 activated CYP2B6, CYP2C9, and CYP3A4 expression to increase metabolic capability [40]. We further discussed the clinical or cell development study of HNF4A and NR1I3. We increased the following four aspects of analysis. First, we downloaded the large-scale liver cancer RNAseq data from TCGA public database (http://cancergenome.nih.gov/); we used the MARS method of R package “DEGseq” to analyze the 423 samples (373 liver cancer samples and 50 normal samples) and found HNF5A (𝑞 value < 1.0𝑒 − 6) and NR1I3 (𝑞 value < 1.0𝑒 − 6) were significantly differentially expressed in the liver cancer dataset; thus, they may be used for the study of liver cancer. Second, we searched PubMed database and confirmed that HNF4 and NR1I3 were associated with liver cancer. Previous studies have demonstrated that downregulation of HNF4A was associated with hepatocellular carcinoma (HCC) progression in rodents and humans. In addition, the study of Huang et al. suggested that NR1I3 was a primary regulator of drug metabolism and detoxification and NR1I3 activation or transient strictly limited liver growth [43–45]. Third, we extracted the direct neighbours of this pair of TFs in the protein-protein interaction network and performed the pathway enrichment analysis by DAVID (https://david.ncifcrf.gov/). Results suggested that these genes were significantly enriched in some well-known

13 clinical or cell development related pathways of KEGG such as “pathways in cancer” (hsa05200, 𝑝 = 1.33𝑒 − 04) and “cell cycle pathway” (hsa04110, 𝑝 = 2.88𝑒 − 04). Finally, we tested the Pearson correlation coefficient of this pair of TFs, and the result showed that they were significant positive correlation (𝑟 = 0.4257, 𝑝 < 2.2𝑒−16), which indicated that these two TFs may collectively regulate their downstream target genes; thus, they played important role in the initiation and progression of liver cancer. In summary, the above analysis result indicated that this pair of TFs (HNF4A and NR1I3) may be potential clinical biomarkers of liver cancer. To confirm the validity of our results, we also constructed the TSN with 𝑘 = 4 (𝑝 < 0.01) (Table S6–S9). First, we compared the two networks and found that the two networks were robust. There were 1470 edges between 99 TFs and 343 subpathways in the TSN with 𝑘 = 4, while there were 1490 edges between 89 TFs and 468 subpathways in the TSN with 𝑘 = 3. The overlap of TFs (86 TFs) in the two networks was statistically significant (hypergeometric test 𝑝 < 1.0𝑒 − 6). In the TSN with 𝑘 = 4, there were 343 subpathways corresponding to 95 pathways, and the overlap of pathways (87 pathways) in the two networks was statistically significant (hypergeometric test 𝑝 < 1.0𝑒 − 6). Thus, these two networks were robust. We then repeated some of the analyses and found that the results of the two networks were similar. We compared one-way clustering results of the two networks and found that the results of two networks were similar. The clustering results of TSN with 𝑘 = 4 (Figures S1, S2, and S3) showed that the subpathways regulated by TFs with common motif, common family, and common tissue tended to cluster in the same pathway class. In particular, we verified the consistent function of subpathways regulated by the TFs with common motif, common family, and common tissue by calculating Jaccard coefficient and found the results of two networks were similar. In the TSN with 𝑘 = 4, the average Jaccard coefficient of the real common-motif, commonfamily, and common-tissue TF pairs was significantly higher than the random TF pairs by shuffling TSN 1000 times at the subpathway level, pathway level, and pathway class level (Table S10). As a control, TSN with 𝑘 = 4 presented the similarity of results when compared with that of 𝑘 = 3. These results indicated that our conclusions were robust in different threshold of subpathway networks. In summary, our study is significant for understanding the TFs underlying the regulatory function in complex biological systems.

Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments This work was supported by the National High Technology Research and Development Program of China (863 Program, Grant no. 2014AA02110), the National Program on Key Basic Research Project (973 Program, Grant no. 2014CB910504), the National Natural Science Foundation of China (Grants

14 nos. 91439117 and 61473106), and the Research Project of Health Department of Heilongjiang Province (2014-418).

References [1] B. Xu, D. E. Schones, Y. Wang, H. Liang, and G. Li, “A structuralbased strategy for recognition of transcription factor binding sites,” PLoS ONE, vol. 8, no. 1, Article ID e52460, 2013. [2] M. Xu and Z. Su, “A novel alignment-free method for comparing transcription factor binding site motifs,” PLoS ONE, vol. 56, no. 1, article e8797, 2010. [3] G.-H. Wei, G. Badis, M. F. Berger et al., “Genome-wide analysis of ETS-family DNA-binding in vitro and in vivo,” The EMBO Journal, vol. 29, no. 13, pp. 2147–2160, 2010. [4] S. Zhang, X. Zhou, C. Du, and Z. Su, “SPIC: a novel similarity metric for comparing transcription factor binding site motifs based on information contents,” BMC Systems Biology, vol. 7, supplement 2, article S14, 2013. [5] A. S. Bais, N. Kaminski, and P. V. Benos, “Finding subtypes of transcription factor motif pairs with distinct regulatory roles,” Nucleic Acids Research, vol. 39, no. 11, article e76, 2011. [6] P. Li, G. Xu, G. Li, and M. Wu, “Function and mechanism of tumor suppressor gene LRRC4/NGL-2,” Molecular Cancer, vol. 13, article 266, 2014. [7] D. M. Budden, D. G. Hurley, J. Cursons, J. F. Markham, M. J. Davis, and E. J. Crampin, “Predicting expression: the complementary power of histone modification and transcription factor binding data,” Epigenetics & Chromatin, vol. 7, article 36, 2014. [8] R. F. Lowdon, B. Zhang, M. Bilenky et al., “Regulatory network decoded from epigenomes of surface ectoderm-derived cell types,” Nature Communications, vol. 5, article 5442, 2014. [9] W. Czaja, K. Y. Miller, M. K. Skinner, and B. L. Miller, “Structural and functional conservation of fungal MatA and human SRY sex-determining proteins,” Nature Communications, vol. 5, article 5434, 2014. [10] A. Vandenbon, Y. Kumagai, S. Akira, and D. M. Standley, “A novel unbiased measure for motif co-occurrence predicts combinatorial regulation of transcription,” BMC Genomics, vol. 13, supplement 7, article S11, 2012. [11] R. Torella, J. Li, E. Kinrade et al., “A combination of computational and experimental approaches identifies DNA sequence constraints associated with target site binding specificity of the transcription factor CSL,” Nucleic Acids Research, vol. 42, no. 16, pp. 10550–10563, 2014. [12] L. H. Wang, N. H. Ing, S. Y. Tsai, B. W. O’Malley, and M. J. Tsai, “The COUP-TFs compose a family of functionally related transcription factors,” Gene Expression, vol. 1, no. 3, pp. 207–216, 1991. [13] S. Danisman, A. D. J. van Dijk, A. Bimbo et al., “Analysis of functional redundancies within the Arabidopsis TCP transcription factor family,” Journal of Experimental Botany, vol. 64, no. 18, pp. 5673–5685, 2013. [14] B. Huang, Z. T. Qi, Z. Xu, and P. Nie, “Global characterization of interferon regulatory factor (IRF) genes in vertebrates: glimpse of the diversification in evolution,” BMC Immunology, vol. 11, article 22, 2010. [15] D.-X. Ding, F.-F. Tian, J.-L. Guo et al., “Dynamic expression patterns of ATF3 and p53 in the hippocampus of a pentylenetetrazole-induced kindling model,” Molecular Medicine Reports, vol. 10, no. 2, pp. 645–651, 2014.

BioMed Research International [16] M. Kann, E. Bae, M. O. Lenz et al., “WT1 targets Gas1 to maintain nephron progenitor cells by modulating FGF signals,” Development, vol. 142, no. 7, pp. 1254–1266, 2015. [17] M. Yoshida, K. Hata, R. Takashima et al., “The transcription factor Foxc1 is necessary for Ihh–Gli2-regulated endochondral ossification,” Nature Communications, vol. 6, article 6653, 2015. [18] Y. Zhi, Y. Duan, X. Zhou et al., “NF-kappaB signaling pathway confers neuroblastoma cells migration and invasion ability via the regulation of CXCR4,” Medical Science Monitor, vol. 20, pp. 2746–2752, 2014. [19] Q. Li, G. Liu, D. Shao et al., “Mucin1 mediates autocrine transforming growth factor beta signaling through activating the cJun N-terminal kinase/activator protein 1 pathway in human hepatocellular carcinoma cells,” The International Journal of Biochemistry and Cell Biology, vol. 59, pp. 116–125, 2015. [20] J. Shan, W. Donelan, J. N. Hayner, F. Zhang, E. E. Dudenhausen, and M. S. Kilberg, “MAPK signaling triggers transcriptional induction of cFOS during amino acid limitation of HepG2 cells,” Biochimica et Biophysica Acta, vol. 1853, no. 3, pp. 539–548, 2015. [21] J. V. Almaden, R. Tsui, Y. C. Liu et al., “A pathway switch directs BAFF signaling to distinct NF𝜅B transcription factors in maturing and proliferating B cells,” Cell Reports, vol. 9, no. 6, pp. 2098–2111, 2014. [22] A. Limonciel, K. Moenks, S. Stanzel et al., “Transcriptomics hit the target: monitoring of ligand-activated and stress response pathways for chemical testing,” Toxicology in Vitro, 2015. [23] T. Judeh, C. Johnson, A. Kumar, and D. Zhu, “TEAK: topology enrichment analysis framework for detecting activated biological subpathways,” Nucleic Acids Research, vol. 41, no. 3, pp. 1425– 1437, 2013. [24] C. Li, X. Li, Y. Miao et al., “SubpathwayMiner: a software package for flexible identification of pathways,” Nucleic Acids Research, vol. 37, no. 19, article e131, 2009. [25] X. Li, C. Li, D. Shang et al., “The implications of relationships between human diseases and metabolic subpathways,” PLoS ONE, vol. 6, no. 6, Article ID e21131, 2011. [26] C. Li, D. Shang, Y. Wang et al., “Characterizing the network of drugs and their affected metabolic subpathways,” PLoS ONE, vol. 7, no. 10, Article ID e47326, 2012. [27] P. Khatri, M. Sirota, and A. J. Butte, “Ten years of pathway analysis: current approaches and outstanding challenges,” PLoS Computational Biology, vol. 8, no. 2, Article ID e1002375, 2012. [28] V. Matys, E. Fricke, R. Geffers et al., “TRANSFAC: transcriptional regulation, from patterns to profiles,” Nucleic Acids Research, vol. 31, no. 1, pp. 374–378, 2003. [29] H. Ogata, S. Goto, K. Sato, W. Fujibuchi, H. Bono, and M. Kanehisa, “KEGG: kyoto encyclopedia of genes and genomes,” Nucleic Acids Research, vol. 27, no. 1, pp. 29–34, 1999. [30] K. Beishline and J. Azizkhan-Clifford, “Sp1 and the ‘hallmarks of cancer’,” The FEBS Journal, vol. 282, pp. 224–258, 2015. [31] J.-P. Zhang, H. Zhang, H.-B. Wang et al., “Down-regulation of Sp1 suppresses cell proliferation, clonogenicity and the expressions of stem cell markers in nasopharyngeal carcinoma,” Journal of Translational Medicine, vol. 12, article 222, 2014. [32] M. L. Tornesello, C. Annunziata, L. Buonaguro, S. Losito, S. Greggi, and F. M. Buonaguro, “TP53 and PIK3CA gene mutations in adenocarcinoma, squamous cell carcinoma and high-grade intraepithelial neoplasia of the cervix,” Journal of Translational Medicine, vol. 12, article 255, 2014. [33] N. Yamamoto, Y. Takemori, M. Sakurai, K. Sugiyama, and H. Sakurai, “Differential recognition of heat shock elements by

BioMed Research International

[34]

[35]

[36]

[37]

[38]

[39]

[40]

[41]

[42]

[43]

[44]

[45]

members of the heat shock transcription factor family,” The FEBS Journal, vol. 276, no. 7, pp. 1962–1974, 2009. V. P. Ramirez, W. Krueger, and B. J. Aneskievich, “TNIP1 reduction of HSPA6 gene expression occurs in promoter regions lacking binding sites for known TNIP1-repressed transcription factors,” Gene, vol. 555, no. 2, pp. 430–437, 2015. S. Mahony and P. V. Benos, “STAMP: a web tool for exploring DNA-binding motif similarities,” Nucleic Acids Research, vol. 35, no. 2, pp. W253–W258, 2007. M. Heinig, E. Petretto, C. Wallace et al., “A trans-acting locus regulates an anti-viral expression network and type 1 diabetes risk,” Nature, vol. 467, pp. 460–464, 2010. F. Allagnat, F. Christulia, F. Ortis et al., “Sustained production of spliced X-box binding protein 1 (XBP1) induces pancreatic beta cell dysfunction and apoptosis,” Diabetologia, vol. 53, no. 6, pp. 1120–1130, 2010. R. G. Tirona, W. Lee, B. F. Leake et al., “The orphan nuclear receptor HNF4𝛼 determines PXR- and CAR-mediated xenobiotic induction of CYP3A4,” Nature Medicine, vol. 9, no. 2, pp. 220–224, 2003. R. Rana, Y. Chen, S. S. Ferguson, G. E. Kissling, S. Surapureddi, and J. A. Goldstein, “Hepatocyte nuclear factor 4𝛼 regulates rifampicin-mediated induction of CYP2C genes in primary cultures of human hepatocytes,” Drug Metabolism & Disposition, vol. 38, no. 4, pp. 591–599, 2010. J. M. Pascussi, S. Gerbal-Chaloin, L. Drocourt, P. Maurel, and M. J. Vilarem, “The expression of CYP2B6, CYP2C9 and CYP3A4 genes: a tangle of networks of nuclear and steroid receptors,” Biochimica et Biophysica Acta, vol. 1619, no. 3, pp. 243–253, 2003. G. Gu, I. Barone, L. Gelsomino et al., “Oldenlandia diffusa extracts exert antiproliferative and apoptotic effects on human breast cancer cells through ER𝛼/Sp1-mediated p53 activation,” Journal of Cellular Physiology, vol. 227, no. 10, pp. 3363–3372, 2012. B. Roson-Burgo, F. Sanchez-Guijo, C. Del Ca˜nizo, and J. De Las Rivas, “Transcriptomic portrait of human Mesenchymal Stromal/Stem cells isolated from bone marrow and placenta,” BMC Genomics, vol. 15, article 910, 2014. N. L. Lazarevich, O. A. Cheremnova, E. V. Varga et al., “Progression of HCC in mice is associated with a downregulation in the expression of hepatocyte nuclear factors,” Hepatology, vol. 39, no. 4, pp. 1038–1047, 2004. B.-F. Ning, J. Ding, C. Yin et al., “Hepatocyte nuclear factor 4𝛼 suppresses the development of hepatocellular carcinoma,” Cancer Research, vol. 70, no. 19, pp. 7640–7651, 2010. W. Huang, J. Zhang, M. Washington et al., “Xenobiotic stress induces hepatomegaly and liver tumors via the nuclear receptor constitutive androstane receptor,” Molecular Endocrinology, vol. 19, no. 6, pp. 1646–1653, 2005.

15