Genome-Wide Analysis of Hepatic Gene Silencing in Hepatoma Cell Variants

Genomics 100 (2012) 176–183

Contents lists available at SciVerse ScienceDirect

Genomics

journal homepage: www.elsevier.com/locate/ygeno

Genome-wide analysis of hepatic gene silencing in hepatoma cell variants

Gary A. Bulla ⁎, Caitlin M. Aylmer, Adele L. Dust, Jeffrey L. Kurkewich, Leon K. Mire, Arnold B. Estanda

Department of Biological Sciences, Eastern Illinois University, 600 Lincoln Avenue, Charleston, IL 61920, USA article info abstract

Article history: Genome-wide gene expression proﬁling was carried out on rat hepatoma cells and compared to proﬁles of Received 21 December 2011 hepatoma “variant” cell lines derived via a stringent selection protocol that enriches for rare cells (b1in Accepted 22 May 2012 100,000 cells) that fail to drive liver function. Results show 132 genes that are strongly (>5-fold) repressed Available online 30 May 2012 in each of the four variant cell lines tested. An additional 68 genes were repressed in 3 of 4 variant cell lines. Importantly, several of the repressed genes are members of transcriptional activation pathways, suggesting Keywords: that they may contribute to maintaining the hepatic phenotype. Ectopic expression of the HNF1A gene in a Hepatoma Cell variants variant cell line resulted in activation of 56 genes, 37 of which were included in the repressed data set. Hepatic These data suggest that a high level of reprogramming occurs when hepatoma cells convert to a non- Microarray differentiated phenotype, a process that can be partially reversed by the introduction of transcription factors. © 2012 Elsevier Inc. All rights reserved.

1. Introduction groups, indicating that although defects may be distinct, common regulatory pathways are impacted. Genetic programs involved in the establishment and maintenance of Extensive remodeling of chromatin has been shown around the liver-specific gene expression have been identified, yet the contribution Serpina1 locus by ectopicexpressionofHNF1α in the H11 cells and interplay of these pathways leading to and maintaining the differ- [16,17], suggesting that single transcription factors can lead to remodel- entiated state remain unclear. Regulatory circuitry within mammalian ing of large regions of DNA. Importantly, hepatic gene expression could hepatocytes has been mapped and includes several master regulatory be rescued in one such variant cell line, M38, by introduction of human genes that encode proteins which bind to and activate liver-specific chromosome 12 [18], although it was not possible to identify the locus genes. These hepatocyte-enriched factors include a homeodomain pro- on chromosome 12 responsible for the rescue event. Previous studies tein (HNF1α), a winged helix factor (FOXA2/HNF3β), an orphan nuclear on hepatoma variant cells were limited to a gene-by-gene analysis. receptor (HNF4), and a onecut homeodomain protein (ONECUT1/HNF6) Thus, additional characterization of these cells via whole genome profil- [1–4]. These factors appear to act synergistically to activate expression ing was merited in order to identify the extent of genomic repro- of hundreds of hepatic genes [4]. Many of these factors operate in gramming that occurred in the variant cell types as well as to uncover autoregulatory loops to activate expression of each other [4–6].Addi- candidate genes responsible for maintaining the hepatic phenotype. tional regulatory circuitry has also been mapped in hepatocytes [7–9]. Inordertoidentifygeneproductsthatmayplayakeyroleininitiat- Unfortunately, the identification and role of additional genetic networks ing and/or maintaining the liver phenotype, we profiled genome-wide that determine hepatic identity has remained elusive. expression in hepatoma cells and hepatoma-derived variant cell lines. Analysis of cultured hepatoma cell lines and their dedifferentiated Our strategy was to compare gene expression profiles of a hepatoma pa- derivatives have long provided information regarding molecular rental cell line with that of a panel of 4 independently derived hepatoma pathways involved in repression and activation of liver-specific variant cell lines. Our goal was to determine the full extent of repro- gene expression [6,10–13]. Previous studies have shown that stable gramming that occurs in the variant cell types leading to loss of hepatic transfection of transcription factors HNF4 or HNF1α into certain hep- function as well as to identify candidate genes that are responsible for atoma variant cell lines (M38 and H11 cells) resulted in re-activation maintaining regulatory circuitry controlling liver function. of the silenced α1AT (Serpina1) gene and other hepatic genes, but introduction of these factors into other variant cell lines (HS2 or M29 2. Materials and methods cells) failed to rescue hepatic function [5,14,15], suggesting that these variant cell lines may have distinct defects. Fusion of various 2.1. Cell lines and culture conditions combinations of variant cell lines failed to establish complementation All cell lines described are derived from the rat liver tumor line H4IIEC3. Fg14 cells are an adenine phosphoribosyltransferase-positive + – + ⁎ Corresponding author. Fax: +1 217 581 7141. (APRT ), xanthine guanine phosphoribosyltransferase-positive (GPT ) − E-mail address: [email protected] (G.A. Bulla). cell line derived from the APRT and hypoxanthine–guanine

− phosphoribosyltransferase-negative (HPRT )Fado-2cellsbystable with the test RNA samples contained 19 μl sterile nuclease-free H2O, transfection of Gpt and Aprt transgenes driven by the human 3 μl RNA, 25 μl MasterAmp 2X RT-PCR PreMix, 2 μl Oligo DT and 1 μl α1-antitrypisin (α1AT, SERPINA1) gene promoter (−640 to −2bp). Moloney Murine Leukemia Virus Reverse Transcriptase (MMLV-RT).

Dedifferentiated cell lines, including M29, H11, HS2 and M38 cells, The control mix contained 23 μl sterile nuclease-free H2O, 25 μl were derived from the Fg14 cells by negative selection against both MasterAmp 2X RT-PCR PreMix, 1 μl MMVL-RT, and 1 μl control mix. Aprt and Gpt transgene gene expression using 20 μg/ml 2,6- The mixes were stored on ice before starting the reaction. Samples diaminopurine (DAP) and 30 μg/ml 6-thioxanthine, respectively [19]. were then incubated at 37 °C for 30 min, and qPCR was run using M29 and M38 cells have a reversion rate to APRT+ of approximately the synthesized cDNA. All primer pairs used were complementary 10−5, while the H11 cells revert at a frequency of 10−3 and HS2 cells at to exonic sequences that spanned intronic regions in order to limit b10−8.Allcellsweremaintainedin1:1Ham'sF12/Dulbecco'smodiﬁed the possibility of genomic DNA ampliﬁcation giving rise to false pos- Eagle's medium (FDV) containing 5% fetal bovine serum (FBS) (Gibco) itive signals for RNA production. and 5 μg/100 ml penicillin–streptomycin (Gibco) at 37 °C in a humid 5% The cDNA quality of each cell line was tested with the reference

CO2 chamber. gene Gapdh. Triplicate assays were used for each reaction and raw threshold values (Ct) were averaged following ampliﬁcation. For the 2.2. RNA extraction test RNA, triplicates of 20 ng cDNA was added to 10 μl of Fast SYBR Green (Applied Biosystems), 1 μl of a primer working stock containing RNA was extracted from nearly-conﬂuent monolayers using a both forward and reverse (IDTDNA) with a concentration of ~50 μM

Qiagen RNeasy Mini Kit (Cat #74104) following the kit protocol and 4 μlofRNase-freeH2O (Qiagen). Primer sequences used for with the addition of a DNaseI (Cat #79254) digestion step, as per pro- RT-qPCR are as follows: Gapdh: for- 5′ tgattctacccacggcaagtt 3′,rev- tocol. Brieﬂy, approximately 107 cells were lysed with RLT buffer, 5′ tgatgggtttcccattgatga 3′;Alb:for-5′ catcctgaaccgtctgtgtg 3′,rev-5′ samples spun through a Qiagen column shredder to ensure homoge- tttccaccaaggacccacta 3′;Fgb:for-5′ gaccatccacaacggcatgt 3′,rev-5′ nization, and RNA collected on an RNEasy column. Samples were cgccgccatcttcttttgag 3′; SerpinA1: for- 5′ cctataccgggagctggtccat 3′, washed and then digested with DNase I for 15 min at RT. RNA was rev- 5′ ttgcgagtgtcacccttgct 3′;Hnf4a:for-5′ tgagcctggaggattacatcaac further washed then eluted. RNA integrity was determined by 28S 3′,rev-5′ aactggatctgcatcatctg 3′; Hhex: for- 5′ aaaggaaaggcggtcaag- and 18S rRNA visualization following gel electrophoresis in MOPS tgaggt 3′,rev-5′ atcttggccagacgctttctctca 3′; Wnt4: for- 5′ agaag- with 1% agarose-2.2 M formaldehyde gels, as described [20]. RNA pu- ctcaaaggcctgatccaga 3′,rev-5′ actggtattgcgactcctcaatgg 3′. The control rity and concentration were determined by nano-drop spectropho- for the RT-qPCR contained 5 μl of RNase-free H2O, in place of cDNA, tometry at 260 and 280 nm. Triplicate samples of RNA extracted 10 μl of SYBR Green, 4 μl of RNase-free H2Oand1μl of primer working from independent cultures of each cell line (with the exception of stock. The Applied Biosystems StepOne thermocycler was set for 40 cy- the M29 cells) were used for microarray analysis. cles with a denaturing temperature of 95 °C for 3 s and annealing/

extension temperatures at 5 °C above the Tm of the primers for 30 s. 2.3. Microarray analysis Also, a ﬁnal melt curve step was performed at 95 °C for 15 s, down to 60 °C for 60 s and a ﬁnal step at 95 °C again for 15 s.

For Illumina microarray analysis, samples were prepared and ana- After amplification, the raw Ct values were averaged for each lyzed by the W. M. Keck Center for Comparative and Functional Geno- specific cell line. The amplified target genes in each hepatoma var- mics in the Roy J. Carver Biotechnology Center at the University of iant were normalized to the respective Gapdh Ct‐value, generating Illinois at Urbana-Champaign. RNA quality was determined using the adelta-Ctvalue(Δ Ct).Folddifferenceingeneexpressionwasde- Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Briefly, termined with a delta delta‐Ct (ΔΔ Ct)calculation.TheΔ Ct of the 300 ng of high quality total RNA were primed with an oligo(dT) primer control cell line (Fg14) was subtracted from the hepatoma variant bearing a T7 promoter, and reverse transcribed in the first-strand cDNA Δ Ct value, generating a ΔΔ Ct value. Using a log base 2 scale, synthesis using the Illumina Total Prep RNA Amplification Kit (Ambion, the calculated difference was placed into the ΔΔ Ct equation: Inc., Austin, TX). Single stranded cDNA was then converted into double 2(ΔCt (sample a) − experimental) − (ΔCt − control). stranded cDNA according to the manufacturer's instructions. The double-stranded cDNA was purified and served as a template in the 2.5. Testing candidate genes for hepatic gene rescue 14-hour in vitro transcription (IVT) reaction. Prior to hybridization, the synthesized biotin labeled cRNA was cleaned up using the same Expression vectors containing candidate genes (purchased from Amplification Kit. After the quality control assessment, 750 ng of Origene, Inc.) were introduced into the H11 variant cell line by lipo- cRNA from each experimental sample along with hybridization con- fection. Briefly, plasmid DNA was mixed with the lipofection reagent trols was hybridized for 16 h to the RatRef-12 v1 Expression BeadChips (Lipofectamine LTX with PLUS reagent; Invitrogen, Inc.) as per kit (Illumina, Inc., San Diego, CA) in the 58 °C Illumina Hybridization Oven. protocol and incubated with freshly plated H11 cells in 6-well dishes. Washing, staining with streptavidin-Cy3 (GE Healthcare Bio-Sciences, After 8 h, the transfection mixtures were replaced with complete me- Piscataway, NJ), and scanning were performed according to the dium. After 48 h, cells were split 1:20 into complete medium plus Illumina Whole-Genome Gene Expression Direct Hybridization Assay 50 μg/ml G418. After 2–3 weeks, G418 resistant clones were pooled Guide (Part#11322355, Rev. A). Two BeadChips were used in this (10–50 clones per pool), expanded, and RNA extracted as described study, each containing twelve arrays. The arrays were scanned using above. RNA was reverse transcribed into cDNA for use in RT-qPCR. an Illumina BeadArray Reader. Each array image was visually screened In addition, pooled cells were plated into complete medium with to discount for signal artifacts, scratches or debris. The images were and without the addition of Adenine–Aminopterin–Thymidine analyzed using the GeneExpression Module (version 1.8.0) of the (AAT) to select for cells which had re-expressed the SERPINA1-Aprt Illumina GenomeStudio software. transgene. Finally, select RNAs were used for single Illumina whole genome microarrays as described above and compared to triplicate 2.4. Reverse transcription quantitative polymerase chain reaction microarrays using RNA isolated from non-transfected H11 cells. (RT-qPCR) 3. Results The standard two-step RT-PCR protocol in the MasterAmp High Fidelity RT-PCR kit (Epicentre Biotechnologies) was used for the syn- Whole genome expression profiling of hepatoma variant cell lines thesis of cDNA from purified RNA. Reverse transcription reactions was carried out in order to identify pathways responsible for the loss 178 G.A. Bulla et al. / Genomics 100 (2012) 176–183 of liver function in these cells. Fg14 hepatoma cells and four previous- difference must be shared by at least 2 of the four variant cell lines ly characterized hepatoma variant cell lines (designated M29, HS2, tested. These criteria suggest a very low number of false positives H11 and M38) derived from Fg14 cells have been previously de- that could interfere with the analysis. scribed [1,13–15,19]. These variant cell lines were derived, without the use of mutagens, at a low frequency (~10− 6) from Fg14 cells based on their inability to drive expression of a SERPINA–Aprt trans- 3.2. Microarray validation gene (see Materials and Methods, Fig. 1). These hepatoma variant cells are relatively stable, but have been shown to revert back to the Validation of microarray results was carried out using reverse- hepatoma phenotype at frequencies ranging from 10− 3 to b10− 8 transcription quantitative PCR (RT-qPCR) on select genes. Expression (based on reactivation of the SERPINA1-Aprt transgene and a few levels for several genes were normalized to Gapdh signals (which var- key liver-specific genes). ied by less than 2-fold) and comparisons between the Fg14 hepatoma In order to determine viability of the whole-genome approach in parental cells and the variant cells were made. Genes monitored in- evaluating the loss of liver-specific gene expression in the cell vari- cluded those encoding transcriptional activation pathway proteins ants, triplicate whole-genome microarray experiments were carried (Wnt4, Hnf4a, Hnf1a and Hhex) and downstream genes (Alb, Fgb, out using RNA isolated from parental and variant cell lines. RNA was Serpina1). Representative samples from each of the cell line RNAs reverse transcribed into cDNA, which was then transcribed into tested were converted to cDNA and amplified in the presence of SYBR RNA, labeled with biotin, and hybridized to the Illumina RatRef-12 Green. Triplicate reactions were done and ΔΔCT values determined. v1Beadchips microarrays. The microarrays were then stained with The data shows strong repression of each of these genes across streptavidin-Cy3 and scanned for fluorescence. most variant cell lines compared to those of the hepatoma parental cells (Fig. 2B). In general, strong agreement was observed between 3.1. Microarray assay reproducibility the fold differences observed in the microarray data (Fig. 2A) and those derived from qPCR. However, the actual fold differences in the To examine reproducibility of the microarray assay, we compared qPCR tended to be much more robust, indicating that the microarray biological replicates for each cell line as well as variability for gene data underrepresented the actual fold differences. This difference is targets within a single microarray (normally 25 to 70 replicate exaggerated in cases in which the hepatoma levels of florescence beads per assay) using P value analysis. Results show that fluorescent are modest (see Table 2 for examples). values of ~100–200 units typically had P values of >0.05, while ex- We also compared our microarray data sets with previously pub- pression values for all genes examined further in the current study lished data from normal rat liver using the same Illumina RatRef-12 (with a >5-fold reduction in expression comparing Fg14 to the vari- Expression BeadChips microarray assays [21] (kind gift of Lie Guo, ant cell types) showed P values of b0.001 in the hepatoma parental National Center of Toxicological Research, FDA) normalized to line and had typical fluorescent values of 200–30,000 units (results Gapdh levels. As expected, the Fg14 tumor cells showed a large de- not shown). gree of misregulated gene expression compared to the normal rat In order to normalize day-to-day variation being responsible for liver, with ~1100 genes showing 5-fold differences in expression. changes in gene expression profiles, RNA was extracted from each However, expression levels of Hnf4a, Hnf1a, Foxa1, Onecut1 and cell line on separate days. Analysis of microarray signals showed Cebp were within 2-fold compared to the Fg14 cells (data not that variation in signal strength for each control gene was less than shown). 2-fold for housekeeping genes Actb and Gapdh across all arrays (data not shown). Variability of expression among replicates was then considered in order to estimate false positive rates. The ratio of A Fold Repression of Gene Expression (Microarray) averaged gene expression across each set of three replicate arrays H11 was done for each cell line. We observed a modest number of genes 100 with >5-fold signal variability within replicates. Among the M38 ~22,500 targets, the number of genes with signal strength that dif- M29 fered by >5-fold between replicates ranged from 0 to 5 genes for 10

the Fg14, H11 and M38 replicates, although 22 genes were identiﬁed (log10) HS2 for the HS2 cells (Data not shown). Based on these data, we applied two criteria when comparing average gene expression between hep- Fold Repression 1 atoma parental cells and the variant cell lines: 1) gene expression Alb Fgb Serpina1 Wnt4 Hnf1a Foxa2 Hnf4 Hhex Gene Name values in the variant cell lines must show a >5-fold difference in average expression compared to parental cells and 2) this >5-fold B Fold Repression of Gene Expression (qPCR) ON ON OFF H11 1. SERPINA1-Aprt 10000 M38 2. SERPINA1-Gpt 1000 M29 HS2 Fado-2 Fg-14 H11, M38 HS2, M29 100 AAT, DAP, Variants HAT 6TX (log10) - APRT + - 10

APRT APRT Fold Repression - + - GPT GPT GPT 1 + + - Alb Fgb Wnt4 Hnf1a Foxa2 Hnf4 Hhex SERPINA1 SERPINA1 SERPINA1 Serpina1 Gene Name Fig. 1. Derivation of hepatoma variant cell lines. Fg-14 cells contain two selectable trans-genes under the control of the liver-speciﬁc human SERPINA1 (alpha-1 anti- Fig. 2. Comparison of microarray data and qPCR data. Fold repression in the hepatoma trypsin) gene promoter. Selection against expression of these genes resulted in the variant cell lines compared to those in Fg14 hepatoma cells is shown for A) Microarray production of hepatoma variant cell lines (e.g. HS2, M38, H11 and M29) which lack ex- analysis and B) qPCR analysis. Fold differences are shown on a Log10 scale. The upper pression of several liver genes. graph shows two orders of magnitude, while the lower graph shows 4 orders. G.A. Bulla et al. / Genomics 100 (2012) 176–183 179

5-10X 150 10-100X

100 >100X

50 # of genes

0 M38 HS2 H11 M29 Cell Line

Fig. 3. Gene repression in hepatoma variant cells. A. The number of genes with 5–10 fold, >10-fold and >100-fold repression levels (compared to Fg14 values) is shown for each cell line. Each cell line showed similar gene repression proﬁles. B. The number of genes repressed >5-fold in all 4 hepatoma variant cell lines (center number), shared by 3 of 4 cell lines (next set of values around center number), 2 of 4 cell lines and those unique to one cell line. A total of 355 genes were identiﬁed as >5-fold repressed in at least one cell line, with 132 repressed in all four cell lines.

3.3. Analysis of gene repression in hepatoma variants repressed in each of the four cell lines, as were other genes in the same serine protease gene cluster such as Serpina6 and Serpina10. Genes that were repressed by >5-fold in the variant cells com- Other repressed gene families included 11 CYP genes, and 7 members pared to the Fg14 parental cells were assessed. Results show a large of the UDP glycosyltransferase gene superfamily (Table 1). number of genes that are strongly repressed in the variant cell lines. Each of the variant cell lines showed a similar number of repressed genes, with >5-fold repressed genes ranging from 221 to 281 genes 3.4. Transcription factor silencing (Fig. 3A). Remarkably, a large number of >10-fold repressed genes fi were also noted, ranging from 92 to 114 genes. Lastly, a small number Previous reports have clari ed the role of core transcriptional reg- – of genes (6–8) were repressed >100-fold in each cell line. ulatory proteins in driving hepatic gene expression [2 4]. Four master fi As described above, due to observed variability in expression values regulators in human hepatocytes have been identi ed, including α β between replicate samples, we asked which genes were repressed >5- Hepatocyte Nuclear Factors HNF1 , HNF3 (FOXA2), HNF4 and fold in at least two of four variant cell lines. A large number of genes ONECUT1 (HNF6). Combined chromatin immunoprecipitation and were identified (268 genes) that are repressed >5-fold in at least two microarray analysis was used to show that these factors operate com- of the four variant cell lines tested (Fig. 3B). Of these, 50% (132) were re- binatorially to activate hepatic gene expression, each binding to – pressed in all four variant cell lines. Remarkably, 65 genes were re- 800 4000 gene promoters [4]. Due to loss of expression of several he- pressed >10-fold in all four variant cell lines (data not shown). In patic genes, and based on previous results analyzing variant cells, we addition, another 69 genes were shared between only two of the four expected these transcription factor genes to be repressed in the cell cell lines and 126 genes were repressed in only one cell line. Not surpris- variants. Unexpectedly, Hnf1a, Hnf4a, and Cebpa showed only modest ingly, the majority of the unique genes were identified in the HS2 cells levels of repression (2-to 4.5 fold) and Foxa2 showed no repression. fi (which showed the highest level of replicate variability) and M29 cells Only Hnf6 (Onecut1) showed a strong repression pro le (Table 2). (whose expression profile was limited to a single microarray). Of the 200 genes that were repressed >5-fold, many of these are 3.5. Identification of candidate genes well-described markers of liver function, including Alb, Agt, Asl, Aldob, Ass1, Pah, Fgg, Fgb, Hpx, Pah, Pck1, Tat, and Ttr. As expected, To identify candidate genes, we required that two criteria were because the human SERPINA1 promoter sequence was incorporated met: 1) >5-fold repression of gene expression in 2 of 4 variant cell into the selection strategy, the rat Serpina1 gene was also strongly lines compared to those levels in the parental Fg14 cells and 2) the

Table 1 Genes repressed >5-fold in 3 of 4 hepatoma variant cell lines. Of the 200 targets identiﬁed, only those with identiﬁed genes or LOC numbers (157 total) are shown.

ABCC2⁎ BHMT⁎ CYP2B15⁎ FBP2 HHEX MAOB REEP6 TFPI⁎ ABCG5⁎ C2⁎ CYP2C23⁎ FETUB⁎ HMGCS2⁎ MDK RNASE4⁎ THPO ADH⁎ C3⁎ CYP2C6 FGB⁎ HNRPAB METTL7B S100G⁎ TMEM176B⁎ ADM2⁎ C4BPA⁎ CYP2D2⁎ FGG⁎ HPN MTTP⁎ SEC16B TTR AGMAT⁎ C4BPB CYP2D5 FMO1⁎ HPX⁎ MUCDHL⁎ SECTM1 UGT1A1⁎ AGT⁎ C5 CYP2E1⁎ FUT2 HRSP12 OASL SELENBP1⁎ UGT2B⁎ AGXT2 C9⁎ CYP2F4 G6PC⁎ HSD17B13⁎ ONECUT1⁎ SERPINA1⁎ UGT2B17 AHSG⁎ CCND1 CYP3A18⁎ GC⁎ HSD17B2 OPLAH SERPINA10⁎ UGT2B36⁎ AKR7A3 CDH17⁎ CYP4F1⁎ GCLM IGFBP1⁎ PAH⁎ SERPINA3N UGT2B36 ALB CESL1 CYP8B1⁎ GLIPR1 ITIH3 PCK1⁎ SERPINA6⁎ UGT2B37⁎ ALDH1A7 CFI⁎ DEPDC7 GNG11 ITM2A PCSK6 SERPINC1⁎ UGT2B5⁎ ALDOB⁎ CHDH DIO1 GPD1 KMO⁎ PCSK9 SERPING1 LOC688266 AMT⁎ CPA1 ELA1 GPT⁎ KNG1⁎ PLG SLC16A6 LOC691259⁎ ANG1 CPN1 ENPP3 GPX3 KRT8⁎ PON1⁎ SLC17A1⁎ APON⁎ CREB3L3⁎ EPHX1⁎ GSTA2⁎ LCAT PPP1R14A SLC22A7 AQP11 CRYAB F2⁎ GULO LGALS2⁎ PTGDS SOAT2⁎ AQP8 CTH FABP1⁎ GYS2 LGALS4⁎ PTGR1 SPR ASL CTRC FAM25A⁎ HAAO⁎ LINCR PZP⁎ SSTR2 ASS CXCL11 FBLIM1 HAL⁎ LIPC⁎ QPRT⁎ SULT1B1 BAAT CYP17A1⁎ FBP1⁎ HAMP⁎ MAF RBP1⁎ TAT⁎

⁎ Average repression level >10-fold. 180 G.A. Bulla et al. / Genomics 100 (2012) 176–183

Table 2 30 (67%) of the genes for which promoter sequences are known (45 Expression levels of transcription factors in the Fg14 cells and variant cell lines as de- genes known, 11 unknown) contain HNF1α binding sites (Fig. 4C). termined by microarray data. Average signals obtained for Fg14 and H11 cells for a Not unexpectedly, the majority (37 of 56) of genes were identified set of liver-enriched transcription factors are shown. In addition, fold differences in expression values are given for each cell line. in our original analysis of gene repression in the variant cells, with 12 of the remaining genes being expressed at modest levels (b500 Average values Fold repression units) in the Fg14 cells, precluding identification using our stringent Transcription factor Fg14 H11 M38 H11 HS2 M29 5-fold cut-off. Cebpa 250.0 100.7 2.1 2.5 1.5 2.1 Foxa2 3417.9 5632.0 0.6 0.6 1.1 1.1 3.7. Comparison of gene profiles with that of hepatoma×fibroblast whole Hnf1a 455.2 107.7 3.4 4.2 4.5 4.2 cell hybrids Hnf4 247.0 88.6 2.9 2.8 2.9 2.7 Onecut1 2011.9 134.6 10.7 14.9 5.3 16.9 Genome-wide repression of liver-specific gene expression in hepatoma×fibroblast somatic cell hybrids was recently reported [22]. known or proposed function of the gene product influences transcrip- We compared repression profiles of the cell hybrids with that of the tion (either as a transcription factor or is a member of a signaling hepatoma variant cell lines in the current study with regard to candi- pathway that alters transcription). Of the 268 genes that met the date genes identified. Three candidate genes (Hhex, Igfbp1, and first criteria, 19 of the genes identified were of unknown function. Onecut) were identified in both data sets, but the majority of candidate We identified 14 genes that fit these criteria (Table 3). genes were non-overlapping. Both data sets showed 3–4 fold repres- Importantly, several of the identified genes are members of tran- sion of transcription factors Hnf4a and Hnf1a, a level of repression scriptional activation pathways, suggesting that they contribute to that did not meet the 5-fold cutoff used in this study (Fig. 5). maintaining the hepatic phenotype. Repressed genes include liver- enriched factors (Hhex, Onecut1), two developmental pluripotency- 4. Discussion associated genes (Dppa4 and Dppa5), Creg1 (cellular repressor of E1A-stimulated genes), Creb3L3 (cAMP responsive element binding Dedifferentiated hepatoma cells have long been studied in order protein 3-like 3), and Tcfap2b (AP-2 beta), which has been reported to define pathways that determine liver identity [6,12,23–25]. The to regulate the expression of genes required for development of ecto- hepatoma variants used in this study were selected based on their in- dermal tissues. These data suggest that a high level of reprogramming ability to drive the human SERPINA1 promoter fused to selectable occurred in variant cells, leading to a dramatic loss of liver‐specific marker genes Aprt and Gpt [19]. These cells have been reported to gene expression as well as activation of a smaller set of genes. lack expression of Hnf1a and Hnf4a as well as several downstream genes [1,14,18]. Also, defective apoptotic signaling has been observed 3.6. Forced expression of HNF1α in H11 variant cells partially rescues the in these cells [24,26]. As expected from previous reports [14,15,18], hepatic phenotype the rat Serpina1 gene was strongly repressed (55–90-fold) in each of the variant cell lines tested compared to the parental Fg14 cells. We next tested several candidate genes for their ability to restore Unexpectedly, Hnf1a and Hnf4a (encoding transcription factors hepatic gene expression. As a first approach, we asked whether the reported to activate Serpina1 expression), did not meet our criteria SERPINA1-aprt transgene (used to select the variants) was reacti- of 5-fold repression, instead showing a modest 3 to 4-fold repression. vated, allowing survival in AAT selective medium. Of the eight genes This is likely due to the low level expression of (300–400 units) of tested (Onecut1, Hnf1a, Igfbp1, Hhex, Creb3l3, Dppa4, Mdk and these factors in the hepatoma parental cells as measured by the Creg1), only Hnf1a-transfected cells survived in AAT selection microarrays. Because signal to noise levels in the microarray ap- (Fig. 4A), in agreement with previously published data 19. qPCR proach 100 units, fold differences would be expected to be muted. In- analysis verified that Hnf1a expression was increased 100-fold to deed, qPCR analysis of relative expression levels showed a ~100-fold levels observed in the original hepatoma cells. Elevated levels of reduction in expression of these genes. Thus we conclude that our Fgb, Hhex, and Serpina1 RNA were also observed (Fig. 4B). Single criteria may not be optimal when considering genes expressed at whole genome microarray analysis of the Hnf1a-transfected cells low levels, and we likely failed to detect some candidate genes was compared to triplicate H11 data. Remarkably, 56 genes were using the 5-fold repression criteria. found to be activated at least 3 fold in the Hnf1a transfected cells We observed very little variability in expression levels when com- that were not activated in H11 cells transfected with control genes paring replicate microarrays for some cell lines (Fg14, H11, M38), but (Fig. 4C). Promoter analysis using Genecard analysis indicates that substantial variability in the HS2 replicates. This latter variability

Table 3 List of candidate genes identiﬁed from list of repressed genes.

Fold repression Gene name Description

M38 HS2 H11 M29

6.2 1.7 3.6 5.5 BHLHB2 Basic helix–loop–helix domain containing, Class 2B 12.9 15.0 14.6 13.7 CREB3L3 cAMP responsive element binding protein 3-like 3 3.6 6.4 2.7 5.9 CREG E1A-stimulated genes (predicted) 17.4 18.7 18.7 16.7 DPPA4 Developmental pluripotency associated 4 (LOC680293) (predicted) 12.2 11.9 11.9 10.1 DPPA5 Similar to DPPA5 developmental pluripotency associated 5 (RGD1564306) (predicted) 9.3 8.5 7.9 8.8 GAS2 Growth arrest-speciﬁc 2. (RGD1563167) (predicted) 5.1 6.0 2.8 5.8 HHEX Hematopoietically-expressed homeobox protein 118.6 128.0 131.9 107.4 IGFBP1 Insulin like growth factor binding protein 1 8.3 9.6 0.1 7.4 MDK Midkine (neurite growth‐promoting factor 2) 10.7 5.3 14.9 16.9 ONECUT1 One cut homeobox 1 7.0 14.9 2.1 43.3 RNF125 Ring Finger protein 125 (predicted) 9.4 10.3 10.6 5.3 SEC16b SEC16 homolog B. (S. cerevisiae homolog) 19.1 19.7 19.7 19.0 STRA8 Stimulated by Retinoic Acid 8 (RGD1562852) (predicted) 3.5 9.5 1.0 21.0 TCFAP2B Transcription factor AP-2 beta G.A. Bulla et al. / Genomics 100 (2012) 176–183 181

A Cell hybrids Variants SERPINA1 promoter APRT -680 +1 BHLHB2 CREB3L3 FOXA2 CREG GHR HHEX DPPA4 GHRH IGFBP1 DPPA5 LEF1 ONECUT1 GAS2 WIF1 (HNF4A) MDK WNT4 (HNF1A) RNF125 SEC16B STRA8 TCFAP2B

B 100 Fig. 5. Comparison of candidate genes identiﬁed from whole genome analysis of hepatoma×ﬁbroblast whole cell hybrids [22] and the current study of hepatoma vari- qPCR data ants. Genes shown were repressed >5 fold compared to hepatoma parental cells and are members of transcriptional activation pathways. Hnf4 and Hnf1a were only repressed 3–4 fold in both cell hybrids and variant cells, but are included due to previous studies showing their role in activation of liver gene expression. 10

higher false positive rate (though it could not be measured directly). Indeed, the M29 cells had 51 repressed genes identified that were Fold Activation (log10) unique to the M29 cells. We reasoned that the genes identified that 1 Serpina1 Fgb Hhex Hnf1a were shared between variant cell lines would limit the number of false positives. However, we also realize that this strategy could also result in a failure to identify genes driving this process, as each cell Microarray data 10.9 11.7 1.9 16.8 (Fold activation) line has a different reversion rate, as well as response to the introduction of Hnf1a [14,15,18]. Therefore, each variant cell line likely has distinct defects that lead to loss of liver gene expression. C 30 with an HNF1α binding site 56 genes It has been previously reported that these variant cell lines are α activated >3-fold 15 without an HNF1 binding site able to revert back to hepatoma cells when positive selection on the 11 unknown Aprt or Gpt genes is applied [14]. In each case, surviving cells rescued expression of Hnf4a, Hnf1a and Serpina1 genes. Thus, the robust phe- AFP* CYP2C23 LGALS2* SERPINA1** notypes of the variant cells do not appear to be fixed, but rather cells AGT CYP2E1 LGALs2 SERPINA6 are able to revert to hepatoma phenotypes at low frequency. The re- AKR7A3* DDAH1 LSS SLC40A1 sults presented here show that expression levels of over 300 genes ALDOB DNASE1L3 LOC300024** SOSTDC1** APOM EPB4.1L4A#* LOC363060 TRPM6# are dramatically affected (~60 of these by greater than 10-fold). We C4BPA** EPCAM LOC497769 TTR observed 132 genes that were repressed in all four cell lines, and an C9 F2 MAF UGDH additional 68 genes shared among 3 of 4 cell lines. Genes affected in- CCL20 FGB** MAOB UGT1A1 CDH17* FXYD2** MSMO UGT2B cluded a large set of well-studied liver-enriched genes (e.g. Alb, Fgb, CFI** G6PC NKD1# UGT2B7* Ttr, Pck1, Tat, Pah) as well as many genes not previously described CHDH GC** PTGR1 UGT2B17 as being liver-specific. In addition, 6 members of the serpin gene fam- COL5A2 GSTA2* RTP3# UGT2B36 CPN1 HNF1A** S100A11 UGT2B5* ily were repressed, as were 11 members of the CYP family proteins CTGF IGFBP1* SCP2 YC2 and 7 members of the UDP glycosyltransferase gene superfamily. Marsden and Fournier [27] showed that the serpin gene cluster * Activation level >5-fold **Activation level >10-fold at 14q32.1 contains three liver-specific serpin genes, SERPINA1, # Predicted gene identity SERPINA6 and SERPINA10. Deletion of a cluster of DNaseI hypersensi- tive sites in this locus resulted in loss of expression of each gene and Fig. 4. Testing candidate genes for hepatic gene rescue. Expression vectors containing candidate genes and a neo marker gene were stably introduced into H11 variant cells decreased histone acetylation that suggested extensive remodeling of using G418 selection. Clones were pooled and either plated or expanded and RNA the chromosomal region [28]. Each of these genes was repressed extracted. A) Pooled tansfectants were plated into complete medium with and without >10-fold in all four variant cell lines tested in the current study, con- AAT at two densities (200 and 2000 cells per well) to select for re-expression of the sistent with the idea of common regulatory control of these genes. In SERPINA1-Aprt transgene. B) qPCR analysis was carried out on RNA extracted from α the Hnf1a-transfected cells to monitor expression of HNF1a as well as other addition, introduction of the HNF4/HNF1 transactivation cascade liver-enriched genes. Fold activation over levels in non-transfected H11 cells shown. into the H11 variant cells resulted in extensive remodeling and re- Microarray levels (described in C) are included for these genes. C) Whole-genome mi- expression of an introduced human serpin locus [29]. As described croarray analysis was carried on the Hnf1a-transfected cells and genes identified that above, microarray data showed only modest repression on these fac- were activated >3-fold (ignoring genes activated by introduction of control genes). tors in the variant cell lines, yet qPCR showed robust repression of Promoter analysis (GeneCard) was used to identify which genes contain HNF1α binding sites in proximal promoter regions. Hnf1a. Other well-known liver-enriched factors, Foxa2 and Cepba, showed less than 2.5 fold repression in any variant cell line, although Onecut1 showed strong repression in all variant cell lines. likely led to the identification of a number of false positives in the HS2 We screened the panel of genes that were repressed greater than cells (a false positive rate of 20 genes expected from replicate analy- 5-fold in at least two of the four cell variants for reported gene regu- sis). Indeed, HS2 cells showed 48 unique repressed genes, compared latory activity. We reasoned that genes responsible for the observed to 4 and 25 unique genes in the M38 and H11 cells, respectively. Like- repression phenotype would likely also be repressed in the variant wise, because we had only one sample of M29 cells, we expected a cells. Fourteen candidate genes were identified that encode proteins 182 G.A. Bulla et al. / Genomics 100 (2012) 176–183 involved in gene regulation, including Bhlhe40, Creb3l3, Creg, Dppa4, will be carried out to determine whether other proteins can play a Dppa5, Gas2, Hhex, Igfbp1, Mdk, Onecut1, Rnf125, Sec16b, Stra8 and role in reactivation of the liver phenotype. Tcfap2b. Each of these genes were moderately to highly expressed In conclusion, whole-genome analysis gene repression in hepato- in hepatoma cells (600U–10,000U) but strongly repressed in the var- ma variant cell lines indicates that distinct gene sets are repressed in iant cell lines. Hhex is a homeodomain transcription factor expressed each variant cell line, but that a large degree of overlap is observed. in the anterior definitive endoderm [30] and is required for proper Genetic rescue experiments have shown that HNF1a can restore mammalian development. Loss-of-function Hhex knockout mice die much of the lost liver gene expression. Further testing of candidate during embryogenesis with several defects, including altered liver de- genes is currently being conducted in order to identify genes that velopment [31]. The role of Igfbp1 has been shown to both disrupt play a critical role in establishing and maintaining hepatic function and promote IGF-1 action [32]. Creg1 antagonizes adenoviral E1A ac- in the mammalian liver. tivity and has been suggested to affect transcriptional control of cell growth and differentiation [33], although reports in liver function Acknowledgments have not been described. Bhlhe40 (Dec1/Stra13) is a basic helix– – loop helix (bHLH) transcriptional repressor widely expressed in This work was supported by the National Science Foundation both embryonic and adult tissues and has been reported to strongly grant # 0841653 to G. A. Bulla. We thank Lie Guo, National Center suppress Pepck expression in H4IIE hepatoma cells [34] as well as af- of Toxicological Research, FDA, for generously providing microarray fect circadian rhythms [35]. Dppa4 and 5 are developmental data generated from normal rat liver samples. pluripotency associated genes reported to be active in embryogenesis, including induction of mouse embryonic stem cells into primitive References ectoderm lineage [36,37]. Gas2 is a protein thought to be involved in growth arrest as well as development [38] and was one of the four [1] G.A. Bulla, Selective loss of the hepatic phenotype due to the absence of a tran- genes (Gas2, Igfbp2, Nr1i3, and Ccne2) that were differentially scriptional activation pathway, Somat. Cell Mol. Genet. 23 (1997) 185–201. [2] S. Cereghini, Liver-enriched transcription factors and hepatocyte differentiation, expressed in tetraploid and octoploid cells [39]. Mdk (Midkine) is a FASEB J. 10 (1996) 267–282. heparin-binding growth factor reported to affect the proliferation [3] R.H. Costa, V.V. Kalinichenko, A.X. Holterman, X. Wang, Transcription factors in liver and migration of several cell types [40] and has been shown to play development, differentiation and regeneration, Hepatology 38 (2003) 1331–1347. a role in liver regeneration [41]. RNF125 is a ubiquitin ligase that [4] D.T. Odom, R.D. Dowell, E.S. Jacobsen, L. Nekludova, P.A. Rolfe, T.W. Danford, D.K. Gifford, E. Fraenkel, G.I. Bell, R.A. Young, Core transcriptional regulatory circuitry has also been reported to be a positive regulator of T-cell activation in human hepatocytes, Mol. Syst. Biol. 2 (2006) 2006.0017. [42] and to inhibit HIV-1 viral transcription [43]. Stra8 is a retinoic [5] G.A. Bulla, Hepatocyte nuclear factor-4 prevents silencing of hepatocyte nuclear fi acid-inducible gene active in early mammalian embryogenesis [44]. factor-1 expression in hepatoma × broblast cell hybrids, Nucleic Acids Res. 25 (1997) 2501–2508. Finally, Tfap2b (AP-2 related transcription factor) is a transcription [6] C.J. Kuo, P.B. Conley, L. Chen, F.M. Sladek, J.E. Darnell Jr., G.R. Crabtree, A transcrip- activator involved in early embryogenesis [45]. tional hierarchy involved in mammalian cell-type specification, Nature 355 We previously reported a genome-wide analysis of gene silencing (1992) 457–461. fi [7] S.A. Duncan, M.A. Navas, D. Dufort, J. Rossant, M. Stoffel, Regulation of a transcrip- in hepatoma× broblast whole cell hybrids [22]. We reported ap- tion factor network required for differentiation and metabolism, Science 281 proximately 300 genes that were repressed >5-fold. We noted a (1998) 692–695. large degree of overlap of genes identified in the current study of [8] C.S. Lee, J.R. Friedman, J.T. Fulmer, K.H. Kaestner, The initiation of liver development is dependent on Foxa transcription factors, Nature 435 (2005) 944–947. cell variants, although a comprehensive comparison is not possible [9] D.T. Odom, N. Zizlsperger, D.B. Gordon, G.W. Bell, N.J. Rinaldi, H.L. Murray, T.L. due to the use of different microarray formats. However, certain con- Volkert, J. Schreiber, P.A. Rolfe, D.K. Gifford, E. Fraenkel, G.I. Bell, R.A. Young, Con- clusions can be made. First, the majority of well-known markers of trol of pancreas and liver gene expression by HNF transcription factors, Science 303 (2004) 1378–1381. liver function are repressed in both cell systems. The remarkable [10] D. Cassio, E. Rogier, G. Feldmann, M.C. Weiss, Plasma-protein production by rat numbers of genes that are strongly repressed in both cell models sug- hepatoma cells in culture, their variants and revertants, Differentiation 30 gest that extensive remodeling occurs upon loss of liver function. No- (1986) 220–228. tably, only a modest number of candidate genes (Oncecut1, Hhex, [11] J. Deschatrette, E.E. Moore, M. Dubois, M.C. Weiss, Dedifferentiated variants of a rat hepatoma: reversion analysis, Cell 19 (1980) 1043–1051. Igfbp1 as well as Hnf1a and Hnf4a if the threshold is lowered to [12] J. Deschatrette, M.C. Weiss, Characterization of differentiated and 3-fold) are shared between these two cell model systems. Candidate dedifferentiated clones from a rat hepatoma, Biochimie 56 (1974) 1603–1611. genes identified in the previous study fell into three categories: Wnt [13] G.A. Bulla, E. Givens, S. Brown, B. Oladiran, D. Kraus, A common regulatory locus affects both HNF4/HNF1a pathway activation and sensitivity to LPS-mediated ap- pathway, growth hormone signaling and the HNF network [22]. Can- optosis in rat hepatoma cells, J. Cell Sci. 114 (2001) 1205–1212. didate genes identified in the current study included 14 genes that [14] G.A. Bulla, R.E. Fournier, Genetic analysis of a transcriptional activation pathway appear to have little in common (see Table 3), except that they appear by using hepatoma cell variants, Mol. Cell. Biol. 14 (1994) 7086–7094. [15] G.A. Bulla, D.A. Kraus, Dissociation of the hepatic phenotype from HNF4 and to be involved in transcriptional and/or developmental regulatory HNF1a expression, Biosci. Rep. 24 (2005) 595–608. pathways. [16] P. Rollini, R.E. Fournier, The HNF-4/HNF-1alpha transactivation cascade regulates This suggests that distinct strategies are used to silence gene ex- gene activity and chromatin structure of the human serine protease inhibitor gene cluster at 14q32.1, Proc. Natl. Acad. Sci. U. S. A. 96 (1999) 10308–10313. pression in these two cell models. Given the very distinct strategies [17] P. Rollini, L. Xu, R.E. Fournier, Partial activation of gene activity and chromatin re- used to generate cell types with repressed liver gene expression, per- modeling of the human 14q32.1 serpin gene cluster by HNF-1 alpha and HNF-4 in haps this is not too surprising. However, the majority of the hundreds fibroblast microcell hybrids, Somat. Cell Mol. Genet. 25 (1999) 207–221. fi [18] G.A. Bulla, J.R. Batanian, H.M. Young, N. Bradley, N. Kazi, B.R. Haddad, Rescue of of downstream liver-speci c genes are repressed in both model sys- the HNF4/HNF1 pathway in hepatoma variant cells containing human chromo- tems, indicating that the same liver-specific pathways are impacted some 12, Genomics 54 (1998) 398–407. (data not shown). [19] G.A. Bulla, R.E. Fournier, Direct selection of hepatoma cell variants deficient in – Our approach was to use rat whole genome microarrays to analyze alpha 1-antitrypsin gene expression, Somat. Cell Mol. Genet. 18 (1992) 361 370. [20] G.A. Bulla, Extinction of a1-antitrypsin expression in cell hybrids is independent the well characterized rat hepatoma cells and hepatoma variants to of HNF1a and HNF4 and involves both promoter and internal DNA sequences, understand the full extent of gene silencing in cell variants as well Nucleic Acids Res. 27 (1999) 1190–1197. as to identify candidate genes involved in the process. Our results [21] L. Guo, Q. Xia, S. Dial, P.C. Chan, P. Fu, Analysis of gene expression changes of drug metabolizing enzymes in the livers of F344 rats following oral treatment with suggest that gene silencing is extensive, with hundreds of genes kava extract, Food Chem. Toxicol. 47 (2009) 433–442. strongly repressed in the variant cells. The ability of HNF1α to acti- [22] G.A. Bulla, Q. Luong, S. Shrestha, S. Reeb, S. Hickman, Genome-wide analysis of vate expression of 60 genes, most with HNF1a binding sites in their hepatic gene silencing in mammalian cell hybrids, Genomics 96 (2010) 323–332. [23] S. Cereghini, M. Blumenfeld, M. Yaniv, A liver-specific factor essential for albumin promoters, suggests that a single factor can cause dramatic repro- transcription differs between differentiated and dedifferentiated rat hepatoma gramming of the variant cells. Testing of additional candidate genes cells, Genes Dev. 2 (1988) 957–974. G.A. Bulla et al. / Genomics 100 (2012) 176–183 183

[24] D. Kraus, G.A. Bulla, Defective NF-kB signaling in dedifferentiated hepatoma cells, [36] H. Masaki, T. Nishida, S. Kitajima, K. Asahina, H. Teraoka, Developmental Somat. Cell Mol. Genet. 27 (2002) 275–286. pluripotency-associated 4 (DPPA4) localized in active chromatin inhibits mouse [25] G.F. Spath, M.C. Weiss, Hepatocyte nuclear factor 4 expression overcomes repres- embryonic stem cell differentiation into a primitive ectoderm lineage, J. Biol. sion of the hepatic phenotype in dedifferentiated hepatoma cells, Mol. Cell. Biol. Chem. 282 (2007) 33034–33042. 17 (1997) 1913–1922. [37] S.K. Kim, M.R. Suh, H.S. Yoon, J.B. Lee, S.K. Oh, S.Y. Moon, S.H. Moon, J.Y. Lee, J.H. [26] A.K. Schmitz, D.M. Kraus, G.A. Bulla, Tissue-specificity of apoptosis in hepatoma Hwang, W.J. Cho, K.S. Kim, Identification of developmental pluripotency derived cell lines, Apoptosis 9 (2004) 369–375. associated 5 expression in human pluripotent stem cells, Stem Cells 23 (2005) [27] M.D. Marsden, R.E. Fournier, Organization and expression of the human serpin 458–462. gene cluster at 14q32.1, Front. Biosci. 10 (2005) 1768–1778. [38] K.K. Lee, M.K. Tang, D.T. Yew, P.H. Chow, S.P. Yee, C. Schneider, C. Brancolini, gas2 [28] E.W. Baxter, W.J. Cummings, R.E. Fournier, Formation of a large, complex domain is a multifunctional gene involved in the regulation of apoptosis and of histone hyperacetylation at human 14q32.1 requires the serpin locus control chondrogenesis in the developing mouse limb, Dev. Biol. 207 (1999) 14–25. region, Nucleic Acids Res. 33 (2005) 3313–3322. [39] P. Lu, S. Prost, H. Caldwell, J.D. Tugwood, G.R. Betton, D.J. Harrison, Microarray [29] P. Rollini, R.E.K. Fournier, The HNF4/HNF1a transactivation cascade regulates analysis of gene expression of mouse hepatocytes of different ploidy, Mamm. gene activity and chromatin structure of the human serine protease inhibitor Genome 18 (2007) 617–626. cluster at 14q32.1, Proc. Natl. Acad. Sci. U. S. A. 96 (1999) 10308–10313. [40] Y. Xu, X. Qu, X. Zhang, Y. Luo, Y. Zhang, K. Hou, Y. Liu, Midkine positively [30] P.Q. Thomas, A. Brown, R.S. Beddington, Hex: a homeobox gene revealing peri- regulates the proliferation of human gastric cancer cells, Cancer Lett. 279 implantation asymmetry in the mouse embryo and an early transient marker of (2009) 137–144. endothelial cell precursors, Development 125 (1998) 85–94. [41] K. Ochiai, H. Muramatsu, S. Yamamoto, H. Ando, T. Muramatsu, The role of mid- [31] H. Hallaq, E. Pinter, J. Enciso, J. McGrath, C. Zeiss, M. Brueckner, J. Madri, H.C. kine and pleiotrophin in liver regeneration, Liver Int. 24 (2004) 484–491. Jacobs, C.M. Wilson, H. Vasavada, X. Jiang, C.W. Bogue, A null mutation of Hhex re- [42] A.L. Giannini, Y. Gao, M.J. Bijlmakers, T-cell regulator RNF125/TRAC-1 belongs to a sults in abnormal cardiac development, defective vasculogenesis and elevated novel family of ubiquitin ligases with zinc fingers and a ubiquitin-binding Vegfa levels, Development 131 (2004) 5197–5209. domain, Biochem. J. 410 (2008) 101–111. [32] V. Hwa, Y. Oh, R.G. Rosenfeld, The insulin-like growth factor-binding protein [43] S. Shoji-Kawata, Q. Zhong, M. Kameoka, Y. Iwabu, S. Sapsutthipas, R.B. Luftig, K. (IGFBP) superfamily, Endocr. Rev. 20 (1999) 761–787. Ikuta, The RING finger ubiquitin ligase RNF125/TRAC-1 down-modulates HIV-1 [33] E. Veal, M. Eisenstein, Z.H. Tseng, G. Gill, A cellular repressor of E1A-stimulated replication in primary human peripheral blood mononuclear cells, Virology 368 genes that inhibits activation by E2F, Mol. Cell. Biol. 18 (1998) 5032–5041. (2007) 191–204. [34] K. Yamada, H. Ogata-Kawata, K. Matsuura, K. Miyamoto, SHARP-2/Stra13/DEC1 as [44] V. Sapin, P. Bouillet, M. Oulad-Abdelghani, B. Dastugue, P. Chambon, P. Dolle, Dif- a potential repressor of phosphoenolpyruvate carboxykinase gene expression, ferential expression of retinoic acid-inducible (Stra) genes during mouse placen- FEBS Lett. 579 (2005) 1509–1514. tation, Mech. Dev. 92 (2000) 295–299. [35] A. Nakashima, T. Kawamoto, K.K. Honda, T. Ueshima, M. Noshiro, T. Iwata, K. [45] M. Moser, A. Imhof, A. Pscherer, R. Bauer, W. Amselgruber, F. Sinowatz, F. Fujimoto, H. Kubo, S. Honma, N. Yorioka, N. Kohno, Y. Kato, DEC1 modulates the Hofstadter, R. Schule, R. Buettner, Cloning and characterization of a second AP-2 circadian phase of clock gene expression, Mol. Cell. Biol. 28 (2008) 4080–4092. transcription factor: AP-2 beta, Development 121 (1995) 2779–2788.